Commit | Line | Data |
---|---|---|
5b497af4 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
51580e79 | 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 3 | * Copyright (c) 2016 Facebook |
fd978bf7 | 4 | * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io |
51580e79 | 5 | */ |
838e9690 | 6 | #include <uapi/linux/btf.h> |
aef2feda | 7 | #include <linux/bpf-cgroup.h> |
51580e79 AS |
8 | #include <linux/kernel.h> |
9 | #include <linux/types.h> | |
10 | #include <linux/slab.h> | |
11 | #include <linux/bpf.h> | |
838e9690 | 12 | #include <linux/btf.h> |
58e2af8b | 13 | #include <linux/bpf_verifier.h> |
51580e79 AS |
14 | #include <linux/filter.h> |
15 | #include <net/netlink.h> | |
16 | #include <linux/file.h> | |
17 | #include <linux/vmalloc.h> | |
ebb676da | 18 | #include <linux/stringify.h> |
cc8b0b92 AS |
19 | #include <linux/bsearch.h> |
20 | #include <linux/sort.h> | |
c195651e | 21 | #include <linux/perf_event.h> |
d9762e84 | 22 | #include <linux/ctype.h> |
6ba43b76 | 23 | #include <linux/error-injection.h> |
9e4e01df | 24 | #include <linux/bpf_lsm.h> |
1e6c62a8 | 25 | #include <linux/btf_ids.h> |
47e34cb7 | 26 | #include <linux/poison.h> |
bd5314f8 | 27 | #include <linux/module.h> |
f42bcd16 | 28 | #include <linux/cpumask.h> |
1fda5bb6 | 29 | #include <linux/bpf_mem_alloc.h> |
680ee045 | 30 | #include <net/xdp.h> |
51580e79 | 31 | |
f4ac7e0b JK |
32 | #include "disasm.h" |
33 | ||
00176a34 | 34 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
91cc1a99 | 35 | #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ |
00176a34 JK |
36 | [_id] = & _name ## _verifier_ops, |
37 | #define BPF_MAP_TYPE(_id, _ops) | |
f2e10bff | 38 | #define BPF_LINK_TYPE(_id, _name) |
00176a34 JK |
39 | #include <linux/bpf_types.h> |
40 | #undef BPF_PROG_TYPE | |
41 | #undef BPF_MAP_TYPE | |
f2e10bff | 42 | #undef BPF_LINK_TYPE |
00176a34 JK |
43 | }; |
44 | ||
1fda5bb6 YS |
45 | struct bpf_mem_alloc bpf_global_percpu_ma; |
46 | static bool bpf_global_percpu_ma_set; | |
47 | ||
51580e79 AS |
48 | /* bpf_check() is a static code analyzer that walks eBPF program |
49 | * instruction by instruction and updates register/stack state. | |
50 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
51 | * | |
52 | * The first pass is depth-first-search to check that the program is a DAG. | |
53 | * It rejects the following programs: | |
54 | * - larger than BPF_MAXINSNS insns | |
55 | * - if loop is present (detected via back-edge) | |
56 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
57 | * - out of bounds or malformed jumps | |
58 | * The second pass is all possible path descent from the 1st insn. | |
8fb33b60 | 59 | * Since it's analyzing all paths through the program, the length of the |
eba38a96 | 60 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
61 | * insn is less then 4K, but there are too many branches that change stack/regs. |
62 | * Number of 'branches to be analyzed' is limited to 1k | |
63 | * | |
64 | * On entry to each instruction, each register has a type, and the instruction | |
65 | * changes the types of the registers depending on instruction semantics. | |
66 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
67 | * copied to R1. | |
68 | * | |
69 | * All registers are 64-bit. | |
70 | * R0 - return register | |
71 | * R1-R5 argument passing registers | |
72 | * R6-R9 callee saved registers | |
73 | * R10 - frame pointer read-only | |
74 | * | |
75 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
76 | * and has type PTR_TO_CTX. | |
77 | * | |
78 | * Verifier tracks arithmetic operations on pointers in case: | |
79 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
80 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
81 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
82 | * and 2nd arithmetic instruction is pattern matched to recognize | |
83 | * that it wants to construct a pointer to some element within stack. | |
84 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
85 | * (and -20 constant is saved for further stack bounds checking). | |
86 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
87 | * | |
f1174f77 | 88 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 89 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 90 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
91 | * |
92 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
93 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
94 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
95 | * |
96 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
97 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
98 | * | |
99 | * registers used to pass values to function calls are checked against | |
100 | * function argument constraints. | |
101 | * | |
102 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
103 | * It means that the register type passed to this function must be | |
104 | * PTR_TO_STACK and it will be used inside the function as | |
105 | * 'pointer to map element key' | |
106 | * | |
107 | * For example the argument constraints for bpf_map_lookup_elem(): | |
108 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
109 | * .arg1_type = ARG_CONST_MAP_PTR, | |
110 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
111 | * | |
112 | * ret_type says that this function returns 'pointer to map elem value or null' | |
113 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
114 | * 2nd argument should be a pointer to stack, which will be used inside | |
115 | * the helper function as a pointer to map element key. | |
116 | * | |
117 | * On the kernel side the helper function looks like: | |
118 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
119 | * { | |
120 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
121 | * void *key = (void *) (unsigned long) r2; | |
122 | * void *value; | |
123 | * | |
124 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
125 | * [key, key + map->key_size) bytes are valid and were initialized on | |
126 | * the stack of eBPF program. | |
127 | * } | |
128 | * | |
129 | * Corresponding eBPF program may look like: | |
130 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
131 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
132 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
133 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
134 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
135 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
136 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
137 | * | |
138 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
139 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
140 | * and were initialized prior to this call. | |
141 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
142 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
143 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
8fb33b60 | 144 | * returns either pointer to map value or NULL. |
51580e79 AS |
145 | * |
146 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
147 | * insn, the register holding that pointer in the true branch changes state to | |
148 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
149 | * branch. See check_cond_jmp_op(). | |
150 | * | |
151 | * After the call R0 is set to return type of the function and registers R1-R5 | |
152 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
153 | * |
154 | * The following reference types represent a potential reference to a kernel | |
155 | * resource which, after first being allocated, must be checked and freed by | |
156 | * the BPF program: | |
157 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
158 | * | |
159 | * When the verifier sees a helper call return a reference type, it allocates a | |
160 | * pointer id for the reference and stores it in the current function state. | |
161 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
162 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
163 | * passes through a NULL-check conditional. For the branch wherein the state is | |
164 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
165 | * |
166 | * For each helper function that allocates a reference, such as | |
167 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
168 | * bpf_sk_release(). When a reference type passes into the release function, | |
169 | * the verifier also releases the reference. If any unchecked or unreleased | |
170 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
171 | */ |
172 | ||
17a52670 | 173 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 174 | struct bpf_verifier_stack_elem { |
17a52670 AS |
175 | /* verifer state is 'st' |
176 | * before processing instruction 'insn_idx' | |
177 | * and after processing instruction 'prev_insn_idx' | |
178 | */ | |
58e2af8b | 179 | struct bpf_verifier_state st; |
17a52670 AS |
180 | int insn_idx; |
181 | int prev_insn_idx; | |
58e2af8b | 182 | struct bpf_verifier_stack_elem *next; |
6f8a57cc AN |
183 | /* length of verifier log at the time this state was pushed on stack */ |
184 | u32 log_pos; | |
cbd35700 AS |
185 | }; |
186 | ||
b285fcb7 | 187 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 188 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 189 | |
d2e4c1e6 DB |
190 | #define BPF_MAP_KEY_POISON (1ULL << 63) |
191 | #define BPF_MAP_KEY_SEEN (1ULL << 62) | |
192 | ||
c93552c4 DB |
193 | #define BPF_MAP_PTR_UNPRIV 1UL |
194 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
195 | POISON_POINTER_DELTA)) | |
196 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
197 | ||
bc34dee6 JK |
198 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx); |
199 | static int release_reference(struct bpf_verifier_env *env, int ref_obj_id); | |
6a3cd331 | 200 | static void invalidate_non_owning_refs(struct bpf_verifier_env *env); |
5d92ddc3 | 201 | static bool in_rbtree_lock_required_cb(struct bpf_verifier_env *env); |
6a3cd331 DM |
202 | static int ref_set_non_owning(struct bpf_verifier_env *env, |
203 | struct bpf_reg_state *reg); | |
1cf3bfc6 IL |
204 | static void specialize_kfunc(struct bpf_verifier_env *env, |
205 | u32 func_id, u16 offset, unsigned long *addr); | |
51302c95 | 206 | static bool is_trusted_reg(const struct bpf_reg_state *reg); |
bc34dee6 | 207 | |
c93552c4 DB |
208 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) |
209 | { | |
d2e4c1e6 | 210 | return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON; |
c93552c4 DB |
211 | } |
212 | ||
213 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
214 | { | |
d2e4c1e6 | 215 | return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV; |
c93552c4 DB |
216 | } |
217 | ||
218 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
219 | const struct bpf_map *map, bool unpriv) | |
220 | { | |
221 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
222 | unpriv |= bpf_map_ptr_unpriv(aux); | |
d2e4c1e6 DB |
223 | aux->map_ptr_state = (unsigned long)map | |
224 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
225 | } | |
226 | ||
227 | static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux) | |
228 | { | |
229 | return aux->map_key_state & BPF_MAP_KEY_POISON; | |
230 | } | |
231 | ||
232 | static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux) | |
233 | { | |
234 | return !(aux->map_key_state & BPF_MAP_KEY_SEEN); | |
235 | } | |
236 | ||
237 | static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux) | |
238 | { | |
239 | return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON); | |
240 | } | |
241 | ||
242 | static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state) | |
243 | { | |
244 | bool poisoned = bpf_map_key_poisoned(aux); | |
245 | ||
246 | aux->map_key_state = state | BPF_MAP_KEY_SEEN | | |
247 | (poisoned ? BPF_MAP_KEY_POISON : 0ULL); | |
c93552c4 | 248 | } |
fad73a1a | 249 | |
fde2a388 AN |
250 | static bool bpf_helper_call(const struct bpf_insn *insn) |
251 | { | |
252 | return insn->code == (BPF_JMP | BPF_CALL) && | |
253 | insn->src_reg == 0; | |
254 | } | |
255 | ||
23a2d70c YS |
256 | static bool bpf_pseudo_call(const struct bpf_insn *insn) |
257 | { | |
258 | return insn->code == (BPF_JMP | BPF_CALL) && | |
259 | insn->src_reg == BPF_PSEUDO_CALL; | |
260 | } | |
261 | ||
e6ac2450 MKL |
262 | static bool bpf_pseudo_kfunc_call(const struct bpf_insn *insn) |
263 | { | |
264 | return insn->code == (BPF_JMP | BPF_CALL) && | |
265 | insn->src_reg == BPF_PSEUDO_KFUNC_CALL; | |
266 | } | |
267 | ||
33ff9823 DB |
268 | struct bpf_call_arg_meta { |
269 | struct bpf_map *map_ptr; | |
435faee1 | 270 | bool raw_mode; |
36bbef52 | 271 | bool pkt_access; |
8f14852e | 272 | u8 release_regno; |
435faee1 DB |
273 | int regno; |
274 | int access_size; | |
457f4436 | 275 | int mem_size; |
10060503 | 276 | u64 msize_max_value; |
1b986589 | 277 | int ref_obj_id; |
f8064ab9 | 278 | int dynptr_id; |
3e8ce298 | 279 | int map_uid; |
d83525ca | 280 | int func_id; |
22dc4a0f | 281 | struct btf *btf; |
eaa6bcb7 | 282 | u32 btf_id; |
22dc4a0f | 283 | struct btf *ret_btf; |
eaa6bcb7 | 284 | u32 ret_btf_id; |
69c087ba | 285 | u32 subprogno; |
aa3496ac | 286 | struct btf_field *kptr_field; |
33ff9823 DB |
287 | }; |
288 | ||
d0e1ac22 AN |
289 | struct bpf_kfunc_call_arg_meta { |
290 | /* In parameters */ | |
291 | struct btf *btf; | |
292 | u32 func_id; | |
293 | u32 kfunc_flags; | |
294 | const struct btf_type *func_proto; | |
295 | const char *func_name; | |
296 | /* Out parameters */ | |
297 | u32 ref_obj_id; | |
298 | u8 release_regno; | |
299 | bool r0_rdonly; | |
300 | u32 ret_btf_id; | |
301 | u64 r0_size; | |
302 | u32 subprogno; | |
303 | struct { | |
304 | u64 value; | |
305 | bool found; | |
306 | } arg_constant; | |
4d585f48 | 307 | |
7793fc3b | 308 | /* arg_{btf,btf_id,owning_ref} are used by kfunc-specific handling, |
4d585f48 DM |
309 | * generally to pass info about user-defined local kptr types to later |
310 | * verification logic | |
36d8bdf7 | 311 | * bpf_obj_drop/bpf_percpu_obj_drop |
4d585f48 DM |
312 | * Record the local kptr type to be drop'd |
313 | * bpf_refcount_acquire (via KF_ARG_PTR_TO_REFCOUNTED_KPTR arg type) | |
7793fc3b DM |
314 | * Record the local kptr type to be refcount_incr'd and use |
315 | * arg_owning_ref to determine whether refcount_acquire should be | |
316 | * fallible | |
4d585f48 DM |
317 | */ |
318 | struct btf *arg_btf; | |
319 | u32 arg_btf_id; | |
7793fc3b | 320 | bool arg_owning_ref; |
4d585f48 | 321 | |
d0e1ac22 AN |
322 | struct { |
323 | struct btf_field *field; | |
324 | } arg_list_head; | |
325 | struct { | |
326 | struct btf_field *field; | |
327 | } arg_rbtree_root; | |
328 | struct { | |
329 | enum bpf_dynptr_type type; | |
330 | u32 id; | |
361f129f | 331 | u32 ref_obj_id; |
d0e1ac22 | 332 | } initialized_dynptr; |
06accc87 AN |
333 | struct { |
334 | u8 spi; | |
335 | u8 frameno; | |
336 | } iter; | |
d0e1ac22 AN |
337 | u64 mem_size; |
338 | }; | |
339 | ||
8580ac94 AS |
340 | struct btf *btf_vmlinux; |
341 | ||
cbd35700 | 342 | static DEFINE_MUTEX(bpf_verifier_lock); |
1fda5bb6 | 343 | static DEFINE_MUTEX(bpf_percpu_ma_lock); |
cbd35700 | 344 | |
d9762e84 MKL |
345 | static const struct bpf_line_info * |
346 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
347 | { | |
348 | const struct bpf_line_info *linfo; | |
349 | const struct bpf_prog *prog; | |
350 | u32 i, nr_linfo; | |
351 | ||
352 | prog = env->prog; | |
353 | nr_linfo = prog->aux->nr_linfo; | |
354 | ||
355 | if (!nr_linfo || insn_off >= prog->len) | |
356 | return NULL; | |
357 | ||
358 | linfo = prog->aux->linfo; | |
359 | for (i = 1; i < nr_linfo; i++) | |
360 | if (insn_off < linfo[i].insn_off) | |
361 | break; | |
362 | ||
363 | return &linfo[i - 1]; | |
364 | } | |
365 | ||
abe08840 JO |
366 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) |
367 | { | |
77d2e05a | 368 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
369 | va_list args; |
370 | ||
77d2e05a MKL |
371 | if (!bpf_verifier_log_needed(&env->log)) |
372 | return; | |
373 | ||
abe08840 | 374 | va_start(args, fmt); |
77d2e05a | 375 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
376 | va_end(args); |
377 | } | |
cbd35700 | 378 | |
d9762e84 MKL |
379 | static const char *ltrim(const char *s) |
380 | { | |
381 | while (isspace(*s)) | |
382 | s++; | |
383 | ||
384 | return s; | |
385 | } | |
386 | ||
387 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
388 | u32 insn_off, | |
389 | const char *prefix_fmt, ...) | |
390 | { | |
391 | const struct bpf_line_info *linfo; | |
392 | ||
393 | if (!bpf_verifier_log_needed(&env->log)) | |
394 | return; | |
395 | ||
396 | linfo = find_linfo(env, insn_off); | |
397 | if (!linfo || linfo == env->prev_linfo) | |
398 | return; | |
399 | ||
400 | if (prefix_fmt) { | |
401 | va_list args; | |
402 | ||
403 | va_start(args, prefix_fmt); | |
404 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
405 | va_end(args); | |
406 | } | |
407 | ||
408 | verbose(env, "%s\n", | |
409 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
410 | linfo->line_off))); | |
411 | ||
412 | env->prev_linfo = linfo; | |
413 | } | |
414 | ||
bc2591d6 YS |
415 | static void verbose_invalid_scalar(struct bpf_verifier_env *env, |
416 | struct bpf_reg_state *reg, | |
417 | struct tnum *range, const char *ctx, | |
418 | const char *reg_name) | |
419 | { | |
420 | char tn_buf[48]; | |
421 | ||
422 | verbose(env, "At %s the register %s ", ctx, reg_name); | |
423 | if (!tnum_is_unknown(reg->var_off)) { | |
424 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
425 | verbose(env, "has value %s", tn_buf); | |
426 | } else { | |
427 | verbose(env, "has unknown scalar value"); | |
428 | } | |
429 | tnum_strn(tn_buf, sizeof(tn_buf), *range); | |
430 | verbose(env, " should have been in %s\n", tn_buf); | |
431 | } | |
432 | ||
de8f3a83 DB |
433 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
434 | { | |
0c9a7a7e | 435 | type = base_type(type); |
de8f3a83 DB |
436 | return type == PTR_TO_PACKET || |
437 | type == PTR_TO_PACKET_META; | |
438 | } | |
439 | ||
46f8bc92 MKL |
440 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
441 | { | |
442 | return type == PTR_TO_SOCKET || | |
655a51e5 | 443 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
444 | type == PTR_TO_TCP_SOCK || |
445 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
446 | } |
447 | ||
1057d299 AS |
448 | static bool type_may_be_null(u32 type) |
449 | { | |
450 | return type & PTR_MAYBE_NULL; | |
451 | } | |
452 | ||
51302c95 | 453 | static bool reg_not_null(const struct bpf_reg_state *reg) |
cac616db | 454 | { |
51302c95 DV |
455 | enum bpf_reg_type type; |
456 | ||
457 | type = reg->type; | |
1057d299 AS |
458 | if (type_may_be_null(type)) |
459 | return false; | |
460 | ||
461 | type = base_type(type); | |
cac616db JF |
462 | return type == PTR_TO_SOCKET || |
463 | type == PTR_TO_TCP_SOCK || | |
464 | type == PTR_TO_MAP_VALUE || | |
69c087ba | 465 | type == PTR_TO_MAP_KEY || |
d5271c5b | 466 | type == PTR_TO_SOCK_COMMON || |
51302c95 | 467 | (type == PTR_TO_BTF_ID && is_trusted_reg(reg)) || |
d5271c5b | 468 | type == PTR_TO_MEM; |
cac616db JF |
469 | } |
470 | ||
d8939cb0 DM |
471 | static bool type_is_ptr_alloc_obj(u32 type) |
472 | { | |
473 | return base_type(type) == PTR_TO_BTF_ID && type_flag(type) & MEM_ALLOC; | |
474 | } | |
475 | ||
6a3cd331 DM |
476 | static bool type_is_non_owning_ref(u32 type) |
477 | { | |
478 | return type_is_ptr_alloc_obj(type) && type_flag(type) & NON_OWN_REF; | |
479 | } | |
480 | ||
4e814da0 KKD |
481 | static struct btf_record *reg_btf_record(const struct bpf_reg_state *reg) |
482 | { | |
483 | struct btf_record *rec = NULL; | |
484 | struct btf_struct_meta *meta; | |
485 | ||
486 | if (reg->type == PTR_TO_MAP_VALUE) { | |
487 | rec = reg->map_ptr->record; | |
d8939cb0 | 488 | } else if (type_is_ptr_alloc_obj(reg->type)) { |
4e814da0 KKD |
489 | meta = btf_find_struct_meta(reg->btf, reg->btf_id); |
490 | if (meta) | |
491 | rec = meta->record; | |
492 | } | |
493 | return rec; | |
494 | } | |
495 | ||
fde2a388 AN |
496 | static bool subprog_is_global(const struct bpf_verifier_env *env, int subprog) |
497 | { | |
498 | struct bpf_func_info_aux *aux = env->prog->aux->func_info_aux; | |
499 | ||
500 | return aux && aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
501 | } | |
502 | ||
d83525ca AS |
503 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
504 | { | |
4e814da0 | 505 | return btf_record_has_field(reg_btf_record(reg), BPF_SPIN_LOCK); |
cba368c1 MKL |
506 | } |
507 | ||
20b2aff4 HL |
508 | static bool type_is_rdonly_mem(u32 type) |
509 | { | |
510 | return type & MEM_RDONLY; | |
cba368c1 MKL |
511 | } |
512 | ||
64d85290 JS |
513 | static bool is_acquire_function(enum bpf_func_id func_id, |
514 | const struct bpf_map *map) | |
515 | { | |
516 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
517 | ||
518 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
519 | func_id == BPF_FUNC_sk_lookup_udp || | |
457f4436 | 520 | func_id == BPF_FUNC_skc_lookup_tcp || |
c0a5a21c KKD |
521 | func_id == BPF_FUNC_ringbuf_reserve || |
522 | func_id == BPF_FUNC_kptr_xchg) | |
64d85290 JS |
523 | return true; |
524 | ||
525 | if (func_id == BPF_FUNC_map_lookup_elem && | |
526 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
527 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
528 | return true; | |
529 | ||
530 | return false; | |
46f8bc92 MKL |
531 | } |
532 | ||
1b986589 MKL |
533 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
534 | { | |
535 | return func_id == BPF_FUNC_tcp_sock || | |
1df8f55a MKL |
536 | func_id == BPF_FUNC_sk_fullsock || |
537 | func_id == BPF_FUNC_skc_to_tcp_sock || | |
538 | func_id == BPF_FUNC_skc_to_tcp6_sock || | |
539 | func_id == BPF_FUNC_skc_to_udp6_sock || | |
3bc253c2 | 540 | func_id == BPF_FUNC_skc_to_mptcp_sock || |
1df8f55a MKL |
541 | func_id == BPF_FUNC_skc_to_tcp_timewait_sock || |
542 | func_id == BPF_FUNC_skc_to_tcp_request_sock; | |
1b986589 MKL |
543 | } |
544 | ||
88374342 | 545 | static bool is_dynptr_ref_function(enum bpf_func_id func_id) |
b2d8ef19 DM |
546 | { |
547 | return func_id == BPF_FUNC_dynptr_data; | |
548 | } | |
549 | ||
ab5cfac1 | 550 | static bool is_sync_callback_calling_kfunc(u32 btf_id); |
f18b03fa | 551 | static bool is_bpf_throw_kfunc(struct bpf_insn *insn); |
fde2a388 | 552 | |
ab5cfac1 | 553 | static bool is_sync_callback_calling_function(enum bpf_func_id func_id) |
be2ef816 AN |
554 | { |
555 | return func_id == BPF_FUNC_for_each_map_elem || | |
be2ef816 AN |
556 | func_id == BPF_FUNC_find_vma || |
557 | func_id == BPF_FUNC_loop || | |
558 | func_id == BPF_FUNC_user_ringbuf_drain; | |
559 | } | |
560 | ||
fde2a388 AN |
561 | static bool is_async_callback_calling_function(enum bpf_func_id func_id) |
562 | { | |
563 | return func_id == BPF_FUNC_timer_set_callback; | |
564 | } | |
565 | ||
ab5cfac1 EZ |
566 | static bool is_callback_calling_function(enum bpf_func_id func_id) |
567 | { | |
568 | return is_sync_callback_calling_function(func_id) || | |
569 | is_async_callback_calling_function(func_id); | |
570 | } | |
571 | ||
572 | static bool is_sync_callback_calling_insn(struct bpf_insn *insn) | |
573 | { | |
574 | return (bpf_helper_call(insn) && is_sync_callback_calling_function(insn->imm)) || | |
575 | (bpf_pseudo_kfunc_call(insn) && is_sync_callback_calling_kfunc(insn->imm)); | |
576 | } | |
577 | ||
9bb00b28 YS |
578 | static bool is_storage_get_function(enum bpf_func_id func_id) |
579 | { | |
580 | return func_id == BPF_FUNC_sk_storage_get || | |
581 | func_id == BPF_FUNC_inode_storage_get || | |
582 | func_id == BPF_FUNC_task_storage_get || | |
583 | func_id == BPF_FUNC_cgrp_storage_get; | |
584 | } | |
585 | ||
b2d8ef19 DM |
586 | static bool helper_multiple_ref_obj_use(enum bpf_func_id func_id, |
587 | const struct bpf_map *map) | |
588 | { | |
589 | int ref_obj_uses = 0; | |
590 | ||
591 | if (is_ptr_cast_function(func_id)) | |
592 | ref_obj_uses++; | |
593 | if (is_acquire_function(func_id, map)) | |
594 | ref_obj_uses++; | |
88374342 | 595 | if (is_dynptr_ref_function(func_id)) |
b2d8ef19 DM |
596 | ref_obj_uses++; |
597 | ||
598 | return ref_obj_uses > 1; | |
599 | } | |
600 | ||
39491867 BJ |
601 | static bool is_cmpxchg_insn(const struct bpf_insn *insn) |
602 | { | |
603 | return BPF_CLASS(insn->code) == BPF_STX && | |
604 | BPF_MODE(insn->code) == BPF_ATOMIC && | |
605 | insn->imm == BPF_CMPXCHG; | |
606 | } | |
607 | ||
c25b2ae1 HL |
608 | /* string representation of 'enum bpf_reg_type' |
609 | * | |
610 | * Note that reg_type_str() can not appear more than once in a single verbose() | |
611 | * statement. | |
612 | */ | |
613 | static const char *reg_type_str(struct bpf_verifier_env *env, | |
614 | enum bpf_reg_type type) | |
615 | { | |
ef66c547 | 616 | char postfix[16] = {0}, prefix[64] = {0}; |
c25b2ae1 HL |
617 | static const char * const str[] = { |
618 | [NOT_INIT] = "?", | |
7df5072c | 619 | [SCALAR_VALUE] = "scalar", |
c25b2ae1 HL |
620 | [PTR_TO_CTX] = "ctx", |
621 | [CONST_PTR_TO_MAP] = "map_ptr", | |
622 | [PTR_TO_MAP_VALUE] = "map_value", | |
623 | [PTR_TO_STACK] = "fp", | |
624 | [PTR_TO_PACKET] = "pkt", | |
625 | [PTR_TO_PACKET_META] = "pkt_meta", | |
626 | [PTR_TO_PACKET_END] = "pkt_end", | |
627 | [PTR_TO_FLOW_KEYS] = "flow_keys", | |
628 | [PTR_TO_SOCKET] = "sock", | |
629 | [PTR_TO_SOCK_COMMON] = "sock_common", | |
630 | [PTR_TO_TCP_SOCK] = "tcp_sock", | |
631 | [PTR_TO_TP_BUFFER] = "tp_buffer", | |
632 | [PTR_TO_XDP_SOCK] = "xdp_sock", | |
633 | [PTR_TO_BTF_ID] = "ptr_", | |
c25b2ae1 | 634 | [PTR_TO_MEM] = "mem", |
20b2aff4 | 635 | [PTR_TO_BUF] = "buf", |
c25b2ae1 HL |
636 | [PTR_TO_FUNC] = "func", |
637 | [PTR_TO_MAP_KEY] = "map_key", | |
27060531 | 638 | [CONST_PTR_TO_DYNPTR] = "dynptr_ptr", |
c25b2ae1 HL |
639 | }; |
640 | ||
641 | if (type & PTR_MAYBE_NULL) { | |
5844101a | 642 | if (base_type(type) == PTR_TO_BTF_ID) |
c25b2ae1 HL |
643 | strncpy(postfix, "or_null_", 16); |
644 | else | |
645 | strncpy(postfix, "_or_null", 16); | |
646 | } | |
647 | ||
9bb00b28 | 648 | snprintf(prefix, sizeof(prefix), "%s%s%s%s%s%s%s", |
ef66c547 DV |
649 | type & MEM_RDONLY ? "rdonly_" : "", |
650 | type & MEM_RINGBUF ? "ringbuf_" : "", | |
651 | type & MEM_USER ? "user_" : "", | |
652 | type & MEM_PERCPU ? "percpu_" : "", | |
9bb00b28 | 653 | type & MEM_RCU ? "rcu_" : "", |
3f00c523 DV |
654 | type & PTR_UNTRUSTED ? "untrusted_" : "", |
655 | type & PTR_TRUSTED ? "trusted_" : "" | |
ef66c547 | 656 | ); |
20b2aff4 | 657 | |
d9439c21 | 658 | snprintf(env->tmp_str_buf, TMP_STR_BUF_LEN, "%s%s%s", |
20b2aff4 | 659 | prefix, str[base_type(type)], postfix); |
d9439c21 | 660 | return env->tmp_str_buf; |
c25b2ae1 | 661 | } |
17a52670 | 662 | |
8efea21d EC |
663 | static char slot_type_char[] = { |
664 | [STACK_INVALID] = '?', | |
665 | [STACK_SPILL] = 'r', | |
666 | [STACK_MISC] = 'm', | |
667 | [STACK_ZERO] = '0', | |
97e03f52 | 668 | [STACK_DYNPTR] = 'd', |
06accc87 | 669 | [STACK_ITER] = 'i', |
8efea21d EC |
670 | }; |
671 | ||
4e92024a AS |
672 | static void print_liveness(struct bpf_verifier_env *env, |
673 | enum bpf_reg_liveness live) | |
674 | { | |
9242b5f5 | 675 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
676 | verbose(env, "_"); |
677 | if (live & REG_LIVE_READ) | |
678 | verbose(env, "r"); | |
679 | if (live & REG_LIVE_WRITTEN) | |
680 | verbose(env, "w"); | |
9242b5f5 AS |
681 | if (live & REG_LIVE_DONE) |
682 | verbose(env, "D"); | |
4e92024a AS |
683 | } |
684 | ||
79168a66 | 685 | static int __get_spi(s32 off) |
97e03f52 JK |
686 | { |
687 | return (-off - 1) / BPF_REG_SIZE; | |
688 | } | |
689 | ||
f5b625e5 KKD |
690 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
691 | const struct bpf_reg_state *reg) | |
692 | { | |
693 | struct bpf_verifier_state *cur = env->cur_state; | |
694 | ||
695 | return cur->frame[reg->frameno]; | |
696 | } | |
697 | ||
97e03f52 JK |
698 | static bool is_spi_bounds_valid(struct bpf_func_state *state, int spi, int nr_slots) |
699 | { | |
f5b625e5 | 700 | int allocated_slots = state->allocated_stack / BPF_REG_SIZE; |
97e03f52 | 701 | |
f5b625e5 KKD |
702 | /* We need to check that slots between [spi - nr_slots + 1, spi] are |
703 | * within [0, allocated_stack). | |
704 | * | |
705 | * Please note that the spi grows downwards. For example, a dynptr | |
706 | * takes the size of two stack slots; the first slot will be at | |
707 | * spi and the second slot will be at spi - 1. | |
708 | */ | |
709 | return spi - nr_slots + 1 >= 0 && spi < allocated_slots; | |
97e03f52 JK |
710 | } |
711 | ||
a461f5ad AN |
712 | static int stack_slot_obj_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
713 | const char *obj_kind, int nr_slots) | |
f4d7e40a | 714 | { |
79168a66 | 715 | int off, spi; |
f4d7e40a | 716 | |
79168a66 | 717 | if (!tnum_is_const(reg->var_off)) { |
a461f5ad | 718 | verbose(env, "%s has to be at a constant offset\n", obj_kind); |
79168a66 KKD |
719 | return -EINVAL; |
720 | } | |
721 | ||
722 | off = reg->off + reg->var_off.value; | |
723 | if (off % BPF_REG_SIZE) { | |
a461f5ad | 724 | verbose(env, "cannot pass in %s at an offset=%d\n", obj_kind, off); |
79168a66 KKD |
725 | return -EINVAL; |
726 | } | |
727 | ||
728 | spi = __get_spi(off); | |
a461f5ad AN |
729 | if (spi + 1 < nr_slots) { |
730 | verbose(env, "cannot pass in %s at an offset=%d\n", obj_kind, off); | |
79168a66 KKD |
731 | return -EINVAL; |
732 | } | |
97e03f52 | 733 | |
a461f5ad | 734 | if (!is_spi_bounds_valid(func(env, reg), spi, nr_slots)) |
f5b625e5 KKD |
735 | return -ERANGE; |
736 | return spi; | |
f4d7e40a AS |
737 | } |
738 | ||
a461f5ad AN |
739 | static int dynptr_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
740 | { | |
741 | return stack_slot_obj_get_spi(env, reg, "dynptr", BPF_DYNPTR_NR_SLOTS); | |
742 | } | |
743 | ||
06accc87 AN |
744 | static int iter_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg, int nr_slots) |
745 | { | |
746 | return stack_slot_obj_get_spi(env, reg, "iter", nr_slots); | |
747 | } | |
748 | ||
b32a5dae | 749 | static const char *btf_type_name(const struct btf *btf, u32 id) |
9e15db66 | 750 | { |
22dc4a0f | 751 | return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off); |
9e15db66 AS |
752 | } |
753 | ||
d54e0f6c AN |
754 | static const char *dynptr_type_str(enum bpf_dynptr_type type) |
755 | { | |
756 | switch (type) { | |
757 | case BPF_DYNPTR_TYPE_LOCAL: | |
758 | return "local"; | |
759 | case BPF_DYNPTR_TYPE_RINGBUF: | |
760 | return "ringbuf"; | |
761 | case BPF_DYNPTR_TYPE_SKB: | |
762 | return "skb"; | |
763 | case BPF_DYNPTR_TYPE_XDP: | |
764 | return "xdp"; | |
765 | case BPF_DYNPTR_TYPE_INVALID: | |
766 | return "<invalid>"; | |
767 | default: | |
768 | WARN_ONCE(1, "unknown dynptr type %d\n", type); | |
769 | return "<unknown>"; | |
770 | } | |
771 | } | |
772 | ||
06accc87 AN |
773 | static const char *iter_type_str(const struct btf *btf, u32 btf_id) |
774 | { | |
775 | if (!btf || btf_id == 0) | |
776 | return "<invalid>"; | |
777 | ||
778 | /* we already validated that type is valid and has conforming name */ | |
b32a5dae | 779 | return btf_type_name(btf, btf_id) + sizeof(ITER_PREFIX) - 1; |
06accc87 AN |
780 | } |
781 | ||
782 | static const char *iter_state_str(enum bpf_iter_state state) | |
783 | { | |
784 | switch (state) { | |
785 | case BPF_ITER_STATE_ACTIVE: | |
786 | return "active"; | |
787 | case BPF_ITER_STATE_DRAINED: | |
788 | return "drained"; | |
789 | case BPF_ITER_STATE_INVALID: | |
790 | return "<invalid>"; | |
791 | default: | |
792 | WARN_ONCE(1, "unknown iter state %d\n", state); | |
793 | return "<unknown>"; | |
794 | } | |
795 | } | |
796 | ||
0f55f9ed CL |
797 | static void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno) |
798 | { | |
799 | env->scratched_regs |= 1U << regno; | |
800 | } | |
801 | ||
802 | static void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi) | |
803 | { | |
343e5375 | 804 | env->scratched_stack_slots |= 1ULL << spi; |
0f55f9ed CL |
805 | } |
806 | ||
807 | static bool reg_scratched(const struct bpf_verifier_env *env, u32 regno) | |
808 | { | |
809 | return (env->scratched_regs >> regno) & 1; | |
810 | } | |
811 | ||
812 | static bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno) | |
813 | { | |
814 | return (env->scratched_stack_slots >> regno) & 1; | |
815 | } | |
816 | ||
817 | static bool verifier_state_scratched(const struct bpf_verifier_env *env) | |
818 | { | |
819 | return env->scratched_regs || env->scratched_stack_slots; | |
820 | } | |
821 | ||
822 | static void mark_verifier_state_clean(struct bpf_verifier_env *env) | |
823 | { | |
824 | env->scratched_regs = 0U; | |
343e5375 | 825 | env->scratched_stack_slots = 0ULL; |
0f55f9ed CL |
826 | } |
827 | ||
828 | /* Used for printing the entire verifier state. */ | |
829 | static void mark_verifier_state_scratched(struct bpf_verifier_env *env) | |
830 | { | |
831 | env->scratched_regs = ~0U; | |
343e5375 | 832 | env->scratched_stack_slots = ~0ULL; |
0f55f9ed CL |
833 | } |
834 | ||
97e03f52 JK |
835 | static enum bpf_dynptr_type arg_to_dynptr_type(enum bpf_arg_type arg_type) |
836 | { | |
837 | switch (arg_type & DYNPTR_TYPE_FLAG_MASK) { | |
838 | case DYNPTR_TYPE_LOCAL: | |
839 | return BPF_DYNPTR_TYPE_LOCAL; | |
bc34dee6 JK |
840 | case DYNPTR_TYPE_RINGBUF: |
841 | return BPF_DYNPTR_TYPE_RINGBUF; | |
b5964b96 JK |
842 | case DYNPTR_TYPE_SKB: |
843 | return BPF_DYNPTR_TYPE_SKB; | |
05421aec JK |
844 | case DYNPTR_TYPE_XDP: |
845 | return BPF_DYNPTR_TYPE_XDP; | |
97e03f52 JK |
846 | default: |
847 | return BPF_DYNPTR_TYPE_INVALID; | |
848 | } | |
849 | } | |
850 | ||
66e3a13e JK |
851 | static enum bpf_type_flag get_dynptr_type_flag(enum bpf_dynptr_type type) |
852 | { | |
853 | switch (type) { | |
854 | case BPF_DYNPTR_TYPE_LOCAL: | |
855 | return DYNPTR_TYPE_LOCAL; | |
856 | case BPF_DYNPTR_TYPE_RINGBUF: | |
857 | return DYNPTR_TYPE_RINGBUF; | |
858 | case BPF_DYNPTR_TYPE_SKB: | |
859 | return DYNPTR_TYPE_SKB; | |
860 | case BPF_DYNPTR_TYPE_XDP: | |
861 | return DYNPTR_TYPE_XDP; | |
862 | default: | |
863 | return 0; | |
864 | } | |
865 | } | |
866 | ||
bc34dee6 JK |
867 | static bool dynptr_type_refcounted(enum bpf_dynptr_type type) |
868 | { | |
869 | return type == BPF_DYNPTR_TYPE_RINGBUF; | |
870 | } | |
871 | ||
27060531 KKD |
872 | static void __mark_dynptr_reg(struct bpf_reg_state *reg, |
873 | enum bpf_dynptr_type type, | |
f8064ab9 | 874 | bool first_slot, int dynptr_id); |
27060531 KKD |
875 | |
876 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, | |
877 | struct bpf_reg_state *reg); | |
878 | ||
f8064ab9 KKD |
879 | static void mark_dynptr_stack_regs(struct bpf_verifier_env *env, |
880 | struct bpf_reg_state *sreg1, | |
27060531 KKD |
881 | struct bpf_reg_state *sreg2, |
882 | enum bpf_dynptr_type type) | |
883 | { | |
f8064ab9 KKD |
884 | int id = ++env->id_gen; |
885 | ||
886 | __mark_dynptr_reg(sreg1, type, true, id); | |
887 | __mark_dynptr_reg(sreg2, type, false, id); | |
27060531 KKD |
888 | } |
889 | ||
f8064ab9 KKD |
890 | static void mark_dynptr_cb_reg(struct bpf_verifier_env *env, |
891 | struct bpf_reg_state *reg, | |
27060531 KKD |
892 | enum bpf_dynptr_type type) |
893 | { | |
f8064ab9 | 894 | __mark_dynptr_reg(reg, type, true, ++env->id_gen); |
27060531 KKD |
895 | } |
896 | ||
ef8fc7a0 KKD |
897 | static int destroy_if_dynptr_stack_slot(struct bpf_verifier_env *env, |
898 | struct bpf_func_state *state, int spi); | |
27060531 | 899 | |
97e03f52 | 900 | static int mark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
361f129f | 901 | enum bpf_arg_type arg_type, int insn_idx, int clone_ref_obj_id) |
97e03f52 JK |
902 | { |
903 | struct bpf_func_state *state = func(env, reg); | |
904 | enum bpf_dynptr_type type; | |
361f129f | 905 | int spi, i, err; |
97e03f52 | 906 | |
79168a66 KKD |
907 | spi = dynptr_get_spi(env, reg); |
908 | if (spi < 0) | |
909 | return spi; | |
97e03f52 | 910 | |
379d4ba8 KKD |
911 | /* We cannot assume both spi and spi - 1 belong to the same dynptr, |
912 | * hence we need to call destroy_if_dynptr_stack_slot twice for both, | |
913 | * to ensure that for the following example: | |
914 | * [d1][d1][d2][d2] | |
915 | * spi 3 2 1 0 | |
916 | * So marking spi = 2 should lead to destruction of both d1 and d2. In | |
917 | * case they do belong to same dynptr, second call won't see slot_type | |
918 | * as STACK_DYNPTR and will simply skip destruction. | |
919 | */ | |
920 | err = destroy_if_dynptr_stack_slot(env, state, spi); | |
921 | if (err) | |
922 | return err; | |
923 | err = destroy_if_dynptr_stack_slot(env, state, spi - 1); | |
924 | if (err) | |
925 | return err; | |
97e03f52 JK |
926 | |
927 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
928 | state->stack[spi].slot_type[i] = STACK_DYNPTR; | |
929 | state->stack[spi - 1].slot_type[i] = STACK_DYNPTR; | |
930 | } | |
931 | ||
932 | type = arg_to_dynptr_type(arg_type); | |
933 | if (type == BPF_DYNPTR_TYPE_INVALID) | |
934 | return -EINVAL; | |
935 | ||
f8064ab9 | 936 | mark_dynptr_stack_regs(env, &state->stack[spi].spilled_ptr, |
27060531 | 937 | &state->stack[spi - 1].spilled_ptr, type); |
97e03f52 | 938 | |
bc34dee6 JK |
939 | if (dynptr_type_refcounted(type)) { |
940 | /* The id is used to track proper releasing */ | |
361f129f JK |
941 | int id; |
942 | ||
943 | if (clone_ref_obj_id) | |
944 | id = clone_ref_obj_id; | |
945 | else | |
946 | id = acquire_reference_state(env, insn_idx); | |
947 | ||
bc34dee6 JK |
948 | if (id < 0) |
949 | return id; | |
950 | ||
27060531 KKD |
951 | state->stack[spi].spilled_ptr.ref_obj_id = id; |
952 | state->stack[spi - 1].spilled_ptr.ref_obj_id = id; | |
bc34dee6 JK |
953 | } |
954 | ||
d6fefa11 KKD |
955 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; |
956 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
957 | ||
97e03f52 JK |
958 | return 0; |
959 | } | |
960 | ||
361f129f | 961 | static void invalidate_dynptr(struct bpf_verifier_env *env, struct bpf_func_state *state, int spi) |
97e03f52 | 962 | { |
361f129f | 963 | int i; |
97e03f52 JK |
964 | |
965 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
966 | state->stack[spi].slot_type[i] = STACK_INVALID; | |
967 | state->stack[spi - 1].slot_type[i] = STACK_INVALID; | |
968 | } | |
969 | ||
27060531 KKD |
970 | __mark_reg_not_init(env, &state->stack[spi].spilled_ptr); |
971 | __mark_reg_not_init(env, &state->stack[spi - 1].spilled_ptr); | |
d6fefa11 KKD |
972 | |
973 | /* Why do we need to set REG_LIVE_WRITTEN for STACK_INVALID slot? | |
974 | * | |
975 | * While we don't allow reading STACK_INVALID, it is still possible to | |
976 | * do <8 byte writes marking some but not all slots as STACK_MISC. Then, | |
977 | * helpers or insns can do partial read of that part without failing, | |
978 | * but check_stack_range_initialized, check_stack_read_var_off, and | |
979 | * check_stack_read_fixed_off will do mark_reg_read for all 8-bytes of | |
980 | * the slot conservatively. Hence we need to prevent those liveness | |
981 | * marking walks. | |
982 | * | |
983 | * This was not a problem before because STACK_INVALID is only set by | |
984 | * default (where the default reg state has its reg->parent as NULL), or | |
985 | * in clean_live_states after REG_LIVE_DONE (at which point | |
986 | * mark_reg_read won't walk reg->parent chain), but not randomly during | |
987 | * verifier state exploration (like we did above). Hence, for our case | |
988 | * parentage chain will still be live (i.e. reg->parent may be | |
989 | * non-NULL), while earlier reg->parent was NULL, so we need | |
990 | * REG_LIVE_WRITTEN to screen off read marker propagation when it is | |
991 | * done later on reads or by mark_dynptr_read as well to unnecessary | |
992 | * mark registers in verifier state. | |
993 | */ | |
994 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
995 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
361f129f JK |
996 | } |
997 | ||
998 | static int unmark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
999 | { | |
1000 | struct bpf_func_state *state = func(env, reg); | |
1001 | int spi, ref_obj_id, i; | |
1002 | ||
1003 | spi = dynptr_get_spi(env, reg); | |
1004 | if (spi < 0) | |
1005 | return spi; | |
1006 | ||
1007 | if (!dynptr_type_refcounted(state->stack[spi].spilled_ptr.dynptr.type)) { | |
1008 | invalidate_dynptr(env, state, spi); | |
1009 | return 0; | |
1010 | } | |
1011 | ||
1012 | ref_obj_id = state->stack[spi].spilled_ptr.ref_obj_id; | |
1013 | ||
1014 | /* If the dynptr has a ref_obj_id, then we need to invalidate | |
1015 | * two things: | |
1016 | * | |
1017 | * 1) Any dynptrs with a matching ref_obj_id (clones) | |
1018 | * 2) Any slices derived from this dynptr. | |
1019 | */ | |
1020 | ||
1021 | /* Invalidate any slices associated with this dynptr */ | |
1022 | WARN_ON_ONCE(release_reference(env, ref_obj_id)); | |
1023 | ||
1024 | /* Invalidate any dynptr clones */ | |
1025 | for (i = 1; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
1026 | if (state->stack[i].spilled_ptr.ref_obj_id != ref_obj_id) | |
1027 | continue; | |
1028 | ||
1029 | /* it should always be the case that if the ref obj id | |
1030 | * matches then the stack slot also belongs to a | |
1031 | * dynptr | |
1032 | */ | |
1033 | if (state->stack[i].slot_type[0] != STACK_DYNPTR) { | |
1034 | verbose(env, "verifier internal error: misconfigured ref_obj_id\n"); | |
1035 | return -EFAULT; | |
1036 | } | |
1037 | if (state->stack[i].spilled_ptr.dynptr.first_slot) | |
1038 | invalidate_dynptr(env, state, i); | |
1039 | } | |
d6fefa11 | 1040 | |
97e03f52 JK |
1041 | return 0; |
1042 | } | |
1043 | ||
ef8fc7a0 KKD |
1044 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1045 | struct bpf_reg_state *reg); | |
1046 | ||
dbd8d228 KKD |
1047 | static void mark_reg_invalid(const struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
1048 | { | |
1049 | if (!env->allow_ptr_leaks) | |
1050 | __mark_reg_not_init(env, reg); | |
1051 | else | |
1052 | __mark_reg_unknown(env, reg); | |
1053 | } | |
1054 | ||
ef8fc7a0 KKD |
1055 | static int destroy_if_dynptr_stack_slot(struct bpf_verifier_env *env, |
1056 | struct bpf_func_state *state, int spi) | |
97e03f52 | 1057 | { |
f8064ab9 KKD |
1058 | struct bpf_func_state *fstate; |
1059 | struct bpf_reg_state *dreg; | |
1060 | int i, dynptr_id; | |
27060531 | 1061 | |
ef8fc7a0 KKD |
1062 | /* We always ensure that STACK_DYNPTR is never set partially, |
1063 | * hence just checking for slot_type[0] is enough. This is | |
1064 | * different for STACK_SPILL, where it may be only set for | |
1065 | * 1 byte, so code has to use is_spilled_reg. | |
1066 | */ | |
1067 | if (state->stack[spi].slot_type[0] != STACK_DYNPTR) | |
1068 | return 0; | |
97e03f52 | 1069 | |
ef8fc7a0 KKD |
1070 | /* Reposition spi to first slot */ |
1071 | if (!state->stack[spi].spilled_ptr.dynptr.first_slot) | |
1072 | spi = spi + 1; | |
1073 | ||
1074 | if (dynptr_type_refcounted(state->stack[spi].spilled_ptr.dynptr.type)) { | |
1075 | verbose(env, "cannot overwrite referenced dynptr\n"); | |
1076 | return -EINVAL; | |
1077 | } | |
1078 | ||
1079 | mark_stack_slot_scratched(env, spi); | |
1080 | mark_stack_slot_scratched(env, spi - 1); | |
97e03f52 | 1081 | |
ef8fc7a0 | 1082 | /* Writing partially to one dynptr stack slot destroys both. */ |
97e03f52 | 1083 | for (i = 0; i < BPF_REG_SIZE; i++) { |
ef8fc7a0 KKD |
1084 | state->stack[spi].slot_type[i] = STACK_INVALID; |
1085 | state->stack[spi - 1].slot_type[i] = STACK_INVALID; | |
97e03f52 JK |
1086 | } |
1087 | ||
f8064ab9 KKD |
1088 | dynptr_id = state->stack[spi].spilled_ptr.id; |
1089 | /* Invalidate any slices associated with this dynptr */ | |
1090 | bpf_for_each_reg_in_vstate(env->cur_state, fstate, dreg, ({ | |
1091 | /* Dynptr slices are only PTR_TO_MEM_OR_NULL and PTR_TO_MEM */ | |
1092 | if (dreg->type != (PTR_TO_MEM | PTR_MAYBE_NULL) && dreg->type != PTR_TO_MEM) | |
1093 | continue; | |
dbd8d228 KKD |
1094 | if (dreg->dynptr_id == dynptr_id) |
1095 | mark_reg_invalid(env, dreg); | |
f8064ab9 | 1096 | })); |
ef8fc7a0 KKD |
1097 | |
1098 | /* Do not release reference state, we are destroying dynptr on stack, | |
1099 | * not using some helper to release it. Just reset register. | |
1100 | */ | |
1101 | __mark_reg_not_init(env, &state->stack[spi].spilled_ptr); | |
1102 | __mark_reg_not_init(env, &state->stack[spi - 1].spilled_ptr); | |
1103 | ||
1104 | /* Same reason as unmark_stack_slots_dynptr above */ | |
1105 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1106 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1107 | ||
1108 | return 0; | |
1109 | } | |
1110 | ||
7e0dac28 | 1111 | static bool is_dynptr_reg_valid_uninit(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
97e03f52 | 1112 | { |
7e0dac28 JK |
1113 | int spi; |
1114 | ||
27060531 KKD |
1115 | if (reg->type == CONST_PTR_TO_DYNPTR) |
1116 | return false; | |
97e03f52 | 1117 | |
7e0dac28 JK |
1118 | spi = dynptr_get_spi(env, reg); |
1119 | ||
1120 | /* -ERANGE (i.e. spi not falling into allocated stack slots) isn't an | |
1121 | * error because this just means the stack state hasn't been updated yet. | |
1122 | * We will do check_mem_access to check and update stack bounds later. | |
f5b625e5 | 1123 | */ |
7e0dac28 JK |
1124 | if (spi < 0 && spi != -ERANGE) |
1125 | return false; | |
1126 | ||
1127 | /* We don't need to check if the stack slots are marked by previous | |
1128 | * dynptr initializations because we allow overwriting existing unreferenced | |
1129 | * STACK_DYNPTR slots, see mark_stack_slots_dynptr which calls | |
1130 | * destroy_if_dynptr_stack_slot to ensure dynptr objects at the slots we are | |
1131 | * touching are completely destructed before we reinitialize them for a new | |
1132 | * one. For referenced ones, destroy_if_dynptr_stack_slot returns an error early | |
1133 | * instead of delaying it until the end where the user will get "Unreleased | |
379d4ba8 KKD |
1134 | * reference" error. |
1135 | */ | |
97e03f52 JK |
1136 | return true; |
1137 | } | |
1138 | ||
7e0dac28 | 1139 | static bool is_dynptr_reg_valid_init(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
97e03f52 JK |
1140 | { |
1141 | struct bpf_func_state *state = func(env, reg); | |
7e0dac28 | 1142 | int i, spi; |
97e03f52 | 1143 | |
7e0dac28 JK |
1144 | /* This already represents first slot of initialized bpf_dynptr. |
1145 | * | |
1146 | * CONST_PTR_TO_DYNPTR already has fixed and var_off as 0 due to | |
1147 | * check_func_arg_reg_off's logic, so we don't need to check its | |
1148 | * offset and alignment. | |
1149 | */ | |
27060531 KKD |
1150 | if (reg->type == CONST_PTR_TO_DYNPTR) |
1151 | return true; | |
1152 | ||
7e0dac28 | 1153 | spi = dynptr_get_spi(env, reg); |
79168a66 KKD |
1154 | if (spi < 0) |
1155 | return false; | |
f5b625e5 | 1156 | if (!state->stack[spi].spilled_ptr.dynptr.first_slot) |
97e03f52 JK |
1157 | return false; |
1158 | ||
1159 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
1160 | if (state->stack[spi].slot_type[i] != STACK_DYNPTR || | |
1161 | state->stack[spi - 1].slot_type[i] != STACK_DYNPTR) | |
1162 | return false; | |
1163 | } | |
1164 | ||
e9e315b4 RS |
1165 | return true; |
1166 | } | |
1167 | ||
6b75bd3d KKD |
1168 | static bool is_dynptr_type_expected(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
1169 | enum bpf_arg_type arg_type) | |
e9e315b4 RS |
1170 | { |
1171 | struct bpf_func_state *state = func(env, reg); | |
1172 | enum bpf_dynptr_type dynptr_type; | |
27060531 | 1173 | int spi; |
e9e315b4 | 1174 | |
97e03f52 JK |
1175 | /* ARG_PTR_TO_DYNPTR takes any type of dynptr */ |
1176 | if (arg_type == ARG_PTR_TO_DYNPTR) | |
1177 | return true; | |
1178 | ||
e9e315b4 | 1179 | dynptr_type = arg_to_dynptr_type(arg_type); |
27060531 KKD |
1180 | if (reg->type == CONST_PTR_TO_DYNPTR) { |
1181 | return reg->dynptr.type == dynptr_type; | |
1182 | } else { | |
79168a66 KKD |
1183 | spi = dynptr_get_spi(env, reg); |
1184 | if (spi < 0) | |
1185 | return false; | |
27060531 KKD |
1186 | return state->stack[spi].spilled_ptr.dynptr.type == dynptr_type; |
1187 | } | |
97e03f52 JK |
1188 | } |
1189 | ||
06accc87 AN |
1190 | static void __mark_reg_known_zero(struct bpf_reg_state *reg); |
1191 | ||
dfab99df CZ |
1192 | static bool in_rcu_cs(struct bpf_verifier_env *env); |
1193 | ||
1194 | static bool is_kfunc_rcu_protected(struct bpf_kfunc_call_arg_meta *meta); | |
1195 | ||
06accc87 | 1196 | static int mark_stack_slots_iter(struct bpf_verifier_env *env, |
dfab99df | 1197 | struct bpf_kfunc_call_arg_meta *meta, |
06accc87 AN |
1198 | struct bpf_reg_state *reg, int insn_idx, |
1199 | struct btf *btf, u32 btf_id, int nr_slots) | |
1200 | { | |
1201 | struct bpf_func_state *state = func(env, reg); | |
1202 | int spi, i, j, id; | |
1203 | ||
1204 | spi = iter_get_spi(env, reg, nr_slots); | |
1205 | if (spi < 0) | |
1206 | return spi; | |
1207 | ||
1208 | id = acquire_reference_state(env, insn_idx); | |
1209 | if (id < 0) | |
1210 | return id; | |
1211 | ||
1212 | for (i = 0; i < nr_slots; i++) { | |
1213 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1214 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1215 | ||
1216 | __mark_reg_known_zero(st); | |
1217 | st->type = PTR_TO_STACK; /* we don't have dedicated reg type */ | |
dfab99df CZ |
1218 | if (is_kfunc_rcu_protected(meta)) { |
1219 | if (in_rcu_cs(env)) | |
1220 | st->type |= MEM_RCU; | |
1221 | else | |
1222 | st->type |= PTR_UNTRUSTED; | |
1223 | } | |
06accc87 AN |
1224 | st->live |= REG_LIVE_WRITTEN; |
1225 | st->ref_obj_id = i == 0 ? id : 0; | |
1226 | st->iter.btf = btf; | |
1227 | st->iter.btf_id = btf_id; | |
1228 | st->iter.state = BPF_ITER_STATE_ACTIVE; | |
1229 | st->iter.depth = 0; | |
1230 | ||
1231 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1232 | slot->slot_type[j] = STACK_ITER; | |
1233 | ||
1234 | mark_stack_slot_scratched(env, spi - i); | |
1235 | } | |
1236 | ||
1237 | return 0; | |
1238 | } | |
1239 | ||
1240 | static int unmark_stack_slots_iter(struct bpf_verifier_env *env, | |
1241 | struct bpf_reg_state *reg, int nr_slots) | |
1242 | { | |
1243 | struct bpf_func_state *state = func(env, reg); | |
1244 | int spi, i, j; | |
1245 | ||
1246 | spi = iter_get_spi(env, reg, nr_slots); | |
1247 | if (spi < 0) | |
1248 | return spi; | |
1249 | ||
1250 | for (i = 0; i < nr_slots; i++) { | |
1251 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1252 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1253 | ||
1254 | if (i == 0) | |
1255 | WARN_ON_ONCE(release_reference(env, st->ref_obj_id)); | |
1256 | ||
1257 | __mark_reg_not_init(env, st); | |
1258 | ||
1259 | /* see unmark_stack_slots_dynptr() for why we need to set REG_LIVE_WRITTEN */ | |
1260 | st->live |= REG_LIVE_WRITTEN; | |
1261 | ||
1262 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1263 | slot->slot_type[j] = STACK_INVALID; | |
1264 | ||
1265 | mark_stack_slot_scratched(env, spi - i); | |
1266 | } | |
1267 | ||
1268 | return 0; | |
1269 | } | |
1270 | ||
1271 | static bool is_iter_reg_valid_uninit(struct bpf_verifier_env *env, | |
1272 | struct bpf_reg_state *reg, int nr_slots) | |
1273 | { | |
1274 | struct bpf_func_state *state = func(env, reg); | |
1275 | int spi, i, j; | |
1276 | ||
1277 | /* For -ERANGE (i.e. spi not falling into allocated stack slots), we | |
1278 | * will do check_mem_access to check and update stack bounds later, so | |
1279 | * return true for that case. | |
1280 | */ | |
1281 | spi = iter_get_spi(env, reg, nr_slots); | |
1282 | if (spi == -ERANGE) | |
1283 | return true; | |
1284 | if (spi < 0) | |
1285 | return false; | |
1286 | ||
1287 | for (i = 0; i < nr_slots; i++) { | |
1288 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1289 | ||
1290 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1291 | if (slot->slot_type[j] == STACK_ITER) | |
1292 | return false; | |
1293 | } | |
1294 | ||
1295 | return true; | |
1296 | } | |
1297 | ||
dfab99df | 1298 | static int is_iter_reg_valid_init(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
06accc87 AN |
1299 | struct btf *btf, u32 btf_id, int nr_slots) |
1300 | { | |
1301 | struct bpf_func_state *state = func(env, reg); | |
1302 | int spi, i, j; | |
1303 | ||
1304 | spi = iter_get_spi(env, reg, nr_slots); | |
1305 | if (spi < 0) | |
dfab99df | 1306 | return -EINVAL; |
06accc87 AN |
1307 | |
1308 | for (i = 0; i < nr_slots; i++) { | |
1309 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1310 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1311 | ||
dfab99df CZ |
1312 | if (st->type & PTR_UNTRUSTED) |
1313 | return -EPROTO; | |
06accc87 AN |
1314 | /* only main (first) slot has ref_obj_id set */ |
1315 | if (i == 0 && !st->ref_obj_id) | |
dfab99df | 1316 | return -EINVAL; |
06accc87 | 1317 | if (i != 0 && st->ref_obj_id) |
dfab99df | 1318 | return -EINVAL; |
06accc87 | 1319 | if (st->iter.btf != btf || st->iter.btf_id != btf_id) |
dfab99df | 1320 | return -EINVAL; |
06accc87 AN |
1321 | |
1322 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1323 | if (slot->slot_type[j] != STACK_ITER) | |
dfab99df | 1324 | return -EINVAL; |
06accc87 AN |
1325 | } |
1326 | ||
dfab99df | 1327 | return 0; |
06accc87 AN |
1328 | } |
1329 | ||
1330 | /* Check if given stack slot is "special": | |
1331 | * - spilled register state (STACK_SPILL); | |
1332 | * - dynptr state (STACK_DYNPTR); | |
1333 | * - iter state (STACK_ITER). | |
1334 | */ | |
1335 | static bool is_stack_slot_special(const struct bpf_stack_state *stack) | |
1336 | { | |
1337 | enum bpf_stack_slot_type type = stack->slot_type[BPF_REG_SIZE - 1]; | |
1338 | ||
1339 | switch (type) { | |
1340 | case STACK_SPILL: | |
1341 | case STACK_DYNPTR: | |
1342 | case STACK_ITER: | |
1343 | return true; | |
1344 | case STACK_INVALID: | |
1345 | case STACK_MISC: | |
1346 | case STACK_ZERO: | |
1347 | return false; | |
1348 | default: | |
1349 | WARN_ONCE(1, "unknown stack slot type %d\n", type); | |
1350 | return true; | |
1351 | } | |
1352 | } | |
1353 | ||
27113c59 MKL |
1354 | /* The reg state of a pointer or a bounded scalar was saved when |
1355 | * it was spilled to the stack. | |
1356 | */ | |
1357 | static bool is_spilled_reg(const struct bpf_stack_state *stack) | |
1358 | { | |
1359 | return stack->slot_type[BPF_REG_SIZE - 1] == STACK_SPILL; | |
1360 | } | |
1361 | ||
407958a0 AN |
1362 | static bool is_spilled_scalar_reg(const struct bpf_stack_state *stack) |
1363 | { | |
1364 | return stack->slot_type[BPF_REG_SIZE - 1] == STACK_SPILL && | |
1365 | stack->spilled_ptr.type == SCALAR_VALUE; | |
1366 | } | |
1367 | ||
354e8f19 MKL |
1368 | static void scrub_spilled_slot(u8 *stype) |
1369 | { | |
1370 | if (*stype != STACK_INVALID) | |
1371 | *stype = STACK_MISC; | |
1372 | } | |
1373 | ||
72f8a1de AN |
1374 | static void print_scalar_ranges(struct bpf_verifier_env *env, |
1375 | const struct bpf_reg_state *reg, | |
1376 | const char **sep) | |
1377 | { | |
1378 | struct { | |
1379 | const char *name; | |
1380 | u64 val; | |
1381 | bool omit; | |
1382 | } minmaxs[] = { | |
1383 | {"smin", reg->smin_value, reg->smin_value == S64_MIN}, | |
1384 | {"smax", reg->smax_value, reg->smax_value == S64_MAX}, | |
1385 | {"umin", reg->umin_value, reg->umin_value == 0}, | |
1386 | {"umax", reg->umax_value, reg->umax_value == U64_MAX}, | |
1387 | {"smin32", (s64)reg->s32_min_value, reg->s32_min_value == S32_MIN}, | |
1388 | {"smax32", (s64)reg->s32_max_value, reg->s32_max_value == S32_MAX}, | |
1389 | {"umin32", reg->u32_min_value, reg->u32_min_value == 0}, | |
1390 | {"umax32", reg->u32_max_value, reg->u32_max_value == U32_MAX}, | |
1391 | }, *m1, *m2, *mend = &minmaxs[ARRAY_SIZE(minmaxs)]; | |
1392 | bool neg1, neg2; | |
1393 | ||
1394 | for (m1 = &minmaxs[0]; m1 < mend; m1++) { | |
1395 | if (m1->omit) | |
1396 | continue; | |
1397 | ||
1398 | neg1 = m1->name[0] == 's' && (s64)m1->val < 0; | |
1399 | ||
1400 | verbose(env, "%s%s=", *sep, m1->name); | |
1401 | *sep = ","; | |
1402 | ||
1403 | for (m2 = m1 + 2; m2 < mend; m2 += 2) { | |
1404 | if (m2->omit || m2->val != m1->val) | |
1405 | continue; | |
1406 | /* don't mix negatives with positives */ | |
1407 | neg2 = m2->name[0] == 's' && (s64)m2->val < 0; | |
1408 | if (neg2 != neg1) | |
1409 | continue; | |
1410 | m2->omit = true; | |
1411 | verbose(env, "%s=", m2->name); | |
1412 | } | |
1413 | ||
1414 | verbose(env, m1->name[0] == 's' ? "%lld" : "%llu", m1->val); | |
1415 | } | |
1416 | } | |
1417 | ||
61bd5218 | 1418 | static void print_verifier_state(struct bpf_verifier_env *env, |
0f55f9ed CL |
1419 | const struct bpf_func_state *state, |
1420 | bool print_all) | |
17a52670 | 1421 | { |
f4d7e40a | 1422 | const struct bpf_reg_state *reg; |
17a52670 AS |
1423 | enum bpf_reg_type t; |
1424 | int i; | |
1425 | ||
f4d7e40a AS |
1426 | if (state->frameno) |
1427 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 1428 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
1429 | reg = &state->regs[i]; |
1430 | t = reg->type; | |
17a52670 AS |
1431 | if (t == NOT_INIT) |
1432 | continue; | |
0f55f9ed CL |
1433 | if (!print_all && !reg_scratched(env, i)) |
1434 | continue; | |
4e92024a AS |
1435 | verbose(env, " R%d", i); |
1436 | print_liveness(env, reg->live); | |
7df5072c | 1437 | verbose(env, "="); |
b5dc0163 AS |
1438 | if (t == SCALAR_VALUE && reg->precise) |
1439 | verbose(env, "P"); | |
f1174f77 EC |
1440 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
1441 | tnum_is_const(reg->var_off)) { | |
1442 | /* reg->off should be 0 for SCALAR_VALUE */ | |
7df5072c | 1443 | verbose(env, "%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
61bd5218 | 1444 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 1445 | } else { |
7df5072c ML |
1446 | const char *sep = ""; |
1447 | ||
1448 | verbose(env, "%s", reg_type_str(env, t)); | |
5844101a | 1449 | if (base_type(t) == PTR_TO_BTF_ID) |
b32a5dae | 1450 | verbose(env, "%s", btf_type_name(reg->btf, reg->btf_id)); |
7df5072c ML |
1451 | verbose(env, "("); |
1452 | /* | |
1453 | * _a stands for append, was shortened to avoid multiline statements below. | |
1454 | * This macro is used to output a comma separated list of attributes. | |
1455 | */ | |
1456 | #define verbose_a(fmt, ...) ({ verbose(env, "%s" fmt, sep, __VA_ARGS__); sep = ","; }) | |
1457 | ||
1458 | if (reg->id) | |
1459 | verbose_a("id=%d", reg->id); | |
a28ace78 | 1460 | if (reg->ref_obj_id) |
7df5072c | 1461 | verbose_a("ref_obj_id=%d", reg->ref_obj_id); |
6a3cd331 DM |
1462 | if (type_is_non_owning_ref(reg->type)) |
1463 | verbose_a("%s", "non_own_ref"); | |
f1174f77 | 1464 | if (t != SCALAR_VALUE) |
7df5072c | 1465 | verbose_a("off=%d", reg->off); |
de8f3a83 | 1466 | if (type_is_pkt_pointer(t)) |
7df5072c | 1467 | verbose_a("r=%d", reg->range); |
c25b2ae1 HL |
1468 | else if (base_type(t) == CONST_PTR_TO_MAP || |
1469 | base_type(t) == PTR_TO_MAP_KEY || | |
1470 | base_type(t) == PTR_TO_MAP_VALUE) | |
7df5072c ML |
1471 | verbose_a("ks=%d,vs=%d", |
1472 | reg->map_ptr->key_size, | |
1473 | reg->map_ptr->value_size); | |
7d1238f2 EC |
1474 | if (tnum_is_const(reg->var_off)) { |
1475 | /* Typically an immediate SCALAR_VALUE, but | |
1476 | * could be a pointer whose offset is too big | |
1477 | * for reg->off | |
1478 | */ | |
7df5072c | 1479 | verbose_a("imm=%llx", reg->var_off.value); |
7d1238f2 | 1480 | } else { |
72f8a1de | 1481 | print_scalar_ranges(env, reg, &sep); |
7d1238f2 EC |
1482 | if (!tnum_is_unknown(reg->var_off)) { |
1483 | char tn_buf[48]; | |
f1174f77 | 1484 | |
7d1238f2 | 1485 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
7df5072c | 1486 | verbose_a("var_off=%s", tn_buf); |
7d1238f2 | 1487 | } |
f1174f77 | 1488 | } |
7df5072c ML |
1489 | #undef verbose_a |
1490 | ||
61bd5218 | 1491 | verbose(env, ")"); |
f1174f77 | 1492 | } |
17a52670 | 1493 | } |
638f5b90 | 1494 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
1495 | char types_buf[BPF_REG_SIZE + 1]; |
1496 | bool valid = false; | |
1497 | int j; | |
1498 | ||
1499 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
1500 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
1501 | valid = true; | |
d54e0f6c | 1502 | types_buf[j] = slot_type_char[state->stack[i].slot_type[j]]; |
8efea21d EC |
1503 | } |
1504 | types_buf[BPF_REG_SIZE] = 0; | |
1505 | if (!valid) | |
1506 | continue; | |
0f55f9ed CL |
1507 | if (!print_all && !stack_slot_scratched(env, i)) |
1508 | continue; | |
d54e0f6c AN |
1509 | switch (state->stack[i].slot_type[BPF_REG_SIZE - 1]) { |
1510 | case STACK_SPILL: | |
b5dc0163 AS |
1511 | reg = &state->stack[i].spilled_ptr; |
1512 | t = reg->type; | |
d54e0f6c AN |
1513 | |
1514 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1515 | print_liveness(env, reg->live); | |
7df5072c | 1516 | verbose(env, "=%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
b5dc0163 AS |
1517 | if (t == SCALAR_VALUE && reg->precise) |
1518 | verbose(env, "P"); | |
1519 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
1520 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
d54e0f6c AN |
1521 | break; |
1522 | case STACK_DYNPTR: | |
1523 | i += BPF_DYNPTR_NR_SLOTS - 1; | |
1524 | reg = &state->stack[i].spilled_ptr; | |
1525 | ||
1526 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1527 | print_liveness(env, reg->live); | |
1528 | verbose(env, "=dynptr_%s", dynptr_type_str(reg->dynptr.type)); | |
1529 | if (reg->ref_obj_id) | |
1530 | verbose(env, "(ref_id=%d)", reg->ref_obj_id); | |
1531 | break; | |
06accc87 AN |
1532 | case STACK_ITER: |
1533 | /* only main slot has ref_obj_id set; skip others */ | |
1534 | reg = &state->stack[i].spilled_ptr; | |
1535 | if (!reg->ref_obj_id) | |
1536 | continue; | |
1537 | ||
1538 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1539 | print_liveness(env, reg->live); | |
1540 | verbose(env, "=iter_%s(ref_id=%d,state=%s,depth=%u)", | |
1541 | iter_type_str(reg->iter.btf, reg->iter.btf_id), | |
1542 | reg->ref_obj_id, iter_state_str(reg->iter.state), | |
1543 | reg->iter.depth); | |
1544 | break; | |
d54e0f6c AN |
1545 | case STACK_MISC: |
1546 | case STACK_ZERO: | |
1547 | default: | |
1548 | reg = &state->stack[i].spilled_ptr; | |
1549 | ||
1550 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1551 | types_buf[j] = slot_type_char[state->stack[i].slot_type[j]]; | |
1552 | types_buf[BPF_REG_SIZE] = 0; | |
1553 | ||
1554 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1555 | print_liveness(env, reg->live); | |
8efea21d | 1556 | verbose(env, "=%s", types_buf); |
d54e0f6c | 1557 | break; |
b5dc0163 | 1558 | } |
17a52670 | 1559 | } |
fd978bf7 JS |
1560 | if (state->acquired_refs && state->refs[0].id) { |
1561 | verbose(env, " refs=%d", state->refs[0].id); | |
1562 | for (i = 1; i < state->acquired_refs; i++) | |
1563 | if (state->refs[i].id) | |
1564 | verbose(env, ",%d", state->refs[i].id); | |
1565 | } | |
bfc6bb74 AS |
1566 | if (state->in_callback_fn) |
1567 | verbose(env, " cb"); | |
1568 | if (state->in_async_callback_fn) | |
1569 | verbose(env, " async_cb"); | |
61bd5218 | 1570 | verbose(env, "\n"); |
1a8a315f AN |
1571 | if (!print_all) |
1572 | mark_verifier_state_clean(env); | |
17a52670 AS |
1573 | } |
1574 | ||
2e576648 CL |
1575 | static inline u32 vlog_alignment(u32 pos) |
1576 | { | |
1577 | return round_up(max(pos + BPF_LOG_MIN_ALIGNMENT / 2, BPF_LOG_ALIGNMENT), | |
1578 | BPF_LOG_MIN_ALIGNMENT) - pos - 1; | |
1579 | } | |
1580 | ||
1581 | static void print_insn_state(struct bpf_verifier_env *env, | |
1582 | const struct bpf_func_state *state) | |
1583 | { | |
12166409 | 1584 | if (env->prev_log_pos && env->prev_log_pos == env->log.end_pos) { |
2e576648 | 1585 | /* remove new line character */ |
12166409 AN |
1586 | bpf_vlog_reset(&env->log, env->prev_log_pos - 1); |
1587 | verbose(env, "%*c;", vlog_alignment(env->prev_insn_print_pos), ' '); | |
2e576648 CL |
1588 | } else { |
1589 | verbose(env, "%d:", env->insn_idx); | |
1590 | } | |
1591 | print_verifier_state(env, state, false); | |
17a52670 AS |
1592 | } |
1593 | ||
c69431aa LB |
1594 | /* copy array src of length n * size bytes to dst. dst is reallocated if it's too |
1595 | * small to hold src. This is different from krealloc since we don't want to preserve | |
1596 | * the contents of dst. | |
1597 | * | |
1598 | * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could | |
1599 | * not be allocated. | |
638f5b90 | 1600 | */ |
c69431aa | 1601 | static void *copy_array(void *dst, const void *src, size_t n, size_t size, gfp_t flags) |
638f5b90 | 1602 | { |
45435d8d KC |
1603 | size_t alloc_bytes; |
1604 | void *orig = dst; | |
c69431aa LB |
1605 | size_t bytes; |
1606 | ||
1607 | if (ZERO_OR_NULL_PTR(src)) | |
1608 | goto out; | |
1609 | ||
1610 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
1611 | return NULL; | |
1612 | ||
45435d8d KC |
1613 | alloc_bytes = max(ksize(orig), kmalloc_size_roundup(bytes)); |
1614 | dst = krealloc(orig, alloc_bytes, flags); | |
1615 | if (!dst) { | |
1616 | kfree(orig); | |
1617 | return NULL; | |
c69431aa LB |
1618 | } |
1619 | ||
1620 | memcpy(dst, src, bytes); | |
1621 | out: | |
1622 | return dst ? dst : ZERO_SIZE_PTR; | |
1623 | } | |
1624 | ||
1625 | /* resize an array from old_n items to new_n items. the array is reallocated if it's too | |
1626 | * small to hold new_n items. new items are zeroed out if the array grows. | |
1627 | * | |
1628 | * Contrary to krealloc_array, does not free arr if new_n is zero. | |
1629 | */ | |
1630 | static void *realloc_array(void *arr, size_t old_n, size_t new_n, size_t size) | |
1631 | { | |
ceb35b66 | 1632 | size_t alloc_size; |
42378a9c KC |
1633 | void *new_arr; |
1634 | ||
c69431aa LB |
1635 | if (!new_n || old_n == new_n) |
1636 | goto out; | |
1637 | ||
ceb35b66 KC |
1638 | alloc_size = kmalloc_size_roundup(size_mul(new_n, size)); |
1639 | new_arr = krealloc(arr, alloc_size, GFP_KERNEL); | |
42378a9c KC |
1640 | if (!new_arr) { |
1641 | kfree(arr); | |
c69431aa | 1642 | return NULL; |
42378a9c KC |
1643 | } |
1644 | arr = new_arr; | |
c69431aa LB |
1645 | |
1646 | if (new_n > old_n) | |
1647 | memset(arr + old_n * size, 0, (new_n - old_n) * size); | |
1648 | ||
1649 | out: | |
1650 | return arr ? arr : ZERO_SIZE_PTR; | |
1651 | } | |
1652 | ||
1653 | static int copy_reference_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1654 | { | |
1655 | dst->refs = copy_array(dst->refs, src->refs, src->acquired_refs, | |
1656 | sizeof(struct bpf_reference_state), GFP_KERNEL); | |
1657 | if (!dst->refs) | |
1658 | return -ENOMEM; | |
1659 | ||
1660 | dst->acquired_refs = src->acquired_refs; | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | static int copy_stack_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1665 | { | |
1666 | size_t n = src->allocated_stack / BPF_REG_SIZE; | |
1667 | ||
1668 | dst->stack = copy_array(dst->stack, src->stack, n, sizeof(struct bpf_stack_state), | |
1669 | GFP_KERNEL); | |
1670 | if (!dst->stack) | |
1671 | return -ENOMEM; | |
1672 | ||
1673 | dst->allocated_stack = src->allocated_stack; | |
1674 | return 0; | |
1675 | } | |
1676 | ||
1677 | static int resize_reference_state(struct bpf_func_state *state, size_t n) | |
1678 | { | |
1679 | state->refs = realloc_array(state->refs, state->acquired_refs, n, | |
1680 | sizeof(struct bpf_reference_state)); | |
1681 | if (!state->refs) | |
1682 | return -ENOMEM; | |
1683 | ||
1684 | state->acquired_refs = n; | |
1685 | return 0; | |
1686 | } | |
1687 | ||
1688 | static int grow_stack_state(struct bpf_func_state *state, int size) | |
1689 | { | |
1690 | size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE; | |
1691 | ||
1692 | if (old_n >= n) | |
1693 | return 0; | |
1694 | ||
1695 | state->stack = realloc_array(state->stack, old_n, n, sizeof(struct bpf_stack_state)); | |
1696 | if (!state->stack) | |
1697 | return -ENOMEM; | |
1698 | ||
1699 | state->allocated_stack = size; | |
1700 | return 0; | |
fd978bf7 JS |
1701 | } |
1702 | ||
1703 | /* Acquire a pointer id from the env and update the state->refs to include | |
1704 | * this new pointer reference. | |
1705 | * On success, returns a valid pointer id to associate with the register | |
1706 | * On failure, returns a negative errno. | |
638f5b90 | 1707 | */ |
fd978bf7 | 1708 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 1709 | { |
fd978bf7 JS |
1710 | struct bpf_func_state *state = cur_func(env); |
1711 | int new_ofs = state->acquired_refs; | |
1712 | int id, err; | |
1713 | ||
c69431aa | 1714 | err = resize_reference_state(state, state->acquired_refs + 1); |
fd978bf7 JS |
1715 | if (err) |
1716 | return err; | |
1717 | id = ++env->id_gen; | |
1718 | state->refs[new_ofs].id = id; | |
1719 | state->refs[new_ofs].insn_idx = insn_idx; | |
9d9d00ac | 1720 | state->refs[new_ofs].callback_ref = state->in_callback_fn ? state->frameno : 0; |
638f5b90 | 1721 | |
fd978bf7 JS |
1722 | return id; |
1723 | } | |
1724 | ||
1725 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 1726 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
1727 | { |
1728 | int i, last_idx; | |
1729 | ||
fd978bf7 JS |
1730 | last_idx = state->acquired_refs - 1; |
1731 | for (i = 0; i < state->acquired_refs; i++) { | |
1732 | if (state->refs[i].id == ptr_id) { | |
9d9d00ac KKD |
1733 | /* Cannot release caller references in callbacks */ |
1734 | if (state->in_callback_fn && state->refs[i].callback_ref != state->frameno) | |
1735 | return -EINVAL; | |
fd978bf7 JS |
1736 | if (last_idx && i != last_idx) |
1737 | memcpy(&state->refs[i], &state->refs[last_idx], | |
1738 | sizeof(*state->refs)); | |
1739 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
1740 | state->acquired_refs--; | |
638f5b90 | 1741 | return 0; |
638f5b90 | 1742 | } |
638f5b90 | 1743 | } |
46f8bc92 | 1744 | return -EINVAL; |
fd978bf7 JS |
1745 | } |
1746 | ||
f4d7e40a AS |
1747 | static void free_func_state(struct bpf_func_state *state) |
1748 | { | |
5896351e AS |
1749 | if (!state) |
1750 | return; | |
fd978bf7 | 1751 | kfree(state->refs); |
f4d7e40a AS |
1752 | kfree(state->stack); |
1753 | kfree(state); | |
1754 | } | |
1755 | ||
b5dc0163 AS |
1756 | static void clear_jmp_history(struct bpf_verifier_state *state) |
1757 | { | |
1758 | kfree(state->jmp_history); | |
1759 | state->jmp_history = NULL; | |
1760 | state->jmp_history_cnt = 0; | |
1761 | } | |
1762 | ||
1969db47 AS |
1763 | static void free_verifier_state(struct bpf_verifier_state *state, |
1764 | bool free_self) | |
638f5b90 | 1765 | { |
f4d7e40a AS |
1766 | int i; |
1767 | ||
1768 | for (i = 0; i <= state->curframe; i++) { | |
1769 | free_func_state(state->frame[i]); | |
1770 | state->frame[i] = NULL; | |
1771 | } | |
b5dc0163 | 1772 | clear_jmp_history(state); |
1969db47 AS |
1773 | if (free_self) |
1774 | kfree(state); | |
638f5b90 AS |
1775 | } |
1776 | ||
1777 | /* copy verifier state from src to dst growing dst stack space | |
1778 | * when necessary to accommodate larger src stack | |
1779 | */ | |
f4d7e40a AS |
1780 | static int copy_func_state(struct bpf_func_state *dst, |
1781 | const struct bpf_func_state *src) | |
638f5b90 AS |
1782 | { |
1783 | int err; | |
1784 | ||
fd978bf7 JS |
1785 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); |
1786 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
1787 | if (err) |
1788 | return err; | |
638f5b90 AS |
1789 | return copy_stack_state(dst, src); |
1790 | } | |
1791 | ||
f4d7e40a AS |
1792 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
1793 | const struct bpf_verifier_state *src) | |
1794 | { | |
1795 | struct bpf_func_state *dst; | |
1796 | int i, err; | |
1797 | ||
06ab6a50 LB |
1798 | dst_state->jmp_history = copy_array(dst_state->jmp_history, src->jmp_history, |
1799 | src->jmp_history_cnt, sizeof(struct bpf_idx_pair), | |
1800 | GFP_USER); | |
1801 | if (!dst_state->jmp_history) | |
1802 | return -ENOMEM; | |
b5dc0163 AS |
1803 | dst_state->jmp_history_cnt = src->jmp_history_cnt; |
1804 | ||
f18b03fa KKD |
1805 | /* if dst has more stack frames then src frame, free them, this is also |
1806 | * necessary in case of exceptional exits using bpf_throw. | |
1807 | */ | |
f4d7e40a AS |
1808 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { |
1809 | free_func_state(dst_state->frame[i]); | |
1810 | dst_state->frame[i] = NULL; | |
1811 | } | |
979d63d5 | 1812 | dst_state->speculative = src->speculative; |
9bb00b28 | 1813 | dst_state->active_rcu_lock = src->active_rcu_lock; |
f4d7e40a | 1814 | dst_state->curframe = src->curframe; |
d0d78c1d KKD |
1815 | dst_state->active_lock.ptr = src->active_lock.ptr; |
1816 | dst_state->active_lock.id = src->active_lock.id; | |
2589726d AS |
1817 | dst_state->branches = src->branches; |
1818 | dst_state->parent = src->parent; | |
b5dc0163 AS |
1819 | dst_state->first_insn_idx = src->first_insn_idx; |
1820 | dst_state->last_insn_idx = src->last_insn_idx; | |
2793a8b0 | 1821 | dst_state->dfs_depth = src->dfs_depth; |
ab5cfac1 | 1822 | dst_state->callback_unroll_depth = src->callback_unroll_depth; |
2a099282 | 1823 | dst_state->used_as_loop_entry = src->used_as_loop_entry; |
f4d7e40a AS |
1824 | for (i = 0; i <= src->curframe; i++) { |
1825 | dst = dst_state->frame[i]; | |
1826 | if (!dst) { | |
1827 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
1828 | if (!dst) | |
1829 | return -ENOMEM; | |
1830 | dst_state->frame[i] = dst; | |
1831 | } | |
1832 | err = copy_func_state(dst, src->frame[i]); | |
1833 | if (err) | |
1834 | return err; | |
1835 | } | |
1836 | return 0; | |
1837 | } | |
1838 | ||
3c4e420c EZ |
1839 | static u32 state_htab_size(struct bpf_verifier_env *env) |
1840 | { | |
1841 | return env->prog->len; | |
1842 | } | |
1843 | ||
1844 | static struct bpf_verifier_state_list **explored_state(struct bpf_verifier_env *env, int idx) | |
1845 | { | |
1846 | struct bpf_verifier_state *cur = env->cur_state; | |
1847 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
1848 | ||
1849 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
1850 | } | |
1851 | ||
4c97259a EZ |
1852 | static bool same_callsites(struct bpf_verifier_state *a, struct bpf_verifier_state *b) |
1853 | { | |
1854 | int fr; | |
1855 | ||
1856 | if (a->curframe != b->curframe) | |
1857 | return false; | |
1858 | ||
1859 | for (fr = a->curframe; fr >= 0; fr--) | |
1860 | if (a->frame[fr]->callsite != b->frame[fr]->callsite) | |
1861 | return false; | |
1862 | ||
1863 | return true; | |
1864 | } | |
1865 | ||
2a099282 EZ |
1866 | /* Open coded iterators allow back-edges in the state graph in order to |
1867 | * check unbounded loops that iterators. | |
1868 | * | |
1869 | * In is_state_visited() it is necessary to know if explored states are | |
1870 | * part of some loops in order to decide whether non-exact states | |
1871 | * comparison could be used: | |
1872 | * - non-exact states comparison establishes sub-state relation and uses | |
1873 | * read and precision marks to do so, these marks are propagated from | |
1874 | * children states and thus are not guaranteed to be final in a loop; | |
1875 | * - exact states comparison just checks if current and explored states | |
1876 | * are identical (and thus form a back-edge). | |
1877 | * | |
1878 | * Paper "A New Algorithm for Identifying Loops in Decompilation" | |
1879 | * by Tao Wei, Jian Mao, Wei Zou and Yu Chen [1] presents a convenient | |
1880 | * algorithm for loop structure detection and gives an overview of | |
1881 | * relevant terminology. It also has helpful illustrations. | |
1882 | * | |
1883 | * [1] https://api.semanticscholar.org/CorpusID:15784067 | |
1884 | * | |
1885 | * We use a similar algorithm but because loop nested structure is | |
1886 | * irrelevant for verifier ours is significantly simpler and resembles | |
1887 | * strongly connected components algorithm from Sedgewick's textbook. | |
1888 | * | |
1889 | * Define topmost loop entry as a first node of the loop traversed in a | |
1890 | * depth first search starting from initial state. The goal of the loop | |
1891 | * tracking algorithm is to associate topmost loop entries with states | |
1892 | * derived from these entries. | |
1893 | * | |
1894 | * For each step in the DFS states traversal algorithm needs to identify | |
1895 | * the following situations: | |
1896 | * | |
1897 | * initial initial initial | |
1898 | * | | | | |
1899 | * V V V | |
1900 | * ... ... .---------> hdr | |
1901 | * | | | | | |
1902 | * V V | V | |
1903 | * cur .-> succ | .------... | |
1904 | * | | | | | | | |
1905 | * V | V | V V | |
1906 | * succ '-- cur | ... ... | |
1907 | * | | | | |
1908 | * | V V | |
1909 | * | succ <- cur | |
1910 | * | | | |
1911 | * | V | |
1912 | * | ... | |
1913 | * | | | |
1914 | * '----' | |
1915 | * | |
1916 | * (A) successor state of cur (B) successor state of cur or it's entry | |
1917 | * not yet traversed are in current DFS path, thus cur and succ | |
1918 | * are members of the same outermost loop | |
1919 | * | |
1920 | * initial initial | |
1921 | * | | | |
1922 | * V V | |
1923 | * ... ... | |
1924 | * | | | |
1925 | * V V | |
1926 | * .------... .------... | |
1927 | * | | | | | |
1928 | * V V V V | |
1929 | * .-> hdr ... ... ... | |
1930 | * | | | | | | |
1931 | * | V V V V | |
1932 | * | succ <- cur succ <- cur | |
1933 | * | | | | |
1934 | * | V V | |
1935 | * | ... ... | |
1936 | * | | | | |
1937 | * '----' exit | |
1938 | * | |
1939 | * (C) successor state of cur is a part of some loop but this loop | |
1940 | * does not include cur or successor state is not in a loop at all. | |
1941 | * | |
1942 | * Algorithm could be described as the following python code: | |
1943 | * | |
1944 | * traversed = set() # Set of traversed nodes | |
1945 | * entries = {} # Mapping from node to loop entry | |
1946 | * depths = {} # Depth level assigned to graph node | |
1947 | * path = set() # Current DFS path | |
1948 | * | |
1949 | * # Find outermost loop entry known for n | |
1950 | * def get_loop_entry(n): | |
1951 | * h = entries.get(n, None) | |
1952 | * while h in entries and entries[h] != h: | |
1953 | * h = entries[h] | |
1954 | * return h | |
1955 | * | |
1956 | * # Update n's loop entry if h's outermost entry comes | |
1957 | * # before n's outermost entry in current DFS path. | |
1958 | * def update_loop_entry(n, h): | |
1959 | * n1 = get_loop_entry(n) or n | |
1960 | * h1 = get_loop_entry(h) or h | |
1961 | * if h1 in path and depths[h1] <= depths[n1]: | |
1962 | * entries[n] = h1 | |
1963 | * | |
1964 | * def dfs(n, depth): | |
1965 | * traversed.add(n) | |
1966 | * path.add(n) | |
1967 | * depths[n] = depth | |
1968 | * for succ in G.successors(n): | |
1969 | * if succ not in traversed: | |
1970 | * # Case A: explore succ and update cur's loop entry | |
1971 | * # only if succ's entry is in current DFS path. | |
1972 | * dfs(succ, depth + 1) | |
1973 | * h = get_loop_entry(succ) | |
1974 | * update_loop_entry(n, h) | |
1975 | * else: | |
1976 | * # Case B or C depending on `h1 in path` check in update_loop_entry(). | |
1977 | * update_loop_entry(n, succ) | |
1978 | * path.remove(n) | |
1979 | * | |
1980 | * To adapt this algorithm for use with verifier: | |
1981 | * - use st->branch == 0 as a signal that DFS of succ had been finished | |
1982 | * and cur's loop entry has to be updated (case A), handle this in | |
1983 | * update_branch_counts(); | |
1984 | * - use st->branch > 0 as a signal that st is in the current DFS path; | |
1985 | * - handle cases B and C in is_state_visited(); | |
1986 | * - update topmost loop entry for intermediate states in get_loop_entry(). | |
1987 | */ | |
1988 | static struct bpf_verifier_state *get_loop_entry(struct bpf_verifier_state *st) | |
1989 | { | |
1990 | struct bpf_verifier_state *topmost = st->loop_entry, *old; | |
1991 | ||
1992 | while (topmost && topmost->loop_entry && topmost != topmost->loop_entry) | |
1993 | topmost = topmost->loop_entry; | |
1994 | /* Update loop entries for intermediate states to avoid this | |
1995 | * traversal in future get_loop_entry() calls. | |
1996 | */ | |
1997 | while (st && st->loop_entry != topmost) { | |
1998 | old = st->loop_entry; | |
1999 | st->loop_entry = topmost; | |
2000 | st = old; | |
2001 | } | |
2002 | return topmost; | |
2003 | } | |
2004 | ||
2005 | static void update_loop_entry(struct bpf_verifier_state *cur, struct bpf_verifier_state *hdr) | |
2006 | { | |
2007 | struct bpf_verifier_state *cur1, *hdr1; | |
2008 | ||
2009 | cur1 = get_loop_entry(cur) ?: cur; | |
2010 | hdr1 = get_loop_entry(hdr) ?: hdr; | |
2011 | /* The head1->branches check decides between cases B and C in | |
2012 | * comment for get_loop_entry(). If hdr1->branches == 0 then | |
2013 | * head's topmost loop entry is not in current DFS path, | |
2014 | * hence 'cur' and 'hdr' are not in the same loop and there is | |
2015 | * no need to update cur->loop_entry. | |
2016 | */ | |
2017 | if (hdr1->branches && hdr1->dfs_depth <= cur1->dfs_depth) { | |
2018 | cur->loop_entry = hdr; | |
2019 | hdr->used_as_loop_entry = true; | |
2020 | } | |
2021 | } | |
2022 | ||
2589726d AS |
2023 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
2024 | { | |
2025 | while (st) { | |
2026 | u32 br = --st->branches; | |
2027 | ||
2a099282 EZ |
2028 | /* br == 0 signals that DFS exploration for 'st' is finished, |
2029 | * thus it is necessary to update parent's loop entry if it | |
2030 | * turned out that st is a part of some loop. | |
2031 | * This is a part of 'case A' in get_loop_entry() comment. | |
2032 | */ | |
2033 | if (br == 0 && st->parent && st->loop_entry) | |
2034 | update_loop_entry(st->parent, st->loop_entry); | |
2035 | ||
2589726d AS |
2036 | /* WARN_ON(br > 1) technically makes sense here, |
2037 | * but see comment in push_stack(), hence: | |
2038 | */ | |
2039 | WARN_ONCE((int)br < 0, | |
2040 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
2041 | br); | |
2042 | if (br) | |
2043 | break; | |
2044 | st = st->parent; | |
2045 | } | |
2046 | } | |
2047 | ||
638f5b90 | 2048 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 2049 | int *insn_idx, bool pop_log) |
638f5b90 AS |
2050 | { |
2051 | struct bpf_verifier_state *cur = env->cur_state; | |
2052 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
2053 | int err; | |
17a52670 AS |
2054 | |
2055 | if (env->head == NULL) | |
638f5b90 | 2056 | return -ENOENT; |
17a52670 | 2057 | |
638f5b90 AS |
2058 | if (cur) { |
2059 | err = copy_verifier_state(cur, &head->st); | |
2060 | if (err) | |
2061 | return err; | |
2062 | } | |
6f8a57cc AN |
2063 | if (pop_log) |
2064 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
2065 | if (insn_idx) |
2066 | *insn_idx = head->insn_idx; | |
17a52670 | 2067 | if (prev_insn_idx) |
638f5b90 AS |
2068 | *prev_insn_idx = head->prev_insn_idx; |
2069 | elem = head->next; | |
1969db47 | 2070 | free_verifier_state(&head->st, false); |
638f5b90 | 2071 | kfree(head); |
17a52670 AS |
2072 | env->head = elem; |
2073 | env->stack_size--; | |
638f5b90 | 2074 | return 0; |
17a52670 AS |
2075 | } |
2076 | ||
58e2af8b | 2077 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
2078 | int insn_idx, int prev_insn_idx, |
2079 | bool speculative) | |
17a52670 | 2080 | { |
638f5b90 | 2081 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 2082 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 2083 | int err; |
17a52670 | 2084 | |
638f5b90 | 2085 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
2086 | if (!elem) |
2087 | goto err; | |
2088 | ||
17a52670 AS |
2089 | elem->insn_idx = insn_idx; |
2090 | elem->prev_insn_idx = prev_insn_idx; | |
2091 | elem->next = env->head; | |
12166409 | 2092 | elem->log_pos = env->log.end_pos; |
17a52670 AS |
2093 | env->head = elem; |
2094 | env->stack_size++; | |
1969db47 AS |
2095 | err = copy_verifier_state(&elem->st, cur); |
2096 | if (err) | |
2097 | goto err; | |
979d63d5 | 2098 | elem->st.speculative |= speculative; |
b285fcb7 AS |
2099 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
2100 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
2101 | env->stack_size); | |
17a52670 AS |
2102 | goto err; |
2103 | } | |
2589726d AS |
2104 | if (elem->st.parent) { |
2105 | ++elem->st.parent->branches; | |
2106 | /* WARN_ON(branches > 2) technically makes sense here, | |
2107 | * but | |
2108 | * 1. speculative states will bump 'branches' for non-branch | |
2109 | * instructions | |
2110 | * 2. is_state_visited() heuristics may decide not to create | |
2111 | * a new state for a sequence of branches and all such current | |
2112 | * and cloned states will be pointing to a single parent state | |
2113 | * which might have large 'branches' count. | |
2114 | */ | |
2115 | } | |
17a52670 AS |
2116 | return &elem->st; |
2117 | err: | |
5896351e AS |
2118 | free_verifier_state(env->cur_state, true); |
2119 | env->cur_state = NULL; | |
17a52670 | 2120 | /* pop all elements and return */ |
6f8a57cc | 2121 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
2122 | return NULL; |
2123 | } | |
2124 | ||
2125 | #define CALLER_SAVED_REGS 6 | |
2126 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
2127 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
2128 | }; | |
2129 | ||
e688c3db AS |
2130 | /* This helper doesn't clear reg->id */ |
2131 | static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
b03c9f9f | 2132 | { |
b03c9f9f EC |
2133 | reg->var_off = tnum_const(imm); |
2134 | reg->smin_value = (s64)imm; | |
2135 | reg->smax_value = (s64)imm; | |
2136 | reg->umin_value = imm; | |
2137 | reg->umax_value = imm; | |
3f50f132 JF |
2138 | |
2139 | reg->s32_min_value = (s32)imm; | |
2140 | reg->s32_max_value = (s32)imm; | |
2141 | reg->u32_min_value = (u32)imm; | |
2142 | reg->u32_max_value = (u32)imm; | |
2143 | } | |
2144 | ||
e688c3db AS |
2145 | /* Mark the unknown part of a register (variable offset or scalar value) as |
2146 | * known to have the value @imm. | |
2147 | */ | |
2148 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
2149 | { | |
a73bf9f2 | 2150 | /* Clear off and union(map_ptr, range) */ |
e688c3db AS |
2151 | memset(((u8 *)reg) + sizeof(reg->type), 0, |
2152 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
a73bf9f2 AN |
2153 | reg->id = 0; |
2154 | reg->ref_obj_id = 0; | |
e688c3db AS |
2155 | ___mark_reg_known(reg, imm); |
2156 | } | |
2157 | ||
3f50f132 JF |
2158 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) |
2159 | { | |
2160 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
2161 | reg->s32_min_value = (s32)imm; | |
2162 | reg->s32_max_value = (s32)imm; | |
2163 | reg->u32_min_value = (u32)imm; | |
2164 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
2165 | } |
2166 | ||
f1174f77 EC |
2167 | /* Mark the 'variable offset' part of a register as zero. This should be |
2168 | * used only on registers holding a pointer type. | |
2169 | */ | |
2170 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 2171 | { |
b03c9f9f | 2172 | __mark_reg_known(reg, 0); |
f1174f77 | 2173 | } |
a9789ef9 | 2174 | |
cc2b14d5 AS |
2175 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
2176 | { | |
2177 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
2178 | reg->type = SCALAR_VALUE; |
2179 | } | |
2180 | ||
61bd5218 JK |
2181 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
2182 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
2183 | { |
2184 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 2185 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
2186 | /* Something bad happened, let's kill all regs */ |
2187 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 2188 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
2189 | return; |
2190 | } | |
2191 | __mark_reg_known_zero(regs + regno); | |
2192 | } | |
2193 | ||
27060531 | 2194 | static void __mark_dynptr_reg(struct bpf_reg_state *reg, enum bpf_dynptr_type type, |
f8064ab9 | 2195 | bool first_slot, int dynptr_id) |
27060531 KKD |
2196 | { |
2197 | /* reg->type has no meaning for STACK_DYNPTR, but when we set reg for | |
2198 | * callback arguments, it does need to be CONST_PTR_TO_DYNPTR, so simply | |
2199 | * set it unconditionally as it is ignored for STACK_DYNPTR anyway. | |
2200 | */ | |
2201 | __mark_reg_known_zero(reg); | |
2202 | reg->type = CONST_PTR_TO_DYNPTR; | |
f8064ab9 KKD |
2203 | /* Give each dynptr a unique id to uniquely associate slices to it. */ |
2204 | reg->id = dynptr_id; | |
27060531 KKD |
2205 | reg->dynptr.type = type; |
2206 | reg->dynptr.first_slot = first_slot; | |
2207 | } | |
2208 | ||
4ddb7416 DB |
2209 | static void mark_ptr_not_null_reg(struct bpf_reg_state *reg) |
2210 | { | |
c25b2ae1 | 2211 | if (base_type(reg->type) == PTR_TO_MAP_VALUE) { |
4ddb7416 DB |
2212 | const struct bpf_map *map = reg->map_ptr; |
2213 | ||
2214 | if (map->inner_map_meta) { | |
2215 | reg->type = CONST_PTR_TO_MAP; | |
2216 | reg->map_ptr = map->inner_map_meta; | |
3e8ce298 AS |
2217 | /* transfer reg's id which is unique for every map_lookup_elem |
2218 | * as UID of the inner map. | |
2219 | */ | |
db559117 | 2220 | if (btf_record_has_field(map->inner_map_meta->record, BPF_TIMER)) |
34d11a44 | 2221 | reg->map_uid = reg->id; |
4ddb7416 DB |
2222 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { |
2223 | reg->type = PTR_TO_XDP_SOCK; | |
2224 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || | |
2225 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
2226 | reg->type = PTR_TO_SOCKET; | |
2227 | } else { | |
2228 | reg->type = PTR_TO_MAP_VALUE; | |
2229 | } | |
c25b2ae1 | 2230 | return; |
4ddb7416 | 2231 | } |
c25b2ae1 HL |
2232 | |
2233 | reg->type &= ~PTR_MAYBE_NULL; | |
4ddb7416 DB |
2234 | } |
2235 | ||
5d92ddc3 DM |
2236 | static void mark_reg_graph_node(struct bpf_reg_state *regs, u32 regno, |
2237 | struct btf_field_graph_root *ds_head) | |
2238 | { | |
2239 | __mark_reg_known_zero(®s[regno]); | |
2240 | regs[regno].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
2241 | regs[regno].btf = ds_head->btf; | |
2242 | regs[regno].btf_id = ds_head->value_btf_id; | |
2243 | regs[regno].off = ds_head->node_offset; | |
2244 | } | |
2245 | ||
de8f3a83 DB |
2246 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
2247 | { | |
2248 | return type_is_pkt_pointer(reg->type); | |
2249 | } | |
2250 | ||
2251 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
2252 | { | |
2253 | return reg_is_pkt_pointer(reg) || | |
2254 | reg->type == PTR_TO_PACKET_END; | |
2255 | } | |
2256 | ||
66e3a13e JK |
2257 | static bool reg_is_dynptr_slice_pkt(const struct bpf_reg_state *reg) |
2258 | { | |
2259 | return base_type(reg->type) == PTR_TO_MEM && | |
2260 | (reg->type & DYNPTR_TYPE_SKB || reg->type & DYNPTR_TYPE_XDP); | |
2261 | } | |
2262 | ||
de8f3a83 DB |
2263 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ |
2264 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
2265 | enum bpf_reg_type which) | |
2266 | { | |
2267 | /* The register can already have a range from prior markings. | |
2268 | * This is fine as long as it hasn't been advanced from its | |
2269 | * origin. | |
2270 | */ | |
2271 | return reg->type == which && | |
2272 | reg->id == 0 && | |
2273 | reg->off == 0 && | |
2274 | tnum_equals_const(reg->var_off, 0); | |
2275 | } | |
2276 | ||
3f50f132 JF |
2277 | /* Reset the min/max bounds of a register */ |
2278 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
2279 | { | |
2280 | reg->smin_value = S64_MIN; | |
2281 | reg->smax_value = S64_MAX; | |
2282 | reg->umin_value = 0; | |
2283 | reg->umax_value = U64_MAX; | |
2284 | ||
2285 | reg->s32_min_value = S32_MIN; | |
2286 | reg->s32_max_value = S32_MAX; | |
2287 | reg->u32_min_value = 0; | |
2288 | reg->u32_max_value = U32_MAX; | |
2289 | } | |
2290 | ||
2291 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
2292 | { | |
2293 | reg->smin_value = S64_MIN; | |
2294 | reg->smax_value = S64_MAX; | |
2295 | reg->umin_value = 0; | |
2296 | reg->umax_value = U64_MAX; | |
2297 | } | |
2298 | ||
2299 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
2300 | { | |
2301 | reg->s32_min_value = S32_MIN; | |
2302 | reg->s32_max_value = S32_MAX; | |
2303 | reg->u32_min_value = 0; | |
2304 | reg->u32_max_value = U32_MAX; | |
2305 | } | |
2306 | ||
2307 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
2308 | { | |
2309 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
2310 | ||
2311 | /* min signed is max(sign bit) | min(other bits) */ | |
2312 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
2313 | var32_off.value | (var32_off.mask & S32_MIN)); | |
2314 | /* max signed is min(sign bit) | max(other bits) */ | |
2315 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
2316 | var32_off.value | (var32_off.mask & S32_MAX)); | |
2317 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
2318 | reg->u32_max_value = min(reg->u32_max_value, | |
2319 | (u32)(var32_off.value | var32_off.mask)); | |
2320 | } | |
2321 | ||
2322 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
2323 | { |
2324 | /* min signed is max(sign bit) | min(other bits) */ | |
2325 | reg->smin_value = max_t(s64, reg->smin_value, | |
2326 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
2327 | /* max signed is min(sign bit) | max(other bits) */ | |
2328 | reg->smax_value = min_t(s64, reg->smax_value, | |
2329 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
2330 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
2331 | reg->umax_value = min(reg->umax_value, | |
2332 | reg->var_off.value | reg->var_off.mask); | |
2333 | } | |
2334 | ||
3f50f132 JF |
2335 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
2336 | { | |
2337 | __update_reg32_bounds(reg); | |
2338 | __update_reg64_bounds(reg); | |
2339 | } | |
2340 | ||
b03c9f9f | 2341 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
2342 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
2343 | { | |
2344 | /* Learn sign from signed bounds. | |
2345 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
2346 | * are the same, so combine. This works even in the negative case, e.g. | |
2347 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
2348 | */ | |
2349 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
2350 | reg->s32_min_value = reg->u32_min_value = | |
2351 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
2352 | reg->s32_max_value = reg->u32_max_value = | |
2353 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
2354 | return; | |
2355 | } | |
2356 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
2357 | * boundary, so we must be careful. | |
2358 | */ | |
2359 | if ((s32)reg->u32_max_value >= 0) { | |
2360 | /* Positive. We can't learn anything from the smin, but smax | |
2361 | * is positive, hence safe. | |
2362 | */ | |
2363 | reg->s32_min_value = reg->u32_min_value; | |
2364 | reg->s32_max_value = reg->u32_max_value = | |
2365 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
2366 | } else if ((s32)reg->u32_min_value < 0) { | |
2367 | /* Negative. We can't learn anything from the smax, but smin | |
2368 | * is negative, hence safe. | |
2369 | */ | |
2370 | reg->s32_min_value = reg->u32_min_value = | |
2371 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
2372 | reg->s32_max_value = reg->u32_max_value; | |
2373 | } | |
2374 | } | |
2375 | ||
2376 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
2377 | { |
2378 | /* Learn sign from signed bounds. | |
2379 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
2380 | * are the same, so combine. This works even in the negative case, e.g. | |
2381 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
2382 | */ | |
2383 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
2384 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
2385 | reg->umin_value); | |
2386 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
2387 | reg->umax_value); | |
2388 | return; | |
2389 | } | |
2390 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
2391 | * boundary, so we must be careful. | |
2392 | */ | |
2393 | if ((s64)reg->umax_value >= 0) { | |
2394 | /* Positive. We can't learn anything from the smin, but smax | |
2395 | * is positive, hence safe. | |
2396 | */ | |
2397 | reg->smin_value = reg->umin_value; | |
2398 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
2399 | reg->umax_value); | |
2400 | } else if ((s64)reg->umin_value < 0) { | |
2401 | /* Negative. We can't learn anything from the smax, but smin | |
2402 | * is negative, hence safe. | |
2403 | */ | |
2404 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
2405 | reg->umin_value); | |
2406 | reg->smax_value = reg->umax_value; | |
2407 | } | |
2408 | } | |
2409 | ||
3f50f132 JF |
2410 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
2411 | { | |
2412 | __reg32_deduce_bounds(reg); | |
2413 | __reg64_deduce_bounds(reg); | |
2414 | } | |
2415 | ||
b03c9f9f EC |
2416 | /* Attempts to improve var_off based on unsigned min/max information */ |
2417 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
2418 | { | |
3f50f132 JF |
2419 | struct tnum var64_off = tnum_intersect(reg->var_off, |
2420 | tnum_range(reg->umin_value, | |
2421 | reg->umax_value)); | |
7be14c1c DB |
2422 | struct tnum var32_off = tnum_intersect(tnum_subreg(var64_off), |
2423 | tnum_range(reg->u32_min_value, | |
2424 | reg->u32_max_value)); | |
3f50f132 JF |
2425 | |
2426 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
2427 | } |
2428 | ||
3844d153 DB |
2429 | static void reg_bounds_sync(struct bpf_reg_state *reg) |
2430 | { | |
2431 | /* We might have learned new bounds from the var_off. */ | |
2432 | __update_reg_bounds(reg); | |
2433 | /* We might have learned something about the sign bit. */ | |
2434 | __reg_deduce_bounds(reg); | |
2435 | /* We might have learned some bits from the bounds. */ | |
2436 | __reg_bound_offset(reg); | |
2437 | /* Intersecting with the old var_off might have improved our bounds | |
2438 | * slightly, e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2439 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2440 | */ | |
2441 | __update_reg_bounds(reg); | |
2442 | } | |
2443 | ||
e572ff80 DB |
2444 | static bool __reg32_bound_s64(s32 a) |
2445 | { | |
2446 | return a >= 0 && a <= S32_MAX; | |
2447 | } | |
2448 | ||
3f50f132 | 2449 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 2450 | { |
3f50f132 JF |
2451 | reg->umin_value = reg->u32_min_value; |
2452 | reg->umax_value = reg->u32_max_value; | |
e572ff80 DB |
2453 | |
2454 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds but must | |
2455 | * be positive otherwise set to worse case bounds and refine later | |
2456 | * from tnum. | |
3f50f132 | 2457 | */ |
e572ff80 DB |
2458 | if (__reg32_bound_s64(reg->s32_min_value) && |
2459 | __reg32_bound_s64(reg->s32_max_value)) { | |
3a71dc36 | 2460 | reg->smin_value = reg->s32_min_value; |
e572ff80 DB |
2461 | reg->smax_value = reg->s32_max_value; |
2462 | } else { | |
3a71dc36 | 2463 | reg->smin_value = 0; |
e572ff80 DB |
2464 | reg->smax_value = U32_MAX; |
2465 | } | |
3f50f132 JF |
2466 | } |
2467 | ||
2468 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
2469 | { | |
2470 | /* special case when 64-bit register has upper 32-bit register | |
2471 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
2472 | * allowing us to use 32-bit bounds directly, | |
2473 | */ | |
2474 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
2475 | __reg_assign_32_into_64(reg); | |
2476 | } else { | |
2477 | /* Otherwise the best we can do is push lower 32bit known and | |
2478 | * unknown bits into register (var_off set from jmp logic) | |
2479 | * then learn as much as possible from the 64-bit tnum | |
2480 | * known and unknown bits. The previous smin/smax bounds are | |
2481 | * invalid here because of jmp32 compare so mark them unknown | |
2482 | * so they do not impact tnum bounds calculation. | |
2483 | */ | |
2484 | __mark_reg64_unbounded(reg); | |
3f50f132 | 2485 | } |
3844d153 | 2486 | reg_bounds_sync(reg); |
3f50f132 JF |
2487 | } |
2488 | ||
2489 | static bool __reg64_bound_s32(s64 a) | |
2490 | { | |
388e2c0b | 2491 | return a >= S32_MIN && a <= S32_MAX; |
3f50f132 JF |
2492 | } |
2493 | ||
2494 | static bool __reg64_bound_u32(u64 a) | |
2495 | { | |
b9979db8 | 2496 | return a >= U32_MIN && a <= U32_MAX; |
3f50f132 JF |
2497 | } |
2498 | ||
2499 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
2500 | { | |
2501 | __mark_reg32_unbounded(reg); | |
b0270958 | 2502 | if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) { |
3f50f132 | 2503 | reg->s32_min_value = (s32)reg->smin_value; |
3f50f132 | 2504 | reg->s32_max_value = (s32)reg->smax_value; |
b0270958 | 2505 | } |
10bf4e83 | 2506 | if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value)) { |
3f50f132 | 2507 | reg->u32_min_value = (u32)reg->umin_value; |
3f50f132 | 2508 | reg->u32_max_value = (u32)reg->umax_value; |
10bf4e83 | 2509 | } |
3844d153 | 2510 | reg_bounds_sync(reg); |
b03c9f9f EC |
2511 | } |
2512 | ||
f1174f77 | 2513 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
2514 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
2515 | struct bpf_reg_state *reg) | |
f1174f77 | 2516 | { |
a9c676bc | 2517 | /* |
a73bf9f2 | 2518 | * Clear type, off, and union(map_ptr, range) and |
a9c676bc AS |
2519 | * padding between 'type' and union |
2520 | */ | |
2521 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 2522 | reg->type = SCALAR_VALUE; |
a73bf9f2 AN |
2523 | reg->id = 0; |
2524 | reg->ref_obj_id = 0; | |
f1174f77 | 2525 | reg->var_off = tnum_unknown; |
f4d7e40a | 2526 | reg->frameno = 0; |
be2ef816 | 2527 | reg->precise = !env->bpf_capable; |
b03c9f9f | 2528 | __mark_reg_unbounded(reg); |
f1174f77 EC |
2529 | } |
2530 | ||
61bd5218 JK |
2531 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
2532 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
2533 | { |
2534 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 2535 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
2536 | /* Something bad happened, let's kill all regs except FP */ |
2537 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 2538 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
2539 | return; |
2540 | } | |
f54c7898 | 2541 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
2542 | } |
2543 | ||
f54c7898 DB |
2544 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
2545 | struct bpf_reg_state *reg) | |
f1174f77 | 2546 | { |
f54c7898 | 2547 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
2548 | reg->type = NOT_INIT; |
2549 | } | |
2550 | ||
61bd5218 JK |
2551 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
2552 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
2553 | { |
2554 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 2555 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
2556 | /* Something bad happened, let's kill all regs except FP */ |
2557 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 2558 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
2559 | return; |
2560 | } | |
f54c7898 | 2561 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
2562 | } |
2563 | ||
41c48f3a AI |
2564 | static void mark_btf_ld_reg(struct bpf_verifier_env *env, |
2565 | struct bpf_reg_state *regs, u32 regno, | |
22dc4a0f | 2566 | enum bpf_reg_type reg_type, |
c6f1bfe8 YS |
2567 | struct btf *btf, u32 btf_id, |
2568 | enum bpf_type_flag flag) | |
41c48f3a AI |
2569 | { |
2570 | if (reg_type == SCALAR_VALUE) { | |
2571 | mark_reg_unknown(env, regs, regno); | |
2572 | return; | |
2573 | } | |
2574 | mark_reg_known_zero(env, regs, regno); | |
c6f1bfe8 | 2575 | regs[regno].type = PTR_TO_BTF_ID | flag; |
22dc4a0f | 2576 | regs[regno].btf = btf; |
41c48f3a AI |
2577 | regs[regno].btf_id = btf_id; |
2578 | } | |
2579 | ||
5327ed3d | 2580 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 2581 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 2582 | struct bpf_func_state *state) |
17a52670 | 2583 | { |
f4d7e40a | 2584 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
2585 | int i; |
2586 | ||
dc503a8a | 2587 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 2588 | mark_reg_not_init(env, regs, i); |
dc503a8a | 2589 | regs[i].live = REG_LIVE_NONE; |
679c782d | 2590 | regs[i].parent = NULL; |
5327ed3d | 2591 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 2592 | } |
17a52670 AS |
2593 | |
2594 | /* frame pointer */ | |
f1174f77 | 2595 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 2596 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 2597 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
2598 | } |
2599 | ||
f4d7e40a AS |
2600 | #define BPF_MAIN_FUNC (-1) |
2601 | static void init_func_state(struct bpf_verifier_env *env, | |
2602 | struct bpf_func_state *state, | |
2603 | int callsite, int frameno, int subprogno) | |
2604 | { | |
2605 | state->callsite = callsite; | |
2606 | state->frameno = frameno; | |
2607 | state->subprogno = subprogno; | |
1bfe26fb | 2608 | state->callback_ret_range = tnum_range(0, 0); |
f4d7e40a | 2609 | init_reg_state(env, state); |
0f55f9ed | 2610 | mark_verifier_state_scratched(env); |
f4d7e40a AS |
2611 | } |
2612 | ||
bfc6bb74 AS |
2613 | /* Similar to push_stack(), but for async callbacks */ |
2614 | static struct bpf_verifier_state *push_async_cb(struct bpf_verifier_env *env, | |
2615 | int insn_idx, int prev_insn_idx, | |
2616 | int subprog) | |
2617 | { | |
2618 | struct bpf_verifier_stack_elem *elem; | |
2619 | struct bpf_func_state *frame; | |
2620 | ||
2621 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); | |
2622 | if (!elem) | |
2623 | goto err; | |
2624 | ||
2625 | elem->insn_idx = insn_idx; | |
2626 | elem->prev_insn_idx = prev_insn_idx; | |
2627 | elem->next = env->head; | |
12166409 | 2628 | elem->log_pos = env->log.end_pos; |
bfc6bb74 AS |
2629 | env->head = elem; |
2630 | env->stack_size++; | |
2631 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { | |
2632 | verbose(env, | |
2633 | "The sequence of %d jumps is too complex for async cb.\n", | |
2634 | env->stack_size); | |
2635 | goto err; | |
2636 | } | |
2637 | /* Unlike push_stack() do not copy_verifier_state(). | |
2638 | * The caller state doesn't matter. | |
2639 | * This is async callback. It starts in a fresh stack. | |
2640 | * Initialize it similar to do_check_common(). | |
2641 | */ | |
2642 | elem->st.branches = 1; | |
2643 | frame = kzalloc(sizeof(*frame), GFP_KERNEL); | |
2644 | if (!frame) | |
2645 | goto err; | |
2646 | init_func_state(env, frame, | |
2647 | BPF_MAIN_FUNC /* callsite */, | |
2648 | 0 /* frameno within this callchain */, | |
2649 | subprog /* subprog number within this prog */); | |
2650 | elem->st.frame[0] = frame; | |
2651 | return &elem->st; | |
2652 | err: | |
2653 | free_verifier_state(env->cur_state, true); | |
2654 | env->cur_state = NULL; | |
2655 | /* pop all elements and return */ | |
2656 | while (!pop_stack(env, NULL, NULL, false)); | |
2657 | return NULL; | |
2658 | } | |
2659 | ||
2660 | ||
17a52670 AS |
2661 | enum reg_arg_type { |
2662 | SRC_OP, /* register is used as source operand */ | |
2663 | DST_OP, /* register is used as destination operand */ | |
2664 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
2665 | }; | |
2666 | ||
cc8b0b92 AS |
2667 | static int cmp_subprogs(const void *a, const void *b) |
2668 | { | |
9c8105bd JW |
2669 | return ((struct bpf_subprog_info *)a)->start - |
2670 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
2671 | } |
2672 | ||
2673 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
2674 | { | |
9c8105bd | 2675 | struct bpf_subprog_info *p; |
cc8b0b92 | 2676 | |
9c8105bd JW |
2677 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
2678 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
2679 | if (!p) |
2680 | return -ENOENT; | |
9c8105bd | 2681 | return p - env->subprog_info; |
cc8b0b92 AS |
2682 | |
2683 | } | |
2684 | ||
2685 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
2686 | { | |
2687 | int insn_cnt = env->prog->len; | |
2688 | int ret; | |
2689 | ||
2690 | if (off >= insn_cnt || off < 0) { | |
2691 | verbose(env, "call to invalid destination\n"); | |
2692 | return -EINVAL; | |
2693 | } | |
2694 | ret = find_subprog(env, off); | |
2695 | if (ret >= 0) | |
282a0f46 | 2696 | return ret; |
4cb3d99c | 2697 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
2698 | verbose(env, "too many subprograms\n"); |
2699 | return -E2BIG; | |
2700 | } | |
e6ac2450 | 2701 | /* determine subprog starts. The end is one before the next starts */ |
9c8105bd JW |
2702 | env->subprog_info[env->subprog_cnt++].start = off; |
2703 | sort(env->subprog_info, env->subprog_cnt, | |
2704 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
282a0f46 | 2705 | return env->subprog_cnt - 1; |
cc8b0b92 AS |
2706 | } |
2707 | ||
b9ae0c9d KKD |
2708 | static int bpf_find_exception_callback_insn_off(struct bpf_verifier_env *env) |
2709 | { | |
2710 | struct bpf_prog_aux *aux = env->prog->aux; | |
2711 | struct btf *btf = aux->btf; | |
2712 | const struct btf_type *t; | |
2713 | u32 main_btf_id, id; | |
2714 | const char *name; | |
2715 | int ret, i; | |
2716 | ||
2717 | /* Non-zero func_info_cnt implies valid btf */ | |
2718 | if (!aux->func_info_cnt) | |
2719 | return 0; | |
2720 | main_btf_id = aux->func_info[0].type_id; | |
2721 | ||
2722 | t = btf_type_by_id(btf, main_btf_id); | |
2723 | if (!t) { | |
2724 | verbose(env, "invalid btf id for main subprog in func_info\n"); | |
2725 | return -EINVAL; | |
2726 | } | |
2727 | ||
2728 | name = btf_find_decl_tag_value(btf, t, -1, "exception_callback:"); | |
2729 | if (IS_ERR(name)) { | |
2730 | ret = PTR_ERR(name); | |
2731 | /* If there is no tag present, there is no exception callback */ | |
2732 | if (ret == -ENOENT) | |
2733 | ret = 0; | |
2734 | else if (ret == -EEXIST) | |
2735 | verbose(env, "multiple exception callback tags for main subprog\n"); | |
2736 | return ret; | |
2737 | } | |
2738 | ||
2739 | ret = btf_find_by_name_kind(btf, name, BTF_KIND_FUNC); | |
2740 | if (ret < 0) { | |
2741 | verbose(env, "exception callback '%s' could not be found in BTF\n", name); | |
2742 | return ret; | |
2743 | } | |
2744 | id = ret; | |
2745 | t = btf_type_by_id(btf, id); | |
2746 | if (btf_func_linkage(t) != BTF_FUNC_GLOBAL) { | |
2747 | verbose(env, "exception callback '%s' must have global linkage\n", name); | |
2748 | return -EINVAL; | |
2749 | } | |
2750 | ret = 0; | |
2751 | for (i = 0; i < aux->func_info_cnt; i++) { | |
2752 | if (aux->func_info[i].type_id != id) | |
2753 | continue; | |
2754 | ret = aux->func_info[i].insn_off; | |
2755 | /* Further func_info and subprog checks will also happen | |
2756 | * later, so assume this is the right insn_off for now. | |
2757 | */ | |
2758 | if (!ret) { | |
2759 | verbose(env, "invalid exception callback insn_off in func_info: 0\n"); | |
2760 | ret = -EINVAL; | |
2761 | } | |
2762 | } | |
2763 | if (!ret) { | |
2764 | verbose(env, "exception callback type id not found in func_info\n"); | |
2765 | ret = -EINVAL; | |
2766 | } | |
2767 | return ret; | |
2768 | } | |
2769 | ||
2357672c KKD |
2770 | #define MAX_KFUNC_DESCS 256 |
2771 | #define MAX_KFUNC_BTFS 256 | |
2772 | ||
e6ac2450 MKL |
2773 | struct bpf_kfunc_desc { |
2774 | struct btf_func_model func_model; | |
2775 | u32 func_id; | |
2776 | s32 imm; | |
2357672c | 2777 | u16 offset; |
1cf3bfc6 | 2778 | unsigned long addr; |
2357672c KKD |
2779 | }; |
2780 | ||
2781 | struct bpf_kfunc_btf { | |
2782 | struct btf *btf; | |
2783 | struct module *module; | |
2784 | u16 offset; | |
e6ac2450 MKL |
2785 | }; |
2786 | ||
e6ac2450 | 2787 | struct bpf_kfunc_desc_tab { |
1cf3bfc6 IL |
2788 | /* Sorted by func_id (BTF ID) and offset (fd_array offset) during |
2789 | * verification. JITs do lookups by bpf_insn, where func_id may not be | |
2790 | * available, therefore at the end of verification do_misc_fixups() | |
2791 | * sorts this by imm and offset. | |
2792 | */ | |
e6ac2450 MKL |
2793 | struct bpf_kfunc_desc descs[MAX_KFUNC_DESCS]; |
2794 | u32 nr_descs; | |
2795 | }; | |
2796 | ||
2357672c KKD |
2797 | struct bpf_kfunc_btf_tab { |
2798 | struct bpf_kfunc_btf descs[MAX_KFUNC_BTFS]; | |
2799 | u32 nr_descs; | |
2800 | }; | |
2801 | ||
2802 | static int kfunc_desc_cmp_by_id_off(const void *a, const void *b) | |
e6ac2450 MKL |
2803 | { |
2804 | const struct bpf_kfunc_desc *d0 = a; | |
2805 | const struct bpf_kfunc_desc *d1 = b; | |
2806 | ||
2807 | /* func_id is not greater than BTF_MAX_TYPE */ | |
2357672c KKD |
2808 | return d0->func_id - d1->func_id ?: d0->offset - d1->offset; |
2809 | } | |
2810 | ||
2811 | static int kfunc_btf_cmp_by_off(const void *a, const void *b) | |
2812 | { | |
2813 | const struct bpf_kfunc_btf *d0 = a; | |
2814 | const struct bpf_kfunc_btf *d1 = b; | |
2815 | ||
2816 | return d0->offset - d1->offset; | |
e6ac2450 MKL |
2817 | } |
2818 | ||
2819 | static const struct bpf_kfunc_desc * | |
2357672c | 2820 | find_kfunc_desc(const struct bpf_prog *prog, u32 func_id, u16 offset) |
e6ac2450 MKL |
2821 | { |
2822 | struct bpf_kfunc_desc desc = { | |
2823 | .func_id = func_id, | |
2357672c | 2824 | .offset = offset, |
e6ac2450 MKL |
2825 | }; |
2826 | struct bpf_kfunc_desc_tab *tab; | |
2827 | ||
2828 | tab = prog->aux->kfunc_tab; | |
2829 | return bsearch(&desc, tab->descs, tab->nr_descs, | |
2357672c KKD |
2830 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_id_off); |
2831 | } | |
2832 | ||
1cf3bfc6 IL |
2833 | int bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id, |
2834 | u16 btf_fd_idx, u8 **func_addr) | |
2835 | { | |
2836 | const struct bpf_kfunc_desc *desc; | |
2837 | ||
2838 | desc = find_kfunc_desc(prog, func_id, btf_fd_idx); | |
2839 | if (!desc) | |
2840 | return -EFAULT; | |
2841 | ||
2842 | *func_addr = (u8 *)desc->addr; | |
2843 | return 0; | |
2844 | } | |
2845 | ||
2357672c | 2846 | static struct btf *__find_kfunc_desc_btf(struct bpf_verifier_env *env, |
b202d844 | 2847 | s16 offset) |
2357672c KKD |
2848 | { |
2849 | struct bpf_kfunc_btf kf_btf = { .offset = offset }; | |
2850 | struct bpf_kfunc_btf_tab *tab; | |
2851 | struct bpf_kfunc_btf *b; | |
2852 | struct module *mod; | |
2853 | struct btf *btf; | |
2854 | int btf_fd; | |
2855 | ||
2856 | tab = env->prog->aux->kfunc_btf_tab; | |
2857 | b = bsearch(&kf_btf, tab->descs, tab->nr_descs, | |
2858 | sizeof(tab->descs[0]), kfunc_btf_cmp_by_off); | |
2859 | if (!b) { | |
2860 | if (tab->nr_descs == MAX_KFUNC_BTFS) { | |
2861 | verbose(env, "too many different module BTFs\n"); | |
2862 | return ERR_PTR(-E2BIG); | |
2863 | } | |
2864 | ||
2865 | if (bpfptr_is_null(env->fd_array)) { | |
2866 | verbose(env, "kfunc offset > 0 without fd_array is invalid\n"); | |
2867 | return ERR_PTR(-EPROTO); | |
2868 | } | |
2869 | ||
2870 | if (copy_from_bpfptr_offset(&btf_fd, env->fd_array, | |
2871 | offset * sizeof(btf_fd), | |
2872 | sizeof(btf_fd))) | |
2873 | return ERR_PTR(-EFAULT); | |
2874 | ||
2875 | btf = btf_get_by_fd(btf_fd); | |
588cd7ef KKD |
2876 | if (IS_ERR(btf)) { |
2877 | verbose(env, "invalid module BTF fd specified\n"); | |
2357672c | 2878 | return btf; |
588cd7ef | 2879 | } |
2357672c KKD |
2880 | |
2881 | if (!btf_is_module(btf)) { | |
2882 | verbose(env, "BTF fd for kfunc is not a module BTF\n"); | |
2883 | btf_put(btf); | |
2884 | return ERR_PTR(-EINVAL); | |
2885 | } | |
2886 | ||
2887 | mod = btf_try_get_module(btf); | |
2888 | if (!mod) { | |
2889 | btf_put(btf); | |
2890 | return ERR_PTR(-ENXIO); | |
2891 | } | |
2892 | ||
2893 | b = &tab->descs[tab->nr_descs++]; | |
2894 | b->btf = btf; | |
2895 | b->module = mod; | |
2896 | b->offset = offset; | |
2897 | ||
2898 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2899 | kfunc_btf_cmp_by_off, NULL); | |
2900 | } | |
2357672c | 2901 | return b->btf; |
e6ac2450 MKL |
2902 | } |
2903 | ||
2357672c KKD |
2904 | void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab) |
2905 | { | |
2906 | if (!tab) | |
2907 | return; | |
2908 | ||
2909 | while (tab->nr_descs--) { | |
2910 | module_put(tab->descs[tab->nr_descs].module); | |
2911 | btf_put(tab->descs[tab->nr_descs].btf); | |
2912 | } | |
2913 | kfree(tab); | |
2914 | } | |
2915 | ||
43bf0878 | 2916 | static struct btf *find_kfunc_desc_btf(struct bpf_verifier_env *env, s16 offset) |
2357672c | 2917 | { |
2357672c KKD |
2918 | if (offset) { |
2919 | if (offset < 0) { | |
2920 | /* In the future, this can be allowed to increase limit | |
2921 | * of fd index into fd_array, interpreted as u16. | |
2922 | */ | |
2923 | verbose(env, "negative offset disallowed for kernel module function call\n"); | |
2924 | return ERR_PTR(-EINVAL); | |
2925 | } | |
2926 | ||
b202d844 | 2927 | return __find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2928 | } |
2929 | return btf_vmlinux ?: ERR_PTR(-ENOENT); | |
e6ac2450 MKL |
2930 | } |
2931 | ||
2357672c | 2932 | static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset) |
e6ac2450 MKL |
2933 | { |
2934 | const struct btf_type *func, *func_proto; | |
2357672c | 2935 | struct bpf_kfunc_btf_tab *btf_tab; |
e6ac2450 MKL |
2936 | struct bpf_kfunc_desc_tab *tab; |
2937 | struct bpf_prog_aux *prog_aux; | |
2938 | struct bpf_kfunc_desc *desc; | |
2939 | const char *func_name; | |
2357672c | 2940 | struct btf *desc_btf; |
8cbf062a | 2941 | unsigned long call_imm; |
e6ac2450 MKL |
2942 | unsigned long addr; |
2943 | int err; | |
2944 | ||
2945 | prog_aux = env->prog->aux; | |
2946 | tab = prog_aux->kfunc_tab; | |
2357672c | 2947 | btf_tab = prog_aux->kfunc_btf_tab; |
e6ac2450 MKL |
2948 | if (!tab) { |
2949 | if (!btf_vmlinux) { | |
2950 | verbose(env, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
2951 | return -ENOTSUPP; | |
2952 | } | |
2953 | ||
2954 | if (!env->prog->jit_requested) { | |
2955 | verbose(env, "JIT is required for calling kernel function\n"); | |
2956 | return -ENOTSUPP; | |
2957 | } | |
2958 | ||
2959 | if (!bpf_jit_supports_kfunc_call()) { | |
2960 | verbose(env, "JIT does not support calling kernel function\n"); | |
2961 | return -ENOTSUPP; | |
2962 | } | |
2963 | ||
2964 | if (!env->prog->gpl_compatible) { | |
2965 | verbose(env, "cannot call kernel function from non-GPL compatible program\n"); | |
2966 | return -EINVAL; | |
2967 | } | |
2968 | ||
2969 | tab = kzalloc(sizeof(*tab), GFP_KERNEL); | |
2970 | if (!tab) | |
2971 | return -ENOMEM; | |
2972 | prog_aux->kfunc_tab = tab; | |
2973 | } | |
2974 | ||
a5d82727 KKD |
2975 | /* func_id == 0 is always invalid, but instead of returning an error, be |
2976 | * conservative and wait until the code elimination pass before returning | |
2977 | * error, so that invalid calls that get pruned out can be in BPF programs | |
2978 | * loaded from userspace. It is also required that offset be untouched | |
2979 | * for such calls. | |
2980 | */ | |
2981 | if (!func_id && !offset) | |
2982 | return 0; | |
2983 | ||
2357672c KKD |
2984 | if (!btf_tab && offset) { |
2985 | btf_tab = kzalloc(sizeof(*btf_tab), GFP_KERNEL); | |
2986 | if (!btf_tab) | |
2987 | return -ENOMEM; | |
2988 | prog_aux->kfunc_btf_tab = btf_tab; | |
2989 | } | |
2990 | ||
43bf0878 | 2991 | desc_btf = find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2992 | if (IS_ERR(desc_btf)) { |
2993 | verbose(env, "failed to find BTF for kernel function\n"); | |
2994 | return PTR_ERR(desc_btf); | |
2995 | } | |
2996 | ||
2997 | if (find_kfunc_desc(env->prog, func_id, offset)) | |
e6ac2450 MKL |
2998 | return 0; |
2999 | ||
3000 | if (tab->nr_descs == MAX_KFUNC_DESCS) { | |
3001 | verbose(env, "too many different kernel function calls\n"); | |
3002 | return -E2BIG; | |
3003 | } | |
3004 | ||
2357672c | 3005 | func = btf_type_by_id(desc_btf, func_id); |
e6ac2450 MKL |
3006 | if (!func || !btf_type_is_func(func)) { |
3007 | verbose(env, "kernel btf_id %u is not a function\n", | |
3008 | func_id); | |
3009 | return -EINVAL; | |
3010 | } | |
2357672c | 3011 | func_proto = btf_type_by_id(desc_btf, func->type); |
e6ac2450 MKL |
3012 | if (!func_proto || !btf_type_is_func_proto(func_proto)) { |
3013 | verbose(env, "kernel function btf_id %u does not have a valid func_proto\n", | |
3014 | func_id); | |
3015 | return -EINVAL; | |
3016 | } | |
3017 | ||
2357672c | 3018 | func_name = btf_name_by_offset(desc_btf, func->name_off); |
e6ac2450 MKL |
3019 | addr = kallsyms_lookup_name(func_name); |
3020 | if (!addr) { | |
3021 | verbose(env, "cannot find address for kernel function %s\n", | |
3022 | func_name); | |
3023 | return -EINVAL; | |
3024 | } | |
1cf3bfc6 | 3025 | specialize_kfunc(env, func_id, offset, &addr); |
e6ac2450 | 3026 | |
1cf3bfc6 IL |
3027 | if (bpf_jit_supports_far_kfunc_call()) { |
3028 | call_imm = func_id; | |
3029 | } else { | |
3030 | call_imm = BPF_CALL_IMM(addr); | |
3031 | /* Check whether the relative offset overflows desc->imm */ | |
3032 | if ((unsigned long)(s32)call_imm != call_imm) { | |
3033 | verbose(env, "address of kernel function %s is out of range\n", | |
3034 | func_name); | |
3035 | return -EINVAL; | |
3036 | } | |
8cbf062a HT |
3037 | } |
3038 | ||
3d76a4d3 SF |
3039 | if (bpf_dev_bound_kfunc_id(func_id)) { |
3040 | err = bpf_dev_bound_kfunc_check(&env->log, prog_aux); | |
3041 | if (err) | |
3042 | return err; | |
3043 | } | |
3044 | ||
e6ac2450 MKL |
3045 | desc = &tab->descs[tab->nr_descs++]; |
3046 | desc->func_id = func_id; | |
8cbf062a | 3047 | desc->imm = call_imm; |
2357672c | 3048 | desc->offset = offset; |
1cf3bfc6 | 3049 | desc->addr = addr; |
2357672c | 3050 | err = btf_distill_func_proto(&env->log, desc_btf, |
e6ac2450 MKL |
3051 | func_proto, func_name, |
3052 | &desc->func_model); | |
3053 | if (!err) | |
3054 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2357672c | 3055 | kfunc_desc_cmp_by_id_off, NULL); |
e6ac2450 MKL |
3056 | return err; |
3057 | } | |
3058 | ||
1cf3bfc6 | 3059 | static int kfunc_desc_cmp_by_imm_off(const void *a, const void *b) |
e6ac2450 MKL |
3060 | { |
3061 | const struct bpf_kfunc_desc *d0 = a; | |
3062 | const struct bpf_kfunc_desc *d1 = b; | |
3063 | ||
1cf3bfc6 IL |
3064 | if (d0->imm != d1->imm) |
3065 | return d0->imm < d1->imm ? -1 : 1; | |
3066 | if (d0->offset != d1->offset) | |
3067 | return d0->offset < d1->offset ? -1 : 1; | |
e6ac2450 MKL |
3068 | return 0; |
3069 | } | |
3070 | ||
1cf3bfc6 | 3071 | static void sort_kfunc_descs_by_imm_off(struct bpf_prog *prog) |
e6ac2450 MKL |
3072 | { |
3073 | struct bpf_kfunc_desc_tab *tab; | |
3074 | ||
3075 | tab = prog->aux->kfunc_tab; | |
3076 | if (!tab) | |
3077 | return; | |
3078 | ||
3079 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
1cf3bfc6 | 3080 | kfunc_desc_cmp_by_imm_off, NULL); |
e6ac2450 MKL |
3081 | } |
3082 | ||
3083 | bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) | |
3084 | { | |
3085 | return !!prog->aux->kfunc_tab; | |
3086 | } | |
3087 | ||
3088 | const struct btf_func_model * | |
3089 | bpf_jit_find_kfunc_model(const struct bpf_prog *prog, | |
3090 | const struct bpf_insn *insn) | |
3091 | { | |
3092 | const struct bpf_kfunc_desc desc = { | |
3093 | .imm = insn->imm, | |
1cf3bfc6 | 3094 | .offset = insn->off, |
e6ac2450 MKL |
3095 | }; |
3096 | const struct bpf_kfunc_desc *res; | |
3097 | struct bpf_kfunc_desc_tab *tab; | |
3098 | ||
3099 | tab = prog->aux->kfunc_tab; | |
3100 | res = bsearch(&desc, tab->descs, tab->nr_descs, | |
1cf3bfc6 | 3101 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm_off); |
e6ac2450 MKL |
3102 | |
3103 | return res ? &res->func_model : NULL; | |
3104 | } | |
3105 | ||
3106 | static int add_subprog_and_kfunc(struct bpf_verifier_env *env) | |
cc8b0b92 | 3107 | { |
9c8105bd | 3108 | struct bpf_subprog_info *subprog = env->subprog_info; |
b9ae0c9d | 3109 | int i, ret, insn_cnt = env->prog->len, ex_cb_insn; |
cc8b0b92 | 3110 | struct bpf_insn *insn = env->prog->insnsi; |
cc8b0b92 | 3111 | |
f910cefa JW |
3112 | /* Add entry function. */ |
3113 | ret = add_subprog(env, 0); | |
e6ac2450 | 3114 | if (ret) |
f910cefa JW |
3115 | return ret; |
3116 | ||
e6ac2450 MKL |
3117 | for (i = 0; i < insn_cnt; i++, insn++) { |
3118 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn) && | |
3119 | !bpf_pseudo_kfunc_call(insn)) | |
cc8b0b92 | 3120 | continue; |
e6ac2450 | 3121 | |
2c78ee89 | 3122 | if (!env->bpf_capable) { |
e6ac2450 | 3123 | verbose(env, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); |
cc8b0b92 AS |
3124 | return -EPERM; |
3125 | } | |
e6ac2450 | 3126 | |
3990ed4c | 3127 | if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn)) |
e6ac2450 | 3128 | ret = add_subprog(env, i + insn->imm + 1); |
3990ed4c | 3129 | else |
2357672c | 3130 | ret = add_kfunc_call(env, insn->imm, insn->off); |
e6ac2450 | 3131 | |
cc8b0b92 AS |
3132 | if (ret < 0) |
3133 | return ret; | |
3134 | } | |
3135 | ||
b9ae0c9d KKD |
3136 | ret = bpf_find_exception_callback_insn_off(env); |
3137 | if (ret < 0) | |
3138 | return ret; | |
3139 | ex_cb_insn = ret; | |
3140 | ||
3141 | /* If ex_cb_insn > 0, this means that the main program has a subprog | |
3142 | * marked using BTF decl tag to serve as the exception callback. | |
3143 | */ | |
3144 | if (ex_cb_insn) { | |
3145 | ret = add_subprog(env, ex_cb_insn); | |
3146 | if (ret < 0) | |
3147 | return ret; | |
3148 | for (i = 1; i < env->subprog_cnt; i++) { | |
3149 | if (env->subprog_info[i].start != ex_cb_insn) | |
3150 | continue; | |
3151 | env->exception_callback_subprog = i; | |
3152 | break; | |
3153 | } | |
3154 | } | |
3155 | ||
4cb3d99c JW |
3156 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
3157 | * logic. 'subprog_cnt' should not be increased. | |
3158 | */ | |
3159 | subprog[env->subprog_cnt].start = insn_cnt; | |
3160 | ||
06ee7115 | 3161 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 3162 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 3163 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 | 3164 | |
e6ac2450 MKL |
3165 | return 0; |
3166 | } | |
3167 | ||
3168 | static int check_subprogs(struct bpf_verifier_env *env) | |
3169 | { | |
3170 | int i, subprog_start, subprog_end, off, cur_subprog = 0; | |
3171 | struct bpf_subprog_info *subprog = env->subprog_info; | |
3172 | struct bpf_insn *insn = env->prog->insnsi; | |
3173 | int insn_cnt = env->prog->len; | |
3174 | ||
cc8b0b92 | 3175 | /* now check that all jumps are within the same subprog */ |
4cb3d99c JW |
3176 | subprog_start = subprog[cur_subprog].start; |
3177 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
3178 | for (i = 0; i < insn_cnt; i++) { |
3179 | u8 code = insn[i].code; | |
3180 | ||
7f6e4312 | 3181 | if (code == (BPF_JMP | BPF_CALL) && |
df2ccc18 IL |
3182 | insn[i].src_reg == 0 && |
3183 | insn[i].imm == BPF_FUNC_tail_call) | |
7f6e4312 | 3184 | subprog[cur_subprog].has_tail_call = true; |
09b28d76 AS |
3185 | if (BPF_CLASS(code) == BPF_LD && |
3186 | (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND)) | |
3187 | subprog[cur_subprog].has_ld_abs = true; | |
092ed096 | 3188 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
3189 | goto next; |
3190 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
3191 | goto next; | |
4cd58e9a YS |
3192 | if (code == (BPF_JMP32 | BPF_JA)) |
3193 | off = i + insn[i].imm + 1; | |
3194 | else | |
3195 | off = i + insn[i].off + 1; | |
cc8b0b92 AS |
3196 | if (off < subprog_start || off >= subprog_end) { |
3197 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
3198 | return -EINVAL; | |
3199 | } | |
3200 | next: | |
3201 | if (i == subprog_end - 1) { | |
3202 | /* to avoid fall-through from one subprog into another | |
3203 | * the last insn of the subprog should be either exit | |
f18b03fa | 3204 | * or unconditional jump back or bpf_throw call |
cc8b0b92 AS |
3205 | */ |
3206 | if (code != (BPF_JMP | BPF_EXIT) && | |
4cd58e9a | 3207 | code != (BPF_JMP32 | BPF_JA) && |
cc8b0b92 AS |
3208 | code != (BPF_JMP | BPF_JA)) { |
3209 | verbose(env, "last insn is not an exit or jmp\n"); | |
3210 | return -EINVAL; | |
3211 | } | |
3212 | subprog_start = subprog_end; | |
4cb3d99c JW |
3213 | cur_subprog++; |
3214 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 3215 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
3216 | } |
3217 | } | |
3218 | return 0; | |
3219 | } | |
3220 | ||
679c782d EC |
3221 | /* Parentage chain of this register (or stack slot) should take care of all |
3222 | * issues like callee-saved registers, stack slot allocation time, etc. | |
3223 | */ | |
f4d7e40a | 3224 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 3225 | const struct bpf_reg_state *state, |
5327ed3d | 3226 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
3227 | { |
3228 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 3229 | int cnt = 0; |
dc503a8a EC |
3230 | |
3231 | while (parent) { | |
3232 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 3233 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 3234 | break; |
9242b5f5 AS |
3235 | if (parent->live & REG_LIVE_DONE) { |
3236 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
c25b2ae1 | 3237 | reg_type_str(env, parent->type), |
9242b5f5 AS |
3238 | parent->var_off.value, parent->off); |
3239 | return -EFAULT; | |
3240 | } | |
5327ed3d JW |
3241 | /* The first condition is more likely to be true than the |
3242 | * second, checked it first. | |
3243 | */ | |
3244 | if ((parent->live & REG_LIVE_READ) == flag || | |
3245 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
3246 | /* The parentage chain never changes and |
3247 | * this parent was already marked as LIVE_READ. | |
3248 | * There is no need to keep walking the chain again and | |
3249 | * keep re-marking all parents as LIVE_READ. | |
3250 | * This case happens when the same register is read | |
3251 | * multiple times without writes into it in-between. | |
5327ed3d JW |
3252 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
3253 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
3254 | */ |
3255 | break; | |
dc503a8a | 3256 | /* ... then we depend on parent's value */ |
5327ed3d JW |
3257 | parent->live |= flag; |
3258 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
3259 | if (flag == REG_LIVE_READ64) | |
3260 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
3261 | state = parent; |
3262 | parent = state->parent; | |
f4d7e40a | 3263 | writes = true; |
06ee7115 | 3264 | cnt++; |
dc503a8a | 3265 | } |
06ee7115 AS |
3266 | |
3267 | if (env->longest_mark_read_walk < cnt) | |
3268 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 3269 | return 0; |
dc503a8a EC |
3270 | } |
3271 | ||
d6fefa11 KKD |
3272 | static int mark_dynptr_read(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
3273 | { | |
3274 | struct bpf_func_state *state = func(env, reg); | |
3275 | int spi, ret; | |
3276 | ||
3277 | /* For CONST_PTR_TO_DYNPTR, it must have already been done by | |
3278 | * check_reg_arg in check_helper_call and mark_btf_func_reg_size in | |
3279 | * check_kfunc_call. | |
3280 | */ | |
3281 | if (reg->type == CONST_PTR_TO_DYNPTR) | |
3282 | return 0; | |
79168a66 KKD |
3283 | spi = dynptr_get_spi(env, reg); |
3284 | if (spi < 0) | |
3285 | return spi; | |
d6fefa11 KKD |
3286 | /* Caller ensures dynptr is valid and initialized, which means spi is in |
3287 | * bounds and spi is the first dynptr slot. Simply mark stack slot as | |
3288 | * read. | |
3289 | */ | |
3290 | ret = mark_reg_read(env, &state->stack[spi].spilled_ptr, | |
3291 | state->stack[spi].spilled_ptr.parent, REG_LIVE_READ64); | |
3292 | if (ret) | |
3293 | return ret; | |
3294 | return mark_reg_read(env, &state->stack[spi - 1].spilled_ptr, | |
3295 | state->stack[spi - 1].spilled_ptr.parent, REG_LIVE_READ64); | |
3296 | } | |
3297 | ||
06accc87 AN |
3298 | static int mark_iter_read(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
3299 | int spi, int nr_slots) | |
3300 | { | |
3301 | struct bpf_func_state *state = func(env, reg); | |
3302 | int err, i; | |
3303 | ||
3304 | for (i = 0; i < nr_slots; i++) { | |
3305 | struct bpf_reg_state *st = &state->stack[spi - i].spilled_ptr; | |
3306 | ||
3307 | err = mark_reg_read(env, st, st->parent, REG_LIVE_READ64); | |
3308 | if (err) | |
3309 | return err; | |
3310 | ||
3311 | mark_stack_slot_scratched(env, spi - i); | |
3312 | } | |
3313 | ||
3314 | return 0; | |
3315 | } | |
3316 | ||
5327ed3d JW |
3317 | /* This function is supposed to be used by the following 32-bit optimization |
3318 | * code only. It returns TRUE if the source or destination register operates | |
3319 | * on 64-bit, otherwise return FALSE. | |
3320 | */ | |
3321 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
3322 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
3323 | { | |
3324 | u8 code, class, op; | |
3325 | ||
3326 | code = insn->code; | |
3327 | class = BPF_CLASS(code); | |
3328 | op = BPF_OP(code); | |
3329 | if (class == BPF_JMP) { | |
3330 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
3331 | * conservatively. | |
3332 | */ | |
3333 | if (op == BPF_EXIT) | |
3334 | return true; | |
3335 | if (op == BPF_CALL) { | |
3336 | /* BPF to BPF call will reach here because of marking | |
3337 | * caller saved clobber with DST_OP_NO_MARK for which we | |
3338 | * don't care the register def because they are anyway | |
3339 | * marked as NOT_INIT already. | |
3340 | */ | |
3341 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
3342 | return false; | |
3343 | /* Helper call will reach here because of arg type | |
3344 | * check, conservatively return TRUE. | |
3345 | */ | |
3346 | if (t == SRC_OP) | |
3347 | return true; | |
3348 | ||
3349 | return false; | |
3350 | } | |
3351 | } | |
3352 | ||
0845c3db YS |
3353 | if (class == BPF_ALU64 && op == BPF_END && (insn->imm == 16 || insn->imm == 32)) |
3354 | return false; | |
3355 | ||
5327ed3d | 3356 | if (class == BPF_ALU64 || class == BPF_JMP || |
5327ed3d JW |
3357 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) |
3358 | return true; | |
3359 | ||
3360 | if (class == BPF_ALU || class == BPF_JMP32) | |
3361 | return false; | |
3362 | ||
3363 | if (class == BPF_LDX) { | |
3364 | if (t != SRC_OP) | |
577c06af | 3365 | return BPF_SIZE(code) == BPF_DW || BPF_MODE(code) == BPF_MEMSX; |
5327ed3d JW |
3366 | /* LDX source must be ptr. */ |
3367 | return true; | |
3368 | } | |
3369 | ||
3370 | if (class == BPF_STX) { | |
83a28819 IL |
3371 | /* BPF_STX (including atomic variants) has multiple source |
3372 | * operands, one of which is a ptr. Check whether the caller is | |
3373 | * asking about it. | |
3374 | */ | |
3375 | if (t == SRC_OP && reg->type != SCALAR_VALUE) | |
5327ed3d JW |
3376 | return true; |
3377 | return BPF_SIZE(code) == BPF_DW; | |
3378 | } | |
3379 | ||
3380 | if (class == BPF_LD) { | |
3381 | u8 mode = BPF_MODE(code); | |
3382 | ||
3383 | /* LD_IMM64 */ | |
3384 | if (mode == BPF_IMM) | |
3385 | return true; | |
3386 | ||
3387 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
3388 | if (t != SRC_OP) | |
3389 | return false; | |
3390 | ||
3391 | /* Implicit ctx ptr. */ | |
3392 | if (regno == BPF_REG_6) | |
3393 | return true; | |
3394 | ||
3395 | /* Explicit source could be any width. */ | |
3396 | return true; | |
3397 | } | |
3398 | ||
3399 | if (class == BPF_ST) | |
3400 | /* The only source register for BPF_ST is a ptr. */ | |
3401 | return true; | |
3402 | ||
3403 | /* Conservatively return true at default. */ | |
3404 | return true; | |
3405 | } | |
3406 | ||
83a28819 IL |
3407 | /* Return the regno defined by the insn, or -1. */ |
3408 | static int insn_def_regno(const struct bpf_insn *insn) | |
b325fbca | 3409 | { |
83a28819 IL |
3410 | switch (BPF_CLASS(insn->code)) { |
3411 | case BPF_JMP: | |
3412 | case BPF_JMP32: | |
3413 | case BPF_ST: | |
3414 | return -1; | |
3415 | case BPF_STX: | |
3416 | if (BPF_MODE(insn->code) == BPF_ATOMIC && | |
3417 | (insn->imm & BPF_FETCH)) { | |
3418 | if (insn->imm == BPF_CMPXCHG) | |
3419 | return BPF_REG_0; | |
3420 | else | |
3421 | return insn->src_reg; | |
3422 | } else { | |
3423 | return -1; | |
3424 | } | |
3425 | default: | |
3426 | return insn->dst_reg; | |
3427 | } | |
b325fbca JW |
3428 | } |
3429 | ||
3430 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
3431 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
3432 | { | |
83a28819 IL |
3433 | int dst_reg = insn_def_regno(insn); |
3434 | ||
3435 | if (dst_reg == -1) | |
b325fbca JW |
3436 | return false; |
3437 | ||
83a28819 | 3438 | return !is_reg64(env, insn, dst_reg, NULL, DST_OP); |
b325fbca JW |
3439 | } |
3440 | ||
5327ed3d JW |
3441 | static void mark_insn_zext(struct bpf_verifier_env *env, |
3442 | struct bpf_reg_state *reg) | |
3443 | { | |
3444 | s32 def_idx = reg->subreg_def; | |
3445 | ||
3446 | if (def_idx == DEF_NOT_SUBREG) | |
3447 | return; | |
3448 | ||
3449 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
3450 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
3451 | reg->subreg_def = DEF_NOT_SUBREG; | |
3452 | } | |
3453 | ||
683b96f9 EZ |
3454 | static int __check_reg_arg(struct bpf_verifier_env *env, struct bpf_reg_state *regs, u32 regno, |
3455 | enum reg_arg_type t) | |
17a52670 | 3456 | { |
5327ed3d | 3457 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
683b96f9 | 3458 | struct bpf_reg_state *reg; |
5327ed3d | 3459 | bool rw64; |
dc503a8a | 3460 | |
17a52670 | 3461 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 3462 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
3463 | return -EINVAL; |
3464 | } | |
3465 | ||
0f55f9ed CL |
3466 | mark_reg_scratched(env, regno); |
3467 | ||
c342dc10 | 3468 | reg = ®s[regno]; |
5327ed3d | 3469 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
3470 | if (t == SRC_OP) { |
3471 | /* check whether register used as source operand can be read */ | |
c342dc10 | 3472 | if (reg->type == NOT_INIT) { |
61bd5218 | 3473 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
3474 | return -EACCES; |
3475 | } | |
679c782d | 3476 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
3477 | if (regno == BPF_REG_FP) |
3478 | return 0; | |
3479 | ||
5327ed3d JW |
3480 | if (rw64) |
3481 | mark_insn_zext(env, reg); | |
3482 | ||
3483 | return mark_reg_read(env, reg, reg->parent, | |
3484 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
3485 | } else { |
3486 | /* check whether register used as dest operand can be written to */ | |
3487 | if (regno == BPF_REG_FP) { | |
61bd5218 | 3488 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
3489 | return -EACCES; |
3490 | } | |
c342dc10 | 3491 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 3492 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 3493 | if (t == DST_OP) |
61bd5218 | 3494 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
3495 | } |
3496 | return 0; | |
3497 | } | |
3498 | ||
683b96f9 EZ |
3499 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
3500 | enum reg_arg_type t) | |
3501 | { | |
3502 | struct bpf_verifier_state *vstate = env->cur_state; | |
3503 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3504 | ||
3505 | return __check_reg_arg(env, state->regs, regno, t); | |
3506 | } | |
3507 | ||
bffdeaa8 AN |
3508 | static void mark_jmp_point(struct bpf_verifier_env *env, int idx) |
3509 | { | |
3510 | env->insn_aux_data[idx].jmp_point = true; | |
3511 | } | |
3512 | ||
3513 | static bool is_jmp_point(struct bpf_verifier_env *env, int insn_idx) | |
3514 | { | |
3515 | return env->insn_aux_data[insn_idx].jmp_point; | |
3516 | } | |
3517 | ||
b5dc0163 AS |
3518 | /* for any branch, call, exit record the history of jmps in the given state */ |
3519 | static int push_jmp_history(struct bpf_verifier_env *env, | |
3520 | struct bpf_verifier_state *cur) | |
3521 | { | |
3522 | u32 cnt = cur->jmp_history_cnt; | |
3523 | struct bpf_idx_pair *p; | |
ceb35b66 | 3524 | size_t alloc_size; |
b5dc0163 | 3525 | |
bffdeaa8 AN |
3526 | if (!is_jmp_point(env, env->insn_idx)) |
3527 | return 0; | |
3528 | ||
b5dc0163 | 3529 | cnt++; |
ceb35b66 KC |
3530 | alloc_size = kmalloc_size_roundup(size_mul(cnt, sizeof(*p))); |
3531 | p = krealloc(cur->jmp_history, alloc_size, GFP_USER); | |
b5dc0163 AS |
3532 | if (!p) |
3533 | return -ENOMEM; | |
3534 | p[cnt - 1].idx = env->insn_idx; | |
3535 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
3536 | cur->jmp_history = p; | |
3537 | cur->jmp_history_cnt = cnt; | |
3538 | return 0; | |
3539 | } | |
3540 | ||
3541 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
3542 | * history then previous instruction came from straight line execution. | |
4bb7ea94 AN |
3543 | * Return -ENOENT if we exhausted all instructions within given state. |
3544 | * | |
3545 | * It's legal to have a bit of a looping with the same starting and ending | |
3546 | * insn index within the same state, e.g.: 3->4->5->3, so just because current | |
3547 | * instruction index is the same as state's first_idx doesn't mean we are | |
3548 | * done. If there is still some jump history left, we should keep going. We | |
3549 | * need to take into account that we might have a jump history between given | |
3550 | * state's parent and itself, due to checkpointing. In this case, we'll have | |
3551 | * history entry recording a jump from last instruction of parent state and | |
3552 | * first instruction of given state. | |
b5dc0163 AS |
3553 | */ |
3554 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
3555 | u32 *history) | |
3556 | { | |
3557 | u32 cnt = *history; | |
3558 | ||
4bb7ea94 AN |
3559 | if (i == st->first_insn_idx) { |
3560 | if (cnt == 0) | |
3561 | return -ENOENT; | |
3562 | if (cnt == 1 && st->jmp_history[0].idx == i) | |
3563 | return -ENOENT; | |
3564 | } | |
3565 | ||
b5dc0163 AS |
3566 | if (cnt && st->jmp_history[cnt - 1].idx == i) { |
3567 | i = st->jmp_history[cnt - 1].prev_idx; | |
3568 | (*history)--; | |
3569 | } else { | |
3570 | i--; | |
3571 | } | |
3572 | return i; | |
3573 | } | |
3574 | ||
e6ac2450 MKL |
3575 | static const char *disasm_kfunc_name(void *data, const struct bpf_insn *insn) |
3576 | { | |
3577 | const struct btf_type *func; | |
2357672c | 3578 | struct btf *desc_btf; |
e6ac2450 MKL |
3579 | |
3580 | if (insn->src_reg != BPF_PSEUDO_KFUNC_CALL) | |
3581 | return NULL; | |
3582 | ||
43bf0878 | 3583 | desc_btf = find_kfunc_desc_btf(data, insn->off); |
2357672c KKD |
3584 | if (IS_ERR(desc_btf)) |
3585 | return "<error>"; | |
3586 | ||
3587 | func = btf_type_by_id(desc_btf, insn->imm); | |
3588 | return btf_name_by_offset(desc_btf, func->name_off); | |
e6ac2450 MKL |
3589 | } |
3590 | ||
407958a0 AN |
3591 | static inline void bt_init(struct backtrack_state *bt, u32 frame) |
3592 | { | |
3593 | bt->frame = frame; | |
3594 | } | |
3595 | ||
3596 | static inline void bt_reset(struct backtrack_state *bt) | |
3597 | { | |
3598 | struct bpf_verifier_env *env = bt->env; | |
3599 | ||
3600 | memset(bt, 0, sizeof(*bt)); | |
3601 | bt->env = env; | |
3602 | } | |
3603 | ||
3604 | static inline u32 bt_empty(struct backtrack_state *bt) | |
3605 | { | |
3606 | u64 mask = 0; | |
3607 | int i; | |
3608 | ||
3609 | for (i = 0; i <= bt->frame; i++) | |
3610 | mask |= bt->reg_masks[i] | bt->stack_masks[i]; | |
3611 | ||
3612 | return mask == 0; | |
3613 | } | |
3614 | ||
3615 | static inline int bt_subprog_enter(struct backtrack_state *bt) | |
3616 | { | |
3617 | if (bt->frame == MAX_CALL_FRAMES - 1) { | |
3618 | verbose(bt->env, "BUG subprog enter from frame %d\n", bt->frame); | |
3619 | WARN_ONCE(1, "verifier backtracking bug"); | |
3620 | return -EFAULT; | |
3621 | } | |
3622 | bt->frame++; | |
3623 | return 0; | |
3624 | } | |
3625 | ||
3626 | static inline int bt_subprog_exit(struct backtrack_state *bt) | |
3627 | { | |
3628 | if (bt->frame == 0) { | |
3629 | verbose(bt->env, "BUG subprog exit from frame 0\n"); | |
3630 | WARN_ONCE(1, "verifier backtracking bug"); | |
3631 | return -EFAULT; | |
3632 | } | |
3633 | bt->frame--; | |
3634 | return 0; | |
3635 | } | |
3636 | ||
3637 | static inline void bt_set_frame_reg(struct backtrack_state *bt, u32 frame, u32 reg) | |
3638 | { | |
3639 | bt->reg_masks[frame] |= 1 << reg; | |
3640 | } | |
3641 | ||
3642 | static inline void bt_clear_frame_reg(struct backtrack_state *bt, u32 frame, u32 reg) | |
3643 | { | |
3644 | bt->reg_masks[frame] &= ~(1 << reg); | |
3645 | } | |
3646 | ||
3647 | static inline void bt_set_reg(struct backtrack_state *bt, u32 reg) | |
3648 | { | |
3649 | bt_set_frame_reg(bt, bt->frame, reg); | |
3650 | } | |
3651 | ||
3652 | static inline void bt_clear_reg(struct backtrack_state *bt, u32 reg) | |
3653 | { | |
3654 | bt_clear_frame_reg(bt, bt->frame, reg); | |
3655 | } | |
3656 | ||
3657 | static inline void bt_set_frame_slot(struct backtrack_state *bt, u32 frame, u32 slot) | |
3658 | { | |
3659 | bt->stack_masks[frame] |= 1ull << slot; | |
3660 | } | |
3661 | ||
3662 | static inline void bt_clear_frame_slot(struct backtrack_state *bt, u32 frame, u32 slot) | |
3663 | { | |
3664 | bt->stack_masks[frame] &= ~(1ull << slot); | |
3665 | } | |
3666 | ||
3667 | static inline void bt_set_slot(struct backtrack_state *bt, u32 slot) | |
3668 | { | |
3669 | bt_set_frame_slot(bt, bt->frame, slot); | |
3670 | } | |
3671 | ||
3672 | static inline void bt_clear_slot(struct backtrack_state *bt, u32 slot) | |
3673 | { | |
3674 | bt_clear_frame_slot(bt, bt->frame, slot); | |
3675 | } | |
3676 | ||
3677 | static inline u32 bt_frame_reg_mask(struct backtrack_state *bt, u32 frame) | |
3678 | { | |
3679 | return bt->reg_masks[frame]; | |
3680 | } | |
3681 | ||
3682 | static inline u32 bt_reg_mask(struct backtrack_state *bt) | |
3683 | { | |
3684 | return bt->reg_masks[bt->frame]; | |
3685 | } | |
3686 | ||
3687 | static inline u64 bt_frame_stack_mask(struct backtrack_state *bt, u32 frame) | |
3688 | { | |
3689 | return bt->stack_masks[frame]; | |
3690 | } | |
3691 | ||
3692 | static inline u64 bt_stack_mask(struct backtrack_state *bt) | |
3693 | { | |
3694 | return bt->stack_masks[bt->frame]; | |
3695 | } | |
3696 | ||
3697 | static inline bool bt_is_reg_set(struct backtrack_state *bt, u32 reg) | |
3698 | { | |
3699 | return bt->reg_masks[bt->frame] & (1 << reg); | |
3700 | } | |
3701 | ||
3702 | static inline bool bt_is_slot_set(struct backtrack_state *bt, u32 slot) | |
3703 | { | |
3704 | return bt->stack_masks[bt->frame] & (1ull << slot); | |
3705 | } | |
3706 | ||
d9439c21 AN |
3707 | /* format registers bitmask, e.g., "r0,r2,r4" for 0x15 mask */ |
3708 | static void fmt_reg_mask(char *buf, ssize_t buf_sz, u32 reg_mask) | |
3709 | { | |
3710 | DECLARE_BITMAP(mask, 64); | |
3711 | bool first = true; | |
3712 | int i, n; | |
3713 | ||
3714 | buf[0] = '\0'; | |
3715 | ||
3716 | bitmap_from_u64(mask, reg_mask); | |
3717 | for_each_set_bit(i, mask, 32) { | |
3718 | n = snprintf(buf, buf_sz, "%sr%d", first ? "" : ",", i); | |
3719 | first = false; | |
3720 | buf += n; | |
3721 | buf_sz -= n; | |
3722 | if (buf_sz < 0) | |
3723 | break; | |
3724 | } | |
3725 | } | |
3726 | /* format stack slots bitmask, e.g., "-8,-24,-40" for 0x15 mask */ | |
3727 | static void fmt_stack_mask(char *buf, ssize_t buf_sz, u64 stack_mask) | |
3728 | { | |
3729 | DECLARE_BITMAP(mask, 64); | |
3730 | bool first = true; | |
3731 | int i, n; | |
3732 | ||
3733 | buf[0] = '\0'; | |
3734 | ||
3735 | bitmap_from_u64(mask, stack_mask); | |
3736 | for_each_set_bit(i, mask, 64) { | |
3737 | n = snprintf(buf, buf_sz, "%s%d", first ? "" : ",", -(i + 1) * 8); | |
3738 | first = false; | |
3739 | buf += n; | |
3740 | buf_sz -= n; | |
3741 | if (buf_sz < 0) | |
3742 | break; | |
3743 | } | |
3744 | } | |
3745 | ||
ab5cfac1 EZ |
3746 | static bool calls_callback(struct bpf_verifier_env *env, int insn_idx); |
3747 | ||
b5dc0163 AS |
3748 | /* For given verifier state backtrack_insn() is called from the last insn to |
3749 | * the first insn. Its purpose is to compute a bitmask of registers and | |
3750 | * stack slots that needs precision in the parent verifier state. | |
fde2a388 AN |
3751 | * |
3752 | * @idx is an index of the instruction we are currently processing; | |
3753 | * @subseq_idx is an index of the subsequent instruction that: | |
3754 | * - *would be* executed next, if jump history is viewed in forward order; | |
3755 | * - *was* processed previously during backtracking. | |
b5dc0163 | 3756 | */ |
fde2a388 | 3757 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, int subseq_idx, |
407958a0 | 3758 | struct backtrack_state *bt) |
b5dc0163 AS |
3759 | { |
3760 | const struct bpf_insn_cbs cbs = { | |
e6ac2450 | 3761 | .cb_call = disasm_kfunc_name, |
b5dc0163 AS |
3762 | .cb_print = verbose, |
3763 | .private_data = env, | |
3764 | }; | |
3765 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
3766 | u8 class = BPF_CLASS(insn->code); | |
3767 | u8 opcode = BPF_OP(insn->code); | |
3768 | u8 mode = BPF_MODE(insn->code); | |
407958a0 AN |
3769 | u32 dreg = insn->dst_reg; |
3770 | u32 sreg = insn->src_reg; | |
fde2a388 | 3771 | u32 spi, i; |
b5dc0163 AS |
3772 | |
3773 | if (insn->code == 0) | |
3774 | return 0; | |
496f3324 | 3775 | if (env->log.level & BPF_LOG_LEVEL2) { |
d9439c21 AN |
3776 | fmt_reg_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, bt_reg_mask(bt)); |
3777 | verbose(env, "mark_precise: frame%d: regs=%s ", | |
3778 | bt->frame, env->tmp_str_buf); | |
3779 | fmt_stack_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, bt_stack_mask(bt)); | |
3780 | verbose(env, "stack=%s before ", env->tmp_str_buf); | |
b5dc0163 AS |
3781 | verbose(env, "%d: ", idx); |
3782 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
3783 | } | |
3784 | ||
3785 | if (class == BPF_ALU || class == BPF_ALU64) { | |
407958a0 | 3786 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 | 3787 | return 0; |
291d044f SHY |
3788 | if (opcode == BPF_END || opcode == BPF_NEG) { |
3789 | /* sreg is reserved and unused | |
3790 | * dreg still need precision before this insn | |
3791 | */ | |
3792 | return 0; | |
3793 | } else if (opcode == BPF_MOV) { | |
b5dc0163 | 3794 | if (BPF_SRC(insn->code) == BPF_X) { |
8100928c | 3795 | /* dreg = sreg or dreg = (s8, s16, s32)sreg |
b5dc0163 AS |
3796 | * dreg needs precision after this insn |
3797 | * sreg needs precision before this insn | |
3798 | */ | |
407958a0 AN |
3799 | bt_clear_reg(bt, dreg); |
3800 | bt_set_reg(bt, sreg); | |
b5dc0163 AS |
3801 | } else { |
3802 | /* dreg = K | |
3803 | * dreg needs precision after this insn. | |
3804 | * Corresponding register is already marked | |
3805 | * as precise=true in this verifier state. | |
3806 | * No further markings in parent are necessary | |
3807 | */ | |
407958a0 | 3808 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
3809 | } |
3810 | } else { | |
3811 | if (BPF_SRC(insn->code) == BPF_X) { | |
3812 | /* dreg += sreg | |
3813 | * both dreg and sreg need precision | |
3814 | * before this insn | |
3815 | */ | |
407958a0 | 3816 | bt_set_reg(bt, sreg); |
b5dc0163 AS |
3817 | } /* else dreg += K |
3818 | * dreg still needs precision before this insn | |
3819 | */ | |
3820 | } | |
3821 | } else if (class == BPF_LDX) { | |
407958a0 | 3822 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 | 3823 | return 0; |
407958a0 | 3824 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
3825 | |
3826 | /* scalars can only be spilled into stack w/o losing precision. | |
3827 | * Load from any other memory can be zero extended. | |
3828 | * The desire to keep that precision is already indicated | |
3829 | * by 'precise' mark in corresponding register of this state. | |
3830 | * No further tracking necessary. | |
3831 | */ | |
3832 | if (insn->src_reg != BPF_REG_FP) | |
3833 | return 0; | |
b5dc0163 AS |
3834 | |
3835 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
3836 | * that [fp - off] slot contains scalar that needs to be | |
3837 | * tracked with precision | |
3838 | */ | |
3839 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
3840 | if (spi >= 64) { | |
3841 | verbose(env, "BUG spi %d\n", spi); | |
3842 | WARN_ONCE(1, "verifier backtracking bug"); | |
3843 | return -EFAULT; | |
3844 | } | |
407958a0 | 3845 | bt_set_slot(bt, spi); |
b3b50f05 | 3846 | } else if (class == BPF_STX || class == BPF_ST) { |
407958a0 | 3847 | if (bt_is_reg_set(bt, dreg)) |
b3b50f05 | 3848 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
3849 | * to access memory. It means backtracking |
3850 | * encountered a case of pointer subtraction. | |
3851 | */ | |
3852 | return -ENOTSUPP; | |
3853 | /* scalars can only be spilled into stack */ | |
3854 | if (insn->dst_reg != BPF_REG_FP) | |
3855 | return 0; | |
b5dc0163 AS |
3856 | spi = (-insn->off - 1) / BPF_REG_SIZE; |
3857 | if (spi >= 64) { | |
3858 | verbose(env, "BUG spi %d\n", spi); | |
3859 | WARN_ONCE(1, "verifier backtracking bug"); | |
3860 | return -EFAULT; | |
3861 | } | |
407958a0 | 3862 | if (!bt_is_slot_set(bt, spi)) |
b5dc0163 | 3863 | return 0; |
407958a0 | 3864 | bt_clear_slot(bt, spi); |
b3b50f05 | 3865 | if (class == BPF_STX) |
407958a0 | 3866 | bt_set_reg(bt, sreg); |
b5dc0163 | 3867 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
fde2a388 AN |
3868 | if (bpf_pseudo_call(insn)) { |
3869 | int subprog_insn_idx, subprog; | |
3870 | ||
3871 | subprog_insn_idx = idx + insn->imm + 1; | |
3872 | subprog = find_subprog(env, subprog_insn_idx); | |
3873 | if (subprog < 0) | |
3874 | return -EFAULT; | |
3875 | ||
3876 | if (subprog_is_global(env, subprog)) { | |
3877 | /* check that jump history doesn't have any | |
3878 | * extra instructions from subprog; the next | |
3879 | * instruction after call to global subprog | |
3880 | * should be literally next instruction in | |
3881 | * caller program | |
3882 | */ | |
3883 | WARN_ONCE(idx + 1 != subseq_idx, "verifier backtracking bug"); | |
3884 | /* r1-r5 are invalidated after subprog call, | |
3885 | * so for global func call it shouldn't be set | |
3886 | * anymore | |
3887 | */ | |
3888 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { | |
3889 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3890 | WARN_ONCE(1, "verifier backtracking bug"); | |
3891 | return -EFAULT; | |
3892 | } | |
3893 | /* global subprog always sets R0 */ | |
3894 | bt_clear_reg(bt, BPF_REG_0); | |
3895 | return 0; | |
3896 | } else { | |
3897 | /* static subprog call instruction, which | |
3898 | * means that we are exiting current subprog, | |
3899 | * so only r1-r5 could be still requested as | |
3900 | * precise, r0 and r6-r10 or any stack slot in | |
3901 | * the current frame should be zero by now | |
3902 | */ | |
3903 | if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) { | |
3904 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3905 | WARN_ONCE(1, "verifier backtracking bug"); | |
3906 | return -EFAULT; | |
3907 | } | |
3908 | /* we don't track register spills perfectly, | |
3909 | * so fallback to force-precise instead of failing */ | |
3910 | if (bt_stack_mask(bt) != 0) | |
3911 | return -ENOTSUPP; | |
3912 | /* propagate r1-r5 to the caller */ | |
3913 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
3914 | if (bt_is_reg_set(bt, i)) { | |
3915 | bt_clear_reg(bt, i); | |
3916 | bt_set_frame_reg(bt, bt->frame - 1, i); | |
3917 | } | |
3918 | } | |
3919 | if (bt_subprog_exit(bt)) | |
3920 | return -EFAULT; | |
3921 | return 0; | |
3922 | } | |
ab5cfac1 EZ |
3923 | } else if (is_sync_callback_calling_insn(insn) && idx != subseq_idx - 1) { |
3924 | /* exit from callback subprog to callback-calling helper or | |
3925 | * kfunc call. Use idx/subseq_idx check to discern it from | |
3926 | * straight line code backtracking. | |
3927 | * Unlike the subprog call handling above, we shouldn't | |
3928 | * propagate precision of r1-r5 (if any requested), as they are | |
3929 | * not actually arguments passed directly to callback subprogs | |
be2ef816 | 3930 | */ |
fde2a388 AN |
3931 | if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) { |
3932 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3933 | WARN_ONCE(1, "verifier backtracking bug"); | |
3934 | return -EFAULT; | |
3935 | } | |
3936 | if (bt_stack_mask(bt) != 0) | |
be2ef816 | 3937 | return -ENOTSUPP; |
fde2a388 AN |
3938 | /* clear r1-r5 in callback subprog's mask */ |
3939 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
3940 | bt_clear_reg(bt, i); | |
3941 | if (bt_subprog_exit(bt)) | |
3942 | return -EFAULT; | |
3943 | return 0; | |
3944 | } else if (opcode == BPF_CALL) { | |
d3178e8a HS |
3945 | /* kfunc with imm==0 is invalid and fixup_kfunc_call will |
3946 | * catch this error later. Make backtracking conservative | |
3947 | * with ENOTSUPP. | |
3948 | */ | |
3949 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL && insn->imm == 0) | |
3950 | return -ENOTSUPP; | |
b5dc0163 | 3951 | /* regular helper call sets R0 */ |
407958a0 AN |
3952 | bt_clear_reg(bt, BPF_REG_0); |
3953 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { | |
b5dc0163 AS |
3954 | /* if backtracing was looking for registers R1-R5 |
3955 | * they should have been found already. | |
3956 | */ | |
407958a0 | 3957 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); |
b5dc0163 AS |
3958 | WARN_ONCE(1, "verifier backtracking bug"); |
3959 | return -EFAULT; | |
3960 | } | |
3961 | } else if (opcode == BPF_EXIT) { | |
fde2a388 AN |
3962 | bool r0_precise; |
3963 | ||
ab5cfac1 EZ |
3964 | /* Backtracking to a nested function call, 'idx' is a part of |
3965 | * the inner frame 'subseq_idx' is a part of the outer frame. | |
3966 | * In case of a regular function call, instructions giving | |
3967 | * precision to registers R1-R5 should have been found already. | |
3968 | * In case of a callback, it is ok to have R1-R5 marked for | |
3969 | * backtracking, as these registers are set by the function | |
3970 | * invoking callback. | |
3971 | */ | |
3972 | if (subseq_idx >= 0 && calls_callback(env, subseq_idx)) | |
3973 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
3974 | bt_clear_reg(bt, i); | |
fde2a388 | 3975 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { |
fde2a388 AN |
3976 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); |
3977 | WARN_ONCE(1, "verifier backtracking bug"); | |
3978 | return -EFAULT; | |
3979 | } | |
3980 | ||
3981 | /* BPF_EXIT in subprog or callback always returns | |
3982 | * right after the call instruction, so by checking | |
3983 | * whether the instruction at subseq_idx-1 is subprog | |
3984 | * call or not we can distinguish actual exit from | |
3985 | * *subprog* from exit from *callback*. In the former | |
3986 | * case, we need to propagate r0 precision, if | |
3987 | * necessary. In the former we never do that. | |
3988 | */ | |
3989 | r0_precise = subseq_idx - 1 >= 0 && | |
3990 | bpf_pseudo_call(&env->prog->insnsi[subseq_idx - 1]) && | |
3991 | bt_is_reg_set(bt, BPF_REG_0); | |
3992 | ||
3993 | bt_clear_reg(bt, BPF_REG_0); | |
3994 | if (bt_subprog_enter(bt)) | |
3995 | return -EFAULT; | |
3996 | ||
3997 | if (r0_precise) | |
3998 | bt_set_reg(bt, BPF_REG_0); | |
3999 | /* r6-r9 and stack slots will stay set in caller frame | |
4000 | * bitmasks until we return back from callee(s) | |
4001 | */ | |
4002 | return 0; | |
71b547f5 | 4003 | } else if (BPF_SRC(insn->code) == BPF_X) { |
407958a0 | 4004 | if (!bt_is_reg_set(bt, dreg) && !bt_is_reg_set(bt, sreg)) |
71b547f5 DB |
4005 | return 0; |
4006 | /* dreg <cond> sreg | |
4007 | * Both dreg and sreg need precision before | |
4008 | * this insn. If only sreg was marked precise | |
4009 | * before it would be equally necessary to | |
4010 | * propagate it to dreg. | |
4011 | */ | |
407958a0 AN |
4012 | bt_set_reg(bt, dreg); |
4013 | bt_set_reg(bt, sreg); | |
71b547f5 DB |
4014 | /* else dreg <cond> K |
4015 | * Only dreg still needs precision before | |
4016 | * this insn, so for the K-based conditional | |
4017 | * there is nothing new to be marked. | |
4018 | */ | |
b5dc0163 AS |
4019 | } |
4020 | } else if (class == BPF_LD) { | |
407958a0 | 4021 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 | 4022 | return 0; |
407958a0 | 4023 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
4024 | /* It's ld_imm64 or ld_abs or ld_ind. |
4025 | * For ld_imm64 no further tracking of precision | |
4026 | * into parent is necessary | |
4027 | */ | |
4028 | if (mode == BPF_IND || mode == BPF_ABS) | |
4029 | /* to be analyzed */ | |
4030 | return -ENOTSUPP; | |
b5dc0163 AS |
4031 | } |
4032 | return 0; | |
4033 | } | |
4034 | ||
4035 | /* the scalar precision tracking algorithm: | |
4036 | * . at the start all registers have precise=false. | |
4037 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
4038 | * . once precise value of the scalar register is used in: | |
4039 | * . ptr + scalar alu | |
4040 | * . if (scalar cond K|scalar) | |
4041 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
4042 | * backtrack through the verifier states and mark all registers and | |
4043 | * stack slots with spilled constants that these scalar regisers | |
4044 | * should be precise. | |
4045 | * . during state pruning two registers (or spilled stack slots) | |
4046 | * are equivalent if both are not precise. | |
4047 | * | |
4048 | * Note the verifier cannot simply walk register parentage chain, | |
4049 | * since many different registers and stack slots could have been | |
4050 | * used to compute single precise scalar. | |
4051 | * | |
4052 | * The approach of starting with precise=true for all registers and then | |
4053 | * backtrack to mark a register as not precise when the verifier detects | |
4054 | * that program doesn't care about specific value (e.g., when helper | |
4055 | * takes register as ARG_ANYTHING parameter) is not safe. | |
4056 | * | |
4057 | * It's ok to walk single parentage chain of the verifier states. | |
4058 | * It's possible that this backtracking will go all the way till 1st insn. | |
4059 | * All other branches will be explored for needing precision later. | |
4060 | * | |
4061 | * The backtracking needs to deal with cases like: | |
4062 | * R8=map_value(id=0,off=0,ks=4,vs=1952,imm=0) R9_w=map_value(id=0,off=40,ks=4,vs=1952,imm=0) | |
4063 | * r9 -= r8 | |
4064 | * r5 = r9 | |
4065 | * if r5 > 0x79f goto pc+7 | |
4066 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
4067 | * r5 += 1 | |
4068 | * ... | |
4069 | * call bpf_perf_event_output#25 | |
4070 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
4071 | * | |
4072 | * and this case: | |
4073 | * r6 = 1 | |
4074 | * call foo // uses callee's r6 inside to compute r0 | |
4075 | * r0 += r6 | |
4076 | * if r0 == 0 goto | |
4077 | * | |
4078 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
4079 | * | |
4080 | * Also if parent's curframe > frame where backtracking started, | |
4081 | * the verifier need to mark registers in both frames, otherwise callees | |
4082 | * may incorrectly prune callers. This is similar to | |
4083 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
4084 | * | |
4085 | * For now backtracking falls back into conservative marking. | |
4086 | */ | |
4087 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
4088 | struct bpf_verifier_state *st) | |
4089 | { | |
4090 | struct bpf_func_state *func; | |
4091 | struct bpf_reg_state *reg; | |
4092 | int i, j; | |
4093 | ||
d9439c21 AN |
4094 | if (env->log.level & BPF_LOG_LEVEL2) { |
4095 | verbose(env, "mark_precise: frame%d: falling back to forcing all scalars precise\n", | |
4096 | st->curframe); | |
4097 | } | |
4098 | ||
b5dc0163 AS |
4099 | /* big hammer: mark all scalars precise in this path. |
4100 | * pop_stack may still get !precise scalars. | |
f63181b6 AN |
4101 | * We also skip current state and go straight to first parent state, |
4102 | * because precision markings in current non-checkpointed state are | |
4103 | * not needed. See why in the comment in __mark_chain_precision below. | |
b5dc0163 | 4104 | */ |
f63181b6 | 4105 | for (st = st->parent; st; st = st->parent) { |
b5dc0163 AS |
4106 | for (i = 0; i <= st->curframe; i++) { |
4107 | func = st->frame[i]; | |
4108 | for (j = 0; j < BPF_REG_FP; j++) { | |
4109 | reg = &func->regs[j]; | |
d9439c21 | 4110 | if (reg->type != SCALAR_VALUE || reg->precise) |
b5dc0163 AS |
4111 | continue; |
4112 | reg->precise = true; | |
d9439c21 AN |
4113 | if (env->log.level & BPF_LOG_LEVEL2) { |
4114 | verbose(env, "force_precise: frame%d: forcing r%d to be precise\n", | |
4115 | i, j); | |
4116 | } | |
b5dc0163 AS |
4117 | } |
4118 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
27113c59 | 4119 | if (!is_spilled_reg(&func->stack[j])) |
b5dc0163 AS |
4120 | continue; |
4121 | reg = &func->stack[j].spilled_ptr; | |
d9439c21 | 4122 | if (reg->type != SCALAR_VALUE || reg->precise) |
b5dc0163 AS |
4123 | continue; |
4124 | reg->precise = true; | |
d9439c21 AN |
4125 | if (env->log.level & BPF_LOG_LEVEL2) { |
4126 | verbose(env, "force_precise: frame%d: forcing fp%d to be precise\n", | |
4127 | i, -(j + 1) * 8); | |
4128 | } | |
b5dc0163 AS |
4129 | } |
4130 | } | |
f63181b6 | 4131 | } |
b5dc0163 AS |
4132 | } |
4133 | ||
7a830b53 AN |
4134 | static void mark_all_scalars_imprecise(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
4135 | { | |
4136 | struct bpf_func_state *func; | |
4137 | struct bpf_reg_state *reg; | |
4138 | int i, j; | |
4139 | ||
4140 | for (i = 0; i <= st->curframe; i++) { | |
4141 | func = st->frame[i]; | |
4142 | for (j = 0; j < BPF_REG_FP; j++) { | |
4143 | reg = &func->regs[j]; | |
4144 | if (reg->type != SCALAR_VALUE) | |
4145 | continue; | |
4146 | reg->precise = false; | |
4147 | } | |
4148 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
4149 | if (!is_spilled_reg(&func->stack[j])) | |
4150 | continue; | |
4151 | reg = &func->stack[j].spilled_ptr; | |
4152 | if (reg->type != SCALAR_VALUE) | |
4153 | continue; | |
4154 | reg->precise = false; | |
4155 | } | |
4156 | } | |
4157 | } | |
4158 | ||
904e6ddf EZ |
4159 | static bool idset_contains(struct bpf_idset *s, u32 id) |
4160 | { | |
4161 | u32 i; | |
4162 | ||
4163 | for (i = 0; i < s->count; ++i) | |
4164 | if (s->ids[i] == id) | |
4165 | return true; | |
4166 | ||
4167 | return false; | |
4168 | } | |
4169 | ||
4170 | static int idset_push(struct bpf_idset *s, u32 id) | |
4171 | { | |
4172 | if (WARN_ON_ONCE(s->count >= ARRAY_SIZE(s->ids))) | |
4173 | return -EFAULT; | |
4174 | s->ids[s->count++] = id; | |
4175 | return 0; | |
4176 | } | |
4177 | ||
4178 | static void idset_reset(struct bpf_idset *s) | |
4179 | { | |
4180 | s->count = 0; | |
4181 | } | |
4182 | ||
4183 | /* Collect a set of IDs for all registers currently marked as precise in env->bt. | |
4184 | * Mark all registers with these IDs as precise. | |
4185 | */ | |
4186 | static int mark_precise_scalar_ids(struct bpf_verifier_env *env, struct bpf_verifier_state *st) | |
4187 | { | |
4188 | struct bpf_idset *precise_ids = &env->idset_scratch; | |
4189 | struct backtrack_state *bt = &env->bt; | |
4190 | struct bpf_func_state *func; | |
4191 | struct bpf_reg_state *reg; | |
4192 | DECLARE_BITMAP(mask, 64); | |
4193 | int i, fr; | |
4194 | ||
4195 | idset_reset(precise_ids); | |
4196 | ||
4197 | for (fr = bt->frame; fr >= 0; fr--) { | |
4198 | func = st->frame[fr]; | |
4199 | ||
4200 | bitmap_from_u64(mask, bt_frame_reg_mask(bt, fr)); | |
4201 | for_each_set_bit(i, mask, 32) { | |
4202 | reg = &func->regs[i]; | |
4203 | if (!reg->id || reg->type != SCALAR_VALUE) | |
4204 | continue; | |
4205 | if (idset_push(precise_ids, reg->id)) | |
4206 | return -EFAULT; | |
4207 | } | |
4208 | ||
4209 | bitmap_from_u64(mask, bt_frame_stack_mask(bt, fr)); | |
4210 | for_each_set_bit(i, mask, 64) { | |
4211 | if (i >= func->allocated_stack / BPF_REG_SIZE) | |
4212 | break; | |
4213 | if (!is_spilled_scalar_reg(&func->stack[i])) | |
4214 | continue; | |
4215 | reg = &func->stack[i].spilled_ptr; | |
4216 | if (!reg->id) | |
4217 | continue; | |
4218 | if (idset_push(precise_ids, reg->id)) | |
4219 | return -EFAULT; | |
4220 | } | |
4221 | } | |
4222 | ||
4223 | for (fr = 0; fr <= st->curframe; ++fr) { | |
4224 | func = st->frame[fr]; | |
4225 | ||
4226 | for (i = BPF_REG_0; i < BPF_REG_10; ++i) { | |
4227 | reg = &func->regs[i]; | |
4228 | if (!reg->id) | |
4229 | continue; | |
4230 | if (!idset_contains(precise_ids, reg->id)) | |
4231 | continue; | |
4232 | bt_set_frame_reg(bt, fr, i); | |
4233 | } | |
4234 | for (i = 0; i < func->allocated_stack / BPF_REG_SIZE; ++i) { | |
4235 | if (!is_spilled_scalar_reg(&func->stack[i])) | |
4236 | continue; | |
4237 | reg = &func->stack[i].spilled_ptr; | |
4238 | if (!reg->id) | |
4239 | continue; | |
4240 | if (!idset_contains(precise_ids, reg->id)) | |
4241 | continue; | |
4242 | bt_set_frame_slot(bt, fr, i); | |
4243 | } | |
4244 | } | |
4245 | ||
4246 | return 0; | |
4247 | } | |
4248 | ||
f63181b6 AN |
4249 | /* |
4250 | * __mark_chain_precision() backtracks BPF program instruction sequence and | |
4251 | * chain of verifier states making sure that register *regno* (if regno >= 0) | |
4252 | * and/or stack slot *spi* (if spi >= 0) are marked as precisely tracked | |
4253 | * SCALARS, as well as any other registers and slots that contribute to | |
4254 | * a tracked state of given registers/stack slots, depending on specific BPF | |
4255 | * assembly instructions (see backtrack_insns() for exact instruction handling | |
4256 | * logic). This backtracking relies on recorded jmp_history and is able to | |
4257 | * traverse entire chain of parent states. This process ends only when all the | |
4258 | * necessary registers/slots and their transitive dependencies are marked as | |
4259 | * precise. | |
4260 | * | |
4261 | * One important and subtle aspect is that precise marks *do not matter* in | |
4262 | * the currently verified state (current state). It is important to understand | |
4263 | * why this is the case. | |
4264 | * | |
4265 | * First, note that current state is the state that is not yet "checkpointed", | |
4266 | * i.e., it is not yet put into env->explored_states, and it has no children | |
4267 | * states as well. It's ephemeral, and can end up either a) being discarded if | |
4268 | * compatible explored state is found at some point or BPF_EXIT instruction is | |
4269 | * reached or b) checkpointed and put into env->explored_states, branching out | |
4270 | * into one or more children states. | |
4271 | * | |
4272 | * In the former case, precise markings in current state are completely | |
4273 | * ignored by state comparison code (see regsafe() for details). Only | |
4274 | * checkpointed ("old") state precise markings are important, and if old | |
4275 | * state's register/slot is precise, regsafe() assumes current state's | |
4276 | * register/slot as precise and checks value ranges exactly and precisely. If | |
4277 | * states turn out to be compatible, current state's necessary precise | |
4278 | * markings and any required parent states' precise markings are enforced | |
4279 | * after the fact with propagate_precision() logic, after the fact. But it's | |
4280 | * important to realize that in this case, even after marking current state | |
4281 | * registers/slots as precise, we immediately discard current state. So what | |
4282 | * actually matters is any of the precise markings propagated into current | |
4283 | * state's parent states, which are always checkpointed (due to b) case above). | |
4284 | * As such, for scenario a) it doesn't matter if current state has precise | |
4285 | * markings set or not. | |
4286 | * | |
4287 | * Now, for the scenario b), checkpointing and forking into child(ren) | |
4288 | * state(s). Note that before current state gets to checkpointing step, any | |
4289 | * processed instruction always assumes precise SCALAR register/slot | |
4290 | * knowledge: if precise value or range is useful to prune jump branch, BPF | |
4291 | * verifier takes this opportunity enthusiastically. Similarly, when | |
4292 | * register's value is used to calculate offset or memory address, exact | |
4293 | * knowledge of SCALAR range is assumed, checked, and enforced. So, similar to | |
4294 | * what we mentioned above about state comparison ignoring precise markings | |
4295 | * during state comparison, BPF verifier ignores and also assumes precise | |
4296 | * markings *at will* during instruction verification process. But as verifier | |
4297 | * assumes precision, it also propagates any precision dependencies across | |
4298 | * parent states, which are not yet finalized, so can be further restricted | |
4299 | * based on new knowledge gained from restrictions enforced by their children | |
4300 | * states. This is so that once those parent states are finalized, i.e., when | |
4301 | * they have no more active children state, state comparison logic in | |
4302 | * is_state_visited() would enforce strict and precise SCALAR ranges, if | |
4303 | * required for correctness. | |
4304 | * | |
4305 | * To build a bit more intuition, note also that once a state is checkpointed, | |
4306 | * the path we took to get to that state is not important. This is crucial | |
4307 | * property for state pruning. When state is checkpointed and finalized at | |
4308 | * some instruction index, it can be correctly and safely used to "short | |
4309 | * circuit" any *compatible* state that reaches exactly the same instruction | |
4310 | * index. I.e., if we jumped to that instruction from a completely different | |
4311 | * code path than original finalized state was derived from, it doesn't | |
4312 | * matter, current state can be discarded because from that instruction | |
4313 | * forward having a compatible state will ensure we will safely reach the | |
4314 | * exit. States describe preconditions for further exploration, but completely | |
4315 | * forget the history of how we got here. | |
4316 | * | |
4317 | * This also means that even if we needed precise SCALAR range to get to | |
4318 | * finalized state, but from that point forward *that same* SCALAR register is | |
4319 | * never used in a precise context (i.e., it's precise value is not needed for | |
4320 | * correctness), it's correct and safe to mark such register as "imprecise" | |
4321 | * (i.e., precise marking set to false). This is what we rely on when we do | |
4322 | * not set precise marking in current state. If no child state requires | |
4323 | * precision for any given SCALAR register, it's safe to dictate that it can | |
4324 | * be imprecise. If any child state does require this register to be precise, | |
4325 | * we'll mark it precise later retroactively during precise markings | |
4326 | * propagation from child state to parent states. | |
7a830b53 AN |
4327 | * |
4328 | * Skipping precise marking setting in current state is a mild version of | |
4329 | * relying on the above observation. But we can utilize this property even | |
4330 | * more aggressively by proactively forgetting any precise marking in the | |
4331 | * current state (which we inherited from the parent state), right before we | |
4332 | * checkpoint it and branch off into new child state. This is done by | |
4333 | * mark_all_scalars_imprecise() to hopefully get more permissive and generic | |
4334 | * finalized states which help in short circuiting more future states. | |
f63181b6 | 4335 | */ |
f655badf | 4336 | static int __mark_chain_precision(struct bpf_verifier_env *env, int regno) |
b5dc0163 | 4337 | { |
407958a0 | 4338 | struct backtrack_state *bt = &env->bt; |
b5dc0163 AS |
4339 | struct bpf_verifier_state *st = env->cur_state; |
4340 | int first_idx = st->first_insn_idx; | |
4341 | int last_idx = env->insn_idx; | |
d84b1a67 | 4342 | int subseq_idx = -1; |
b5dc0163 AS |
4343 | struct bpf_func_state *func; |
4344 | struct bpf_reg_state *reg; | |
b5dc0163 | 4345 | bool skip_first = true; |
d84b1a67 | 4346 | int i, fr, err; |
b5dc0163 | 4347 | |
2c78ee89 | 4348 | if (!env->bpf_capable) |
b5dc0163 AS |
4349 | return 0; |
4350 | ||
407958a0 | 4351 | /* set frame number from which we are starting to backtrack */ |
f655badf | 4352 | bt_init(bt, env->cur_state->curframe); |
407958a0 | 4353 | |
f63181b6 AN |
4354 | /* Do sanity checks against current state of register and/or stack |
4355 | * slot, but don't set precise flag in current state, as precision | |
4356 | * tracking in the current state is unnecessary. | |
4357 | */ | |
f655badf | 4358 | func = st->frame[bt->frame]; |
a3ce685d AS |
4359 | if (regno >= 0) { |
4360 | reg = &func->regs[regno]; | |
4361 | if (reg->type != SCALAR_VALUE) { | |
4362 | WARN_ONCE(1, "backtracing misuse"); | |
4363 | return -EFAULT; | |
4364 | } | |
407958a0 | 4365 | bt_set_reg(bt, regno); |
b5dc0163 | 4366 | } |
b5dc0163 | 4367 | |
407958a0 | 4368 | if (bt_empty(bt)) |
a3ce685d | 4369 | return 0; |
be2ef816 | 4370 | |
b5dc0163 AS |
4371 | for (;;) { |
4372 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
4373 | u32 history = st->jmp_history_cnt; |
4374 | ||
d9439c21 | 4375 | if (env->log.level & BPF_LOG_LEVEL2) { |
d84b1a67 AN |
4376 | verbose(env, "mark_precise: frame%d: last_idx %d first_idx %d subseq_idx %d \n", |
4377 | bt->frame, last_idx, first_idx, subseq_idx); | |
d9439c21 | 4378 | } |
be2ef816 | 4379 | |
904e6ddf EZ |
4380 | /* If some register with scalar ID is marked as precise, |
4381 | * make sure that all registers sharing this ID are also precise. | |
4382 | * This is needed to estimate effect of find_equal_scalars(). | |
4383 | * Do this at the last instruction of each state, | |
4384 | * bpf_reg_state::id fields are valid for these instructions. | |
4385 | * | |
4386 | * Allows to track precision in situation like below: | |
4387 | * | |
4388 | * r2 = unknown value | |
4389 | * ... | |
4390 | * --- state #0 --- | |
4391 | * ... | |
4392 | * r1 = r2 // r1 and r2 now share the same ID | |
4393 | * ... | |
4394 | * --- state #1 {r1.id = A, r2.id = A} --- | |
4395 | * ... | |
4396 | * if (r2 > 10) goto exit; // find_equal_scalars() assigns range to r1 | |
4397 | * ... | |
4398 | * --- state #2 {r1.id = A, r2.id = A} --- | |
4399 | * r3 = r10 | |
4400 | * r3 += r1 // need to mark both r1 and r2 | |
4401 | */ | |
4402 | if (mark_precise_scalar_ids(env, st)) | |
4403 | return -EFAULT; | |
4404 | ||
be2ef816 AN |
4405 | if (last_idx < 0) { |
4406 | /* we are at the entry into subprog, which | |
4407 | * is expected for global funcs, but only if | |
4408 | * requested precise registers are R1-R5 | |
4409 | * (which are global func's input arguments) | |
4410 | */ | |
4411 | if (st->curframe == 0 && | |
4412 | st->frame[0]->subprogno > 0 && | |
4413 | st->frame[0]->callsite == BPF_MAIN_FUNC && | |
407958a0 AN |
4414 | bt_stack_mask(bt) == 0 && |
4415 | (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) == 0) { | |
4416 | bitmap_from_u64(mask, bt_reg_mask(bt)); | |
be2ef816 AN |
4417 | for_each_set_bit(i, mask, 32) { |
4418 | reg = &st->frame[0]->regs[i]; | |
81335f90 AN |
4419 | bt_clear_reg(bt, i); |
4420 | if (reg->type == SCALAR_VALUE) | |
4421 | reg->precise = true; | |
be2ef816 AN |
4422 | } |
4423 | return 0; | |
4424 | } | |
4425 | ||
407958a0 AN |
4426 | verbose(env, "BUG backtracking func entry subprog %d reg_mask %x stack_mask %llx\n", |
4427 | st->frame[0]->subprogno, bt_reg_mask(bt), bt_stack_mask(bt)); | |
be2ef816 AN |
4428 | WARN_ONCE(1, "verifier backtracking bug"); |
4429 | return -EFAULT; | |
4430 | } | |
4431 | ||
d84b1a67 | 4432 | for (i = last_idx;;) { |
b5dc0163 AS |
4433 | if (skip_first) { |
4434 | err = 0; | |
4435 | skip_first = false; | |
4436 | } else { | |
d84b1a67 | 4437 | err = backtrack_insn(env, i, subseq_idx, bt); |
b5dc0163 AS |
4438 | } |
4439 | if (err == -ENOTSUPP) { | |
c50c0b57 | 4440 | mark_all_scalars_precise(env, env->cur_state); |
407958a0 | 4441 | bt_reset(bt); |
b5dc0163 AS |
4442 | return 0; |
4443 | } else if (err) { | |
4444 | return err; | |
4445 | } | |
407958a0 | 4446 | if (bt_empty(bt)) |
b5dc0163 AS |
4447 | /* Found assignment(s) into tracked register in this state. |
4448 | * Since this state is already marked, just return. | |
4449 | * Nothing to be tracked further in the parent state. | |
4450 | */ | |
4451 | return 0; | |
d84b1a67 | 4452 | subseq_idx = i; |
b5dc0163 | 4453 | i = get_prev_insn_idx(st, i, &history); |
4bb7ea94 AN |
4454 | if (i == -ENOENT) |
4455 | break; | |
b5dc0163 AS |
4456 | if (i >= env->prog->len) { |
4457 | /* This can happen if backtracking reached insn 0 | |
4458 | * and there are still reg_mask or stack_mask | |
4459 | * to backtrack. | |
4460 | * It means the backtracking missed the spot where | |
4461 | * particular register was initialized with a constant. | |
4462 | */ | |
4463 | verbose(env, "BUG backtracking idx %d\n", i); | |
4464 | WARN_ONCE(1, "verifier backtracking bug"); | |
4465 | return -EFAULT; | |
4466 | } | |
4467 | } | |
4468 | st = st->parent; | |
4469 | if (!st) | |
4470 | break; | |
4471 | ||
1ef22b68 AN |
4472 | for (fr = bt->frame; fr >= 0; fr--) { |
4473 | func = st->frame[fr]; | |
4474 | bitmap_from_u64(mask, bt_frame_reg_mask(bt, fr)); | |
4475 | for_each_set_bit(i, mask, 32) { | |
4476 | reg = &func->regs[i]; | |
4477 | if (reg->type != SCALAR_VALUE) { | |
4478 | bt_clear_frame_reg(bt, fr, i); | |
4479 | continue; | |
4480 | } | |
4481 | if (reg->precise) | |
4482 | bt_clear_frame_reg(bt, fr, i); | |
4483 | else | |
4484 | reg->precise = true; | |
a3ce685d | 4485 | } |
b5dc0163 | 4486 | |
1ef22b68 AN |
4487 | bitmap_from_u64(mask, bt_frame_stack_mask(bt, fr)); |
4488 | for_each_set_bit(i, mask, 64) { | |
4489 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
4490 | /* the sequence of instructions: | |
4491 | * 2: (bf) r3 = r10 | |
4492 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
4493 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
4494 | * doesn't contain jmps. It's backtracked | |
4495 | * as a single block. | |
4496 | * During backtracking insn 3 is not recognized as | |
4497 | * stack access, so at the end of backtracking | |
4498 | * stack slot fp-8 is still marked in stack_mask. | |
4499 | * However the parent state may not have accessed | |
4500 | * fp-8 and it's "unallocated" stack space. | |
4501 | * In such case fallback to conservative. | |
4502 | */ | |
c50c0b57 | 4503 | mark_all_scalars_precise(env, env->cur_state); |
1ef22b68 AN |
4504 | bt_reset(bt); |
4505 | return 0; | |
4506 | } | |
b5dc0163 | 4507 | |
1ef22b68 AN |
4508 | if (!is_spilled_scalar_reg(&func->stack[i])) { |
4509 | bt_clear_frame_slot(bt, fr, i); | |
4510 | continue; | |
4511 | } | |
4512 | reg = &func->stack[i].spilled_ptr; | |
4513 | if (reg->precise) | |
4514 | bt_clear_frame_slot(bt, fr, i); | |
4515 | else | |
4516 | reg->precise = true; | |
4517 | } | |
4518 | if (env->log.level & BPF_LOG_LEVEL2) { | |
4519 | fmt_reg_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, | |
4520 | bt_frame_reg_mask(bt, fr)); | |
4521 | verbose(env, "mark_precise: frame%d: parent state regs=%s ", | |
4522 | fr, env->tmp_str_buf); | |
4523 | fmt_stack_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, | |
4524 | bt_frame_stack_mask(bt, fr)); | |
4525 | verbose(env, "stack=%s: ", env->tmp_str_buf); | |
4526 | print_verifier_state(env, func, true); | |
a3ce685d | 4527 | } |
b5dc0163 AS |
4528 | } |
4529 | ||
407958a0 | 4530 | if (bt_empty(bt)) |
c50c0b57 | 4531 | return 0; |
b5dc0163 | 4532 | |
d84b1a67 | 4533 | subseq_idx = first_idx; |
b5dc0163 AS |
4534 | last_idx = st->last_insn_idx; |
4535 | first_idx = st->first_insn_idx; | |
4536 | } | |
c50c0b57 AN |
4537 | |
4538 | /* if we still have requested precise regs or slots, we missed | |
4539 | * something (e.g., stack access through non-r10 register), so | |
4540 | * fallback to marking all precise | |
4541 | */ | |
4542 | if (!bt_empty(bt)) { | |
4543 | mark_all_scalars_precise(env, env->cur_state); | |
4544 | bt_reset(bt); | |
4545 | } | |
4546 | ||
b5dc0163 AS |
4547 | return 0; |
4548 | } | |
4549 | ||
eb1f7f71 | 4550 | int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
a3ce685d | 4551 | { |
f655badf | 4552 | return __mark_chain_precision(env, regno); |
a3ce685d AS |
4553 | } |
4554 | ||
f655badf AN |
4555 | /* mark_chain_precision_batch() assumes that env->bt is set in the caller to |
4556 | * desired reg and stack masks across all relevant frames | |
4557 | */ | |
4558 | static int mark_chain_precision_batch(struct bpf_verifier_env *env) | |
a3ce685d | 4559 | { |
f655badf | 4560 | return __mark_chain_precision(env, -1); |
a3ce685d | 4561 | } |
b5dc0163 | 4562 | |
1be7f75d AS |
4563 | static bool is_spillable_regtype(enum bpf_reg_type type) |
4564 | { | |
c25b2ae1 | 4565 | switch (base_type(type)) { |
1be7f75d | 4566 | case PTR_TO_MAP_VALUE: |
1be7f75d AS |
4567 | case PTR_TO_STACK: |
4568 | case PTR_TO_CTX: | |
969bf05e | 4569 | case PTR_TO_PACKET: |
de8f3a83 | 4570 | case PTR_TO_PACKET_META: |
969bf05e | 4571 | case PTR_TO_PACKET_END: |
d58e468b | 4572 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 4573 | case CONST_PTR_TO_MAP: |
c64b7983 | 4574 | case PTR_TO_SOCKET: |
46f8bc92 | 4575 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 4576 | case PTR_TO_TCP_SOCK: |
fada7fdc | 4577 | case PTR_TO_XDP_SOCK: |
65726b5b | 4578 | case PTR_TO_BTF_ID: |
20b2aff4 | 4579 | case PTR_TO_BUF: |
744ea4e3 | 4580 | case PTR_TO_MEM: |
69c087ba YS |
4581 | case PTR_TO_FUNC: |
4582 | case PTR_TO_MAP_KEY: | |
1be7f75d AS |
4583 | return true; |
4584 | default: | |
4585 | return false; | |
4586 | } | |
4587 | } | |
4588 | ||
cc2b14d5 AS |
4589 | /* Does this register contain a constant zero? */ |
4590 | static bool register_is_null(struct bpf_reg_state *reg) | |
4591 | { | |
4592 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
4593 | } | |
4594 | ||
f7cf25b2 AS |
4595 | static bool register_is_const(struct bpf_reg_state *reg) |
4596 | { | |
4597 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
4598 | } | |
4599 | ||
5689d49b YS |
4600 | static bool __is_scalar_unbounded(struct bpf_reg_state *reg) |
4601 | { | |
4602 | return tnum_is_unknown(reg->var_off) && | |
4603 | reg->smin_value == S64_MIN && reg->smax_value == S64_MAX && | |
4604 | reg->umin_value == 0 && reg->umax_value == U64_MAX && | |
4605 | reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX && | |
4606 | reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX; | |
4607 | } | |
4608 | ||
4609 | static bool register_is_bounded(struct bpf_reg_state *reg) | |
4610 | { | |
4611 | return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg); | |
4612 | } | |
4613 | ||
6e7e63cb JH |
4614 | static bool __is_pointer_value(bool allow_ptr_leaks, |
4615 | const struct bpf_reg_state *reg) | |
4616 | { | |
4617 | if (allow_ptr_leaks) | |
4618 | return false; | |
4619 | ||
4620 | return reg->type != SCALAR_VALUE; | |
4621 | } | |
4622 | ||
71f656a5 EZ |
4623 | /* Copy src state preserving dst->parent and dst->live fields */ |
4624 | static void copy_register_state(struct bpf_reg_state *dst, const struct bpf_reg_state *src) | |
4625 | { | |
4626 | struct bpf_reg_state *parent = dst->parent; | |
4627 | enum bpf_reg_liveness live = dst->live; | |
4628 | ||
4629 | *dst = *src; | |
4630 | dst->parent = parent; | |
4631 | dst->live = live; | |
4632 | } | |
4633 | ||
f7cf25b2 | 4634 | static void save_register_state(struct bpf_func_state *state, |
354e8f19 MKL |
4635 | int spi, struct bpf_reg_state *reg, |
4636 | int size) | |
f7cf25b2 AS |
4637 | { |
4638 | int i; | |
4639 | ||
71f656a5 | 4640 | copy_register_state(&state->stack[spi].spilled_ptr, reg); |
354e8f19 MKL |
4641 | if (size == BPF_REG_SIZE) |
4642 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
f7cf25b2 | 4643 | |
354e8f19 MKL |
4644 | for (i = BPF_REG_SIZE; i > BPF_REG_SIZE - size; i--) |
4645 | state->stack[spi].slot_type[i - 1] = STACK_SPILL; | |
f7cf25b2 | 4646 | |
354e8f19 MKL |
4647 | /* size < 8 bytes spill */ |
4648 | for (; i; i--) | |
4649 | scrub_spilled_slot(&state->stack[spi].slot_type[i - 1]); | |
f7cf25b2 AS |
4650 | } |
4651 | ||
ecdf985d EZ |
4652 | static bool is_bpf_st_mem(struct bpf_insn *insn) |
4653 | { | |
4654 | return BPF_CLASS(insn->code) == BPF_ST && BPF_MODE(insn->code) == BPF_MEM; | |
4655 | } | |
4656 | ||
01f810ac | 4657 | /* check_stack_{read,write}_fixed_off functions track spill/fill of registers, |
17a52670 AS |
4658 | * stack boundary and alignment are checked in check_mem_access() |
4659 | */ | |
01f810ac AM |
4660 | static int check_stack_write_fixed_off(struct bpf_verifier_env *env, |
4661 | /* stack frame we're writing to */ | |
4662 | struct bpf_func_state *state, | |
4663 | int off, int size, int value_regno, | |
4664 | int insn_idx) | |
17a52670 | 4665 | { |
f4d7e40a | 4666 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 4667 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
ecdf985d | 4668 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; |
f7cf25b2 | 4669 | struct bpf_reg_state *reg = NULL; |
ecdf985d | 4670 | u32 dst_reg = insn->dst_reg; |
638f5b90 | 4671 | |
c69431aa | 4672 | err = grow_stack_state(state, round_up(slot + 1, BPF_REG_SIZE)); |
638f5b90 AS |
4673 | if (err) |
4674 | return err; | |
9c399760 AS |
4675 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
4676 | * so it's aligned access and [off, off + size) are within stack limits | |
4677 | */ | |
638f5b90 AS |
4678 | if (!env->allow_ptr_leaks && |
4679 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
4680 | size != BPF_REG_SIZE) { | |
4681 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
4682 | return -EACCES; | |
4683 | } | |
17a52670 | 4684 | |
f4d7e40a | 4685 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
4686 | if (value_regno >= 0) |
4687 | reg = &cur->regs[value_regno]; | |
2039f26f DB |
4688 | if (!env->bypass_spec_v4) { |
4689 | bool sanitize = reg && is_spillable_regtype(reg->type); | |
4690 | ||
4691 | for (i = 0; i < size; i++) { | |
e4f4db47 LG |
4692 | u8 type = state->stack[spi].slot_type[i]; |
4693 | ||
4694 | if (type != STACK_MISC && type != STACK_ZERO) { | |
2039f26f DB |
4695 | sanitize = true; |
4696 | break; | |
4697 | } | |
4698 | } | |
4699 | ||
4700 | if (sanitize) | |
4701 | env->insn_aux_data[insn_idx].sanitize_stack_spill = true; | |
4702 | } | |
17a52670 | 4703 | |
ef8fc7a0 KKD |
4704 | err = destroy_if_dynptr_stack_slot(env, state, spi); |
4705 | if (err) | |
4706 | return err; | |
4707 | ||
0f55f9ed | 4708 | mark_stack_slot_scratched(env, spi); |
354e8f19 | 4709 | if (reg && !(off % BPF_REG_SIZE) && register_is_bounded(reg) && |
2c78ee89 | 4710 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
4711 | if (dst_reg != BPF_REG_FP) { |
4712 | /* The backtracking logic can only recognize explicit | |
4713 | * stack slot address like [fp - 8]. Other spill of | |
8fb33b60 | 4714 | * scalar via different register has to be conservative. |
b5dc0163 AS |
4715 | * Backtrack from here and mark all registers as precise |
4716 | * that contributed into 'reg' being a constant. | |
4717 | */ | |
4718 | err = mark_chain_precision(env, value_regno); | |
4719 | if (err) | |
4720 | return err; | |
4721 | } | |
354e8f19 | 4722 | save_register_state(state, spi, reg, size); |
713274f1 MM |
4723 | /* Break the relation on a narrowing spill. */ |
4724 | if (fls64(reg->umax_value) > BITS_PER_BYTE * size) | |
4725 | state->stack[spi].spilled_ptr.id = 0; | |
ecdf985d EZ |
4726 | } else if (!reg && !(off % BPF_REG_SIZE) && is_bpf_st_mem(insn) && |
4727 | insn->imm != 0 && env->bpf_capable) { | |
4728 | struct bpf_reg_state fake_reg = {}; | |
4729 | ||
811c3636 | 4730 | __mark_reg_known(&fake_reg, insn->imm); |
ecdf985d EZ |
4731 | fake_reg.type = SCALAR_VALUE; |
4732 | save_register_state(state, spi, &fake_reg, size); | |
f7cf25b2 | 4733 | } else if (reg && is_spillable_regtype(reg->type)) { |
17a52670 | 4734 | /* register containing pointer is being spilled into stack */ |
9c399760 | 4735 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 4736 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 4737 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
4738 | return -EACCES; |
4739 | } | |
f7cf25b2 | 4740 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
4741 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
4742 | return -EINVAL; | |
4743 | } | |
354e8f19 | 4744 | save_register_state(state, spi, reg, size); |
9c399760 | 4745 | } else { |
cc2b14d5 AS |
4746 | u8 type = STACK_MISC; |
4747 | ||
679c782d EC |
4748 | /* regular write of data into stack destroys any spilled ptr */ |
4749 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
06accc87 AN |
4750 | /* Mark slots as STACK_MISC if they belonged to spilled ptr/dynptr/iter. */ |
4751 | if (is_stack_slot_special(&state->stack[spi])) | |
0bae2d4d | 4752 | for (i = 0; i < BPF_REG_SIZE; i++) |
354e8f19 | 4753 | scrub_spilled_slot(&state->stack[spi].slot_type[i]); |
9c399760 | 4754 | |
cc2b14d5 AS |
4755 | /* only mark the slot as written if all 8 bytes were written |
4756 | * otherwise read propagation may incorrectly stop too soon | |
4757 | * when stack slots are partially written. | |
4758 | * This heuristic means that read propagation will be | |
4759 | * conservative, since it will add reg_live_read marks | |
4760 | * to stack slots all the way to first state when programs | |
4761 | * writes+reads less than 8 bytes | |
4762 | */ | |
4763 | if (size == BPF_REG_SIZE) | |
4764 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
4765 | ||
4766 | /* when we zero initialize stack slots mark them as such */ | |
ecdf985d EZ |
4767 | if ((reg && register_is_null(reg)) || |
4768 | (!reg && is_bpf_st_mem(insn) && insn->imm == 0)) { | |
b5dc0163 AS |
4769 | /* backtracking doesn't work for STACK_ZERO yet. */ |
4770 | err = mark_chain_precision(env, value_regno); | |
4771 | if (err) | |
4772 | return err; | |
cc2b14d5 | 4773 | type = STACK_ZERO; |
b5dc0163 | 4774 | } |
cc2b14d5 | 4775 | |
0bae2d4d | 4776 | /* Mark slots affected by this stack write. */ |
9c399760 | 4777 | for (i = 0; i < size; i++) |
638f5b90 | 4778 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 4779 | type; |
17a52670 AS |
4780 | } |
4781 | return 0; | |
4782 | } | |
4783 | ||
01f810ac AM |
4784 | /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is |
4785 | * known to contain a variable offset. | |
4786 | * This function checks whether the write is permitted and conservatively | |
4787 | * tracks the effects of the write, considering that each stack slot in the | |
4788 | * dynamic range is potentially written to. | |
4789 | * | |
4790 | * 'off' includes 'regno->off'. | |
4791 | * 'value_regno' can be -1, meaning that an unknown value is being written to | |
4792 | * the stack. | |
4793 | * | |
4794 | * Spilled pointers in range are not marked as written because we don't know | |
4795 | * what's going to be actually written. This means that read propagation for | |
4796 | * future reads cannot be terminated by this write. | |
4797 | * | |
4798 | * For privileged programs, uninitialized stack slots are considered | |
4799 | * initialized by this write (even though we don't know exactly what offsets | |
4800 | * are going to be written to). The idea is that we don't want the verifier to | |
4801 | * reject future reads that access slots written to through variable offsets. | |
4802 | */ | |
4803 | static int check_stack_write_var_off(struct bpf_verifier_env *env, | |
4804 | /* func where register points to */ | |
4805 | struct bpf_func_state *state, | |
4806 | int ptr_regno, int off, int size, | |
4807 | int value_regno, int insn_idx) | |
4808 | { | |
4809 | struct bpf_func_state *cur; /* state of the current function */ | |
4810 | int min_off, max_off; | |
4811 | int i, err; | |
4812 | struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL; | |
31ff2135 | 4813 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; |
01f810ac AM |
4814 | bool writing_zero = false; |
4815 | /* set if the fact that we're writing a zero is used to let any | |
4816 | * stack slots remain STACK_ZERO | |
4817 | */ | |
4818 | bool zero_used = false; | |
4819 | ||
4820 | cur = env->cur_state->frame[env->cur_state->curframe]; | |
4821 | ptr_reg = &cur->regs[ptr_regno]; | |
4822 | min_off = ptr_reg->smin_value + off; | |
4823 | max_off = ptr_reg->smax_value + off + size; | |
4824 | if (value_regno >= 0) | |
4825 | value_reg = &cur->regs[value_regno]; | |
31ff2135 EZ |
4826 | if ((value_reg && register_is_null(value_reg)) || |
4827 | (!value_reg && is_bpf_st_mem(insn) && insn->imm == 0)) | |
01f810ac AM |
4828 | writing_zero = true; |
4829 | ||
c69431aa | 4830 | err = grow_stack_state(state, round_up(-min_off, BPF_REG_SIZE)); |
01f810ac AM |
4831 | if (err) |
4832 | return err; | |
4833 | ||
ef8fc7a0 KKD |
4834 | for (i = min_off; i < max_off; i++) { |
4835 | int spi; | |
4836 | ||
4837 | spi = __get_spi(i); | |
4838 | err = destroy_if_dynptr_stack_slot(env, state, spi); | |
4839 | if (err) | |
4840 | return err; | |
4841 | } | |
01f810ac AM |
4842 | |
4843 | /* Variable offset writes destroy any spilled pointers in range. */ | |
4844 | for (i = min_off; i < max_off; i++) { | |
4845 | u8 new_type, *stype; | |
4846 | int slot, spi; | |
4847 | ||
4848 | slot = -i - 1; | |
4849 | spi = slot / BPF_REG_SIZE; | |
4850 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
0f55f9ed | 4851 | mark_stack_slot_scratched(env, spi); |
01f810ac | 4852 | |
f5e477a8 KKD |
4853 | if (!env->allow_ptr_leaks && *stype != STACK_MISC && *stype != STACK_ZERO) { |
4854 | /* Reject the write if range we may write to has not | |
4855 | * been initialized beforehand. If we didn't reject | |
4856 | * here, the ptr status would be erased below (even | |
4857 | * though not all slots are actually overwritten), | |
4858 | * possibly opening the door to leaks. | |
4859 | * | |
4860 | * We do however catch STACK_INVALID case below, and | |
4861 | * only allow reading possibly uninitialized memory | |
4862 | * later for CAP_PERFMON, as the write may not happen to | |
4863 | * that slot. | |
01f810ac AM |
4864 | */ |
4865 | verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d", | |
4866 | insn_idx, i); | |
4867 | return -EINVAL; | |
4868 | } | |
4869 | ||
4870 | /* Erase all spilled pointers. */ | |
4871 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
4872 | ||
4873 | /* Update the slot type. */ | |
4874 | new_type = STACK_MISC; | |
4875 | if (writing_zero && *stype == STACK_ZERO) { | |
4876 | new_type = STACK_ZERO; | |
4877 | zero_used = true; | |
4878 | } | |
4879 | /* If the slot is STACK_INVALID, we check whether it's OK to | |
4880 | * pretend that it will be initialized by this write. The slot | |
4881 | * might not actually be written to, and so if we mark it as | |
4882 | * initialized future reads might leak uninitialized memory. | |
4883 | * For privileged programs, we will accept such reads to slots | |
4884 | * that may or may not be written because, if we're reject | |
4885 | * them, the error would be too confusing. | |
4886 | */ | |
4887 | if (*stype == STACK_INVALID && !env->allow_uninit_stack) { | |
4888 | verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d", | |
4889 | insn_idx, i); | |
4890 | return -EINVAL; | |
4891 | } | |
4892 | *stype = new_type; | |
4893 | } | |
4894 | if (zero_used) { | |
4895 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
4896 | err = mark_chain_precision(env, value_regno); | |
4897 | if (err) | |
4898 | return err; | |
4899 | } | |
4900 | return 0; | |
4901 | } | |
4902 | ||
4903 | /* When register 'dst_regno' is assigned some values from stack[min_off, | |
4904 | * max_off), we set the register's type according to the types of the | |
4905 | * respective stack slots. If all the stack values are known to be zeros, then | |
4906 | * so is the destination reg. Otherwise, the register is considered to be | |
4907 | * SCALAR. This function does not deal with register filling; the caller must | |
4908 | * ensure that all spilled registers in the stack range have been marked as | |
4909 | * read. | |
4910 | */ | |
4911 | static void mark_reg_stack_read(struct bpf_verifier_env *env, | |
4912 | /* func where src register points to */ | |
4913 | struct bpf_func_state *ptr_state, | |
4914 | int min_off, int max_off, int dst_regno) | |
4915 | { | |
4916 | struct bpf_verifier_state *vstate = env->cur_state; | |
4917 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
4918 | int i, slot, spi; | |
4919 | u8 *stype; | |
4920 | int zeros = 0; | |
4921 | ||
4922 | for (i = min_off; i < max_off; i++) { | |
4923 | slot = -i - 1; | |
4924 | spi = slot / BPF_REG_SIZE; | |
e0bf4622 | 4925 | mark_stack_slot_scratched(env, spi); |
01f810ac AM |
4926 | stype = ptr_state->stack[spi].slot_type; |
4927 | if (stype[slot % BPF_REG_SIZE] != STACK_ZERO) | |
4928 | break; | |
4929 | zeros++; | |
4930 | } | |
4931 | if (zeros == max_off - min_off) { | |
4932 | /* any access_size read into register is zero extended, | |
4933 | * so the whole register == const_zero | |
4934 | */ | |
4935 | __mark_reg_const_zero(&state->regs[dst_regno]); | |
4936 | /* backtracking doesn't support STACK_ZERO yet, | |
4937 | * so mark it precise here, so that later | |
4938 | * backtracking can stop here. | |
4939 | * Backtracking may not need this if this register | |
4940 | * doesn't participate in pointer adjustment. | |
4941 | * Forward propagation of precise flag is not | |
4942 | * necessary either. This mark is only to stop | |
4943 | * backtracking. Any register that contributed | |
4944 | * to const 0 was marked precise before spill. | |
4945 | */ | |
4946 | state->regs[dst_regno].precise = true; | |
4947 | } else { | |
4948 | /* have read misc data from the stack */ | |
4949 | mark_reg_unknown(env, state->regs, dst_regno); | |
4950 | } | |
4951 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
4952 | } | |
4953 | ||
4954 | /* Read the stack at 'off' and put the results into the register indicated by | |
4955 | * 'dst_regno'. It handles reg filling if the addressed stack slot is a | |
4956 | * spilled reg. | |
4957 | * | |
4958 | * 'dst_regno' can be -1, meaning that the read value is not going to a | |
4959 | * register. | |
4960 | * | |
4961 | * The access is assumed to be within the current stack bounds. | |
4962 | */ | |
4963 | static int check_stack_read_fixed_off(struct bpf_verifier_env *env, | |
4964 | /* func where src register points to */ | |
4965 | struct bpf_func_state *reg_state, | |
4966 | int off, int size, int dst_regno) | |
17a52670 | 4967 | { |
f4d7e40a AS |
4968 | struct bpf_verifier_state *vstate = env->cur_state; |
4969 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 4970 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 4971 | struct bpf_reg_state *reg; |
354e8f19 | 4972 | u8 *stype, type; |
17a52670 | 4973 | |
f4d7e40a | 4974 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 4975 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 4976 | |
e0bf4622 AN |
4977 | mark_stack_slot_scratched(env, spi); |
4978 | ||
27113c59 | 4979 | if (is_spilled_reg(®_state->stack[spi])) { |
f30d4968 MKL |
4980 | u8 spill_size = 1; |
4981 | ||
4982 | for (i = BPF_REG_SIZE - 1; i > 0 && stype[i - 1] == STACK_SPILL; i--) | |
4983 | spill_size++; | |
354e8f19 | 4984 | |
f30d4968 | 4985 | if (size != BPF_REG_SIZE || spill_size != BPF_REG_SIZE) { |
f7cf25b2 AS |
4986 | if (reg->type != SCALAR_VALUE) { |
4987 | verbose_linfo(env, env->insn_idx, "; "); | |
4988 | verbose(env, "invalid size of register fill\n"); | |
4989 | return -EACCES; | |
4990 | } | |
354e8f19 MKL |
4991 | |
4992 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
4993 | if (dst_regno < 0) | |
4994 | return 0; | |
4995 | ||
f30d4968 | 4996 | if (!(off % BPF_REG_SIZE) && size == spill_size) { |
354e8f19 MKL |
4997 | /* The earlier check_reg_arg() has decided the |
4998 | * subreg_def for this insn. Save it first. | |
4999 | */ | |
5000 | s32 subreg_def = state->regs[dst_regno].subreg_def; | |
5001 | ||
71f656a5 | 5002 | copy_register_state(&state->regs[dst_regno], reg); |
354e8f19 MKL |
5003 | state->regs[dst_regno].subreg_def = subreg_def; |
5004 | } else { | |
5005 | for (i = 0; i < size; i++) { | |
5006 | type = stype[(slot - i) % BPF_REG_SIZE]; | |
5007 | if (type == STACK_SPILL) | |
5008 | continue; | |
5009 | if (type == STACK_MISC) | |
5010 | continue; | |
6715df8d EZ |
5011 | if (type == STACK_INVALID && env->allow_uninit_stack) |
5012 | continue; | |
354e8f19 MKL |
5013 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
5014 | off, i, size); | |
5015 | return -EACCES; | |
5016 | } | |
01f810ac | 5017 | mark_reg_unknown(env, state->regs, dst_regno); |
f7cf25b2 | 5018 | } |
354e8f19 | 5019 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
f7cf25b2 | 5020 | return 0; |
17a52670 | 5021 | } |
17a52670 | 5022 | |
01f810ac | 5023 | if (dst_regno >= 0) { |
17a52670 | 5024 | /* restore register state from stack */ |
71f656a5 | 5025 | copy_register_state(&state->regs[dst_regno], reg); |
2f18f62e AS |
5026 | /* mark reg as written since spilled pointer state likely |
5027 | * has its liveness marks cleared by is_state_visited() | |
5028 | * which resets stack/reg liveness for state transitions | |
5029 | */ | |
01f810ac | 5030 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
6e7e63cb | 5031 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
01f810ac | 5032 | /* If dst_regno==-1, the caller is asking us whether |
6e7e63cb JH |
5033 | * it is acceptable to use this value as a SCALAR_VALUE |
5034 | * (e.g. for XADD). | |
5035 | * We must not allow unprivileged callers to do that | |
5036 | * with spilled pointers. | |
5037 | */ | |
5038 | verbose(env, "leaking pointer from stack off %d\n", | |
5039 | off); | |
5040 | return -EACCES; | |
dc503a8a | 5041 | } |
f7cf25b2 | 5042 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 AS |
5043 | } else { |
5044 | for (i = 0; i < size; i++) { | |
01f810ac AM |
5045 | type = stype[(slot - i) % BPF_REG_SIZE]; |
5046 | if (type == STACK_MISC) | |
cc2b14d5 | 5047 | continue; |
01f810ac | 5048 | if (type == STACK_ZERO) |
cc2b14d5 | 5049 | continue; |
6715df8d EZ |
5050 | if (type == STACK_INVALID && env->allow_uninit_stack) |
5051 | continue; | |
cc2b14d5 AS |
5052 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
5053 | off, i, size); | |
5054 | return -EACCES; | |
5055 | } | |
f7cf25b2 | 5056 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
01f810ac AM |
5057 | if (dst_regno >= 0) |
5058 | mark_reg_stack_read(env, reg_state, off, off + size, dst_regno); | |
17a52670 | 5059 | } |
f7cf25b2 | 5060 | return 0; |
17a52670 AS |
5061 | } |
5062 | ||
61df10c7 | 5063 | enum bpf_access_src { |
01f810ac AM |
5064 | ACCESS_DIRECT = 1, /* the access is performed by an instruction */ |
5065 | ACCESS_HELPER = 2, /* the access is performed by a helper */ | |
5066 | }; | |
5067 | ||
5068 | static int check_stack_range_initialized(struct bpf_verifier_env *env, | |
5069 | int regno, int off, int access_size, | |
5070 | bool zero_size_allowed, | |
61df10c7 | 5071 | enum bpf_access_src type, |
01f810ac AM |
5072 | struct bpf_call_arg_meta *meta); |
5073 | ||
5074 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) | |
5075 | { | |
5076 | return cur_regs(env) + regno; | |
5077 | } | |
5078 | ||
5079 | /* Read the stack at 'ptr_regno + off' and put the result into the register | |
5080 | * 'dst_regno'. | |
5081 | * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'), | |
5082 | * but not its variable offset. | |
5083 | * 'size' is assumed to be <= reg size and the access is assumed to be aligned. | |
5084 | * | |
5085 | * As opposed to check_stack_read_fixed_off, this function doesn't deal with | |
5086 | * filling registers (i.e. reads of spilled register cannot be detected when | |
5087 | * the offset is not fixed). We conservatively mark 'dst_regno' as containing | |
5088 | * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable | |
5089 | * offset; for a fixed offset check_stack_read_fixed_off should be used | |
5090 | * instead. | |
5091 | */ | |
5092 | static int check_stack_read_var_off(struct bpf_verifier_env *env, | |
5093 | int ptr_regno, int off, int size, int dst_regno) | |
e4298d25 | 5094 | { |
01f810ac AM |
5095 | /* The state of the source register. */ |
5096 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
5097 | struct bpf_func_state *ptr_state = func(env, reg); | |
5098 | int err; | |
5099 | int min_off, max_off; | |
5100 | ||
5101 | /* Note that we pass a NULL meta, so raw access will not be permitted. | |
e4298d25 | 5102 | */ |
01f810ac AM |
5103 | err = check_stack_range_initialized(env, ptr_regno, off, size, |
5104 | false, ACCESS_DIRECT, NULL); | |
5105 | if (err) | |
5106 | return err; | |
5107 | ||
5108 | min_off = reg->smin_value + off; | |
5109 | max_off = reg->smax_value + off; | |
5110 | mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno); | |
5111 | return 0; | |
5112 | } | |
5113 | ||
5114 | /* check_stack_read dispatches to check_stack_read_fixed_off or | |
5115 | * check_stack_read_var_off. | |
5116 | * | |
5117 | * The caller must ensure that the offset falls within the allocated stack | |
5118 | * bounds. | |
5119 | * | |
5120 | * 'dst_regno' is a register which will receive the value from the stack. It | |
5121 | * can be -1, meaning that the read value is not going to a register. | |
5122 | */ | |
5123 | static int check_stack_read(struct bpf_verifier_env *env, | |
5124 | int ptr_regno, int off, int size, | |
5125 | int dst_regno) | |
5126 | { | |
5127 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
5128 | struct bpf_func_state *state = func(env, reg); | |
5129 | int err; | |
5130 | /* Some accesses are only permitted with a static offset. */ | |
5131 | bool var_off = !tnum_is_const(reg->var_off); | |
5132 | ||
5133 | /* The offset is required to be static when reads don't go to a | |
5134 | * register, in order to not leak pointers (see | |
5135 | * check_stack_read_fixed_off). | |
5136 | */ | |
5137 | if (dst_regno < 0 && var_off) { | |
e4298d25 DB |
5138 | char tn_buf[48]; |
5139 | ||
5140 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac | 5141 | verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n", |
e4298d25 DB |
5142 | tn_buf, off, size); |
5143 | return -EACCES; | |
5144 | } | |
01f810ac AM |
5145 | /* Variable offset is prohibited for unprivileged mode for simplicity |
5146 | * since it requires corresponding support in Spectre masking for stack | |
082cdc69 LG |
5147 | * ALU. See also retrieve_ptr_limit(). The check in |
5148 | * check_stack_access_for_ptr_arithmetic() called by | |
5149 | * adjust_ptr_min_max_vals() prevents users from creating stack pointers | |
5150 | * with variable offsets, therefore no check is required here. Further, | |
5151 | * just checking it here would be insufficient as speculative stack | |
5152 | * writes could still lead to unsafe speculative behaviour. | |
01f810ac | 5153 | */ |
01f810ac AM |
5154 | if (!var_off) { |
5155 | off += reg->var_off.value; | |
5156 | err = check_stack_read_fixed_off(env, state, off, size, | |
5157 | dst_regno); | |
5158 | } else { | |
5159 | /* Variable offset stack reads need more conservative handling | |
5160 | * than fixed offset ones. Note that dst_regno >= 0 on this | |
5161 | * branch. | |
5162 | */ | |
5163 | err = check_stack_read_var_off(env, ptr_regno, off, size, | |
5164 | dst_regno); | |
5165 | } | |
5166 | return err; | |
5167 | } | |
5168 | ||
5169 | ||
5170 | /* check_stack_write dispatches to check_stack_write_fixed_off or | |
5171 | * check_stack_write_var_off. | |
5172 | * | |
5173 | * 'ptr_regno' is the register used as a pointer into the stack. | |
5174 | * 'off' includes 'ptr_regno->off', but not its variable offset (if any). | |
5175 | * 'value_regno' is the register whose value we're writing to the stack. It can | |
5176 | * be -1, meaning that we're not writing from a register. | |
5177 | * | |
5178 | * The caller must ensure that the offset falls within the maximum stack size. | |
5179 | */ | |
5180 | static int check_stack_write(struct bpf_verifier_env *env, | |
5181 | int ptr_regno, int off, int size, | |
5182 | int value_regno, int insn_idx) | |
5183 | { | |
5184 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
5185 | struct bpf_func_state *state = func(env, reg); | |
5186 | int err; | |
5187 | ||
5188 | if (tnum_is_const(reg->var_off)) { | |
5189 | off += reg->var_off.value; | |
5190 | err = check_stack_write_fixed_off(env, state, off, size, | |
5191 | value_regno, insn_idx); | |
5192 | } else { | |
5193 | /* Variable offset stack reads need more conservative handling | |
5194 | * than fixed offset ones. | |
5195 | */ | |
5196 | err = check_stack_write_var_off(env, state, | |
5197 | ptr_regno, off, size, | |
5198 | value_regno, insn_idx); | |
5199 | } | |
5200 | return err; | |
e4298d25 DB |
5201 | } |
5202 | ||
591fe988 DB |
5203 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
5204 | int off, int size, enum bpf_access_type type) | |
5205 | { | |
5206 | struct bpf_reg_state *regs = cur_regs(env); | |
5207 | struct bpf_map *map = regs[regno].map_ptr; | |
5208 | u32 cap = bpf_map_flags_to_cap(map); | |
5209 | ||
5210 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
5211 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
5212 | map->value_size, off, size); | |
5213 | return -EACCES; | |
5214 | } | |
5215 | ||
5216 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
5217 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
5218 | map->value_size, off, size); | |
5219 | return -EACCES; | |
5220 | } | |
5221 | ||
5222 | return 0; | |
5223 | } | |
5224 | ||
457f4436 AN |
5225 | /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */ |
5226 | static int __check_mem_access(struct bpf_verifier_env *env, int regno, | |
5227 | int off, int size, u32 mem_size, | |
5228 | bool zero_size_allowed) | |
17a52670 | 5229 | { |
457f4436 AN |
5230 | bool size_ok = size > 0 || (size == 0 && zero_size_allowed); |
5231 | struct bpf_reg_state *reg; | |
5232 | ||
5233 | if (off >= 0 && size_ok && (u64)off + size <= mem_size) | |
5234 | return 0; | |
17a52670 | 5235 | |
457f4436 AN |
5236 | reg = &cur_regs(env)[regno]; |
5237 | switch (reg->type) { | |
69c087ba YS |
5238 | case PTR_TO_MAP_KEY: |
5239 | verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n", | |
5240 | mem_size, off, size); | |
5241 | break; | |
457f4436 | 5242 | case PTR_TO_MAP_VALUE: |
61bd5218 | 5243 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
457f4436 AN |
5244 | mem_size, off, size); |
5245 | break; | |
5246 | case PTR_TO_PACKET: | |
5247 | case PTR_TO_PACKET_META: | |
5248 | case PTR_TO_PACKET_END: | |
5249 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", | |
5250 | off, size, regno, reg->id, off, mem_size); | |
5251 | break; | |
5252 | case PTR_TO_MEM: | |
5253 | default: | |
5254 | verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n", | |
5255 | mem_size, off, size); | |
17a52670 | 5256 | } |
457f4436 AN |
5257 | |
5258 | return -EACCES; | |
17a52670 AS |
5259 | } |
5260 | ||
457f4436 AN |
5261 | /* check read/write into a memory region with possible variable offset */ |
5262 | static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno, | |
5263 | int off, int size, u32 mem_size, | |
5264 | bool zero_size_allowed) | |
dbcfe5f7 | 5265 | { |
f4d7e40a AS |
5266 | struct bpf_verifier_state *vstate = env->cur_state; |
5267 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
5268 | struct bpf_reg_state *reg = &state->regs[regno]; |
5269 | int err; | |
5270 | ||
457f4436 | 5271 | /* We may have adjusted the register pointing to memory region, so we |
f1174f77 EC |
5272 | * need to try adding each of min_value and max_value to off |
5273 | * to make sure our theoretical access will be safe. | |
2e576648 CL |
5274 | * |
5275 | * The minimum value is only important with signed | |
dbcfe5f7 GB |
5276 | * comparisons where we can't assume the floor of a |
5277 | * value is 0. If we are using signed variables for our | |
5278 | * index'es we need to make sure that whatever we use | |
5279 | * will have a set floor within our range. | |
5280 | */ | |
b7137c4e DB |
5281 | if (reg->smin_value < 0 && |
5282 | (reg->smin_value == S64_MIN || | |
5283 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
5284 | reg->smin_value + off < 0)) { | |
61bd5218 | 5285 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
5286 | regno); |
5287 | return -EACCES; | |
5288 | } | |
457f4436 AN |
5289 | err = __check_mem_access(env, regno, reg->smin_value + off, size, |
5290 | mem_size, zero_size_allowed); | |
dbcfe5f7 | 5291 | if (err) { |
457f4436 | 5292 | verbose(env, "R%d min value is outside of the allowed memory range\n", |
61bd5218 | 5293 | regno); |
dbcfe5f7 GB |
5294 | return err; |
5295 | } | |
5296 | ||
b03c9f9f EC |
5297 | /* If we haven't set a max value then we need to bail since we can't be |
5298 | * sure we won't do bad things. | |
5299 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 5300 | */ |
b03c9f9f | 5301 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
457f4436 | 5302 | verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n", |
dbcfe5f7 GB |
5303 | regno); |
5304 | return -EACCES; | |
5305 | } | |
457f4436 AN |
5306 | err = __check_mem_access(env, regno, reg->umax_value + off, size, |
5307 | mem_size, zero_size_allowed); | |
5308 | if (err) { | |
5309 | verbose(env, "R%d max value is outside of the allowed memory range\n", | |
61bd5218 | 5310 | regno); |
457f4436 AN |
5311 | return err; |
5312 | } | |
5313 | ||
5314 | return 0; | |
5315 | } | |
d83525ca | 5316 | |
e9147b44 KKD |
5317 | static int __check_ptr_off_reg(struct bpf_verifier_env *env, |
5318 | const struct bpf_reg_state *reg, int regno, | |
5319 | bool fixed_off_ok) | |
5320 | { | |
5321 | /* Access to this pointer-typed register or passing it to a helper | |
5322 | * is only allowed in its original, unmodified form. | |
5323 | */ | |
5324 | ||
5325 | if (reg->off < 0) { | |
5326 | verbose(env, "negative offset %s ptr R%d off=%d disallowed\n", | |
5327 | reg_type_str(env, reg->type), regno, reg->off); | |
5328 | return -EACCES; | |
5329 | } | |
5330 | ||
5331 | if (!fixed_off_ok && reg->off) { | |
5332 | verbose(env, "dereference of modified %s ptr R%d off=%d disallowed\n", | |
5333 | reg_type_str(env, reg->type), regno, reg->off); | |
5334 | return -EACCES; | |
5335 | } | |
5336 | ||
5337 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
5338 | char tn_buf[48]; | |
5339 | ||
5340 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
5341 | verbose(env, "variable %s access var_off=%s disallowed\n", | |
5342 | reg_type_str(env, reg->type), tn_buf); | |
5343 | return -EACCES; | |
5344 | } | |
5345 | ||
5346 | return 0; | |
5347 | } | |
5348 | ||
5349 | int check_ptr_off_reg(struct bpf_verifier_env *env, | |
5350 | const struct bpf_reg_state *reg, int regno) | |
5351 | { | |
5352 | return __check_ptr_off_reg(env, reg, regno, false); | |
5353 | } | |
5354 | ||
61df10c7 | 5355 | static int map_kptr_match_type(struct bpf_verifier_env *env, |
aa3496ac | 5356 | struct btf_field *kptr_field, |
61df10c7 KKD |
5357 | struct bpf_reg_state *reg, u32 regno) |
5358 | { | |
b32a5dae | 5359 | const char *targ_name = btf_type_name(kptr_field->kptr.btf, kptr_field->kptr.btf_id); |
ab6c637a | 5360 | int perm_flags; |
61df10c7 KKD |
5361 | const char *reg_name = ""; |
5362 | ||
ab6c637a YS |
5363 | if (btf_is_kernel(reg->btf)) { |
5364 | perm_flags = PTR_MAYBE_NULL | PTR_TRUSTED | MEM_RCU; | |
5365 | ||
5366 | /* Only unreferenced case accepts untrusted pointers */ | |
5367 | if (kptr_field->type == BPF_KPTR_UNREF) | |
5368 | perm_flags |= PTR_UNTRUSTED; | |
5369 | } else { | |
5370 | perm_flags = PTR_MAYBE_NULL | MEM_ALLOC; | |
36d8bdf7 YS |
5371 | if (kptr_field->type == BPF_KPTR_PERCPU) |
5372 | perm_flags |= MEM_PERCPU; | |
ab6c637a | 5373 | } |
6efe152d KKD |
5374 | |
5375 | if (base_type(reg->type) != PTR_TO_BTF_ID || (type_flag(reg->type) & ~perm_flags)) | |
61df10c7 KKD |
5376 | goto bad_type; |
5377 | ||
61df10c7 | 5378 | /* We need to verify reg->type and reg->btf, before accessing reg->btf */ |
b32a5dae | 5379 | reg_name = btf_type_name(reg->btf, reg->btf_id); |
61df10c7 | 5380 | |
c0a5a21c KKD |
5381 | /* For ref_ptr case, release function check should ensure we get one |
5382 | * referenced PTR_TO_BTF_ID, and that its fixed offset is 0. For the | |
5383 | * normal store of unreferenced kptr, we must ensure var_off is zero. | |
5384 | * Since ref_ptr cannot be accessed directly by BPF insns, checks for | |
5385 | * reg->off and reg->ref_obj_id are not needed here. | |
5386 | */ | |
61df10c7 KKD |
5387 | if (__check_ptr_off_reg(env, reg, regno, true)) |
5388 | return -EACCES; | |
5389 | ||
ab6c637a | 5390 | /* A full type match is needed, as BTF can be vmlinux, module or prog BTF, and |
61df10c7 KKD |
5391 | * we also need to take into account the reg->off. |
5392 | * | |
5393 | * We want to support cases like: | |
5394 | * | |
5395 | * struct foo { | |
5396 | * struct bar br; | |
5397 | * struct baz bz; | |
5398 | * }; | |
5399 | * | |
5400 | * struct foo *v; | |
5401 | * v = func(); // PTR_TO_BTF_ID | |
5402 | * val->foo = v; // reg->off is zero, btf and btf_id match type | |
5403 | * val->bar = &v->br; // reg->off is still zero, but we need to retry with | |
5404 | * // first member type of struct after comparison fails | |
5405 | * val->baz = &v->bz; // reg->off is non-zero, so struct needs to be walked | |
5406 | * // to match type | |
5407 | * | |
5408 | * In the kptr_ref case, check_func_arg_reg_off already ensures reg->off | |
2ab3b380 KKD |
5409 | * is zero. We must also ensure that btf_struct_ids_match does not walk |
5410 | * the struct to match type against first member of struct, i.e. reject | |
5411 | * second case from above. Hence, when type is BPF_KPTR_REF, we set | |
5412 | * strict mode to true for type match. | |
61df10c7 KKD |
5413 | */ |
5414 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
aa3496ac | 5415 | kptr_field->kptr.btf, kptr_field->kptr.btf_id, |
36d8bdf7 | 5416 | kptr_field->type != BPF_KPTR_UNREF)) |
61df10c7 KKD |
5417 | goto bad_type; |
5418 | return 0; | |
5419 | bad_type: | |
5420 | verbose(env, "invalid kptr access, R%d type=%s%s ", regno, | |
5421 | reg_type_str(env, reg->type), reg_name); | |
6efe152d | 5422 | verbose(env, "expected=%s%s", reg_type_str(env, PTR_TO_BTF_ID), targ_name); |
aa3496ac | 5423 | if (kptr_field->type == BPF_KPTR_UNREF) |
6efe152d KKD |
5424 | verbose(env, " or %s%s\n", reg_type_str(env, PTR_TO_BTF_ID | PTR_UNTRUSTED), |
5425 | targ_name); | |
5426 | else | |
5427 | verbose(env, "\n"); | |
61df10c7 KKD |
5428 | return -EINVAL; |
5429 | } | |
5430 | ||
20c09d92 AS |
5431 | /* The non-sleepable programs and sleepable programs with explicit bpf_rcu_read_lock() |
5432 | * can dereference RCU protected pointers and result is PTR_TRUSTED. | |
5433 | */ | |
5434 | static bool in_rcu_cs(struct bpf_verifier_env *env) | |
5435 | { | |
5861d1e8 DM |
5436 | return env->cur_state->active_rcu_lock || |
5437 | env->cur_state->active_lock.ptr || | |
5438 | !env->prog->aux->sleepable; | |
20c09d92 AS |
5439 | } |
5440 | ||
5441 | /* Once GCC supports btf_type_tag the following mechanism will be replaced with tag check */ | |
5442 | BTF_SET_START(rcu_protected_types) | |
5443 | BTF_ID(struct, prog_test_ref_kfunc) | |
05670f81 | 5444 | #ifdef CONFIG_CGROUPS |
20c09d92 | 5445 | BTF_ID(struct, cgroup) |
05670f81 | 5446 | #endif |
63d2d83d | 5447 | BTF_ID(struct, bpf_cpumask) |
d02c48fa | 5448 | BTF_ID(struct, task_struct) |
20c09d92 AS |
5449 | BTF_SET_END(rcu_protected_types) |
5450 | ||
5451 | static bool rcu_protected_object(const struct btf *btf, u32 btf_id) | |
5452 | { | |
5453 | if (!btf_is_kernel(btf)) | |
5454 | return false; | |
5455 | return btf_id_set_contains(&rcu_protected_types, btf_id); | |
5456 | } | |
5457 | ||
5458 | static bool rcu_safe_kptr(const struct btf_field *field) | |
5459 | { | |
5460 | const struct btf_field_kptr *kptr = &field->kptr; | |
5461 | ||
36d8bdf7 YS |
5462 | return field->type == BPF_KPTR_PERCPU || |
5463 | (field->type == BPF_KPTR_REF && rcu_protected_object(kptr->btf, kptr->btf_id)); | |
5464 | } | |
5465 | ||
5466 | static u32 btf_ld_kptr_type(struct bpf_verifier_env *env, struct btf_field *kptr_field) | |
5467 | { | |
5468 | if (rcu_safe_kptr(kptr_field) && in_rcu_cs(env)) { | |
5469 | if (kptr_field->type != BPF_KPTR_PERCPU) | |
5470 | return PTR_MAYBE_NULL | MEM_RCU; | |
5471 | return PTR_MAYBE_NULL | MEM_RCU | MEM_PERCPU; | |
5472 | } | |
5473 | return PTR_MAYBE_NULL | PTR_UNTRUSTED; | |
20c09d92 AS |
5474 | } |
5475 | ||
61df10c7 KKD |
5476 | static int check_map_kptr_access(struct bpf_verifier_env *env, u32 regno, |
5477 | int value_regno, int insn_idx, | |
aa3496ac | 5478 | struct btf_field *kptr_field) |
61df10c7 KKD |
5479 | { |
5480 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; | |
5481 | int class = BPF_CLASS(insn->code); | |
5482 | struct bpf_reg_state *val_reg; | |
5483 | ||
5484 | /* Things we already checked for in check_map_access and caller: | |
5485 | * - Reject cases where variable offset may touch kptr | |
5486 | * - size of access (must be BPF_DW) | |
5487 | * - tnum_is_const(reg->var_off) | |
aa3496ac | 5488 | * - kptr_field->offset == off + reg->var_off.value |
61df10c7 KKD |
5489 | */ |
5490 | /* Only BPF_[LDX,STX,ST] | BPF_MEM | BPF_DW is supported */ | |
5491 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
5492 | verbose(env, "kptr in map can only be accessed using BPF_MEM instruction mode\n"); | |
5493 | return -EACCES; | |
5494 | } | |
5495 | ||
6efe152d KKD |
5496 | /* We only allow loading referenced kptr, since it will be marked as |
5497 | * untrusted, similar to unreferenced kptr. | |
5498 | */ | |
36d8bdf7 YS |
5499 | if (class != BPF_LDX && |
5500 | (kptr_field->type == BPF_KPTR_REF || kptr_field->type == BPF_KPTR_PERCPU)) { | |
6efe152d | 5501 | verbose(env, "store to referenced kptr disallowed\n"); |
c0a5a21c KKD |
5502 | return -EACCES; |
5503 | } | |
5504 | ||
61df10c7 KKD |
5505 | if (class == BPF_LDX) { |
5506 | val_reg = reg_state(env, value_regno); | |
5507 | /* We can simply mark the value_regno receiving the pointer | |
5508 | * value from map as PTR_TO_BTF_ID, with the correct type. | |
5509 | */ | |
aa3496ac | 5510 | mark_btf_ld_reg(env, cur_regs(env), value_regno, PTR_TO_BTF_ID, kptr_field->kptr.btf, |
36d8bdf7 | 5511 | kptr_field->kptr.btf_id, btf_ld_kptr_type(env, kptr_field)); |
61df10c7 KKD |
5512 | /* For mark_ptr_or_null_reg */ |
5513 | val_reg->id = ++env->id_gen; | |
5514 | } else if (class == BPF_STX) { | |
5515 | val_reg = reg_state(env, value_regno); | |
5516 | if (!register_is_null(val_reg) && | |
aa3496ac | 5517 | map_kptr_match_type(env, kptr_field, val_reg, value_regno)) |
61df10c7 KKD |
5518 | return -EACCES; |
5519 | } else if (class == BPF_ST) { | |
5520 | if (insn->imm) { | |
5521 | verbose(env, "BPF_ST imm must be 0 when storing to kptr at off=%u\n", | |
aa3496ac | 5522 | kptr_field->offset); |
61df10c7 KKD |
5523 | return -EACCES; |
5524 | } | |
5525 | } else { | |
5526 | verbose(env, "kptr in map can only be accessed using BPF_LDX/BPF_STX/BPF_ST\n"); | |
5527 | return -EACCES; | |
5528 | } | |
5529 | return 0; | |
5530 | } | |
5531 | ||
457f4436 AN |
5532 | /* check read/write into a map element with possible variable offset */ |
5533 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
61df10c7 KKD |
5534 | int off, int size, bool zero_size_allowed, |
5535 | enum bpf_access_src src) | |
457f4436 AN |
5536 | { |
5537 | struct bpf_verifier_state *vstate = env->cur_state; | |
5538 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5539 | struct bpf_reg_state *reg = &state->regs[regno]; | |
5540 | struct bpf_map *map = reg->map_ptr; | |
aa3496ac KKD |
5541 | struct btf_record *rec; |
5542 | int err, i; | |
457f4436 AN |
5543 | |
5544 | err = check_mem_region_access(env, regno, off, size, map->value_size, | |
5545 | zero_size_allowed); | |
5546 | if (err) | |
5547 | return err; | |
5548 | ||
aa3496ac KKD |
5549 | if (IS_ERR_OR_NULL(map->record)) |
5550 | return 0; | |
5551 | rec = map->record; | |
5552 | for (i = 0; i < rec->cnt; i++) { | |
5553 | struct btf_field *field = &rec->fields[i]; | |
5554 | u32 p = field->offset; | |
d83525ca | 5555 | |
db559117 KKD |
5556 | /* If any part of a field can be touched by load/store, reject |
5557 | * this program. To check that [x1, x2) overlaps with [y1, y2), | |
d83525ca AS |
5558 | * it is sufficient to check x1 < y2 && y1 < x2. |
5559 | */ | |
aa3496ac KKD |
5560 | if (reg->smin_value + off < p + btf_field_type_size(field->type) && |
5561 | p < reg->umax_value + off + size) { | |
5562 | switch (field->type) { | |
5563 | case BPF_KPTR_UNREF: | |
5564 | case BPF_KPTR_REF: | |
36d8bdf7 | 5565 | case BPF_KPTR_PERCPU: |
61df10c7 KKD |
5566 | if (src != ACCESS_DIRECT) { |
5567 | verbose(env, "kptr cannot be accessed indirectly by helper\n"); | |
5568 | return -EACCES; | |
5569 | } | |
5570 | if (!tnum_is_const(reg->var_off)) { | |
5571 | verbose(env, "kptr access cannot have variable offset\n"); | |
5572 | return -EACCES; | |
5573 | } | |
5574 | if (p != off + reg->var_off.value) { | |
5575 | verbose(env, "kptr access misaligned expected=%u off=%llu\n", | |
5576 | p, off + reg->var_off.value); | |
5577 | return -EACCES; | |
5578 | } | |
5579 | if (size != bpf_size_to_bytes(BPF_DW)) { | |
5580 | verbose(env, "kptr access size must be BPF_DW\n"); | |
5581 | return -EACCES; | |
5582 | } | |
5583 | break; | |
aa3496ac | 5584 | default: |
db559117 KKD |
5585 | verbose(env, "%s cannot be accessed directly by load/store\n", |
5586 | btf_field_type_name(field->type)); | |
aa3496ac | 5587 | return -EACCES; |
61df10c7 KKD |
5588 | } |
5589 | } | |
5590 | } | |
aa3496ac | 5591 | return 0; |
dbcfe5f7 GB |
5592 | } |
5593 | ||
969bf05e AS |
5594 | #define MAX_PACKET_OFF 0xffff |
5595 | ||
58e2af8b | 5596 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
5597 | const struct bpf_call_arg_meta *meta, |
5598 | enum bpf_access_type t) | |
4acf6c0b | 5599 | { |
7e40781c UP |
5600 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
5601 | ||
5602 | switch (prog_type) { | |
5d66fa7d | 5603 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
5604 | case BPF_PROG_TYPE_LWT_IN: |
5605 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 5606 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 5607 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 5608 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 5609 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
5610 | if (t == BPF_WRITE) |
5611 | return false; | |
8731745e | 5612 | fallthrough; |
5d66fa7d DB |
5613 | |
5614 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
5615 | case BPF_PROG_TYPE_SCHED_CLS: |
5616 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 5617 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 5618 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 5619 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 5620 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
5621 | if (meta) |
5622 | return meta->pkt_access; | |
5623 | ||
5624 | env->seen_direct_write = true; | |
4acf6c0b | 5625 | return true; |
0d01da6a SF |
5626 | |
5627 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
5628 | if (t == BPF_WRITE) | |
5629 | env->seen_direct_write = true; | |
5630 | ||
5631 | return true; | |
5632 | ||
4acf6c0b BB |
5633 | default: |
5634 | return false; | |
5635 | } | |
5636 | } | |
5637 | ||
f1174f77 | 5638 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 5639 | int size, bool zero_size_allowed) |
f1174f77 | 5640 | { |
638f5b90 | 5641 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
5642 | struct bpf_reg_state *reg = ®s[regno]; |
5643 | int err; | |
5644 | ||
5645 | /* We may have added a variable offset to the packet pointer; but any | |
5646 | * reg->range we have comes after that. We are only checking the fixed | |
5647 | * offset. | |
5648 | */ | |
5649 | ||
5650 | /* We don't allow negative numbers, because we aren't tracking enough | |
5651 | * detail to prove they're safe. | |
5652 | */ | |
b03c9f9f | 5653 | if (reg->smin_value < 0) { |
61bd5218 | 5654 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
5655 | regno); |
5656 | return -EACCES; | |
5657 | } | |
6d94e741 AS |
5658 | |
5659 | err = reg->range < 0 ? -EINVAL : | |
5660 | __check_mem_access(env, regno, off, size, reg->range, | |
457f4436 | 5661 | zero_size_allowed); |
f1174f77 | 5662 | if (err) { |
61bd5218 | 5663 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
5664 | return err; |
5665 | } | |
e647815a | 5666 | |
457f4436 | 5667 | /* __check_mem_access has made sure "off + size - 1" is within u16. |
e647815a JW |
5668 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, |
5669 | * otherwise find_good_pkt_pointers would have refused to set range info | |
457f4436 | 5670 | * that __check_mem_access would have rejected this pkt access. |
e647815a JW |
5671 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. |
5672 | */ | |
5673 | env->prog->aux->max_pkt_offset = | |
5674 | max_t(u32, env->prog->aux->max_pkt_offset, | |
5675 | off + reg->umax_value + size - 1); | |
5676 | ||
f1174f77 EC |
5677 | return err; |
5678 | } | |
5679 | ||
5680 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 5681 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 | 5682 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
22dc4a0f | 5683 | struct btf **btf, u32 *btf_id) |
17a52670 | 5684 | { |
f96da094 DB |
5685 | struct bpf_insn_access_aux info = { |
5686 | .reg_type = *reg_type, | |
9e15db66 | 5687 | .log = &env->log, |
f96da094 | 5688 | }; |
31fd8581 | 5689 | |
4f9218aa | 5690 | if (env->ops->is_valid_access && |
5e43f899 | 5691 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
5692 | /* A non zero info.ctx_field_size indicates that this field is a |
5693 | * candidate for later verifier transformation to load the whole | |
5694 | * field and then apply a mask when accessed with a narrower | |
5695 | * access than actual ctx access size. A zero info.ctx_field_size | |
5696 | * will only allow for whole field access and rejects any other | |
5697 | * type of narrower access. | |
31fd8581 | 5698 | */ |
23994631 | 5699 | *reg_type = info.reg_type; |
31fd8581 | 5700 | |
c25b2ae1 | 5701 | if (base_type(*reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 5702 | *btf = info.btf; |
9e15db66 | 5703 | *btf_id = info.btf_id; |
22dc4a0f | 5704 | } else { |
9e15db66 | 5705 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
22dc4a0f | 5706 | } |
32bbe007 AS |
5707 | /* remember the offset of last byte accessed in ctx */ |
5708 | if (env->prog->aux->max_ctx_offset < off + size) | |
5709 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 5710 | return 0; |
32bbe007 | 5711 | } |
17a52670 | 5712 | |
61bd5218 | 5713 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
5714 | return -EACCES; |
5715 | } | |
5716 | ||
d58e468b PP |
5717 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
5718 | int size) | |
5719 | { | |
5720 | if (size < 0 || off < 0 || | |
5721 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
5722 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
5723 | off, size); | |
5724 | return -EACCES; | |
5725 | } | |
5726 | return 0; | |
5727 | } | |
5728 | ||
5f456649 MKL |
5729 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
5730 | u32 regno, int off, int size, | |
5731 | enum bpf_access_type t) | |
c64b7983 JS |
5732 | { |
5733 | struct bpf_reg_state *regs = cur_regs(env); | |
5734 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 5735 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 5736 | bool valid; |
c64b7983 JS |
5737 | |
5738 | if (reg->smin_value < 0) { | |
5739 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
5740 | regno); | |
5741 | return -EACCES; | |
5742 | } | |
5743 | ||
46f8bc92 MKL |
5744 | switch (reg->type) { |
5745 | case PTR_TO_SOCK_COMMON: | |
5746 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
5747 | break; | |
5748 | case PTR_TO_SOCKET: | |
5749 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
5750 | break; | |
655a51e5 MKL |
5751 | case PTR_TO_TCP_SOCK: |
5752 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
5753 | break; | |
fada7fdc JL |
5754 | case PTR_TO_XDP_SOCK: |
5755 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
5756 | break; | |
46f8bc92 MKL |
5757 | default: |
5758 | valid = false; | |
c64b7983 JS |
5759 | } |
5760 | ||
5f456649 | 5761 | |
46f8bc92 MKL |
5762 | if (valid) { |
5763 | env->insn_aux_data[insn_idx].ctx_field_size = | |
5764 | info.ctx_field_size; | |
5765 | return 0; | |
5766 | } | |
5767 | ||
5768 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
c25b2ae1 | 5769 | regno, reg_type_str(env, reg->type), off, size); |
46f8bc92 MKL |
5770 | |
5771 | return -EACCES; | |
c64b7983 JS |
5772 | } |
5773 | ||
4cabc5b1 DB |
5774 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
5775 | { | |
2a159c6f | 5776 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
5777 | } |
5778 | ||
f37a8cb8 DB |
5779 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
5780 | { | |
2a159c6f | 5781 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 5782 | |
46f8bc92 MKL |
5783 | return reg->type == PTR_TO_CTX; |
5784 | } | |
5785 | ||
5786 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
5787 | { | |
5788 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
5789 | ||
5790 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
5791 | } |
5792 | ||
ca369602 DB |
5793 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
5794 | { | |
2a159c6f | 5795 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
5796 | |
5797 | return type_is_pkt_pointer(reg->type); | |
5798 | } | |
5799 | ||
4b5defde DB |
5800 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
5801 | { | |
5802 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
5803 | ||
5804 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
5805 | return reg->type == PTR_TO_FLOW_KEYS; | |
5806 | } | |
5807 | ||
831deb29 AP |
5808 | static u32 *reg2btf_ids[__BPF_REG_TYPE_MAX] = { |
5809 | #ifdef CONFIG_NET | |
5810 | [PTR_TO_SOCKET] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK], | |
5811 | [PTR_TO_SOCK_COMMON] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
5812 | [PTR_TO_TCP_SOCK] = &btf_sock_ids[BTF_SOCK_TYPE_TCP], | |
5813 | #endif | |
5ba190c2 | 5814 | [CONST_PTR_TO_MAP] = btf_bpf_map_id, |
831deb29 AP |
5815 | }; |
5816 | ||
9bb00b28 YS |
5817 | static bool is_trusted_reg(const struct bpf_reg_state *reg) |
5818 | { | |
5819 | /* A referenced register is always trusted. */ | |
5820 | if (reg->ref_obj_id) | |
5821 | return true; | |
5822 | ||
831deb29 AP |
5823 | /* Types listed in the reg2btf_ids are always trusted */ |
5824 | if (reg2btf_ids[base_type(reg->type)]) | |
5825 | return true; | |
5826 | ||
9bb00b28 | 5827 | /* If a register is not referenced, it is trusted if it has the |
fca1aa75 | 5828 | * MEM_ALLOC or PTR_TRUSTED type modifiers, and no others. Some of the |
9bb00b28 YS |
5829 | * other type modifiers may be safe, but we elect to take an opt-in |
5830 | * approach here as some (e.g. PTR_UNTRUSTED and PTR_MAYBE_NULL) are | |
5831 | * not. | |
5832 | * | |
5833 | * Eventually, we should make PTR_TRUSTED the single source of truth | |
5834 | * for whether a register is trusted. | |
5835 | */ | |
5836 | return type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS && | |
5837 | !bpf_type_has_unsafe_modifiers(reg->type); | |
5838 | } | |
5839 | ||
fca1aa75 YS |
5840 | static bool is_rcu_reg(const struct bpf_reg_state *reg) |
5841 | { | |
5842 | return reg->type & MEM_RCU; | |
5843 | } | |
5844 | ||
afeebf9f AS |
5845 | static void clear_trusted_flags(enum bpf_type_flag *flag) |
5846 | { | |
5847 | *flag &= ~(BPF_REG_TRUSTED_MODIFIERS | MEM_RCU); | |
5848 | } | |
5849 | ||
61bd5218 JK |
5850 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
5851 | const struct bpf_reg_state *reg, | |
d1174416 | 5852 | int off, int size, bool strict) |
969bf05e | 5853 | { |
f1174f77 | 5854 | struct tnum reg_off; |
e07b98d9 | 5855 | int ip_align; |
d1174416 DM |
5856 | |
5857 | /* Byte size accesses are always allowed. */ | |
5858 | if (!strict || size == 1) | |
5859 | return 0; | |
5860 | ||
e4eda884 DM |
5861 | /* For platforms that do not have a Kconfig enabling |
5862 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
5863 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
5864 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
5865 | * to this code only in strict mode where we want to emulate | |
5866 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
5867 | * unconditional IP align value of '2'. | |
e07b98d9 | 5868 | */ |
e4eda884 | 5869 | ip_align = 2; |
f1174f77 EC |
5870 | |
5871 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
5872 | if (!tnum_is_aligned(reg_off, size)) { | |
5873 | char tn_buf[48]; | |
5874 | ||
5875 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
5876 | verbose(env, |
5877 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 5878 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
5879 | return -EACCES; |
5880 | } | |
79adffcd | 5881 | |
969bf05e AS |
5882 | return 0; |
5883 | } | |
5884 | ||
61bd5218 JK |
5885 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
5886 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
5887 | const char *pointer_desc, |
5888 | int off, int size, bool strict) | |
79adffcd | 5889 | { |
f1174f77 EC |
5890 | struct tnum reg_off; |
5891 | ||
5892 | /* Byte size accesses are always allowed. */ | |
5893 | if (!strict || size == 1) | |
5894 | return 0; | |
5895 | ||
5896 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
5897 | if (!tnum_is_aligned(reg_off, size)) { | |
5898 | char tn_buf[48]; | |
5899 | ||
5900 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 5901 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 5902 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
5903 | return -EACCES; |
5904 | } | |
5905 | ||
969bf05e AS |
5906 | return 0; |
5907 | } | |
5908 | ||
e07b98d9 | 5909 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
5910 | const struct bpf_reg_state *reg, int off, |
5911 | int size, bool strict_alignment_once) | |
79adffcd | 5912 | { |
ca369602 | 5913 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 5914 | const char *pointer_desc = ""; |
d1174416 | 5915 | |
79adffcd DB |
5916 | switch (reg->type) { |
5917 | case PTR_TO_PACKET: | |
de8f3a83 DB |
5918 | case PTR_TO_PACKET_META: |
5919 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
5920 | * right in front, treat it the very same way. | |
5921 | */ | |
61bd5218 | 5922 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
5923 | case PTR_TO_FLOW_KEYS: |
5924 | pointer_desc = "flow keys "; | |
5925 | break; | |
69c087ba YS |
5926 | case PTR_TO_MAP_KEY: |
5927 | pointer_desc = "key "; | |
5928 | break; | |
f1174f77 EC |
5929 | case PTR_TO_MAP_VALUE: |
5930 | pointer_desc = "value "; | |
5931 | break; | |
5932 | case PTR_TO_CTX: | |
5933 | pointer_desc = "context "; | |
5934 | break; | |
5935 | case PTR_TO_STACK: | |
5936 | pointer_desc = "stack "; | |
01f810ac AM |
5937 | /* The stack spill tracking logic in check_stack_write_fixed_off() |
5938 | * and check_stack_read_fixed_off() relies on stack accesses being | |
a5ec6ae1 JH |
5939 | * aligned. |
5940 | */ | |
5941 | strict = true; | |
f1174f77 | 5942 | break; |
c64b7983 JS |
5943 | case PTR_TO_SOCKET: |
5944 | pointer_desc = "sock "; | |
5945 | break; | |
46f8bc92 MKL |
5946 | case PTR_TO_SOCK_COMMON: |
5947 | pointer_desc = "sock_common "; | |
5948 | break; | |
655a51e5 MKL |
5949 | case PTR_TO_TCP_SOCK: |
5950 | pointer_desc = "tcp_sock "; | |
5951 | break; | |
fada7fdc JL |
5952 | case PTR_TO_XDP_SOCK: |
5953 | pointer_desc = "xdp_sock "; | |
5954 | break; | |
79adffcd | 5955 | default: |
f1174f77 | 5956 | break; |
79adffcd | 5957 | } |
61bd5218 JK |
5958 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
5959 | strict); | |
79adffcd DB |
5960 | } |
5961 | ||
f4d7e40a AS |
5962 | static int update_stack_depth(struct bpf_verifier_env *env, |
5963 | const struct bpf_func_state *func, | |
5964 | int off) | |
5965 | { | |
9c8105bd | 5966 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
5967 | |
5968 | if (stack >= -off) | |
5969 | return 0; | |
5970 | ||
5971 | /* update known max for given subprogram */ | |
9c8105bd | 5972 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
5973 | return 0; |
5974 | } | |
f4d7e40a | 5975 | |
70a87ffe AS |
5976 | /* starting from main bpf function walk all instructions of the function |
5977 | * and recursively walk all callees that given function can call. | |
5978 | * Ignore jump and exit insns. | |
5979 | * Since recursion is prevented by check_cfg() this algorithm | |
5980 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
5981 | */ | |
b5e9ad52 | 5982 | static int check_max_stack_depth_subprog(struct bpf_verifier_env *env, int idx) |
70a87ffe | 5983 | { |
9c8105bd | 5984 | struct bpf_subprog_info *subprog = env->subprog_info; |
70a87ffe | 5985 | struct bpf_insn *insn = env->prog->insnsi; |
b5e9ad52 | 5986 | int depth = 0, frame = 0, i, subprog_end; |
ebf7d1f5 | 5987 | bool tail_call_reachable = false; |
70a87ffe AS |
5988 | int ret_insn[MAX_CALL_FRAMES]; |
5989 | int ret_prog[MAX_CALL_FRAMES]; | |
ebf7d1f5 | 5990 | int j; |
f4d7e40a | 5991 | |
b5e9ad52 | 5992 | i = subprog[idx].start; |
70a87ffe | 5993 | process_func: |
7f6e4312 MF |
5994 | /* protect against potential stack overflow that might happen when |
5995 | * bpf2bpf calls get combined with tailcalls. Limit the caller's stack | |
5996 | * depth for such case down to 256 so that the worst case scenario | |
5997 | * would result in 8k stack size (32 which is tailcall limit * 256 = | |
5998 | * 8k). | |
5999 | * | |
6000 | * To get the idea what might happen, see an example: | |
6001 | * func1 -> sub rsp, 128 | |
6002 | * subfunc1 -> sub rsp, 256 | |
6003 | * tailcall1 -> add rsp, 256 | |
6004 | * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320) | |
6005 | * subfunc2 -> sub rsp, 64 | |
6006 | * subfunc22 -> sub rsp, 128 | |
6007 | * tailcall2 -> add rsp, 128 | |
6008 | * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416) | |
6009 | * | |
6010 | * tailcall will unwind the current stack frame but it will not get rid | |
6011 | * of caller's stack as shown on the example above. | |
6012 | */ | |
6013 | if (idx && subprog[idx].has_tail_call && depth >= 256) { | |
6014 | verbose(env, | |
6015 | "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n", | |
6016 | depth); | |
6017 | return -EACCES; | |
6018 | } | |
70a87ffe AS |
6019 | /* round up to 32-bytes, since this is granularity |
6020 | * of interpreter stack size | |
6021 | */ | |
9c8105bd | 6022 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 6023 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 6024 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 6025 | frame + 1, depth); |
f4d7e40a AS |
6026 | return -EACCES; |
6027 | } | |
70a87ffe | 6028 | continue_func: |
4cb3d99c | 6029 | subprog_end = subprog[idx + 1].start; |
70a87ffe | 6030 | for (; i < subprog_end; i++) { |
ba7b3e7d | 6031 | int next_insn, sidx; |
7ddc80a4 | 6032 | |
f18b03fa KKD |
6033 | if (bpf_pseudo_kfunc_call(insn + i) && !insn[i].off) { |
6034 | bool err = false; | |
6035 | ||
6036 | if (!is_bpf_throw_kfunc(insn + i)) | |
6037 | continue; | |
6038 | if (subprog[idx].is_cb) | |
6039 | err = true; | |
6040 | for (int c = 0; c < frame && !err; c++) { | |
6041 | if (subprog[ret_prog[c]].is_cb) { | |
6042 | err = true; | |
6043 | break; | |
6044 | } | |
6045 | } | |
6046 | if (!err) | |
6047 | continue; | |
6048 | verbose(env, | |
6049 | "bpf_throw kfunc (insn %d) cannot be called from callback subprog %d\n", | |
6050 | i, idx); | |
6051 | return -EINVAL; | |
6052 | } | |
6053 | ||
69c087ba | 6054 | if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i)) |
70a87ffe AS |
6055 | continue; |
6056 | /* remember insn and function to return to */ | |
6057 | ret_insn[frame] = i + 1; | |
9c8105bd | 6058 | ret_prog[frame] = idx; |
70a87ffe AS |
6059 | |
6060 | /* find the callee */ | |
7ddc80a4 | 6061 | next_insn = i + insn[i].imm + 1; |
ba7b3e7d KKD |
6062 | sidx = find_subprog(env, next_insn); |
6063 | if (sidx < 0) { | |
70a87ffe | 6064 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
7ddc80a4 | 6065 | next_insn); |
70a87ffe AS |
6066 | return -EFAULT; |
6067 | } | |
ba7b3e7d KKD |
6068 | if (subprog[sidx].is_async_cb) { |
6069 | if (subprog[sidx].has_tail_call) { | |
7ddc80a4 AS |
6070 | verbose(env, "verifier bug. subprog has tail_call and async cb\n"); |
6071 | return -EFAULT; | |
6072 | } | |
5415ccd5 KKD |
6073 | /* async callbacks don't increase bpf prog stack size unless called directly */ |
6074 | if (!bpf_pseudo_call(insn + i)) | |
6075 | continue; | |
b9ae0c9d KKD |
6076 | if (subprog[sidx].is_exception_cb) { |
6077 | verbose(env, "insn %d cannot call exception cb directly\n", i); | |
6078 | return -EINVAL; | |
6079 | } | |
7ddc80a4 AS |
6080 | } |
6081 | i = next_insn; | |
ba7b3e7d | 6082 | idx = sidx; |
ebf7d1f5 MF |
6083 | |
6084 | if (subprog[idx].has_tail_call) | |
6085 | tail_call_reachable = true; | |
6086 | ||
70a87ffe AS |
6087 | frame++; |
6088 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
6089 | verbose(env, "the call stack of %d frames is too deep !\n", |
6090 | frame); | |
6091 | return -E2BIG; | |
70a87ffe AS |
6092 | } |
6093 | goto process_func; | |
6094 | } | |
ebf7d1f5 MF |
6095 | /* if tail call got detected across bpf2bpf calls then mark each of the |
6096 | * currently present subprog frames as tail call reachable subprogs; | |
6097 | * this info will be utilized by JIT so that we will be preserving the | |
6098 | * tail call counter throughout bpf2bpf calls combined with tailcalls | |
6099 | */ | |
6100 | if (tail_call_reachable) | |
b9ae0c9d KKD |
6101 | for (j = 0; j < frame; j++) { |
6102 | if (subprog[ret_prog[j]].is_exception_cb) { | |
6103 | verbose(env, "cannot tail call within exception cb\n"); | |
6104 | return -EINVAL; | |
6105 | } | |
ebf7d1f5 | 6106 | subprog[ret_prog[j]].tail_call_reachable = true; |
b9ae0c9d | 6107 | } |
5dd0a6b8 DB |
6108 | if (subprog[0].tail_call_reachable) |
6109 | env->prog->aux->tail_call_reachable = true; | |
ebf7d1f5 | 6110 | |
70a87ffe AS |
6111 | /* end of for() loop means the last insn of the 'subprog' |
6112 | * was reached. Doesn't matter whether it was JA or EXIT | |
6113 | */ | |
6114 | if (frame == 0) | |
6115 | return 0; | |
9c8105bd | 6116 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
6117 | frame--; |
6118 | i = ret_insn[frame]; | |
9c8105bd | 6119 | idx = ret_prog[frame]; |
70a87ffe | 6120 | goto continue_func; |
f4d7e40a AS |
6121 | } |
6122 | ||
b5e9ad52 KKD |
6123 | static int check_max_stack_depth(struct bpf_verifier_env *env) |
6124 | { | |
6125 | struct bpf_subprog_info *si = env->subprog_info; | |
6126 | int ret; | |
6127 | ||
6128 | for (int i = 0; i < env->subprog_cnt; i++) { | |
6129 | if (!i || si[i].is_async_cb) { | |
6130 | ret = check_max_stack_depth_subprog(env, i); | |
6131 | if (ret < 0) | |
6132 | return ret; | |
6133 | } | |
6134 | continue; | |
6135 | } | |
6136 | return 0; | |
6137 | } | |
6138 | ||
19d28fbd | 6139 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
6140 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
6141 | const struct bpf_insn *insn, int idx) | |
6142 | { | |
6143 | int start = idx + insn->imm + 1, subprog; | |
6144 | ||
6145 | subprog = find_subprog(env, start); | |
6146 | if (subprog < 0) { | |
6147 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
6148 | start); | |
6149 | return -EFAULT; | |
6150 | } | |
9c8105bd | 6151 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 6152 | } |
19d28fbd | 6153 | #endif |
1ea47e01 | 6154 | |
afbf21dc YS |
6155 | static int __check_buffer_access(struct bpf_verifier_env *env, |
6156 | const char *buf_info, | |
6157 | const struct bpf_reg_state *reg, | |
6158 | int regno, int off, int size) | |
9df1c28b MM |
6159 | { |
6160 | if (off < 0) { | |
6161 | verbose(env, | |
4fc00b79 | 6162 | "R%d invalid %s buffer access: off=%d, size=%d\n", |
afbf21dc | 6163 | regno, buf_info, off, size); |
9df1c28b MM |
6164 | return -EACCES; |
6165 | } | |
6166 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
6167 | char tn_buf[48]; | |
6168 | ||
6169 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6170 | verbose(env, | |
4fc00b79 | 6171 | "R%d invalid variable buffer offset: off=%d, var_off=%s\n", |
9df1c28b MM |
6172 | regno, off, tn_buf); |
6173 | return -EACCES; | |
6174 | } | |
afbf21dc YS |
6175 | |
6176 | return 0; | |
6177 | } | |
6178 | ||
6179 | static int check_tp_buffer_access(struct bpf_verifier_env *env, | |
6180 | const struct bpf_reg_state *reg, | |
6181 | int regno, int off, int size) | |
6182 | { | |
6183 | int err; | |
6184 | ||
6185 | err = __check_buffer_access(env, "tracepoint", reg, regno, off, size); | |
6186 | if (err) | |
6187 | return err; | |
6188 | ||
9df1c28b MM |
6189 | if (off + size > env->prog->aux->max_tp_access) |
6190 | env->prog->aux->max_tp_access = off + size; | |
6191 | ||
6192 | return 0; | |
6193 | } | |
6194 | ||
afbf21dc YS |
6195 | static int check_buffer_access(struct bpf_verifier_env *env, |
6196 | const struct bpf_reg_state *reg, | |
6197 | int regno, int off, int size, | |
6198 | bool zero_size_allowed, | |
afbf21dc YS |
6199 | u32 *max_access) |
6200 | { | |
44e9a741 | 6201 | const char *buf_info = type_is_rdonly_mem(reg->type) ? "rdonly" : "rdwr"; |
afbf21dc YS |
6202 | int err; |
6203 | ||
6204 | err = __check_buffer_access(env, buf_info, reg, regno, off, size); | |
6205 | if (err) | |
6206 | return err; | |
6207 | ||
6208 | if (off + size > *max_access) | |
6209 | *max_access = off + size; | |
6210 | ||
6211 | return 0; | |
6212 | } | |
6213 | ||
3f50f132 JF |
6214 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
6215 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
6216 | { | |
6217 | reg->var_off = tnum_subreg(reg->var_off); | |
6218 | __reg_assign_32_into_64(reg); | |
6219 | } | |
9df1c28b | 6220 | |
0c17d1d2 JH |
6221 | /* truncate register to smaller size (in bytes) |
6222 | * must be called with size < BPF_REG_SIZE | |
6223 | */ | |
6224 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
6225 | { | |
6226 | u64 mask; | |
6227 | ||
6228 | /* clear high bits in bit representation */ | |
6229 | reg->var_off = tnum_cast(reg->var_off, size); | |
6230 | ||
6231 | /* fix arithmetic bounds */ | |
6232 | mask = ((u64)1 << (size * 8)) - 1; | |
6233 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
6234 | reg->umin_value &= mask; | |
6235 | reg->umax_value &= mask; | |
6236 | } else { | |
6237 | reg->umin_value = 0; | |
6238 | reg->umax_value = mask; | |
6239 | } | |
6240 | reg->smin_value = reg->umin_value; | |
6241 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
6242 | |
6243 | /* If size is smaller than 32bit register the 32bit register | |
6244 | * values are also truncated so we push 64-bit bounds into | |
6245 | * 32-bit bounds. Above were truncated < 32-bits already. | |
6246 | */ | |
6247 | if (size >= 4) | |
6248 | return; | |
6249 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
6250 | } |
6251 | ||
1f9a1ea8 YS |
6252 | static void set_sext64_default_val(struct bpf_reg_state *reg, int size) |
6253 | { | |
6254 | if (size == 1) { | |
6255 | reg->smin_value = reg->s32_min_value = S8_MIN; | |
6256 | reg->smax_value = reg->s32_max_value = S8_MAX; | |
6257 | } else if (size == 2) { | |
6258 | reg->smin_value = reg->s32_min_value = S16_MIN; | |
6259 | reg->smax_value = reg->s32_max_value = S16_MAX; | |
6260 | } else { | |
6261 | /* size == 4 */ | |
6262 | reg->smin_value = reg->s32_min_value = S32_MIN; | |
6263 | reg->smax_value = reg->s32_max_value = S32_MAX; | |
6264 | } | |
6265 | reg->umin_value = reg->u32_min_value = 0; | |
6266 | reg->umax_value = U64_MAX; | |
6267 | reg->u32_max_value = U32_MAX; | |
6268 | reg->var_off = tnum_unknown; | |
6269 | } | |
6270 | ||
6271 | static void coerce_reg_to_size_sx(struct bpf_reg_state *reg, int size) | |
6272 | { | |
6273 | s64 init_s64_max, init_s64_min, s64_max, s64_min, u64_cval; | |
6274 | u64 top_smax_value, top_smin_value; | |
6275 | u64 num_bits = size * 8; | |
6276 | ||
6277 | if (tnum_is_const(reg->var_off)) { | |
6278 | u64_cval = reg->var_off.value; | |
6279 | if (size == 1) | |
6280 | reg->var_off = tnum_const((s8)u64_cval); | |
6281 | else if (size == 2) | |
6282 | reg->var_off = tnum_const((s16)u64_cval); | |
6283 | else | |
6284 | /* size == 4 */ | |
6285 | reg->var_off = tnum_const((s32)u64_cval); | |
6286 | ||
6287 | u64_cval = reg->var_off.value; | |
6288 | reg->smax_value = reg->smin_value = u64_cval; | |
6289 | reg->umax_value = reg->umin_value = u64_cval; | |
6290 | reg->s32_max_value = reg->s32_min_value = u64_cval; | |
6291 | reg->u32_max_value = reg->u32_min_value = u64_cval; | |
6292 | return; | |
6293 | } | |
6294 | ||
6295 | top_smax_value = ((u64)reg->smax_value >> num_bits) << num_bits; | |
6296 | top_smin_value = ((u64)reg->smin_value >> num_bits) << num_bits; | |
6297 | ||
6298 | if (top_smax_value != top_smin_value) | |
6299 | goto out; | |
6300 | ||
6301 | /* find the s64_min and s64_min after sign extension */ | |
6302 | if (size == 1) { | |
6303 | init_s64_max = (s8)reg->smax_value; | |
6304 | init_s64_min = (s8)reg->smin_value; | |
6305 | } else if (size == 2) { | |
6306 | init_s64_max = (s16)reg->smax_value; | |
6307 | init_s64_min = (s16)reg->smin_value; | |
6308 | } else { | |
6309 | init_s64_max = (s32)reg->smax_value; | |
6310 | init_s64_min = (s32)reg->smin_value; | |
6311 | } | |
6312 | ||
6313 | s64_max = max(init_s64_max, init_s64_min); | |
6314 | s64_min = min(init_s64_max, init_s64_min); | |
6315 | ||
6316 | /* both of s64_max/s64_min positive or negative */ | |
09fedc73 | 6317 | if ((s64_max >= 0) == (s64_min >= 0)) { |
1f9a1ea8 YS |
6318 | reg->smin_value = reg->s32_min_value = s64_min; |
6319 | reg->smax_value = reg->s32_max_value = s64_max; | |
6320 | reg->umin_value = reg->u32_min_value = s64_min; | |
6321 | reg->umax_value = reg->u32_max_value = s64_max; | |
6322 | reg->var_off = tnum_range(s64_min, s64_max); | |
6323 | return; | |
6324 | } | |
6325 | ||
6326 | out: | |
6327 | set_sext64_default_val(reg, size); | |
6328 | } | |
6329 | ||
8100928c YS |
6330 | static void set_sext32_default_val(struct bpf_reg_state *reg, int size) |
6331 | { | |
6332 | if (size == 1) { | |
6333 | reg->s32_min_value = S8_MIN; | |
6334 | reg->s32_max_value = S8_MAX; | |
6335 | } else { | |
6336 | /* size == 2 */ | |
6337 | reg->s32_min_value = S16_MIN; | |
6338 | reg->s32_max_value = S16_MAX; | |
6339 | } | |
6340 | reg->u32_min_value = 0; | |
6341 | reg->u32_max_value = U32_MAX; | |
6342 | } | |
6343 | ||
6344 | static void coerce_subreg_to_size_sx(struct bpf_reg_state *reg, int size) | |
6345 | { | |
6346 | s32 init_s32_max, init_s32_min, s32_max, s32_min, u32_val; | |
6347 | u32 top_smax_value, top_smin_value; | |
6348 | u32 num_bits = size * 8; | |
6349 | ||
6350 | if (tnum_is_const(reg->var_off)) { | |
6351 | u32_val = reg->var_off.value; | |
6352 | if (size == 1) | |
6353 | reg->var_off = tnum_const((s8)u32_val); | |
6354 | else | |
6355 | reg->var_off = tnum_const((s16)u32_val); | |
6356 | ||
6357 | u32_val = reg->var_off.value; | |
6358 | reg->s32_min_value = reg->s32_max_value = u32_val; | |
6359 | reg->u32_min_value = reg->u32_max_value = u32_val; | |
6360 | return; | |
6361 | } | |
6362 | ||
6363 | top_smax_value = ((u32)reg->s32_max_value >> num_bits) << num_bits; | |
6364 | top_smin_value = ((u32)reg->s32_min_value >> num_bits) << num_bits; | |
6365 | ||
6366 | if (top_smax_value != top_smin_value) | |
6367 | goto out; | |
6368 | ||
6369 | /* find the s32_min and s32_min after sign extension */ | |
6370 | if (size == 1) { | |
6371 | init_s32_max = (s8)reg->s32_max_value; | |
6372 | init_s32_min = (s8)reg->s32_min_value; | |
6373 | } else { | |
6374 | /* size == 2 */ | |
6375 | init_s32_max = (s16)reg->s32_max_value; | |
6376 | init_s32_min = (s16)reg->s32_min_value; | |
6377 | } | |
6378 | s32_max = max(init_s32_max, init_s32_min); | |
6379 | s32_min = min(init_s32_max, init_s32_min); | |
6380 | ||
09fedc73 | 6381 | if ((s32_min >= 0) == (s32_max >= 0)) { |
8100928c YS |
6382 | reg->s32_min_value = s32_min; |
6383 | reg->s32_max_value = s32_max; | |
6384 | reg->u32_min_value = (u32)s32_min; | |
6385 | reg->u32_max_value = (u32)s32_max; | |
6386 | return; | |
6387 | } | |
6388 | ||
6389 | out: | |
6390 | set_sext32_default_val(reg, size); | |
6391 | } | |
6392 | ||
a23740ec AN |
6393 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
6394 | { | |
353050be DB |
6395 | /* A map is considered read-only if the following condition are true: |
6396 | * | |
6397 | * 1) BPF program side cannot change any of the map content. The | |
6398 | * BPF_F_RDONLY_PROG flag is throughout the lifetime of a map | |
6399 | * and was set at map creation time. | |
6400 | * 2) The map value(s) have been initialized from user space by a | |
6401 | * loader and then "frozen", such that no new map update/delete | |
6402 | * operations from syscall side are possible for the rest of | |
6403 | * the map's lifetime from that point onwards. | |
6404 | * 3) Any parallel/pending map update/delete operations from syscall | |
6405 | * side have been completed. Only after that point, it's safe to | |
6406 | * assume that map value(s) are immutable. | |
6407 | */ | |
6408 | return (map->map_flags & BPF_F_RDONLY_PROG) && | |
6409 | READ_ONCE(map->frozen) && | |
6410 | !bpf_map_write_active(map); | |
a23740ec AN |
6411 | } |
6412 | ||
1f9a1ea8 YS |
6413 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val, |
6414 | bool is_ldsx) | |
a23740ec AN |
6415 | { |
6416 | void *ptr; | |
6417 | u64 addr; | |
6418 | int err; | |
6419 | ||
6420 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
6421 | if (err) | |
6422 | return err; | |
2dedd7d2 | 6423 | ptr = (void *)(long)addr + off; |
a23740ec AN |
6424 | |
6425 | switch (size) { | |
6426 | case sizeof(u8): | |
1f9a1ea8 | 6427 | *val = is_ldsx ? (s64)*(s8 *)ptr : (u64)*(u8 *)ptr; |
a23740ec AN |
6428 | break; |
6429 | case sizeof(u16): | |
1f9a1ea8 | 6430 | *val = is_ldsx ? (s64)*(s16 *)ptr : (u64)*(u16 *)ptr; |
a23740ec AN |
6431 | break; |
6432 | case sizeof(u32): | |
1f9a1ea8 | 6433 | *val = is_ldsx ? (s64)*(s32 *)ptr : (u64)*(u32 *)ptr; |
a23740ec AN |
6434 | break; |
6435 | case sizeof(u64): | |
6436 | *val = *(u64 *)ptr; | |
6437 | break; | |
6438 | default: | |
6439 | return -EINVAL; | |
6440 | } | |
6441 | return 0; | |
6442 | } | |
6443 | ||
6fcd486b | 6444 | #define BTF_TYPE_SAFE_RCU(__type) __PASTE(__type, __safe_rcu) |
30ee9821 | 6445 | #define BTF_TYPE_SAFE_RCU_OR_NULL(__type) __PASTE(__type, __safe_rcu_or_null) |
6fcd486b | 6446 | #define BTF_TYPE_SAFE_TRUSTED(__type) __PASTE(__type, __safe_trusted) |
57539b1c | 6447 | |
6fcd486b AS |
6448 | /* |
6449 | * Allow list few fields as RCU trusted or full trusted. | |
6450 | * This logic doesn't allow mix tagging and will be removed once GCC supports | |
6451 | * btf_type_tag. | |
6452 | */ | |
6453 | ||
6454 | /* RCU trusted: these fields are trusted in RCU CS and never NULL */ | |
6455 | BTF_TYPE_SAFE_RCU(struct task_struct) { | |
57539b1c | 6456 | const cpumask_t *cpus_ptr; |
8d093b4e | 6457 | struct css_set __rcu *cgroups; |
6fcd486b AS |
6458 | struct task_struct __rcu *real_parent; |
6459 | struct task_struct *group_leader; | |
8d093b4e AS |
6460 | }; |
6461 | ||
30ee9821 AS |
6462 | BTF_TYPE_SAFE_RCU(struct cgroup) { |
6463 | /* cgrp->kn is always accessible as documented in kernel/cgroup/cgroup.c */ | |
6464 | struct kernfs_node *kn; | |
6465 | }; | |
6466 | ||
6fcd486b | 6467 | BTF_TYPE_SAFE_RCU(struct css_set) { |
8d093b4e | 6468 | struct cgroup *dfl_cgrp; |
57539b1c DV |
6469 | }; |
6470 | ||
30ee9821 AS |
6471 | /* RCU trusted: these fields are trusted in RCU CS and can be NULL */ |
6472 | BTF_TYPE_SAFE_RCU_OR_NULL(struct mm_struct) { | |
6473 | struct file __rcu *exe_file; | |
6474 | }; | |
6475 | ||
6476 | /* skb->sk, req->sk are not RCU protected, but we mark them as such | |
6477 | * because bpf prog accessible sockets are SOCK_RCU_FREE. | |
6478 | */ | |
6479 | BTF_TYPE_SAFE_RCU_OR_NULL(struct sk_buff) { | |
6480 | struct sock *sk; | |
6481 | }; | |
6482 | ||
6483 | BTF_TYPE_SAFE_RCU_OR_NULL(struct request_sock) { | |
6484 | struct sock *sk; | |
6485 | }; | |
6486 | ||
6fcd486b AS |
6487 | /* full trusted: these fields are trusted even outside of RCU CS and never NULL */ |
6488 | BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta) { | |
63260df1 | 6489 | struct seq_file *seq; |
6fcd486b AS |
6490 | }; |
6491 | ||
6492 | BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task) { | |
63260df1 AS |
6493 | struct bpf_iter_meta *meta; |
6494 | struct task_struct *task; | |
6fcd486b AS |
6495 | }; |
6496 | ||
6497 | BTF_TYPE_SAFE_TRUSTED(struct linux_binprm) { | |
6498 | struct file *file; | |
6499 | }; | |
6500 | ||
6501 | BTF_TYPE_SAFE_TRUSTED(struct file) { | |
6502 | struct inode *f_inode; | |
6503 | }; | |
6504 | ||
6505 | BTF_TYPE_SAFE_TRUSTED(struct dentry) { | |
6506 | /* no negative dentry-s in places where bpf can see it */ | |
6507 | struct inode *d_inode; | |
6508 | }; | |
6509 | ||
6510 | BTF_TYPE_SAFE_TRUSTED(struct socket) { | |
6511 | struct sock *sk; | |
6512 | }; | |
6513 | ||
6514 | static bool type_is_rcu(struct bpf_verifier_env *env, | |
6515 | struct bpf_reg_state *reg, | |
63260df1 | 6516 | const char *field_name, u32 btf_id) |
57539b1c | 6517 | { |
6fcd486b | 6518 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct task_struct)); |
30ee9821 | 6519 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct cgroup)); |
6fcd486b | 6520 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct css_set)); |
57539b1c | 6521 | |
63260df1 | 6522 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_rcu"); |
6fcd486b | 6523 | } |
57539b1c | 6524 | |
30ee9821 AS |
6525 | static bool type_is_rcu_or_null(struct bpf_verifier_env *env, |
6526 | struct bpf_reg_state *reg, | |
6527 | const char *field_name, u32 btf_id) | |
6528 | { | |
6529 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct mm_struct)); | |
6530 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct sk_buff)); | |
6531 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct request_sock)); | |
6532 | ||
6533 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_rcu_or_null"); | |
6534 | } | |
6535 | ||
6fcd486b AS |
6536 | static bool type_is_trusted(struct bpf_verifier_env *env, |
6537 | struct bpf_reg_state *reg, | |
63260df1 | 6538 | const char *field_name, u32 btf_id) |
6fcd486b AS |
6539 | { |
6540 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta)); | |
6541 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task)); | |
6542 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct linux_binprm)); | |
6543 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct file)); | |
6544 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct dentry)); | |
6545 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct socket)); | |
6546 | ||
63260df1 | 6547 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_trusted"); |
57539b1c DV |
6548 | } |
6549 | ||
9e15db66 AS |
6550 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
6551 | struct bpf_reg_state *regs, | |
6552 | int regno, int off, int size, | |
6553 | enum bpf_access_type atype, | |
6554 | int value_regno) | |
6555 | { | |
6556 | struct bpf_reg_state *reg = regs + regno; | |
22dc4a0f AN |
6557 | const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id); |
6558 | const char *tname = btf_name_by_offset(reg->btf, t->name_off); | |
63260df1 | 6559 | const char *field_name = NULL; |
c6f1bfe8 | 6560 | enum bpf_type_flag flag = 0; |
b7e852a9 | 6561 | u32 btf_id = 0; |
9e15db66 AS |
6562 | int ret; |
6563 | ||
c67cae55 AS |
6564 | if (!env->allow_ptr_leaks) { |
6565 | verbose(env, | |
6566 | "'struct %s' access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", | |
6567 | tname); | |
6568 | return -EPERM; | |
6569 | } | |
6570 | if (!env->prog->gpl_compatible && btf_is_kernel(reg->btf)) { | |
6571 | verbose(env, | |
6572 | "Cannot access kernel 'struct %s' from non-GPL compatible program\n", | |
6573 | tname); | |
6574 | return -EINVAL; | |
6575 | } | |
9e15db66 AS |
6576 | if (off < 0) { |
6577 | verbose(env, | |
6578 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
6579 | regno, tname, off); | |
6580 | return -EACCES; | |
6581 | } | |
6582 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
6583 | char tn_buf[48]; | |
6584 | ||
6585 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6586 | verbose(env, | |
6587 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
6588 | regno, tname, off, tn_buf); | |
6589 | return -EACCES; | |
6590 | } | |
6591 | ||
c6f1bfe8 YS |
6592 | if (reg->type & MEM_USER) { |
6593 | verbose(env, | |
6594 | "R%d is ptr_%s access user memory: off=%d\n", | |
6595 | regno, tname, off); | |
6596 | return -EACCES; | |
6597 | } | |
6598 | ||
5844101a HL |
6599 | if (reg->type & MEM_PERCPU) { |
6600 | verbose(env, | |
6601 | "R%d is ptr_%s access percpu memory: off=%d\n", | |
6602 | regno, tname, off); | |
6603 | return -EACCES; | |
6604 | } | |
6605 | ||
7d64c513 | 6606 | if (env->ops->btf_struct_access && !type_is_alloc(reg->type) && atype == BPF_WRITE) { |
282de143 KKD |
6607 | if (!btf_is_kernel(reg->btf)) { |
6608 | verbose(env, "verifier internal error: reg->btf must be kernel btf\n"); | |
6609 | return -EFAULT; | |
6610 | } | |
b7e852a9 | 6611 | ret = env->ops->btf_struct_access(&env->log, reg, off, size); |
27ae7997 | 6612 | } else { |
282de143 KKD |
6613 | /* Writes are permitted with default btf_struct_access for |
6614 | * program allocated objects (which always have ref_obj_id > 0), | |
6615 | * but not for untrusted PTR_TO_BTF_ID | MEM_ALLOC. | |
6616 | */ | |
503e4def | 6617 | if (atype != BPF_READ && !type_is_ptr_alloc_obj(reg->type)) { |
27ae7997 MKL |
6618 | verbose(env, "only read is supported\n"); |
6619 | return -EACCES; | |
6620 | } | |
6621 | ||
6a3cd331 | 6622 | if (type_is_alloc(reg->type) && !type_is_non_owning_ref(reg->type) && |
01cc55af | 6623 | !(reg->type & MEM_RCU) && !reg->ref_obj_id) { |
282de143 KKD |
6624 | verbose(env, "verifier internal error: ref_obj_id for allocated object must be non-zero\n"); |
6625 | return -EFAULT; | |
6626 | } | |
6627 | ||
63260df1 | 6628 | ret = btf_struct_access(&env->log, reg, off, size, atype, &btf_id, &flag, &field_name); |
27ae7997 MKL |
6629 | } |
6630 | ||
9e15db66 AS |
6631 | if (ret < 0) |
6632 | return ret; | |
6633 | ||
6fcd486b AS |
6634 | if (ret != PTR_TO_BTF_ID) { |
6635 | /* just mark; */ | |
6efe152d | 6636 | |
6fcd486b AS |
6637 | } else if (type_flag(reg->type) & PTR_UNTRUSTED) { |
6638 | /* If this is an untrusted pointer, all pointers formed by walking it | |
6639 | * also inherit the untrusted flag. | |
6640 | */ | |
6641 | flag = PTR_UNTRUSTED; | |
6642 | ||
6643 | } else if (is_trusted_reg(reg) || is_rcu_reg(reg)) { | |
6644 | /* By default any pointer obtained from walking a trusted pointer is no | |
6645 | * longer trusted, unless the field being accessed has explicitly been | |
6646 | * marked as inheriting its parent's state of trust (either full or RCU). | |
6647 | * For example: | |
6648 | * 'cgroups' pointer is untrusted if task->cgroups dereference | |
6649 | * happened in a sleepable program outside of bpf_rcu_read_lock() | |
6650 | * section. In a non-sleepable program it's trusted while in RCU CS (aka MEM_RCU). | |
6651 | * Note bpf_rcu_read_unlock() converts MEM_RCU pointers to PTR_UNTRUSTED. | |
6652 | * | |
6653 | * A regular RCU-protected pointer with __rcu tag can also be deemed | |
6654 | * trusted if we are in an RCU CS. Such pointer can be NULL. | |
20c09d92 | 6655 | */ |
63260df1 | 6656 | if (type_is_trusted(env, reg, field_name, btf_id)) { |
6fcd486b AS |
6657 | flag |= PTR_TRUSTED; |
6658 | } else if (in_rcu_cs(env) && !type_may_be_null(reg->type)) { | |
63260df1 | 6659 | if (type_is_rcu(env, reg, field_name, btf_id)) { |
6fcd486b AS |
6660 | /* ignore __rcu tag and mark it MEM_RCU */ |
6661 | flag |= MEM_RCU; | |
30ee9821 AS |
6662 | } else if (flag & MEM_RCU || |
6663 | type_is_rcu_or_null(env, reg, field_name, btf_id)) { | |
6fcd486b | 6664 | /* __rcu tagged pointers can be NULL */ |
30ee9821 | 6665 | flag |= MEM_RCU | PTR_MAYBE_NULL; |
7ce4dc3e YS |
6666 | |
6667 | /* We always trust them */ | |
6668 | if (type_is_rcu_or_null(env, reg, field_name, btf_id) && | |
6669 | flag & PTR_UNTRUSTED) | |
6670 | flag &= ~PTR_UNTRUSTED; | |
6fcd486b AS |
6671 | } else if (flag & (MEM_PERCPU | MEM_USER)) { |
6672 | /* keep as-is */ | |
6673 | } else { | |
afeebf9f AS |
6674 | /* walking unknown pointers yields old deprecated PTR_TO_BTF_ID */ |
6675 | clear_trusted_flags(&flag); | |
6fcd486b AS |
6676 | } |
6677 | } else { | |
6678 | /* | |
6679 | * If not in RCU CS or MEM_RCU pointer can be NULL then | |
6680 | * aggressively mark as untrusted otherwise such | |
6681 | * pointers will be plain PTR_TO_BTF_ID without flags | |
6682 | * and will be allowed to be passed into helpers for | |
6683 | * compat reasons. | |
6684 | */ | |
6685 | flag = PTR_UNTRUSTED; | |
6686 | } | |
20c09d92 | 6687 | } else { |
6fcd486b | 6688 | /* Old compat. Deprecated */ |
afeebf9f | 6689 | clear_trusted_flags(&flag); |
20c09d92 | 6690 | } |
3f00c523 | 6691 | |
41c48f3a | 6692 | if (atype == BPF_READ && value_regno >= 0) |
c6f1bfe8 | 6693 | mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id, flag); |
41c48f3a AI |
6694 | |
6695 | return 0; | |
6696 | } | |
6697 | ||
6698 | static int check_ptr_to_map_access(struct bpf_verifier_env *env, | |
6699 | struct bpf_reg_state *regs, | |
6700 | int regno, int off, int size, | |
6701 | enum bpf_access_type atype, | |
6702 | int value_regno) | |
6703 | { | |
6704 | struct bpf_reg_state *reg = regs + regno; | |
6705 | struct bpf_map *map = reg->map_ptr; | |
6728aea7 | 6706 | struct bpf_reg_state map_reg; |
c6f1bfe8 | 6707 | enum bpf_type_flag flag = 0; |
41c48f3a AI |
6708 | const struct btf_type *t; |
6709 | const char *tname; | |
6710 | u32 btf_id; | |
6711 | int ret; | |
6712 | ||
6713 | if (!btf_vmlinux) { | |
6714 | verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
6715 | return -ENOTSUPP; | |
6716 | } | |
6717 | ||
6718 | if (!map->ops->map_btf_id || !*map->ops->map_btf_id) { | |
6719 | verbose(env, "map_ptr access not supported for map type %d\n", | |
6720 | map->map_type); | |
6721 | return -ENOTSUPP; | |
6722 | } | |
6723 | ||
6724 | t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id); | |
6725 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
6726 | ||
c67cae55 | 6727 | if (!env->allow_ptr_leaks) { |
41c48f3a | 6728 | verbose(env, |
c67cae55 | 6729 | "'struct %s' access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", |
41c48f3a AI |
6730 | tname); |
6731 | return -EPERM; | |
9e15db66 | 6732 | } |
27ae7997 | 6733 | |
41c48f3a AI |
6734 | if (off < 0) { |
6735 | verbose(env, "R%d is %s invalid negative access: off=%d\n", | |
6736 | regno, tname, off); | |
6737 | return -EACCES; | |
6738 | } | |
6739 | ||
6740 | if (atype != BPF_READ) { | |
6741 | verbose(env, "only read from %s is supported\n", tname); | |
6742 | return -EACCES; | |
6743 | } | |
6744 | ||
6728aea7 KKD |
6745 | /* Simulate access to a PTR_TO_BTF_ID */ |
6746 | memset(&map_reg, 0, sizeof(map_reg)); | |
6747 | mark_btf_ld_reg(env, &map_reg, 0, PTR_TO_BTF_ID, btf_vmlinux, *map->ops->map_btf_id, 0); | |
63260df1 | 6748 | ret = btf_struct_access(&env->log, &map_reg, off, size, atype, &btf_id, &flag, NULL); |
41c48f3a AI |
6749 | if (ret < 0) |
6750 | return ret; | |
6751 | ||
6752 | if (value_regno >= 0) | |
c6f1bfe8 | 6753 | mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id, flag); |
41c48f3a | 6754 | |
9e15db66 AS |
6755 | return 0; |
6756 | } | |
6757 | ||
01f810ac AM |
6758 | /* Check that the stack access at the given offset is within bounds. The |
6759 | * maximum valid offset is -1. | |
6760 | * | |
6761 | * The minimum valid offset is -MAX_BPF_STACK for writes, and | |
6762 | * -state->allocated_stack for reads. | |
6763 | */ | |
6764 | static int check_stack_slot_within_bounds(int off, | |
6765 | struct bpf_func_state *state, | |
6766 | enum bpf_access_type t) | |
6767 | { | |
6768 | int min_valid_off; | |
6769 | ||
6770 | if (t == BPF_WRITE) | |
6771 | min_valid_off = -MAX_BPF_STACK; | |
6772 | else | |
6773 | min_valid_off = -state->allocated_stack; | |
6774 | ||
6775 | if (off < min_valid_off || off > -1) | |
6776 | return -EACCES; | |
6777 | return 0; | |
6778 | } | |
6779 | ||
6780 | /* Check that the stack access at 'regno + off' falls within the maximum stack | |
6781 | * bounds. | |
6782 | * | |
6783 | * 'off' includes `regno->offset`, but not its dynamic part (if any). | |
6784 | */ | |
6785 | static int check_stack_access_within_bounds( | |
6786 | struct bpf_verifier_env *env, | |
6787 | int regno, int off, int access_size, | |
61df10c7 | 6788 | enum bpf_access_src src, enum bpf_access_type type) |
01f810ac AM |
6789 | { |
6790 | struct bpf_reg_state *regs = cur_regs(env); | |
6791 | struct bpf_reg_state *reg = regs + regno; | |
6792 | struct bpf_func_state *state = func(env, reg); | |
6793 | int min_off, max_off; | |
6794 | int err; | |
6795 | char *err_extra; | |
6796 | ||
6797 | if (src == ACCESS_HELPER) | |
6798 | /* We don't know if helpers are reading or writing (or both). */ | |
6799 | err_extra = " indirect access to"; | |
6800 | else if (type == BPF_READ) | |
6801 | err_extra = " read from"; | |
6802 | else | |
6803 | err_extra = " write to"; | |
6804 | ||
6805 | if (tnum_is_const(reg->var_off)) { | |
6806 | min_off = reg->var_off.value + off; | |
6807 | if (access_size > 0) | |
6808 | max_off = min_off + access_size - 1; | |
6809 | else | |
6810 | max_off = min_off; | |
6811 | } else { | |
6812 | if (reg->smax_value >= BPF_MAX_VAR_OFF || | |
6813 | reg->smin_value <= -BPF_MAX_VAR_OFF) { | |
6814 | verbose(env, "invalid unbounded variable-offset%s stack R%d\n", | |
6815 | err_extra, regno); | |
6816 | return -EACCES; | |
6817 | } | |
6818 | min_off = reg->smin_value + off; | |
6819 | if (access_size > 0) | |
6820 | max_off = reg->smax_value + off + access_size - 1; | |
6821 | else | |
6822 | max_off = min_off; | |
6823 | } | |
6824 | ||
6825 | err = check_stack_slot_within_bounds(min_off, state, type); | |
6826 | if (!err) | |
6827 | err = check_stack_slot_within_bounds(max_off, state, type); | |
6828 | ||
6829 | if (err) { | |
6830 | if (tnum_is_const(reg->var_off)) { | |
6831 | verbose(env, "invalid%s stack R%d off=%d size=%d\n", | |
6832 | err_extra, regno, off, access_size); | |
6833 | } else { | |
6834 | char tn_buf[48]; | |
6835 | ||
6836 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6837 | verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n", | |
6838 | err_extra, regno, tn_buf, access_size); | |
6839 | } | |
6840 | } | |
6841 | return err; | |
6842 | } | |
41c48f3a | 6843 | |
17a52670 AS |
6844 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
6845 | * if t==write, value_regno is a register which value is stored into memory | |
6846 | * if t==read, value_regno is a register which will receive the value from memory | |
6847 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
6848 | * if t==read && value_regno==-1, don't care what we read from memory | |
6849 | */ | |
ca369602 DB |
6850 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
6851 | int off, int bpf_size, enum bpf_access_type t, | |
1f9a1ea8 | 6852 | int value_regno, bool strict_alignment_once, bool is_ldsx) |
17a52670 | 6853 | { |
638f5b90 AS |
6854 | struct bpf_reg_state *regs = cur_regs(env); |
6855 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 6856 | struct bpf_func_state *state; |
17a52670 AS |
6857 | int size, err = 0; |
6858 | ||
6859 | size = bpf_size_to_bytes(bpf_size); | |
6860 | if (size < 0) | |
6861 | return size; | |
6862 | ||
f1174f77 | 6863 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 6864 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
6865 | if (err) |
6866 | return err; | |
17a52670 | 6867 | |
f1174f77 EC |
6868 | /* for access checks, reg->off is just part of off */ |
6869 | off += reg->off; | |
6870 | ||
69c087ba YS |
6871 | if (reg->type == PTR_TO_MAP_KEY) { |
6872 | if (t == BPF_WRITE) { | |
6873 | verbose(env, "write to change key R%d not allowed\n", regno); | |
6874 | return -EACCES; | |
6875 | } | |
6876 | ||
6877 | err = check_mem_region_access(env, regno, off, size, | |
6878 | reg->map_ptr->key_size, false); | |
6879 | if (err) | |
6880 | return err; | |
6881 | if (value_regno >= 0) | |
6882 | mark_reg_unknown(env, regs, value_regno); | |
6883 | } else if (reg->type == PTR_TO_MAP_VALUE) { | |
aa3496ac | 6884 | struct btf_field *kptr_field = NULL; |
61df10c7 | 6885 | |
1be7f75d AS |
6886 | if (t == BPF_WRITE && value_regno >= 0 && |
6887 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 6888 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
6889 | return -EACCES; |
6890 | } | |
591fe988 DB |
6891 | err = check_map_access_type(env, regno, off, size, t); |
6892 | if (err) | |
6893 | return err; | |
61df10c7 KKD |
6894 | err = check_map_access(env, regno, off, size, false, ACCESS_DIRECT); |
6895 | if (err) | |
6896 | return err; | |
6897 | if (tnum_is_const(reg->var_off)) | |
aa3496ac KKD |
6898 | kptr_field = btf_record_find(reg->map_ptr->record, |
6899 | off + reg->var_off.value, BPF_KPTR); | |
6900 | if (kptr_field) { | |
6901 | err = check_map_kptr_access(env, regno, value_regno, insn_idx, kptr_field); | |
61df10c7 | 6902 | } else if (t == BPF_READ && value_regno >= 0) { |
a23740ec AN |
6903 | struct bpf_map *map = reg->map_ptr; |
6904 | ||
6905 | /* if map is read-only, track its contents as scalars */ | |
6906 | if (tnum_is_const(reg->var_off) && | |
6907 | bpf_map_is_rdonly(map) && | |
6908 | map->ops->map_direct_value_addr) { | |
6909 | int map_off = off + reg->var_off.value; | |
6910 | u64 val = 0; | |
6911 | ||
6912 | err = bpf_map_direct_read(map, map_off, size, | |
1f9a1ea8 | 6913 | &val, is_ldsx); |
a23740ec AN |
6914 | if (err) |
6915 | return err; | |
6916 | ||
6917 | regs[value_regno].type = SCALAR_VALUE; | |
6918 | __mark_reg_known(®s[value_regno], val); | |
6919 | } else { | |
6920 | mark_reg_unknown(env, regs, value_regno); | |
6921 | } | |
6922 | } | |
34d3a78c HL |
6923 | } else if (base_type(reg->type) == PTR_TO_MEM) { |
6924 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
6925 | ||
6926 | if (type_may_be_null(reg->type)) { | |
6927 | verbose(env, "R%d invalid mem access '%s'\n", regno, | |
6928 | reg_type_str(env, reg->type)); | |
6929 | return -EACCES; | |
6930 | } | |
6931 | ||
6932 | if (t == BPF_WRITE && rdonly_mem) { | |
6933 | verbose(env, "R%d cannot write into %s\n", | |
6934 | regno, reg_type_str(env, reg->type)); | |
6935 | return -EACCES; | |
6936 | } | |
6937 | ||
457f4436 AN |
6938 | if (t == BPF_WRITE && value_regno >= 0 && |
6939 | is_pointer_value(env, value_regno)) { | |
6940 | verbose(env, "R%d leaks addr into mem\n", value_regno); | |
6941 | return -EACCES; | |
6942 | } | |
34d3a78c | 6943 | |
457f4436 AN |
6944 | err = check_mem_region_access(env, regno, off, size, |
6945 | reg->mem_size, false); | |
34d3a78c | 6946 | if (!err && value_regno >= 0 && (t == BPF_READ || rdonly_mem)) |
457f4436 | 6947 | mark_reg_unknown(env, regs, value_regno); |
1a0dc1ac | 6948 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 6949 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
22dc4a0f | 6950 | struct btf *btf = NULL; |
9e15db66 | 6951 | u32 btf_id = 0; |
19de99f7 | 6952 | |
1be7f75d AS |
6953 | if (t == BPF_WRITE && value_regno >= 0 && |
6954 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 6955 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
6956 | return -EACCES; |
6957 | } | |
f1174f77 | 6958 | |
be80a1d3 | 6959 | err = check_ptr_off_reg(env, reg, regno); |
58990d1f DB |
6960 | if (err < 0) |
6961 | return err; | |
6962 | ||
c6f1bfe8 YS |
6963 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf, |
6964 | &btf_id); | |
9e15db66 AS |
6965 | if (err) |
6966 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 6967 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 6968 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
6969 | * PTR_TO_PACKET[_META,_END]. In the latter |
6970 | * case, we know the offset is zero. | |
f1174f77 | 6971 | */ |
46f8bc92 | 6972 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 6973 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 6974 | } else { |
638f5b90 | 6975 | mark_reg_known_zero(env, regs, |
61bd5218 | 6976 | value_regno); |
c25b2ae1 | 6977 | if (type_may_be_null(reg_type)) |
46f8bc92 | 6978 | regs[value_regno].id = ++env->id_gen; |
5327ed3d JW |
6979 | /* A load of ctx field could have different |
6980 | * actual load size with the one encoded in the | |
6981 | * insn. When the dst is PTR, it is for sure not | |
6982 | * a sub-register. | |
6983 | */ | |
6984 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
c25b2ae1 | 6985 | if (base_type(reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 6986 | regs[value_regno].btf = btf; |
9e15db66 | 6987 | regs[value_regno].btf_id = btf_id; |
22dc4a0f | 6988 | } |
46f8bc92 | 6989 | } |
638f5b90 | 6990 | regs[value_regno].type = reg_type; |
969bf05e | 6991 | } |
17a52670 | 6992 | |
f1174f77 | 6993 | } else if (reg->type == PTR_TO_STACK) { |
01f810ac AM |
6994 | /* Basic bounds checks. */ |
6995 | err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t); | |
e4298d25 DB |
6996 | if (err) |
6997 | return err; | |
8726679a | 6998 | |
f4d7e40a AS |
6999 | state = func(env, reg); |
7000 | err = update_stack_depth(env, state, off); | |
7001 | if (err) | |
7002 | return err; | |
8726679a | 7003 | |
01f810ac AM |
7004 | if (t == BPF_READ) |
7005 | err = check_stack_read(env, regno, off, size, | |
61bd5218 | 7006 | value_regno); |
01f810ac AM |
7007 | else |
7008 | err = check_stack_write(env, regno, off, size, | |
7009 | value_regno, insn_idx); | |
de8f3a83 | 7010 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 7011 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 7012 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
7013 | return -EACCES; |
7014 | } | |
4acf6c0b BB |
7015 | if (t == BPF_WRITE && value_regno >= 0 && |
7016 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
7017 | verbose(env, "R%d leaks addr into packet\n", |
7018 | value_regno); | |
4acf6c0b BB |
7019 | return -EACCES; |
7020 | } | |
9fd29c08 | 7021 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 7022 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 7023 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
7024 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
7025 | if (t == BPF_WRITE && value_regno >= 0 && | |
7026 | is_pointer_value(env, value_regno)) { | |
7027 | verbose(env, "R%d leaks addr into flow keys\n", | |
7028 | value_regno); | |
7029 | return -EACCES; | |
7030 | } | |
7031 | ||
7032 | err = check_flow_keys_access(env, off, size); | |
7033 | if (!err && t == BPF_READ && value_regno >= 0) | |
7034 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 7035 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 7036 | if (t == BPF_WRITE) { |
46f8bc92 | 7037 | verbose(env, "R%d cannot write into %s\n", |
c25b2ae1 | 7038 | regno, reg_type_str(env, reg->type)); |
c64b7983 JS |
7039 | return -EACCES; |
7040 | } | |
5f456649 | 7041 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
7042 | if (!err && value_regno >= 0) |
7043 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
7044 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
7045 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
7046 | if (!err && t == BPF_READ && value_regno >= 0) | |
7047 | mark_reg_unknown(env, regs, value_regno); | |
bff61f6f HL |
7048 | } else if (base_type(reg->type) == PTR_TO_BTF_ID && |
7049 | !type_may_be_null(reg->type)) { | |
9e15db66 AS |
7050 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, |
7051 | value_regno); | |
41c48f3a AI |
7052 | } else if (reg->type == CONST_PTR_TO_MAP) { |
7053 | err = check_ptr_to_map_access(env, regs, regno, off, size, t, | |
7054 | value_regno); | |
20b2aff4 HL |
7055 | } else if (base_type(reg->type) == PTR_TO_BUF) { |
7056 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
20b2aff4 HL |
7057 | u32 *max_access; |
7058 | ||
7059 | if (rdonly_mem) { | |
7060 | if (t == BPF_WRITE) { | |
7061 | verbose(env, "R%d cannot write into %s\n", | |
7062 | regno, reg_type_str(env, reg->type)); | |
7063 | return -EACCES; | |
7064 | } | |
20b2aff4 HL |
7065 | max_access = &env->prog->aux->max_rdonly_access; |
7066 | } else { | |
20b2aff4 | 7067 | max_access = &env->prog->aux->max_rdwr_access; |
afbf21dc | 7068 | } |
20b2aff4 | 7069 | |
f6dfbe31 | 7070 | err = check_buffer_access(env, reg, regno, off, size, false, |
44e9a741 | 7071 | max_access); |
20b2aff4 HL |
7072 | |
7073 | if (!err && value_regno >= 0 && (rdonly_mem || t == BPF_READ)) | |
afbf21dc | 7074 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 7075 | } else { |
61bd5218 | 7076 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
c25b2ae1 | 7077 | reg_type_str(env, reg->type)); |
17a52670 AS |
7078 | return -EACCES; |
7079 | } | |
969bf05e | 7080 | |
f1174f77 | 7081 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 7082 | regs[value_regno].type == SCALAR_VALUE) { |
1f9a1ea8 YS |
7083 | if (!is_ldsx) |
7084 | /* b/h/w load zero-extends, mark upper bits as known 0 */ | |
7085 | coerce_reg_to_size(®s[value_regno], size); | |
7086 | else | |
7087 | coerce_reg_to_size_sx(®s[value_regno], size); | |
969bf05e | 7088 | } |
17a52670 AS |
7089 | return err; |
7090 | } | |
7091 | ||
91c960b0 | 7092 | static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 7093 | { |
5ffa2550 | 7094 | int load_reg; |
17a52670 AS |
7095 | int err; |
7096 | ||
5ca419f2 BJ |
7097 | switch (insn->imm) { |
7098 | case BPF_ADD: | |
7099 | case BPF_ADD | BPF_FETCH: | |
981f94c3 BJ |
7100 | case BPF_AND: |
7101 | case BPF_AND | BPF_FETCH: | |
7102 | case BPF_OR: | |
7103 | case BPF_OR | BPF_FETCH: | |
7104 | case BPF_XOR: | |
7105 | case BPF_XOR | BPF_FETCH: | |
5ffa2550 BJ |
7106 | case BPF_XCHG: |
7107 | case BPF_CMPXCHG: | |
5ca419f2 BJ |
7108 | break; |
7109 | default: | |
91c960b0 BJ |
7110 | verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm); |
7111 | return -EINVAL; | |
7112 | } | |
7113 | ||
7114 | if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) { | |
7115 | verbose(env, "invalid atomic operand size\n"); | |
17a52670 AS |
7116 | return -EINVAL; |
7117 | } | |
7118 | ||
7119 | /* check src1 operand */ | |
dc503a8a | 7120 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
7121 | if (err) |
7122 | return err; | |
7123 | ||
7124 | /* check src2 operand */ | |
dc503a8a | 7125 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
7126 | if (err) |
7127 | return err; | |
7128 | ||
5ffa2550 BJ |
7129 | if (insn->imm == BPF_CMPXCHG) { |
7130 | /* Check comparison of R0 with memory location */ | |
a82fe085 DB |
7131 | const u32 aux_reg = BPF_REG_0; |
7132 | ||
7133 | err = check_reg_arg(env, aux_reg, SRC_OP); | |
5ffa2550 BJ |
7134 | if (err) |
7135 | return err; | |
a82fe085 DB |
7136 | |
7137 | if (is_pointer_value(env, aux_reg)) { | |
7138 | verbose(env, "R%d leaks addr into mem\n", aux_reg); | |
7139 | return -EACCES; | |
7140 | } | |
5ffa2550 BJ |
7141 | } |
7142 | ||
6bdf6abc | 7143 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 7144 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
7145 | return -EACCES; |
7146 | } | |
7147 | ||
ca369602 | 7148 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 7149 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
7150 | is_flow_key_reg(env, insn->dst_reg) || |
7151 | is_sk_reg(env, insn->dst_reg)) { | |
91c960b0 | 7152 | verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n", |
2a159c6f | 7153 | insn->dst_reg, |
c25b2ae1 | 7154 | reg_type_str(env, reg_state(env, insn->dst_reg)->type)); |
f37a8cb8 DB |
7155 | return -EACCES; |
7156 | } | |
7157 | ||
37086bfd BJ |
7158 | if (insn->imm & BPF_FETCH) { |
7159 | if (insn->imm == BPF_CMPXCHG) | |
7160 | load_reg = BPF_REG_0; | |
7161 | else | |
7162 | load_reg = insn->src_reg; | |
7163 | ||
7164 | /* check and record load of old value */ | |
7165 | err = check_reg_arg(env, load_reg, DST_OP); | |
7166 | if (err) | |
7167 | return err; | |
7168 | } else { | |
7169 | /* This instruction accesses a memory location but doesn't | |
7170 | * actually load it into a register. | |
7171 | */ | |
7172 | load_reg = -1; | |
7173 | } | |
7174 | ||
7d3baf0a DB |
7175 | /* Check whether we can read the memory, with second call for fetch |
7176 | * case to simulate the register fill. | |
7177 | */ | |
31fd8581 | 7178 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
1f9a1ea8 | 7179 | BPF_SIZE(insn->code), BPF_READ, -1, true, false); |
7d3baf0a DB |
7180 | if (!err && load_reg >= 0) |
7181 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, | |
7182 | BPF_SIZE(insn->code), BPF_READ, load_reg, | |
1f9a1ea8 | 7183 | true, false); |
17a52670 AS |
7184 | if (err) |
7185 | return err; | |
7186 | ||
7d3baf0a | 7187 | /* Check whether we can write into the same memory. */ |
5ca419f2 | 7188 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
1f9a1ea8 | 7189 | BPF_SIZE(insn->code), BPF_WRITE, -1, true, false); |
5ca419f2 BJ |
7190 | if (err) |
7191 | return err; | |
7192 | ||
5ca419f2 | 7193 | return 0; |
17a52670 AS |
7194 | } |
7195 | ||
01f810ac AM |
7196 | /* When register 'regno' is used to read the stack (either directly or through |
7197 | * a helper function) make sure that it's within stack boundary and, depending | |
7198 | * on the access type, that all elements of the stack are initialized. | |
7199 | * | |
7200 | * 'off' includes 'regno->off', but not its dynamic part (if any). | |
7201 | * | |
7202 | * All registers that have been spilled on the stack in the slots within the | |
7203 | * read offsets are marked as read. | |
7204 | */ | |
7205 | static int check_stack_range_initialized( | |
7206 | struct bpf_verifier_env *env, int regno, int off, | |
7207 | int access_size, bool zero_size_allowed, | |
61df10c7 | 7208 | enum bpf_access_src type, struct bpf_call_arg_meta *meta) |
2011fccf AI |
7209 | { |
7210 | struct bpf_reg_state *reg = reg_state(env, regno); | |
01f810ac AM |
7211 | struct bpf_func_state *state = func(env, reg); |
7212 | int err, min_off, max_off, i, j, slot, spi; | |
7213 | char *err_extra = type == ACCESS_HELPER ? " indirect" : ""; | |
7214 | enum bpf_access_type bounds_check_type; | |
7215 | /* Some accesses can write anything into the stack, others are | |
7216 | * read-only. | |
7217 | */ | |
7218 | bool clobber = false; | |
2011fccf | 7219 | |
01f810ac AM |
7220 | if (access_size == 0 && !zero_size_allowed) { |
7221 | verbose(env, "invalid zero-sized read\n"); | |
2011fccf AI |
7222 | return -EACCES; |
7223 | } | |
2011fccf | 7224 | |
01f810ac AM |
7225 | if (type == ACCESS_HELPER) { |
7226 | /* The bounds checks for writes are more permissive than for | |
7227 | * reads. However, if raw_mode is not set, we'll do extra | |
7228 | * checks below. | |
7229 | */ | |
7230 | bounds_check_type = BPF_WRITE; | |
7231 | clobber = true; | |
7232 | } else { | |
7233 | bounds_check_type = BPF_READ; | |
7234 | } | |
7235 | err = check_stack_access_within_bounds(env, regno, off, access_size, | |
7236 | type, bounds_check_type); | |
7237 | if (err) | |
7238 | return err; | |
7239 | ||
17a52670 | 7240 | |
2011fccf | 7241 | if (tnum_is_const(reg->var_off)) { |
01f810ac | 7242 | min_off = max_off = reg->var_off.value + off; |
2011fccf | 7243 | } else { |
088ec26d AI |
7244 | /* Variable offset is prohibited for unprivileged mode for |
7245 | * simplicity since it requires corresponding support in | |
7246 | * Spectre masking for stack ALU. | |
7247 | * See also retrieve_ptr_limit(). | |
7248 | */ | |
2c78ee89 | 7249 | if (!env->bypass_spec_v1) { |
088ec26d | 7250 | char tn_buf[48]; |
f1174f77 | 7251 | |
088ec26d | 7252 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
01f810ac AM |
7253 | verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n", |
7254 | regno, err_extra, tn_buf); | |
088ec26d AI |
7255 | return -EACCES; |
7256 | } | |
f2bcd05e AI |
7257 | /* Only initialized buffer on stack is allowed to be accessed |
7258 | * with variable offset. With uninitialized buffer it's hard to | |
7259 | * guarantee that whole memory is marked as initialized on | |
7260 | * helper return since specific bounds are unknown what may | |
7261 | * cause uninitialized stack leaking. | |
7262 | */ | |
7263 | if (meta && meta->raw_mode) | |
7264 | meta = NULL; | |
7265 | ||
01f810ac AM |
7266 | min_off = reg->smin_value + off; |
7267 | max_off = reg->smax_value + off; | |
17a52670 AS |
7268 | } |
7269 | ||
435faee1 | 7270 | if (meta && meta->raw_mode) { |
ef8fc7a0 KKD |
7271 | /* Ensure we won't be overwriting dynptrs when simulating byte |
7272 | * by byte access in check_helper_call using meta.access_size. | |
7273 | * This would be a problem if we have a helper in the future | |
7274 | * which takes: | |
7275 | * | |
7276 | * helper(uninit_mem, len, dynptr) | |
7277 | * | |
7278 | * Now, uninint_mem may overlap with dynptr pointer. Hence, it | |
7279 | * may end up writing to dynptr itself when touching memory from | |
7280 | * arg 1. This can be relaxed on a case by case basis for known | |
7281 | * safe cases, but reject due to the possibilitiy of aliasing by | |
7282 | * default. | |
7283 | */ | |
7284 | for (i = min_off; i < max_off + access_size; i++) { | |
7285 | int stack_off = -i - 1; | |
7286 | ||
7287 | spi = __get_spi(i); | |
7288 | /* raw_mode may write past allocated_stack */ | |
7289 | if (state->allocated_stack <= stack_off) | |
7290 | continue; | |
7291 | if (state->stack[spi].slot_type[stack_off % BPF_REG_SIZE] == STACK_DYNPTR) { | |
7292 | verbose(env, "potential write to dynptr at off=%d disallowed\n", i); | |
7293 | return -EACCES; | |
7294 | } | |
7295 | } | |
435faee1 DB |
7296 | meta->access_size = access_size; |
7297 | meta->regno = regno; | |
7298 | return 0; | |
7299 | } | |
7300 | ||
2011fccf | 7301 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
7302 | u8 *stype; |
7303 | ||
2011fccf | 7304 | slot = -i - 1; |
638f5b90 | 7305 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
7306 | if (state->allocated_stack <= slot) |
7307 | goto err; | |
7308 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
7309 | if (*stype == STACK_MISC) | |
7310 | goto mark; | |
6715df8d EZ |
7311 | if ((*stype == STACK_ZERO) || |
7312 | (*stype == STACK_INVALID && env->allow_uninit_stack)) { | |
01f810ac AM |
7313 | if (clobber) { |
7314 | /* helper can write anything into the stack */ | |
7315 | *stype = STACK_MISC; | |
7316 | } | |
cc2b14d5 | 7317 | goto mark; |
17a52670 | 7318 | } |
1d68f22b | 7319 | |
27113c59 | 7320 | if (is_spilled_reg(&state->stack[spi]) && |
cd17d38f YS |
7321 | (state->stack[spi].spilled_ptr.type == SCALAR_VALUE || |
7322 | env->allow_ptr_leaks)) { | |
01f810ac AM |
7323 | if (clobber) { |
7324 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); | |
7325 | for (j = 0; j < BPF_REG_SIZE; j++) | |
354e8f19 | 7326 | scrub_spilled_slot(&state->stack[spi].slot_type[j]); |
01f810ac | 7327 | } |
f7cf25b2 AS |
7328 | goto mark; |
7329 | } | |
7330 | ||
cc2b14d5 | 7331 | err: |
2011fccf | 7332 | if (tnum_is_const(reg->var_off)) { |
01f810ac AM |
7333 | verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n", |
7334 | err_extra, regno, min_off, i - min_off, access_size); | |
2011fccf AI |
7335 | } else { |
7336 | char tn_buf[48]; | |
7337 | ||
7338 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac AM |
7339 | verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n", |
7340 | err_extra, regno, tn_buf, i - min_off, access_size); | |
2011fccf | 7341 | } |
cc2b14d5 AS |
7342 | return -EACCES; |
7343 | mark: | |
7344 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
7345 | * the whole slot to be marked as 'read' | |
7346 | */ | |
679c782d | 7347 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
7348 | state->stack[spi].spilled_ptr.parent, |
7349 | REG_LIVE_READ64); | |
261f4664 KKD |
7350 | /* We do not set REG_LIVE_WRITTEN for stack slot, as we can not |
7351 | * be sure that whether stack slot is written to or not. Hence, | |
7352 | * we must still conservatively propagate reads upwards even if | |
7353 | * helper may write to the entire memory range. | |
7354 | */ | |
17a52670 | 7355 | } |
2011fccf | 7356 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
7357 | } |
7358 | ||
06c1c049 GB |
7359 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
7360 | int access_size, bool zero_size_allowed, | |
7361 | struct bpf_call_arg_meta *meta) | |
7362 | { | |
638f5b90 | 7363 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
20b2aff4 | 7364 | u32 *max_access; |
06c1c049 | 7365 | |
20b2aff4 | 7366 | switch (base_type(reg->type)) { |
06c1c049 | 7367 | case PTR_TO_PACKET: |
de8f3a83 | 7368 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
7369 | return check_packet_access(env, regno, reg->off, access_size, |
7370 | zero_size_allowed); | |
69c087ba | 7371 | case PTR_TO_MAP_KEY: |
7b3552d3 KKD |
7372 | if (meta && meta->raw_mode) { |
7373 | verbose(env, "R%d cannot write into %s\n", regno, | |
7374 | reg_type_str(env, reg->type)); | |
7375 | return -EACCES; | |
7376 | } | |
69c087ba YS |
7377 | return check_mem_region_access(env, regno, reg->off, access_size, |
7378 | reg->map_ptr->key_size, false); | |
06c1c049 | 7379 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
7380 | if (check_map_access_type(env, regno, reg->off, access_size, |
7381 | meta && meta->raw_mode ? BPF_WRITE : | |
7382 | BPF_READ)) | |
7383 | return -EACCES; | |
9fd29c08 | 7384 | return check_map_access(env, regno, reg->off, access_size, |
61df10c7 | 7385 | zero_size_allowed, ACCESS_HELPER); |
457f4436 | 7386 | case PTR_TO_MEM: |
97e6d7da KKD |
7387 | if (type_is_rdonly_mem(reg->type)) { |
7388 | if (meta && meta->raw_mode) { | |
7389 | verbose(env, "R%d cannot write into %s\n", regno, | |
7390 | reg_type_str(env, reg->type)); | |
7391 | return -EACCES; | |
7392 | } | |
7393 | } | |
457f4436 AN |
7394 | return check_mem_region_access(env, regno, reg->off, |
7395 | access_size, reg->mem_size, | |
7396 | zero_size_allowed); | |
20b2aff4 HL |
7397 | case PTR_TO_BUF: |
7398 | if (type_is_rdonly_mem(reg->type)) { | |
97e6d7da KKD |
7399 | if (meta && meta->raw_mode) { |
7400 | verbose(env, "R%d cannot write into %s\n", regno, | |
7401 | reg_type_str(env, reg->type)); | |
20b2aff4 | 7402 | return -EACCES; |
97e6d7da | 7403 | } |
20b2aff4 | 7404 | |
20b2aff4 HL |
7405 | max_access = &env->prog->aux->max_rdonly_access; |
7406 | } else { | |
20b2aff4 HL |
7407 | max_access = &env->prog->aux->max_rdwr_access; |
7408 | } | |
afbf21dc YS |
7409 | return check_buffer_access(env, reg, regno, reg->off, |
7410 | access_size, zero_size_allowed, | |
44e9a741 | 7411 | max_access); |
0d004c02 | 7412 | case PTR_TO_STACK: |
01f810ac AM |
7413 | return check_stack_range_initialized( |
7414 | env, | |
7415 | regno, reg->off, access_size, | |
7416 | zero_size_allowed, ACCESS_HELPER, meta); | |
3e30be42 AS |
7417 | case PTR_TO_BTF_ID: |
7418 | return check_ptr_to_btf_access(env, regs, regno, reg->off, | |
7419 | access_size, BPF_READ, -1); | |
15baa55f BT |
7420 | case PTR_TO_CTX: |
7421 | /* in case the function doesn't know how to access the context, | |
7422 | * (because we are in a program of type SYSCALL for example), we | |
7423 | * can not statically check its size. | |
7424 | * Dynamically check it now. | |
7425 | */ | |
7426 | if (!env->ops->convert_ctx_access) { | |
7427 | enum bpf_access_type atype = meta && meta->raw_mode ? BPF_WRITE : BPF_READ; | |
7428 | int offset = access_size - 1; | |
7429 | ||
7430 | /* Allow zero-byte read from PTR_TO_CTX */ | |
7431 | if (access_size == 0) | |
7432 | return zero_size_allowed ? 0 : -EACCES; | |
7433 | ||
7434 | return check_mem_access(env, env->insn_idx, regno, offset, BPF_B, | |
1f9a1ea8 | 7435 | atype, -1, false, false); |
15baa55f BT |
7436 | } |
7437 | ||
7438 | fallthrough; | |
0d004c02 LB |
7439 | default: /* scalar_value or invalid ptr */ |
7440 | /* Allow zero-byte read from NULL, regardless of pointer type */ | |
7441 | if (zero_size_allowed && access_size == 0 && | |
7442 | register_is_null(reg)) | |
7443 | return 0; | |
7444 | ||
c25b2ae1 HL |
7445 | verbose(env, "R%d type=%s ", regno, |
7446 | reg_type_str(env, reg->type)); | |
7447 | verbose(env, "expected=%s\n", reg_type_str(env, PTR_TO_STACK)); | |
0d004c02 | 7448 | return -EACCES; |
06c1c049 GB |
7449 | } |
7450 | } | |
7451 | ||
d583691c KKD |
7452 | static int check_mem_size_reg(struct bpf_verifier_env *env, |
7453 | struct bpf_reg_state *reg, u32 regno, | |
7454 | bool zero_size_allowed, | |
7455 | struct bpf_call_arg_meta *meta) | |
7456 | { | |
7457 | int err; | |
7458 | ||
7459 | /* This is used to refine r0 return value bounds for helpers | |
7460 | * that enforce this value as an upper bound on return values. | |
7461 | * See do_refine_retval_range() for helpers that can refine | |
7462 | * the return value. C type of helper is u32 so we pull register | |
7463 | * bound from umax_value however, if negative verifier errors | |
7464 | * out. Only upper bounds can be learned because retval is an | |
7465 | * int type and negative retvals are allowed. | |
7466 | */ | |
be77354a | 7467 | meta->msize_max_value = reg->umax_value; |
d583691c KKD |
7468 | |
7469 | /* The register is SCALAR_VALUE; the access check | |
7470 | * happens using its boundaries. | |
7471 | */ | |
7472 | if (!tnum_is_const(reg->var_off)) | |
7473 | /* For unprivileged variable accesses, disable raw | |
7474 | * mode so that the program is required to | |
7475 | * initialize all the memory that the helper could | |
7476 | * just partially fill up. | |
7477 | */ | |
7478 | meta = NULL; | |
7479 | ||
7480 | if (reg->smin_value < 0) { | |
7481 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", | |
7482 | regno); | |
7483 | return -EACCES; | |
7484 | } | |
7485 | ||
7486 | if (reg->umin_value == 0) { | |
7487 | err = check_helper_mem_access(env, regno - 1, 0, | |
7488 | zero_size_allowed, | |
7489 | meta); | |
7490 | if (err) | |
7491 | return err; | |
7492 | } | |
7493 | ||
7494 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { | |
7495 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", | |
7496 | regno); | |
7497 | return -EACCES; | |
7498 | } | |
7499 | err = check_helper_mem_access(env, regno - 1, | |
7500 | reg->umax_value, | |
7501 | zero_size_allowed, meta); | |
7502 | if (!err) | |
7503 | err = mark_chain_precision(env, regno); | |
7504 | return err; | |
7505 | } | |
7506 | ||
e5069b9c DB |
7507 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
7508 | u32 regno, u32 mem_size) | |
7509 | { | |
be77354a KKD |
7510 | bool may_be_null = type_may_be_null(reg->type); |
7511 | struct bpf_reg_state saved_reg; | |
7512 | struct bpf_call_arg_meta meta; | |
7513 | int err; | |
7514 | ||
e5069b9c DB |
7515 | if (register_is_null(reg)) |
7516 | return 0; | |
7517 | ||
be77354a KKD |
7518 | memset(&meta, 0, sizeof(meta)); |
7519 | /* Assuming that the register contains a value check if the memory | |
7520 | * access is safe. Temporarily save and restore the register's state as | |
7521 | * the conversion shouldn't be visible to a caller. | |
7522 | */ | |
7523 | if (may_be_null) { | |
7524 | saved_reg = *reg; | |
e5069b9c | 7525 | mark_ptr_not_null_reg(reg); |
e5069b9c DB |
7526 | } |
7527 | ||
be77354a KKD |
7528 | err = check_helper_mem_access(env, regno, mem_size, true, &meta); |
7529 | /* Check access for BPF_WRITE */ | |
7530 | meta.raw_mode = true; | |
7531 | err = err ?: check_helper_mem_access(env, regno, mem_size, true, &meta); | |
7532 | ||
7533 | if (may_be_null) | |
7534 | *reg = saved_reg; | |
7535 | ||
7536 | return err; | |
e5069b9c DB |
7537 | } |
7538 | ||
00b85860 KKD |
7539 | static int check_kfunc_mem_size_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
7540 | u32 regno) | |
d583691c KKD |
7541 | { |
7542 | struct bpf_reg_state *mem_reg = &cur_regs(env)[regno - 1]; | |
7543 | bool may_be_null = type_may_be_null(mem_reg->type); | |
7544 | struct bpf_reg_state saved_reg; | |
be77354a | 7545 | struct bpf_call_arg_meta meta; |
d583691c KKD |
7546 | int err; |
7547 | ||
7548 | WARN_ON_ONCE(regno < BPF_REG_2 || regno > BPF_REG_5); | |
7549 | ||
be77354a KKD |
7550 | memset(&meta, 0, sizeof(meta)); |
7551 | ||
d583691c KKD |
7552 | if (may_be_null) { |
7553 | saved_reg = *mem_reg; | |
7554 | mark_ptr_not_null_reg(mem_reg); | |
7555 | } | |
7556 | ||
be77354a KKD |
7557 | err = check_mem_size_reg(env, reg, regno, true, &meta); |
7558 | /* Check access for BPF_WRITE */ | |
7559 | meta.raw_mode = true; | |
7560 | err = err ?: check_mem_size_reg(env, reg, regno, true, &meta); | |
d583691c KKD |
7561 | |
7562 | if (may_be_null) | |
7563 | *mem_reg = saved_reg; | |
7564 | return err; | |
7565 | } | |
7566 | ||
d83525ca | 7567 | /* Implementation details: |
4e814da0 KKD |
7568 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL. |
7569 | * bpf_obj_new returns PTR_TO_BTF_ID | MEM_ALLOC | PTR_MAYBE_NULL. | |
d83525ca | 7570 | * Two bpf_map_lookups (even with the same key) will have different reg->id. |
4e814da0 KKD |
7571 | * Two separate bpf_obj_new will also have different reg->id. |
7572 | * For traditional PTR_TO_MAP_VALUE or PTR_TO_BTF_ID | MEM_ALLOC, the verifier | |
7573 | * clears reg->id after value_or_null->value transition, since the verifier only | |
7574 | * cares about the range of access to valid map value pointer and doesn't care | |
7575 | * about actual address of the map element. | |
d83525ca AS |
7576 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps |
7577 | * reg->id > 0 after value_or_null->value transition. By doing so | |
7578 | * two bpf_map_lookups will be considered two different pointers that | |
4e814da0 KKD |
7579 | * point to different bpf_spin_locks. Likewise for pointers to allocated objects |
7580 | * returned from bpf_obj_new. | |
d83525ca AS |
7581 | * The verifier allows taking only one bpf_spin_lock at a time to avoid |
7582 | * dead-locks. | |
7583 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
7584 | * reg_is_refcounted() logic. The verifier needs to remember only | |
7585 | * one spin_lock instead of array of acquired_refs. | |
d0d78c1d | 7586 | * cur_state->active_lock remembers which map value element or allocated |
4e814da0 | 7587 | * object got locked and clears it after bpf_spin_unlock. |
d83525ca AS |
7588 | */ |
7589 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
7590 | bool is_lock) | |
7591 | { | |
7592 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7593 | struct bpf_verifier_state *cur = env->cur_state; | |
7594 | bool is_const = tnum_is_const(reg->var_off); | |
d83525ca | 7595 | u64 val = reg->var_off.value; |
4e814da0 KKD |
7596 | struct bpf_map *map = NULL; |
7597 | struct btf *btf = NULL; | |
7598 | struct btf_record *rec; | |
d83525ca | 7599 | |
d83525ca AS |
7600 | if (!is_const) { |
7601 | verbose(env, | |
7602 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
7603 | regno); | |
7604 | return -EINVAL; | |
7605 | } | |
4e814da0 KKD |
7606 | if (reg->type == PTR_TO_MAP_VALUE) { |
7607 | map = reg->map_ptr; | |
7608 | if (!map->btf) { | |
7609 | verbose(env, | |
7610 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
7611 | map->name); | |
7612 | return -EINVAL; | |
7613 | } | |
7614 | } else { | |
7615 | btf = reg->btf; | |
d83525ca | 7616 | } |
4e814da0 KKD |
7617 | |
7618 | rec = reg_btf_record(reg); | |
7619 | if (!btf_record_has_field(rec, BPF_SPIN_LOCK)) { | |
7620 | verbose(env, "%s '%s' has no valid bpf_spin_lock\n", map ? "map" : "local", | |
7621 | map ? map->name : "kptr"); | |
d83525ca AS |
7622 | return -EINVAL; |
7623 | } | |
4e814da0 | 7624 | if (rec->spin_lock_off != val + reg->off) { |
db559117 | 7625 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock' that is at %d\n", |
4e814da0 | 7626 | val + reg->off, rec->spin_lock_off); |
d83525ca AS |
7627 | return -EINVAL; |
7628 | } | |
7629 | if (is_lock) { | |
d0d78c1d | 7630 | if (cur->active_lock.ptr) { |
d83525ca AS |
7631 | verbose(env, |
7632 | "Locking two bpf_spin_locks are not allowed\n"); | |
7633 | return -EINVAL; | |
7634 | } | |
d0d78c1d KKD |
7635 | if (map) |
7636 | cur->active_lock.ptr = map; | |
7637 | else | |
7638 | cur->active_lock.ptr = btf; | |
7639 | cur->active_lock.id = reg->id; | |
d83525ca | 7640 | } else { |
d0d78c1d KKD |
7641 | void *ptr; |
7642 | ||
7643 | if (map) | |
7644 | ptr = map; | |
7645 | else | |
7646 | ptr = btf; | |
7647 | ||
7648 | if (!cur->active_lock.ptr) { | |
d83525ca AS |
7649 | verbose(env, "bpf_spin_unlock without taking a lock\n"); |
7650 | return -EINVAL; | |
7651 | } | |
d0d78c1d KKD |
7652 | if (cur->active_lock.ptr != ptr || |
7653 | cur->active_lock.id != reg->id) { | |
d83525ca AS |
7654 | verbose(env, "bpf_spin_unlock of different lock\n"); |
7655 | return -EINVAL; | |
7656 | } | |
534e86bc | 7657 | |
6a3cd331 | 7658 | invalidate_non_owning_refs(env); |
534e86bc | 7659 | |
6a3cd331 DM |
7660 | cur->active_lock.ptr = NULL; |
7661 | cur->active_lock.id = 0; | |
d83525ca AS |
7662 | } |
7663 | return 0; | |
7664 | } | |
7665 | ||
b00628b1 AS |
7666 | static int process_timer_func(struct bpf_verifier_env *env, int regno, |
7667 | struct bpf_call_arg_meta *meta) | |
7668 | { | |
7669 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7670 | bool is_const = tnum_is_const(reg->var_off); | |
7671 | struct bpf_map *map = reg->map_ptr; | |
7672 | u64 val = reg->var_off.value; | |
7673 | ||
7674 | if (!is_const) { | |
7675 | verbose(env, | |
7676 | "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n", | |
7677 | regno); | |
7678 | return -EINVAL; | |
7679 | } | |
7680 | if (!map->btf) { | |
7681 | verbose(env, "map '%s' has to have BTF in order to use bpf_timer\n", | |
7682 | map->name); | |
7683 | return -EINVAL; | |
7684 | } | |
db559117 KKD |
7685 | if (!btf_record_has_field(map->record, BPF_TIMER)) { |
7686 | verbose(env, "map '%s' has no valid bpf_timer\n", map->name); | |
68134668 AS |
7687 | return -EINVAL; |
7688 | } | |
db559117 | 7689 | if (map->record->timer_off != val + reg->off) { |
68134668 | 7690 | verbose(env, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n", |
db559117 | 7691 | val + reg->off, map->record->timer_off); |
b00628b1 AS |
7692 | return -EINVAL; |
7693 | } | |
7694 | if (meta->map_ptr) { | |
7695 | verbose(env, "verifier bug. Two map pointers in a timer helper\n"); | |
7696 | return -EFAULT; | |
7697 | } | |
3e8ce298 | 7698 | meta->map_uid = reg->map_uid; |
b00628b1 AS |
7699 | meta->map_ptr = map; |
7700 | return 0; | |
7701 | } | |
7702 | ||
c0a5a21c KKD |
7703 | static int process_kptr_func(struct bpf_verifier_env *env, int regno, |
7704 | struct bpf_call_arg_meta *meta) | |
7705 | { | |
7706 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
c0a5a21c | 7707 | struct bpf_map *map_ptr = reg->map_ptr; |
aa3496ac | 7708 | struct btf_field *kptr_field; |
c0a5a21c | 7709 | u32 kptr_off; |
c0a5a21c KKD |
7710 | |
7711 | if (!tnum_is_const(reg->var_off)) { | |
7712 | verbose(env, | |
7713 | "R%d doesn't have constant offset. kptr has to be at the constant offset\n", | |
7714 | regno); | |
7715 | return -EINVAL; | |
7716 | } | |
7717 | if (!map_ptr->btf) { | |
7718 | verbose(env, "map '%s' has to have BTF in order to use bpf_kptr_xchg\n", | |
7719 | map_ptr->name); | |
7720 | return -EINVAL; | |
7721 | } | |
aa3496ac KKD |
7722 | if (!btf_record_has_field(map_ptr->record, BPF_KPTR)) { |
7723 | verbose(env, "map '%s' has no valid kptr\n", map_ptr->name); | |
c0a5a21c KKD |
7724 | return -EINVAL; |
7725 | } | |
7726 | ||
7727 | meta->map_ptr = map_ptr; | |
7728 | kptr_off = reg->off + reg->var_off.value; | |
aa3496ac KKD |
7729 | kptr_field = btf_record_find(map_ptr->record, kptr_off, BPF_KPTR); |
7730 | if (!kptr_field) { | |
c0a5a21c KKD |
7731 | verbose(env, "off=%d doesn't point to kptr\n", kptr_off); |
7732 | return -EACCES; | |
7733 | } | |
36d8bdf7 | 7734 | if (kptr_field->type != BPF_KPTR_REF && kptr_field->type != BPF_KPTR_PERCPU) { |
c0a5a21c KKD |
7735 | verbose(env, "off=%d kptr isn't referenced kptr\n", kptr_off); |
7736 | return -EACCES; | |
7737 | } | |
aa3496ac | 7738 | meta->kptr_field = kptr_field; |
c0a5a21c KKD |
7739 | return 0; |
7740 | } | |
7741 | ||
27060531 KKD |
7742 | /* There are two register types representing a bpf_dynptr, one is PTR_TO_STACK |
7743 | * which points to a stack slot, and the other is CONST_PTR_TO_DYNPTR. | |
7744 | * | |
7745 | * In both cases we deal with the first 8 bytes, but need to mark the next 8 | |
7746 | * bytes as STACK_DYNPTR in case of PTR_TO_STACK. In case of | |
7747 | * CONST_PTR_TO_DYNPTR, we are guaranteed to get the beginning of the object. | |
7748 | * | |
7749 | * Mutability of bpf_dynptr is at two levels, one is at the level of struct | |
7750 | * bpf_dynptr itself, i.e. whether the helper is receiving a pointer to struct | |
7751 | * bpf_dynptr or pointer to const struct bpf_dynptr. In the former case, it can | |
7752 | * mutate the view of the dynptr and also possibly destroy it. In the latter | |
7753 | * case, it cannot mutate the bpf_dynptr itself but it can still mutate the | |
7754 | * memory that dynptr points to. | |
7755 | * | |
7756 | * The verifier will keep track both levels of mutation (bpf_dynptr's in | |
7757 | * reg->type and the memory's in reg->dynptr.type), but there is no support for | |
7758 | * readonly dynptr view yet, hence only the first case is tracked and checked. | |
7759 | * | |
7760 | * This is consistent with how C applies the const modifier to a struct object, | |
7761 | * where the pointer itself inside bpf_dynptr becomes const but not what it | |
7762 | * points to. | |
7763 | * | |
7764 | * Helpers which do not mutate the bpf_dynptr set MEM_RDONLY in their argument | |
7765 | * type, and declare it as 'const struct bpf_dynptr *' in their prototype. | |
7766 | */ | |
1d18feb2 | 7767 | static int process_dynptr_func(struct bpf_verifier_env *env, int regno, int insn_idx, |
361f129f | 7768 | enum bpf_arg_type arg_type, int clone_ref_obj_id) |
6b75bd3d KKD |
7769 | { |
7770 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
1d18feb2 | 7771 | int err; |
6b75bd3d | 7772 | |
27060531 KKD |
7773 | /* MEM_UNINIT and MEM_RDONLY are exclusive, when applied to an |
7774 | * ARG_PTR_TO_DYNPTR (or ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_*): | |
7775 | */ | |
7776 | if ((arg_type & (MEM_UNINIT | MEM_RDONLY)) == (MEM_UNINIT | MEM_RDONLY)) { | |
7777 | verbose(env, "verifier internal error: misconfigured dynptr helper type flags\n"); | |
7778 | return -EFAULT; | |
7779 | } | |
79168a66 | 7780 | |
27060531 KKD |
7781 | /* MEM_UNINIT - Points to memory that is an appropriate candidate for |
7782 | * constructing a mutable bpf_dynptr object. | |
7783 | * | |
7784 | * Currently, this is only possible with PTR_TO_STACK | |
7785 | * pointing to a region of at least 16 bytes which doesn't | |
7786 | * contain an existing bpf_dynptr. | |
7787 | * | |
7788 | * MEM_RDONLY - Points to a initialized bpf_dynptr that will not be | |
7789 | * mutated or destroyed. However, the memory it points to | |
7790 | * may be mutated. | |
7791 | * | |
7792 | * None - Points to a initialized dynptr that can be mutated and | |
7793 | * destroyed, including mutation of the memory it points | |
7794 | * to. | |
6b75bd3d | 7795 | */ |
6b75bd3d | 7796 | if (arg_type & MEM_UNINIT) { |
1d18feb2 JK |
7797 | int i; |
7798 | ||
7e0dac28 | 7799 | if (!is_dynptr_reg_valid_uninit(env, reg)) { |
6b75bd3d KKD |
7800 | verbose(env, "Dynptr has to be an uninitialized dynptr\n"); |
7801 | return -EINVAL; | |
7802 | } | |
7803 | ||
1d18feb2 JK |
7804 | /* we write BPF_DW bits (8 bytes) at a time */ |
7805 | for (i = 0; i < BPF_DYNPTR_SIZE; i += 8) { | |
7806 | err = check_mem_access(env, insn_idx, regno, | |
1f9a1ea8 | 7807 | i, BPF_DW, BPF_WRITE, -1, false, false); |
1d18feb2 JK |
7808 | if (err) |
7809 | return err; | |
6b75bd3d KKD |
7810 | } |
7811 | ||
361f129f | 7812 | err = mark_stack_slots_dynptr(env, reg, arg_type, insn_idx, clone_ref_obj_id); |
27060531 KKD |
7813 | } else /* MEM_RDONLY and None case from above */ { |
7814 | /* For the reg->type == PTR_TO_STACK case, bpf_dynptr is never const */ | |
7815 | if (reg->type == CONST_PTR_TO_DYNPTR && !(arg_type & MEM_RDONLY)) { | |
7816 | verbose(env, "cannot pass pointer to const bpf_dynptr, the helper mutates it\n"); | |
7817 | return -EINVAL; | |
7818 | } | |
7819 | ||
7e0dac28 | 7820 | if (!is_dynptr_reg_valid_init(env, reg)) { |
6b75bd3d KKD |
7821 | verbose(env, |
7822 | "Expected an initialized dynptr as arg #%d\n", | |
7823 | regno); | |
7824 | return -EINVAL; | |
7825 | } | |
7826 | ||
27060531 KKD |
7827 | /* Fold modifiers (in this case, MEM_RDONLY) when checking expected type */ |
7828 | if (!is_dynptr_type_expected(env, reg, arg_type & ~MEM_RDONLY)) { | |
6b75bd3d KKD |
7829 | verbose(env, |
7830 | "Expected a dynptr of type %s as arg #%d\n", | |
d54e0f6c | 7831 | dynptr_type_str(arg_to_dynptr_type(arg_type)), regno); |
6b75bd3d KKD |
7832 | return -EINVAL; |
7833 | } | |
d6fefa11 KKD |
7834 | |
7835 | err = mark_dynptr_read(env, reg); | |
6b75bd3d | 7836 | } |
1d18feb2 | 7837 | return err; |
6b75bd3d KKD |
7838 | } |
7839 | ||
06accc87 AN |
7840 | static u32 iter_ref_obj_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg, int spi) |
7841 | { | |
7842 | struct bpf_func_state *state = func(env, reg); | |
7843 | ||
7844 | return state->stack[spi].spilled_ptr.ref_obj_id; | |
7845 | } | |
7846 | ||
7847 | static bool is_iter_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7848 | { | |
7849 | return meta->kfunc_flags & (KF_ITER_NEW | KF_ITER_NEXT | KF_ITER_DESTROY); | |
7850 | } | |
7851 | ||
7852 | static bool is_iter_new_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7853 | { | |
7854 | return meta->kfunc_flags & KF_ITER_NEW; | |
7855 | } | |
7856 | ||
7857 | static bool is_iter_next_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7858 | { | |
7859 | return meta->kfunc_flags & KF_ITER_NEXT; | |
7860 | } | |
7861 | ||
7862 | static bool is_iter_destroy_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7863 | { | |
7864 | return meta->kfunc_flags & KF_ITER_DESTROY; | |
7865 | } | |
7866 | ||
7867 | static bool is_kfunc_arg_iter(struct bpf_kfunc_call_arg_meta *meta, int arg) | |
7868 | { | |
7869 | /* btf_check_iter_kfuncs() guarantees that first argument of any iter | |
7870 | * kfunc is iter state pointer | |
7871 | */ | |
7872 | return arg == 0 && is_iter_kfunc(meta); | |
7873 | } | |
7874 | ||
7875 | static int process_iter_arg(struct bpf_verifier_env *env, int regno, int insn_idx, | |
7876 | struct bpf_kfunc_call_arg_meta *meta) | |
7877 | { | |
7878 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7879 | const struct btf_type *t; | |
7880 | const struct btf_param *arg; | |
7881 | int spi, err, i, nr_slots; | |
7882 | u32 btf_id; | |
7883 | ||
7884 | /* btf_check_iter_kfuncs() ensures we don't need to validate anything here */ | |
7885 | arg = &btf_params(meta->func_proto)[0]; | |
7886 | t = btf_type_skip_modifiers(meta->btf, arg->type, NULL); /* PTR */ | |
7887 | t = btf_type_skip_modifiers(meta->btf, t->type, &btf_id); /* STRUCT */ | |
7888 | nr_slots = t->size / BPF_REG_SIZE; | |
7889 | ||
06accc87 AN |
7890 | if (is_iter_new_kfunc(meta)) { |
7891 | /* bpf_iter_<type>_new() expects pointer to uninit iter state */ | |
7892 | if (!is_iter_reg_valid_uninit(env, reg, nr_slots)) { | |
7893 | verbose(env, "expected uninitialized iter_%s as arg #%d\n", | |
7894 | iter_type_str(meta->btf, btf_id), regno); | |
7895 | return -EINVAL; | |
7896 | } | |
7897 | ||
7898 | for (i = 0; i < nr_slots * 8; i += BPF_REG_SIZE) { | |
7899 | err = check_mem_access(env, insn_idx, regno, | |
1f9a1ea8 | 7900 | i, BPF_DW, BPF_WRITE, -1, false, false); |
06accc87 AN |
7901 | if (err) |
7902 | return err; | |
7903 | } | |
7904 | ||
dfab99df | 7905 | err = mark_stack_slots_iter(env, meta, reg, insn_idx, meta->btf, btf_id, nr_slots); |
06accc87 AN |
7906 | if (err) |
7907 | return err; | |
7908 | } else { | |
7909 | /* iter_next() or iter_destroy() expect initialized iter state*/ | |
dfab99df CZ |
7910 | err = is_iter_reg_valid_init(env, reg, meta->btf, btf_id, nr_slots); |
7911 | switch (err) { | |
7912 | case 0: | |
7913 | break; | |
7914 | case -EINVAL: | |
06accc87 AN |
7915 | verbose(env, "expected an initialized iter_%s as arg #%d\n", |
7916 | iter_type_str(meta->btf, btf_id), regno); | |
dfab99df CZ |
7917 | return err; |
7918 | case -EPROTO: | |
7919 | verbose(env, "expected an RCU CS when using %s\n", meta->func_name); | |
7920 | return err; | |
7921 | default: | |
7922 | return err; | |
06accc87 AN |
7923 | } |
7924 | ||
b63cbc49 AN |
7925 | spi = iter_get_spi(env, reg, nr_slots); |
7926 | if (spi < 0) | |
7927 | return spi; | |
7928 | ||
06accc87 AN |
7929 | err = mark_iter_read(env, reg, spi, nr_slots); |
7930 | if (err) | |
7931 | return err; | |
7932 | ||
b63cbc49 AN |
7933 | /* remember meta->iter info for process_iter_next_call() */ |
7934 | meta->iter.spi = spi; | |
7935 | meta->iter.frameno = reg->frameno; | |
06accc87 AN |
7936 | meta->ref_obj_id = iter_ref_obj_id(env, reg, spi); |
7937 | ||
7938 | if (is_iter_destroy_kfunc(meta)) { | |
7939 | err = unmark_stack_slots_iter(env, reg, nr_slots); | |
7940 | if (err) | |
7941 | return err; | |
7942 | } | |
7943 | } | |
7944 | ||
7945 | return 0; | |
7946 | } | |
7947 | ||
2793a8b0 EZ |
7948 | /* Look for a previous loop entry at insn_idx: nearest parent state |
7949 | * stopped at insn_idx with callsites matching those in cur->frame. | |
7950 | */ | |
7951 | static struct bpf_verifier_state *find_prev_entry(struct bpf_verifier_env *env, | |
7952 | struct bpf_verifier_state *cur, | |
7953 | int insn_idx) | |
7954 | { | |
7955 | struct bpf_verifier_state_list *sl; | |
7956 | struct bpf_verifier_state *st; | |
7957 | ||
7958 | /* Explored states are pushed in stack order, most recent states come first */ | |
7959 | sl = *explored_state(env, insn_idx); | |
7960 | for (; sl; sl = sl->next) { | |
7961 | /* If st->branches != 0 state is a part of current DFS verification path, | |
7962 | * hence cur & st for a loop. | |
7963 | */ | |
7964 | st = &sl->state; | |
7965 | if (st->insn_idx == insn_idx && st->branches && same_callsites(st, cur) && | |
7966 | st->dfs_depth < cur->dfs_depth) | |
7967 | return st; | |
7968 | } | |
7969 | ||
7970 | return NULL; | |
7971 | } | |
7972 | ||
7973 | static void reset_idmap_scratch(struct bpf_verifier_env *env); | |
7974 | static bool regs_exact(const struct bpf_reg_state *rold, | |
7975 | const struct bpf_reg_state *rcur, | |
7976 | struct bpf_idmap *idmap); | |
7977 | ||
7978 | static void maybe_widen_reg(struct bpf_verifier_env *env, | |
7979 | struct bpf_reg_state *rold, struct bpf_reg_state *rcur, | |
7980 | struct bpf_idmap *idmap) | |
7981 | { | |
7982 | if (rold->type != SCALAR_VALUE) | |
7983 | return; | |
7984 | if (rold->type != rcur->type) | |
7985 | return; | |
7986 | if (rold->precise || rcur->precise || regs_exact(rold, rcur, idmap)) | |
7987 | return; | |
7988 | __mark_reg_unknown(env, rcur); | |
7989 | } | |
7990 | ||
7991 | static int widen_imprecise_scalars(struct bpf_verifier_env *env, | |
7992 | struct bpf_verifier_state *old, | |
7993 | struct bpf_verifier_state *cur) | |
7994 | { | |
7995 | struct bpf_func_state *fold, *fcur; | |
7996 | int i, fr; | |
7997 | ||
7998 | reset_idmap_scratch(env); | |
7999 | for (fr = old->curframe; fr >= 0; fr--) { | |
8000 | fold = old->frame[fr]; | |
8001 | fcur = cur->frame[fr]; | |
8002 | ||
8003 | for (i = 0; i < MAX_BPF_REG; i++) | |
8004 | maybe_widen_reg(env, | |
8005 | &fold->regs[i], | |
8006 | &fcur->regs[i], | |
8007 | &env->idmap_scratch); | |
8008 | ||
8009 | for (i = 0; i < fold->allocated_stack / BPF_REG_SIZE; i++) { | |
8010 | if (!is_spilled_reg(&fold->stack[i]) || | |
8011 | !is_spilled_reg(&fcur->stack[i])) | |
8012 | continue; | |
8013 | ||
8014 | maybe_widen_reg(env, | |
8015 | &fold->stack[i].spilled_ptr, | |
8016 | &fcur->stack[i].spilled_ptr, | |
8017 | &env->idmap_scratch); | |
8018 | } | |
8019 | } | |
8020 | return 0; | |
8021 | } | |
8022 | ||
06accc87 AN |
8023 | /* process_iter_next_call() is called when verifier gets to iterator's next |
8024 | * "method" (e.g., bpf_iter_num_next() for numbers iterator) call. We'll refer | |
8025 | * to it as just "iter_next()" in comments below. | |
8026 | * | |
8027 | * BPF verifier relies on a crucial contract for any iter_next() | |
8028 | * implementation: it should *eventually* return NULL, and once that happens | |
8029 | * it should keep returning NULL. That is, once iterator exhausts elements to | |
8030 | * iterate, it should never reset or spuriously return new elements. | |
8031 | * | |
8032 | * With the assumption of such contract, process_iter_next_call() simulates | |
8033 | * a fork in the verifier state to validate loop logic correctness and safety | |
8034 | * without having to simulate infinite amount of iterations. | |
8035 | * | |
8036 | * In current state, we first assume that iter_next() returned NULL and | |
8037 | * iterator state is set to DRAINED (BPF_ITER_STATE_DRAINED). In such | |
8038 | * conditions we should not form an infinite loop and should eventually reach | |
8039 | * exit. | |
8040 | * | |
8041 | * Besides that, we also fork current state and enqueue it for later | |
8042 | * verification. In a forked state we keep iterator state as ACTIVE | |
8043 | * (BPF_ITER_STATE_ACTIVE) and assume non-NULL return from iter_next(). We | |
8044 | * also bump iteration depth to prevent erroneous infinite loop detection | |
8045 | * later on (see iter_active_depths_differ() comment for details). In this | |
8046 | * state we assume that we'll eventually loop back to another iter_next() | |
8047 | * calls (it could be in exactly same location or in some other instruction, | |
8048 | * it doesn't matter, we don't make any unnecessary assumptions about this, | |
8049 | * everything revolves around iterator state in a stack slot, not which | |
8050 | * instruction is calling iter_next()). When that happens, we either will come | |
8051 | * to iter_next() with equivalent state and can conclude that next iteration | |
8052 | * will proceed in exactly the same way as we just verified, so it's safe to | |
8053 | * assume that loop converges. If not, we'll go on another iteration | |
8054 | * simulation with a different input state, until all possible starting states | |
8055 | * are validated or we reach maximum number of instructions limit. | |
8056 | * | |
8057 | * This way, we will either exhaustively discover all possible input states | |
8058 | * that iterator loop can start with and eventually will converge, or we'll | |
8059 | * effectively regress into bounded loop simulation logic and either reach | |
8060 | * maximum number of instructions if loop is not provably convergent, or there | |
8061 | * is some statically known limit on number of iterations (e.g., if there is | |
8062 | * an explicit `if n > 100 then break;` statement somewhere in the loop). | |
8063 | * | |
2793a8b0 EZ |
8064 | * Iteration convergence logic in is_state_visited() relies on exact |
8065 | * states comparison, which ignores read and precision marks. | |
8066 | * This is necessary because read and precision marks are not finalized | |
8067 | * while in the loop. Exact comparison might preclude convergence for | |
8068 | * simple programs like below: | |
8069 | * | |
8070 | * i = 0; | |
8071 | * while(iter_next(&it)) | |
8072 | * i++; | |
8073 | * | |
8074 | * At each iteration step i++ would produce a new distinct state and | |
8075 | * eventually instruction processing limit would be reached. | |
8076 | * | |
8077 | * To avoid such behavior speculatively forget (widen) range for | |
8078 | * imprecise scalar registers, if those registers were not precise at the | |
8079 | * end of the previous iteration and do not match exactly. | |
8080 | * | |
8081 | * This is a conservative heuristic that allows to verify wide range of programs, | |
8082 | * however it precludes verification of programs that conjure an | |
8083 | * imprecise value on the first loop iteration and use it as precise on a second. | |
8084 | * For example, the following safe program would fail to verify: | |
8085 | * | |
8086 | * struct bpf_num_iter it; | |
8087 | * int arr[10]; | |
8088 | * int i = 0, a = 0; | |
8089 | * bpf_iter_num_new(&it, 0, 10); | |
8090 | * while (bpf_iter_num_next(&it)) { | |
8091 | * if (a == 0) { | |
8092 | * a = 1; | |
8093 | * i = 7; // Because i changed verifier would forget | |
8094 | * // it's range on second loop entry. | |
8095 | * } else { | |
8096 | * arr[i] = 42; // This would fail to verify. | |
8097 | * } | |
8098 | * } | |
8099 | * bpf_iter_num_destroy(&it); | |
06accc87 AN |
8100 | */ |
8101 | static int process_iter_next_call(struct bpf_verifier_env *env, int insn_idx, | |
8102 | struct bpf_kfunc_call_arg_meta *meta) | |
8103 | { | |
2793a8b0 | 8104 | struct bpf_verifier_state *cur_st = env->cur_state, *queued_st, *prev_st; |
06accc87 AN |
8105 | struct bpf_func_state *cur_fr = cur_st->frame[cur_st->curframe], *queued_fr; |
8106 | struct bpf_reg_state *cur_iter, *queued_iter; | |
8107 | int iter_frameno = meta->iter.frameno; | |
8108 | int iter_spi = meta->iter.spi; | |
8109 | ||
8110 | BTF_TYPE_EMIT(struct bpf_iter); | |
8111 | ||
8112 | cur_iter = &env->cur_state->frame[iter_frameno]->stack[iter_spi].spilled_ptr; | |
8113 | ||
8114 | if (cur_iter->iter.state != BPF_ITER_STATE_ACTIVE && | |
8115 | cur_iter->iter.state != BPF_ITER_STATE_DRAINED) { | |
8116 | verbose(env, "verifier internal error: unexpected iterator state %d (%s)\n", | |
8117 | cur_iter->iter.state, iter_state_str(cur_iter->iter.state)); | |
8118 | return -EFAULT; | |
8119 | } | |
8120 | ||
8121 | if (cur_iter->iter.state == BPF_ITER_STATE_ACTIVE) { | |
2793a8b0 EZ |
8122 | /* Because iter_next() call is a checkpoint is_state_visitied() |
8123 | * should guarantee parent state with same call sites and insn_idx. | |
8124 | */ | |
8125 | if (!cur_st->parent || cur_st->parent->insn_idx != insn_idx || | |
8126 | !same_callsites(cur_st->parent, cur_st)) { | |
8127 | verbose(env, "bug: bad parent state for iter next call"); | |
8128 | return -EFAULT; | |
8129 | } | |
8130 | /* Note cur_st->parent in the call below, it is necessary to skip | |
8131 | * checkpoint created for cur_st by is_state_visited() | |
8132 | * right at this instruction. | |
8133 | */ | |
8134 | prev_st = find_prev_entry(env, cur_st->parent, insn_idx); | |
06accc87 AN |
8135 | /* branch out active iter state */ |
8136 | queued_st = push_stack(env, insn_idx + 1, insn_idx, false); | |
8137 | if (!queued_st) | |
8138 | return -ENOMEM; | |
8139 | ||
8140 | queued_iter = &queued_st->frame[iter_frameno]->stack[iter_spi].spilled_ptr; | |
8141 | queued_iter->iter.state = BPF_ITER_STATE_ACTIVE; | |
8142 | queued_iter->iter.depth++; | |
2793a8b0 EZ |
8143 | if (prev_st) |
8144 | widen_imprecise_scalars(env, prev_st, queued_st); | |
06accc87 AN |
8145 | |
8146 | queued_fr = queued_st->frame[queued_st->curframe]; | |
8147 | mark_ptr_not_null_reg(&queued_fr->regs[BPF_REG_0]); | |
8148 | } | |
8149 | ||
8150 | /* switch to DRAINED state, but keep the depth unchanged */ | |
8151 | /* mark current iter state as drained and assume returned NULL */ | |
8152 | cur_iter->iter.state = BPF_ITER_STATE_DRAINED; | |
8153 | __mark_reg_const_zero(&cur_fr->regs[BPF_REG_0]); | |
8154 | ||
8155 | return 0; | |
8156 | } | |
8157 | ||
90133415 DB |
8158 | static bool arg_type_is_mem_size(enum bpf_arg_type type) |
8159 | { | |
8160 | return type == ARG_CONST_SIZE || | |
8161 | type == ARG_CONST_SIZE_OR_ZERO; | |
8162 | } | |
8163 | ||
8f14852e KKD |
8164 | static bool arg_type_is_release(enum bpf_arg_type type) |
8165 | { | |
8166 | return type & OBJ_RELEASE; | |
8167 | } | |
8168 | ||
97e03f52 JK |
8169 | static bool arg_type_is_dynptr(enum bpf_arg_type type) |
8170 | { | |
8171 | return base_type(type) == ARG_PTR_TO_DYNPTR; | |
8172 | } | |
8173 | ||
57c3bb72 AI |
8174 | static int int_ptr_type_to_size(enum bpf_arg_type type) |
8175 | { | |
8176 | if (type == ARG_PTR_TO_INT) | |
8177 | return sizeof(u32); | |
8178 | else if (type == ARG_PTR_TO_LONG) | |
8179 | return sizeof(u64); | |
8180 | ||
8181 | return -EINVAL; | |
8182 | } | |
8183 | ||
912f442c LB |
8184 | static int resolve_map_arg_type(struct bpf_verifier_env *env, |
8185 | const struct bpf_call_arg_meta *meta, | |
8186 | enum bpf_arg_type *arg_type) | |
8187 | { | |
8188 | if (!meta->map_ptr) { | |
8189 | /* kernel subsystem misconfigured verifier */ | |
8190 | verbose(env, "invalid map_ptr to access map->type\n"); | |
8191 | return -EACCES; | |
8192 | } | |
8193 | ||
8194 | switch (meta->map_ptr->map_type) { | |
8195 | case BPF_MAP_TYPE_SOCKMAP: | |
8196 | case BPF_MAP_TYPE_SOCKHASH: | |
8197 | if (*arg_type == ARG_PTR_TO_MAP_VALUE) { | |
6550f2dd | 8198 | *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON; |
912f442c LB |
8199 | } else { |
8200 | verbose(env, "invalid arg_type for sockmap/sockhash\n"); | |
8201 | return -EINVAL; | |
8202 | } | |
8203 | break; | |
9330986c JK |
8204 | case BPF_MAP_TYPE_BLOOM_FILTER: |
8205 | if (meta->func_id == BPF_FUNC_map_peek_elem) | |
8206 | *arg_type = ARG_PTR_TO_MAP_VALUE; | |
8207 | break; | |
912f442c LB |
8208 | default: |
8209 | break; | |
8210 | } | |
8211 | return 0; | |
8212 | } | |
8213 | ||
f79e7ea5 LB |
8214 | struct bpf_reg_types { |
8215 | const enum bpf_reg_type types[10]; | |
1df8f55a | 8216 | u32 *btf_id; |
f79e7ea5 LB |
8217 | }; |
8218 | ||
f79e7ea5 LB |
8219 | static const struct bpf_reg_types sock_types = { |
8220 | .types = { | |
8221 | PTR_TO_SOCK_COMMON, | |
8222 | PTR_TO_SOCKET, | |
8223 | PTR_TO_TCP_SOCK, | |
8224 | PTR_TO_XDP_SOCK, | |
8225 | }, | |
8226 | }; | |
8227 | ||
49a2a4d4 | 8228 | #ifdef CONFIG_NET |
1df8f55a MKL |
8229 | static const struct bpf_reg_types btf_id_sock_common_types = { |
8230 | .types = { | |
8231 | PTR_TO_SOCK_COMMON, | |
8232 | PTR_TO_SOCKET, | |
8233 | PTR_TO_TCP_SOCK, | |
8234 | PTR_TO_XDP_SOCK, | |
8235 | PTR_TO_BTF_ID, | |
3f00c523 | 8236 | PTR_TO_BTF_ID | PTR_TRUSTED, |
1df8f55a MKL |
8237 | }, |
8238 | .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
8239 | }; | |
49a2a4d4 | 8240 | #endif |
1df8f55a | 8241 | |
f79e7ea5 LB |
8242 | static const struct bpf_reg_types mem_types = { |
8243 | .types = { | |
8244 | PTR_TO_STACK, | |
8245 | PTR_TO_PACKET, | |
8246 | PTR_TO_PACKET_META, | |
69c087ba | 8247 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
8248 | PTR_TO_MAP_VALUE, |
8249 | PTR_TO_MEM, | |
894f2a8b | 8250 | PTR_TO_MEM | MEM_RINGBUF, |
20b2aff4 | 8251 | PTR_TO_BUF, |
3e30be42 | 8252 | PTR_TO_BTF_ID | PTR_TRUSTED, |
f79e7ea5 LB |
8253 | }, |
8254 | }; | |
8255 | ||
8256 | static const struct bpf_reg_types int_ptr_types = { | |
8257 | .types = { | |
8258 | PTR_TO_STACK, | |
8259 | PTR_TO_PACKET, | |
8260 | PTR_TO_PACKET_META, | |
69c087ba | 8261 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
8262 | PTR_TO_MAP_VALUE, |
8263 | }, | |
8264 | }; | |
8265 | ||
4e814da0 KKD |
8266 | static const struct bpf_reg_types spin_lock_types = { |
8267 | .types = { | |
8268 | PTR_TO_MAP_VALUE, | |
8269 | PTR_TO_BTF_ID | MEM_ALLOC, | |
8270 | } | |
8271 | }; | |
8272 | ||
f79e7ea5 LB |
8273 | static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } }; |
8274 | static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } }; | |
8275 | static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } }; | |
894f2a8b | 8276 | static const struct bpf_reg_types ringbuf_mem_types = { .types = { PTR_TO_MEM | MEM_RINGBUF } }; |
f79e7ea5 | 8277 | static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } }; |
3f00c523 DV |
8278 | static const struct bpf_reg_types btf_ptr_types = { |
8279 | .types = { | |
8280 | PTR_TO_BTF_ID, | |
8281 | PTR_TO_BTF_ID | PTR_TRUSTED, | |
fca1aa75 | 8282 | PTR_TO_BTF_ID | MEM_RCU, |
3f00c523 DV |
8283 | }, |
8284 | }; | |
8285 | static const struct bpf_reg_types percpu_btf_ptr_types = { | |
8286 | .types = { | |
8287 | PTR_TO_BTF_ID | MEM_PERCPU, | |
01cc55af | 8288 | PTR_TO_BTF_ID | MEM_PERCPU | MEM_RCU, |
3f00c523 DV |
8289 | PTR_TO_BTF_ID | MEM_PERCPU | PTR_TRUSTED, |
8290 | } | |
8291 | }; | |
69c087ba YS |
8292 | static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } }; |
8293 | static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } }; | |
fff13c4b | 8294 | static const struct bpf_reg_types const_str_ptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
b00628b1 | 8295 | static const struct bpf_reg_types timer_types = { .types = { PTR_TO_MAP_VALUE } }; |
c0a5a21c | 8296 | static const struct bpf_reg_types kptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
20571567 DV |
8297 | static const struct bpf_reg_types dynptr_types = { |
8298 | .types = { | |
8299 | PTR_TO_STACK, | |
27060531 | 8300 | CONST_PTR_TO_DYNPTR, |
20571567 DV |
8301 | } |
8302 | }; | |
f79e7ea5 | 8303 | |
0789e13b | 8304 | static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = { |
d1673304 DM |
8305 | [ARG_PTR_TO_MAP_KEY] = &mem_types, |
8306 | [ARG_PTR_TO_MAP_VALUE] = &mem_types, | |
f79e7ea5 LB |
8307 | [ARG_CONST_SIZE] = &scalar_types, |
8308 | [ARG_CONST_SIZE_OR_ZERO] = &scalar_types, | |
8309 | [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types, | |
8310 | [ARG_CONST_MAP_PTR] = &const_map_ptr_types, | |
8311 | [ARG_PTR_TO_CTX] = &context_types, | |
f79e7ea5 | 8312 | [ARG_PTR_TO_SOCK_COMMON] = &sock_types, |
49a2a4d4 | 8313 | #ifdef CONFIG_NET |
1df8f55a | 8314 | [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types, |
49a2a4d4 | 8315 | #endif |
f79e7ea5 | 8316 | [ARG_PTR_TO_SOCKET] = &fullsock_types, |
f79e7ea5 LB |
8317 | [ARG_PTR_TO_BTF_ID] = &btf_ptr_types, |
8318 | [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types, | |
8319 | [ARG_PTR_TO_MEM] = &mem_types, | |
894f2a8b | 8320 | [ARG_PTR_TO_RINGBUF_MEM] = &ringbuf_mem_types, |
f79e7ea5 LB |
8321 | [ARG_PTR_TO_INT] = &int_ptr_types, |
8322 | [ARG_PTR_TO_LONG] = &int_ptr_types, | |
eaa6bcb7 | 8323 | [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types, |
69c087ba | 8324 | [ARG_PTR_TO_FUNC] = &func_ptr_types, |
48946bd6 | 8325 | [ARG_PTR_TO_STACK] = &stack_ptr_types, |
fff13c4b | 8326 | [ARG_PTR_TO_CONST_STR] = &const_str_ptr_types, |
b00628b1 | 8327 | [ARG_PTR_TO_TIMER] = &timer_types, |
c0a5a21c | 8328 | [ARG_PTR_TO_KPTR] = &kptr_types, |
20571567 | 8329 | [ARG_PTR_TO_DYNPTR] = &dynptr_types, |
f79e7ea5 LB |
8330 | }; |
8331 | ||
8332 | static int check_reg_type(struct bpf_verifier_env *env, u32 regno, | |
a968d5e2 | 8333 | enum bpf_arg_type arg_type, |
c0a5a21c KKD |
8334 | const u32 *arg_btf_id, |
8335 | struct bpf_call_arg_meta *meta) | |
f79e7ea5 LB |
8336 | { |
8337 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
8338 | enum bpf_reg_type expected, type = reg->type; | |
a968d5e2 | 8339 | const struct bpf_reg_types *compatible; |
f79e7ea5 LB |
8340 | int i, j; |
8341 | ||
48946bd6 | 8342 | compatible = compatible_reg_types[base_type(arg_type)]; |
a968d5e2 MKL |
8343 | if (!compatible) { |
8344 | verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type); | |
8345 | return -EFAULT; | |
8346 | } | |
8347 | ||
216e3cd2 HL |
8348 | /* ARG_PTR_TO_MEM + RDONLY is compatible with PTR_TO_MEM and PTR_TO_MEM + RDONLY, |
8349 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM and NOT with PTR_TO_MEM + RDONLY | |
8350 | * | |
8351 | * Same for MAYBE_NULL: | |
8352 | * | |
8353 | * ARG_PTR_TO_MEM + MAYBE_NULL is compatible with PTR_TO_MEM and PTR_TO_MEM + MAYBE_NULL, | |
8354 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM but NOT with PTR_TO_MEM + MAYBE_NULL | |
8355 | * | |
2012c867 DR |
8356 | * ARG_PTR_TO_MEM is compatible with PTR_TO_MEM that is tagged with a dynptr type. |
8357 | * | |
216e3cd2 HL |
8358 | * Therefore we fold these flags depending on the arg_type before comparison. |
8359 | */ | |
8360 | if (arg_type & MEM_RDONLY) | |
8361 | type &= ~MEM_RDONLY; | |
8362 | if (arg_type & PTR_MAYBE_NULL) | |
8363 | type &= ~PTR_MAYBE_NULL; | |
2012c867 DR |
8364 | if (base_type(arg_type) == ARG_PTR_TO_MEM) |
8365 | type &= ~DYNPTR_TYPE_FLAG_MASK; | |
216e3cd2 | 8366 | |
36d8bdf7 | 8367 | if (meta->func_id == BPF_FUNC_kptr_xchg && type_is_alloc(type)) { |
738c96d5 | 8368 | type &= ~MEM_ALLOC; |
36d8bdf7 YS |
8369 | type &= ~MEM_PERCPU; |
8370 | } | |
738c96d5 | 8371 | |
f79e7ea5 LB |
8372 | for (i = 0; i < ARRAY_SIZE(compatible->types); i++) { |
8373 | expected = compatible->types[i]; | |
8374 | if (expected == NOT_INIT) | |
8375 | break; | |
8376 | ||
8377 | if (type == expected) | |
a968d5e2 | 8378 | goto found; |
f79e7ea5 LB |
8379 | } |
8380 | ||
216e3cd2 | 8381 | verbose(env, "R%d type=%s expected=", regno, reg_type_str(env, reg->type)); |
f79e7ea5 | 8382 | for (j = 0; j + 1 < i; j++) |
c25b2ae1 HL |
8383 | verbose(env, "%s, ", reg_type_str(env, compatible->types[j])); |
8384 | verbose(env, "%s\n", reg_type_str(env, compatible->types[j])); | |
f79e7ea5 | 8385 | return -EACCES; |
a968d5e2 MKL |
8386 | |
8387 | found: | |
da03e43a KKD |
8388 | if (base_type(reg->type) != PTR_TO_BTF_ID) |
8389 | return 0; | |
8390 | ||
3e30be42 AS |
8391 | if (compatible == &mem_types) { |
8392 | if (!(arg_type & MEM_RDONLY)) { | |
8393 | verbose(env, | |
8394 | "%s() may write into memory pointed by R%d type=%s\n", | |
8395 | func_id_name(meta->func_id), | |
8396 | regno, reg_type_str(env, reg->type)); | |
8397 | return -EACCES; | |
8398 | } | |
8399 | return 0; | |
8400 | } | |
8401 | ||
da03e43a KKD |
8402 | switch ((int)reg->type) { |
8403 | case PTR_TO_BTF_ID: | |
8404 | case PTR_TO_BTF_ID | PTR_TRUSTED: | |
8405 | case PTR_TO_BTF_ID | MEM_RCU: | |
add68b84 AS |
8406 | case PTR_TO_BTF_ID | PTR_MAYBE_NULL: |
8407 | case PTR_TO_BTF_ID | PTR_MAYBE_NULL | MEM_RCU: | |
da03e43a | 8408 | { |
2ab3b380 KKD |
8409 | /* For bpf_sk_release, it needs to match against first member |
8410 | * 'struct sock_common', hence make an exception for it. This | |
8411 | * allows bpf_sk_release to work for multiple socket types. | |
8412 | */ | |
8413 | bool strict_type_match = arg_type_is_release(arg_type) && | |
8414 | meta->func_id != BPF_FUNC_sk_release; | |
8415 | ||
add68b84 AS |
8416 | if (type_may_be_null(reg->type) && |
8417 | (!type_may_be_null(arg_type) || arg_type_is_release(arg_type))) { | |
8418 | verbose(env, "Possibly NULL pointer passed to helper arg%d\n", regno); | |
8419 | return -EACCES; | |
8420 | } | |
8421 | ||
1df8f55a MKL |
8422 | if (!arg_btf_id) { |
8423 | if (!compatible->btf_id) { | |
8424 | verbose(env, "verifier internal error: missing arg compatible BTF ID\n"); | |
8425 | return -EFAULT; | |
8426 | } | |
8427 | arg_btf_id = compatible->btf_id; | |
8428 | } | |
8429 | ||
c0a5a21c | 8430 | if (meta->func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac | 8431 | if (map_kptr_match_type(env, meta->kptr_field, reg, regno)) |
c0a5a21c | 8432 | return -EACCES; |
47e34cb7 DM |
8433 | } else { |
8434 | if (arg_btf_id == BPF_PTR_POISON) { | |
8435 | verbose(env, "verifier internal error:"); | |
8436 | verbose(env, "R%d has non-overwritten BPF_PTR_POISON type\n", | |
8437 | regno); | |
8438 | return -EACCES; | |
8439 | } | |
8440 | ||
8441 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
8442 | btf_vmlinux, *arg_btf_id, | |
8443 | strict_type_match)) { | |
8444 | verbose(env, "R%d is of type %s but %s is expected\n", | |
b32a5dae DM |
8445 | regno, btf_type_name(reg->btf, reg->btf_id), |
8446 | btf_type_name(btf_vmlinux, *arg_btf_id)); | |
47e34cb7 DM |
8447 | return -EACCES; |
8448 | } | |
a968d5e2 | 8449 | } |
da03e43a KKD |
8450 | break; |
8451 | } | |
8452 | case PTR_TO_BTF_ID | MEM_ALLOC: | |
36d8bdf7 | 8453 | case PTR_TO_BTF_ID | MEM_PERCPU | MEM_ALLOC: |
738c96d5 DM |
8454 | if (meta->func_id != BPF_FUNC_spin_lock && meta->func_id != BPF_FUNC_spin_unlock && |
8455 | meta->func_id != BPF_FUNC_kptr_xchg) { | |
4e814da0 KKD |
8456 | verbose(env, "verifier internal error: unimplemented handling of MEM_ALLOC\n"); |
8457 | return -EFAULT; | |
8458 | } | |
ab6c637a YS |
8459 | if (meta->func_id == BPF_FUNC_kptr_xchg) { |
8460 | if (map_kptr_match_type(env, meta->kptr_field, reg, regno)) | |
8461 | return -EACCES; | |
8462 | } | |
da03e43a KKD |
8463 | break; |
8464 | case PTR_TO_BTF_ID | MEM_PERCPU: | |
01cc55af | 8465 | case PTR_TO_BTF_ID | MEM_PERCPU | MEM_RCU: |
da03e43a KKD |
8466 | case PTR_TO_BTF_ID | MEM_PERCPU | PTR_TRUSTED: |
8467 | /* Handled by helper specific checks */ | |
8468 | break; | |
8469 | default: | |
8470 | verbose(env, "verifier internal error: invalid PTR_TO_BTF_ID register for type match\n"); | |
8471 | return -EFAULT; | |
a968d5e2 | 8472 | } |
a968d5e2 | 8473 | return 0; |
f79e7ea5 LB |
8474 | } |
8475 | ||
6a3cd331 DM |
8476 | static struct btf_field * |
8477 | reg_find_field_offset(const struct bpf_reg_state *reg, s32 off, u32 fields) | |
8478 | { | |
8479 | struct btf_field *field; | |
8480 | struct btf_record *rec; | |
8481 | ||
8482 | rec = reg_btf_record(reg); | |
8483 | if (!rec) | |
8484 | return NULL; | |
8485 | ||
8486 | field = btf_record_find(rec, off, fields); | |
8487 | if (!field) | |
8488 | return NULL; | |
8489 | ||
8490 | return field; | |
8491 | } | |
8492 | ||
25b35dd2 KKD |
8493 | int check_func_arg_reg_off(struct bpf_verifier_env *env, |
8494 | const struct bpf_reg_state *reg, int regno, | |
8f14852e | 8495 | enum bpf_arg_type arg_type) |
25b35dd2 | 8496 | { |
184c9bdb | 8497 | u32 type = reg->type; |
25b35dd2 | 8498 | |
184c9bdb KKD |
8499 | /* When referenced register is passed to release function, its fixed |
8500 | * offset must be 0. | |
8501 | * | |
8502 | * We will check arg_type_is_release reg has ref_obj_id when storing | |
8503 | * meta->release_regno. | |
8504 | */ | |
8505 | if (arg_type_is_release(arg_type)) { | |
8506 | /* ARG_PTR_TO_DYNPTR with OBJ_RELEASE is a bit special, as it | |
8507 | * may not directly point to the object being released, but to | |
8508 | * dynptr pointing to such object, which might be at some offset | |
8509 | * on the stack. In that case, we simply to fallback to the | |
8510 | * default handling. | |
8511 | */ | |
8512 | if (arg_type_is_dynptr(arg_type) && type == PTR_TO_STACK) | |
8513 | return 0; | |
6a3cd331 | 8514 | |
184c9bdb KKD |
8515 | /* Doing check_ptr_off_reg check for the offset will catch this |
8516 | * because fixed_off_ok is false, but checking here allows us | |
8517 | * to give the user a better error message. | |
8518 | */ | |
8519 | if (reg->off) { | |
8520 | verbose(env, "R%d must have zero offset when passed to release func or trusted arg to kfunc\n", | |
8521 | regno); | |
8522 | return -EINVAL; | |
8523 | } | |
8524 | return __check_ptr_off_reg(env, reg, regno, false); | |
8525 | } | |
8526 | ||
8527 | switch (type) { | |
8528 | /* Pointer types where both fixed and variable offset is explicitly allowed: */ | |
97e03f52 | 8529 | case PTR_TO_STACK: |
25b35dd2 KKD |
8530 | case PTR_TO_PACKET: |
8531 | case PTR_TO_PACKET_META: | |
8532 | case PTR_TO_MAP_KEY: | |
8533 | case PTR_TO_MAP_VALUE: | |
8534 | case PTR_TO_MEM: | |
8535 | case PTR_TO_MEM | MEM_RDONLY: | |
894f2a8b | 8536 | case PTR_TO_MEM | MEM_RINGBUF: |
25b35dd2 KKD |
8537 | case PTR_TO_BUF: |
8538 | case PTR_TO_BUF | MEM_RDONLY: | |
97e03f52 | 8539 | case SCALAR_VALUE: |
184c9bdb | 8540 | return 0; |
25b35dd2 KKD |
8541 | /* All the rest must be rejected, except PTR_TO_BTF_ID which allows |
8542 | * fixed offset. | |
8543 | */ | |
8544 | case PTR_TO_BTF_ID: | |
282de143 | 8545 | case PTR_TO_BTF_ID | MEM_ALLOC: |
3f00c523 | 8546 | case PTR_TO_BTF_ID | PTR_TRUSTED: |
fca1aa75 | 8547 | case PTR_TO_BTF_ID | MEM_RCU: |
6a3cd331 | 8548 | case PTR_TO_BTF_ID | MEM_ALLOC | NON_OWN_REF: |
0816b8c6 | 8549 | case PTR_TO_BTF_ID | MEM_ALLOC | NON_OWN_REF | MEM_RCU: |
24d5bb80 | 8550 | /* When referenced PTR_TO_BTF_ID is passed to release function, |
184c9bdb KKD |
8551 | * its fixed offset must be 0. In the other cases, fixed offset |
8552 | * can be non-zero. This was already checked above. So pass | |
8553 | * fixed_off_ok as true to allow fixed offset for all other | |
8554 | * cases. var_off always must be 0 for PTR_TO_BTF_ID, hence we | |
8555 | * still need to do checks instead of returning. | |
24d5bb80 | 8556 | */ |
184c9bdb | 8557 | return __check_ptr_off_reg(env, reg, regno, true); |
25b35dd2 | 8558 | default: |
184c9bdb | 8559 | return __check_ptr_off_reg(env, reg, regno, false); |
25b35dd2 | 8560 | } |
25b35dd2 KKD |
8561 | } |
8562 | ||
485ec51e JK |
8563 | static struct bpf_reg_state *get_dynptr_arg_reg(struct bpf_verifier_env *env, |
8564 | const struct bpf_func_proto *fn, | |
8565 | struct bpf_reg_state *regs) | |
8566 | { | |
8567 | struct bpf_reg_state *state = NULL; | |
8568 | int i; | |
8569 | ||
8570 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) | |
8571 | if (arg_type_is_dynptr(fn->arg_type[i])) { | |
8572 | if (state) { | |
8573 | verbose(env, "verifier internal error: multiple dynptr args\n"); | |
8574 | return NULL; | |
8575 | } | |
8576 | state = ®s[BPF_REG_1 + i]; | |
8577 | } | |
8578 | ||
8579 | if (!state) | |
8580 | verbose(env, "verifier internal error: no dynptr arg found\n"); | |
8581 | ||
8582 | return state; | |
8583 | } | |
8584 | ||
f8064ab9 | 8585 | static int dynptr_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
34d4ef57 JK |
8586 | { |
8587 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8588 | int spi; |
34d4ef57 | 8589 | |
27060531 | 8590 | if (reg->type == CONST_PTR_TO_DYNPTR) |
f8064ab9 KKD |
8591 | return reg->id; |
8592 | spi = dynptr_get_spi(env, reg); | |
8593 | if (spi < 0) | |
8594 | return spi; | |
8595 | return state->stack[spi].spilled_ptr.id; | |
8596 | } | |
8597 | ||
79168a66 | 8598 | static int dynptr_ref_obj_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
34d4ef57 JK |
8599 | { |
8600 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8601 | int spi; |
27060531 | 8602 | |
27060531 KKD |
8603 | if (reg->type == CONST_PTR_TO_DYNPTR) |
8604 | return reg->ref_obj_id; | |
79168a66 KKD |
8605 | spi = dynptr_get_spi(env, reg); |
8606 | if (spi < 0) | |
8607 | return spi; | |
27060531 | 8608 | return state->stack[spi].spilled_ptr.ref_obj_id; |
34d4ef57 JK |
8609 | } |
8610 | ||
b5964b96 JK |
8611 | static enum bpf_dynptr_type dynptr_get_type(struct bpf_verifier_env *env, |
8612 | struct bpf_reg_state *reg) | |
8613 | { | |
8614 | struct bpf_func_state *state = func(env, reg); | |
8615 | int spi; | |
8616 | ||
8617 | if (reg->type == CONST_PTR_TO_DYNPTR) | |
8618 | return reg->dynptr.type; | |
8619 | ||
8620 | spi = __get_spi(reg->off); | |
8621 | if (spi < 0) { | |
8622 | verbose(env, "verifier internal error: invalid spi when querying dynptr type\n"); | |
8623 | return BPF_DYNPTR_TYPE_INVALID; | |
8624 | } | |
8625 | ||
8626 | return state->stack[spi].spilled_ptr.dynptr.type; | |
8627 | } | |
8628 | ||
af7ec138 YS |
8629 | static int check_func_arg(struct bpf_verifier_env *env, u32 arg, |
8630 | struct bpf_call_arg_meta *meta, | |
1d18feb2 JK |
8631 | const struct bpf_func_proto *fn, |
8632 | int insn_idx) | |
17a52670 | 8633 | { |
af7ec138 | 8634 | u32 regno = BPF_REG_1 + arg; |
638f5b90 | 8635 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
af7ec138 | 8636 | enum bpf_arg_type arg_type = fn->arg_type[arg]; |
f79e7ea5 | 8637 | enum bpf_reg_type type = reg->type; |
508362ac | 8638 | u32 *arg_btf_id = NULL; |
17a52670 AS |
8639 | int err = 0; |
8640 | ||
80f1d68c | 8641 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
8642 | return 0; |
8643 | ||
dc503a8a EC |
8644 | err = check_reg_arg(env, regno, SRC_OP); |
8645 | if (err) | |
8646 | return err; | |
17a52670 | 8647 | |
1be7f75d AS |
8648 | if (arg_type == ARG_ANYTHING) { |
8649 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
8650 | verbose(env, "R%d leaks addr into helper function\n", |
8651 | regno); | |
1be7f75d AS |
8652 | return -EACCES; |
8653 | } | |
80f1d68c | 8654 | return 0; |
1be7f75d | 8655 | } |
80f1d68c | 8656 | |
de8f3a83 | 8657 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 8658 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 8659 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
8660 | return -EACCES; |
8661 | } | |
8662 | ||
16d1e00c | 8663 | if (base_type(arg_type) == ARG_PTR_TO_MAP_VALUE) { |
912f442c LB |
8664 | err = resolve_map_arg_type(env, meta, &arg_type); |
8665 | if (err) | |
8666 | return err; | |
8667 | } | |
8668 | ||
48946bd6 | 8669 | if (register_is_null(reg) && type_may_be_null(arg_type)) |
fd1b0d60 LB |
8670 | /* A NULL register has a SCALAR_VALUE type, so skip |
8671 | * type checking. | |
8672 | */ | |
8673 | goto skip_type_check; | |
8674 | ||
508362ac | 8675 | /* arg_btf_id and arg_size are in a union. */ |
4e814da0 KKD |
8676 | if (base_type(arg_type) == ARG_PTR_TO_BTF_ID || |
8677 | base_type(arg_type) == ARG_PTR_TO_SPIN_LOCK) | |
508362ac MM |
8678 | arg_btf_id = fn->arg_btf_id[arg]; |
8679 | ||
8680 | err = check_reg_type(env, regno, arg_type, arg_btf_id, meta); | |
f79e7ea5 LB |
8681 | if (err) |
8682 | return err; | |
8683 | ||
8f14852e | 8684 | err = check_func_arg_reg_off(env, reg, regno, arg_type); |
25b35dd2 KKD |
8685 | if (err) |
8686 | return err; | |
d7b9454a | 8687 | |
fd1b0d60 | 8688 | skip_type_check: |
8f14852e | 8689 | if (arg_type_is_release(arg_type)) { |
bc34dee6 JK |
8690 | if (arg_type_is_dynptr(arg_type)) { |
8691 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8692 | int spi; |
bc34dee6 | 8693 | |
27060531 KKD |
8694 | /* Only dynptr created on stack can be released, thus |
8695 | * the get_spi and stack state checks for spilled_ptr | |
8696 | * should only be done before process_dynptr_func for | |
8697 | * PTR_TO_STACK. | |
8698 | */ | |
8699 | if (reg->type == PTR_TO_STACK) { | |
79168a66 | 8700 | spi = dynptr_get_spi(env, reg); |
f5b625e5 | 8701 | if (spi < 0 || !state->stack[spi].spilled_ptr.ref_obj_id) { |
27060531 KKD |
8702 | verbose(env, "arg %d is an unacquired reference\n", regno); |
8703 | return -EINVAL; | |
8704 | } | |
8705 | } else { | |
8706 | verbose(env, "cannot release unowned const bpf_dynptr\n"); | |
bc34dee6 JK |
8707 | return -EINVAL; |
8708 | } | |
8709 | } else if (!reg->ref_obj_id && !register_is_null(reg)) { | |
8f14852e KKD |
8710 | verbose(env, "R%d must be referenced when passed to release function\n", |
8711 | regno); | |
8712 | return -EINVAL; | |
8713 | } | |
8714 | if (meta->release_regno) { | |
8715 | verbose(env, "verifier internal error: more than one release argument\n"); | |
8716 | return -EFAULT; | |
8717 | } | |
8718 | meta->release_regno = regno; | |
8719 | } | |
8720 | ||
02f7c958 | 8721 | if (reg->ref_obj_id) { |
457f4436 AN |
8722 | if (meta->ref_obj_id) { |
8723 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
8724 | regno, reg->ref_obj_id, | |
8725 | meta->ref_obj_id); | |
8726 | return -EFAULT; | |
8727 | } | |
8728 | meta->ref_obj_id = reg->ref_obj_id; | |
17a52670 AS |
8729 | } |
8730 | ||
8ab4cdcf JK |
8731 | switch (base_type(arg_type)) { |
8732 | case ARG_CONST_MAP_PTR: | |
17a52670 | 8733 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ |
3e8ce298 AS |
8734 | if (meta->map_ptr) { |
8735 | /* Use map_uid (which is unique id of inner map) to reject: | |
8736 | * inner_map1 = bpf_map_lookup_elem(outer_map, key1) | |
8737 | * inner_map2 = bpf_map_lookup_elem(outer_map, key2) | |
8738 | * if (inner_map1 && inner_map2) { | |
8739 | * timer = bpf_map_lookup_elem(inner_map1); | |
8740 | * if (timer) | |
8741 | * // mismatch would have been allowed | |
8742 | * bpf_timer_init(timer, inner_map2); | |
8743 | * } | |
8744 | * | |
8745 | * Comparing map_ptr is enough to distinguish normal and outer maps. | |
8746 | */ | |
8747 | if (meta->map_ptr != reg->map_ptr || | |
8748 | meta->map_uid != reg->map_uid) { | |
8749 | verbose(env, | |
8750 | "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n", | |
8751 | meta->map_uid, reg->map_uid); | |
8752 | return -EINVAL; | |
8753 | } | |
b00628b1 | 8754 | } |
33ff9823 | 8755 | meta->map_ptr = reg->map_ptr; |
3e8ce298 | 8756 | meta->map_uid = reg->map_uid; |
8ab4cdcf JK |
8757 | break; |
8758 | case ARG_PTR_TO_MAP_KEY: | |
17a52670 AS |
8759 | /* bpf_map_xxx(..., map_ptr, ..., key) call: |
8760 | * check that [key, key + map->key_size) are within | |
8761 | * stack limits and initialized | |
8762 | */ | |
33ff9823 | 8763 | if (!meta->map_ptr) { |
17a52670 AS |
8764 | /* in function declaration map_ptr must come before |
8765 | * map_key, so that it's verified and known before | |
8766 | * we have to check map_key here. Otherwise it means | |
8767 | * that kernel subsystem misconfigured verifier | |
8768 | */ | |
61bd5218 | 8769 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
8770 | return -EACCES; |
8771 | } | |
d71962f3 PC |
8772 | err = check_helper_mem_access(env, regno, |
8773 | meta->map_ptr->key_size, false, | |
8774 | NULL); | |
8ab4cdcf JK |
8775 | break; |
8776 | case ARG_PTR_TO_MAP_VALUE: | |
48946bd6 HL |
8777 | if (type_may_be_null(arg_type) && register_is_null(reg)) |
8778 | return 0; | |
8779 | ||
17a52670 AS |
8780 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
8781 | * check [value, value + map->value_size) validity | |
8782 | */ | |
33ff9823 | 8783 | if (!meta->map_ptr) { |
17a52670 | 8784 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 8785 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
8786 | return -EACCES; |
8787 | } | |
16d1e00c | 8788 | meta->raw_mode = arg_type & MEM_UNINIT; |
d71962f3 PC |
8789 | err = check_helper_mem_access(env, regno, |
8790 | meta->map_ptr->value_size, false, | |
2ea864c5 | 8791 | meta); |
8ab4cdcf JK |
8792 | break; |
8793 | case ARG_PTR_TO_PERCPU_BTF_ID: | |
eaa6bcb7 HL |
8794 | if (!reg->btf_id) { |
8795 | verbose(env, "Helper has invalid btf_id in R%d\n", regno); | |
8796 | return -EACCES; | |
8797 | } | |
22dc4a0f | 8798 | meta->ret_btf = reg->btf; |
eaa6bcb7 | 8799 | meta->ret_btf_id = reg->btf_id; |
8ab4cdcf JK |
8800 | break; |
8801 | case ARG_PTR_TO_SPIN_LOCK: | |
5d92ddc3 DM |
8802 | if (in_rbtree_lock_required_cb(env)) { |
8803 | verbose(env, "can't spin_{lock,unlock} in rbtree cb\n"); | |
8804 | return -EACCES; | |
8805 | } | |
c18f0b6a | 8806 | if (meta->func_id == BPF_FUNC_spin_lock) { |
ac50fe51 KKD |
8807 | err = process_spin_lock(env, regno, true); |
8808 | if (err) | |
8809 | return err; | |
c18f0b6a | 8810 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { |
ac50fe51 KKD |
8811 | err = process_spin_lock(env, regno, false); |
8812 | if (err) | |
8813 | return err; | |
c18f0b6a LB |
8814 | } else { |
8815 | verbose(env, "verifier internal error\n"); | |
8816 | return -EFAULT; | |
8817 | } | |
8ab4cdcf JK |
8818 | break; |
8819 | case ARG_PTR_TO_TIMER: | |
ac50fe51 KKD |
8820 | err = process_timer_func(env, regno, meta); |
8821 | if (err) | |
8822 | return err; | |
8ab4cdcf JK |
8823 | break; |
8824 | case ARG_PTR_TO_FUNC: | |
69c087ba | 8825 | meta->subprogno = reg->subprogno; |
8ab4cdcf JK |
8826 | break; |
8827 | case ARG_PTR_TO_MEM: | |
a2bbe7cc LB |
8828 | /* The access to this pointer is only checked when we hit the |
8829 | * next is_mem_size argument below. | |
8830 | */ | |
16d1e00c | 8831 | meta->raw_mode = arg_type & MEM_UNINIT; |
508362ac MM |
8832 | if (arg_type & MEM_FIXED_SIZE) { |
8833 | err = check_helper_mem_access(env, regno, | |
8834 | fn->arg_size[arg], false, | |
8835 | meta); | |
8836 | } | |
8ab4cdcf JK |
8837 | break; |
8838 | case ARG_CONST_SIZE: | |
8839 | err = check_mem_size_reg(env, reg, regno, false, meta); | |
8840 | break; | |
8841 | case ARG_CONST_SIZE_OR_ZERO: | |
8842 | err = check_mem_size_reg(env, reg, regno, true, meta); | |
8843 | break; | |
8844 | case ARG_PTR_TO_DYNPTR: | |
361f129f | 8845 | err = process_dynptr_func(env, regno, insn_idx, arg_type, 0); |
ac50fe51 KKD |
8846 | if (err) |
8847 | return err; | |
8ab4cdcf JK |
8848 | break; |
8849 | case ARG_CONST_ALLOC_SIZE_OR_ZERO: | |
457f4436 | 8850 | if (!tnum_is_const(reg->var_off)) { |
28a8add6 | 8851 | verbose(env, "R%d is not a known constant'\n", |
457f4436 AN |
8852 | regno); |
8853 | return -EACCES; | |
8854 | } | |
8855 | meta->mem_size = reg->var_off.value; | |
2fc31465 KKD |
8856 | err = mark_chain_precision(env, regno); |
8857 | if (err) | |
8858 | return err; | |
8ab4cdcf JK |
8859 | break; |
8860 | case ARG_PTR_TO_INT: | |
8861 | case ARG_PTR_TO_LONG: | |
8862 | { | |
57c3bb72 AI |
8863 | int size = int_ptr_type_to_size(arg_type); |
8864 | ||
8865 | err = check_helper_mem_access(env, regno, size, false, meta); | |
8866 | if (err) | |
8867 | return err; | |
8868 | err = check_ptr_alignment(env, reg, 0, size, true); | |
8ab4cdcf JK |
8869 | break; |
8870 | } | |
8871 | case ARG_PTR_TO_CONST_STR: | |
8872 | { | |
fff13c4b FR |
8873 | struct bpf_map *map = reg->map_ptr; |
8874 | int map_off; | |
8875 | u64 map_addr; | |
8876 | char *str_ptr; | |
8877 | ||
a8fad73e | 8878 | if (!bpf_map_is_rdonly(map)) { |
fff13c4b FR |
8879 | verbose(env, "R%d does not point to a readonly map'\n", regno); |
8880 | return -EACCES; | |
8881 | } | |
8882 | ||
8883 | if (!tnum_is_const(reg->var_off)) { | |
8884 | verbose(env, "R%d is not a constant address'\n", regno); | |
8885 | return -EACCES; | |
8886 | } | |
8887 | ||
8888 | if (!map->ops->map_direct_value_addr) { | |
8889 | verbose(env, "no direct value access support for this map type\n"); | |
8890 | return -EACCES; | |
8891 | } | |
8892 | ||
8893 | err = check_map_access(env, regno, reg->off, | |
61df10c7 KKD |
8894 | map->value_size - reg->off, false, |
8895 | ACCESS_HELPER); | |
fff13c4b FR |
8896 | if (err) |
8897 | return err; | |
8898 | ||
8899 | map_off = reg->off + reg->var_off.value; | |
8900 | err = map->ops->map_direct_value_addr(map, &map_addr, map_off); | |
8901 | if (err) { | |
8902 | verbose(env, "direct value access on string failed\n"); | |
8903 | return err; | |
8904 | } | |
8905 | ||
8906 | str_ptr = (char *)(long)(map_addr); | |
8907 | if (!strnchr(str_ptr + map_off, map->value_size - map_off, 0)) { | |
8908 | verbose(env, "string is not zero-terminated\n"); | |
8909 | return -EINVAL; | |
8910 | } | |
8ab4cdcf JK |
8911 | break; |
8912 | } | |
8913 | case ARG_PTR_TO_KPTR: | |
ac50fe51 KKD |
8914 | err = process_kptr_func(env, regno, meta); |
8915 | if (err) | |
8916 | return err; | |
8ab4cdcf | 8917 | break; |
17a52670 AS |
8918 | } |
8919 | ||
8920 | return err; | |
8921 | } | |
8922 | ||
0126240f LB |
8923 | static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) |
8924 | { | |
8925 | enum bpf_attach_type eatype = env->prog->expected_attach_type; | |
7e40781c | 8926 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
0126240f LB |
8927 | |
8928 | if (func_id != BPF_FUNC_map_update_elem) | |
8929 | return false; | |
8930 | ||
8931 | /* It's not possible to get access to a locked struct sock in these | |
8932 | * contexts, so updating is safe. | |
8933 | */ | |
8934 | switch (type) { | |
8935 | case BPF_PROG_TYPE_TRACING: | |
8936 | if (eatype == BPF_TRACE_ITER) | |
8937 | return true; | |
8938 | break; | |
8939 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
8940 | case BPF_PROG_TYPE_SCHED_CLS: | |
8941 | case BPF_PROG_TYPE_SCHED_ACT: | |
8942 | case BPF_PROG_TYPE_XDP: | |
8943 | case BPF_PROG_TYPE_SK_REUSEPORT: | |
8944 | case BPF_PROG_TYPE_FLOW_DISSECTOR: | |
8945 | case BPF_PROG_TYPE_SK_LOOKUP: | |
8946 | return true; | |
8947 | default: | |
8948 | break; | |
8949 | } | |
8950 | ||
8951 | verbose(env, "cannot update sockmap in this context\n"); | |
8952 | return false; | |
8953 | } | |
8954 | ||
e411901c MF |
8955 | static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env) |
8956 | { | |
95acd881 TA |
8957 | return env->prog->jit_requested && |
8958 | bpf_jit_supports_subprog_tailcalls(); | |
e411901c MF |
8959 | } |
8960 | ||
61bd5218 JK |
8961 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
8962 | struct bpf_map *map, int func_id) | |
35578d79 | 8963 | { |
35578d79 KX |
8964 | if (!map) |
8965 | return 0; | |
8966 | ||
6aff67c8 AS |
8967 | /* We need a two way check, first is from map perspective ... */ |
8968 | switch (map->map_type) { | |
8969 | case BPF_MAP_TYPE_PROG_ARRAY: | |
8970 | if (func_id != BPF_FUNC_tail_call) | |
8971 | goto error; | |
8972 | break; | |
8973 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
8974 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 8975 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 8976 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
8977 | func_id != BPF_FUNC_perf_event_read_value && |
8978 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
8979 | goto error; |
8980 | break; | |
457f4436 AN |
8981 | case BPF_MAP_TYPE_RINGBUF: |
8982 | if (func_id != BPF_FUNC_ringbuf_output && | |
8983 | func_id != BPF_FUNC_ringbuf_reserve && | |
bc34dee6 JK |
8984 | func_id != BPF_FUNC_ringbuf_query && |
8985 | func_id != BPF_FUNC_ringbuf_reserve_dynptr && | |
8986 | func_id != BPF_FUNC_ringbuf_submit_dynptr && | |
8987 | func_id != BPF_FUNC_ringbuf_discard_dynptr) | |
457f4436 AN |
8988 | goto error; |
8989 | break; | |
583c1f42 | 8990 | case BPF_MAP_TYPE_USER_RINGBUF: |
20571567 DV |
8991 | if (func_id != BPF_FUNC_user_ringbuf_drain) |
8992 | goto error; | |
8993 | break; | |
6aff67c8 AS |
8994 | case BPF_MAP_TYPE_STACK_TRACE: |
8995 | if (func_id != BPF_FUNC_get_stackid) | |
8996 | goto error; | |
8997 | break; | |
4ed8ec52 | 8998 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 8999 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 9000 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
9001 | goto error; |
9002 | break; | |
cd339431 | 9003 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 9004 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
9005 | if (func_id != BPF_FUNC_get_local_storage) |
9006 | goto error; | |
9007 | break; | |
546ac1ff | 9008 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 9009 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
9010 | if (func_id != BPF_FUNC_redirect_map && |
9011 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
9012 | goto error; |
9013 | break; | |
fbfc504a BT |
9014 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
9015 | * appear. | |
9016 | */ | |
6710e112 JDB |
9017 | case BPF_MAP_TYPE_CPUMAP: |
9018 | if (func_id != BPF_FUNC_redirect_map) | |
9019 | goto error; | |
9020 | break; | |
fada7fdc JL |
9021 | case BPF_MAP_TYPE_XSKMAP: |
9022 | if (func_id != BPF_FUNC_redirect_map && | |
9023 | func_id != BPF_FUNC_map_lookup_elem) | |
9024 | goto error; | |
9025 | break; | |
56f668df | 9026 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 9027 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
9028 | if (func_id != BPF_FUNC_map_lookup_elem) |
9029 | goto error; | |
16a43625 | 9030 | break; |
174a79ff JF |
9031 | case BPF_MAP_TYPE_SOCKMAP: |
9032 | if (func_id != BPF_FUNC_sk_redirect_map && | |
9033 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 9034 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 9035 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 | 9036 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
9037 | func_id != BPF_FUNC_map_lookup_elem && |
9038 | !may_update_sockmap(env, func_id)) | |
174a79ff JF |
9039 | goto error; |
9040 | break; | |
81110384 JF |
9041 | case BPF_MAP_TYPE_SOCKHASH: |
9042 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
9043 | func_id != BPF_FUNC_sock_hash_update && | |
9044 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 9045 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 | 9046 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
9047 | func_id != BPF_FUNC_map_lookup_elem && |
9048 | !may_update_sockmap(env, func_id)) | |
81110384 JF |
9049 | goto error; |
9050 | break; | |
2dbb9b9e MKL |
9051 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
9052 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
9053 | goto error; | |
9054 | break; | |
f1a2e44a MV |
9055 | case BPF_MAP_TYPE_QUEUE: |
9056 | case BPF_MAP_TYPE_STACK: | |
9057 | if (func_id != BPF_FUNC_map_peek_elem && | |
9058 | func_id != BPF_FUNC_map_pop_elem && | |
9059 | func_id != BPF_FUNC_map_push_elem) | |
9060 | goto error; | |
9061 | break; | |
6ac99e8f MKL |
9062 | case BPF_MAP_TYPE_SK_STORAGE: |
9063 | if (func_id != BPF_FUNC_sk_storage_get && | |
9db44fdd KKD |
9064 | func_id != BPF_FUNC_sk_storage_delete && |
9065 | func_id != BPF_FUNC_kptr_xchg) | |
6ac99e8f MKL |
9066 | goto error; |
9067 | break; | |
8ea63684 KS |
9068 | case BPF_MAP_TYPE_INODE_STORAGE: |
9069 | if (func_id != BPF_FUNC_inode_storage_get && | |
9db44fdd KKD |
9070 | func_id != BPF_FUNC_inode_storage_delete && |
9071 | func_id != BPF_FUNC_kptr_xchg) | |
8ea63684 KS |
9072 | goto error; |
9073 | break; | |
4cf1bc1f KS |
9074 | case BPF_MAP_TYPE_TASK_STORAGE: |
9075 | if (func_id != BPF_FUNC_task_storage_get && | |
9db44fdd KKD |
9076 | func_id != BPF_FUNC_task_storage_delete && |
9077 | func_id != BPF_FUNC_kptr_xchg) | |
4cf1bc1f KS |
9078 | goto error; |
9079 | break; | |
c4bcfb38 YS |
9080 | case BPF_MAP_TYPE_CGRP_STORAGE: |
9081 | if (func_id != BPF_FUNC_cgrp_storage_get && | |
9db44fdd KKD |
9082 | func_id != BPF_FUNC_cgrp_storage_delete && |
9083 | func_id != BPF_FUNC_kptr_xchg) | |
c4bcfb38 YS |
9084 | goto error; |
9085 | break; | |
9330986c JK |
9086 | case BPF_MAP_TYPE_BLOOM_FILTER: |
9087 | if (func_id != BPF_FUNC_map_peek_elem && | |
9088 | func_id != BPF_FUNC_map_push_elem) | |
9089 | goto error; | |
9090 | break; | |
6aff67c8 AS |
9091 | default: |
9092 | break; | |
9093 | } | |
9094 | ||
9095 | /* ... and second from the function itself. */ | |
9096 | switch (func_id) { | |
9097 | case BPF_FUNC_tail_call: | |
9098 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
9099 | goto error; | |
e411901c MF |
9100 | if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) { |
9101 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
f4d7e40a AS |
9102 | return -EINVAL; |
9103 | } | |
6aff67c8 AS |
9104 | break; |
9105 | case BPF_FUNC_perf_event_read: | |
9106 | case BPF_FUNC_perf_event_output: | |
908432ca | 9107 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 9108 | case BPF_FUNC_skb_output: |
d831ee84 | 9109 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
9110 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
9111 | goto error; | |
9112 | break; | |
5b029a32 DB |
9113 | case BPF_FUNC_ringbuf_output: |
9114 | case BPF_FUNC_ringbuf_reserve: | |
9115 | case BPF_FUNC_ringbuf_query: | |
bc34dee6 JK |
9116 | case BPF_FUNC_ringbuf_reserve_dynptr: |
9117 | case BPF_FUNC_ringbuf_submit_dynptr: | |
9118 | case BPF_FUNC_ringbuf_discard_dynptr: | |
5b029a32 DB |
9119 | if (map->map_type != BPF_MAP_TYPE_RINGBUF) |
9120 | goto error; | |
9121 | break; | |
20571567 DV |
9122 | case BPF_FUNC_user_ringbuf_drain: |
9123 | if (map->map_type != BPF_MAP_TYPE_USER_RINGBUF) | |
9124 | goto error; | |
9125 | break; | |
6aff67c8 AS |
9126 | case BPF_FUNC_get_stackid: |
9127 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
9128 | goto error; | |
9129 | break; | |
60d20f91 | 9130 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 9131 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
9132 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
9133 | goto error; | |
9134 | break; | |
97f91a7c | 9135 | case BPF_FUNC_redirect_map: |
9c270af3 | 9136 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 9137 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
9138 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
9139 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
9140 | goto error; |
9141 | break; | |
174a79ff | 9142 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 9143 | case BPF_FUNC_msg_redirect_map: |
81110384 | 9144 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
9145 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
9146 | goto error; | |
9147 | break; | |
81110384 JF |
9148 | case BPF_FUNC_sk_redirect_hash: |
9149 | case BPF_FUNC_msg_redirect_hash: | |
9150 | case BPF_FUNC_sock_hash_update: | |
9151 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
9152 | goto error; |
9153 | break; | |
cd339431 | 9154 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
9155 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
9156 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
9157 | goto error; |
9158 | break; | |
2dbb9b9e | 9159 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
9160 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
9161 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
9162 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
9163 | goto error; |
9164 | break; | |
f1a2e44a | 9165 | case BPF_FUNC_map_pop_elem: |
f1a2e44a MV |
9166 | if (map->map_type != BPF_MAP_TYPE_QUEUE && |
9167 | map->map_type != BPF_MAP_TYPE_STACK) | |
9168 | goto error; | |
9169 | break; | |
9330986c JK |
9170 | case BPF_FUNC_map_peek_elem: |
9171 | case BPF_FUNC_map_push_elem: | |
9172 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
9173 | map->map_type != BPF_MAP_TYPE_STACK && | |
9174 | map->map_type != BPF_MAP_TYPE_BLOOM_FILTER) | |
9175 | goto error; | |
9176 | break; | |
07343110 FZ |
9177 | case BPF_FUNC_map_lookup_percpu_elem: |
9178 | if (map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY && | |
9179 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && | |
9180 | map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH) | |
9181 | goto error; | |
9182 | break; | |
6ac99e8f MKL |
9183 | case BPF_FUNC_sk_storage_get: |
9184 | case BPF_FUNC_sk_storage_delete: | |
9185 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
9186 | goto error; | |
9187 | break; | |
8ea63684 KS |
9188 | case BPF_FUNC_inode_storage_get: |
9189 | case BPF_FUNC_inode_storage_delete: | |
9190 | if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE) | |
9191 | goto error; | |
9192 | break; | |
4cf1bc1f KS |
9193 | case BPF_FUNC_task_storage_get: |
9194 | case BPF_FUNC_task_storage_delete: | |
9195 | if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) | |
9196 | goto error; | |
9197 | break; | |
c4bcfb38 YS |
9198 | case BPF_FUNC_cgrp_storage_get: |
9199 | case BPF_FUNC_cgrp_storage_delete: | |
9200 | if (map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) | |
9201 | goto error; | |
9202 | break; | |
6aff67c8 AS |
9203 | default: |
9204 | break; | |
35578d79 KX |
9205 | } |
9206 | ||
9207 | return 0; | |
6aff67c8 | 9208 | error: |
61bd5218 | 9209 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 9210 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 9211 | return -EINVAL; |
35578d79 KX |
9212 | } |
9213 | ||
90133415 | 9214 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
9215 | { |
9216 | int count = 0; | |
9217 | ||
39f19ebb | 9218 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 9219 | count++; |
39f19ebb | 9220 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 9221 | count++; |
39f19ebb | 9222 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 9223 | count++; |
39f19ebb | 9224 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 9225 | count++; |
39f19ebb | 9226 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
9227 | count++; |
9228 | ||
90133415 DB |
9229 | /* We only support one arg being in raw mode at the moment, |
9230 | * which is sufficient for the helper functions we have | |
9231 | * right now. | |
9232 | */ | |
9233 | return count <= 1; | |
9234 | } | |
9235 | ||
508362ac | 9236 | static bool check_args_pair_invalid(const struct bpf_func_proto *fn, int arg) |
90133415 | 9237 | { |
508362ac MM |
9238 | bool is_fixed = fn->arg_type[arg] & MEM_FIXED_SIZE; |
9239 | bool has_size = fn->arg_size[arg] != 0; | |
9240 | bool is_next_size = false; | |
9241 | ||
9242 | if (arg + 1 < ARRAY_SIZE(fn->arg_type)) | |
9243 | is_next_size = arg_type_is_mem_size(fn->arg_type[arg + 1]); | |
9244 | ||
9245 | if (base_type(fn->arg_type[arg]) != ARG_PTR_TO_MEM) | |
9246 | return is_next_size; | |
9247 | ||
9248 | return has_size == is_next_size || is_next_size == is_fixed; | |
90133415 DB |
9249 | } |
9250 | ||
9251 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
9252 | { | |
9253 | /* bpf_xxx(..., buf, len) call will access 'len' | |
9254 | * bytes from memory 'buf'. Both arg types need | |
9255 | * to be paired, so make sure there's no buggy | |
9256 | * helper function specification. | |
9257 | */ | |
9258 | if (arg_type_is_mem_size(fn->arg1_type) || | |
508362ac MM |
9259 | check_args_pair_invalid(fn, 0) || |
9260 | check_args_pair_invalid(fn, 1) || | |
9261 | check_args_pair_invalid(fn, 2) || | |
9262 | check_args_pair_invalid(fn, 3) || | |
9263 | check_args_pair_invalid(fn, 4)) | |
90133415 DB |
9264 | return false; |
9265 | ||
9266 | return true; | |
9267 | } | |
9268 | ||
9436ef6e LB |
9269 | static bool check_btf_id_ok(const struct bpf_func_proto *fn) |
9270 | { | |
9271 | int i; | |
9272 | ||
1df8f55a | 9273 | for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) { |
4e814da0 KKD |
9274 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_BTF_ID) |
9275 | return !!fn->arg_btf_id[i]; | |
9276 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_SPIN_LOCK) | |
9277 | return fn->arg_btf_id[i] == BPF_PTR_POISON; | |
508362ac MM |
9278 | if (base_type(fn->arg_type[i]) != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i] && |
9279 | /* arg_btf_id and arg_size are in a union. */ | |
9280 | (base_type(fn->arg_type[i]) != ARG_PTR_TO_MEM || | |
9281 | !(fn->arg_type[i] & MEM_FIXED_SIZE))) | |
1df8f55a MKL |
9282 | return false; |
9283 | } | |
9284 | ||
9436ef6e LB |
9285 | return true; |
9286 | } | |
9287 | ||
0c9a7a7e | 9288 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
9289 | { |
9290 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 9291 | check_arg_pair_ok(fn) && |
b2d8ef19 | 9292 | check_btf_id_ok(fn) ? 0 : -EINVAL; |
435faee1 DB |
9293 | } |
9294 | ||
de8f3a83 DB |
9295 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
9296 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
66e3a13e JK |
9297 | * |
9298 | * This also applies to dynptr slices belonging to skb and xdp dynptrs, | |
9299 | * since these slices point to packet data. | |
f1174f77 | 9300 | */ |
b239da34 | 9301 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 9302 | { |
b239da34 KKD |
9303 | struct bpf_func_state *state; |
9304 | struct bpf_reg_state *reg; | |
969bf05e | 9305 | |
b239da34 | 9306 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
66e3a13e | 9307 | if (reg_is_pkt_pointer_any(reg) || reg_is_dynptr_slice_pkt(reg)) |
dbd8d228 | 9308 | mark_reg_invalid(env, reg); |
b239da34 | 9309 | })); |
f4d7e40a AS |
9310 | } |
9311 | ||
6d94e741 AS |
9312 | enum { |
9313 | AT_PKT_END = -1, | |
9314 | BEYOND_PKT_END = -2, | |
9315 | }; | |
9316 | ||
9317 | static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open) | |
9318 | { | |
9319 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
9320 | struct bpf_reg_state *reg = &state->regs[regn]; | |
9321 | ||
9322 | if (reg->type != PTR_TO_PACKET) | |
9323 | /* PTR_TO_PACKET_META is not supported yet */ | |
9324 | return; | |
9325 | ||
9326 | /* The 'reg' is pkt > pkt_end or pkt >= pkt_end. | |
9327 | * How far beyond pkt_end it goes is unknown. | |
9328 | * if (!range_open) it's the case of pkt >= pkt_end | |
9329 | * if (range_open) it's the case of pkt > pkt_end | |
9330 | * hence this pointer is at least 1 byte bigger than pkt_end | |
9331 | */ | |
9332 | if (range_open) | |
9333 | reg->range = BEYOND_PKT_END; | |
9334 | else | |
9335 | reg->range = AT_PKT_END; | |
9336 | } | |
9337 | ||
fd978bf7 JS |
9338 | /* The pointer with the specified id has released its reference to kernel |
9339 | * resources. Identify all copies of the same pointer and clear the reference. | |
9340 | */ | |
9341 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 9342 | int ref_obj_id) |
fd978bf7 | 9343 | { |
b239da34 KKD |
9344 | struct bpf_func_state *state; |
9345 | struct bpf_reg_state *reg; | |
1b986589 | 9346 | int err; |
fd978bf7 | 9347 | |
1b986589 MKL |
9348 | err = release_reference_state(cur_func(env), ref_obj_id); |
9349 | if (err) | |
9350 | return err; | |
9351 | ||
b239da34 | 9352 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
dbd8d228 KKD |
9353 | if (reg->ref_obj_id == ref_obj_id) |
9354 | mark_reg_invalid(env, reg); | |
b239da34 | 9355 | })); |
fd978bf7 | 9356 | |
1b986589 | 9357 | return 0; |
fd978bf7 JS |
9358 | } |
9359 | ||
6a3cd331 DM |
9360 | static void invalidate_non_owning_refs(struct bpf_verifier_env *env) |
9361 | { | |
9362 | struct bpf_func_state *unused; | |
9363 | struct bpf_reg_state *reg; | |
9364 | ||
9365 | bpf_for_each_reg_in_vstate(env->cur_state, unused, reg, ({ | |
9366 | if (type_is_non_owning_ref(reg->type)) | |
dbd8d228 | 9367 | mark_reg_invalid(env, reg); |
6a3cd331 DM |
9368 | })); |
9369 | } | |
9370 | ||
51c39bb1 AS |
9371 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
9372 | struct bpf_reg_state *regs) | |
9373 | { | |
9374 | int i; | |
9375 | ||
9376 | /* after the call registers r0 - r5 were scratched */ | |
9377 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
9378 | mark_reg_not_init(env, regs, caller_saved[i]); | |
683b96f9 | 9379 | __check_reg_arg(env, regs, caller_saved[i], DST_OP_NO_MARK); |
51c39bb1 AS |
9380 | } |
9381 | } | |
9382 | ||
14351375 YS |
9383 | typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env, |
9384 | struct bpf_func_state *caller, | |
9385 | struct bpf_func_state *callee, | |
9386 | int insn_idx); | |
9387 | ||
be2ef816 AN |
9388 | static int set_callee_state(struct bpf_verifier_env *env, |
9389 | struct bpf_func_state *caller, | |
9390 | struct bpf_func_state *callee, int insn_idx); | |
9391 | ||
58124a98 EZ |
9392 | static int setup_func_entry(struct bpf_verifier_env *env, int subprog, int callsite, |
9393 | set_callee_state_fn set_callee_state_cb, | |
9394 | struct bpf_verifier_state *state) | |
f4d7e40a | 9395 | { |
f4d7e40a | 9396 | struct bpf_func_state *caller, *callee; |
14351375 | 9397 | int err; |
f4d7e40a | 9398 | |
aada9ce6 | 9399 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 9400 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 9401 | state->curframe + 2); |
f4d7e40a AS |
9402 | return -E2BIG; |
9403 | } | |
9404 | ||
f4d7e40a AS |
9405 | if (state->frame[state->curframe + 1]) { |
9406 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
9407 | state->curframe + 1); | |
9408 | return -EFAULT; | |
9409 | } | |
9410 | ||
58124a98 EZ |
9411 | caller = state->frame[state->curframe]; |
9412 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); | |
9413 | if (!callee) | |
9414 | return -ENOMEM; | |
9415 | state->frame[state->curframe + 1] = callee; | |
9416 | ||
9417 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
9418 | * into its own stack before reading from it. | |
9419 | * callee can read/write into caller's stack | |
9420 | */ | |
9421 | init_func_state(env, callee, | |
9422 | /* remember the callsite, it will be used by bpf_exit */ | |
9423 | callsite, | |
9424 | state->curframe + 1 /* frameno within this callchain */, | |
9425 | subprog /* subprog number within this prog */); | |
9426 | /* Transfer references to the callee */ | |
9427 | err = copy_reference_state(callee, caller); | |
9428 | err = err ?: set_callee_state_cb(env, caller, callee, callsite); | |
9429 | if (err) | |
9430 | goto err_out; | |
9431 | ||
9432 | /* only increment it after check_reg_arg() finished */ | |
9433 | state->curframe++; | |
9434 | ||
9435 | return 0; | |
9436 | ||
9437 | err_out: | |
9438 | free_func_state(callee); | |
9439 | state->frame[state->curframe + 1] = NULL; | |
9440 | return err; | |
9441 | } | |
9442 | ||
ab5cfac1 EZ |
9443 | static int push_callback_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
9444 | int insn_idx, int subprog, | |
9445 | set_callee_state_fn set_callee_state_cb) | |
58124a98 | 9446 | { |
ab5cfac1 | 9447 | struct bpf_verifier_state *state = env->cur_state, *callback_state; |
58124a98 EZ |
9448 | struct bpf_func_state *caller, *callee; |
9449 | int err; | |
9450 | ||
9451 | caller = state->frame[state->curframe]; | |
95f2f26f | 9452 | err = btf_check_subprog_call(env, subprog, caller->regs); |
51c39bb1 AS |
9453 | if (err == -EFAULT) |
9454 | return err; | |
51c39bb1 | 9455 | |
be2ef816 AN |
9456 | /* set_callee_state is used for direct subprog calls, but we are |
9457 | * interested in validating only BPF helpers that can call subprogs as | |
9458 | * callbacks | |
9459 | */ | |
ab5cfac1 EZ |
9460 | env->subprog_info[subprog].is_cb = true; |
9461 | if (bpf_pseudo_kfunc_call(insn) && | |
9462 | !is_sync_callback_calling_kfunc(insn->imm)) { | |
9463 | verbose(env, "verifier bug: kfunc %s#%d not marked as callback-calling\n", | |
9464 | func_id_name(insn->imm), insn->imm); | |
9465 | return -EFAULT; | |
9466 | } else if (!bpf_pseudo_kfunc_call(insn) && | |
9467 | !is_callback_calling_function(insn->imm)) { /* helper */ | |
9468 | verbose(env, "verifier bug: helper %s#%d not marked as callback-calling\n", | |
9469 | func_id_name(insn->imm), insn->imm); | |
9470 | return -EFAULT; | |
be2ef816 AN |
9471 | } |
9472 | ||
bfc6bb74 | 9473 | if (insn->code == (BPF_JMP | BPF_CALL) && |
a5bebc4f | 9474 | insn->src_reg == 0 && |
bfc6bb74 AS |
9475 | insn->imm == BPF_FUNC_timer_set_callback) { |
9476 | struct bpf_verifier_state *async_cb; | |
9477 | ||
9478 | /* there is no real recursion here. timer callbacks are async */ | |
7ddc80a4 | 9479 | env->subprog_info[subprog].is_async_cb = true; |
bfc6bb74 | 9480 | async_cb = push_async_cb(env, env->subprog_info[subprog].start, |
ab5cfac1 | 9481 | insn_idx, subprog); |
bfc6bb74 AS |
9482 | if (!async_cb) |
9483 | return -EFAULT; | |
9484 | callee = async_cb->frame[0]; | |
9485 | callee->async_entry_cnt = caller->async_entry_cnt + 1; | |
9486 | ||
9487 | /* Convert bpf_timer_set_callback() args into timer callback args */ | |
ab5cfac1 | 9488 | err = set_callee_state_cb(env, caller, callee, insn_idx); |
bfc6bb74 AS |
9489 | if (err) |
9490 | return err; | |
9491 | ||
ab5cfac1 EZ |
9492 | return 0; |
9493 | } | |
9494 | ||
9495 | /* for callback functions enqueue entry to callback and | |
9496 | * proceed with next instruction within current frame. | |
9497 | */ | |
9498 | callback_state = push_stack(env, env->subprog_info[subprog].start, insn_idx, false); | |
9499 | if (!callback_state) | |
9500 | return -ENOMEM; | |
9501 | ||
9502 | err = setup_func_entry(env, subprog, insn_idx, set_callee_state_cb, | |
9503 | callback_state); | |
9504 | if (err) | |
9505 | return err; | |
9506 | ||
9507 | callback_state->callback_unroll_depth++; | |
bb124da6 EZ |
9508 | callback_state->frame[callback_state->curframe - 1]->callback_depth++; |
9509 | caller->callback_depth = 0; | |
ab5cfac1 EZ |
9510 | return 0; |
9511 | } | |
9512 | ||
9513 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
9514 | int *insn_idx) | |
9515 | { | |
9516 | struct bpf_verifier_state *state = env->cur_state; | |
9517 | struct bpf_func_state *caller; | |
9518 | int err, subprog, target_insn; | |
9519 | ||
9520 | target_insn = *insn_idx + insn->imm + 1; | |
9521 | subprog = find_subprog(env, target_insn); | |
9522 | if (subprog < 0) { | |
9523 | verbose(env, "verifier bug. No program starts at insn %d\n", target_insn); | |
9524 | return -EFAULT; | |
9525 | } | |
9526 | ||
9527 | caller = state->frame[state->curframe]; | |
9528 | err = btf_check_subprog_call(env, subprog, caller->regs); | |
9529 | if (err == -EFAULT) | |
9530 | return err; | |
9531 | if (subprog_is_global(env, subprog)) { | |
9532 | if (err) { | |
9533 | verbose(env, "Caller passes invalid args into func#%d\n", subprog); | |
9534 | return err; | |
9535 | } | |
9536 | ||
9537 | if (env->log.level & BPF_LOG_LEVEL) | |
9538 | verbose(env, "Func#%d is global and valid. Skipping.\n", subprog); | |
bfc6bb74 | 9539 | clear_caller_saved_regs(env, caller->regs); |
ab5cfac1 EZ |
9540 | |
9541 | /* All global functions return a 64-bit SCALAR_VALUE */ | |
bfc6bb74 AS |
9542 | mark_reg_unknown(env, caller->regs, BPF_REG_0); |
9543 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
ab5cfac1 | 9544 | |
bfc6bb74 AS |
9545 | /* continue with next insn after call */ |
9546 | return 0; | |
9547 | } | |
9548 | ||
ab5cfac1 EZ |
9549 | /* for regular function entry setup new frame and continue |
9550 | * from that frame. | |
9551 | */ | |
9552 | err = setup_func_entry(env, subprog, *insn_idx, set_callee_state, state); | |
fd978bf7 | 9553 | if (err) |
58124a98 | 9554 | return err; |
f4d7e40a | 9555 | |
51c39bb1 | 9556 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a | 9557 | |
f4d7e40a | 9558 | /* and go analyze first insn of the callee */ |
14351375 | 9559 | *insn_idx = env->subprog_info[subprog].start - 1; |
f4d7e40a | 9560 | |
06ee7115 | 9561 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 9562 | verbose(env, "caller:\n"); |
0f55f9ed | 9563 | print_verifier_state(env, caller, true); |
f4d7e40a | 9564 | verbose(env, "callee:\n"); |
58124a98 | 9565 | print_verifier_state(env, state->frame[state->curframe], true); |
f4d7e40a | 9566 | } |
eb86559a | 9567 | |
58124a98 | 9568 | return 0; |
f4d7e40a AS |
9569 | } |
9570 | ||
314ee05e YS |
9571 | int map_set_for_each_callback_args(struct bpf_verifier_env *env, |
9572 | struct bpf_func_state *caller, | |
9573 | struct bpf_func_state *callee) | |
9574 | { | |
9575 | /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn, | |
9576 | * void *callback_ctx, u64 flags); | |
9577 | * callback_fn(struct bpf_map *map, void *key, void *value, | |
9578 | * void *callback_ctx); | |
9579 | */ | |
9580 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
9581 | ||
9582 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
9583 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9584 | callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9585 | ||
9586 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
9587 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
9588 | callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9589 | ||
9590 | /* pointer to stack or null */ | |
9591 | callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3]; | |
9592 | ||
9593 | /* unused */ | |
9594 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9595 | return 0; | |
9596 | } | |
9597 | ||
14351375 YS |
9598 | static int set_callee_state(struct bpf_verifier_env *env, |
9599 | struct bpf_func_state *caller, | |
9600 | struct bpf_func_state *callee, int insn_idx) | |
9601 | { | |
9602 | int i; | |
9603 | ||
9604 | /* copy r1 - r5 args that callee can access. The copy includes parent | |
9605 | * pointers, which connects us up to the liveness chain | |
9606 | */ | |
9607 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
9608 | callee->regs[i] = caller->regs[i]; | |
9609 | return 0; | |
9610 | } | |
9611 | ||
69c087ba YS |
9612 | static int set_map_elem_callback_state(struct bpf_verifier_env *env, |
9613 | struct bpf_func_state *caller, | |
9614 | struct bpf_func_state *callee, | |
9615 | int insn_idx) | |
9616 | { | |
9617 | struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx]; | |
9618 | struct bpf_map *map; | |
9619 | int err; | |
9620 | ||
9621 | if (bpf_map_ptr_poisoned(insn_aux)) { | |
9622 | verbose(env, "tail_call abusing map_ptr\n"); | |
9623 | return -EINVAL; | |
9624 | } | |
9625 | ||
9626 | map = BPF_MAP_PTR(insn_aux->map_ptr_state); | |
9627 | if (!map->ops->map_set_for_each_callback_args || | |
9628 | !map->ops->map_for_each_callback) { | |
9629 | verbose(env, "callback function not allowed for map\n"); | |
9630 | return -ENOTSUPP; | |
9631 | } | |
9632 | ||
9633 | err = map->ops->map_set_for_each_callback_args(env, caller, callee); | |
9634 | if (err) | |
9635 | return err; | |
9636 | ||
9637 | callee->in_callback_fn = true; | |
1bfe26fb | 9638 | callee->callback_ret_range = tnum_range(0, 1); |
69c087ba YS |
9639 | return 0; |
9640 | } | |
9641 | ||
e6f2dd0f JK |
9642 | static int set_loop_callback_state(struct bpf_verifier_env *env, |
9643 | struct bpf_func_state *caller, | |
9644 | struct bpf_func_state *callee, | |
9645 | int insn_idx) | |
9646 | { | |
9647 | /* bpf_loop(u32 nr_loops, void *callback_fn, void *callback_ctx, | |
9648 | * u64 flags); | |
9649 | * callback_fn(u32 index, void *callback_ctx); | |
9650 | */ | |
9651 | callee->regs[BPF_REG_1].type = SCALAR_VALUE; | |
9652 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; | |
9653 | ||
9654 | /* unused */ | |
9655 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9656 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9657 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9658 | ||
9659 | callee->in_callback_fn = true; | |
1bfe26fb | 9660 | callee->callback_ret_range = tnum_range(0, 1); |
e6f2dd0f JK |
9661 | return 0; |
9662 | } | |
9663 | ||
b00628b1 AS |
9664 | static int set_timer_callback_state(struct bpf_verifier_env *env, |
9665 | struct bpf_func_state *caller, | |
9666 | struct bpf_func_state *callee, | |
9667 | int insn_idx) | |
9668 | { | |
9669 | struct bpf_map *map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9670 | ||
9671 | /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn); | |
9672 | * callback_fn(struct bpf_map *map, void *key, void *value); | |
9673 | */ | |
9674 | callee->regs[BPF_REG_1].type = CONST_PTR_TO_MAP; | |
9675 | __mark_reg_known_zero(&callee->regs[BPF_REG_1]); | |
9676 | callee->regs[BPF_REG_1].map_ptr = map_ptr; | |
9677 | ||
9678 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
9679 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9680 | callee->regs[BPF_REG_2].map_ptr = map_ptr; | |
9681 | ||
9682 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
9683 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
9684 | callee->regs[BPF_REG_3].map_ptr = map_ptr; | |
9685 | ||
9686 | /* unused */ | |
9687 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9688 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
bfc6bb74 | 9689 | callee->in_async_callback_fn = true; |
1bfe26fb | 9690 | callee->callback_ret_range = tnum_range(0, 1); |
b00628b1 AS |
9691 | return 0; |
9692 | } | |
9693 | ||
7c7e3d31 SL |
9694 | static int set_find_vma_callback_state(struct bpf_verifier_env *env, |
9695 | struct bpf_func_state *caller, | |
9696 | struct bpf_func_state *callee, | |
9697 | int insn_idx) | |
9698 | { | |
9699 | /* bpf_find_vma(struct task_struct *task, u64 addr, | |
9700 | * void *callback_fn, void *callback_ctx, u64 flags) | |
9701 | * (callback_fn)(struct task_struct *task, | |
9702 | * struct vm_area_struct *vma, void *callback_ctx); | |
9703 | */ | |
9704 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
9705 | ||
9706 | callee->regs[BPF_REG_2].type = PTR_TO_BTF_ID; | |
9707 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9708 | callee->regs[BPF_REG_2].btf = btf_vmlinux; | |
d19ddb47 | 9709 | callee->regs[BPF_REG_2].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA], |
7c7e3d31 SL |
9710 | |
9711 | /* pointer to stack or null */ | |
9712 | callee->regs[BPF_REG_3] = caller->regs[BPF_REG_4]; | |
9713 | ||
9714 | /* unused */ | |
9715 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9716 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9717 | callee->in_callback_fn = true; | |
1bfe26fb | 9718 | callee->callback_ret_range = tnum_range(0, 1); |
7c7e3d31 SL |
9719 | return 0; |
9720 | } | |
9721 | ||
20571567 DV |
9722 | static int set_user_ringbuf_callback_state(struct bpf_verifier_env *env, |
9723 | struct bpf_func_state *caller, | |
9724 | struct bpf_func_state *callee, | |
9725 | int insn_idx) | |
9726 | { | |
9727 | /* bpf_user_ringbuf_drain(struct bpf_map *map, void *callback_fn, void | |
9728 | * callback_ctx, u64 flags); | |
27060531 | 9729 | * callback_fn(const struct bpf_dynptr_t* dynptr, void *callback_ctx); |
20571567 DV |
9730 | */ |
9731 | __mark_reg_not_init(env, &callee->regs[BPF_REG_0]); | |
f8064ab9 | 9732 | mark_dynptr_cb_reg(env, &callee->regs[BPF_REG_1], BPF_DYNPTR_TYPE_LOCAL); |
20571567 DV |
9733 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; |
9734 | ||
9735 | /* unused */ | |
9736 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9737 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9738 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9739 | ||
9740 | callee->in_callback_fn = true; | |
c92a7a52 | 9741 | callee->callback_ret_range = tnum_range(0, 1); |
20571567 DV |
9742 | return 0; |
9743 | } | |
9744 | ||
5d92ddc3 DM |
9745 | static int set_rbtree_add_callback_state(struct bpf_verifier_env *env, |
9746 | struct bpf_func_state *caller, | |
9747 | struct bpf_func_state *callee, | |
9748 | int insn_idx) | |
9749 | { | |
d2dcc67d | 9750 | /* void bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node, |
5d92ddc3 DM |
9751 | * bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b)); |
9752 | * | |
d2dcc67d | 9753 | * 'struct bpf_rb_node *node' arg to bpf_rbtree_add_impl is the same PTR_TO_BTF_ID w/ offset |
5d92ddc3 DM |
9754 | * that 'less' callback args will be receiving. However, 'node' arg was release_reference'd |
9755 | * by this point, so look at 'root' | |
9756 | */ | |
9757 | struct btf_field *field; | |
9758 | ||
9759 | field = reg_find_field_offset(&caller->regs[BPF_REG_1], caller->regs[BPF_REG_1].off, | |
9760 | BPF_RB_ROOT); | |
9761 | if (!field || !field->graph_root.value_btf_id) | |
9762 | return -EFAULT; | |
9763 | ||
9764 | mark_reg_graph_node(callee->regs, BPF_REG_1, &field->graph_root); | |
9765 | ref_set_non_owning(env, &callee->regs[BPF_REG_1]); | |
9766 | mark_reg_graph_node(callee->regs, BPF_REG_2, &field->graph_root); | |
9767 | ref_set_non_owning(env, &callee->regs[BPF_REG_2]); | |
9768 | ||
9769 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9770 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9771 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9772 | callee->in_callback_fn = true; | |
9773 | callee->callback_ret_range = tnum_range(0, 1); | |
9774 | return 0; | |
9775 | } | |
9776 | ||
9777 | static bool is_rbtree_lock_required_kfunc(u32 btf_id); | |
9778 | ||
9779 | /* Are we currently verifying the callback for a rbtree helper that must | |
9780 | * be called with lock held? If so, no need to complain about unreleased | |
9781 | * lock | |
9782 | */ | |
9783 | static bool in_rbtree_lock_required_cb(struct bpf_verifier_env *env) | |
9784 | { | |
9785 | struct bpf_verifier_state *state = env->cur_state; | |
9786 | struct bpf_insn *insn = env->prog->insnsi; | |
9787 | struct bpf_func_state *callee; | |
9788 | int kfunc_btf_id; | |
9789 | ||
9790 | if (!state->curframe) | |
9791 | return false; | |
9792 | ||
9793 | callee = state->frame[state->curframe]; | |
9794 | ||
9795 | if (!callee->in_callback_fn) | |
9796 | return false; | |
9797 | ||
9798 | kfunc_btf_id = insn[callee->callsite].imm; | |
9799 | return is_rbtree_lock_required_kfunc(kfunc_btf_id); | |
9800 | } | |
9801 | ||
f4d7e40a AS |
9802 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) |
9803 | { | |
cafe2c21 | 9804 | struct bpf_verifier_state *state = env->cur_state, *prev_st; |
f4d7e40a AS |
9805 | struct bpf_func_state *caller, *callee; |
9806 | struct bpf_reg_state *r0; | |
cafe2c21 | 9807 | bool in_callback_fn; |
fd978bf7 | 9808 | int err; |
f4d7e40a AS |
9809 | |
9810 | callee = state->frame[state->curframe]; | |
9811 | r0 = &callee->regs[BPF_REG_0]; | |
9812 | if (r0->type == PTR_TO_STACK) { | |
9813 | /* technically it's ok to return caller's stack pointer | |
9814 | * (or caller's caller's pointer) back to the caller, | |
9815 | * since these pointers are valid. Only current stack | |
9816 | * pointer will be invalid as soon as function exits, | |
9817 | * but let's be conservative | |
9818 | */ | |
9819 | verbose(env, "cannot return stack pointer to the caller\n"); | |
9820 | return -EINVAL; | |
9821 | } | |
9822 | ||
eb86559a | 9823 | caller = state->frame[state->curframe - 1]; |
69c087ba YS |
9824 | if (callee->in_callback_fn) { |
9825 | /* enforce R0 return value range [0, 1]. */ | |
1bfe26fb | 9826 | struct tnum range = callee->callback_ret_range; |
69c087ba YS |
9827 | |
9828 | if (r0->type != SCALAR_VALUE) { | |
9829 | verbose(env, "R0 not a scalar value\n"); | |
9830 | return -EACCES; | |
9831 | } | |
9832 | if (!tnum_in(range, r0->var_off)) { | |
9833 | verbose_invalid_scalar(env, r0, &range, "callback return", "R0"); | |
9834 | return -EINVAL; | |
9835 | } | |
ab5cfac1 EZ |
9836 | if (!calls_callback(env, callee->callsite)) { |
9837 | verbose(env, "BUG: in callback at %d, callsite %d !calls_callback\n", | |
9838 | *insn_idx, callee->callsite); | |
9839 | return -EFAULT; | |
9840 | } | |
69c087ba YS |
9841 | } else { |
9842 | /* return to the caller whatever r0 had in the callee */ | |
9843 | caller->regs[BPF_REG_0] = *r0; | |
9844 | } | |
f4d7e40a | 9845 | |
9d9d00ac KKD |
9846 | /* callback_fn frame should have released its own additions to parent's |
9847 | * reference state at this point, or check_reference_leak would | |
9848 | * complain, hence it must be the same as the caller. There is no need | |
9849 | * to copy it back. | |
9850 | */ | |
9851 | if (!callee->in_callback_fn) { | |
9852 | /* Transfer references to the caller */ | |
9853 | err = copy_reference_state(caller, callee); | |
9854 | if (err) | |
9855 | return err; | |
9856 | } | |
fd978bf7 | 9857 | |
ab5cfac1 EZ |
9858 | /* for callbacks like bpf_loop or bpf_for_each_map_elem go back to callsite, |
9859 | * there function call logic would reschedule callback visit. If iteration | |
9860 | * converges is_state_visited() would prune that visit eventually. | |
9861 | */ | |
cafe2c21 EZ |
9862 | in_callback_fn = callee->in_callback_fn; |
9863 | if (in_callback_fn) | |
ab5cfac1 EZ |
9864 | *insn_idx = callee->callsite; |
9865 | else | |
9866 | *insn_idx = callee->callsite + 1; | |
9867 | ||
06ee7115 | 9868 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 9869 | verbose(env, "returning from callee:\n"); |
0f55f9ed | 9870 | print_verifier_state(env, callee, true); |
f4d7e40a | 9871 | verbose(env, "to caller at %d:\n", *insn_idx); |
0f55f9ed | 9872 | print_verifier_state(env, caller, true); |
f4d7e40a | 9873 | } |
f18b03fa KKD |
9874 | /* clear everything in the callee. In case of exceptional exits using |
9875 | * bpf_throw, this will be done by copy_verifier_state for extra frames. */ | |
f4d7e40a | 9876 | free_func_state(callee); |
eb86559a | 9877 | state->frame[state->curframe--] = NULL; |
cafe2c21 EZ |
9878 | |
9879 | /* for callbacks widen imprecise scalars to make programs like below verify: | |
9880 | * | |
9881 | * struct ctx { int i; } | |
9882 | * void cb(int idx, struct ctx *ctx) { ctx->i++; ... } | |
9883 | * ... | |
9884 | * struct ctx = { .i = 0; } | |
9885 | * bpf_loop(100, cb, &ctx, 0); | |
9886 | * | |
9887 | * This is similar to what is done in process_iter_next_call() for open | |
9888 | * coded iterators. | |
9889 | */ | |
9890 | prev_st = in_callback_fn ? find_prev_entry(env, state, *insn_idx) : NULL; | |
9891 | if (prev_st) { | |
9892 | err = widen_imprecise_scalars(env, prev_st, state); | |
9893 | if (err) | |
9894 | return err; | |
9895 | } | |
f4d7e40a AS |
9896 | return 0; |
9897 | } | |
9898 | ||
849fa506 YS |
9899 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
9900 | int func_id, | |
9901 | struct bpf_call_arg_meta *meta) | |
9902 | { | |
9903 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
9904 | ||
f42bcd16 | 9905 | if (ret_type != RET_INTEGER) |
849fa506 YS |
9906 | return; |
9907 | ||
f42bcd16 AN |
9908 | switch (func_id) { |
9909 | case BPF_FUNC_get_stack: | |
9910 | case BPF_FUNC_get_task_stack: | |
9911 | case BPF_FUNC_probe_read_str: | |
9912 | case BPF_FUNC_probe_read_kernel_str: | |
9913 | case BPF_FUNC_probe_read_user_str: | |
9914 | ret_reg->smax_value = meta->msize_max_value; | |
9915 | ret_reg->s32_max_value = meta->msize_max_value; | |
9916 | ret_reg->smin_value = -MAX_ERRNO; | |
9917 | ret_reg->s32_min_value = -MAX_ERRNO; | |
9918 | reg_bounds_sync(ret_reg); | |
9919 | break; | |
9920 | case BPF_FUNC_get_smp_processor_id: | |
9921 | ret_reg->umax_value = nr_cpu_ids - 1; | |
9922 | ret_reg->u32_max_value = nr_cpu_ids - 1; | |
9923 | ret_reg->smax_value = nr_cpu_ids - 1; | |
9924 | ret_reg->s32_max_value = nr_cpu_ids - 1; | |
9925 | ret_reg->umin_value = 0; | |
9926 | ret_reg->u32_min_value = 0; | |
9927 | ret_reg->smin_value = 0; | |
9928 | ret_reg->s32_min_value = 0; | |
9929 | reg_bounds_sync(ret_reg); | |
9930 | break; | |
9931 | } | |
849fa506 YS |
9932 | } |
9933 | ||
c93552c4 DB |
9934 | static int |
9935 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
9936 | int func_id, int insn_idx) | |
9937 | { | |
9938 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 9939 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
9940 | |
9941 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
9942 | func_id != BPF_FUNC_map_lookup_elem && |
9943 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
9944 | func_id != BPF_FUNC_map_delete_elem && |
9945 | func_id != BPF_FUNC_map_push_elem && | |
9946 | func_id != BPF_FUNC_map_pop_elem && | |
69c087ba | 9947 | func_id != BPF_FUNC_map_peek_elem && |
e6a4750f | 9948 | func_id != BPF_FUNC_for_each_map_elem && |
07343110 FZ |
9949 | func_id != BPF_FUNC_redirect_map && |
9950 | func_id != BPF_FUNC_map_lookup_percpu_elem) | |
c93552c4 | 9951 | return 0; |
09772d92 | 9952 | |
591fe988 | 9953 | if (map == NULL) { |
c93552c4 DB |
9954 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
9955 | return -EINVAL; | |
9956 | } | |
9957 | ||
591fe988 DB |
9958 | /* In case of read-only, some additional restrictions |
9959 | * need to be applied in order to prevent altering the | |
9960 | * state of the map from program side. | |
9961 | */ | |
9962 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
9963 | (func_id == BPF_FUNC_map_delete_elem || | |
9964 | func_id == BPF_FUNC_map_update_elem || | |
9965 | func_id == BPF_FUNC_map_push_elem || | |
9966 | func_id == BPF_FUNC_map_pop_elem)) { | |
9967 | verbose(env, "write into map forbidden\n"); | |
9968 | return -EACCES; | |
9969 | } | |
9970 | ||
d2e4c1e6 | 9971 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 9972 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 9973 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 9974 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 9975 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 9976 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
9977 | return 0; |
9978 | } | |
9979 | ||
d2e4c1e6 DB |
9980 | static int |
9981 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
9982 | int func_id, int insn_idx) | |
9983 | { | |
9984 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
9985 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
9986 | struct bpf_map *map = meta->map_ptr; | |
a657182a | 9987 | u64 val, max; |
cc52d914 | 9988 | int err; |
d2e4c1e6 DB |
9989 | |
9990 | if (func_id != BPF_FUNC_tail_call) | |
9991 | return 0; | |
9992 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
9993 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
9994 | return -EINVAL; | |
9995 | } | |
9996 | ||
d2e4c1e6 | 9997 | reg = ®s[BPF_REG_3]; |
a657182a DB |
9998 | val = reg->var_off.value; |
9999 | max = map->max_entries; | |
d2e4c1e6 | 10000 | |
a657182a | 10001 | if (!(register_is_const(reg) && val < max)) { |
d2e4c1e6 DB |
10002 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); |
10003 | return 0; | |
10004 | } | |
10005 | ||
cc52d914 DB |
10006 | err = mark_chain_precision(env, BPF_REG_3); |
10007 | if (err) | |
10008 | return err; | |
d2e4c1e6 DB |
10009 | if (bpf_map_key_unseen(aux)) |
10010 | bpf_map_key_store(aux, val); | |
10011 | else if (!bpf_map_key_poisoned(aux) && | |
10012 | bpf_map_key_immediate(aux) != val) | |
10013 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
10014 | return 0; | |
10015 | } | |
10016 | ||
f18b03fa | 10017 | static int check_reference_leak(struct bpf_verifier_env *env, bool exception_exit) |
fd978bf7 JS |
10018 | { |
10019 | struct bpf_func_state *state = cur_func(env); | |
9d9d00ac | 10020 | bool refs_lingering = false; |
fd978bf7 JS |
10021 | int i; |
10022 | ||
f18b03fa | 10023 | if (!exception_exit && state->frameno && !state->in_callback_fn) |
9d9d00ac KKD |
10024 | return 0; |
10025 | ||
fd978bf7 | 10026 | for (i = 0; i < state->acquired_refs; i++) { |
f18b03fa | 10027 | if (!exception_exit && state->in_callback_fn && state->refs[i].callback_ref != state->frameno) |
9d9d00ac | 10028 | continue; |
fd978bf7 JS |
10029 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", |
10030 | state->refs[i].id, state->refs[i].insn_idx); | |
9d9d00ac | 10031 | refs_lingering = true; |
fd978bf7 | 10032 | } |
9d9d00ac | 10033 | return refs_lingering ? -EINVAL : 0; |
fd978bf7 JS |
10034 | } |
10035 | ||
7b15523a FR |
10036 | static int check_bpf_snprintf_call(struct bpf_verifier_env *env, |
10037 | struct bpf_reg_state *regs) | |
10038 | { | |
10039 | struct bpf_reg_state *fmt_reg = ®s[BPF_REG_3]; | |
10040 | struct bpf_reg_state *data_len_reg = ®s[BPF_REG_5]; | |
10041 | struct bpf_map *fmt_map = fmt_reg->map_ptr; | |
78aa1cc9 | 10042 | struct bpf_bprintf_data data = {}; |
7b15523a FR |
10043 | int err, fmt_map_off, num_args; |
10044 | u64 fmt_addr; | |
10045 | char *fmt; | |
10046 | ||
10047 | /* data must be an array of u64 */ | |
10048 | if (data_len_reg->var_off.value % 8) | |
10049 | return -EINVAL; | |
10050 | num_args = data_len_reg->var_off.value / 8; | |
10051 | ||
10052 | /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const | |
10053 | * and map_direct_value_addr is set. | |
10054 | */ | |
10055 | fmt_map_off = fmt_reg->off + fmt_reg->var_off.value; | |
10056 | err = fmt_map->ops->map_direct_value_addr(fmt_map, &fmt_addr, | |
10057 | fmt_map_off); | |
8e8ee109 FR |
10058 | if (err) { |
10059 | verbose(env, "verifier bug\n"); | |
10060 | return -EFAULT; | |
10061 | } | |
7b15523a FR |
10062 | fmt = (char *)(long)fmt_addr + fmt_map_off; |
10063 | ||
10064 | /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we | |
10065 | * can focus on validating the format specifiers. | |
10066 | */ | |
78aa1cc9 | 10067 | err = bpf_bprintf_prepare(fmt, UINT_MAX, NULL, num_args, &data); |
7b15523a FR |
10068 | if (err < 0) |
10069 | verbose(env, "Invalid format string\n"); | |
10070 | ||
10071 | return err; | |
10072 | } | |
10073 | ||
9b99edca JO |
10074 | static int check_get_func_ip(struct bpf_verifier_env *env) |
10075 | { | |
9b99edca JO |
10076 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
10077 | int func_id = BPF_FUNC_get_func_ip; | |
10078 | ||
10079 | if (type == BPF_PROG_TYPE_TRACING) { | |
f92c1e18 | 10080 | if (!bpf_prog_has_trampoline(env->prog)) { |
9b99edca JO |
10081 | verbose(env, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n", |
10082 | func_id_name(func_id), func_id); | |
10083 | return -ENOTSUPP; | |
10084 | } | |
10085 | return 0; | |
9ffd9f3f JO |
10086 | } else if (type == BPF_PROG_TYPE_KPROBE) { |
10087 | return 0; | |
9b99edca JO |
10088 | } |
10089 | ||
10090 | verbose(env, "func %s#%d not supported for program type %d\n", | |
10091 | func_id_name(func_id), func_id, type); | |
10092 | return -ENOTSUPP; | |
10093 | } | |
10094 | ||
1ade2371 EZ |
10095 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
10096 | { | |
10097 | return &env->insn_aux_data[env->insn_idx]; | |
10098 | } | |
10099 | ||
10100 | static bool loop_flag_is_zero(struct bpf_verifier_env *env) | |
10101 | { | |
10102 | struct bpf_reg_state *regs = cur_regs(env); | |
10103 | struct bpf_reg_state *reg = ®s[BPF_REG_4]; | |
10104 | bool reg_is_null = register_is_null(reg); | |
10105 | ||
10106 | if (reg_is_null) | |
10107 | mark_chain_precision(env, BPF_REG_4); | |
10108 | ||
10109 | return reg_is_null; | |
10110 | } | |
10111 | ||
10112 | static void update_loop_inline_state(struct bpf_verifier_env *env, u32 subprogno) | |
10113 | { | |
10114 | struct bpf_loop_inline_state *state = &cur_aux(env)->loop_inline_state; | |
10115 | ||
10116 | if (!state->initialized) { | |
10117 | state->initialized = 1; | |
10118 | state->fit_for_inline = loop_flag_is_zero(env); | |
10119 | state->callback_subprogno = subprogno; | |
10120 | return; | |
10121 | } | |
10122 | ||
10123 | if (!state->fit_for_inline) | |
10124 | return; | |
10125 | ||
10126 | state->fit_for_inline = (loop_flag_is_zero(env) && | |
10127 | state->callback_subprogno == subprogno); | |
10128 | } | |
10129 | ||
69c087ba YS |
10130 | static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
10131 | int *insn_idx_p) | |
17a52670 | 10132 | { |
aef9d4a3 | 10133 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
01cc55af | 10134 | bool returns_cpu_specific_alloc_ptr = false; |
17a52670 | 10135 | const struct bpf_func_proto *fn = NULL; |
3c480732 | 10136 | enum bpf_return_type ret_type; |
c25b2ae1 | 10137 | enum bpf_type_flag ret_flag; |
638f5b90 | 10138 | struct bpf_reg_state *regs; |
33ff9823 | 10139 | struct bpf_call_arg_meta meta; |
69c087ba | 10140 | int insn_idx = *insn_idx_p; |
969bf05e | 10141 | bool changes_data; |
69c087ba | 10142 | int i, err, func_id; |
17a52670 AS |
10143 | |
10144 | /* find function prototype */ | |
69c087ba | 10145 | func_id = insn->imm; |
17a52670 | 10146 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { |
61bd5218 JK |
10147 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
10148 | func_id); | |
17a52670 AS |
10149 | return -EINVAL; |
10150 | } | |
10151 | ||
00176a34 | 10152 | if (env->ops->get_func_proto) |
5e43f899 | 10153 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 10154 | if (!fn) { |
61bd5218 JK |
10155 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
10156 | func_id); | |
17a52670 AS |
10157 | return -EINVAL; |
10158 | } | |
10159 | ||
10160 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 10161 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 10162 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
10163 | return -EINVAL; |
10164 | } | |
10165 | ||
eae2e83e JO |
10166 | if (fn->allowed && !fn->allowed(env->prog)) { |
10167 | verbose(env, "helper call is not allowed in probe\n"); | |
10168 | return -EINVAL; | |
10169 | } | |
10170 | ||
01685c5b YS |
10171 | if (!env->prog->aux->sleepable && fn->might_sleep) { |
10172 | verbose(env, "helper call might sleep in a non-sleepable prog\n"); | |
10173 | return -EINVAL; | |
10174 | } | |
10175 | ||
04514d13 | 10176 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 10177 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
10178 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
10179 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
10180 | func_id_name(func_id), func_id); | |
10181 | return -EINVAL; | |
10182 | } | |
969bf05e | 10183 | |
33ff9823 | 10184 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 10185 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 10186 | |
0c9a7a7e | 10187 | err = check_func_proto(fn, func_id); |
435faee1 | 10188 | if (err) { |
61bd5218 | 10189 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 10190 | func_id_name(func_id), func_id); |
435faee1 DB |
10191 | return err; |
10192 | } | |
10193 | ||
9bb00b28 YS |
10194 | if (env->cur_state->active_rcu_lock) { |
10195 | if (fn->might_sleep) { | |
10196 | verbose(env, "sleepable helper %s#%d in rcu_read_lock region\n", | |
10197 | func_id_name(func_id), func_id); | |
10198 | return -EINVAL; | |
10199 | } | |
10200 | ||
10201 | if (env->prog->aux->sleepable && is_storage_get_function(func_id)) | |
10202 | env->insn_aux_data[insn_idx].storage_get_func_atomic = true; | |
10203 | } | |
10204 | ||
d83525ca | 10205 | meta.func_id = func_id; |
17a52670 | 10206 | /* check args */ |
523a4cf4 | 10207 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { |
1d18feb2 | 10208 | err = check_func_arg(env, i, &meta, fn, insn_idx); |
a7658e1a AS |
10209 | if (err) |
10210 | return err; | |
10211 | } | |
17a52670 | 10212 | |
c93552c4 DB |
10213 | err = record_func_map(env, &meta, func_id, insn_idx); |
10214 | if (err) | |
10215 | return err; | |
10216 | ||
d2e4c1e6 DB |
10217 | err = record_func_key(env, &meta, func_id, insn_idx); |
10218 | if (err) | |
10219 | return err; | |
10220 | ||
435faee1 DB |
10221 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
10222 | * is inferred from register state. | |
10223 | */ | |
10224 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 | 10225 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
1f9a1ea8 | 10226 | BPF_WRITE, -1, false, false); |
435faee1 DB |
10227 | if (err) |
10228 | return err; | |
10229 | } | |
10230 | ||
8f14852e KKD |
10231 | regs = cur_regs(env); |
10232 | ||
10233 | if (meta.release_regno) { | |
10234 | err = -EINVAL; | |
27060531 KKD |
10235 | /* This can only be set for PTR_TO_STACK, as CONST_PTR_TO_DYNPTR cannot |
10236 | * be released by any dynptr helper. Hence, unmark_stack_slots_dynptr | |
10237 | * is safe to do directly. | |
10238 | */ | |
10239 | if (arg_type_is_dynptr(fn->arg_type[meta.release_regno - BPF_REG_1])) { | |
10240 | if (regs[meta.release_regno].type == CONST_PTR_TO_DYNPTR) { | |
10241 | verbose(env, "verifier internal error: CONST_PTR_TO_DYNPTR cannot be released\n"); | |
10242 | return -EFAULT; | |
10243 | } | |
97e03f52 | 10244 | err = unmark_stack_slots_dynptr(env, ®s[meta.release_regno]); |
5b221ecb YS |
10245 | } else if (func_id == BPF_FUNC_kptr_xchg && meta.ref_obj_id) { |
10246 | u32 ref_obj_id = meta.ref_obj_id; | |
10247 | bool in_rcu = in_rcu_cs(env); | |
10248 | struct bpf_func_state *state; | |
10249 | struct bpf_reg_state *reg; | |
10250 | ||
10251 | err = release_reference_state(cur_func(env), ref_obj_id); | |
10252 | if (!err) { | |
10253 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ | |
10254 | if (reg->ref_obj_id == ref_obj_id) { | |
10255 | if (in_rcu && (reg->type & MEM_ALLOC) && (reg->type & MEM_PERCPU)) { | |
10256 | reg->ref_obj_id = 0; | |
10257 | reg->type &= ~MEM_ALLOC; | |
10258 | reg->type |= MEM_RCU; | |
10259 | } else { | |
10260 | mark_reg_invalid(env, reg); | |
10261 | } | |
10262 | } | |
10263 | })); | |
10264 | } | |
27060531 | 10265 | } else if (meta.ref_obj_id) { |
8f14852e | 10266 | err = release_reference(env, meta.ref_obj_id); |
27060531 KKD |
10267 | } else if (register_is_null(®s[meta.release_regno])) { |
10268 | /* meta.ref_obj_id can only be 0 if register that is meant to be | |
10269 | * released is NULL, which must be > R0. | |
10270 | */ | |
8f14852e | 10271 | err = 0; |
27060531 | 10272 | } |
46f8bc92 MKL |
10273 | if (err) { |
10274 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
10275 | func_id_name(func_id), func_id); | |
fd978bf7 | 10276 | return err; |
46f8bc92 | 10277 | } |
fd978bf7 JS |
10278 | } |
10279 | ||
e6f2dd0f JK |
10280 | switch (func_id) { |
10281 | case BPF_FUNC_tail_call: | |
f18b03fa | 10282 | err = check_reference_leak(env, false); |
e6f2dd0f JK |
10283 | if (err) { |
10284 | verbose(env, "tail_call would lead to reference leak\n"); | |
10285 | return err; | |
10286 | } | |
10287 | break; | |
10288 | case BPF_FUNC_get_local_storage: | |
10289 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
10290 | * this is required because get_local_storage() can't return an error. | |
10291 | */ | |
10292 | if (!register_is_null(®s[BPF_REG_2])) { | |
10293 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
10294 | return -EINVAL; | |
10295 | } | |
10296 | break; | |
10297 | case BPF_FUNC_for_each_map_elem: | |
ab5cfac1 EZ |
10298 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, |
10299 | set_map_elem_callback_state); | |
e6f2dd0f JK |
10300 | break; |
10301 | case BPF_FUNC_timer_set_callback: | |
ab5cfac1 EZ |
10302 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, |
10303 | set_timer_callback_state); | |
e6f2dd0f JK |
10304 | break; |
10305 | case BPF_FUNC_find_vma: | |
ab5cfac1 EZ |
10306 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, |
10307 | set_find_vma_callback_state); | |
e6f2dd0f JK |
10308 | break; |
10309 | case BPF_FUNC_snprintf: | |
7b15523a | 10310 | err = check_bpf_snprintf_call(env, regs); |
e6f2dd0f JK |
10311 | break; |
10312 | case BPF_FUNC_loop: | |
1ade2371 | 10313 | update_loop_inline_state(env, meta.subprogno); |
bb124da6 EZ |
10314 | /* Verifier relies on R1 value to determine if bpf_loop() iteration |
10315 | * is finished, thus mark it precise. | |
10316 | */ | |
10317 | err = mark_chain_precision(env, BPF_REG_1); | |
10318 | if (err) | |
10319 | return err; | |
10320 | if (cur_func(env)->callback_depth < regs[BPF_REG_1].umax_value) { | |
10321 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, | |
10322 | set_loop_callback_state); | |
10323 | } else { | |
10324 | cur_func(env)->callback_depth = 0; | |
10325 | if (env->log.level & BPF_LOG_LEVEL2) | |
10326 | verbose(env, "frame%d bpf_loop iteration limit reached\n", | |
10327 | env->cur_state->curframe); | |
10328 | } | |
e6f2dd0f | 10329 | break; |
263ae152 JK |
10330 | case BPF_FUNC_dynptr_from_mem: |
10331 | if (regs[BPF_REG_1].type != PTR_TO_MAP_VALUE) { | |
10332 | verbose(env, "Unsupported reg type %s for bpf_dynptr_from_mem data\n", | |
10333 | reg_type_str(env, regs[BPF_REG_1].type)); | |
10334 | return -EACCES; | |
10335 | } | |
69fd337a SF |
10336 | break; |
10337 | case BPF_FUNC_set_retval: | |
aef9d4a3 SF |
10338 | if (prog_type == BPF_PROG_TYPE_LSM && |
10339 | env->prog->expected_attach_type == BPF_LSM_CGROUP) { | |
69fd337a SF |
10340 | if (!env->prog->aux->attach_func_proto->type) { |
10341 | /* Make sure programs that attach to void | |
10342 | * hooks don't try to modify return value. | |
10343 | */ | |
10344 | verbose(env, "BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
10345 | return -EINVAL; | |
10346 | } | |
10347 | } | |
10348 | break; | |
88374342 | 10349 | case BPF_FUNC_dynptr_data: |
485ec51e JK |
10350 | { |
10351 | struct bpf_reg_state *reg; | |
10352 | int id, ref_obj_id; | |
20571567 | 10353 | |
485ec51e JK |
10354 | reg = get_dynptr_arg_reg(env, fn, regs); |
10355 | if (!reg) | |
10356 | return -EFAULT; | |
f8064ab9 | 10357 | |
f8064ab9 | 10358 | |
485ec51e JK |
10359 | if (meta.dynptr_id) { |
10360 | verbose(env, "verifier internal error: meta.dynptr_id already set\n"); | |
10361 | return -EFAULT; | |
88374342 | 10362 | } |
485ec51e JK |
10363 | if (meta.ref_obj_id) { |
10364 | verbose(env, "verifier internal error: meta.ref_obj_id already set\n"); | |
88374342 JK |
10365 | return -EFAULT; |
10366 | } | |
485ec51e JK |
10367 | |
10368 | id = dynptr_id(env, reg); | |
10369 | if (id < 0) { | |
10370 | verbose(env, "verifier internal error: failed to obtain dynptr id\n"); | |
10371 | return id; | |
10372 | } | |
10373 | ||
10374 | ref_obj_id = dynptr_ref_obj_id(env, reg); | |
10375 | if (ref_obj_id < 0) { | |
10376 | verbose(env, "verifier internal error: failed to obtain dynptr ref_obj_id\n"); | |
10377 | return ref_obj_id; | |
10378 | } | |
10379 | ||
10380 | meta.dynptr_id = id; | |
10381 | meta.ref_obj_id = ref_obj_id; | |
10382 | ||
88374342 | 10383 | break; |
485ec51e | 10384 | } |
b5964b96 JK |
10385 | case BPF_FUNC_dynptr_write: |
10386 | { | |
10387 | enum bpf_dynptr_type dynptr_type; | |
10388 | struct bpf_reg_state *reg; | |
10389 | ||
10390 | reg = get_dynptr_arg_reg(env, fn, regs); | |
10391 | if (!reg) | |
10392 | return -EFAULT; | |
10393 | ||
10394 | dynptr_type = dynptr_get_type(env, reg); | |
10395 | if (dynptr_type == BPF_DYNPTR_TYPE_INVALID) | |
10396 | return -EFAULT; | |
10397 | ||
10398 | if (dynptr_type == BPF_DYNPTR_TYPE_SKB) | |
10399 | /* this will trigger clear_all_pkt_pointers(), which will | |
10400 | * invalidate all dynptr slices associated with the skb | |
10401 | */ | |
10402 | changes_data = true; | |
10403 | ||
10404 | break; | |
10405 | } | |
01cc55af YS |
10406 | case BPF_FUNC_per_cpu_ptr: |
10407 | case BPF_FUNC_this_cpu_ptr: | |
10408 | { | |
10409 | struct bpf_reg_state *reg = ®s[BPF_REG_1]; | |
10410 | const struct btf_type *type; | |
10411 | ||
10412 | if (reg->type & MEM_RCU) { | |
10413 | type = btf_type_by_id(reg->btf, reg->btf_id); | |
10414 | if (!type || !btf_type_is_struct(type)) { | |
10415 | verbose(env, "Helper has invalid btf/btf_id in R1\n"); | |
10416 | return -EFAULT; | |
10417 | } | |
10418 | returns_cpu_specific_alloc_ptr = true; | |
10419 | env->insn_aux_data[insn_idx].call_with_percpu_alloc_ptr = true; | |
10420 | } | |
10421 | break; | |
10422 | } | |
20571567 | 10423 | case BPF_FUNC_user_ringbuf_drain: |
ab5cfac1 EZ |
10424 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, |
10425 | set_user_ringbuf_callback_state); | |
20571567 | 10426 | break; |
7b15523a FR |
10427 | } |
10428 | ||
e6f2dd0f JK |
10429 | if (err) |
10430 | return err; | |
10431 | ||
17a52670 | 10432 | /* reset caller saved regs */ |
dc503a8a | 10433 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 10434 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
10435 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
10436 | } | |
17a52670 | 10437 | |
5327ed3d JW |
10438 | /* helper call returns 64-bit value. */ |
10439 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
10440 | ||
dc503a8a | 10441 | /* update return register (already marked as written above) */ |
3c480732 | 10442 | ret_type = fn->ret_type; |
0c9a7a7e JK |
10443 | ret_flag = type_flag(ret_type); |
10444 | ||
10445 | switch (base_type(ret_type)) { | |
10446 | case RET_INTEGER: | |
f1174f77 | 10447 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 10448 | mark_reg_unknown(env, regs, BPF_REG_0); |
0c9a7a7e JK |
10449 | break; |
10450 | case RET_VOID: | |
17a52670 | 10451 | regs[BPF_REG_0].type = NOT_INIT; |
0c9a7a7e JK |
10452 | break; |
10453 | case RET_PTR_TO_MAP_VALUE: | |
f1174f77 | 10454 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 10455 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
10456 | /* remember map_ptr, so that check_map_access() |
10457 | * can check 'value_size' boundary of memory access | |
10458 | * to map element returned from bpf_map_lookup_elem() | |
10459 | */ | |
33ff9823 | 10460 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
10461 | verbose(env, |
10462 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
10463 | return -EINVAL; |
10464 | } | |
33ff9823 | 10465 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
3e8ce298 | 10466 | regs[BPF_REG_0].map_uid = meta.map_uid; |
c25b2ae1 HL |
10467 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE | ret_flag; |
10468 | if (!type_may_be_null(ret_type) && | |
db559117 | 10469 | btf_record_has_field(meta.map_ptr->record, BPF_SPIN_LOCK)) { |
c25b2ae1 | 10470 | regs[BPF_REG_0].id = ++env->id_gen; |
4d31f301 | 10471 | } |
0c9a7a7e JK |
10472 | break; |
10473 | case RET_PTR_TO_SOCKET: | |
c64b7983 | 10474 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 10475 | regs[BPF_REG_0].type = PTR_TO_SOCKET | ret_flag; |
0c9a7a7e JK |
10476 | break; |
10477 | case RET_PTR_TO_SOCK_COMMON: | |
85a51f8c | 10478 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 10479 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON | ret_flag; |
0c9a7a7e JK |
10480 | break; |
10481 | case RET_PTR_TO_TCP_SOCK: | |
655a51e5 | 10482 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 10483 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK | ret_flag; |
0c9a7a7e | 10484 | break; |
2de2669b | 10485 | case RET_PTR_TO_MEM: |
457f4436 | 10486 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 10487 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
457f4436 | 10488 | regs[BPF_REG_0].mem_size = meta.mem_size; |
0c9a7a7e JK |
10489 | break; |
10490 | case RET_PTR_TO_MEM_OR_BTF_ID: | |
10491 | { | |
eaa6bcb7 HL |
10492 | const struct btf_type *t; |
10493 | ||
10494 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
22dc4a0f | 10495 | t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL); |
eaa6bcb7 HL |
10496 | if (!btf_type_is_struct(t)) { |
10497 | u32 tsize; | |
10498 | const struct btf_type *ret; | |
10499 | const char *tname; | |
10500 | ||
10501 | /* resolve the type size of ksym. */ | |
22dc4a0f | 10502 | ret = btf_resolve_size(meta.ret_btf, t, &tsize); |
eaa6bcb7 | 10503 | if (IS_ERR(ret)) { |
22dc4a0f | 10504 | tname = btf_name_by_offset(meta.ret_btf, t->name_off); |
eaa6bcb7 HL |
10505 | verbose(env, "unable to resolve the size of type '%s': %ld\n", |
10506 | tname, PTR_ERR(ret)); | |
10507 | return -EINVAL; | |
10508 | } | |
c25b2ae1 | 10509 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
eaa6bcb7 HL |
10510 | regs[BPF_REG_0].mem_size = tsize; |
10511 | } else { | |
01cc55af YS |
10512 | if (returns_cpu_specific_alloc_ptr) { |
10513 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC | MEM_RCU; | |
10514 | } else { | |
10515 | /* MEM_RDONLY may be carried from ret_flag, but it | |
10516 | * doesn't apply on PTR_TO_BTF_ID. Fold it, otherwise | |
10517 | * it will confuse the check of PTR_TO_BTF_ID in | |
10518 | * check_mem_access(). | |
10519 | */ | |
10520 | ret_flag &= ~MEM_RDONLY; | |
10521 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; | |
10522 | } | |
34d3a78c | 10523 | |
22dc4a0f | 10524 | regs[BPF_REG_0].btf = meta.ret_btf; |
eaa6bcb7 HL |
10525 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; |
10526 | } | |
0c9a7a7e JK |
10527 | break; |
10528 | } | |
10529 | case RET_PTR_TO_BTF_ID: | |
10530 | { | |
c0a5a21c | 10531 | struct btf *ret_btf; |
af7ec138 YS |
10532 | int ret_btf_id; |
10533 | ||
10534 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
c25b2ae1 | 10535 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; |
c0a5a21c | 10536 | if (func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac KKD |
10537 | ret_btf = meta.kptr_field->kptr.btf; |
10538 | ret_btf_id = meta.kptr_field->kptr.btf_id; | |
36d8bdf7 | 10539 | if (!btf_is_kernel(ret_btf)) { |
738c96d5 | 10540 | regs[BPF_REG_0].type |= MEM_ALLOC; |
36d8bdf7 YS |
10541 | if (meta.kptr_field->type == BPF_KPTR_PERCPU) |
10542 | regs[BPF_REG_0].type |= MEM_PERCPU; | |
10543 | } | |
c0a5a21c | 10544 | } else { |
47e34cb7 DM |
10545 | if (fn->ret_btf_id == BPF_PTR_POISON) { |
10546 | verbose(env, "verifier internal error:"); | |
10547 | verbose(env, "func %s has non-overwritten BPF_PTR_POISON return type\n", | |
10548 | func_id_name(func_id)); | |
10549 | return -EINVAL; | |
10550 | } | |
c0a5a21c KKD |
10551 | ret_btf = btf_vmlinux; |
10552 | ret_btf_id = *fn->ret_btf_id; | |
10553 | } | |
af7ec138 | 10554 | if (ret_btf_id == 0) { |
3c480732 HL |
10555 | verbose(env, "invalid return type %u of func %s#%d\n", |
10556 | base_type(ret_type), func_id_name(func_id), | |
10557 | func_id); | |
af7ec138 YS |
10558 | return -EINVAL; |
10559 | } | |
c0a5a21c | 10560 | regs[BPF_REG_0].btf = ret_btf; |
af7ec138 | 10561 | regs[BPF_REG_0].btf_id = ret_btf_id; |
0c9a7a7e JK |
10562 | break; |
10563 | } | |
10564 | default: | |
3c480732 HL |
10565 | verbose(env, "unknown return type %u of func %s#%d\n", |
10566 | base_type(ret_type), func_id_name(func_id), func_id); | |
17a52670 AS |
10567 | return -EINVAL; |
10568 | } | |
04fd61ab | 10569 | |
c25b2ae1 | 10570 | if (type_may_be_null(regs[BPF_REG_0].type)) |
93c230e3 MKL |
10571 | regs[BPF_REG_0].id = ++env->id_gen; |
10572 | ||
b2d8ef19 DM |
10573 | if (helper_multiple_ref_obj_use(func_id, meta.map_ptr)) { |
10574 | verbose(env, "verifier internal error: func %s#%d sets ref_obj_id more than once\n", | |
10575 | func_id_name(func_id), func_id); | |
10576 | return -EFAULT; | |
10577 | } | |
10578 | ||
f8064ab9 KKD |
10579 | if (is_dynptr_ref_function(func_id)) |
10580 | regs[BPF_REG_0].dynptr_id = meta.dynptr_id; | |
10581 | ||
88374342 | 10582 | if (is_ptr_cast_function(func_id) || is_dynptr_ref_function(func_id)) { |
1b986589 MKL |
10583 | /* For release_reference() */ |
10584 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 10585 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
10586 | int id = acquire_reference_state(env, insn_idx); |
10587 | ||
10588 | if (id < 0) | |
10589 | return id; | |
10590 | /* For mark_ptr_or_null_reg() */ | |
10591 | regs[BPF_REG_0].id = id; | |
10592 | /* For release_reference() */ | |
10593 | regs[BPF_REG_0].ref_obj_id = id; | |
10594 | } | |
1b986589 | 10595 | |
849fa506 YS |
10596 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
10597 | ||
61bd5218 | 10598 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
10599 | if (err) |
10600 | return err; | |
04fd61ab | 10601 | |
fa28dcb8 SL |
10602 | if ((func_id == BPF_FUNC_get_stack || |
10603 | func_id == BPF_FUNC_get_task_stack) && | |
10604 | !env->prog->has_callchain_buf) { | |
c195651e YS |
10605 | const char *err_str; |
10606 | ||
10607 | #ifdef CONFIG_PERF_EVENTS | |
10608 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
10609 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
10610 | #else | |
10611 | err = -ENOTSUPP; | |
10612 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
10613 | #endif | |
10614 | if (err) { | |
10615 | verbose(env, err_str, func_id_name(func_id), func_id); | |
10616 | return err; | |
10617 | } | |
10618 | ||
10619 | env->prog->has_callchain_buf = true; | |
10620 | } | |
10621 | ||
5d99cb2c SL |
10622 | if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack) |
10623 | env->prog->call_get_stack = true; | |
10624 | ||
9b99edca JO |
10625 | if (func_id == BPF_FUNC_get_func_ip) { |
10626 | if (check_get_func_ip(env)) | |
10627 | return -ENOTSUPP; | |
10628 | env->prog->call_get_func_ip = true; | |
10629 | } | |
10630 | ||
969bf05e AS |
10631 | if (changes_data) |
10632 | clear_all_pkt_pointers(env); | |
10633 | return 0; | |
10634 | } | |
10635 | ||
e6ac2450 MKL |
10636 | /* mark_btf_func_reg_size() is used when the reg size is determined by |
10637 | * the BTF func_proto's return value size and argument. | |
10638 | */ | |
10639 | static void mark_btf_func_reg_size(struct bpf_verifier_env *env, u32 regno, | |
10640 | size_t reg_size) | |
10641 | { | |
10642 | struct bpf_reg_state *reg = &cur_regs(env)[regno]; | |
10643 | ||
10644 | if (regno == BPF_REG_0) { | |
10645 | /* Function return value */ | |
10646 | reg->live |= REG_LIVE_WRITTEN; | |
10647 | reg->subreg_def = reg_size == sizeof(u64) ? | |
10648 | DEF_NOT_SUBREG : env->insn_idx + 1; | |
10649 | } else { | |
10650 | /* Function argument */ | |
10651 | if (reg_size == sizeof(u64)) { | |
10652 | mark_insn_zext(env, reg); | |
10653 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
10654 | } else { | |
10655 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ32); | |
10656 | } | |
10657 | } | |
10658 | } | |
10659 | ||
00b85860 KKD |
10660 | static bool is_kfunc_acquire(struct bpf_kfunc_call_arg_meta *meta) |
10661 | { | |
10662 | return meta->kfunc_flags & KF_ACQUIRE; | |
10663 | } | |
a5d82727 | 10664 | |
00b85860 KKD |
10665 | static bool is_kfunc_release(struct bpf_kfunc_call_arg_meta *meta) |
10666 | { | |
10667 | return meta->kfunc_flags & KF_RELEASE; | |
10668 | } | |
e6ac2450 | 10669 | |
00b85860 KKD |
10670 | static bool is_kfunc_trusted_args(struct bpf_kfunc_call_arg_meta *meta) |
10671 | { | |
6c831c46 | 10672 | return (meta->kfunc_flags & KF_TRUSTED_ARGS) || is_kfunc_release(meta); |
00b85860 | 10673 | } |
4dd48c6f | 10674 | |
00b85860 KKD |
10675 | static bool is_kfunc_sleepable(struct bpf_kfunc_call_arg_meta *meta) |
10676 | { | |
10677 | return meta->kfunc_flags & KF_SLEEPABLE; | |
10678 | } | |
5c073f26 | 10679 | |
00b85860 KKD |
10680 | static bool is_kfunc_destructive(struct bpf_kfunc_call_arg_meta *meta) |
10681 | { | |
10682 | return meta->kfunc_flags & KF_DESTRUCTIVE; | |
10683 | } | |
eb1f7f71 | 10684 | |
fca1aa75 YS |
10685 | static bool is_kfunc_rcu(struct bpf_kfunc_call_arg_meta *meta) |
10686 | { | |
10687 | return meta->kfunc_flags & KF_RCU; | |
10688 | } | |
10689 | ||
dfab99df CZ |
10690 | static bool is_kfunc_rcu_protected(struct bpf_kfunc_call_arg_meta *meta) |
10691 | { | |
10692 | return meta->kfunc_flags & KF_RCU_PROTECTED; | |
10693 | } | |
10694 | ||
a50388db KKD |
10695 | static bool __kfunc_param_match_suffix(const struct btf *btf, |
10696 | const struct btf_param *arg, | |
10697 | const char *suffix) | |
00b85860 | 10698 | { |
a50388db | 10699 | int suffix_len = strlen(suffix), len; |
00b85860 | 10700 | const char *param_name; |
e6ac2450 | 10701 | |
00b85860 KKD |
10702 | /* In the future, this can be ported to use BTF tagging */ |
10703 | param_name = btf_name_by_offset(btf, arg->name_off); | |
10704 | if (str_is_empty(param_name)) | |
10705 | return false; | |
10706 | len = strlen(param_name); | |
a50388db | 10707 | if (len < suffix_len) |
00b85860 | 10708 | return false; |
a50388db KKD |
10709 | param_name += len - suffix_len; |
10710 | return !strncmp(param_name, suffix, suffix_len); | |
10711 | } | |
5c073f26 | 10712 | |
a50388db KKD |
10713 | static bool is_kfunc_arg_mem_size(const struct btf *btf, |
10714 | const struct btf_param *arg, | |
10715 | const struct bpf_reg_state *reg) | |
10716 | { | |
10717 | const struct btf_type *t; | |
5c073f26 | 10718 | |
a50388db KKD |
10719 | t = btf_type_skip_modifiers(btf, arg->type, NULL); |
10720 | if (!btf_type_is_scalar(t) || reg->type != SCALAR_VALUE) | |
00b85860 | 10721 | return false; |
eb1f7f71 | 10722 | |
a50388db KKD |
10723 | return __kfunc_param_match_suffix(btf, arg, "__sz"); |
10724 | } | |
eb1f7f71 | 10725 | |
66e3a13e JK |
10726 | static bool is_kfunc_arg_const_mem_size(const struct btf *btf, |
10727 | const struct btf_param *arg, | |
10728 | const struct bpf_reg_state *reg) | |
10729 | { | |
10730 | const struct btf_type *t; | |
10731 | ||
10732 | t = btf_type_skip_modifiers(btf, arg->type, NULL); | |
10733 | if (!btf_type_is_scalar(t) || reg->type != SCALAR_VALUE) | |
10734 | return false; | |
10735 | ||
10736 | return __kfunc_param_match_suffix(btf, arg, "__szk"); | |
10737 | } | |
10738 | ||
3bda08b6 DR |
10739 | static bool is_kfunc_arg_optional(const struct btf *btf, const struct btf_param *arg) |
10740 | { | |
10741 | return __kfunc_param_match_suffix(btf, arg, "__opt"); | |
10742 | } | |
10743 | ||
a50388db KKD |
10744 | static bool is_kfunc_arg_constant(const struct btf *btf, const struct btf_param *arg) |
10745 | { | |
10746 | return __kfunc_param_match_suffix(btf, arg, "__k"); | |
00b85860 | 10747 | } |
eb1f7f71 | 10748 | |
958cf2e2 KKD |
10749 | static bool is_kfunc_arg_ignore(const struct btf *btf, const struct btf_param *arg) |
10750 | { | |
10751 | return __kfunc_param_match_suffix(btf, arg, "__ign"); | |
10752 | } | |
5c073f26 | 10753 | |
ac9f0605 KKD |
10754 | static bool is_kfunc_arg_alloc_obj(const struct btf *btf, const struct btf_param *arg) |
10755 | { | |
10756 | return __kfunc_param_match_suffix(btf, arg, "__alloc"); | |
10757 | } | |
e6ac2450 | 10758 | |
d96d937d JK |
10759 | static bool is_kfunc_arg_uninit(const struct btf *btf, const struct btf_param *arg) |
10760 | { | |
10761 | return __kfunc_param_match_suffix(btf, arg, "__uninit"); | |
10762 | } | |
10763 | ||
7c50b1cb DM |
10764 | static bool is_kfunc_arg_refcounted_kptr(const struct btf *btf, const struct btf_param *arg) |
10765 | { | |
10766 | return __kfunc_param_match_suffix(btf, arg, "__refcounted_kptr"); | |
10767 | } | |
10768 | ||
cb3ecf79 CZ |
10769 | static bool is_kfunc_arg_nullable(const struct btf *btf, const struct btf_param *arg) |
10770 | { | |
10771 | return __kfunc_param_match_suffix(btf, arg, "__nullable"); | |
10772 | } | |
10773 | ||
00b85860 KKD |
10774 | static bool is_kfunc_arg_scalar_with_name(const struct btf *btf, |
10775 | const struct btf_param *arg, | |
10776 | const char *name) | |
10777 | { | |
10778 | int len, target_len = strlen(name); | |
10779 | const char *param_name; | |
e6ac2450 | 10780 | |
00b85860 KKD |
10781 | param_name = btf_name_by_offset(btf, arg->name_off); |
10782 | if (str_is_empty(param_name)) | |
10783 | return false; | |
10784 | len = strlen(param_name); | |
10785 | if (len != target_len) | |
10786 | return false; | |
10787 | if (strcmp(param_name, name)) | |
10788 | return false; | |
e6ac2450 | 10789 | |
00b85860 | 10790 | return true; |
e6ac2450 MKL |
10791 | } |
10792 | ||
00b85860 KKD |
10793 | enum { |
10794 | KF_ARG_DYNPTR_ID, | |
8cab76ec KKD |
10795 | KF_ARG_LIST_HEAD_ID, |
10796 | KF_ARG_LIST_NODE_ID, | |
cd6791b4 DM |
10797 | KF_ARG_RB_ROOT_ID, |
10798 | KF_ARG_RB_NODE_ID, | |
00b85860 | 10799 | }; |
b03c9f9f | 10800 | |
00b85860 KKD |
10801 | BTF_ID_LIST(kf_arg_btf_ids) |
10802 | BTF_ID(struct, bpf_dynptr_kern) | |
8cab76ec KKD |
10803 | BTF_ID(struct, bpf_list_head) |
10804 | BTF_ID(struct, bpf_list_node) | |
bd1279ae DM |
10805 | BTF_ID(struct, bpf_rb_root) |
10806 | BTF_ID(struct, bpf_rb_node) | |
b03c9f9f | 10807 | |
8cab76ec KKD |
10808 | static bool __is_kfunc_ptr_arg_type(const struct btf *btf, |
10809 | const struct btf_param *arg, int type) | |
3f50f132 | 10810 | { |
00b85860 KKD |
10811 | const struct btf_type *t; |
10812 | u32 res_id; | |
3f50f132 | 10813 | |
00b85860 KKD |
10814 | t = btf_type_skip_modifiers(btf, arg->type, NULL); |
10815 | if (!t) | |
10816 | return false; | |
10817 | if (!btf_type_is_ptr(t)) | |
10818 | return false; | |
10819 | t = btf_type_skip_modifiers(btf, t->type, &res_id); | |
10820 | if (!t) | |
10821 | return false; | |
8cab76ec | 10822 | return btf_types_are_same(btf, res_id, btf_vmlinux, kf_arg_btf_ids[type]); |
3f50f132 JF |
10823 | } |
10824 | ||
8cab76ec | 10825 | static bool is_kfunc_arg_dynptr(const struct btf *btf, const struct btf_param *arg) |
b03c9f9f | 10826 | { |
8cab76ec | 10827 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_DYNPTR_ID); |
969bf05e AS |
10828 | } |
10829 | ||
8cab76ec | 10830 | static bool is_kfunc_arg_list_head(const struct btf *btf, const struct btf_param *arg) |
3f50f132 | 10831 | { |
8cab76ec | 10832 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_HEAD_ID); |
3f50f132 JF |
10833 | } |
10834 | ||
8cab76ec | 10835 | static bool is_kfunc_arg_list_node(const struct btf *btf, const struct btf_param *arg) |
bb7f0f98 | 10836 | { |
8cab76ec | 10837 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_NODE_ID); |
00b85860 KKD |
10838 | } |
10839 | ||
cd6791b4 DM |
10840 | static bool is_kfunc_arg_rbtree_root(const struct btf *btf, const struct btf_param *arg) |
10841 | { | |
10842 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_RB_ROOT_ID); | |
10843 | } | |
10844 | ||
10845 | static bool is_kfunc_arg_rbtree_node(const struct btf *btf, const struct btf_param *arg) | |
10846 | { | |
10847 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_RB_NODE_ID); | |
10848 | } | |
10849 | ||
5d92ddc3 DM |
10850 | static bool is_kfunc_arg_callback(struct bpf_verifier_env *env, const struct btf *btf, |
10851 | const struct btf_param *arg) | |
10852 | { | |
10853 | const struct btf_type *t; | |
10854 | ||
10855 | t = btf_type_resolve_func_ptr(btf, arg->type, NULL); | |
10856 | if (!t) | |
10857 | return false; | |
10858 | ||
10859 | return true; | |
10860 | } | |
10861 | ||
00b85860 KKD |
10862 | /* Returns true if struct is composed of scalars, 4 levels of nesting allowed */ |
10863 | static bool __btf_type_is_scalar_struct(struct bpf_verifier_env *env, | |
10864 | const struct btf *btf, | |
10865 | const struct btf_type *t, int rec) | |
10866 | { | |
10867 | const struct btf_type *member_type; | |
10868 | const struct btf_member *member; | |
10869 | u32 i; | |
10870 | ||
10871 | if (!btf_type_is_struct(t)) | |
10872 | return false; | |
10873 | ||
10874 | for_each_member(i, t, member) { | |
10875 | const struct btf_array *array; | |
10876 | ||
10877 | member_type = btf_type_skip_modifiers(btf, member->type, NULL); | |
10878 | if (btf_type_is_struct(member_type)) { | |
10879 | if (rec >= 3) { | |
10880 | verbose(env, "max struct nesting depth exceeded\n"); | |
10881 | return false; | |
10882 | } | |
10883 | if (!__btf_type_is_scalar_struct(env, btf, member_type, rec + 1)) | |
10884 | return false; | |
10885 | continue; | |
10886 | } | |
10887 | if (btf_type_is_array(member_type)) { | |
10888 | array = btf_array(member_type); | |
10889 | if (!array->nelems) | |
10890 | return false; | |
10891 | member_type = btf_type_skip_modifiers(btf, array->type, NULL); | |
10892 | if (!btf_type_is_scalar(member_type)) | |
10893 | return false; | |
10894 | continue; | |
10895 | } | |
10896 | if (!btf_type_is_scalar(member_type)) | |
10897 | return false; | |
10898 | } | |
10899 | return true; | |
10900 | } | |
10901 | ||
00b85860 KKD |
10902 | enum kfunc_ptr_arg_type { |
10903 | KF_ARG_PTR_TO_CTX, | |
7c50b1cb DM |
10904 | KF_ARG_PTR_TO_ALLOC_BTF_ID, /* Allocated object */ |
10905 | KF_ARG_PTR_TO_REFCOUNTED_KPTR, /* Refcounted local kptr */ | |
00b85860 | 10906 | KF_ARG_PTR_TO_DYNPTR, |
06accc87 | 10907 | KF_ARG_PTR_TO_ITER, |
8cab76ec KKD |
10908 | KF_ARG_PTR_TO_LIST_HEAD, |
10909 | KF_ARG_PTR_TO_LIST_NODE, | |
7c50b1cb | 10910 | KF_ARG_PTR_TO_BTF_ID, /* Also covers reg2btf_ids conversions */ |
00b85860 | 10911 | KF_ARG_PTR_TO_MEM, |
7c50b1cb | 10912 | KF_ARG_PTR_TO_MEM_SIZE, /* Size derived from next argument, skip it */ |
5d92ddc3 | 10913 | KF_ARG_PTR_TO_CALLBACK, |
cd6791b4 DM |
10914 | KF_ARG_PTR_TO_RB_ROOT, |
10915 | KF_ARG_PTR_TO_RB_NODE, | |
cb3ecf79 | 10916 | KF_ARG_PTR_TO_NULL, |
00b85860 KKD |
10917 | }; |
10918 | ||
ac9f0605 KKD |
10919 | enum special_kfunc_type { |
10920 | KF_bpf_obj_new_impl, | |
10921 | KF_bpf_obj_drop_impl, | |
7c50b1cb | 10922 | KF_bpf_refcount_acquire_impl, |
d2dcc67d DM |
10923 | KF_bpf_list_push_front_impl, |
10924 | KF_bpf_list_push_back_impl, | |
8cab76ec KKD |
10925 | KF_bpf_list_pop_front, |
10926 | KF_bpf_list_pop_back, | |
fd264ca0 | 10927 | KF_bpf_cast_to_kern_ctx, |
a35b9af4 | 10928 | KF_bpf_rdonly_cast, |
9bb00b28 YS |
10929 | KF_bpf_rcu_read_lock, |
10930 | KF_bpf_rcu_read_unlock, | |
bd1279ae | 10931 | KF_bpf_rbtree_remove, |
d2dcc67d | 10932 | KF_bpf_rbtree_add_impl, |
bd1279ae | 10933 | KF_bpf_rbtree_first, |
b5964b96 | 10934 | KF_bpf_dynptr_from_skb, |
05421aec | 10935 | KF_bpf_dynptr_from_xdp, |
66e3a13e JK |
10936 | KF_bpf_dynptr_slice, |
10937 | KF_bpf_dynptr_slice_rdwr, | |
361f129f | 10938 | KF_bpf_dynptr_clone, |
36d8bdf7 YS |
10939 | KF_bpf_percpu_obj_new_impl, |
10940 | KF_bpf_percpu_obj_drop_impl, | |
f18b03fa | 10941 | KF_bpf_throw, |
9c66dc94 | 10942 | KF_bpf_iter_css_task_new, |
ac9f0605 KKD |
10943 | }; |
10944 | ||
10945 | BTF_SET_START(special_kfunc_set) | |
10946 | BTF_ID(func, bpf_obj_new_impl) | |
10947 | BTF_ID(func, bpf_obj_drop_impl) | |
7c50b1cb | 10948 | BTF_ID(func, bpf_refcount_acquire_impl) |
d2dcc67d DM |
10949 | BTF_ID(func, bpf_list_push_front_impl) |
10950 | BTF_ID(func, bpf_list_push_back_impl) | |
8cab76ec KKD |
10951 | BTF_ID(func, bpf_list_pop_front) |
10952 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 10953 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 10954 | BTF_ID(func, bpf_rdonly_cast) |
bd1279ae | 10955 | BTF_ID(func, bpf_rbtree_remove) |
d2dcc67d | 10956 | BTF_ID(func, bpf_rbtree_add_impl) |
bd1279ae | 10957 | BTF_ID(func, bpf_rbtree_first) |
b5964b96 | 10958 | BTF_ID(func, bpf_dynptr_from_skb) |
05421aec | 10959 | BTF_ID(func, bpf_dynptr_from_xdp) |
66e3a13e JK |
10960 | BTF_ID(func, bpf_dynptr_slice) |
10961 | BTF_ID(func, bpf_dynptr_slice_rdwr) | |
361f129f | 10962 | BTF_ID(func, bpf_dynptr_clone) |
36d8bdf7 YS |
10963 | BTF_ID(func, bpf_percpu_obj_new_impl) |
10964 | BTF_ID(func, bpf_percpu_obj_drop_impl) | |
f18b03fa | 10965 | BTF_ID(func, bpf_throw) |
05670f81 | 10966 | #ifdef CONFIG_CGROUPS |
9c66dc94 | 10967 | BTF_ID(func, bpf_iter_css_task_new) |
05670f81 | 10968 | #endif |
ac9f0605 KKD |
10969 | BTF_SET_END(special_kfunc_set) |
10970 | ||
10971 | BTF_ID_LIST(special_kfunc_list) | |
10972 | BTF_ID(func, bpf_obj_new_impl) | |
10973 | BTF_ID(func, bpf_obj_drop_impl) | |
7c50b1cb | 10974 | BTF_ID(func, bpf_refcount_acquire_impl) |
d2dcc67d DM |
10975 | BTF_ID(func, bpf_list_push_front_impl) |
10976 | BTF_ID(func, bpf_list_push_back_impl) | |
8cab76ec KKD |
10977 | BTF_ID(func, bpf_list_pop_front) |
10978 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 10979 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 10980 | BTF_ID(func, bpf_rdonly_cast) |
9bb00b28 YS |
10981 | BTF_ID(func, bpf_rcu_read_lock) |
10982 | BTF_ID(func, bpf_rcu_read_unlock) | |
bd1279ae | 10983 | BTF_ID(func, bpf_rbtree_remove) |
d2dcc67d | 10984 | BTF_ID(func, bpf_rbtree_add_impl) |
bd1279ae | 10985 | BTF_ID(func, bpf_rbtree_first) |
b5964b96 | 10986 | BTF_ID(func, bpf_dynptr_from_skb) |
05421aec | 10987 | BTF_ID(func, bpf_dynptr_from_xdp) |
66e3a13e JK |
10988 | BTF_ID(func, bpf_dynptr_slice) |
10989 | BTF_ID(func, bpf_dynptr_slice_rdwr) | |
361f129f | 10990 | BTF_ID(func, bpf_dynptr_clone) |
36d8bdf7 YS |
10991 | BTF_ID(func, bpf_percpu_obj_new_impl) |
10992 | BTF_ID(func, bpf_percpu_obj_drop_impl) | |
f18b03fa | 10993 | BTF_ID(func, bpf_throw) |
05670f81 | 10994 | #ifdef CONFIG_CGROUPS |
9c66dc94 | 10995 | BTF_ID(func, bpf_iter_css_task_new) |
05670f81 MB |
10996 | #else |
10997 | BTF_ID_UNUSED | |
10998 | #endif | |
9bb00b28 | 10999 | |
7793fc3b DM |
11000 | static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta) |
11001 | { | |
11002 | if (meta->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl] && | |
11003 | meta->arg_owning_ref) { | |
11004 | return false; | |
11005 | } | |
11006 | ||
11007 | return meta->kfunc_flags & KF_RET_NULL; | |
11008 | } | |
11009 | ||
9bb00b28 YS |
11010 | static bool is_kfunc_bpf_rcu_read_lock(struct bpf_kfunc_call_arg_meta *meta) |
11011 | { | |
11012 | return meta->func_id == special_kfunc_list[KF_bpf_rcu_read_lock]; | |
11013 | } | |
11014 | ||
11015 | static bool is_kfunc_bpf_rcu_read_unlock(struct bpf_kfunc_call_arg_meta *meta) | |
11016 | { | |
11017 | return meta->func_id == special_kfunc_list[KF_bpf_rcu_read_unlock]; | |
11018 | } | |
ac9f0605 | 11019 | |
00b85860 KKD |
11020 | static enum kfunc_ptr_arg_type |
11021 | get_kfunc_ptr_arg_type(struct bpf_verifier_env *env, | |
11022 | struct bpf_kfunc_call_arg_meta *meta, | |
11023 | const struct btf_type *t, const struct btf_type *ref_t, | |
11024 | const char *ref_tname, const struct btf_param *args, | |
11025 | int argno, int nargs) | |
11026 | { | |
11027 | u32 regno = argno + 1; | |
11028 | struct bpf_reg_state *regs = cur_regs(env); | |
11029 | struct bpf_reg_state *reg = ®s[regno]; | |
11030 | bool arg_mem_size = false; | |
11031 | ||
fd264ca0 YS |
11032 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) |
11033 | return KF_ARG_PTR_TO_CTX; | |
11034 | ||
00b85860 KKD |
11035 | /* In this function, we verify the kfunc's BTF as per the argument type, |
11036 | * leaving the rest of the verification with respect to the register | |
11037 | * type to our caller. When a set of conditions hold in the BTF type of | |
11038 | * arguments, we resolve it to a known kfunc_ptr_arg_type. | |
11039 | */ | |
11040 | if (btf_get_prog_ctx_type(&env->log, meta->btf, t, resolve_prog_type(env->prog), argno)) | |
11041 | return KF_ARG_PTR_TO_CTX; | |
11042 | ||
ac9f0605 KKD |
11043 | if (is_kfunc_arg_alloc_obj(meta->btf, &args[argno])) |
11044 | return KF_ARG_PTR_TO_ALLOC_BTF_ID; | |
11045 | ||
7c50b1cb DM |
11046 | if (is_kfunc_arg_refcounted_kptr(meta->btf, &args[argno])) |
11047 | return KF_ARG_PTR_TO_REFCOUNTED_KPTR; | |
00b85860 KKD |
11048 | |
11049 | if (is_kfunc_arg_dynptr(meta->btf, &args[argno])) | |
11050 | return KF_ARG_PTR_TO_DYNPTR; | |
11051 | ||
06accc87 AN |
11052 | if (is_kfunc_arg_iter(meta, argno)) |
11053 | return KF_ARG_PTR_TO_ITER; | |
11054 | ||
8cab76ec KKD |
11055 | if (is_kfunc_arg_list_head(meta->btf, &args[argno])) |
11056 | return KF_ARG_PTR_TO_LIST_HEAD; | |
11057 | ||
11058 | if (is_kfunc_arg_list_node(meta->btf, &args[argno])) | |
11059 | return KF_ARG_PTR_TO_LIST_NODE; | |
11060 | ||
cd6791b4 DM |
11061 | if (is_kfunc_arg_rbtree_root(meta->btf, &args[argno])) |
11062 | return KF_ARG_PTR_TO_RB_ROOT; | |
11063 | ||
11064 | if (is_kfunc_arg_rbtree_node(meta->btf, &args[argno])) | |
11065 | return KF_ARG_PTR_TO_RB_NODE; | |
11066 | ||
00b85860 KKD |
11067 | if ((base_type(reg->type) == PTR_TO_BTF_ID || reg2btf_ids[base_type(reg->type)])) { |
11068 | if (!btf_type_is_struct(ref_t)) { | |
11069 | verbose(env, "kernel function %s args#%d pointer type %s %s is not supported\n", | |
11070 | meta->func_name, argno, btf_type_str(ref_t), ref_tname); | |
11071 | return -EINVAL; | |
11072 | } | |
11073 | return KF_ARG_PTR_TO_BTF_ID; | |
11074 | } | |
11075 | ||
5d92ddc3 DM |
11076 | if (is_kfunc_arg_callback(env, meta->btf, &args[argno])) |
11077 | return KF_ARG_PTR_TO_CALLBACK; | |
11078 | ||
cb3ecf79 CZ |
11079 | if (is_kfunc_arg_nullable(meta->btf, &args[argno]) && register_is_null(reg)) |
11080 | return KF_ARG_PTR_TO_NULL; | |
66e3a13e JK |
11081 | |
11082 | if (argno + 1 < nargs && | |
11083 | (is_kfunc_arg_mem_size(meta->btf, &args[argno + 1], ®s[regno + 1]) || | |
11084 | is_kfunc_arg_const_mem_size(meta->btf, &args[argno + 1], ®s[regno + 1]))) | |
00b85860 KKD |
11085 | arg_mem_size = true; |
11086 | ||
11087 | /* This is the catch all argument type of register types supported by | |
11088 | * check_helper_mem_access. However, we only allow when argument type is | |
11089 | * pointer to scalar, or struct composed (recursively) of scalars. When | |
11090 | * arg_mem_size is true, the pointer can be void *. | |
11091 | */ | |
11092 | if (!btf_type_is_scalar(ref_t) && !__btf_type_is_scalar_struct(env, meta->btf, ref_t, 0) && | |
11093 | (arg_mem_size ? !btf_type_is_void(ref_t) : 1)) { | |
11094 | verbose(env, "arg#%d pointer type %s %s must point to %sscalar, or struct with scalar\n", | |
11095 | argno, btf_type_str(ref_t), ref_tname, arg_mem_size ? "void, " : ""); | |
11096 | return -EINVAL; | |
11097 | } | |
11098 | return arg_mem_size ? KF_ARG_PTR_TO_MEM_SIZE : KF_ARG_PTR_TO_MEM; | |
11099 | } | |
11100 | ||
11101 | static int process_kf_arg_ptr_to_btf_id(struct bpf_verifier_env *env, | |
11102 | struct bpf_reg_state *reg, | |
11103 | const struct btf_type *ref_t, | |
11104 | const char *ref_tname, u32 ref_id, | |
11105 | struct bpf_kfunc_call_arg_meta *meta, | |
11106 | int argno) | |
11107 | { | |
11108 | const struct btf_type *reg_ref_t; | |
11109 | bool strict_type_match = false; | |
11110 | const struct btf *reg_btf; | |
11111 | const char *reg_ref_tname; | |
11112 | u32 reg_ref_id; | |
11113 | ||
3f00c523 | 11114 | if (base_type(reg->type) == PTR_TO_BTF_ID) { |
00b85860 KKD |
11115 | reg_btf = reg->btf; |
11116 | reg_ref_id = reg->btf_id; | |
11117 | } else { | |
11118 | reg_btf = btf_vmlinux; | |
11119 | reg_ref_id = *reg2btf_ids[base_type(reg->type)]; | |
11120 | } | |
11121 | ||
b613d335 DV |
11122 | /* Enforce strict type matching for calls to kfuncs that are acquiring |
11123 | * or releasing a reference, or are no-cast aliases. We do _not_ | |
11124 | * enforce strict matching for plain KF_TRUSTED_ARGS kfuncs by default, | |
11125 | * as we want to enable BPF programs to pass types that are bitwise | |
11126 | * equivalent without forcing them to explicitly cast with something | |
11127 | * like bpf_cast_to_kern_ctx(). | |
11128 | * | |
11129 | * For example, say we had a type like the following: | |
11130 | * | |
11131 | * struct bpf_cpumask { | |
11132 | * cpumask_t cpumask; | |
11133 | * refcount_t usage; | |
11134 | * }; | |
11135 | * | |
11136 | * Note that as specified in <linux/cpumask.h>, cpumask_t is typedef'ed | |
11137 | * to a struct cpumask, so it would be safe to pass a struct | |
11138 | * bpf_cpumask * to a kfunc expecting a struct cpumask *. | |
11139 | * | |
11140 | * The philosophy here is similar to how we allow scalars of different | |
11141 | * types to be passed to kfuncs as long as the size is the same. The | |
11142 | * only difference here is that we're simply allowing | |
11143 | * btf_struct_ids_match() to walk the struct at the 0th offset, and | |
11144 | * resolve types. | |
11145 | */ | |
11146 | if (is_kfunc_acquire(meta) || | |
11147 | (is_kfunc_release(meta) && reg->ref_obj_id) || | |
11148 | btf_type_ids_nocast_alias(&env->log, reg_btf, reg_ref_id, meta->btf, ref_id)) | |
00b85860 KKD |
11149 | strict_type_match = true; |
11150 | ||
b613d335 DV |
11151 | WARN_ON_ONCE(is_kfunc_trusted_args(meta) && reg->off); |
11152 | ||
00b85860 KKD |
11153 | reg_ref_t = btf_type_skip_modifiers(reg_btf, reg_ref_id, ®_ref_id); |
11154 | reg_ref_tname = btf_name_by_offset(reg_btf, reg_ref_t->name_off); | |
11155 | if (!btf_struct_ids_match(&env->log, reg_btf, reg_ref_id, reg->off, meta->btf, ref_id, strict_type_match)) { | |
11156 | verbose(env, "kernel function %s args#%d expected pointer to %s %s but R%d has a pointer to %s %s\n", | |
11157 | meta->func_name, argno, btf_type_str(ref_t), ref_tname, argno + 1, | |
11158 | btf_type_str(reg_ref_t), reg_ref_tname); | |
11159 | return -EINVAL; | |
11160 | } | |
11161 | return 0; | |
11162 | } | |
11163 | ||
6a3cd331 | 11164 | static int ref_set_non_owning(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
534e86bc | 11165 | { |
6a3cd331 | 11166 | struct bpf_verifier_state *state = env->cur_state; |
0816b8c6 | 11167 | struct btf_record *rec = reg_btf_record(reg); |
6a3cd331 DM |
11168 | |
11169 | if (!state->active_lock.ptr) { | |
11170 | verbose(env, "verifier internal error: ref_set_non_owning w/o active lock\n"); | |
11171 | return -EFAULT; | |
11172 | } | |
11173 | ||
11174 | if (type_flag(reg->type) & NON_OWN_REF) { | |
11175 | verbose(env, "verifier internal error: NON_OWN_REF already set\n"); | |
11176 | return -EFAULT; | |
11177 | } | |
11178 | ||
11179 | reg->type |= NON_OWN_REF; | |
0816b8c6 DM |
11180 | if (rec->refcount_off >= 0) |
11181 | reg->type |= MEM_RCU; | |
11182 | ||
6a3cd331 DM |
11183 | return 0; |
11184 | } | |
11185 | ||
11186 | static int ref_convert_owning_non_owning(struct bpf_verifier_env *env, u32 ref_obj_id) | |
11187 | { | |
11188 | struct bpf_func_state *state, *unused; | |
534e86bc KKD |
11189 | struct bpf_reg_state *reg; |
11190 | int i; | |
11191 | ||
6a3cd331 DM |
11192 | state = cur_func(env); |
11193 | ||
534e86bc | 11194 | if (!ref_obj_id) { |
6a3cd331 DM |
11195 | verbose(env, "verifier internal error: ref_obj_id is zero for " |
11196 | "owning -> non-owning conversion\n"); | |
534e86bc KKD |
11197 | return -EFAULT; |
11198 | } | |
6a3cd331 | 11199 | |
534e86bc | 11200 | for (i = 0; i < state->acquired_refs; i++) { |
6a3cd331 DM |
11201 | if (state->refs[i].id != ref_obj_id) |
11202 | continue; | |
11203 | ||
11204 | /* Clear ref_obj_id here so release_reference doesn't clobber | |
11205 | * the whole reg | |
11206 | */ | |
11207 | bpf_for_each_reg_in_vstate(env->cur_state, unused, reg, ({ | |
11208 | if (reg->ref_obj_id == ref_obj_id) { | |
11209 | reg->ref_obj_id = 0; | |
11210 | ref_set_non_owning(env, reg); | |
534e86bc | 11211 | } |
6a3cd331 DM |
11212 | })); |
11213 | return 0; | |
534e86bc | 11214 | } |
6a3cd331 | 11215 | |
534e86bc KKD |
11216 | verbose(env, "verifier internal error: ref state missing for ref_obj_id\n"); |
11217 | return -EFAULT; | |
11218 | } | |
11219 | ||
8cab76ec KKD |
11220 | /* Implementation details: |
11221 | * | |
11222 | * Each register points to some region of memory, which we define as an | |
11223 | * allocation. Each allocation may embed a bpf_spin_lock which protects any | |
11224 | * special BPF objects (bpf_list_head, bpf_rb_root, etc.) part of the same | |
11225 | * allocation. The lock and the data it protects are colocated in the same | |
11226 | * memory region. | |
11227 | * | |
11228 | * Hence, everytime a register holds a pointer value pointing to such | |
11229 | * allocation, the verifier preserves a unique reg->id for it. | |
11230 | * | |
11231 | * The verifier remembers the lock 'ptr' and the lock 'id' whenever | |
11232 | * bpf_spin_lock is called. | |
11233 | * | |
11234 | * To enable this, lock state in the verifier captures two values: | |
11235 | * active_lock.ptr = Register's type specific pointer | |
11236 | * active_lock.id = A unique ID for each register pointer value | |
11237 | * | |
11238 | * Currently, PTR_TO_MAP_VALUE and PTR_TO_BTF_ID | MEM_ALLOC are the two | |
11239 | * supported register types. | |
11240 | * | |
11241 | * The active_lock.ptr in case of map values is the reg->map_ptr, and in case of | |
11242 | * allocated objects is the reg->btf pointer. | |
11243 | * | |
11244 | * The active_lock.id is non-unique for maps supporting direct_value_addr, as we | |
11245 | * can establish the provenance of the map value statically for each distinct | |
11246 | * lookup into such maps. They always contain a single map value hence unique | |
11247 | * IDs for each pseudo load pessimizes the algorithm and rejects valid programs. | |
11248 | * | |
11249 | * So, in case of global variables, they use array maps with max_entries = 1, | |
11250 | * hence their active_lock.ptr becomes map_ptr and id = 0 (since they all point | |
11251 | * into the same map value as max_entries is 1, as described above). | |
11252 | * | |
11253 | * In case of inner map lookups, the inner map pointer has same map_ptr as the | |
11254 | * outer map pointer (in verifier context), but each lookup into an inner map | |
11255 | * assigns a fresh reg->id to the lookup, so while lookups into distinct inner | |
11256 | * maps from the same outer map share the same map_ptr as active_lock.ptr, they | |
11257 | * will get different reg->id assigned to each lookup, hence different | |
11258 | * active_lock.id. | |
11259 | * | |
11260 | * In case of allocated objects, active_lock.ptr is the reg->btf, and the | |
11261 | * reg->id is a unique ID preserved after the NULL pointer check on the pointer | |
11262 | * returned from bpf_obj_new. Each allocation receives a new reg->id. | |
11263 | */ | |
11264 | static int check_reg_allocation_locked(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
11265 | { | |
11266 | void *ptr; | |
11267 | u32 id; | |
11268 | ||
11269 | switch ((int)reg->type) { | |
11270 | case PTR_TO_MAP_VALUE: | |
11271 | ptr = reg->map_ptr; | |
11272 | break; | |
11273 | case PTR_TO_BTF_ID | MEM_ALLOC: | |
11274 | ptr = reg->btf; | |
11275 | break; | |
11276 | default: | |
11277 | verbose(env, "verifier internal error: unknown reg type for lock check\n"); | |
11278 | return -EFAULT; | |
11279 | } | |
11280 | id = reg->id; | |
11281 | ||
11282 | if (!env->cur_state->active_lock.ptr) | |
11283 | return -EINVAL; | |
11284 | if (env->cur_state->active_lock.ptr != ptr || | |
11285 | env->cur_state->active_lock.id != id) { | |
11286 | verbose(env, "held lock and object are not in the same allocation\n"); | |
11287 | return -EINVAL; | |
11288 | } | |
11289 | return 0; | |
11290 | } | |
11291 | ||
11292 | static bool is_bpf_list_api_kfunc(u32 btf_id) | |
11293 | { | |
d2dcc67d DM |
11294 | return btf_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
11295 | btf_id == special_kfunc_list[KF_bpf_list_push_back_impl] || | |
8cab76ec KKD |
11296 | btf_id == special_kfunc_list[KF_bpf_list_pop_front] || |
11297 | btf_id == special_kfunc_list[KF_bpf_list_pop_back]; | |
11298 | } | |
11299 | ||
cd6791b4 DM |
11300 | static bool is_bpf_rbtree_api_kfunc(u32 btf_id) |
11301 | { | |
d2dcc67d | 11302 | return btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl] || |
cd6791b4 DM |
11303 | btf_id == special_kfunc_list[KF_bpf_rbtree_remove] || |
11304 | btf_id == special_kfunc_list[KF_bpf_rbtree_first]; | |
11305 | } | |
11306 | ||
11307 | static bool is_bpf_graph_api_kfunc(u32 btf_id) | |
11308 | { | |
7c50b1cb DM |
11309 | return is_bpf_list_api_kfunc(btf_id) || is_bpf_rbtree_api_kfunc(btf_id) || |
11310 | btf_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]; | |
cd6791b4 DM |
11311 | } |
11312 | ||
ab5cfac1 | 11313 | static bool is_sync_callback_calling_kfunc(u32 btf_id) |
5d92ddc3 | 11314 | { |
d2dcc67d | 11315 | return btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl]; |
5d92ddc3 DM |
11316 | } |
11317 | ||
f18b03fa KKD |
11318 | static bool is_bpf_throw_kfunc(struct bpf_insn *insn) |
11319 | { | |
11320 | return bpf_pseudo_kfunc_call(insn) && insn->off == 0 && | |
11321 | insn->imm == special_kfunc_list[KF_bpf_throw]; | |
11322 | } | |
11323 | ||
5d92ddc3 DM |
11324 | static bool is_rbtree_lock_required_kfunc(u32 btf_id) |
11325 | { | |
11326 | return is_bpf_rbtree_api_kfunc(btf_id); | |
11327 | } | |
11328 | ||
cd6791b4 DM |
11329 | static bool check_kfunc_is_graph_root_api(struct bpf_verifier_env *env, |
11330 | enum btf_field_type head_field_type, | |
11331 | u32 kfunc_btf_id) | |
11332 | { | |
11333 | bool ret; | |
11334 | ||
11335 | switch (head_field_type) { | |
11336 | case BPF_LIST_HEAD: | |
11337 | ret = is_bpf_list_api_kfunc(kfunc_btf_id); | |
11338 | break; | |
11339 | case BPF_RB_ROOT: | |
11340 | ret = is_bpf_rbtree_api_kfunc(kfunc_btf_id); | |
11341 | break; | |
11342 | default: | |
11343 | verbose(env, "verifier internal error: unexpected graph root argument type %s\n", | |
11344 | btf_field_type_name(head_field_type)); | |
11345 | return false; | |
11346 | } | |
11347 | ||
11348 | if (!ret) | |
11349 | verbose(env, "verifier internal error: %s head arg for unknown kfunc\n", | |
11350 | btf_field_type_name(head_field_type)); | |
11351 | return ret; | |
11352 | } | |
11353 | ||
11354 | static bool check_kfunc_is_graph_node_api(struct bpf_verifier_env *env, | |
11355 | enum btf_field_type node_field_type, | |
11356 | u32 kfunc_btf_id) | |
8cab76ec | 11357 | { |
cd6791b4 DM |
11358 | bool ret; |
11359 | ||
11360 | switch (node_field_type) { | |
11361 | case BPF_LIST_NODE: | |
d2dcc67d DM |
11362 | ret = (kfunc_btf_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
11363 | kfunc_btf_id == special_kfunc_list[KF_bpf_list_push_back_impl]); | |
cd6791b4 DM |
11364 | break; |
11365 | case BPF_RB_NODE: | |
11366 | ret = (kfunc_btf_id == special_kfunc_list[KF_bpf_rbtree_remove] || | |
d2dcc67d | 11367 | kfunc_btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl]); |
cd6791b4 DM |
11368 | break; |
11369 | default: | |
11370 | verbose(env, "verifier internal error: unexpected graph node argument type %s\n", | |
11371 | btf_field_type_name(node_field_type)); | |
11372 | return false; | |
11373 | } | |
11374 | ||
11375 | if (!ret) | |
11376 | verbose(env, "verifier internal error: %s node arg for unknown kfunc\n", | |
11377 | btf_field_type_name(node_field_type)); | |
11378 | return ret; | |
11379 | } | |
11380 | ||
11381 | static int | |
11382 | __process_kf_arg_ptr_to_graph_root(struct bpf_verifier_env *env, | |
11383 | struct bpf_reg_state *reg, u32 regno, | |
11384 | struct bpf_kfunc_call_arg_meta *meta, | |
11385 | enum btf_field_type head_field_type, | |
11386 | struct btf_field **head_field) | |
11387 | { | |
11388 | const char *head_type_name; | |
8cab76ec KKD |
11389 | struct btf_field *field; |
11390 | struct btf_record *rec; | |
cd6791b4 | 11391 | u32 head_off; |
8cab76ec | 11392 | |
cd6791b4 DM |
11393 | if (meta->btf != btf_vmlinux) { |
11394 | verbose(env, "verifier internal error: unexpected btf mismatch in kfunc call\n"); | |
8cab76ec KKD |
11395 | return -EFAULT; |
11396 | } | |
11397 | ||
cd6791b4 DM |
11398 | if (!check_kfunc_is_graph_root_api(env, head_field_type, meta->func_id)) |
11399 | return -EFAULT; | |
11400 | ||
11401 | head_type_name = btf_field_type_name(head_field_type); | |
8cab76ec KKD |
11402 | if (!tnum_is_const(reg->var_off)) { |
11403 | verbose(env, | |
cd6791b4 DM |
11404 | "R%d doesn't have constant offset. %s has to be at the constant offset\n", |
11405 | regno, head_type_name); | |
8cab76ec KKD |
11406 | return -EINVAL; |
11407 | } | |
11408 | ||
11409 | rec = reg_btf_record(reg); | |
cd6791b4 DM |
11410 | head_off = reg->off + reg->var_off.value; |
11411 | field = btf_record_find(rec, head_off, head_field_type); | |
8cab76ec | 11412 | if (!field) { |
cd6791b4 | 11413 | verbose(env, "%s not found at offset=%u\n", head_type_name, head_off); |
8cab76ec KKD |
11414 | return -EINVAL; |
11415 | } | |
11416 | ||
11417 | /* All functions require bpf_list_head to be protected using a bpf_spin_lock */ | |
11418 | if (check_reg_allocation_locked(env, reg)) { | |
cd6791b4 DM |
11419 | verbose(env, "bpf_spin_lock at off=%d must be held for %s\n", |
11420 | rec->spin_lock_off, head_type_name); | |
8cab76ec KKD |
11421 | return -EINVAL; |
11422 | } | |
11423 | ||
cd6791b4 DM |
11424 | if (*head_field) { |
11425 | verbose(env, "verifier internal error: repeating %s arg\n", head_type_name); | |
8cab76ec KKD |
11426 | return -EFAULT; |
11427 | } | |
cd6791b4 | 11428 | *head_field = field; |
8cab76ec KKD |
11429 | return 0; |
11430 | } | |
11431 | ||
cd6791b4 | 11432 | static int process_kf_arg_ptr_to_list_head(struct bpf_verifier_env *env, |
8cab76ec KKD |
11433 | struct bpf_reg_state *reg, u32 regno, |
11434 | struct bpf_kfunc_call_arg_meta *meta) | |
11435 | { | |
cd6791b4 DM |
11436 | return __process_kf_arg_ptr_to_graph_root(env, reg, regno, meta, BPF_LIST_HEAD, |
11437 | &meta->arg_list_head.field); | |
11438 | } | |
11439 | ||
11440 | static int process_kf_arg_ptr_to_rbtree_root(struct bpf_verifier_env *env, | |
11441 | struct bpf_reg_state *reg, u32 regno, | |
11442 | struct bpf_kfunc_call_arg_meta *meta) | |
11443 | { | |
11444 | return __process_kf_arg_ptr_to_graph_root(env, reg, regno, meta, BPF_RB_ROOT, | |
11445 | &meta->arg_rbtree_root.field); | |
11446 | } | |
11447 | ||
11448 | static int | |
11449 | __process_kf_arg_ptr_to_graph_node(struct bpf_verifier_env *env, | |
11450 | struct bpf_reg_state *reg, u32 regno, | |
11451 | struct bpf_kfunc_call_arg_meta *meta, | |
11452 | enum btf_field_type head_field_type, | |
11453 | enum btf_field_type node_field_type, | |
11454 | struct btf_field **node_field) | |
11455 | { | |
11456 | const char *node_type_name; | |
8cab76ec KKD |
11457 | const struct btf_type *et, *t; |
11458 | struct btf_field *field; | |
cd6791b4 | 11459 | u32 node_off; |
8cab76ec | 11460 | |
cd6791b4 DM |
11461 | if (meta->btf != btf_vmlinux) { |
11462 | verbose(env, "verifier internal error: unexpected btf mismatch in kfunc call\n"); | |
8cab76ec KKD |
11463 | return -EFAULT; |
11464 | } | |
11465 | ||
cd6791b4 DM |
11466 | if (!check_kfunc_is_graph_node_api(env, node_field_type, meta->func_id)) |
11467 | return -EFAULT; | |
11468 | ||
11469 | node_type_name = btf_field_type_name(node_field_type); | |
8cab76ec KKD |
11470 | if (!tnum_is_const(reg->var_off)) { |
11471 | verbose(env, | |
cd6791b4 DM |
11472 | "R%d doesn't have constant offset. %s has to be at the constant offset\n", |
11473 | regno, node_type_name); | |
8cab76ec KKD |
11474 | return -EINVAL; |
11475 | } | |
11476 | ||
cd6791b4 DM |
11477 | node_off = reg->off + reg->var_off.value; |
11478 | field = reg_find_field_offset(reg, node_off, node_field_type); | |
11479 | if (!field || field->offset != node_off) { | |
11480 | verbose(env, "%s not found at offset=%u\n", node_type_name, node_off); | |
8cab76ec KKD |
11481 | return -EINVAL; |
11482 | } | |
11483 | ||
cd6791b4 | 11484 | field = *node_field; |
8cab76ec | 11485 | |
30465003 | 11486 | et = btf_type_by_id(field->graph_root.btf, field->graph_root.value_btf_id); |
8cab76ec | 11487 | t = btf_type_by_id(reg->btf, reg->btf_id); |
30465003 DM |
11488 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, 0, field->graph_root.btf, |
11489 | field->graph_root.value_btf_id, true)) { | |
cd6791b4 | 11490 | verbose(env, "operation on %s expects arg#1 %s at offset=%d " |
8cab76ec | 11491 | "in struct %s, but arg is at offset=%d in struct %s\n", |
cd6791b4 DM |
11492 | btf_field_type_name(head_field_type), |
11493 | btf_field_type_name(node_field_type), | |
30465003 DM |
11494 | field->graph_root.node_offset, |
11495 | btf_name_by_offset(field->graph_root.btf, et->name_off), | |
cd6791b4 | 11496 | node_off, btf_name_by_offset(reg->btf, t->name_off)); |
8cab76ec KKD |
11497 | return -EINVAL; |
11498 | } | |
2140a6e3 DM |
11499 | meta->arg_btf = reg->btf; |
11500 | meta->arg_btf_id = reg->btf_id; | |
8cab76ec | 11501 | |
cd6791b4 DM |
11502 | if (node_off != field->graph_root.node_offset) { |
11503 | verbose(env, "arg#1 offset=%d, but expected %s at offset=%d in struct %s\n", | |
11504 | node_off, btf_field_type_name(node_field_type), | |
11505 | field->graph_root.node_offset, | |
30465003 | 11506 | btf_name_by_offset(field->graph_root.btf, et->name_off)); |
8cab76ec KKD |
11507 | return -EINVAL; |
11508 | } | |
6a3cd331 DM |
11509 | |
11510 | return 0; | |
8cab76ec KKD |
11511 | } |
11512 | ||
cd6791b4 DM |
11513 | static int process_kf_arg_ptr_to_list_node(struct bpf_verifier_env *env, |
11514 | struct bpf_reg_state *reg, u32 regno, | |
11515 | struct bpf_kfunc_call_arg_meta *meta) | |
11516 | { | |
11517 | return __process_kf_arg_ptr_to_graph_node(env, reg, regno, meta, | |
11518 | BPF_LIST_HEAD, BPF_LIST_NODE, | |
11519 | &meta->arg_list_head.field); | |
11520 | } | |
11521 | ||
11522 | static int process_kf_arg_ptr_to_rbtree_node(struct bpf_verifier_env *env, | |
11523 | struct bpf_reg_state *reg, u32 regno, | |
11524 | struct bpf_kfunc_call_arg_meta *meta) | |
11525 | { | |
11526 | return __process_kf_arg_ptr_to_graph_node(env, reg, regno, meta, | |
11527 | BPF_RB_ROOT, BPF_RB_NODE, | |
11528 | &meta->arg_rbtree_root.field); | |
11529 | } | |
11530 | ||
3091b667 CZ |
11531 | /* |
11532 | * css_task iter allowlist is needed to avoid dead locking on css_set_lock. | |
11533 | * LSM hooks and iters (both sleepable and non-sleepable) are safe. | |
11534 | * Any sleepable progs are also safe since bpf_check_attach_target() enforce | |
11535 | * them can only be attached to some specific hook points. | |
11536 | */ | |
9c66dc94 CZ |
11537 | static bool check_css_task_iter_allowlist(struct bpf_verifier_env *env) |
11538 | { | |
11539 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); | |
11540 | ||
11541 | switch (prog_type) { | |
11542 | case BPF_PROG_TYPE_LSM: | |
11543 | return true; | |
3091b667 CZ |
11544 | case BPF_PROG_TYPE_TRACING: |
11545 | if (env->prog->expected_attach_type == BPF_TRACE_ITER) | |
11546 | return true; | |
11547 | fallthrough; | |
9c66dc94 | 11548 | default: |
3091b667 | 11549 | return env->prog->aux->sleepable; |
9c66dc94 CZ |
11550 | } |
11551 | } | |
11552 | ||
1d18feb2 JK |
11553 | static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_arg_meta *meta, |
11554 | int insn_idx) | |
00b85860 KKD |
11555 | { |
11556 | const char *func_name = meta->func_name, *ref_tname; | |
11557 | const struct btf *btf = meta->btf; | |
11558 | const struct btf_param *args; | |
7c50b1cb | 11559 | struct btf_record *rec; |
00b85860 KKD |
11560 | u32 i, nargs; |
11561 | int ret; | |
11562 | ||
11563 | args = (const struct btf_param *)(meta->func_proto + 1); | |
11564 | nargs = btf_type_vlen(meta->func_proto); | |
11565 | if (nargs > MAX_BPF_FUNC_REG_ARGS) { | |
11566 | verbose(env, "Function %s has %d > %d args\n", func_name, nargs, | |
11567 | MAX_BPF_FUNC_REG_ARGS); | |
11568 | return -EINVAL; | |
11569 | } | |
11570 | ||
11571 | /* Check that BTF function arguments match actual types that the | |
11572 | * verifier sees. | |
11573 | */ | |
11574 | for (i = 0; i < nargs; i++) { | |
11575 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[i + 1]; | |
11576 | const struct btf_type *t, *ref_t, *resolve_ret; | |
11577 | enum bpf_arg_type arg_type = ARG_DONTCARE; | |
11578 | u32 regno = i + 1, ref_id, type_size; | |
11579 | bool is_ret_buf_sz = false; | |
11580 | int kf_arg_type; | |
11581 | ||
11582 | t = btf_type_skip_modifiers(btf, args[i].type, NULL); | |
958cf2e2 KKD |
11583 | |
11584 | if (is_kfunc_arg_ignore(btf, &args[i])) | |
11585 | continue; | |
11586 | ||
00b85860 KKD |
11587 | if (btf_type_is_scalar(t)) { |
11588 | if (reg->type != SCALAR_VALUE) { | |
11589 | verbose(env, "R%d is not a scalar\n", regno); | |
11590 | return -EINVAL; | |
11591 | } | |
a50388db KKD |
11592 | |
11593 | if (is_kfunc_arg_constant(meta->btf, &args[i])) { | |
11594 | if (meta->arg_constant.found) { | |
11595 | verbose(env, "verifier internal error: only one constant argument permitted\n"); | |
11596 | return -EFAULT; | |
11597 | } | |
11598 | if (!tnum_is_const(reg->var_off)) { | |
11599 | verbose(env, "R%d must be a known constant\n", regno); | |
11600 | return -EINVAL; | |
11601 | } | |
11602 | ret = mark_chain_precision(env, regno); | |
11603 | if (ret < 0) | |
11604 | return ret; | |
11605 | meta->arg_constant.found = true; | |
11606 | meta->arg_constant.value = reg->var_off.value; | |
11607 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdonly_buf_size")) { | |
00b85860 KKD |
11608 | meta->r0_rdonly = true; |
11609 | is_ret_buf_sz = true; | |
11610 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdwr_buf_size")) { | |
11611 | is_ret_buf_sz = true; | |
11612 | } | |
11613 | ||
11614 | if (is_ret_buf_sz) { | |
11615 | if (meta->r0_size) { | |
11616 | verbose(env, "2 or more rdonly/rdwr_buf_size parameters for kfunc"); | |
11617 | return -EINVAL; | |
11618 | } | |
11619 | ||
11620 | if (!tnum_is_const(reg->var_off)) { | |
11621 | verbose(env, "R%d is not a const\n", regno); | |
11622 | return -EINVAL; | |
11623 | } | |
11624 | ||
11625 | meta->r0_size = reg->var_off.value; | |
11626 | ret = mark_chain_precision(env, regno); | |
11627 | if (ret) | |
11628 | return ret; | |
11629 | } | |
11630 | continue; | |
11631 | } | |
11632 | ||
11633 | if (!btf_type_is_ptr(t)) { | |
11634 | verbose(env, "Unrecognized arg#%d type %s\n", i, btf_type_str(t)); | |
11635 | return -EINVAL; | |
11636 | } | |
11637 | ||
20c09d92 | 11638 | if ((is_kfunc_trusted_args(meta) || is_kfunc_rcu(meta)) && |
cb3ecf79 CZ |
11639 | (register_is_null(reg) || type_may_be_null(reg->type)) && |
11640 | !is_kfunc_arg_nullable(meta->btf, &args[i])) { | |
caf713c3 DV |
11641 | verbose(env, "Possibly NULL pointer passed to trusted arg%d\n", i); |
11642 | return -EACCES; | |
11643 | } | |
11644 | ||
00b85860 KKD |
11645 | if (reg->ref_obj_id) { |
11646 | if (is_kfunc_release(meta) && meta->ref_obj_id) { | |
11647 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
11648 | regno, reg->ref_obj_id, | |
11649 | meta->ref_obj_id); | |
11650 | return -EFAULT; | |
11651 | } | |
11652 | meta->ref_obj_id = reg->ref_obj_id; | |
11653 | if (is_kfunc_release(meta)) | |
11654 | meta->release_regno = regno; | |
11655 | } | |
11656 | ||
11657 | ref_t = btf_type_skip_modifiers(btf, t->type, &ref_id); | |
11658 | ref_tname = btf_name_by_offset(btf, ref_t->name_off); | |
11659 | ||
11660 | kf_arg_type = get_kfunc_ptr_arg_type(env, meta, t, ref_t, ref_tname, args, i, nargs); | |
11661 | if (kf_arg_type < 0) | |
11662 | return kf_arg_type; | |
11663 | ||
11664 | switch (kf_arg_type) { | |
cb3ecf79 CZ |
11665 | case KF_ARG_PTR_TO_NULL: |
11666 | continue; | |
ac9f0605 | 11667 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
00b85860 | 11668 | case KF_ARG_PTR_TO_BTF_ID: |
fca1aa75 | 11669 | if (!is_kfunc_trusted_args(meta) && !is_kfunc_rcu(meta)) |
00b85860 | 11670 | break; |
3f00c523 DV |
11671 | |
11672 | if (!is_trusted_reg(reg)) { | |
fca1aa75 YS |
11673 | if (!is_kfunc_rcu(meta)) { |
11674 | verbose(env, "R%d must be referenced or trusted\n", regno); | |
11675 | return -EINVAL; | |
11676 | } | |
11677 | if (!is_rcu_reg(reg)) { | |
11678 | verbose(env, "R%d must be a rcu pointer\n", regno); | |
11679 | return -EINVAL; | |
11680 | } | |
00b85860 | 11681 | } |
fca1aa75 | 11682 | |
00b85860 KKD |
11683 | fallthrough; |
11684 | case KF_ARG_PTR_TO_CTX: | |
11685 | /* Trusted arguments have the same offset checks as release arguments */ | |
11686 | arg_type |= OBJ_RELEASE; | |
11687 | break; | |
00b85860 | 11688 | case KF_ARG_PTR_TO_DYNPTR: |
06accc87 | 11689 | case KF_ARG_PTR_TO_ITER: |
8cab76ec KKD |
11690 | case KF_ARG_PTR_TO_LIST_HEAD: |
11691 | case KF_ARG_PTR_TO_LIST_NODE: | |
cd6791b4 DM |
11692 | case KF_ARG_PTR_TO_RB_ROOT: |
11693 | case KF_ARG_PTR_TO_RB_NODE: | |
00b85860 KKD |
11694 | case KF_ARG_PTR_TO_MEM: |
11695 | case KF_ARG_PTR_TO_MEM_SIZE: | |
5d92ddc3 | 11696 | case KF_ARG_PTR_TO_CALLBACK: |
7c50b1cb | 11697 | case KF_ARG_PTR_TO_REFCOUNTED_KPTR: |
00b85860 KKD |
11698 | /* Trusted by default */ |
11699 | break; | |
11700 | default: | |
11701 | WARN_ON_ONCE(1); | |
11702 | return -EFAULT; | |
11703 | } | |
11704 | ||
11705 | if (is_kfunc_release(meta) && reg->ref_obj_id) | |
11706 | arg_type |= OBJ_RELEASE; | |
11707 | ret = check_func_arg_reg_off(env, reg, regno, arg_type); | |
11708 | if (ret < 0) | |
11709 | return ret; | |
11710 | ||
11711 | switch (kf_arg_type) { | |
11712 | case KF_ARG_PTR_TO_CTX: | |
11713 | if (reg->type != PTR_TO_CTX) { | |
11714 | verbose(env, "arg#%d expected pointer to ctx, but got %s\n", i, btf_type_str(t)); | |
11715 | return -EINVAL; | |
11716 | } | |
fd264ca0 YS |
11717 | |
11718 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { | |
11719 | ret = get_kern_ctx_btf_id(&env->log, resolve_prog_type(env->prog)); | |
11720 | if (ret < 0) | |
11721 | return -EINVAL; | |
11722 | meta->ret_btf_id = ret; | |
11723 | } | |
00b85860 | 11724 | break; |
ac9f0605 | 11725 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
36d8bdf7 YS |
11726 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC)) { |
11727 | if (meta->func_id != special_kfunc_list[KF_bpf_obj_drop_impl]) { | |
11728 | verbose(env, "arg#%d expected for bpf_obj_drop_impl()\n", i); | |
11729 | return -EINVAL; | |
11730 | } | |
11731 | } else if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC | MEM_PERCPU)) { | |
11732 | if (meta->func_id != special_kfunc_list[KF_bpf_percpu_obj_drop_impl]) { | |
11733 | verbose(env, "arg#%d expected for bpf_percpu_obj_drop_impl()\n", i); | |
11734 | return -EINVAL; | |
11735 | } | |
11736 | } else { | |
ac9f0605 KKD |
11737 | verbose(env, "arg#%d expected pointer to allocated object\n", i); |
11738 | return -EINVAL; | |
11739 | } | |
11740 | if (!reg->ref_obj_id) { | |
11741 | verbose(env, "allocated object must be referenced\n"); | |
11742 | return -EINVAL; | |
11743 | } | |
36d8bdf7 | 11744 | if (meta->btf == btf_vmlinux) { |
4d585f48 DM |
11745 | meta->arg_btf = reg->btf; |
11746 | meta->arg_btf_id = reg->btf_id; | |
ac9f0605 KKD |
11747 | } |
11748 | break; | |
00b85860 | 11749 | case KF_ARG_PTR_TO_DYNPTR: |
d96d937d JK |
11750 | { |
11751 | enum bpf_arg_type dynptr_arg_type = ARG_PTR_TO_DYNPTR; | |
361f129f | 11752 | int clone_ref_obj_id = 0; |
d96d937d | 11753 | |
6b75bd3d | 11754 | if (reg->type != PTR_TO_STACK && |
27060531 | 11755 | reg->type != CONST_PTR_TO_DYNPTR) { |
6b75bd3d | 11756 | verbose(env, "arg#%d expected pointer to stack or dynptr_ptr\n", i); |
00b85860 KKD |
11757 | return -EINVAL; |
11758 | } | |
11759 | ||
d96d937d JK |
11760 | if (reg->type == CONST_PTR_TO_DYNPTR) |
11761 | dynptr_arg_type |= MEM_RDONLY; | |
11762 | ||
11763 | if (is_kfunc_arg_uninit(btf, &args[i])) | |
11764 | dynptr_arg_type |= MEM_UNINIT; | |
11765 | ||
361f129f | 11766 | if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_from_skb]) { |
b5964b96 | 11767 | dynptr_arg_type |= DYNPTR_TYPE_SKB; |
361f129f | 11768 | } else if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_from_xdp]) { |
05421aec | 11769 | dynptr_arg_type |= DYNPTR_TYPE_XDP; |
361f129f JK |
11770 | } else if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_clone] && |
11771 | (dynptr_arg_type & MEM_UNINIT)) { | |
11772 | enum bpf_dynptr_type parent_type = meta->initialized_dynptr.type; | |
11773 | ||
11774 | if (parent_type == BPF_DYNPTR_TYPE_INVALID) { | |
11775 | verbose(env, "verifier internal error: no dynptr type for parent of clone\n"); | |
11776 | return -EFAULT; | |
11777 | } | |
11778 | ||
11779 | dynptr_arg_type |= (unsigned int)get_dynptr_type_flag(parent_type); | |
11780 | clone_ref_obj_id = meta->initialized_dynptr.ref_obj_id; | |
11781 | if (dynptr_type_refcounted(parent_type) && !clone_ref_obj_id) { | |
11782 | verbose(env, "verifier internal error: missing ref obj id for parent of clone\n"); | |
11783 | return -EFAULT; | |
11784 | } | |
11785 | } | |
b5964b96 | 11786 | |
361f129f | 11787 | ret = process_dynptr_func(env, regno, insn_idx, dynptr_arg_type, clone_ref_obj_id); |
6b75bd3d KKD |
11788 | if (ret < 0) |
11789 | return ret; | |
66e3a13e JK |
11790 | |
11791 | if (!(dynptr_arg_type & MEM_UNINIT)) { | |
11792 | int id = dynptr_id(env, reg); | |
11793 | ||
11794 | if (id < 0) { | |
11795 | verbose(env, "verifier internal error: failed to obtain dynptr id\n"); | |
11796 | return id; | |
11797 | } | |
11798 | meta->initialized_dynptr.id = id; | |
11799 | meta->initialized_dynptr.type = dynptr_get_type(env, reg); | |
361f129f | 11800 | meta->initialized_dynptr.ref_obj_id = dynptr_ref_obj_id(env, reg); |
66e3a13e JK |
11801 | } |
11802 | ||
00b85860 | 11803 | break; |
d96d937d | 11804 | } |
06accc87 | 11805 | case KF_ARG_PTR_TO_ITER: |
9c66dc94 CZ |
11806 | if (meta->func_id == special_kfunc_list[KF_bpf_iter_css_task_new]) { |
11807 | if (!check_css_task_iter_allowlist(env)) { | |
3091b667 | 11808 | verbose(env, "css_task_iter is only allowed in bpf_lsm, bpf_iter and sleepable progs\n"); |
9c66dc94 CZ |
11809 | return -EINVAL; |
11810 | } | |
11811 | } | |
06accc87 AN |
11812 | ret = process_iter_arg(env, regno, insn_idx, meta); |
11813 | if (ret < 0) | |
11814 | return ret; | |
11815 | break; | |
8cab76ec KKD |
11816 | case KF_ARG_PTR_TO_LIST_HEAD: |
11817 | if (reg->type != PTR_TO_MAP_VALUE && | |
11818 | reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11819 | verbose(env, "arg#%d expected pointer to map value or allocated object\n", i); | |
11820 | return -EINVAL; | |
11821 | } | |
11822 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC) && !reg->ref_obj_id) { | |
11823 | verbose(env, "allocated object must be referenced\n"); | |
11824 | return -EINVAL; | |
11825 | } | |
11826 | ret = process_kf_arg_ptr_to_list_head(env, reg, regno, meta); | |
11827 | if (ret < 0) | |
11828 | return ret; | |
11829 | break; | |
cd6791b4 DM |
11830 | case KF_ARG_PTR_TO_RB_ROOT: |
11831 | if (reg->type != PTR_TO_MAP_VALUE && | |
11832 | reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11833 | verbose(env, "arg#%d expected pointer to map value or allocated object\n", i); | |
11834 | return -EINVAL; | |
11835 | } | |
11836 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC) && !reg->ref_obj_id) { | |
11837 | verbose(env, "allocated object must be referenced\n"); | |
11838 | return -EINVAL; | |
11839 | } | |
11840 | ret = process_kf_arg_ptr_to_rbtree_root(env, reg, regno, meta); | |
11841 | if (ret < 0) | |
11842 | return ret; | |
11843 | break; | |
8cab76ec KKD |
11844 | case KF_ARG_PTR_TO_LIST_NODE: |
11845 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11846 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
11847 | return -EINVAL; | |
11848 | } | |
11849 | if (!reg->ref_obj_id) { | |
11850 | verbose(env, "allocated object must be referenced\n"); | |
11851 | return -EINVAL; | |
11852 | } | |
11853 | ret = process_kf_arg_ptr_to_list_node(env, reg, regno, meta); | |
11854 | if (ret < 0) | |
11855 | return ret; | |
11856 | break; | |
cd6791b4 | 11857 | case KF_ARG_PTR_TO_RB_NODE: |
a40d3632 DM |
11858 | if (meta->func_id == special_kfunc_list[KF_bpf_rbtree_remove]) { |
11859 | if (!type_is_non_owning_ref(reg->type) || reg->ref_obj_id) { | |
11860 | verbose(env, "rbtree_remove node input must be non-owning ref\n"); | |
11861 | return -EINVAL; | |
11862 | } | |
11863 | if (in_rbtree_lock_required_cb(env)) { | |
11864 | verbose(env, "rbtree_remove not allowed in rbtree cb\n"); | |
11865 | return -EINVAL; | |
11866 | } | |
11867 | } else { | |
11868 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11869 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
11870 | return -EINVAL; | |
11871 | } | |
11872 | if (!reg->ref_obj_id) { | |
11873 | verbose(env, "allocated object must be referenced\n"); | |
11874 | return -EINVAL; | |
11875 | } | |
cd6791b4 | 11876 | } |
a40d3632 | 11877 | |
cd6791b4 DM |
11878 | ret = process_kf_arg_ptr_to_rbtree_node(env, reg, regno, meta); |
11879 | if (ret < 0) | |
11880 | return ret; | |
11881 | break; | |
00b85860 KKD |
11882 | case KF_ARG_PTR_TO_BTF_ID: |
11883 | /* Only base_type is checked, further checks are done here */ | |
3f00c523 | 11884 | if ((base_type(reg->type) != PTR_TO_BTF_ID || |
fca1aa75 | 11885 | (bpf_type_has_unsafe_modifiers(reg->type) && !is_rcu_reg(reg))) && |
3f00c523 DV |
11886 | !reg2btf_ids[base_type(reg->type)]) { |
11887 | verbose(env, "arg#%d is %s ", i, reg_type_str(env, reg->type)); | |
11888 | verbose(env, "expected %s or socket\n", | |
11889 | reg_type_str(env, base_type(reg->type) | | |
11890 | (type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS))); | |
00b85860 KKD |
11891 | return -EINVAL; |
11892 | } | |
11893 | ret = process_kf_arg_ptr_to_btf_id(env, reg, ref_t, ref_tname, ref_id, meta, i); | |
11894 | if (ret < 0) | |
11895 | return ret; | |
11896 | break; | |
11897 | case KF_ARG_PTR_TO_MEM: | |
11898 | resolve_ret = btf_resolve_size(btf, ref_t, &type_size); | |
11899 | if (IS_ERR(resolve_ret)) { | |
11900 | verbose(env, "arg#%d reference type('%s %s') size cannot be determined: %ld\n", | |
11901 | i, btf_type_str(ref_t), ref_tname, PTR_ERR(resolve_ret)); | |
11902 | return -EINVAL; | |
11903 | } | |
11904 | ret = check_mem_reg(env, reg, regno, type_size); | |
11905 | if (ret < 0) | |
11906 | return ret; | |
11907 | break; | |
11908 | case KF_ARG_PTR_TO_MEM_SIZE: | |
66e3a13e | 11909 | { |
3bda08b6 DR |
11910 | struct bpf_reg_state *buff_reg = ®s[regno]; |
11911 | const struct btf_param *buff_arg = &args[i]; | |
66e3a13e JK |
11912 | struct bpf_reg_state *size_reg = ®s[regno + 1]; |
11913 | const struct btf_param *size_arg = &args[i + 1]; | |
11914 | ||
3bda08b6 DR |
11915 | if (!register_is_null(buff_reg) || !is_kfunc_arg_optional(meta->btf, buff_arg)) { |
11916 | ret = check_kfunc_mem_size_reg(env, size_reg, regno + 1); | |
11917 | if (ret < 0) { | |
11918 | verbose(env, "arg#%d arg#%d memory, len pair leads to invalid memory access\n", i, i + 1); | |
11919 | return ret; | |
11920 | } | |
00b85860 | 11921 | } |
66e3a13e JK |
11922 | |
11923 | if (is_kfunc_arg_const_mem_size(meta->btf, size_arg, size_reg)) { | |
11924 | if (meta->arg_constant.found) { | |
11925 | verbose(env, "verifier internal error: only one constant argument permitted\n"); | |
11926 | return -EFAULT; | |
11927 | } | |
11928 | if (!tnum_is_const(size_reg->var_off)) { | |
11929 | verbose(env, "R%d must be a known constant\n", regno + 1); | |
11930 | return -EINVAL; | |
11931 | } | |
11932 | meta->arg_constant.found = true; | |
11933 | meta->arg_constant.value = size_reg->var_off.value; | |
11934 | } | |
11935 | ||
11936 | /* Skip next '__sz' or '__szk' argument */ | |
00b85860 KKD |
11937 | i++; |
11938 | break; | |
66e3a13e | 11939 | } |
5d92ddc3 | 11940 | case KF_ARG_PTR_TO_CALLBACK: |
06d686f7 KKD |
11941 | if (reg->type != PTR_TO_FUNC) { |
11942 | verbose(env, "arg%d expected pointer to func\n", i); | |
11943 | return -EINVAL; | |
11944 | } | |
5d92ddc3 DM |
11945 | meta->subprogno = reg->subprogno; |
11946 | break; | |
7c50b1cb | 11947 | case KF_ARG_PTR_TO_REFCOUNTED_KPTR: |
7793fc3b | 11948 | if (!type_is_ptr_alloc_obj(reg->type)) { |
7c50b1cb DM |
11949 | verbose(env, "arg#%d is neither owning or non-owning ref\n", i); |
11950 | return -EINVAL; | |
11951 | } | |
7793fc3b DM |
11952 | if (!type_is_non_owning_ref(reg->type)) |
11953 | meta->arg_owning_ref = true; | |
7c50b1cb DM |
11954 | |
11955 | rec = reg_btf_record(reg); | |
11956 | if (!rec) { | |
11957 | verbose(env, "verifier internal error: Couldn't find btf_record\n"); | |
11958 | return -EFAULT; | |
11959 | } | |
11960 | ||
11961 | if (rec->refcount_off < 0) { | |
11962 | verbose(env, "arg#%d doesn't point to a type with bpf_refcount field\n", i); | |
11963 | return -EINVAL; | |
11964 | } | |
ba2464c8 | 11965 | |
4d585f48 DM |
11966 | meta->arg_btf = reg->btf; |
11967 | meta->arg_btf_id = reg->btf_id; | |
7c50b1cb | 11968 | break; |
00b85860 KKD |
11969 | } |
11970 | } | |
11971 | ||
11972 | if (is_kfunc_release(meta) && !meta->release_regno) { | |
11973 | verbose(env, "release kernel function %s expects refcounted PTR_TO_BTF_ID\n", | |
11974 | func_name); | |
11975 | return -EINVAL; | |
11976 | } | |
11977 | ||
11978 | return 0; | |
11979 | } | |
11980 | ||
07236eab AN |
11981 | static int fetch_kfunc_meta(struct bpf_verifier_env *env, |
11982 | struct bpf_insn *insn, | |
11983 | struct bpf_kfunc_call_arg_meta *meta, | |
11984 | const char **kfunc_name) | |
e6ac2450 | 11985 | { |
07236eab AN |
11986 | const struct btf_type *func, *func_proto; |
11987 | u32 func_id, *kfunc_flags; | |
11988 | const char *func_name; | |
2357672c | 11989 | struct btf *desc_btf; |
e6ac2450 | 11990 | |
07236eab AN |
11991 | if (kfunc_name) |
11992 | *kfunc_name = NULL; | |
11993 | ||
a5d82727 | 11994 | if (!insn->imm) |
07236eab | 11995 | return -EINVAL; |
a5d82727 | 11996 | |
43bf0878 | 11997 | desc_btf = find_kfunc_desc_btf(env, insn->off); |
2357672c KKD |
11998 | if (IS_ERR(desc_btf)) |
11999 | return PTR_ERR(desc_btf); | |
12000 | ||
e6ac2450 | 12001 | func_id = insn->imm; |
2357672c KKD |
12002 | func = btf_type_by_id(desc_btf, func_id); |
12003 | func_name = btf_name_by_offset(desc_btf, func->name_off); | |
07236eab AN |
12004 | if (kfunc_name) |
12005 | *kfunc_name = func_name; | |
2357672c | 12006 | func_proto = btf_type_by_id(desc_btf, func->type); |
e6ac2450 | 12007 | |
e924e80e | 12008 | kfunc_flags = btf_kfunc_id_set_contains(desc_btf, func_id, env->prog); |
a4703e31 | 12009 | if (!kfunc_flags) { |
e6ac2450 MKL |
12010 | return -EACCES; |
12011 | } | |
00b85860 | 12012 | |
07236eab AN |
12013 | memset(meta, 0, sizeof(*meta)); |
12014 | meta->btf = desc_btf; | |
12015 | meta->func_id = func_id; | |
12016 | meta->kfunc_flags = *kfunc_flags; | |
12017 | meta->func_proto = func_proto; | |
12018 | meta->func_name = func_name; | |
12019 | ||
12020 | return 0; | |
12021 | } | |
12022 | ||
a923819f KKD |
12023 | static int check_return_code(struct bpf_verifier_env *env, int regno); |
12024 | ||
07236eab AN |
12025 | static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
12026 | int *insn_idx_p) | |
12027 | { | |
12028 | const struct btf_type *t, *ptr_type; | |
12029 | u32 i, nargs, ptr_type_id, release_ref_obj_id; | |
12030 | struct bpf_reg_state *regs = cur_regs(env); | |
12031 | const char *func_name, *ptr_type_name; | |
12032 | bool sleepable, rcu_lock, rcu_unlock; | |
12033 | struct bpf_kfunc_call_arg_meta meta; | |
12034 | struct bpf_insn_aux_data *insn_aux; | |
12035 | int err, insn_idx = *insn_idx_p; | |
12036 | const struct btf_param *args; | |
12037 | const struct btf_type *ret_t; | |
12038 | struct btf *desc_btf; | |
12039 | ||
12040 | /* skip for now, but return error when we find this in fixup_kfunc_call */ | |
12041 | if (!insn->imm) | |
12042 | return 0; | |
12043 | ||
12044 | err = fetch_kfunc_meta(env, insn, &meta, &func_name); | |
12045 | if (err == -EACCES && func_name) | |
12046 | verbose(env, "calling kernel function %s is not allowed\n", func_name); | |
12047 | if (err) | |
12048 | return err; | |
12049 | desc_btf = meta.btf; | |
12050 | insn_aux = &env->insn_aux_data[insn_idx]; | |
00b85860 | 12051 | |
06accc87 AN |
12052 | insn_aux->is_iter_next = is_iter_next_kfunc(&meta); |
12053 | ||
00b85860 KKD |
12054 | if (is_kfunc_destructive(&meta) && !capable(CAP_SYS_BOOT)) { |
12055 | verbose(env, "destructive kfunc calls require CAP_SYS_BOOT capability\n"); | |
4dd48c6f AS |
12056 | return -EACCES; |
12057 | } | |
12058 | ||
9bb00b28 YS |
12059 | sleepable = is_kfunc_sleepable(&meta); |
12060 | if (sleepable && !env->prog->aux->sleepable) { | |
00b85860 KKD |
12061 | verbose(env, "program must be sleepable to call sleepable kfunc %s\n", func_name); |
12062 | return -EACCES; | |
12063 | } | |
eb1f7f71 | 12064 | |
ab5cfac1 EZ |
12065 | /* Check the arguments */ |
12066 | err = check_kfunc_args(env, &meta, insn_idx); | |
12067 | if (err < 0) | |
12068 | return err; | |
12069 | ||
12070 | if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
12071 | err = push_callback_call(env, insn, insn_idx, meta.subprogno, | |
12072 | set_rbtree_add_callback_state); | |
12073 | if (err) { | |
12074 | verbose(env, "kfunc %s#%d failed callback verification\n", | |
12075 | func_name, meta.func_id); | |
12076 | return err; | |
12077 | } | |
12078 | } | |
12079 | ||
9bb00b28 YS |
12080 | rcu_lock = is_kfunc_bpf_rcu_read_lock(&meta); |
12081 | rcu_unlock = is_kfunc_bpf_rcu_read_unlock(&meta); | |
9bb00b28 YS |
12082 | |
12083 | if (env->cur_state->active_rcu_lock) { | |
12084 | struct bpf_func_state *state; | |
12085 | struct bpf_reg_state *reg; | |
dfab99df | 12086 | u32 clear_mask = (1 << STACK_SPILL) | (1 << STACK_ITER); |
9bb00b28 | 12087 | |
0816b8c6 DM |
12088 | if (in_rbtree_lock_required_cb(env) && (rcu_lock || rcu_unlock)) { |
12089 | verbose(env, "Calling bpf_rcu_read_{lock,unlock} in unnecessary rbtree callback\n"); | |
12090 | return -EACCES; | |
12091 | } | |
12092 | ||
9bb00b28 YS |
12093 | if (rcu_lock) { |
12094 | verbose(env, "nested rcu read lock (kernel function %s)\n", func_name); | |
12095 | return -EINVAL; | |
12096 | } else if (rcu_unlock) { | |
dfab99df | 12097 | bpf_for_each_reg_in_vstate_mask(env->cur_state, state, reg, clear_mask, ({ |
9bb00b28 | 12098 | if (reg->type & MEM_RCU) { |
fca1aa75 | 12099 | reg->type &= ~(MEM_RCU | PTR_MAYBE_NULL); |
9bb00b28 YS |
12100 | reg->type |= PTR_UNTRUSTED; |
12101 | } | |
12102 | })); | |
12103 | env->cur_state->active_rcu_lock = false; | |
12104 | } else if (sleepable) { | |
12105 | verbose(env, "kernel func %s is sleepable within rcu_read_lock region\n", func_name); | |
12106 | return -EACCES; | |
12107 | } | |
12108 | } else if (rcu_lock) { | |
12109 | env->cur_state->active_rcu_lock = true; | |
12110 | } else if (rcu_unlock) { | |
12111 | verbose(env, "unmatched rcu read unlock (kernel function %s)\n", func_name); | |
12112 | return -EINVAL; | |
12113 | } | |
12114 | ||
5c073f26 | 12115 | /* In case of release function, we get register number of refcounted |
00b85860 | 12116 | * PTR_TO_BTF_ID in bpf_kfunc_arg_meta, do the release now. |
5c073f26 | 12117 | */ |
00b85860 KKD |
12118 | if (meta.release_regno) { |
12119 | err = release_reference(env, regs[meta.release_regno].ref_obj_id); | |
5c073f26 KKD |
12120 | if (err) { |
12121 | verbose(env, "kfunc %s#%d reference has not been acquired before\n", | |
07236eab | 12122 | func_name, meta.func_id); |
5c073f26 KKD |
12123 | return err; |
12124 | } | |
12125 | } | |
e6ac2450 | 12126 | |
d2dcc67d DM |
12127 | if (meta.func_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
12128 | meta.func_id == special_kfunc_list[KF_bpf_list_push_back_impl] || | |
12129 | meta.func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
6a3cd331 | 12130 | release_ref_obj_id = regs[BPF_REG_2].ref_obj_id; |
d2dcc67d | 12131 | insn_aux->insert_off = regs[BPF_REG_2].off; |
2140a6e3 | 12132 | insn_aux->kptr_struct_meta = btf_find_struct_meta(meta.arg_btf, meta.arg_btf_id); |
6a3cd331 DM |
12133 | err = ref_convert_owning_non_owning(env, release_ref_obj_id); |
12134 | if (err) { | |
12135 | verbose(env, "kfunc %s#%d conversion of owning ref to non-owning failed\n", | |
07236eab | 12136 | func_name, meta.func_id); |
6a3cd331 DM |
12137 | return err; |
12138 | } | |
12139 | ||
12140 | err = release_reference(env, release_ref_obj_id); | |
12141 | if (err) { | |
12142 | verbose(env, "kfunc %s#%d reference has not been acquired before\n", | |
07236eab | 12143 | func_name, meta.func_id); |
6a3cd331 DM |
12144 | return err; |
12145 | } | |
12146 | } | |
12147 | ||
f18b03fa KKD |
12148 | if (meta.func_id == special_kfunc_list[KF_bpf_throw]) { |
12149 | if (!bpf_jit_supports_exceptions()) { | |
12150 | verbose(env, "JIT does not support calling kfunc %s#%d\n", | |
12151 | func_name, meta.func_id); | |
12152 | return -ENOTSUPP; | |
12153 | } | |
12154 | env->seen_exception = true; | |
a923819f KKD |
12155 | |
12156 | /* In the case of the default callback, the cookie value passed | |
12157 | * to bpf_throw becomes the return value of the program. | |
12158 | */ | |
12159 | if (!env->exception_callback_subprog) { | |
12160 | err = check_return_code(env, BPF_REG_1); | |
12161 | if (err < 0) | |
12162 | return err; | |
12163 | } | |
f18b03fa KKD |
12164 | } |
12165 | ||
e6ac2450 MKL |
12166 | for (i = 0; i < CALLER_SAVED_REGS; i++) |
12167 | mark_reg_not_init(env, regs, caller_saved[i]); | |
12168 | ||
12169 | /* Check return type */ | |
07236eab | 12170 | t = btf_type_skip_modifiers(desc_btf, meta.func_proto->type, NULL); |
5c073f26 | 12171 | |
00b85860 | 12172 | if (is_kfunc_acquire(&meta) && !btf_type_is_struct_ptr(meta.btf, t)) { |
958cf2e2 | 12173 | /* Only exception is bpf_obj_new_impl */ |
7c50b1cb DM |
12174 | if (meta.btf != btf_vmlinux || |
12175 | (meta.func_id != special_kfunc_list[KF_bpf_obj_new_impl] && | |
36d8bdf7 | 12176 | meta.func_id != special_kfunc_list[KF_bpf_percpu_obj_new_impl] && |
7c50b1cb | 12177 | meta.func_id != special_kfunc_list[KF_bpf_refcount_acquire_impl])) { |
958cf2e2 KKD |
12178 | verbose(env, "acquire kernel function does not return PTR_TO_BTF_ID\n"); |
12179 | return -EINVAL; | |
12180 | } | |
5c073f26 KKD |
12181 | } |
12182 | ||
e6ac2450 MKL |
12183 | if (btf_type_is_scalar(t)) { |
12184 | mark_reg_unknown(env, regs, BPF_REG_0); | |
12185 | mark_btf_func_reg_size(env, BPF_REG_0, t->size); | |
12186 | } else if (btf_type_is_ptr(t)) { | |
958cf2e2 KKD |
12187 | ptr_type = btf_type_skip_modifiers(desc_btf, t->type, &ptr_type_id); |
12188 | ||
12189 | if (meta.btf == btf_vmlinux && btf_id_set_contains(&special_kfunc_set, meta.func_id)) { | |
36d8bdf7 YS |
12190 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_new_impl] || |
12191 | meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) { | |
12192 | struct btf_struct_meta *struct_meta; | |
958cf2e2 KKD |
12193 | struct btf *ret_btf; |
12194 | u32 ret_btf_id; | |
12195 | ||
36d8bdf7 YS |
12196 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_new_impl] && !bpf_global_ma_set) |
12197 | return -ENOMEM; | |
12198 | ||
1fda5bb6 YS |
12199 | if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) { |
12200 | if (!bpf_global_percpu_ma_set) { | |
12201 | mutex_lock(&bpf_percpu_ma_lock); | |
12202 | if (!bpf_global_percpu_ma_set) { | |
12203 | err = bpf_mem_alloc_init(&bpf_global_percpu_ma, 0, true); | |
12204 | if (!err) | |
12205 | bpf_global_percpu_ma_set = true; | |
12206 | } | |
12207 | mutex_unlock(&bpf_percpu_ma_lock); | |
12208 | if (err) | |
12209 | return err; | |
12210 | } | |
12211 | } | |
e181d3f1 | 12212 | |
958cf2e2 KKD |
12213 | if (((u64)(u32)meta.arg_constant.value) != meta.arg_constant.value) { |
12214 | verbose(env, "local type ID argument must be in range [0, U32_MAX]\n"); | |
12215 | return -EINVAL; | |
12216 | } | |
12217 | ||
12218 | ret_btf = env->prog->aux->btf; | |
12219 | ret_btf_id = meta.arg_constant.value; | |
12220 | ||
12221 | /* This may be NULL due to user not supplying a BTF */ | |
12222 | if (!ret_btf) { | |
36d8bdf7 | 12223 | verbose(env, "bpf_obj_new/bpf_percpu_obj_new requires prog BTF\n"); |
958cf2e2 KKD |
12224 | return -EINVAL; |
12225 | } | |
12226 | ||
12227 | ret_t = btf_type_by_id(ret_btf, ret_btf_id); | |
12228 | if (!ret_t || !__btf_type_is_struct(ret_t)) { | |
36d8bdf7 | 12229 | verbose(env, "bpf_obj_new/bpf_percpu_obj_new type ID argument must be of a struct\n"); |
958cf2e2 KKD |
12230 | return -EINVAL; |
12231 | } | |
12232 | ||
36d8bdf7 YS |
12233 | struct_meta = btf_find_struct_meta(ret_btf, ret_btf_id); |
12234 | if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) { | |
12235 | if (!__btf_type_is_scalar_struct(env, ret_btf, ret_t, 0)) { | |
12236 | verbose(env, "bpf_percpu_obj_new type ID argument must be of a struct of scalars\n"); | |
12237 | return -EINVAL; | |
12238 | } | |
12239 | ||
12240 | if (struct_meta) { | |
12241 | verbose(env, "bpf_percpu_obj_new type ID argument must not contain special fields\n"); | |
12242 | return -EINVAL; | |
12243 | } | |
12244 | } | |
12245 | ||
958cf2e2 KKD |
12246 | mark_reg_known_zero(env, regs, BPF_REG_0); |
12247 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
12248 | regs[BPF_REG_0].btf = ret_btf; | |
12249 | regs[BPF_REG_0].btf_id = ret_btf_id; | |
36d8bdf7 YS |
12250 | if (meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) |
12251 | regs[BPF_REG_0].type |= MEM_PERCPU; | |
958cf2e2 | 12252 | |
07236eab | 12253 | insn_aux->obj_new_size = ret_t->size; |
36d8bdf7 | 12254 | insn_aux->kptr_struct_meta = struct_meta; |
7c50b1cb DM |
12255 | } else if (meta.func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]) { |
12256 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12257 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
4d585f48 DM |
12258 | regs[BPF_REG_0].btf = meta.arg_btf; |
12259 | regs[BPF_REG_0].btf_id = meta.arg_btf_id; | |
7c50b1cb DM |
12260 | |
12261 | insn_aux->kptr_struct_meta = | |
4d585f48 DM |
12262 | btf_find_struct_meta(meta.arg_btf, |
12263 | meta.arg_btf_id); | |
8cab76ec KKD |
12264 | } else if (meta.func_id == special_kfunc_list[KF_bpf_list_pop_front] || |
12265 | meta.func_id == special_kfunc_list[KF_bpf_list_pop_back]) { | |
12266 | struct btf_field *field = meta.arg_list_head.field; | |
12267 | ||
a40d3632 DM |
12268 | mark_reg_graph_node(regs, BPF_REG_0, &field->graph_root); |
12269 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_remove] || | |
12270 | meta.func_id == special_kfunc_list[KF_bpf_rbtree_first]) { | |
12271 | struct btf_field *field = meta.arg_rbtree_root.field; | |
12272 | ||
12273 | mark_reg_graph_node(regs, BPF_REG_0, &field->graph_root); | |
fd264ca0 YS |
12274 | } else if (meta.func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { |
12275 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12276 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_TRUSTED; | |
12277 | regs[BPF_REG_0].btf = desc_btf; | |
12278 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; | |
a35b9af4 YS |
12279 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { |
12280 | ret_t = btf_type_by_id(desc_btf, meta.arg_constant.value); | |
12281 | if (!ret_t || !btf_type_is_struct(ret_t)) { | |
12282 | verbose(env, | |
12283 | "kfunc bpf_rdonly_cast type ID argument must be of a struct\n"); | |
12284 | return -EINVAL; | |
12285 | } | |
12286 | ||
12287 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12288 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_UNTRUSTED; | |
12289 | regs[BPF_REG_0].btf = desc_btf; | |
12290 | regs[BPF_REG_0].btf_id = meta.arg_constant.value; | |
66e3a13e JK |
12291 | } else if (meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice] || |
12292 | meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice_rdwr]) { | |
12293 | enum bpf_type_flag type_flag = get_dynptr_type_flag(meta.initialized_dynptr.type); | |
12294 | ||
12295 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12296 | ||
12297 | if (!meta.arg_constant.found) { | |
12298 | verbose(env, "verifier internal error: bpf_dynptr_slice(_rdwr) no constant size\n"); | |
12299 | return -EFAULT; | |
12300 | } | |
12301 | ||
12302 | regs[BPF_REG_0].mem_size = meta.arg_constant.value; | |
12303 | ||
12304 | /* PTR_MAYBE_NULL will be added when is_kfunc_ret_null is checked */ | |
12305 | regs[BPF_REG_0].type = PTR_TO_MEM | type_flag; | |
12306 | ||
12307 | if (meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice]) { | |
12308 | regs[BPF_REG_0].type |= MEM_RDONLY; | |
12309 | } else { | |
12310 | /* this will set env->seen_direct_write to true */ | |
12311 | if (!may_access_direct_pkt_data(env, NULL, BPF_WRITE)) { | |
12312 | verbose(env, "the prog does not allow writes to packet data\n"); | |
12313 | return -EINVAL; | |
12314 | } | |
12315 | } | |
12316 | ||
12317 | if (!meta.initialized_dynptr.id) { | |
12318 | verbose(env, "verifier internal error: no dynptr id\n"); | |
12319 | return -EFAULT; | |
12320 | } | |
12321 | regs[BPF_REG_0].dynptr_id = meta.initialized_dynptr.id; | |
12322 | ||
12323 | /* we don't need to set BPF_REG_0's ref obj id | |
12324 | * because packet slices are not refcounted (see | |
12325 | * dynptr_type_refcounted) | |
12326 | */ | |
958cf2e2 KKD |
12327 | } else { |
12328 | verbose(env, "kernel function %s unhandled dynamic return type\n", | |
12329 | meta.func_name); | |
12330 | return -EFAULT; | |
12331 | } | |
12332 | } else if (!__btf_type_is_struct(ptr_type)) { | |
f4b4eee6 AN |
12333 | if (!meta.r0_size) { |
12334 | __u32 sz; | |
12335 | ||
12336 | if (!IS_ERR(btf_resolve_size(desc_btf, ptr_type, &sz))) { | |
12337 | meta.r0_size = sz; | |
12338 | meta.r0_rdonly = true; | |
12339 | } | |
12340 | } | |
eb1f7f71 BT |
12341 | if (!meta.r0_size) { |
12342 | ptr_type_name = btf_name_by_offset(desc_btf, | |
12343 | ptr_type->name_off); | |
12344 | verbose(env, | |
12345 | "kernel function %s returns pointer type %s %s is not supported\n", | |
12346 | func_name, | |
12347 | btf_type_str(ptr_type), | |
12348 | ptr_type_name); | |
12349 | return -EINVAL; | |
12350 | } | |
12351 | ||
12352 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12353 | regs[BPF_REG_0].type = PTR_TO_MEM; | |
12354 | regs[BPF_REG_0].mem_size = meta.r0_size; | |
12355 | ||
12356 | if (meta.r0_rdonly) | |
12357 | regs[BPF_REG_0].type |= MEM_RDONLY; | |
12358 | ||
12359 | /* Ensures we don't access the memory after a release_reference() */ | |
12360 | if (meta.ref_obj_id) | |
12361 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
12362 | } else { | |
12363 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
12364 | regs[BPF_REG_0].btf = desc_btf; | |
12365 | regs[BPF_REG_0].type = PTR_TO_BTF_ID; | |
12366 | regs[BPF_REG_0].btf_id = ptr_type_id; | |
e6ac2450 | 12367 | } |
958cf2e2 | 12368 | |
00b85860 | 12369 | if (is_kfunc_ret_null(&meta)) { |
5c073f26 KKD |
12370 | regs[BPF_REG_0].type |= PTR_MAYBE_NULL; |
12371 | /* For mark_ptr_or_null_reg, see 93c230e3f5bd6 */ | |
12372 | regs[BPF_REG_0].id = ++env->id_gen; | |
12373 | } | |
e6ac2450 | 12374 | mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *)); |
00b85860 | 12375 | if (is_kfunc_acquire(&meta)) { |
5c073f26 KKD |
12376 | int id = acquire_reference_state(env, insn_idx); |
12377 | ||
12378 | if (id < 0) | |
12379 | return id; | |
00b85860 KKD |
12380 | if (is_kfunc_ret_null(&meta)) |
12381 | regs[BPF_REG_0].id = id; | |
5c073f26 | 12382 | regs[BPF_REG_0].ref_obj_id = id; |
a40d3632 DM |
12383 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_first]) { |
12384 | ref_set_non_owning(env, ®s[BPF_REG_0]); | |
5c073f26 | 12385 | } |
a40d3632 | 12386 | |
00b85860 KKD |
12387 | if (reg_may_point_to_spin_lock(®s[BPF_REG_0]) && !regs[BPF_REG_0].id) |
12388 | regs[BPF_REG_0].id = ++env->id_gen; | |
f6a6a5a9 DM |
12389 | } else if (btf_type_is_void(t)) { |
12390 | if (meta.btf == btf_vmlinux && btf_id_set_contains(&special_kfunc_set, meta.func_id)) { | |
36d8bdf7 YS |
12391 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_drop_impl] || |
12392 | meta.func_id == special_kfunc_list[KF_bpf_percpu_obj_drop_impl]) { | |
f6a6a5a9 | 12393 | insn_aux->kptr_struct_meta = |
4d585f48 DM |
12394 | btf_find_struct_meta(meta.arg_btf, |
12395 | meta.arg_btf_id); | |
f6a6a5a9 DM |
12396 | } |
12397 | } | |
12398 | } | |
e6ac2450 | 12399 | |
07236eab AN |
12400 | nargs = btf_type_vlen(meta.func_proto); |
12401 | args = (const struct btf_param *)(meta.func_proto + 1); | |
e6ac2450 MKL |
12402 | for (i = 0; i < nargs; i++) { |
12403 | u32 regno = i + 1; | |
12404 | ||
2357672c | 12405 | t = btf_type_skip_modifiers(desc_btf, args[i].type, NULL); |
e6ac2450 MKL |
12406 | if (btf_type_is_ptr(t)) |
12407 | mark_btf_func_reg_size(env, regno, sizeof(void *)); | |
12408 | else | |
12409 | /* scalar. ensured by btf_check_kfunc_arg_match() */ | |
12410 | mark_btf_func_reg_size(env, regno, t->size); | |
12411 | } | |
12412 | ||
06accc87 AN |
12413 | if (is_iter_next_kfunc(&meta)) { |
12414 | err = process_iter_next_call(env, insn_idx, &meta); | |
12415 | if (err) | |
12416 | return err; | |
12417 | } | |
12418 | ||
e6ac2450 MKL |
12419 | return 0; |
12420 | } | |
12421 | ||
b03c9f9f EC |
12422 | static bool signed_add_overflows(s64 a, s64 b) |
12423 | { | |
12424 | /* Do the add in u64, where overflow is well-defined */ | |
12425 | s64 res = (s64)((u64)a + (u64)b); | |
12426 | ||
12427 | if (b < 0) | |
12428 | return res > a; | |
12429 | return res < a; | |
12430 | } | |
12431 | ||
bc895e8b | 12432 | static bool signed_add32_overflows(s32 a, s32 b) |
3f50f132 JF |
12433 | { |
12434 | /* Do the add in u32, where overflow is well-defined */ | |
12435 | s32 res = (s32)((u32)a + (u32)b); | |
12436 | ||
12437 | if (b < 0) | |
12438 | return res > a; | |
12439 | return res < a; | |
12440 | } | |
12441 | ||
bc895e8b | 12442 | static bool signed_sub_overflows(s64 a, s64 b) |
b03c9f9f EC |
12443 | { |
12444 | /* Do the sub in u64, where overflow is well-defined */ | |
12445 | s64 res = (s64)((u64)a - (u64)b); | |
12446 | ||
12447 | if (b < 0) | |
12448 | return res < a; | |
12449 | return res > a; | |
969bf05e AS |
12450 | } |
12451 | ||
3f50f132 JF |
12452 | static bool signed_sub32_overflows(s32 a, s32 b) |
12453 | { | |
bc895e8b | 12454 | /* Do the sub in u32, where overflow is well-defined */ |
3f50f132 JF |
12455 | s32 res = (s32)((u32)a - (u32)b); |
12456 | ||
12457 | if (b < 0) | |
12458 | return res < a; | |
12459 | return res > a; | |
12460 | } | |
12461 | ||
bb7f0f98 AS |
12462 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
12463 | const struct bpf_reg_state *reg, | |
12464 | enum bpf_reg_type type) | |
12465 | { | |
12466 | bool known = tnum_is_const(reg->var_off); | |
12467 | s64 val = reg->var_off.value; | |
12468 | s64 smin = reg->smin_value; | |
12469 | ||
12470 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
12471 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
c25b2ae1 | 12472 | reg_type_str(env, type), val); |
bb7f0f98 AS |
12473 | return false; |
12474 | } | |
12475 | ||
12476 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
12477 | verbose(env, "%s pointer offset %d is not allowed\n", | |
c25b2ae1 | 12478 | reg_type_str(env, type), reg->off); |
bb7f0f98 AS |
12479 | return false; |
12480 | } | |
12481 | ||
12482 | if (smin == S64_MIN) { | |
12483 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
c25b2ae1 | 12484 | reg_type_str(env, type)); |
bb7f0f98 AS |
12485 | return false; |
12486 | } | |
12487 | ||
12488 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
12489 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
c25b2ae1 | 12490 | smin, reg_type_str(env, type)); |
bb7f0f98 AS |
12491 | return false; |
12492 | } | |
12493 | ||
12494 | return true; | |
12495 | } | |
12496 | ||
a6aaece0 DB |
12497 | enum { |
12498 | REASON_BOUNDS = -1, | |
12499 | REASON_TYPE = -2, | |
12500 | REASON_PATHS = -3, | |
12501 | REASON_LIMIT = -4, | |
12502 | REASON_STACK = -5, | |
12503 | }; | |
12504 | ||
979d63d5 | 12505 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, |
bb01a1bb | 12506 | u32 *alu_limit, bool mask_to_left) |
979d63d5 | 12507 | { |
7fedb63a | 12508 | u32 max = 0, ptr_limit = 0; |
979d63d5 DB |
12509 | |
12510 | switch (ptr_reg->type) { | |
12511 | case PTR_TO_STACK: | |
1b1597e6 | 12512 | /* Offset 0 is out-of-bounds, but acceptable start for the |
7fedb63a DB |
12513 | * left direction, see BPF_REG_FP. Also, unknown scalar |
12514 | * offset where we would need to deal with min/max bounds is | |
12515 | * currently prohibited for unprivileged. | |
1b1597e6 PK |
12516 | */ |
12517 | max = MAX_BPF_STACK + mask_to_left; | |
7fedb63a | 12518 | ptr_limit = -(ptr_reg->var_off.value + ptr_reg->off); |
b658bbb8 | 12519 | break; |
979d63d5 | 12520 | case PTR_TO_MAP_VALUE: |
1b1597e6 | 12521 | max = ptr_reg->map_ptr->value_size; |
7fedb63a DB |
12522 | ptr_limit = (mask_to_left ? |
12523 | ptr_reg->smin_value : | |
12524 | ptr_reg->umax_value) + ptr_reg->off; | |
b658bbb8 | 12525 | break; |
979d63d5 | 12526 | default: |
a6aaece0 | 12527 | return REASON_TYPE; |
979d63d5 | 12528 | } |
b658bbb8 DB |
12529 | |
12530 | if (ptr_limit >= max) | |
a6aaece0 | 12531 | return REASON_LIMIT; |
b658bbb8 DB |
12532 | *alu_limit = ptr_limit; |
12533 | return 0; | |
979d63d5 DB |
12534 | } |
12535 | ||
d3bd7413 DB |
12536 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
12537 | const struct bpf_insn *insn) | |
12538 | { | |
2c78ee89 | 12539 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
12540 | } |
12541 | ||
12542 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
12543 | u32 alu_state, u32 alu_limit) | |
12544 | { | |
12545 | /* If we arrived here from different branches with different | |
12546 | * state or limits to sanitize, then this won't work. | |
12547 | */ | |
12548 | if (aux->alu_state && | |
12549 | (aux->alu_state != alu_state || | |
12550 | aux->alu_limit != alu_limit)) | |
a6aaece0 | 12551 | return REASON_PATHS; |
d3bd7413 | 12552 | |
e6ac5933 | 12553 | /* Corresponding fixup done in do_misc_fixups(). */ |
d3bd7413 DB |
12554 | aux->alu_state = alu_state; |
12555 | aux->alu_limit = alu_limit; | |
12556 | return 0; | |
12557 | } | |
12558 | ||
12559 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
12560 | struct bpf_insn *insn) | |
12561 | { | |
12562 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
12563 | ||
12564 | if (can_skip_alu_sanitation(env, insn)) | |
12565 | return 0; | |
12566 | ||
12567 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
12568 | } | |
12569 | ||
f5288193 DB |
12570 | static bool sanitize_needed(u8 opcode) |
12571 | { | |
12572 | return opcode == BPF_ADD || opcode == BPF_SUB; | |
12573 | } | |
12574 | ||
3d0220f6 DB |
12575 | struct bpf_sanitize_info { |
12576 | struct bpf_insn_aux_data aux; | |
bb01a1bb | 12577 | bool mask_to_left; |
3d0220f6 DB |
12578 | }; |
12579 | ||
9183671a DB |
12580 | static struct bpf_verifier_state * |
12581 | sanitize_speculative_path(struct bpf_verifier_env *env, | |
12582 | const struct bpf_insn *insn, | |
12583 | u32 next_idx, u32 curr_idx) | |
12584 | { | |
12585 | struct bpf_verifier_state *branch; | |
12586 | struct bpf_reg_state *regs; | |
12587 | ||
12588 | branch = push_stack(env, next_idx, curr_idx, true); | |
12589 | if (branch && insn) { | |
12590 | regs = branch->frame[branch->curframe]->regs; | |
12591 | if (BPF_SRC(insn->code) == BPF_K) { | |
12592 | mark_reg_unknown(env, regs, insn->dst_reg); | |
12593 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
12594 | mark_reg_unknown(env, regs, insn->dst_reg); | |
12595 | mark_reg_unknown(env, regs, insn->src_reg); | |
12596 | } | |
12597 | } | |
12598 | return branch; | |
12599 | } | |
12600 | ||
979d63d5 DB |
12601 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
12602 | struct bpf_insn *insn, | |
12603 | const struct bpf_reg_state *ptr_reg, | |
6f55b2f2 | 12604 | const struct bpf_reg_state *off_reg, |
979d63d5 | 12605 | struct bpf_reg_state *dst_reg, |
3d0220f6 | 12606 | struct bpf_sanitize_info *info, |
7fedb63a | 12607 | const bool commit_window) |
979d63d5 | 12608 | { |
3d0220f6 | 12609 | struct bpf_insn_aux_data *aux = commit_window ? cur_aux(env) : &info->aux; |
979d63d5 | 12610 | struct bpf_verifier_state *vstate = env->cur_state; |
801c6058 | 12611 | bool off_is_imm = tnum_is_const(off_reg->var_off); |
6f55b2f2 | 12612 | bool off_is_neg = off_reg->smin_value < 0; |
979d63d5 DB |
12613 | bool ptr_is_dst_reg = ptr_reg == dst_reg; |
12614 | u8 opcode = BPF_OP(insn->code); | |
12615 | u32 alu_state, alu_limit; | |
12616 | struct bpf_reg_state tmp; | |
12617 | bool ret; | |
f232326f | 12618 | int err; |
979d63d5 | 12619 | |
d3bd7413 | 12620 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
12621 | return 0; |
12622 | ||
12623 | /* We already marked aux for masking from non-speculative | |
12624 | * paths, thus we got here in the first place. We only care | |
12625 | * to explore bad access from here. | |
12626 | */ | |
12627 | if (vstate->speculative) | |
12628 | goto do_sim; | |
12629 | ||
bb01a1bb DB |
12630 | if (!commit_window) { |
12631 | if (!tnum_is_const(off_reg->var_off) && | |
12632 | (off_reg->smin_value < 0) != (off_reg->smax_value < 0)) | |
12633 | return REASON_BOUNDS; | |
12634 | ||
12635 | info->mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
12636 | (opcode == BPF_SUB && !off_is_neg); | |
12637 | } | |
12638 | ||
12639 | err = retrieve_ptr_limit(ptr_reg, &alu_limit, info->mask_to_left); | |
f232326f PK |
12640 | if (err < 0) |
12641 | return err; | |
12642 | ||
7fedb63a DB |
12643 | if (commit_window) { |
12644 | /* In commit phase we narrow the masking window based on | |
12645 | * the observed pointer move after the simulated operation. | |
12646 | */ | |
3d0220f6 DB |
12647 | alu_state = info->aux.alu_state; |
12648 | alu_limit = abs(info->aux.alu_limit - alu_limit); | |
7fedb63a DB |
12649 | } else { |
12650 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
801c6058 | 12651 | alu_state |= off_is_imm ? BPF_ALU_IMMEDIATE : 0; |
7fedb63a DB |
12652 | alu_state |= ptr_is_dst_reg ? |
12653 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
e042aa53 DB |
12654 | |
12655 | /* Limit pruning on unknown scalars to enable deep search for | |
12656 | * potential masking differences from other program paths. | |
12657 | */ | |
12658 | if (!off_is_imm) | |
12659 | env->explore_alu_limits = true; | |
7fedb63a DB |
12660 | } |
12661 | ||
f232326f PK |
12662 | err = update_alu_sanitation_state(aux, alu_state, alu_limit); |
12663 | if (err < 0) | |
12664 | return err; | |
979d63d5 | 12665 | do_sim: |
7fedb63a DB |
12666 | /* If we're in commit phase, we're done here given we already |
12667 | * pushed the truncated dst_reg into the speculative verification | |
12668 | * stack. | |
a7036191 DB |
12669 | * |
12670 | * Also, when register is a known constant, we rewrite register-based | |
12671 | * operation to immediate-based, and thus do not need masking (and as | |
12672 | * a consequence, do not need to simulate the zero-truncation either). | |
7fedb63a | 12673 | */ |
a7036191 | 12674 | if (commit_window || off_is_imm) |
7fedb63a DB |
12675 | return 0; |
12676 | ||
979d63d5 DB |
12677 | /* Simulate and find potential out-of-bounds access under |
12678 | * speculative execution from truncation as a result of | |
12679 | * masking when off was not within expected range. If off | |
12680 | * sits in dst, then we temporarily need to move ptr there | |
12681 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
12682 | * for cases where we use K-based arithmetic in one direction | |
12683 | * and truncated reg-based in the other in order to explore | |
12684 | * bad access. | |
12685 | */ | |
12686 | if (!ptr_is_dst_reg) { | |
12687 | tmp = *dst_reg; | |
71f656a5 | 12688 | copy_register_state(dst_reg, ptr_reg); |
979d63d5 | 12689 | } |
9183671a DB |
12690 | ret = sanitize_speculative_path(env, NULL, env->insn_idx + 1, |
12691 | env->insn_idx); | |
0803278b | 12692 | if (!ptr_is_dst_reg && ret) |
979d63d5 | 12693 | *dst_reg = tmp; |
a6aaece0 DB |
12694 | return !ret ? REASON_STACK : 0; |
12695 | } | |
12696 | ||
fe9a5ca7 DB |
12697 | static void sanitize_mark_insn_seen(struct bpf_verifier_env *env) |
12698 | { | |
12699 | struct bpf_verifier_state *vstate = env->cur_state; | |
12700 | ||
12701 | /* If we simulate paths under speculation, we don't update the | |
12702 | * insn as 'seen' such that when we verify unreachable paths in | |
12703 | * the non-speculative domain, sanitize_dead_code() can still | |
12704 | * rewrite/sanitize them. | |
12705 | */ | |
12706 | if (!vstate->speculative) | |
12707 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; | |
12708 | } | |
12709 | ||
a6aaece0 DB |
12710 | static int sanitize_err(struct bpf_verifier_env *env, |
12711 | const struct bpf_insn *insn, int reason, | |
12712 | const struct bpf_reg_state *off_reg, | |
12713 | const struct bpf_reg_state *dst_reg) | |
12714 | { | |
12715 | static const char *err = "pointer arithmetic with it prohibited for !root"; | |
12716 | const char *op = BPF_OP(insn->code) == BPF_ADD ? "add" : "sub"; | |
12717 | u32 dst = insn->dst_reg, src = insn->src_reg; | |
12718 | ||
12719 | switch (reason) { | |
12720 | case REASON_BOUNDS: | |
12721 | verbose(env, "R%d has unknown scalar with mixed signed bounds, %s\n", | |
12722 | off_reg == dst_reg ? dst : src, err); | |
12723 | break; | |
12724 | case REASON_TYPE: | |
12725 | verbose(env, "R%d has pointer with unsupported alu operation, %s\n", | |
12726 | off_reg == dst_reg ? src : dst, err); | |
12727 | break; | |
12728 | case REASON_PATHS: | |
12729 | verbose(env, "R%d tried to %s from different maps, paths or scalars, %s\n", | |
12730 | dst, op, err); | |
12731 | break; | |
12732 | case REASON_LIMIT: | |
12733 | verbose(env, "R%d tried to %s beyond pointer bounds, %s\n", | |
12734 | dst, op, err); | |
12735 | break; | |
12736 | case REASON_STACK: | |
12737 | verbose(env, "R%d could not be pushed for speculative verification, %s\n", | |
12738 | dst, err); | |
12739 | break; | |
12740 | default: | |
12741 | verbose(env, "verifier internal error: unknown reason (%d)\n", | |
12742 | reason); | |
12743 | break; | |
12744 | } | |
12745 | ||
12746 | return -EACCES; | |
979d63d5 DB |
12747 | } |
12748 | ||
01f810ac AM |
12749 | /* check that stack access falls within stack limits and that 'reg' doesn't |
12750 | * have a variable offset. | |
12751 | * | |
12752 | * Variable offset is prohibited for unprivileged mode for simplicity since it | |
12753 | * requires corresponding support in Spectre masking for stack ALU. See also | |
12754 | * retrieve_ptr_limit(). | |
12755 | * | |
12756 | * | |
12757 | * 'off' includes 'reg->off'. | |
12758 | */ | |
12759 | static int check_stack_access_for_ptr_arithmetic( | |
12760 | struct bpf_verifier_env *env, | |
12761 | int regno, | |
12762 | const struct bpf_reg_state *reg, | |
12763 | int off) | |
12764 | { | |
12765 | if (!tnum_is_const(reg->var_off)) { | |
12766 | char tn_buf[48]; | |
12767 | ||
12768 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
12769 | verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n", | |
12770 | regno, tn_buf, off); | |
12771 | return -EACCES; | |
12772 | } | |
12773 | ||
12774 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
12775 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
12776 | "prohibited for !root; off=%d\n", regno, off); | |
12777 | return -EACCES; | |
12778 | } | |
12779 | ||
12780 | return 0; | |
12781 | } | |
12782 | ||
073815b7 DB |
12783 | static int sanitize_check_bounds(struct bpf_verifier_env *env, |
12784 | const struct bpf_insn *insn, | |
12785 | const struct bpf_reg_state *dst_reg) | |
12786 | { | |
12787 | u32 dst = insn->dst_reg; | |
12788 | ||
12789 | /* For unprivileged we require that resulting offset must be in bounds | |
12790 | * in order to be able to sanitize access later on. | |
12791 | */ | |
12792 | if (env->bypass_spec_v1) | |
12793 | return 0; | |
12794 | ||
12795 | switch (dst_reg->type) { | |
12796 | case PTR_TO_STACK: | |
12797 | if (check_stack_access_for_ptr_arithmetic(env, dst, dst_reg, | |
12798 | dst_reg->off + dst_reg->var_off.value)) | |
12799 | return -EACCES; | |
12800 | break; | |
12801 | case PTR_TO_MAP_VALUE: | |
61df10c7 | 12802 | if (check_map_access(env, dst, dst_reg->off, 1, false, ACCESS_HELPER)) { |
073815b7 DB |
12803 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " |
12804 | "prohibited for !root\n", dst); | |
12805 | return -EACCES; | |
12806 | } | |
12807 | break; | |
12808 | default: | |
12809 | break; | |
12810 | } | |
12811 | ||
12812 | return 0; | |
12813 | } | |
01f810ac | 12814 | |
f1174f77 | 12815 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
12816 | * Caller should also handle BPF_MOV case separately. |
12817 | * If we return -EACCES, caller may want to try again treating pointer as a | |
12818 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
12819 | */ | |
12820 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
12821 | struct bpf_insn *insn, | |
12822 | const struct bpf_reg_state *ptr_reg, | |
12823 | const struct bpf_reg_state *off_reg) | |
969bf05e | 12824 | { |
f4d7e40a AS |
12825 | struct bpf_verifier_state *vstate = env->cur_state; |
12826 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
12827 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 12828 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
12829 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
12830 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
12831 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
12832 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
3d0220f6 | 12833 | struct bpf_sanitize_info info = {}; |
969bf05e | 12834 | u8 opcode = BPF_OP(insn->code); |
24c109bb | 12835 | u32 dst = insn->dst_reg; |
979d63d5 | 12836 | int ret; |
969bf05e | 12837 | |
f1174f77 | 12838 | dst_reg = ®s[dst]; |
969bf05e | 12839 | |
6f16101e DB |
12840 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
12841 | smin_val > smax_val || umin_val > umax_val) { | |
12842 | /* Taint dst register if offset had invalid bounds derived from | |
12843 | * e.g. dead branches. | |
12844 | */ | |
f54c7898 | 12845 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 12846 | return 0; |
f1174f77 EC |
12847 | } |
12848 | ||
12849 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
12850 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
6c693541 YS |
12851 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
12852 | __mark_reg_unknown(env, dst_reg); | |
12853 | return 0; | |
12854 | } | |
12855 | ||
82abbf8d AS |
12856 | verbose(env, |
12857 | "R%d 32-bit pointer arithmetic prohibited\n", | |
12858 | dst); | |
f1174f77 | 12859 | return -EACCES; |
969bf05e AS |
12860 | } |
12861 | ||
c25b2ae1 | 12862 | if (ptr_reg->type & PTR_MAYBE_NULL) { |
aad2eeaf | 12863 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", |
c25b2ae1 | 12864 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 12865 | return -EACCES; |
c25b2ae1 HL |
12866 | } |
12867 | ||
12868 | switch (base_type(ptr_reg->type)) { | |
aad2eeaf | 12869 | case CONST_PTR_TO_MAP: |
7c696732 YS |
12870 | /* smin_val represents the known value */ |
12871 | if (known && smin_val == 0 && opcode == BPF_ADD) | |
12872 | break; | |
8731745e | 12873 | fallthrough; |
aad2eeaf | 12874 | case PTR_TO_PACKET_END: |
c64b7983 | 12875 | case PTR_TO_SOCKET: |
46f8bc92 | 12876 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 12877 | case PTR_TO_TCP_SOCK: |
fada7fdc | 12878 | case PTR_TO_XDP_SOCK: |
aad2eeaf | 12879 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
c25b2ae1 | 12880 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 12881 | return -EACCES; |
aad2eeaf JS |
12882 | default: |
12883 | break; | |
f1174f77 EC |
12884 | } |
12885 | ||
12886 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
12887 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 12888 | */ |
f1174f77 EC |
12889 | dst_reg->type = ptr_reg->type; |
12890 | dst_reg->id = ptr_reg->id; | |
969bf05e | 12891 | |
bb7f0f98 AS |
12892 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
12893 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
12894 | return -EINVAL; | |
12895 | ||
3f50f132 JF |
12896 | /* pointer types do not carry 32-bit bounds at the moment. */ |
12897 | __mark_reg32_unbounded(dst_reg); | |
12898 | ||
7fedb63a DB |
12899 | if (sanitize_needed(opcode)) { |
12900 | ret = sanitize_ptr_alu(env, insn, ptr_reg, off_reg, dst_reg, | |
3d0220f6 | 12901 | &info, false); |
a6aaece0 DB |
12902 | if (ret < 0) |
12903 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
7fedb63a | 12904 | } |
a6aaece0 | 12905 | |
f1174f77 EC |
12906 | switch (opcode) { |
12907 | case BPF_ADD: | |
12908 | /* We can take a fixed offset as long as it doesn't overflow | |
12909 | * the s32 'off' field | |
969bf05e | 12910 | */ |
b03c9f9f EC |
12911 | if (known && (ptr_reg->off + smin_val == |
12912 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 12913 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
12914 | dst_reg->smin_value = smin_ptr; |
12915 | dst_reg->smax_value = smax_ptr; | |
12916 | dst_reg->umin_value = umin_ptr; | |
12917 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 12918 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 12919 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 12920 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
12921 | break; |
12922 | } | |
f1174f77 EC |
12923 | /* A new variable offset is created. Note that off_reg->off |
12924 | * == 0, since it's a scalar. | |
12925 | * dst_reg gets the pointer type and since some positive | |
12926 | * integer value was added to the pointer, give it a new 'id' | |
12927 | * if it's a PTR_TO_PACKET. | |
12928 | * this creates a new 'base' pointer, off_reg (variable) gets | |
12929 | * added into the variable offset, and we copy the fixed offset | |
12930 | * from ptr_reg. | |
969bf05e | 12931 | */ |
b03c9f9f EC |
12932 | if (signed_add_overflows(smin_ptr, smin_val) || |
12933 | signed_add_overflows(smax_ptr, smax_val)) { | |
12934 | dst_reg->smin_value = S64_MIN; | |
12935 | dst_reg->smax_value = S64_MAX; | |
12936 | } else { | |
12937 | dst_reg->smin_value = smin_ptr + smin_val; | |
12938 | dst_reg->smax_value = smax_ptr + smax_val; | |
12939 | } | |
12940 | if (umin_ptr + umin_val < umin_ptr || | |
12941 | umax_ptr + umax_val < umax_ptr) { | |
12942 | dst_reg->umin_value = 0; | |
12943 | dst_reg->umax_value = U64_MAX; | |
12944 | } else { | |
12945 | dst_reg->umin_value = umin_ptr + umin_val; | |
12946 | dst_reg->umax_value = umax_ptr + umax_val; | |
12947 | } | |
f1174f77 EC |
12948 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
12949 | dst_reg->off = ptr_reg->off; | |
0962590e | 12950 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 12951 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
12952 | dst_reg->id = ++env->id_gen; |
12953 | /* something was added to pkt_ptr, set range to zero */ | |
22dc4a0f | 12954 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
f1174f77 EC |
12955 | } |
12956 | break; | |
12957 | case BPF_SUB: | |
12958 | if (dst_reg == off_reg) { | |
12959 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
12960 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
12961 | dst); | |
f1174f77 EC |
12962 | return -EACCES; |
12963 | } | |
12964 | /* We don't allow subtraction from FP, because (according to | |
12965 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
12966 | * be able to deal with it. | |
969bf05e | 12967 | */ |
f1174f77 | 12968 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
12969 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
12970 | dst); | |
f1174f77 EC |
12971 | return -EACCES; |
12972 | } | |
b03c9f9f EC |
12973 | if (known && (ptr_reg->off - smin_val == |
12974 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 12975 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
12976 | dst_reg->smin_value = smin_ptr; |
12977 | dst_reg->smax_value = smax_ptr; | |
12978 | dst_reg->umin_value = umin_ptr; | |
12979 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
12980 | dst_reg->var_off = ptr_reg->var_off; |
12981 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 12982 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 12983 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
12984 | break; |
12985 | } | |
f1174f77 EC |
12986 | /* A new variable offset is created. If the subtrahend is known |
12987 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 12988 | */ |
b03c9f9f EC |
12989 | if (signed_sub_overflows(smin_ptr, smax_val) || |
12990 | signed_sub_overflows(smax_ptr, smin_val)) { | |
12991 | /* Overflow possible, we know nothing */ | |
12992 | dst_reg->smin_value = S64_MIN; | |
12993 | dst_reg->smax_value = S64_MAX; | |
12994 | } else { | |
12995 | dst_reg->smin_value = smin_ptr - smax_val; | |
12996 | dst_reg->smax_value = smax_ptr - smin_val; | |
12997 | } | |
12998 | if (umin_ptr < umax_val) { | |
12999 | /* Overflow possible, we know nothing */ | |
13000 | dst_reg->umin_value = 0; | |
13001 | dst_reg->umax_value = U64_MAX; | |
13002 | } else { | |
13003 | /* Cannot overflow (as long as bounds are consistent) */ | |
13004 | dst_reg->umin_value = umin_ptr - umax_val; | |
13005 | dst_reg->umax_value = umax_ptr - umin_val; | |
13006 | } | |
f1174f77 EC |
13007 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
13008 | dst_reg->off = ptr_reg->off; | |
0962590e | 13009 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 13010 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
13011 | dst_reg->id = ++env->id_gen; |
13012 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 13013 | if (smin_val < 0) |
22dc4a0f | 13014 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
43188702 | 13015 | } |
f1174f77 EC |
13016 | break; |
13017 | case BPF_AND: | |
13018 | case BPF_OR: | |
13019 | case BPF_XOR: | |
82abbf8d AS |
13020 | /* bitwise ops on pointers are troublesome, prohibit. */ |
13021 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
13022 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
13023 | return -EACCES; |
13024 | default: | |
13025 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
13026 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
13027 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 13028 | return -EACCES; |
43188702 JF |
13029 | } |
13030 | ||
bb7f0f98 AS |
13031 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
13032 | return -EINVAL; | |
3844d153 | 13033 | reg_bounds_sync(dst_reg); |
073815b7 DB |
13034 | if (sanitize_check_bounds(env, insn, dst_reg) < 0) |
13035 | return -EACCES; | |
7fedb63a DB |
13036 | if (sanitize_needed(opcode)) { |
13037 | ret = sanitize_ptr_alu(env, insn, dst_reg, off_reg, dst_reg, | |
3d0220f6 | 13038 | &info, true); |
7fedb63a DB |
13039 | if (ret < 0) |
13040 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
0d6303db DB |
13041 | } |
13042 | ||
43188702 JF |
13043 | return 0; |
13044 | } | |
13045 | ||
3f50f132 JF |
13046 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
13047 | struct bpf_reg_state *src_reg) | |
13048 | { | |
13049 | s32 smin_val = src_reg->s32_min_value; | |
13050 | s32 smax_val = src_reg->s32_max_value; | |
13051 | u32 umin_val = src_reg->u32_min_value; | |
13052 | u32 umax_val = src_reg->u32_max_value; | |
13053 | ||
13054 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
13055 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
13056 | dst_reg->s32_min_value = S32_MIN; | |
13057 | dst_reg->s32_max_value = S32_MAX; | |
13058 | } else { | |
13059 | dst_reg->s32_min_value += smin_val; | |
13060 | dst_reg->s32_max_value += smax_val; | |
13061 | } | |
13062 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
13063 | dst_reg->u32_max_value + umax_val < umax_val) { | |
13064 | dst_reg->u32_min_value = 0; | |
13065 | dst_reg->u32_max_value = U32_MAX; | |
13066 | } else { | |
13067 | dst_reg->u32_min_value += umin_val; | |
13068 | dst_reg->u32_max_value += umax_val; | |
13069 | } | |
13070 | } | |
13071 | ||
07cd2631 JF |
13072 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
13073 | struct bpf_reg_state *src_reg) | |
13074 | { | |
13075 | s64 smin_val = src_reg->smin_value; | |
13076 | s64 smax_val = src_reg->smax_value; | |
13077 | u64 umin_val = src_reg->umin_value; | |
13078 | u64 umax_val = src_reg->umax_value; | |
13079 | ||
13080 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
13081 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
13082 | dst_reg->smin_value = S64_MIN; | |
13083 | dst_reg->smax_value = S64_MAX; | |
13084 | } else { | |
13085 | dst_reg->smin_value += smin_val; | |
13086 | dst_reg->smax_value += smax_val; | |
13087 | } | |
13088 | if (dst_reg->umin_value + umin_val < umin_val || | |
13089 | dst_reg->umax_value + umax_val < umax_val) { | |
13090 | dst_reg->umin_value = 0; | |
13091 | dst_reg->umax_value = U64_MAX; | |
13092 | } else { | |
13093 | dst_reg->umin_value += umin_val; | |
13094 | dst_reg->umax_value += umax_val; | |
13095 | } | |
3f50f132 JF |
13096 | } |
13097 | ||
13098 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
13099 | struct bpf_reg_state *src_reg) | |
13100 | { | |
13101 | s32 smin_val = src_reg->s32_min_value; | |
13102 | s32 smax_val = src_reg->s32_max_value; | |
13103 | u32 umin_val = src_reg->u32_min_value; | |
13104 | u32 umax_val = src_reg->u32_max_value; | |
13105 | ||
13106 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
13107 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
13108 | /* Overflow possible, we know nothing */ | |
13109 | dst_reg->s32_min_value = S32_MIN; | |
13110 | dst_reg->s32_max_value = S32_MAX; | |
13111 | } else { | |
13112 | dst_reg->s32_min_value -= smax_val; | |
13113 | dst_reg->s32_max_value -= smin_val; | |
13114 | } | |
13115 | if (dst_reg->u32_min_value < umax_val) { | |
13116 | /* Overflow possible, we know nothing */ | |
13117 | dst_reg->u32_min_value = 0; | |
13118 | dst_reg->u32_max_value = U32_MAX; | |
13119 | } else { | |
13120 | /* Cannot overflow (as long as bounds are consistent) */ | |
13121 | dst_reg->u32_min_value -= umax_val; | |
13122 | dst_reg->u32_max_value -= umin_val; | |
13123 | } | |
07cd2631 JF |
13124 | } |
13125 | ||
13126 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
13127 | struct bpf_reg_state *src_reg) | |
13128 | { | |
13129 | s64 smin_val = src_reg->smin_value; | |
13130 | s64 smax_val = src_reg->smax_value; | |
13131 | u64 umin_val = src_reg->umin_value; | |
13132 | u64 umax_val = src_reg->umax_value; | |
13133 | ||
13134 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
13135 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
13136 | /* Overflow possible, we know nothing */ | |
13137 | dst_reg->smin_value = S64_MIN; | |
13138 | dst_reg->smax_value = S64_MAX; | |
13139 | } else { | |
13140 | dst_reg->smin_value -= smax_val; | |
13141 | dst_reg->smax_value -= smin_val; | |
13142 | } | |
13143 | if (dst_reg->umin_value < umax_val) { | |
13144 | /* Overflow possible, we know nothing */ | |
13145 | dst_reg->umin_value = 0; | |
13146 | dst_reg->umax_value = U64_MAX; | |
13147 | } else { | |
13148 | /* Cannot overflow (as long as bounds are consistent) */ | |
13149 | dst_reg->umin_value -= umax_val; | |
13150 | dst_reg->umax_value -= umin_val; | |
13151 | } | |
3f50f132 JF |
13152 | } |
13153 | ||
13154 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
13155 | struct bpf_reg_state *src_reg) | |
13156 | { | |
13157 | s32 smin_val = src_reg->s32_min_value; | |
13158 | u32 umin_val = src_reg->u32_min_value; | |
13159 | u32 umax_val = src_reg->u32_max_value; | |
13160 | ||
13161 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
13162 | /* Ain't nobody got time to multiply that sign */ | |
13163 | __mark_reg32_unbounded(dst_reg); | |
13164 | return; | |
13165 | } | |
13166 | /* Both values are positive, so we can work with unsigned and | |
13167 | * copy the result to signed (unless it exceeds S32_MAX). | |
13168 | */ | |
13169 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
13170 | /* Potential overflow, we know nothing */ | |
13171 | __mark_reg32_unbounded(dst_reg); | |
13172 | return; | |
13173 | } | |
13174 | dst_reg->u32_min_value *= umin_val; | |
13175 | dst_reg->u32_max_value *= umax_val; | |
13176 | if (dst_reg->u32_max_value > S32_MAX) { | |
13177 | /* Overflow possible, we know nothing */ | |
13178 | dst_reg->s32_min_value = S32_MIN; | |
13179 | dst_reg->s32_max_value = S32_MAX; | |
13180 | } else { | |
13181 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
13182 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
13183 | } | |
07cd2631 JF |
13184 | } |
13185 | ||
13186 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
13187 | struct bpf_reg_state *src_reg) | |
13188 | { | |
13189 | s64 smin_val = src_reg->smin_value; | |
13190 | u64 umin_val = src_reg->umin_value; | |
13191 | u64 umax_val = src_reg->umax_value; | |
13192 | ||
07cd2631 JF |
13193 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
13194 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 13195 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
13196 | return; |
13197 | } | |
13198 | /* Both values are positive, so we can work with unsigned and | |
13199 | * copy the result to signed (unless it exceeds S64_MAX). | |
13200 | */ | |
13201 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
13202 | /* Potential overflow, we know nothing */ | |
3f50f132 | 13203 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
13204 | return; |
13205 | } | |
13206 | dst_reg->umin_value *= umin_val; | |
13207 | dst_reg->umax_value *= umax_val; | |
13208 | if (dst_reg->umax_value > S64_MAX) { | |
13209 | /* Overflow possible, we know nothing */ | |
13210 | dst_reg->smin_value = S64_MIN; | |
13211 | dst_reg->smax_value = S64_MAX; | |
13212 | } else { | |
13213 | dst_reg->smin_value = dst_reg->umin_value; | |
13214 | dst_reg->smax_value = dst_reg->umax_value; | |
13215 | } | |
13216 | } | |
13217 | ||
3f50f132 JF |
13218 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
13219 | struct bpf_reg_state *src_reg) | |
13220 | { | |
13221 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
13222 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
13223 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
13224 | s32 smin_val = src_reg->s32_min_value; | |
13225 | u32 umax_val = src_reg->u32_max_value; | |
13226 | ||
049c4e13 DB |
13227 | if (src_known && dst_known) { |
13228 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 13229 | return; |
049c4e13 | 13230 | } |
3f50f132 JF |
13231 | |
13232 | /* We get our minimum from the var_off, since that's inherently | |
13233 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
13234 | */ | |
13235 | dst_reg->u32_min_value = var32_off.value; | |
13236 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
13237 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
13238 | /* Lose signed bounds when ANDing negative numbers, | |
13239 | * ain't nobody got time for that. | |
13240 | */ | |
13241 | dst_reg->s32_min_value = S32_MIN; | |
13242 | dst_reg->s32_max_value = S32_MAX; | |
13243 | } else { | |
13244 | /* ANDing two positives gives a positive, so safe to | |
13245 | * cast result into s64. | |
13246 | */ | |
13247 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
13248 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
13249 | } | |
3f50f132 JF |
13250 | } |
13251 | ||
07cd2631 JF |
13252 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
13253 | struct bpf_reg_state *src_reg) | |
13254 | { | |
3f50f132 JF |
13255 | bool src_known = tnum_is_const(src_reg->var_off); |
13256 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
13257 | s64 smin_val = src_reg->smin_value; |
13258 | u64 umax_val = src_reg->umax_value; | |
13259 | ||
3f50f132 | 13260 | if (src_known && dst_known) { |
4fbb38a3 | 13261 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
13262 | return; |
13263 | } | |
13264 | ||
07cd2631 JF |
13265 | /* We get our minimum from the var_off, since that's inherently |
13266 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
13267 | */ | |
07cd2631 JF |
13268 | dst_reg->umin_value = dst_reg->var_off.value; |
13269 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
13270 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
13271 | /* Lose signed bounds when ANDing negative numbers, | |
13272 | * ain't nobody got time for that. | |
13273 | */ | |
13274 | dst_reg->smin_value = S64_MIN; | |
13275 | dst_reg->smax_value = S64_MAX; | |
13276 | } else { | |
13277 | /* ANDing two positives gives a positive, so safe to | |
13278 | * cast result into s64. | |
13279 | */ | |
13280 | dst_reg->smin_value = dst_reg->umin_value; | |
13281 | dst_reg->smax_value = dst_reg->umax_value; | |
13282 | } | |
13283 | /* We may learn something more from the var_off */ | |
13284 | __update_reg_bounds(dst_reg); | |
13285 | } | |
13286 | ||
3f50f132 JF |
13287 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
13288 | struct bpf_reg_state *src_reg) | |
13289 | { | |
13290 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
13291 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
13292 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5b9fbeb7 DB |
13293 | s32 smin_val = src_reg->s32_min_value; |
13294 | u32 umin_val = src_reg->u32_min_value; | |
3f50f132 | 13295 | |
049c4e13 DB |
13296 | if (src_known && dst_known) { |
13297 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 13298 | return; |
049c4e13 | 13299 | } |
3f50f132 JF |
13300 | |
13301 | /* We get our maximum from the var_off, and our minimum is the | |
13302 | * maximum of the operands' minima | |
13303 | */ | |
13304 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
13305 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
13306 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
13307 | /* Lose signed bounds when ORing negative numbers, | |
13308 | * ain't nobody got time for that. | |
13309 | */ | |
13310 | dst_reg->s32_min_value = S32_MIN; | |
13311 | dst_reg->s32_max_value = S32_MAX; | |
13312 | } else { | |
13313 | /* ORing two positives gives a positive, so safe to | |
13314 | * cast result into s64. | |
13315 | */ | |
5b9fbeb7 DB |
13316 | dst_reg->s32_min_value = dst_reg->u32_min_value; |
13317 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
3f50f132 JF |
13318 | } |
13319 | } | |
13320 | ||
07cd2631 JF |
13321 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
13322 | struct bpf_reg_state *src_reg) | |
13323 | { | |
3f50f132 JF |
13324 | bool src_known = tnum_is_const(src_reg->var_off); |
13325 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
13326 | s64 smin_val = src_reg->smin_value; |
13327 | u64 umin_val = src_reg->umin_value; | |
13328 | ||
3f50f132 | 13329 | if (src_known && dst_known) { |
4fbb38a3 | 13330 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
13331 | return; |
13332 | } | |
13333 | ||
07cd2631 JF |
13334 | /* We get our maximum from the var_off, and our minimum is the |
13335 | * maximum of the operands' minima | |
13336 | */ | |
07cd2631 JF |
13337 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
13338 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
13339 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
13340 | /* Lose signed bounds when ORing negative numbers, | |
13341 | * ain't nobody got time for that. | |
13342 | */ | |
13343 | dst_reg->smin_value = S64_MIN; | |
13344 | dst_reg->smax_value = S64_MAX; | |
13345 | } else { | |
13346 | /* ORing two positives gives a positive, so safe to | |
13347 | * cast result into s64. | |
13348 | */ | |
13349 | dst_reg->smin_value = dst_reg->umin_value; | |
13350 | dst_reg->smax_value = dst_reg->umax_value; | |
13351 | } | |
13352 | /* We may learn something more from the var_off */ | |
13353 | __update_reg_bounds(dst_reg); | |
13354 | } | |
13355 | ||
2921c90d YS |
13356 | static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg, |
13357 | struct bpf_reg_state *src_reg) | |
13358 | { | |
13359 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
13360 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
13361 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
13362 | s32 smin_val = src_reg->s32_min_value; | |
13363 | ||
049c4e13 DB |
13364 | if (src_known && dst_known) { |
13365 | __mark_reg32_known(dst_reg, var32_off.value); | |
2921c90d | 13366 | return; |
049c4e13 | 13367 | } |
2921c90d YS |
13368 | |
13369 | /* We get both minimum and maximum from the var32_off. */ | |
13370 | dst_reg->u32_min_value = var32_off.value; | |
13371 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
13372 | ||
13373 | if (dst_reg->s32_min_value >= 0 && smin_val >= 0) { | |
13374 | /* XORing two positive sign numbers gives a positive, | |
13375 | * so safe to cast u32 result into s32. | |
13376 | */ | |
13377 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
13378 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
13379 | } else { | |
13380 | dst_reg->s32_min_value = S32_MIN; | |
13381 | dst_reg->s32_max_value = S32_MAX; | |
13382 | } | |
13383 | } | |
13384 | ||
13385 | static void scalar_min_max_xor(struct bpf_reg_state *dst_reg, | |
13386 | struct bpf_reg_state *src_reg) | |
13387 | { | |
13388 | bool src_known = tnum_is_const(src_reg->var_off); | |
13389 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
13390 | s64 smin_val = src_reg->smin_value; | |
13391 | ||
13392 | if (src_known && dst_known) { | |
13393 | /* dst_reg->var_off.value has been updated earlier */ | |
13394 | __mark_reg_known(dst_reg, dst_reg->var_off.value); | |
13395 | return; | |
13396 | } | |
13397 | ||
13398 | /* We get both minimum and maximum from the var_off. */ | |
13399 | dst_reg->umin_value = dst_reg->var_off.value; | |
13400 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
13401 | ||
13402 | if (dst_reg->smin_value >= 0 && smin_val >= 0) { | |
13403 | /* XORing two positive sign numbers gives a positive, | |
13404 | * so safe to cast u64 result into s64. | |
13405 | */ | |
13406 | dst_reg->smin_value = dst_reg->umin_value; | |
13407 | dst_reg->smax_value = dst_reg->umax_value; | |
13408 | } else { | |
13409 | dst_reg->smin_value = S64_MIN; | |
13410 | dst_reg->smax_value = S64_MAX; | |
13411 | } | |
13412 | ||
13413 | __update_reg_bounds(dst_reg); | |
13414 | } | |
13415 | ||
3f50f132 JF |
13416 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
13417 | u64 umin_val, u64 umax_val) | |
07cd2631 | 13418 | { |
07cd2631 JF |
13419 | /* We lose all sign bit information (except what we can pick |
13420 | * up from var_off) | |
13421 | */ | |
3f50f132 JF |
13422 | dst_reg->s32_min_value = S32_MIN; |
13423 | dst_reg->s32_max_value = S32_MAX; | |
13424 | /* If we might shift our top bit out, then we know nothing */ | |
13425 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
13426 | dst_reg->u32_min_value = 0; | |
13427 | dst_reg->u32_max_value = U32_MAX; | |
13428 | } else { | |
13429 | dst_reg->u32_min_value <<= umin_val; | |
13430 | dst_reg->u32_max_value <<= umax_val; | |
13431 | } | |
13432 | } | |
13433 | ||
13434 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
13435 | struct bpf_reg_state *src_reg) | |
13436 | { | |
13437 | u32 umax_val = src_reg->u32_max_value; | |
13438 | u32 umin_val = src_reg->u32_min_value; | |
13439 | /* u32 alu operation will zext upper bits */ | |
13440 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
13441 | ||
13442 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
13443 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
13444 | /* Not required but being careful mark reg64 bounds as unknown so | |
13445 | * that we are forced to pick them up from tnum and zext later and | |
13446 | * if some path skips this step we are still safe. | |
13447 | */ | |
13448 | __mark_reg64_unbounded(dst_reg); | |
13449 | __update_reg32_bounds(dst_reg); | |
13450 | } | |
13451 | ||
13452 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
13453 | u64 umin_val, u64 umax_val) | |
13454 | { | |
13455 | /* Special case <<32 because it is a common compiler pattern to sign | |
13456 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
13457 | * positive we know this shift will also be positive so we can track | |
13458 | * bounds correctly. Otherwise we lose all sign bit information except | |
13459 | * what we can pick up from var_off. Perhaps we can generalize this | |
13460 | * later to shifts of any length. | |
13461 | */ | |
13462 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
13463 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
13464 | else | |
13465 | dst_reg->smax_value = S64_MAX; | |
13466 | ||
13467 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
13468 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
13469 | else | |
13470 | dst_reg->smin_value = S64_MIN; | |
13471 | ||
07cd2631 JF |
13472 | /* If we might shift our top bit out, then we know nothing */ |
13473 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
13474 | dst_reg->umin_value = 0; | |
13475 | dst_reg->umax_value = U64_MAX; | |
13476 | } else { | |
13477 | dst_reg->umin_value <<= umin_val; | |
13478 | dst_reg->umax_value <<= umax_val; | |
13479 | } | |
3f50f132 JF |
13480 | } |
13481 | ||
13482 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
13483 | struct bpf_reg_state *src_reg) | |
13484 | { | |
13485 | u64 umax_val = src_reg->umax_value; | |
13486 | u64 umin_val = src_reg->umin_value; | |
13487 | ||
13488 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
13489 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
13490 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
13491 | ||
07cd2631 JF |
13492 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
13493 | /* We may learn something more from the var_off */ | |
13494 | __update_reg_bounds(dst_reg); | |
13495 | } | |
13496 | ||
3f50f132 JF |
13497 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
13498 | struct bpf_reg_state *src_reg) | |
13499 | { | |
13500 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
13501 | u32 umax_val = src_reg->u32_max_value; | |
13502 | u32 umin_val = src_reg->u32_min_value; | |
13503 | ||
13504 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
13505 | * be negative, then either: | |
13506 | * 1) src_reg might be zero, so the sign bit of the result is | |
13507 | * unknown, so we lose our signed bounds | |
13508 | * 2) it's known negative, thus the unsigned bounds capture the | |
13509 | * signed bounds | |
13510 | * 3) the signed bounds cross zero, so they tell us nothing | |
13511 | * about the result | |
13512 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 13513 | * unsigned bounds capture the signed bounds. |
3f50f132 JF |
13514 | * Thus, in all cases it suffices to blow away our signed bounds |
13515 | * and rely on inferring new ones from the unsigned bounds and | |
13516 | * var_off of the result. | |
13517 | */ | |
13518 | dst_reg->s32_min_value = S32_MIN; | |
13519 | dst_reg->s32_max_value = S32_MAX; | |
13520 | ||
13521 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
13522 | dst_reg->u32_min_value >>= umax_val; | |
13523 | dst_reg->u32_max_value >>= umin_val; | |
13524 | ||
13525 | __mark_reg64_unbounded(dst_reg); | |
13526 | __update_reg32_bounds(dst_reg); | |
13527 | } | |
13528 | ||
07cd2631 JF |
13529 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
13530 | struct bpf_reg_state *src_reg) | |
13531 | { | |
13532 | u64 umax_val = src_reg->umax_value; | |
13533 | u64 umin_val = src_reg->umin_value; | |
13534 | ||
13535 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
13536 | * be negative, then either: | |
13537 | * 1) src_reg might be zero, so the sign bit of the result is | |
13538 | * unknown, so we lose our signed bounds | |
13539 | * 2) it's known negative, thus the unsigned bounds capture the | |
13540 | * signed bounds | |
13541 | * 3) the signed bounds cross zero, so they tell us nothing | |
13542 | * about the result | |
13543 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 13544 | * unsigned bounds capture the signed bounds. |
07cd2631 JF |
13545 | * Thus, in all cases it suffices to blow away our signed bounds |
13546 | * and rely on inferring new ones from the unsigned bounds and | |
13547 | * var_off of the result. | |
13548 | */ | |
13549 | dst_reg->smin_value = S64_MIN; | |
13550 | dst_reg->smax_value = S64_MAX; | |
13551 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
13552 | dst_reg->umin_value >>= umax_val; | |
13553 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
13554 | |
13555 | /* Its not easy to operate on alu32 bounds here because it depends | |
13556 | * on bits being shifted in. Take easy way out and mark unbounded | |
13557 | * so we can recalculate later from tnum. | |
13558 | */ | |
13559 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
13560 | __update_reg_bounds(dst_reg); |
13561 | } | |
13562 | ||
3f50f132 JF |
13563 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
13564 | struct bpf_reg_state *src_reg) | |
07cd2631 | 13565 | { |
3f50f132 | 13566 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
13567 | |
13568 | /* Upon reaching here, src_known is true and | |
13569 | * umax_val is equal to umin_val. | |
13570 | */ | |
3f50f132 JF |
13571 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
13572 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 13573 | |
3f50f132 JF |
13574 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
13575 | ||
13576 | /* blow away the dst_reg umin_value/umax_value and rely on | |
13577 | * dst_reg var_off to refine the result. | |
13578 | */ | |
13579 | dst_reg->u32_min_value = 0; | |
13580 | dst_reg->u32_max_value = U32_MAX; | |
13581 | ||
13582 | __mark_reg64_unbounded(dst_reg); | |
13583 | __update_reg32_bounds(dst_reg); | |
13584 | } | |
13585 | ||
13586 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
13587 | struct bpf_reg_state *src_reg) | |
13588 | { | |
13589 | u64 umin_val = src_reg->umin_value; | |
13590 | ||
13591 | /* Upon reaching here, src_known is true and umax_val is equal | |
13592 | * to umin_val. | |
13593 | */ | |
13594 | dst_reg->smin_value >>= umin_val; | |
13595 | dst_reg->smax_value >>= umin_val; | |
13596 | ||
13597 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
13598 | |
13599 | /* blow away the dst_reg umin_value/umax_value and rely on | |
13600 | * dst_reg var_off to refine the result. | |
13601 | */ | |
13602 | dst_reg->umin_value = 0; | |
13603 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
13604 | |
13605 | /* Its not easy to operate on alu32 bounds here because it depends | |
13606 | * on bits being shifted in from upper 32-bits. Take easy way out | |
13607 | * and mark unbounded so we can recalculate later from tnum. | |
13608 | */ | |
13609 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
13610 | __update_reg_bounds(dst_reg); |
13611 | } | |
13612 | ||
468f6eaf JH |
13613 | /* WARNING: This function does calculations on 64-bit values, but the actual |
13614 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
13615 | * need extra checks in the 32-bit case. | |
13616 | */ | |
f1174f77 EC |
13617 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
13618 | struct bpf_insn *insn, | |
13619 | struct bpf_reg_state *dst_reg, | |
13620 | struct bpf_reg_state src_reg) | |
969bf05e | 13621 | { |
638f5b90 | 13622 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 13623 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 13624 | bool src_known; |
b03c9f9f EC |
13625 | s64 smin_val, smax_val; |
13626 | u64 umin_val, umax_val; | |
3f50f132 JF |
13627 | s32 s32_min_val, s32_max_val; |
13628 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 13629 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
3f50f132 | 13630 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
a6aaece0 | 13631 | int ret; |
b799207e | 13632 | |
b03c9f9f EC |
13633 | smin_val = src_reg.smin_value; |
13634 | smax_val = src_reg.smax_value; | |
13635 | umin_val = src_reg.umin_value; | |
13636 | umax_val = src_reg.umax_value; | |
f23cc643 | 13637 | |
3f50f132 JF |
13638 | s32_min_val = src_reg.s32_min_value; |
13639 | s32_max_val = src_reg.s32_max_value; | |
13640 | u32_min_val = src_reg.u32_min_value; | |
13641 | u32_max_val = src_reg.u32_max_value; | |
13642 | ||
13643 | if (alu32) { | |
13644 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
13645 | if ((src_known && |
13646 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
13647 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
13648 | /* Taint dst register if offset had invalid bounds | |
13649 | * derived from e.g. dead branches. | |
13650 | */ | |
13651 | __mark_reg_unknown(env, dst_reg); | |
13652 | return 0; | |
13653 | } | |
13654 | } else { | |
13655 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
13656 | if ((src_known && |
13657 | (smin_val != smax_val || umin_val != umax_val)) || | |
13658 | smin_val > smax_val || umin_val > umax_val) { | |
13659 | /* Taint dst register if offset had invalid bounds | |
13660 | * derived from e.g. dead branches. | |
13661 | */ | |
13662 | __mark_reg_unknown(env, dst_reg); | |
13663 | return 0; | |
13664 | } | |
6f16101e DB |
13665 | } |
13666 | ||
bb7f0f98 AS |
13667 | if (!src_known && |
13668 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 13669 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
13670 | return 0; |
13671 | } | |
13672 | ||
f5288193 DB |
13673 | if (sanitize_needed(opcode)) { |
13674 | ret = sanitize_val_alu(env, insn); | |
13675 | if (ret < 0) | |
13676 | return sanitize_err(env, insn, ret, NULL, NULL); | |
13677 | } | |
13678 | ||
3f50f132 JF |
13679 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
13680 | * There are two classes of instructions: The first class we track both | |
13681 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
13682 | * greatest amount of precision when alu operations are mixed with jmp32 | |
13683 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
13684 | * and BPF_OR. This is possible because these ops have fairly easy to | |
13685 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
13686 | * See alu32 verifier tests for examples. The second class of | |
13687 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
13688 | * with regards to tracking sign/unsigned bounds because the bits may | |
13689 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
13690 | * the reg unbounded in the subreg bound space and use the resulting | |
13691 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
13692 | */ | |
48461135 JB |
13693 | switch (opcode) { |
13694 | case BPF_ADD: | |
3f50f132 | 13695 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 13696 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 13697 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
13698 | break; |
13699 | case BPF_SUB: | |
3f50f132 | 13700 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 13701 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 13702 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
13703 | break; |
13704 | case BPF_MUL: | |
3f50f132 JF |
13705 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
13706 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 13707 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
13708 | break; |
13709 | case BPF_AND: | |
3f50f132 JF |
13710 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
13711 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 13712 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
13713 | break; |
13714 | case BPF_OR: | |
3f50f132 JF |
13715 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
13716 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 13717 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 | 13718 | break; |
2921c90d YS |
13719 | case BPF_XOR: |
13720 | dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off); | |
13721 | scalar32_min_max_xor(dst_reg, &src_reg); | |
13722 | scalar_min_max_xor(dst_reg, &src_reg); | |
13723 | break; | |
48461135 | 13724 | case BPF_LSH: |
468f6eaf JH |
13725 | if (umax_val >= insn_bitness) { |
13726 | /* Shifts greater than 31 or 63 are undefined. | |
13727 | * This includes shifts by a negative number. | |
b03c9f9f | 13728 | */ |
61bd5218 | 13729 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
13730 | break; |
13731 | } | |
3f50f132 JF |
13732 | if (alu32) |
13733 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
13734 | else | |
13735 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
13736 | break; |
13737 | case BPF_RSH: | |
468f6eaf JH |
13738 | if (umax_val >= insn_bitness) { |
13739 | /* Shifts greater than 31 or 63 are undefined. | |
13740 | * This includes shifts by a negative number. | |
b03c9f9f | 13741 | */ |
61bd5218 | 13742 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
13743 | break; |
13744 | } | |
3f50f132 JF |
13745 | if (alu32) |
13746 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
13747 | else | |
13748 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 13749 | break; |
9cbe1f5a YS |
13750 | case BPF_ARSH: |
13751 | if (umax_val >= insn_bitness) { | |
13752 | /* Shifts greater than 31 or 63 are undefined. | |
13753 | * This includes shifts by a negative number. | |
13754 | */ | |
13755 | mark_reg_unknown(env, regs, insn->dst_reg); | |
13756 | break; | |
13757 | } | |
3f50f132 JF |
13758 | if (alu32) |
13759 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
13760 | else | |
13761 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 13762 | break; |
48461135 | 13763 | default: |
61bd5218 | 13764 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
13765 | break; |
13766 | } | |
13767 | ||
3f50f132 JF |
13768 | /* ALU32 ops are zero extended into 64bit register */ |
13769 | if (alu32) | |
13770 | zext_32_to_64(dst_reg); | |
3844d153 | 13771 | reg_bounds_sync(dst_reg); |
f1174f77 EC |
13772 | return 0; |
13773 | } | |
13774 | ||
13775 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
13776 | * and var_off. | |
13777 | */ | |
13778 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
13779 | struct bpf_insn *insn) | |
13780 | { | |
f4d7e40a AS |
13781 | struct bpf_verifier_state *vstate = env->cur_state; |
13782 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
13783 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
13784 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
13785 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 13786 | int err; |
f1174f77 EC |
13787 | |
13788 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
13789 | src_reg = NULL; |
13790 | if (dst_reg->type != SCALAR_VALUE) | |
13791 | ptr_reg = dst_reg; | |
75748837 AS |
13792 | else |
13793 | /* Make sure ID is cleared otherwise dst_reg min/max could be | |
13794 | * incorrectly propagated into other registers by find_equal_scalars() | |
13795 | */ | |
13796 | dst_reg->id = 0; | |
f1174f77 EC |
13797 | if (BPF_SRC(insn->code) == BPF_X) { |
13798 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
13799 | if (src_reg->type != SCALAR_VALUE) { |
13800 | if (dst_reg->type != SCALAR_VALUE) { | |
13801 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
13802 | * an arbitrary scalar. Disallow all math except |
13803 | * pointer subtraction | |
f1174f77 | 13804 | */ |
dd066823 | 13805 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
13806 | mark_reg_unknown(env, regs, insn->dst_reg); |
13807 | return 0; | |
f1174f77 | 13808 | } |
82abbf8d AS |
13809 | verbose(env, "R%d pointer %s pointer prohibited\n", |
13810 | insn->dst_reg, | |
13811 | bpf_alu_string[opcode >> 4]); | |
13812 | return -EACCES; | |
f1174f77 EC |
13813 | } else { |
13814 | /* scalar += pointer | |
13815 | * This is legal, but we have to reverse our | |
13816 | * src/dest handling in computing the range | |
13817 | */ | |
b5dc0163 AS |
13818 | err = mark_chain_precision(env, insn->dst_reg); |
13819 | if (err) | |
13820 | return err; | |
82abbf8d AS |
13821 | return adjust_ptr_min_max_vals(env, insn, |
13822 | src_reg, dst_reg); | |
f1174f77 EC |
13823 | } |
13824 | } else if (ptr_reg) { | |
13825 | /* pointer += scalar */ | |
b5dc0163 AS |
13826 | err = mark_chain_precision(env, insn->src_reg); |
13827 | if (err) | |
13828 | return err; | |
82abbf8d AS |
13829 | return adjust_ptr_min_max_vals(env, insn, |
13830 | dst_reg, src_reg); | |
a3b666bf AN |
13831 | } else if (dst_reg->precise) { |
13832 | /* if dst_reg is precise, src_reg should be precise as well */ | |
13833 | err = mark_chain_precision(env, insn->src_reg); | |
13834 | if (err) | |
13835 | return err; | |
f1174f77 EC |
13836 | } |
13837 | } else { | |
13838 | /* Pretend the src is a reg with a known value, since we only | |
13839 | * need to be able to read from this state. | |
13840 | */ | |
13841 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 13842 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 13843 | src_reg = &off_reg; |
82abbf8d AS |
13844 | if (ptr_reg) /* pointer += K */ |
13845 | return adjust_ptr_min_max_vals(env, insn, | |
13846 | ptr_reg, src_reg); | |
f1174f77 EC |
13847 | } |
13848 | ||
13849 | /* Got here implies adding two SCALAR_VALUEs */ | |
13850 | if (WARN_ON_ONCE(ptr_reg)) { | |
0f55f9ed | 13851 | print_verifier_state(env, state, true); |
61bd5218 | 13852 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
13853 | return -EINVAL; |
13854 | } | |
13855 | if (WARN_ON(!src_reg)) { | |
0f55f9ed | 13856 | print_verifier_state(env, state, true); |
61bd5218 | 13857 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
13858 | return -EINVAL; |
13859 | } | |
13860 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
13861 | } |
13862 | ||
17a52670 | 13863 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 13864 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 13865 | { |
638f5b90 | 13866 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
13867 | u8 opcode = BPF_OP(insn->code); |
13868 | int err; | |
13869 | ||
13870 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
13871 | if (opcode == BPF_NEG) { | |
395e942d | 13872 | if (BPF_SRC(insn->code) != BPF_K || |
17a52670 AS |
13873 | insn->src_reg != BPF_REG_0 || |
13874 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 13875 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
13876 | return -EINVAL; |
13877 | } | |
13878 | } else { | |
13879 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 | 13880 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
0845c3db YS |
13881 | (BPF_CLASS(insn->code) == BPF_ALU64 && |
13882 | BPF_SRC(insn->code) != BPF_TO_LE)) { | |
61bd5218 | 13883 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
13884 | return -EINVAL; |
13885 | } | |
13886 | } | |
13887 | ||
13888 | /* check src operand */ | |
dc503a8a | 13889 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
13890 | if (err) |
13891 | return err; | |
13892 | ||
1be7f75d | 13893 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 13894 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
13895 | insn->dst_reg); |
13896 | return -EACCES; | |
13897 | } | |
13898 | ||
17a52670 | 13899 | /* check dest operand */ |
dc503a8a | 13900 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
13901 | if (err) |
13902 | return err; | |
13903 | ||
13904 | } else if (opcode == BPF_MOV) { | |
13905 | ||
13906 | if (BPF_SRC(insn->code) == BPF_X) { | |
8100928c | 13907 | if (insn->imm != 0) { |
61bd5218 | 13908 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
13909 | return -EINVAL; |
13910 | } | |
13911 | ||
8100928c YS |
13912 | if (BPF_CLASS(insn->code) == BPF_ALU) { |
13913 | if (insn->off != 0 && insn->off != 8 && insn->off != 16) { | |
13914 | verbose(env, "BPF_MOV uses reserved fields\n"); | |
13915 | return -EINVAL; | |
13916 | } | |
13917 | } else { | |
13918 | if (insn->off != 0 && insn->off != 8 && insn->off != 16 && | |
13919 | insn->off != 32) { | |
13920 | verbose(env, "BPF_MOV uses reserved fields\n"); | |
13921 | return -EINVAL; | |
13922 | } | |
13923 | } | |
13924 | ||
17a52670 | 13925 | /* check src operand */ |
dc503a8a | 13926 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
13927 | if (err) |
13928 | return err; | |
13929 | } else { | |
13930 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 13931 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
13932 | return -EINVAL; |
13933 | } | |
13934 | } | |
13935 | ||
fbeb1603 AF |
13936 | /* check dest operand, mark as required later */ |
13937 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
13938 | if (err) |
13939 | return err; | |
13940 | ||
13941 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
13942 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
13943 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
1ffc85d9 EZ |
13944 | bool need_id = src_reg->type == SCALAR_VALUE && !src_reg->id && |
13945 | !tnum_is_const(src_reg->var_off); | |
e434b8cd | 13946 | |
17a52670 | 13947 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
8100928c YS |
13948 | if (insn->off == 0) { |
13949 | /* case: R1 = R2 | |
13950 | * copy register state to dest reg | |
75748837 | 13951 | */ |
8100928c YS |
13952 | if (need_id) |
13953 | /* Assign src and dst registers the same ID | |
13954 | * that will be used by find_equal_scalars() | |
13955 | * to propagate min/max range. | |
13956 | */ | |
13957 | src_reg->id = ++env->id_gen; | |
13958 | copy_register_state(dst_reg, src_reg); | |
13959 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13960 | dst_reg->subreg_def = DEF_NOT_SUBREG; | |
13961 | } else { | |
13962 | /* case: R1 = (s8, s16 s32)R2 */ | |
db2baf82 YS |
13963 | if (is_pointer_value(env, insn->src_reg)) { |
13964 | verbose(env, | |
13965 | "R%d sign-extension part of pointer\n", | |
13966 | insn->src_reg); | |
13967 | return -EACCES; | |
13968 | } else if (src_reg->type == SCALAR_VALUE) { | |
13969 | bool no_sext; | |
13970 | ||
13971 | no_sext = src_reg->umax_value < (1ULL << (insn->off - 1)); | |
13972 | if (no_sext && need_id) | |
13973 | src_reg->id = ++env->id_gen; | |
13974 | copy_register_state(dst_reg, src_reg); | |
13975 | if (!no_sext) | |
13976 | dst_reg->id = 0; | |
13977 | coerce_reg_to_size_sx(dst_reg, insn->off >> 3); | |
13978 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13979 | dst_reg->subreg_def = DEF_NOT_SUBREG; | |
13980 | } else { | |
13981 | mark_reg_unknown(env, regs, insn->dst_reg); | |
13982 | } | |
8100928c | 13983 | } |
17a52670 | 13984 | } else { |
f1174f77 | 13985 | /* R1 = (u32) R2 */ |
1be7f75d | 13986 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
13987 | verbose(env, |
13988 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
13989 | insn->src_reg); |
13990 | return -EACCES; | |
e434b8cd | 13991 | } else if (src_reg->type == SCALAR_VALUE) { |
8100928c YS |
13992 | if (insn->off == 0) { |
13993 | bool is_src_reg_u32 = src_reg->umax_value <= U32_MAX; | |
13994 | ||
13995 | if (is_src_reg_u32 && need_id) | |
13996 | src_reg->id = ++env->id_gen; | |
13997 | copy_register_state(dst_reg, src_reg); | |
13998 | /* Make sure ID is cleared if src_reg is not in u32 | |
13999 | * range otherwise dst_reg min/max could be incorrectly | |
14000 | * propagated into src_reg by find_equal_scalars() | |
14001 | */ | |
14002 | if (!is_src_reg_u32) | |
14003 | dst_reg->id = 0; | |
14004 | dst_reg->live |= REG_LIVE_WRITTEN; | |
14005 | dst_reg->subreg_def = env->insn_idx + 1; | |
14006 | } else { | |
14007 | /* case: W1 = (s8, s16)W2 */ | |
14008 | bool no_sext = src_reg->umax_value < (1ULL << (insn->off - 1)); | |
14009 | ||
14010 | if (no_sext && need_id) | |
14011 | src_reg->id = ++env->id_gen; | |
14012 | copy_register_state(dst_reg, src_reg); | |
14013 | if (!no_sext) | |
14014 | dst_reg->id = 0; | |
14015 | dst_reg->live |= REG_LIVE_WRITTEN; | |
14016 | dst_reg->subreg_def = env->insn_idx + 1; | |
14017 | coerce_subreg_to_size_sx(dst_reg, insn->off >> 3); | |
14018 | } | |
e434b8cd JW |
14019 | } else { |
14020 | mark_reg_unknown(env, regs, | |
14021 | insn->dst_reg); | |
1be7f75d | 14022 | } |
3f50f132 | 14023 | zext_32_to_64(dst_reg); |
3844d153 | 14024 | reg_bounds_sync(dst_reg); |
17a52670 AS |
14025 | } |
14026 | } else { | |
14027 | /* case: R = imm | |
14028 | * remember the value we stored into this reg | |
14029 | */ | |
fbeb1603 AF |
14030 | /* clear any state __mark_reg_known doesn't set */ |
14031 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 14032 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
14033 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
14034 | __mark_reg_known(regs + insn->dst_reg, | |
14035 | insn->imm); | |
14036 | } else { | |
14037 | __mark_reg_known(regs + insn->dst_reg, | |
14038 | (u32)insn->imm); | |
14039 | } | |
17a52670 AS |
14040 | } |
14041 | ||
14042 | } else if (opcode > BPF_END) { | |
61bd5218 | 14043 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
14044 | return -EINVAL; |
14045 | ||
14046 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
14047 | ||
17a52670 | 14048 | if (BPF_SRC(insn->code) == BPF_X) { |
ec0e2da9 YS |
14049 | if (insn->imm != 0 || insn->off > 1 || |
14050 | (insn->off == 1 && opcode != BPF_MOD && opcode != BPF_DIV)) { | |
61bd5218 | 14051 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
14052 | return -EINVAL; |
14053 | } | |
14054 | /* check src1 operand */ | |
dc503a8a | 14055 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
14056 | if (err) |
14057 | return err; | |
14058 | } else { | |
ec0e2da9 YS |
14059 | if (insn->src_reg != BPF_REG_0 || insn->off > 1 || |
14060 | (insn->off == 1 && opcode != BPF_MOD && opcode != BPF_DIV)) { | |
61bd5218 | 14061 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
14062 | return -EINVAL; |
14063 | } | |
14064 | } | |
14065 | ||
14066 | /* check src2 operand */ | |
dc503a8a | 14067 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
14068 | if (err) |
14069 | return err; | |
14070 | ||
14071 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
14072 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 14073 | verbose(env, "div by zero\n"); |
17a52670 AS |
14074 | return -EINVAL; |
14075 | } | |
14076 | ||
229394e8 RV |
14077 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
14078 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
14079 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
14080 | ||
14081 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 14082 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
14083 | return -EINVAL; |
14084 | } | |
14085 | } | |
14086 | ||
1a0dc1ac | 14087 | /* check dest operand */ |
dc503a8a | 14088 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
14089 | if (err) |
14090 | return err; | |
14091 | ||
f1174f77 | 14092 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
14093 | } |
14094 | ||
14095 | return 0; | |
14096 | } | |
14097 | ||
f4d7e40a | 14098 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 14099 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 14100 | enum bpf_reg_type type, |
fb2a311a | 14101 | bool range_right_open) |
969bf05e | 14102 | { |
b239da34 KKD |
14103 | struct bpf_func_state *state; |
14104 | struct bpf_reg_state *reg; | |
14105 | int new_range; | |
2d2be8ca | 14106 | |
fb2a311a DB |
14107 | if (dst_reg->off < 0 || |
14108 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
14109 | /* This doesn't give us any range */ |
14110 | return; | |
14111 | ||
b03c9f9f EC |
14112 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
14113 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
14114 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
14115 | * than pkt_end, but that's because it's also less than pkt. | |
14116 | */ | |
14117 | return; | |
14118 | ||
fb2a311a DB |
14119 | new_range = dst_reg->off; |
14120 | if (range_right_open) | |
2fa7d94a | 14121 | new_range++; |
fb2a311a DB |
14122 | |
14123 | /* Examples for register markings: | |
2d2be8ca | 14124 | * |
fb2a311a | 14125 | * pkt_data in dst register: |
2d2be8ca DB |
14126 | * |
14127 | * r2 = r3; | |
14128 | * r2 += 8; | |
14129 | * if (r2 > pkt_end) goto <handle exception> | |
14130 | * <access okay> | |
14131 | * | |
b4e432f1 DB |
14132 | * r2 = r3; |
14133 | * r2 += 8; | |
14134 | * if (r2 < pkt_end) goto <access okay> | |
14135 | * <handle exception> | |
14136 | * | |
2d2be8ca DB |
14137 | * Where: |
14138 | * r2 == dst_reg, pkt_end == src_reg | |
14139 | * r2=pkt(id=n,off=8,r=0) | |
14140 | * r3=pkt(id=n,off=0,r=0) | |
14141 | * | |
fb2a311a | 14142 | * pkt_data in src register: |
2d2be8ca DB |
14143 | * |
14144 | * r2 = r3; | |
14145 | * r2 += 8; | |
14146 | * if (pkt_end >= r2) goto <access okay> | |
14147 | * <handle exception> | |
14148 | * | |
b4e432f1 DB |
14149 | * r2 = r3; |
14150 | * r2 += 8; | |
14151 | * if (pkt_end <= r2) goto <handle exception> | |
14152 | * <access okay> | |
14153 | * | |
2d2be8ca DB |
14154 | * Where: |
14155 | * pkt_end == dst_reg, r2 == src_reg | |
14156 | * r2=pkt(id=n,off=8,r=0) | |
14157 | * r3=pkt(id=n,off=0,r=0) | |
14158 | * | |
14159 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
14160 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
14161 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
14162 | * the check. | |
969bf05e | 14163 | */ |
2d2be8ca | 14164 | |
f1174f77 EC |
14165 | /* If our ids match, then we must have the same max_value. And we |
14166 | * don't care about the other reg's fixed offset, since if it's too big | |
14167 | * the range won't allow anything. | |
14168 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
14169 | */ | |
b239da34 KKD |
14170 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
14171 | if (reg->type == type && reg->id == dst_reg->id) | |
14172 | /* keep the maximum range already checked */ | |
14173 | reg->range = max(reg->range, new_range); | |
14174 | })); | |
969bf05e AS |
14175 | } |
14176 | ||
3f50f132 | 14177 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 14178 | { |
3f50f132 JF |
14179 | struct tnum subreg = tnum_subreg(reg->var_off); |
14180 | s32 sval = (s32)val; | |
a72dafaf | 14181 | |
3f50f132 JF |
14182 | switch (opcode) { |
14183 | case BPF_JEQ: | |
14184 | if (tnum_is_const(subreg)) | |
14185 | return !!tnum_equals_const(subreg, val); | |
13fbcee5 YS |
14186 | else if (val < reg->u32_min_value || val > reg->u32_max_value) |
14187 | return 0; | |
42d31dd6 AN |
14188 | else if (sval < reg->s32_min_value || sval > reg->s32_max_value) |
14189 | return 0; | |
3f50f132 JF |
14190 | break; |
14191 | case BPF_JNE: | |
14192 | if (tnum_is_const(subreg)) | |
14193 | return !tnum_equals_const(subreg, val); | |
13fbcee5 YS |
14194 | else if (val < reg->u32_min_value || val > reg->u32_max_value) |
14195 | return 1; | |
42d31dd6 AN |
14196 | else if (sval < reg->s32_min_value || sval > reg->s32_max_value) |
14197 | return 1; | |
3f50f132 JF |
14198 | break; |
14199 | case BPF_JSET: | |
14200 | if ((~subreg.mask & subreg.value) & val) | |
14201 | return 1; | |
14202 | if (!((subreg.mask | subreg.value) & val)) | |
14203 | return 0; | |
14204 | break; | |
14205 | case BPF_JGT: | |
14206 | if (reg->u32_min_value > val) | |
14207 | return 1; | |
14208 | else if (reg->u32_max_value <= val) | |
14209 | return 0; | |
14210 | break; | |
14211 | case BPF_JSGT: | |
14212 | if (reg->s32_min_value > sval) | |
14213 | return 1; | |
ee114dd6 | 14214 | else if (reg->s32_max_value <= sval) |
3f50f132 JF |
14215 | return 0; |
14216 | break; | |
14217 | case BPF_JLT: | |
14218 | if (reg->u32_max_value < val) | |
14219 | return 1; | |
14220 | else if (reg->u32_min_value >= val) | |
14221 | return 0; | |
14222 | break; | |
14223 | case BPF_JSLT: | |
14224 | if (reg->s32_max_value < sval) | |
14225 | return 1; | |
14226 | else if (reg->s32_min_value >= sval) | |
14227 | return 0; | |
14228 | break; | |
14229 | case BPF_JGE: | |
14230 | if (reg->u32_min_value >= val) | |
14231 | return 1; | |
14232 | else if (reg->u32_max_value < val) | |
14233 | return 0; | |
14234 | break; | |
14235 | case BPF_JSGE: | |
14236 | if (reg->s32_min_value >= sval) | |
14237 | return 1; | |
14238 | else if (reg->s32_max_value < sval) | |
14239 | return 0; | |
14240 | break; | |
14241 | case BPF_JLE: | |
14242 | if (reg->u32_max_value <= val) | |
14243 | return 1; | |
14244 | else if (reg->u32_min_value > val) | |
14245 | return 0; | |
14246 | break; | |
14247 | case BPF_JSLE: | |
14248 | if (reg->s32_max_value <= sval) | |
14249 | return 1; | |
14250 | else if (reg->s32_min_value > sval) | |
14251 | return 0; | |
14252 | break; | |
14253 | } | |
4f7b3e82 | 14254 | |
3f50f132 JF |
14255 | return -1; |
14256 | } | |
092ed096 | 14257 | |
3f50f132 JF |
14258 | |
14259 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
14260 | { | |
14261 | s64 sval = (s64)val; | |
a72dafaf | 14262 | |
4f7b3e82 AS |
14263 | switch (opcode) { |
14264 | case BPF_JEQ: | |
14265 | if (tnum_is_const(reg->var_off)) | |
14266 | return !!tnum_equals_const(reg->var_off, val); | |
13fbcee5 YS |
14267 | else if (val < reg->umin_value || val > reg->umax_value) |
14268 | return 0; | |
42d31dd6 AN |
14269 | else if (sval < reg->smin_value || sval > reg->smax_value) |
14270 | return 0; | |
4f7b3e82 AS |
14271 | break; |
14272 | case BPF_JNE: | |
14273 | if (tnum_is_const(reg->var_off)) | |
14274 | return !tnum_equals_const(reg->var_off, val); | |
13fbcee5 YS |
14275 | else if (val < reg->umin_value || val > reg->umax_value) |
14276 | return 1; | |
42d31dd6 AN |
14277 | else if (sval < reg->smin_value || sval > reg->smax_value) |
14278 | return 1; | |
4f7b3e82 | 14279 | break; |
960ea056 JK |
14280 | case BPF_JSET: |
14281 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
14282 | return 1; | |
14283 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
14284 | return 0; | |
14285 | break; | |
4f7b3e82 AS |
14286 | case BPF_JGT: |
14287 | if (reg->umin_value > val) | |
14288 | return 1; | |
14289 | else if (reg->umax_value <= val) | |
14290 | return 0; | |
14291 | break; | |
14292 | case BPF_JSGT: | |
a72dafaf | 14293 | if (reg->smin_value > sval) |
4f7b3e82 | 14294 | return 1; |
ee114dd6 | 14295 | else if (reg->smax_value <= sval) |
4f7b3e82 AS |
14296 | return 0; |
14297 | break; | |
14298 | case BPF_JLT: | |
14299 | if (reg->umax_value < val) | |
14300 | return 1; | |
14301 | else if (reg->umin_value >= val) | |
14302 | return 0; | |
14303 | break; | |
14304 | case BPF_JSLT: | |
a72dafaf | 14305 | if (reg->smax_value < sval) |
4f7b3e82 | 14306 | return 1; |
a72dafaf | 14307 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
14308 | return 0; |
14309 | break; | |
14310 | case BPF_JGE: | |
14311 | if (reg->umin_value >= val) | |
14312 | return 1; | |
14313 | else if (reg->umax_value < val) | |
14314 | return 0; | |
14315 | break; | |
14316 | case BPF_JSGE: | |
a72dafaf | 14317 | if (reg->smin_value >= sval) |
4f7b3e82 | 14318 | return 1; |
a72dafaf | 14319 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
14320 | return 0; |
14321 | break; | |
14322 | case BPF_JLE: | |
14323 | if (reg->umax_value <= val) | |
14324 | return 1; | |
14325 | else if (reg->umin_value > val) | |
14326 | return 0; | |
14327 | break; | |
14328 | case BPF_JSLE: | |
a72dafaf | 14329 | if (reg->smax_value <= sval) |
4f7b3e82 | 14330 | return 1; |
a72dafaf | 14331 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
14332 | return 0; |
14333 | break; | |
14334 | } | |
14335 | ||
14336 | return -1; | |
14337 | } | |
14338 | ||
3f50f132 JF |
14339 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
14340 | * and return: | |
14341 | * 1 - branch will be taken and "goto target" will be executed | |
14342 | * 0 - branch will not be taken and fall-through to next insn | |
14343 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
14344 | * range [0,10] | |
604dca5e | 14345 | */ |
3f50f132 JF |
14346 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
14347 | bool is_jmp32) | |
604dca5e | 14348 | { |
cac616db | 14349 | if (__is_pointer_value(false, reg)) { |
51302c95 | 14350 | if (!reg_not_null(reg)) |
cac616db JF |
14351 | return -1; |
14352 | ||
14353 | /* If pointer is valid tests against zero will fail so we can | |
14354 | * use this to direct branch taken. | |
14355 | */ | |
14356 | if (val != 0) | |
14357 | return -1; | |
14358 | ||
14359 | switch (opcode) { | |
14360 | case BPF_JEQ: | |
14361 | return 0; | |
14362 | case BPF_JNE: | |
14363 | return 1; | |
14364 | default: | |
14365 | return -1; | |
14366 | } | |
14367 | } | |
604dca5e | 14368 | |
3f50f132 JF |
14369 | if (is_jmp32) |
14370 | return is_branch32_taken(reg, val, opcode); | |
14371 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
14372 | } |
14373 | ||
6d94e741 AS |
14374 | static int flip_opcode(u32 opcode) |
14375 | { | |
14376 | /* How can we transform "a <op> b" into "b <op> a"? */ | |
14377 | static const u8 opcode_flip[16] = { | |
14378 | /* these stay the same */ | |
14379 | [BPF_JEQ >> 4] = BPF_JEQ, | |
14380 | [BPF_JNE >> 4] = BPF_JNE, | |
14381 | [BPF_JSET >> 4] = BPF_JSET, | |
14382 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
14383 | [BPF_JGE >> 4] = BPF_JLE, | |
14384 | [BPF_JGT >> 4] = BPF_JLT, | |
14385 | [BPF_JLE >> 4] = BPF_JGE, | |
14386 | [BPF_JLT >> 4] = BPF_JGT, | |
14387 | [BPF_JSGE >> 4] = BPF_JSLE, | |
14388 | [BPF_JSGT >> 4] = BPF_JSLT, | |
14389 | [BPF_JSLE >> 4] = BPF_JSGE, | |
14390 | [BPF_JSLT >> 4] = BPF_JSGT | |
14391 | }; | |
14392 | return opcode_flip[opcode >> 4]; | |
14393 | } | |
14394 | ||
14395 | static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg, | |
14396 | struct bpf_reg_state *src_reg, | |
14397 | u8 opcode) | |
14398 | { | |
14399 | struct bpf_reg_state *pkt; | |
14400 | ||
14401 | if (src_reg->type == PTR_TO_PACKET_END) { | |
14402 | pkt = dst_reg; | |
14403 | } else if (dst_reg->type == PTR_TO_PACKET_END) { | |
14404 | pkt = src_reg; | |
14405 | opcode = flip_opcode(opcode); | |
14406 | } else { | |
14407 | return -1; | |
14408 | } | |
14409 | ||
14410 | if (pkt->range >= 0) | |
14411 | return -1; | |
14412 | ||
14413 | switch (opcode) { | |
14414 | case BPF_JLE: | |
14415 | /* pkt <= pkt_end */ | |
14416 | fallthrough; | |
14417 | case BPF_JGT: | |
14418 | /* pkt > pkt_end */ | |
14419 | if (pkt->range == BEYOND_PKT_END) | |
14420 | /* pkt has at last one extra byte beyond pkt_end */ | |
14421 | return opcode == BPF_JGT; | |
14422 | break; | |
14423 | case BPF_JLT: | |
14424 | /* pkt < pkt_end */ | |
14425 | fallthrough; | |
14426 | case BPF_JGE: | |
14427 | /* pkt >= pkt_end */ | |
14428 | if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END) | |
14429 | return opcode == BPF_JGE; | |
14430 | break; | |
14431 | } | |
14432 | return -1; | |
14433 | } | |
14434 | ||
48461135 JB |
14435 | /* Adjusts the register min/max values in the case that the dst_reg is the |
14436 | * variable register that we are working on, and src_reg is a constant or we're | |
14437 | * simply doing a BPF_K check. | |
f1174f77 | 14438 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
14439 | */ |
14440 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
14441 | struct bpf_reg_state *false_reg, |
14442 | u64 val, u32 val32, | |
092ed096 | 14443 | u8 opcode, bool is_jmp32) |
48461135 | 14444 | { |
3f50f132 JF |
14445 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
14446 | struct tnum false_64off = false_reg->var_off; | |
14447 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
14448 | struct tnum true_64off = true_reg->var_off; | |
14449 | s64 sval = (s64)val; | |
14450 | s32 sval32 = (s32)val32; | |
a72dafaf | 14451 | |
f1174f77 EC |
14452 | /* If the dst_reg is a pointer, we can't learn anything about its |
14453 | * variable offset from the compare (unless src_reg were a pointer into | |
14454 | * the same object, but we don't bother with that. | |
14455 | * Since false_reg and true_reg have the same type by construction, we | |
14456 | * only need to check one of them for pointerness. | |
14457 | */ | |
14458 | if (__is_pointer_value(false, false_reg)) | |
14459 | return; | |
4cabc5b1 | 14460 | |
48461135 | 14461 | switch (opcode) { |
a12ca627 DB |
14462 | /* JEQ/JNE comparison doesn't change the register equivalence. |
14463 | * | |
14464 | * r1 = r2; | |
14465 | * if (r1 == 42) goto label; | |
14466 | * ... | |
14467 | * label: // here both r1 and r2 are known to be 42. | |
14468 | * | |
14469 | * Hence when marking register as known preserve it's ID. | |
14470 | */ | |
48461135 | 14471 | case BPF_JEQ: |
a12ca627 DB |
14472 | if (is_jmp32) { |
14473 | __mark_reg32_known(true_reg, val32); | |
14474 | true_32off = tnum_subreg(true_reg->var_off); | |
14475 | } else { | |
14476 | ___mark_reg_known(true_reg, val); | |
14477 | true_64off = true_reg->var_off; | |
14478 | } | |
14479 | break; | |
48461135 | 14480 | case BPF_JNE: |
a12ca627 DB |
14481 | if (is_jmp32) { |
14482 | __mark_reg32_known(false_reg, val32); | |
14483 | false_32off = tnum_subreg(false_reg->var_off); | |
14484 | } else { | |
14485 | ___mark_reg_known(false_reg, val); | |
14486 | false_64off = false_reg->var_off; | |
14487 | } | |
48461135 | 14488 | break; |
960ea056 | 14489 | case BPF_JSET: |
3f50f132 JF |
14490 | if (is_jmp32) { |
14491 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
14492 | if (is_power_of_2(val32)) | |
14493 | true_32off = tnum_or(true_32off, | |
14494 | tnum_const(val32)); | |
14495 | } else { | |
14496 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
14497 | if (is_power_of_2(val)) | |
14498 | true_64off = tnum_or(true_64off, | |
14499 | tnum_const(val)); | |
14500 | } | |
960ea056 | 14501 | break; |
48461135 | 14502 | case BPF_JGE: |
a72dafaf JW |
14503 | case BPF_JGT: |
14504 | { | |
3f50f132 JF |
14505 | if (is_jmp32) { |
14506 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
14507 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
14508 | ||
14509 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
14510 | false_umax); | |
14511 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
14512 | true_umin); | |
14513 | } else { | |
14514 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
14515 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
14516 | ||
14517 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
14518 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
14519 | } | |
b03c9f9f | 14520 | break; |
a72dafaf | 14521 | } |
48461135 | 14522 | case BPF_JSGE: |
a72dafaf JW |
14523 | case BPF_JSGT: |
14524 | { | |
3f50f132 JF |
14525 | if (is_jmp32) { |
14526 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
14527 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 14528 | |
3f50f132 JF |
14529 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
14530 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
14531 | } else { | |
14532 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
14533 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
14534 | ||
14535 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
14536 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
14537 | } | |
48461135 | 14538 | break; |
a72dafaf | 14539 | } |
b4e432f1 | 14540 | case BPF_JLE: |
a72dafaf JW |
14541 | case BPF_JLT: |
14542 | { | |
3f50f132 JF |
14543 | if (is_jmp32) { |
14544 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
14545 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
14546 | ||
14547 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
14548 | false_umin); | |
14549 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
14550 | true_umax); | |
14551 | } else { | |
14552 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
14553 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
14554 | ||
14555 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
14556 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
14557 | } | |
b4e432f1 | 14558 | break; |
a72dafaf | 14559 | } |
b4e432f1 | 14560 | case BPF_JSLE: |
a72dafaf JW |
14561 | case BPF_JSLT: |
14562 | { | |
3f50f132 JF |
14563 | if (is_jmp32) { |
14564 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
14565 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 14566 | |
3f50f132 JF |
14567 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
14568 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
14569 | } else { | |
14570 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
14571 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
14572 | ||
14573 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
14574 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
14575 | } | |
b4e432f1 | 14576 | break; |
a72dafaf | 14577 | } |
48461135 | 14578 | default: |
0fc31b10 | 14579 | return; |
48461135 JB |
14580 | } |
14581 | ||
3f50f132 JF |
14582 | if (is_jmp32) { |
14583 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
14584 | tnum_subreg(false_32off)); | |
14585 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
14586 | tnum_subreg(true_32off)); | |
14587 | __reg_combine_32_into_64(false_reg); | |
14588 | __reg_combine_32_into_64(true_reg); | |
14589 | } else { | |
14590 | false_reg->var_off = false_64off; | |
14591 | true_reg->var_off = true_64off; | |
14592 | __reg_combine_64_into_32(false_reg); | |
14593 | __reg_combine_64_into_32(true_reg); | |
14594 | } | |
48461135 JB |
14595 | } |
14596 | ||
f1174f77 EC |
14597 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
14598 | * the variable reg. | |
48461135 JB |
14599 | */ |
14600 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
14601 | struct bpf_reg_state *false_reg, |
14602 | u64 val, u32 val32, | |
092ed096 | 14603 | u8 opcode, bool is_jmp32) |
48461135 | 14604 | { |
6d94e741 | 14605 | opcode = flip_opcode(opcode); |
0fc31b10 JH |
14606 | /* This uses zero as "not present in table"; luckily the zero opcode, |
14607 | * BPF_JA, can't get here. | |
b03c9f9f | 14608 | */ |
0fc31b10 | 14609 | if (opcode) |
3f50f132 | 14610 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
14611 | } |
14612 | ||
14613 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
14614 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
14615 | struct bpf_reg_state *dst_reg) | |
14616 | { | |
b03c9f9f EC |
14617 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
14618 | dst_reg->umin_value); | |
14619 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
14620 | dst_reg->umax_value); | |
14621 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
14622 | dst_reg->smin_value); | |
14623 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
14624 | dst_reg->smax_value); | |
f1174f77 EC |
14625 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
14626 | dst_reg->var_off); | |
3844d153 DB |
14627 | reg_bounds_sync(src_reg); |
14628 | reg_bounds_sync(dst_reg); | |
f1174f77 EC |
14629 | } |
14630 | ||
14631 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
14632 | struct bpf_reg_state *true_dst, | |
14633 | struct bpf_reg_state *false_src, | |
14634 | struct bpf_reg_state *false_dst, | |
14635 | u8 opcode) | |
14636 | { | |
14637 | switch (opcode) { | |
14638 | case BPF_JEQ: | |
14639 | __reg_combine_min_max(true_src, true_dst); | |
14640 | break; | |
14641 | case BPF_JNE: | |
14642 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 14643 | break; |
4cabc5b1 | 14644 | } |
48461135 JB |
14645 | } |
14646 | ||
fd978bf7 JS |
14647 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
14648 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 14649 | bool is_null) |
57a09bf0 | 14650 | { |
c25b2ae1 | 14651 | if (type_may_be_null(reg->type) && reg->id == id && |
fca1aa75 | 14652 | (is_rcu_reg(reg) || !WARN_ON_ONCE(!reg->id))) { |
df57f38a KKD |
14653 | /* Old offset (both fixed and variable parts) should have been |
14654 | * known-zero, because we don't allow pointer arithmetic on | |
14655 | * pointers that might be NULL. If we see this happening, don't | |
14656 | * convert the register. | |
14657 | * | |
14658 | * But in some cases, some helpers that return local kptrs | |
14659 | * advance offset for the returned pointer. In those cases, it | |
14660 | * is fine to expect to see reg->off. | |
14661 | */ | |
14662 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || !tnum_equals_const(reg->var_off, 0))) | |
14663 | return; | |
6a3cd331 DM |
14664 | if (!(type_is_ptr_alloc_obj(reg->type) || type_is_non_owning_ref(reg->type)) && |
14665 | WARN_ON_ONCE(reg->off)) | |
e60b0d12 | 14666 | return; |
6a3cd331 | 14667 | |
f1174f77 EC |
14668 | if (is_null) { |
14669 | reg->type = SCALAR_VALUE; | |
1b986589 MKL |
14670 | /* We don't need id and ref_obj_id from this point |
14671 | * onwards anymore, thus we should better reset it, | |
14672 | * so that state pruning has chances to take effect. | |
14673 | */ | |
14674 | reg->id = 0; | |
14675 | reg->ref_obj_id = 0; | |
4ddb7416 DB |
14676 | |
14677 | return; | |
14678 | } | |
14679 | ||
14680 | mark_ptr_not_null_reg(reg); | |
14681 | ||
14682 | if (!reg_may_point_to_spin_lock(reg)) { | |
1b986589 | 14683 | /* For not-NULL ptr, reg->ref_obj_id will be reset |
b239da34 | 14684 | * in release_reference(). |
1b986589 MKL |
14685 | * |
14686 | * reg->id is still used by spin_lock ptr. Other | |
14687 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
14688 | */ |
14689 | reg->id = 0; | |
56f668df | 14690 | } |
57a09bf0 TG |
14691 | } |
14692 | } | |
14693 | ||
14694 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
14695 | * be folded together at some point. | |
14696 | */ | |
840b9615 JS |
14697 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
14698 | bool is_null) | |
57a09bf0 | 14699 | { |
f4d7e40a | 14700 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
b239da34 | 14701 | struct bpf_reg_state *regs = state->regs, *reg; |
1b986589 | 14702 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 14703 | u32 id = regs[regno].id; |
57a09bf0 | 14704 | |
1b986589 MKL |
14705 | if (ref_obj_id && ref_obj_id == id && is_null) |
14706 | /* regs[regno] is in the " == NULL" branch. | |
14707 | * No one could have freed the reference state before | |
14708 | * doing the NULL check. | |
14709 | */ | |
14710 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 14711 | |
b239da34 KKD |
14712 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
14713 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
14714 | })); | |
57a09bf0 TG |
14715 | } |
14716 | ||
5beca081 DB |
14717 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
14718 | struct bpf_reg_state *dst_reg, | |
14719 | struct bpf_reg_state *src_reg, | |
14720 | struct bpf_verifier_state *this_branch, | |
14721 | struct bpf_verifier_state *other_branch) | |
14722 | { | |
14723 | if (BPF_SRC(insn->code) != BPF_X) | |
14724 | return false; | |
14725 | ||
092ed096 JW |
14726 | /* Pointers are always 64-bit. */ |
14727 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
14728 | return false; | |
14729 | ||
5beca081 DB |
14730 | switch (BPF_OP(insn->code)) { |
14731 | case BPF_JGT: | |
14732 | if ((dst_reg->type == PTR_TO_PACKET && | |
14733 | src_reg->type == PTR_TO_PACKET_END) || | |
14734 | (dst_reg->type == PTR_TO_PACKET_META && | |
14735 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
14736 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
14737 | find_good_pkt_pointers(this_branch, dst_reg, | |
14738 | dst_reg->type, false); | |
6d94e741 | 14739 | mark_pkt_end(other_branch, insn->dst_reg, true); |
5beca081 DB |
14740 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
14741 | src_reg->type == PTR_TO_PACKET) || | |
14742 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
14743 | src_reg->type == PTR_TO_PACKET_META)) { | |
14744 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
14745 | find_good_pkt_pointers(other_branch, src_reg, | |
14746 | src_reg->type, true); | |
6d94e741 | 14747 | mark_pkt_end(this_branch, insn->src_reg, false); |
5beca081 DB |
14748 | } else { |
14749 | return false; | |
14750 | } | |
14751 | break; | |
14752 | case BPF_JLT: | |
14753 | if ((dst_reg->type == PTR_TO_PACKET && | |
14754 | src_reg->type == PTR_TO_PACKET_END) || | |
14755 | (dst_reg->type == PTR_TO_PACKET_META && | |
14756 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
14757 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
14758 | find_good_pkt_pointers(other_branch, dst_reg, | |
14759 | dst_reg->type, true); | |
6d94e741 | 14760 | mark_pkt_end(this_branch, insn->dst_reg, false); |
5beca081 DB |
14761 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
14762 | src_reg->type == PTR_TO_PACKET) || | |
14763 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
14764 | src_reg->type == PTR_TO_PACKET_META)) { | |
14765 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
14766 | find_good_pkt_pointers(this_branch, src_reg, | |
14767 | src_reg->type, false); | |
6d94e741 | 14768 | mark_pkt_end(other_branch, insn->src_reg, true); |
5beca081 DB |
14769 | } else { |
14770 | return false; | |
14771 | } | |
14772 | break; | |
14773 | case BPF_JGE: | |
14774 | if ((dst_reg->type == PTR_TO_PACKET && | |
14775 | src_reg->type == PTR_TO_PACKET_END) || | |
14776 | (dst_reg->type == PTR_TO_PACKET_META && | |
14777 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
14778 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
14779 | find_good_pkt_pointers(this_branch, dst_reg, | |
14780 | dst_reg->type, true); | |
6d94e741 | 14781 | mark_pkt_end(other_branch, insn->dst_reg, false); |
5beca081 DB |
14782 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
14783 | src_reg->type == PTR_TO_PACKET) || | |
14784 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
14785 | src_reg->type == PTR_TO_PACKET_META)) { | |
14786 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
14787 | find_good_pkt_pointers(other_branch, src_reg, | |
14788 | src_reg->type, false); | |
6d94e741 | 14789 | mark_pkt_end(this_branch, insn->src_reg, true); |
5beca081 DB |
14790 | } else { |
14791 | return false; | |
14792 | } | |
14793 | break; | |
14794 | case BPF_JLE: | |
14795 | if ((dst_reg->type == PTR_TO_PACKET && | |
14796 | src_reg->type == PTR_TO_PACKET_END) || | |
14797 | (dst_reg->type == PTR_TO_PACKET_META && | |
14798 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
14799 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
14800 | find_good_pkt_pointers(other_branch, dst_reg, | |
14801 | dst_reg->type, false); | |
6d94e741 | 14802 | mark_pkt_end(this_branch, insn->dst_reg, true); |
5beca081 DB |
14803 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
14804 | src_reg->type == PTR_TO_PACKET) || | |
14805 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
14806 | src_reg->type == PTR_TO_PACKET_META)) { | |
14807 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
14808 | find_good_pkt_pointers(this_branch, src_reg, | |
14809 | src_reg->type, true); | |
6d94e741 | 14810 | mark_pkt_end(other_branch, insn->src_reg, false); |
5beca081 DB |
14811 | } else { |
14812 | return false; | |
14813 | } | |
14814 | break; | |
14815 | default: | |
14816 | return false; | |
14817 | } | |
14818 | ||
14819 | return true; | |
14820 | } | |
14821 | ||
75748837 AS |
14822 | static void find_equal_scalars(struct bpf_verifier_state *vstate, |
14823 | struct bpf_reg_state *known_reg) | |
14824 | { | |
14825 | struct bpf_func_state *state; | |
14826 | struct bpf_reg_state *reg; | |
75748837 | 14827 | |
b239da34 KKD |
14828 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
14829 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
71f656a5 | 14830 | copy_register_state(reg, known_reg); |
b239da34 | 14831 | })); |
75748837 AS |
14832 | } |
14833 | ||
58e2af8b | 14834 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
14835 | struct bpf_insn *insn, int *insn_idx) |
14836 | { | |
f4d7e40a AS |
14837 | struct bpf_verifier_state *this_branch = env->cur_state; |
14838 | struct bpf_verifier_state *other_branch; | |
14839 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 14840 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
befae758 | 14841 | struct bpf_reg_state *eq_branch_regs; |
17a52670 | 14842 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 14843 | bool is_jmp32; |
fb8d251e | 14844 | int pred = -1; |
17a52670 AS |
14845 | int err; |
14846 | ||
092ed096 JW |
14847 | /* Only conditional jumps are expected to reach here. */ |
14848 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
14849 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
14850 | return -EINVAL; |
14851 | } | |
14852 | ||
d75e30dd YS |
14853 | /* check src2 operand */ |
14854 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); | |
14855 | if (err) | |
14856 | return err; | |
14857 | ||
14858 | dst_reg = ®s[insn->dst_reg]; | |
17a52670 AS |
14859 | if (BPF_SRC(insn->code) == BPF_X) { |
14860 | if (insn->imm != 0) { | |
092ed096 | 14861 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
14862 | return -EINVAL; |
14863 | } | |
14864 | ||
14865 | /* check src1 operand */ | |
dc503a8a | 14866 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
14867 | if (err) |
14868 | return err; | |
1be7f75d | 14869 | |
d75e30dd YS |
14870 | src_reg = ®s[insn->src_reg]; |
14871 | if (!(reg_is_pkt_pointer_any(dst_reg) && reg_is_pkt_pointer_any(src_reg)) && | |
14872 | is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 14873 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
14874 | insn->src_reg); |
14875 | return -EACCES; | |
14876 | } | |
17a52670 AS |
14877 | } else { |
14878 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 14879 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
14880 | return -EINVAL; |
14881 | } | |
14882 | } | |
14883 | ||
092ed096 | 14884 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 14885 | |
3f50f132 JF |
14886 | if (BPF_SRC(insn->code) == BPF_K) { |
14887 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
14888 | } else if (src_reg->type == SCALAR_VALUE && | |
14889 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
14890 | pred = is_branch_taken(dst_reg, | |
14891 | tnum_subreg(src_reg->var_off).value, | |
14892 | opcode, | |
14893 | is_jmp32); | |
14894 | } else if (src_reg->type == SCALAR_VALUE && | |
14895 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
14896 | pred = is_branch_taken(dst_reg, | |
14897 | src_reg->var_off.value, | |
14898 | opcode, | |
14899 | is_jmp32); | |
953d9f5b YS |
14900 | } else if (dst_reg->type == SCALAR_VALUE && |
14901 | is_jmp32 && tnum_is_const(tnum_subreg(dst_reg->var_off))) { | |
14902 | pred = is_branch_taken(src_reg, | |
14903 | tnum_subreg(dst_reg->var_off).value, | |
14904 | flip_opcode(opcode), | |
14905 | is_jmp32); | |
14906 | } else if (dst_reg->type == SCALAR_VALUE && | |
14907 | !is_jmp32 && tnum_is_const(dst_reg->var_off)) { | |
14908 | pred = is_branch_taken(src_reg, | |
14909 | dst_reg->var_off.value, | |
14910 | flip_opcode(opcode), | |
14911 | is_jmp32); | |
6d94e741 AS |
14912 | } else if (reg_is_pkt_pointer_any(dst_reg) && |
14913 | reg_is_pkt_pointer_any(src_reg) && | |
14914 | !is_jmp32) { | |
14915 | pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode); | |
3f50f132 JF |
14916 | } |
14917 | ||
b5dc0163 | 14918 | if (pred >= 0) { |
cac616db JF |
14919 | /* If we get here with a dst_reg pointer type it is because |
14920 | * above is_branch_taken() special cased the 0 comparison. | |
14921 | */ | |
14922 | if (!__is_pointer_value(false, dst_reg)) | |
14923 | err = mark_chain_precision(env, insn->dst_reg); | |
6d94e741 AS |
14924 | if (BPF_SRC(insn->code) == BPF_X && !err && |
14925 | !__is_pointer_value(false, src_reg)) | |
b5dc0163 AS |
14926 | err = mark_chain_precision(env, insn->src_reg); |
14927 | if (err) | |
14928 | return err; | |
14929 | } | |
9183671a | 14930 | |
fb8d251e | 14931 | if (pred == 1) { |
9183671a DB |
14932 | /* Only follow the goto, ignore fall-through. If needed, push |
14933 | * the fall-through branch for simulation under speculative | |
14934 | * execution. | |
14935 | */ | |
14936 | if (!env->bypass_spec_v1 && | |
14937 | !sanitize_speculative_path(env, insn, *insn_idx + 1, | |
14938 | *insn_idx)) | |
14939 | return -EFAULT; | |
1a8a315f AN |
14940 | if (env->log.level & BPF_LOG_LEVEL) |
14941 | print_insn_state(env, this_branch->frame[this_branch->curframe]); | |
fb8d251e AS |
14942 | *insn_idx += insn->off; |
14943 | return 0; | |
14944 | } else if (pred == 0) { | |
9183671a DB |
14945 | /* Only follow the fall-through branch, since that's where the |
14946 | * program will go. If needed, push the goto branch for | |
14947 | * simulation under speculative execution. | |
fb8d251e | 14948 | */ |
9183671a DB |
14949 | if (!env->bypass_spec_v1 && |
14950 | !sanitize_speculative_path(env, insn, | |
14951 | *insn_idx + insn->off + 1, | |
14952 | *insn_idx)) | |
14953 | return -EFAULT; | |
1a8a315f AN |
14954 | if (env->log.level & BPF_LOG_LEVEL) |
14955 | print_insn_state(env, this_branch->frame[this_branch->curframe]); | |
fb8d251e | 14956 | return 0; |
17a52670 AS |
14957 | } |
14958 | ||
979d63d5 DB |
14959 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
14960 | false); | |
17a52670 AS |
14961 | if (!other_branch) |
14962 | return -EFAULT; | |
f4d7e40a | 14963 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 14964 | |
48461135 JB |
14965 | /* detect if we are comparing against a constant value so we can adjust |
14966 | * our min/max values for our dst register. | |
f1174f77 | 14967 | * this is only legit if both are scalars (or pointers to the same |
befae758 EZ |
14968 | * object, I suppose, see the PTR_MAYBE_NULL related if block below), |
14969 | * because otherwise the different base pointers mean the offsets aren't | |
f1174f77 | 14970 | * comparable. |
48461135 JB |
14971 | */ |
14972 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 14973 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 14974 | |
f1174f77 | 14975 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
14976 | src_reg->type == SCALAR_VALUE) { |
14977 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
14978 | (is_jmp32 && |
14979 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 14980 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 14981 | dst_reg, |
3f50f132 JF |
14982 | src_reg->var_off.value, |
14983 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
14984 | opcode, is_jmp32); |
14985 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
14986 | (is_jmp32 && |
14987 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 14988 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 14989 | src_reg, |
3f50f132 JF |
14990 | dst_reg->var_off.value, |
14991 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
14992 | opcode, is_jmp32); |
14993 | else if (!is_jmp32 && | |
14994 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 14995 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
14996 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
14997 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 14998 | src_reg, dst_reg, opcode); |
e688c3db AS |
14999 | if (src_reg->id && |
15000 | !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) { | |
75748837 AS |
15001 | find_equal_scalars(this_branch, src_reg); |
15002 | find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]); | |
15003 | } | |
15004 | ||
f1174f77 EC |
15005 | } |
15006 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 15007 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
15008 | dst_reg, insn->imm, (u32)insn->imm, |
15009 | opcode, is_jmp32); | |
48461135 JB |
15010 | } |
15011 | ||
e688c3db AS |
15012 | if (dst_reg->type == SCALAR_VALUE && dst_reg->id && |
15013 | !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) { | |
75748837 AS |
15014 | find_equal_scalars(this_branch, dst_reg); |
15015 | find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]); | |
15016 | } | |
15017 | ||
befae758 EZ |
15018 | /* if one pointer register is compared to another pointer |
15019 | * register check if PTR_MAYBE_NULL could be lifted. | |
15020 | * E.g. register A - maybe null | |
15021 | * register B - not null | |
15022 | * for JNE A, B, ... - A is not null in the false branch; | |
15023 | * for JEQ A, B, ... - A is not null in the true branch. | |
8374bfd5 HS |
15024 | * |
15025 | * Since PTR_TO_BTF_ID points to a kernel struct that does | |
15026 | * not need to be null checked by the BPF program, i.e., | |
15027 | * could be null even without PTR_MAYBE_NULL marking, so | |
15028 | * only propagate nullness when neither reg is that type. | |
befae758 EZ |
15029 | */ |
15030 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_X && | |
15031 | __is_pointer_value(false, src_reg) && __is_pointer_value(false, dst_reg) && | |
8374bfd5 HS |
15032 | type_may_be_null(src_reg->type) != type_may_be_null(dst_reg->type) && |
15033 | base_type(src_reg->type) != PTR_TO_BTF_ID && | |
15034 | base_type(dst_reg->type) != PTR_TO_BTF_ID) { | |
befae758 EZ |
15035 | eq_branch_regs = NULL; |
15036 | switch (opcode) { | |
15037 | case BPF_JEQ: | |
15038 | eq_branch_regs = other_branch_regs; | |
15039 | break; | |
15040 | case BPF_JNE: | |
15041 | eq_branch_regs = regs; | |
15042 | break; | |
15043 | default: | |
15044 | /* do nothing */ | |
15045 | break; | |
15046 | } | |
15047 | if (eq_branch_regs) { | |
15048 | if (type_may_be_null(src_reg->type)) | |
15049 | mark_ptr_not_null_reg(&eq_branch_regs[insn->src_reg]); | |
15050 | else | |
15051 | mark_ptr_not_null_reg(&eq_branch_regs[insn->dst_reg]); | |
15052 | } | |
15053 | } | |
15054 | ||
092ed096 JW |
15055 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
15056 | * NOTE: these optimizations below are related with pointer comparison | |
15057 | * which will never be JMP32. | |
15058 | */ | |
15059 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 15060 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
c25b2ae1 | 15061 | type_may_be_null(dst_reg->type)) { |
840b9615 | 15062 | /* Mark all identical registers in each branch as either |
57a09bf0 TG |
15063 | * safe or unknown depending R == 0 or R != 0 conditional. |
15064 | */ | |
840b9615 JS |
15065 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
15066 | opcode == BPF_JNE); | |
15067 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
15068 | opcode == BPF_JEQ); | |
5beca081 DB |
15069 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
15070 | this_branch, other_branch) && | |
15071 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
15072 | verbose(env, "R%d pointer comparison prohibited\n", |
15073 | insn->dst_reg); | |
1be7f75d | 15074 | return -EACCES; |
17a52670 | 15075 | } |
06ee7115 | 15076 | if (env->log.level & BPF_LOG_LEVEL) |
2e576648 | 15077 | print_insn_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
15078 | return 0; |
15079 | } | |
15080 | ||
17a52670 | 15081 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 15082 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 15083 | { |
d8eca5bb | 15084 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 15085 | struct bpf_reg_state *regs = cur_regs(env); |
4976b718 | 15086 | struct bpf_reg_state *dst_reg; |
d8eca5bb | 15087 | struct bpf_map *map; |
17a52670 AS |
15088 | int err; |
15089 | ||
15090 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 15091 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
15092 | return -EINVAL; |
15093 | } | |
15094 | if (insn->off != 0) { | |
61bd5218 | 15095 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
15096 | return -EINVAL; |
15097 | } | |
15098 | ||
dc503a8a | 15099 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
15100 | if (err) |
15101 | return err; | |
15102 | ||
4976b718 | 15103 | dst_reg = ®s[insn->dst_reg]; |
6b173873 | 15104 | if (insn->src_reg == 0) { |
6b173873 JK |
15105 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
15106 | ||
4976b718 | 15107 | dst_reg->type = SCALAR_VALUE; |
b03c9f9f | 15108 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 15109 | return 0; |
6b173873 | 15110 | } |
17a52670 | 15111 | |
d400a6cf DB |
15112 | /* All special src_reg cases are listed below. From this point onwards |
15113 | * we either succeed and assign a corresponding dst_reg->type after | |
15114 | * zeroing the offset, or fail and reject the program. | |
15115 | */ | |
15116 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
4976b718 | 15117 | |
d400a6cf | 15118 | if (insn->src_reg == BPF_PSEUDO_BTF_ID) { |
4976b718 | 15119 | dst_reg->type = aux->btf_var.reg_type; |
34d3a78c | 15120 | switch (base_type(dst_reg->type)) { |
4976b718 HL |
15121 | case PTR_TO_MEM: |
15122 | dst_reg->mem_size = aux->btf_var.mem_size; | |
15123 | break; | |
15124 | case PTR_TO_BTF_ID: | |
22dc4a0f | 15125 | dst_reg->btf = aux->btf_var.btf; |
4976b718 HL |
15126 | dst_reg->btf_id = aux->btf_var.btf_id; |
15127 | break; | |
15128 | default: | |
15129 | verbose(env, "bpf verifier is misconfigured\n"); | |
15130 | return -EFAULT; | |
15131 | } | |
15132 | return 0; | |
15133 | } | |
15134 | ||
69c087ba YS |
15135 | if (insn->src_reg == BPF_PSEUDO_FUNC) { |
15136 | struct bpf_prog_aux *aux = env->prog->aux; | |
3990ed4c MKL |
15137 | u32 subprogno = find_subprog(env, |
15138 | env->insn_idx + insn->imm + 1); | |
69c087ba YS |
15139 | |
15140 | if (!aux->func_info) { | |
15141 | verbose(env, "missing btf func_info\n"); | |
15142 | return -EINVAL; | |
15143 | } | |
15144 | if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) { | |
15145 | verbose(env, "callback function not static\n"); | |
15146 | return -EINVAL; | |
15147 | } | |
15148 | ||
15149 | dst_reg->type = PTR_TO_FUNC; | |
15150 | dst_reg->subprogno = subprogno; | |
15151 | return 0; | |
15152 | } | |
15153 | ||
d8eca5bb | 15154 | map = env->used_maps[aux->map_index]; |
4976b718 | 15155 | dst_reg->map_ptr = map; |
d8eca5bb | 15156 | |
387544bf AS |
15157 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE || |
15158 | insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE) { | |
4976b718 HL |
15159 | dst_reg->type = PTR_TO_MAP_VALUE; |
15160 | dst_reg->off = aux->map_off; | |
d0d78c1d KKD |
15161 | WARN_ON_ONCE(map->max_entries != 1); |
15162 | /* We want reg->id to be same (0) as map_value is not distinct */ | |
387544bf AS |
15163 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD || |
15164 | insn->src_reg == BPF_PSEUDO_MAP_IDX) { | |
4976b718 | 15165 | dst_reg->type = CONST_PTR_TO_MAP; |
d8eca5bb DB |
15166 | } else { |
15167 | verbose(env, "bpf verifier is misconfigured\n"); | |
15168 | return -EINVAL; | |
15169 | } | |
17a52670 | 15170 | |
17a52670 AS |
15171 | return 0; |
15172 | } | |
15173 | ||
96be4325 DB |
15174 | static bool may_access_skb(enum bpf_prog_type type) |
15175 | { | |
15176 | switch (type) { | |
15177 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
15178 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 15179 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
15180 | return true; |
15181 | default: | |
15182 | return false; | |
15183 | } | |
15184 | } | |
15185 | ||
ddd872bc AS |
15186 | /* verify safety of LD_ABS|LD_IND instructions: |
15187 | * - they can only appear in the programs where ctx == skb | |
15188 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
15189 | * preserve R6-R9, and store return value into R0 | |
15190 | * | |
15191 | * Implicit input: | |
15192 | * ctx == skb == R6 == CTX | |
15193 | * | |
15194 | * Explicit input: | |
15195 | * SRC == any register | |
15196 | * IMM == 32-bit immediate | |
15197 | * | |
15198 | * Output: | |
15199 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
15200 | */ | |
58e2af8b | 15201 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 15202 | { |
638f5b90 | 15203 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 15204 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 15205 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
15206 | int i, err; |
15207 | ||
7e40781c | 15208 | if (!may_access_skb(resolve_prog_type(env->prog))) { |
61bd5218 | 15209 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
15210 | return -EINVAL; |
15211 | } | |
15212 | ||
e0cea7ce DB |
15213 | if (!env->ops->gen_ld_abs) { |
15214 | verbose(env, "bpf verifier is misconfigured\n"); | |
15215 | return -EINVAL; | |
15216 | } | |
15217 | ||
ddd872bc | 15218 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 15219 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 15220 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 15221 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
15222 | return -EINVAL; |
15223 | } | |
15224 | ||
15225 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 15226 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
15227 | if (err) |
15228 | return err; | |
15229 | ||
fd978bf7 JS |
15230 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
15231 | * gen_ld_abs() may terminate the program at runtime, leading to | |
15232 | * reference leak. | |
15233 | */ | |
f18b03fa | 15234 | err = check_reference_leak(env, false); |
fd978bf7 JS |
15235 | if (err) { |
15236 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
15237 | return err; | |
15238 | } | |
15239 | ||
d0d78c1d | 15240 | if (env->cur_state->active_lock.ptr) { |
d83525ca AS |
15241 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); |
15242 | return -EINVAL; | |
15243 | } | |
15244 | ||
9bb00b28 YS |
15245 | if (env->cur_state->active_rcu_lock) { |
15246 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_rcu_read_lock-ed region\n"); | |
15247 | return -EINVAL; | |
15248 | } | |
15249 | ||
6d4f151a | 15250 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
15251 | verbose(env, |
15252 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
15253 | return -EINVAL; |
15254 | } | |
15255 | ||
15256 | if (mode == BPF_IND) { | |
15257 | /* check explicit source operand */ | |
dc503a8a | 15258 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
15259 | if (err) |
15260 | return err; | |
15261 | } | |
15262 | ||
be80a1d3 | 15263 | err = check_ptr_off_reg(env, ®s[ctx_reg], ctx_reg); |
6d4f151a DB |
15264 | if (err < 0) |
15265 | return err; | |
15266 | ||
ddd872bc | 15267 | /* reset caller saved regs to unreadable */ |
dc503a8a | 15268 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 15269 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
15270 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
15271 | } | |
ddd872bc AS |
15272 | |
15273 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
15274 | * the value fetched from the packet. |
15275 | * Already marked as written above. | |
ddd872bc | 15276 | */ |
61bd5218 | 15277 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
15278 | /* ld_abs load up to 32-bit skb data. */ |
15279 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
15280 | return 0; |
15281 | } | |
15282 | ||
a923819f | 15283 | static int check_return_code(struct bpf_verifier_env *env, int regno) |
390ee7e2 | 15284 | { |
5cf1e914 | 15285 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 15286 | const struct bpf_prog *prog = env->prog; |
390ee7e2 | 15287 | struct bpf_reg_state *reg; |
82995598 | 15288 | struct tnum range = tnum_range(0, 1), const_0 = tnum_const(0); |
7e40781c | 15289 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
27ae7997 | 15290 | int err; |
bfc6bb74 AS |
15291 | struct bpf_func_state *frame = env->cur_state->frame[0]; |
15292 | const bool is_subprog = frame->subprogno; | |
27ae7997 | 15293 | |
9e4e01df | 15294 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
b9ae0c9d | 15295 | if (!is_subprog || frame->in_exception_callback_fn) { |
d1a6edec SF |
15296 | switch (prog_type) { |
15297 | case BPF_PROG_TYPE_LSM: | |
15298 | if (prog->expected_attach_type == BPF_LSM_CGROUP) | |
15299 | /* See below, can be 0 or 0-1 depending on hook. */ | |
15300 | break; | |
15301 | fallthrough; | |
15302 | case BPF_PROG_TYPE_STRUCT_OPS: | |
15303 | if (!prog->aux->attach_func_proto->type) | |
15304 | return 0; | |
15305 | break; | |
15306 | default: | |
15307 | break; | |
15308 | } | |
15309 | } | |
27ae7997 | 15310 | |
8fb33b60 | 15311 | /* eBPF calling convention is such that R0 is used |
27ae7997 MKL |
15312 | * to return the value from eBPF program. |
15313 | * Make sure that it's readable at this time | |
15314 | * of bpf_exit, which means that program wrote | |
15315 | * something into it earlier | |
15316 | */ | |
a923819f | 15317 | err = check_reg_arg(env, regno, SRC_OP); |
27ae7997 MKL |
15318 | if (err) |
15319 | return err; | |
15320 | ||
a923819f KKD |
15321 | if (is_pointer_value(env, regno)) { |
15322 | verbose(env, "R%d leaks addr as return value\n", regno); | |
27ae7997 MKL |
15323 | return -EACCES; |
15324 | } | |
390ee7e2 | 15325 | |
a923819f | 15326 | reg = cur_regs(env) + regno; |
bfc6bb74 AS |
15327 | |
15328 | if (frame->in_async_callback_fn) { | |
15329 | /* enforce return zero from async callbacks like timer */ | |
15330 | if (reg->type != SCALAR_VALUE) { | |
a923819f KKD |
15331 | verbose(env, "In async callback the register R%d is not a known value (%s)\n", |
15332 | regno, reg_type_str(env, reg->type)); | |
bfc6bb74 AS |
15333 | return -EINVAL; |
15334 | } | |
15335 | ||
82995598 DV |
15336 | if (!tnum_in(const_0, reg->var_off)) { |
15337 | verbose_invalid_scalar(env, reg, &const_0, "async callback", "R0"); | |
bfc6bb74 AS |
15338 | return -EINVAL; |
15339 | } | |
15340 | return 0; | |
15341 | } | |
15342 | ||
b9ae0c9d | 15343 | if (is_subprog && !frame->in_exception_callback_fn) { |
f782e2c3 | 15344 | if (reg->type != SCALAR_VALUE) { |
a923819f KKD |
15345 | verbose(env, "At subprogram exit the register R%d is not a scalar value (%s)\n", |
15346 | regno, reg_type_str(env, reg->type)); | |
f782e2c3 DB |
15347 | return -EINVAL; |
15348 | } | |
15349 | return 0; | |
15350 | } | |
15351 | ||
7e40781c | 15352 | switch (prog_type) { |
983695fa DB |
15353 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
15354 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 | 15355 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
859051dd | 15356 | env->prog->expected_attach_type == BPF_CGROUP_UNIX_RECVMSG || |
1b66d253 DB |
15357 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || |
15358 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
859051dd | 15359 | env->prog->expected_attach_type == BPF_CGROUP_UNIX_GETPEERNAME || |
1b66d253 | 15360 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || |
859051dd DDM |
15361 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME || |
15362 | env->prog->expected_attach_type == BPF_CGROUP_UNIX_GETSOCKNAME) | |
983695fa | 15363 | range = tnum_range(1, 1); |
77241217 SF |
15364 | if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND || |
15365 | env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND) | |
15366 | range = tnum_range(0, 3); | |
ed4ed404 | 15367 | break; |
390ee7e2 | 15368 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 15369 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
15370 | range = tnum_range(0, 3); | |
15371 | enforce_attach_type_range = tnum_range(2, 3); | |
15372 | } | |
ed4ed404 | 15373 | break; |
390ee7e2 AS |
15374 | case BPF_PROG_TYPE_CGROUP_SOCK: |
15375 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 15376 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 15377 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 15378 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 15379 | break; |
15ab09bd AS |
15380 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
15381 | if (!env->prog->aux->attach_btf_id) | |
15382 | return 0; | |
15383 | range = tnum_const(0); | |
15384 | break; | |
15d83c4d | 15385 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
15386 | switch (env->prog->expected_attach_type) { |
15387 | case BPF_TRACE_FENTRY: | |
15388 | case BPF_TRACE_FEXIT: | |
15389 | range = tnum_const(0); | |
15390 | break; | |
15391 | case BPF_TRACE_RAW_TP: | |
15392 | case BPF_MODIFY_RETURN: | |
15d83c4d | 15393 | return 0; |
2ec0616e DB |
15394 | case BPF_TRACE_ITER: |
15395 | break; | |
e92888c7 YS |
15396 | default: |
15397 | return -ENOTSUPP; | |
15398 | } | |
15d83c4d | 15399 | break; |
e9ddbb77 JS |
15400 | case BPF_PROG_TYPE_SK_LOOKUP: |
15401 | range = tnum_range(SK_DROP, SK_PASS); | |
15402 | break; | |
69fd337a SF |
15403 | |
15404 | case BPF_PROG_TYPE_LSM: | |
15405 | if (env->prog->expected_attach_type != BPF_LSM_CGROUP) { | |
15406 | /* Regular BPF_PROG_TYPE_LSM programs can return | |
15407 | * any value. | |
15408 | */ | |
15409 | return 0; | |
15410 | } | |
15411 | if (!env->prog->aux->attach_func_proto->type) { | |
15412 | /* Make sure programs that attach to void | |
15413 | * hooks don't try to modify return value. | |
15414 | */ | |
15415 | range = tnum_range(1, 1); | |
15416 | } | |
15417 | break; | |
15418 | ||
fd9c663b FW |
15419 | case BPF_PROG_TYPE_NETFILTER: |
15420 | range = tnum_range(NF_DROP, NF_ACCEPT); | |
15421 | break; | |
e92888c7 YS |
15422 | case BPF_PROG_TYPE_EXT: |
15423 | /* freplace program can return anything as its return value | |
15424 | * depends on the to-be-replaced kernel func or bpf program. | |
15425 | */ | |
390ee7e2 AS |
15426 | default: |
15427 | return 0; | |
15428 | } | |
15429 | ||
390ee7e2 | 15430 | if (reg->type != SCALAR_VALUE) { |
a923819f KKD |
15431 | verbose(env, "At program exit the register R%d is not a known value (%s)\n", |
15432 | regno, reg_type_str(env, reg->type)); | |
390ee7e2 AS |
15433 | return -EINVAL; |
15434 | } | |
15435 | ||
15436 | if (!tnum_in(range, reg->var_off)) { | |
bc2591d6 | 15437 | verbose_invalid_scalar(env, reg, &range, "program exit", "R0"); |
69fd337a | 15438 | if (prog->expected_attach_type == BPF_LSM_CGROUP && |
d1a6edec | 15439 | prog_type == BPF_PROG_TYPE_LSM && |
69fd337a SF |
15440 | !prog->aux->attach_func_proto->type) |
15441 | verbose(env, "Note, BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
390ee7e2 AS |
15442 | return -EINVAL; |
15443 | } | |
5cf1e914 | 15444 | |
15445 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
15446 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
15447 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
15448 | return 0; |
15449 | } | |
15450 | ||
475fb78f AS |
15451 | /* non-recursive DFS pseudo code |
15452 | * 1 procedure DFS-iterative(G,v): | |
15453 | * 2 label v as discovered | |
15454 | * 3 let S be a stack | |
15455 | * 4 S.push(v) | |
15456 | * 5 while S is not empty | |
b6d20799 | 15457 | * 6 t <- S.peek() |
475fb78f AS |
15458 | * 7 if t is what we're looking for: |
15459 | * 8 return t | |
15460 | * 9 for all edges e in G.adjacentEdges(t) do | |
15461 | * 10 if edge e is already labelled | |
15462 | * 11 continue with the next edge | |
15463 | * 12 w <- G.adjacentVertex(t,e) | |
15464 | * 13 if vertex w is not discovered and not explored | |
15465 | * 14 label e as tree-edge | |
15466 | * 15 label w as discovered | |
15467 | * 16 S.push(w) | |
15468 | * 17 continue at 5 | |
15469 | * 18 else if vertex w is discovered | |
15470 | * 19 label e as back-edge | |
15471 | * 20 else | |
15472 | * 21 // vertex w is explored | |
15473 | * 22 label e as forward- or cross-edge | |
15474 | * 23 label t as explored | |
15475 | * 24 S.pop() | |
15476 | * | |
15477 | * convention: | |
15478 | * 0x10 - discovered | |
15479 | * 0x11 - discovered and fall-through edge labelled | |
15480 | * 0x12 - discovered and fall-through and branch edges labelled | |
15481 | * 0x20 - explored | |
15482 | */ | |
15483 | ||
15484 | enum { | |
15485 | DISCOVERED = 0x10, | |
15486 | EXPLORED = 0x20, | |
15487 | FALLTHROUGH = 1, | |
15488 | BRANCH = 2, | |
15489 | }; | |
15490 | ||
bffdeaa8 | 15491 | static void mark_prune_point(struct bpf_verifier_env *env, int idx) |
5d839021 | 15492 | { |
a8f500af | 15493 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 15494 | } |
f1bca824 | 15495 | |
bffdeaa8 AN |
15496 | static bool is_prune_point(struct bpf_verifier_env *env, int insn_idx) |
15497 | { | |
15498 | return env->insn_aux_data[insn_idx].prune_point; | |
15499 | } | |
15500 | ||
4b5ce570 AN |
15501 | static void mark_force_checkpoint(struct bpf_verifier_env *env, int idx) |
15502 | { | |
15503 | env->insn_aux_data[idx].force_checkpoint = true; | |
15504 | } | |
15505 | ||
15506 | static bool is_force_checkpoint(struct bpf_verifier_env *env, int insn_idx) | |
15507 | { | |
15508 | return env->insn_aux_data[insn_idx].force_checkpoint; | |
15509 | } | |
15510 | ||
ab5cfac1 EZ |
15511 | static void mark_calls_callback(struct bpf_verifier_env *env, int idx) |
15512 | { | |
15513 | env->insn_aux_data[idx].calls_callback = true; | |
15514 | } | |
15515 | ||
15516 | static bool calls_callback(struct bpf_verifier_env *env, int insn_idx) | |
15517 | { | |
15518 | return env->insn_aux_data[insn_idx].calls_callback; | |
15519 | } | |
4b5ce570 | 15520 | |
59e2e27d WAF |
15521 | enum { |
15522 | DONE_EXPLORING = 0, | |
15523 | KEEP_EXPLORING = 1, | |
15524 | }; | |
15525 | ||
475fb78f AS |
15526 | /* t, w, e - match pseudo-code above: |
15527 | * t - index of current instruction | |
15528 | * w - next instruction | |
15529 | * e - edge | |
15530 | */ | |
10e14e96 | 15531 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f | 15532 | { |
7df737e9 AS |
15533 | int *insn_stack = env->cfg.insn_stack; |
15534 | int *insn_state = env->cfg.insn_state; | |
15535 | ||
475fb78f | 15536 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
59e2e27d | 15537 | return DONE_EXPLORING; |
475fb78f AS |
15538 | |
15539 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
59e2e27d | 15540 | return DONE_EXPLORING; |
475fb78f AS |
15541 | |
15542 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 15543 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 15544 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
15545 | return -EINVAL; |
15546 | } | |
15547 | ||
bffdeaa8 | 15548 | if (e == BRANCH) { |
f1bca824 | 15549 | /* mark branch target for state pruning */ |
bffdeaa8 AN |
15550 | mark_prune_point(env, w); |
15551 | mark_jmp_point(env, w); | |
15552 | } | |
f1bca824 | 15553 | |
475fb78f AS |
15554 | if (insn_state[w] == 0) { |
15555 | /* tree-edge */ | |
15556 | insn_state[t] = DISCOVERED | e; | |
15557 | insn_state[w] = DISCOVERED; | |
7df737e9 | 15558 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 15559 | return -E2BIG; |
7df737e9 | 15560 | insn_stack[env->cfg.cur_stack++] = w; |
59e2e27d | 15561 | return KEEP_EXPLORING; |
475fb78f | 15562 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { |
10e14e96 | 15563 | if (env->bpf_capable) |
59e2e27d | 15564 | return DONE_EXPLORING; |
d9762e84 MKL |
15565 | verbose_linfo(env, t, "%d: ", t); |
15566 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 15567 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
15568 | return -EINVAL; |
15569 | } else if (insn_state[w] == EXPLORED) { | |
15570 | /* forward- or cross-edge */ | |
15571 | insn_state[t] = DISCOVERED | e; | |
15572 | } else { | |
61bd5218 | 15573 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
15574 | return -EFAULT; |
15575 | } | |
59e2e27d WAF |
15576 | return DONE_EXPLORING; |
15577 | } | |
15578 | ||
dcb2288b | 15579 | static int visit_func_call_insn(int t, struct bpf_insn *insns, |
efdb22de YS |
15580 | struct bpf_verifier_env *env, |
15581 | bool visit_callee) | |
15582 | { | |
3feb263b | 15583 | int ret, insn_sz; |
efdb22de | 15584 | |
3feb263b | 15585 | insn_sz = bpf_is_ldimm64(&insns[t]) ? 2 : 1; |
10e14e96 | 15586 | ret = push_insn(t, t + insn_sz, FALLTHROUGH, env); |
efdb22de YS |
15587 | if (ret) |
15588 | return ret; | |
15589 | ||
3feb263b | 15590 | mark_prune_point(env, t + insn_sz); |
618945fb | 15591 | /* when we exit from subprog, we need to record non-linear history */ |
3feb263b | 15592 | mark_jmp_point(env, t + insn_sz); |
618945fb | 15593 | |
efdb22de | 15594 | if (visit_callee) { |
bffdeaa8 | 15595 | mark_prune_point(env, t); |
10e14e96 | 15596 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env); |
efdb22de YS |
15597 | } |
15598 | return ret; | |
15599 | } | |
15600 | ||
59e2e27d WAF |
15601 | /* Visits the instruction at index t and returns one of the following: |
15602 | * < 0 - an error occurred | |
15603 | * DONE_EXPLORING - the instruction was fully explored | |
15604 | * KEEP_EXPLORING - there is still work to be done before it is fully explored | |
15605 | */ | |
dcb2288b | 15606 | static int visit_insn(int t, struct bpf_verifier_env *env) |
59e2e27d | 15607 | { |
653ae3a8 | 15608 | struct bpf_insn *insns = env->prog->insnsi, *insn = &insns[t]; |
3feb263b | 15609 | int ret, off, insn_sz; |
59e2e27d | 15610 | |
653ae3a8 | 15611 | if (bpf_pseudo_func(insn)) |
dcb2288b | 15612 | return visit_func_call_insn(t, insns, env, true); |
69c087ba | 15613 | |
59e2e27d | 15614 | /* All non-branch instructions have a single fall-through edge. */ |
653ae3a8 | 15615 | if (BPF_CLASS(insn->code) != BPF_JMP && |
3feb263b AN |
15616 | BPF_CLASS(insn->code) != BPF_JMP32) { |
15617 | insn_sz = bpf_is_ldimm64(insn) ? 2 : 1; | |
10e14e96 | 15618 | return push_insn(t, t + insn_sz, FALLTHROUGH, env); |
3feb263b | 15619 | } |
59e2e27d | 15620 | |
653ae3a8 | 15621 | switch (BPF_OP(insn->code)) { |
59e2e27d WAF |
15622 | case BPF_EXIT: |
15623 | return DONE_EXPLORING; | |
15624 | ||
15625 | case BPF_CALL: | |
c1ee85a9 | 15626 | if (insn->src_reg == 0 && insn->imm == BPF_FUNC_timer_set_callback) |
618945fb AN |
15627 | /* Mark this call insn as a prune point to trigger |
15628 | * is_state_visited() check before call itself is | |
15629 | * processed by __check_func_call(). Otherwise new | |
15630 | * async state will be pushed for further exploration. | |
bfc6bb74 | 15631 | */ |
bffdeaa8 | 15632 | mark_prune_point(env, t); |
ab5cfac1 EZ |
15633 | /* For functions that invoke callbacks it is not known how many times |
15634 | * callback would be called. Verifier models callback calling functions | |
15635 | * by repeatedly visiting callback bodies and returning to origin call | |
15636 | * instruction. | |
15637 | * In order to stop such iteration verifier needs to identify when a | |
15638 | * state identical some state from a previous iteration is reached. | |
15639 | * Check below forces creation of checkpoint before callback calling | |
15640 | * instruction to allow search for such identical states. | |
15641 | */ | |
15642 | if (is_sync_callback_calling_insn(insn)) { | |
15643 | mark_calls_callback(env, t); | |
15644 | mark_force_checkpoint(env, t); | |
15645 | mark_prune_point(env, t); | |
15646 | mark_jmp_point(env, t); | |
15647 | } | |
06accc87 AN |
15648 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
15649 | struct bpf_kfunc_call_arg_meta meta; | |
15650 | ||
15651 | ret = fetch_kfunc_meta(env, insn, &meta, NULL); | |
4b5ce570 | 15652 | if (ret == 0 && is_iter_next_kfunc(&meta)) { |
06accc87 | 15653 | mark_prune_point(env, t); |
4b5ce570 AN |
15654 | /* Checking and saving state checkpoints at iter_next() call |
15655 | * is crucial for fast convergence of open-coded iterator loop | |
15656 | * logic, so we need to force it. If we don't do that, | |
15657 | * is_state_visited() might skip saving a checkpoint, causing | |
15658 | * unnecessarily long sequence of not checkpointed | |
15659 | * instructions and jumps, leading to exhaustion of jump | |
15660 | * history buffer, and potentially other undesired outcomes. | |
15661 | * It is expected that with correct open-coded iterators | |
15662 | * convergence will happen quickly, so we don't run a risk of | |
15663 | * exhausting memory. | |
15664 | */ | |
15665 | mark_force_checkpoint(env, t); | |
15666 | } | |
06accc87 | 15667 | } |
653ae3a8 | 15668 | return visit_func_call_insn(t, insns, env, insn->src_reg == BPF_PSEUDO_CALL); |
59e2e27d WAF |
15669 | |
15670 | case BPF_JA: | |
653ae3a8 | 15671 | if (BPF_SRC(insn->code) != BPF_K) |
59e2e27d WAF |
15672 | return -EINVAL; |
15673 | ||
4cd58e9a YS |
15674 | if (BPF_CLASS(insn->code) == BPF_JMP) |
15675 | off = insn->off; | |
15676 | else | |
15677 | off = insn->imm; | |
15678 | ||
59e2e27d | 15679 | /* unconditional jump with single edge */ |
10e14e96 | 15680 | ret = push_insn(t, t + off + 1, FALLTHROUGH, env); |
59e2e27d WAF |
15681 | if (ret) |
15682 | return ret; | |
15683 | ||
4cd58e9a YS |
15684 | mark_prune_point(env, t + off + 1); |
15685 | mark_jmp_point(env, t + off + 1); | |
59e2e27d WAF |
15686 | |
15687 | return ret; | |
15688 | ||
15689 | default: | |
15690 | /* conditional jump with two edges */ | |
bffdeaa8 | 15691 | mark_prune_point(env, t); |
618945fb | 15692 | |
10e14e96 | 15693 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
59e2e27d WAF |
15694 | if (ret) |
15695 | return ret; | |
15696 | ||
10e14e96 | 15697 | return push_insn(t, t + insn->off + 1, BRANCH, env); |
59e2e27d | 15698 | } |
475fb78f AS |
15699 | } |
15700 | ||
15701 | /* non-recursive depth-first-search to detect loops in BPF program | |
15702 | * loop == back-edge in directed graph | |
15703 | */ | |
58e2af8b | 15704 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f | 15705 | { |
475fb78f | 15706 | int insn_cnt = env->prog->len; |
7df737e9 | 15707 | int *insn_stack, *insn_state; |
b62bf8a5 KKD |
15708 | int ex_insn_beg, i, ret = 0; |
15709 | bool ex_done = false; | |
475fb78f | 15710 | |
7df737e9 | 15711 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
15712 | if (!insn_state) |
15713 | return -ENOMEM; | |
15714 | ||
7df737e9 | 15715 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 15716 | if (!insn_stack) { |
71dde681 | 15717 | kvfree(insn_state); |
475fb78f AS |
15718 | return -ENOMEM; |
15719 | } | |
15720 | ||
15721 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
15722 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 15723 | env->cfg.cur_stack = 1; |
475fb78f | 15724 | |
b62bf8a5 | 15725 | walk_cfg: |
59e2e27d WAF |
15726 | while (env->cfg.cur_stack > 0) { |
15727 | int t = insn_stack[env->cfg.cur_stack - 1]; | |
475fb78f | 15728 | |
dcb2288b | 15729 | ret = visit_insn(t, env); |
59e2e27d WAF |
15730 | switch (ret) { |
15731 | case DONE_EXPLORING: | |
15732 | insn_state[t] = EXPLORED; | |
15733 | env->cfg.cur_stack--; | |
15734 | break; | |
15735 | case KEEP_EXPLORING: | |
15736 | break; | |
15737 | default: | |
15738 | if (ret > 0) { | |
15739 | verbose(env, "visit_insn internal bug\n"); | |
15740 | ret = -EFAULT; | |
475fb78f | 15741 | } |
475fb78f | 15742 | goto err_free; |
59e2e27d | 15743 | } |
475fb78f AS |
15744 | } |
15745 | ||
59e2e27d | 15746 | if (env->cfg.cur_stack < 0) { |
61bd5218 | 15747 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
15748 | ret = -EFAULT; |
15749 | goto err_free; | |
15750 | } | |
475fb78f | 15751 | |
b62bf8a5 KKD |
15752 | if (env->exception_callback_subprog && !ex_done) { |
15753 | ex_insn_beg = env->subprog_info[env->exception_callback_subprog].start; | |
15754 | ||
15755 | insn_state[ex_insn_beg] = DISCOVERED; | |
15756 | insn_stack[0] = ex_insn_beg; | |
15757 | env->cfg.cur_stack = 1; | |
15758 | ex_done = true; | |
15759 | goto walk_cfg; | |
15760 | } | |
15761 | ||
475fb78f | 15762 | for (i = 0; i < insn_cnt; i++) { |
3feb263b AN |
15763 | struct bpf_insn *insn = &env->prog->insnsi[i]; |
15764 | ||
475fb78f | 15765 | if (insn_state[i] != EXPLORED) { |
61bd5218 | 15766 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
15767 | ret = -EINVAL; |
15768 | goto err_free; | |
15769 | } | |
3feb263b AN |
15770 | if (bpf_is_ldimm64(insn)) { |
15771 | if (insn_state[i + 1] != 0) { | |
15772 | verbose(env, "jump into the middle of ldimm64 insn %d\n", i); | |
15773 | ret = -EINVAL; | |
15774 | goto err_free; | |
15775 | } | |
15776 | i++; /* skip second half of ldimm64 */ | |
15777 | } | |
475fb78f AS |
15778 | } |
15779 | ret = 0; /* cfg looks good */ | |
15780 | ||
15781 | err_free: | |
71dde681 AS |
15782 | kvfree(insn_state); |
15783 | kvfree(insn_stack); | |
7df737e9 | 15784 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
15785 | return ret; |
15786 | } | |
15787 | ||
09b28d76 AS |
15788 | static int check_abnormal_return(struct bpf_verifier_env *env) |
15789 | { | |
15790 | int i; | |
15791 | ||
15792 | for (i = 1; i < env->subprog_cnt; i++) { | |
15793 | if (env->subprog_info[i].has_ld_abs) { | |
15794 | verbose(env, "LD_ABS is not allowed in subprogs without BTF\n"); | |
15795 | return -EINVAL; | |
15796 | } | |
15797 | if (env->subprog_info[i].has_tail_call) { | |
15798 | verbose(env, "tail_call is not allowed in subprogs without BTF\n"); | |
15799 | return -EINVAL; | |
15800 | } | |
15801 | } | |
15802 | return 0; | |
15803 | } | |
15804 | ||
838e9690 YS |
15805 | /* The minimum supported BTF func info size */ |
15806 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
15807 | #define MAX_FUNCINFO_REC_SIZE 252 | |
15808 | ||
aaa619eb KKD |
15809 | static int check_btf_func_early(struct bpf_verifier_env *env, |
15810 | const union bpf_attr *attr, | |
15811 | bpfptr_t uattr) | |
838e9690 | 15812 | { |
838e9690 | 15813 | u32 krec_size = sizeof(struct bpf_func_info); |
aaa619eb KKD |
15814 | const struct btf_type *type, *func_proto; |
15815 | u32 i, nfuncs, urec_size, min_size; | |
c454a46b | 15816 | struct bpf_func_info *krecord; |
c454a46b MKL |
15817 | struct bpf_prog *prog; |
15818 | const struct btf *btf; | |
d0b2818e | 15819 | u32 prev_offset = 0; |
aaa619eb | 15820 | bpfptr_t urecord; |
e7ed83d6 | 15821 | int ret = -ENOMEM; |
838e9690 YS |
15822 | |
15823 | nfuncs = attr->func_info_cnt; | |
09b28d76 AS |
15824 | if (!nfuncs) { |
15825 | if (check_abnormal_return(env)) | |
15826 | return -EINVAL; | |
838e9690 | 15827 | return 0; |
09b28d76 | 15828 | } |
838e9690 | 15829 | |
838e9690 YS |
15830 | urec_size = attr->func_info_rec_size; |
15831 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
15832 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
15833 | urec_size % sizeof(u32)) { | |
15834 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
15835 | return -EINVAL; | |
15836 | } | |
15837 | ||
c454a46b MKL |
15838 | prog = env->prog; |
15839 | btf = prog->aux->btf; | |
838e9690 | 15840 | |
af2ac3e1 | 15841 | urecord = make_bpfptr(attr->func_info, uattr.is_kernel); |
838e9690 YS |
15842 | min_size = min_t(u32, krec_size, urec_size); |
15843 | ||
ba64e7d8 | 15844 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
15845 | if (!krecord) |
15846 | return -ENOMEM; | |
ba64e7d8 | 15847 | |
838e9690 YS |
15848 | for (i = 0; i < nfuncs; i++) { |
15849 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
15850 | if (ret) { | |
15851 | if (ret == -E2BIG) { | |
15852 | verbose(env, "nonzero tailing record in func info"); | |
15853 | /* set the size kernel expects so loader can zero | |
15854 | * out the rest of the record. | |
15855 | */ | |
af2ac3e1 AS |
15856 | if (copy_to_bpfptr_offset(uattr, |
15857 | offsetof(union bpf_attr, func_info_rec_size), | |
15858 | &min_size, sizeof(min_size))) | |
838e9690 YS |
15859 | ret = -EFAULT; |
15860 | } | |
c454a46b | 15861 | goto err_free; |
838e9690 YS |
15862 | } |
15863 | ||
af2ac3e1 | 15864 | if (copy_from_bpfptr(&krecord[i], urecord, min_size)) { |
838e9690 | 15865 | ret = -EFAULT; |
c454a46b | 15866 | goto err_free; |
838e9690 YS |
15867 | } |
15868 | ||
d30d42e0 | 15869 | /* check insn_off */ |
09b28d76 | 15870 | ret = -EINVAL; |
838e9690 | 15871 | if (i == 0) { |
d30d42e0 | 15872 | if (krecord[i].insn_off) { |
838e9690 | 15873 | verbose(env, |
d30d42e0 MKL |
15874 | "nonzero insn_off %u for the first func info record", |
15875 | krecord[i].insn_off); | |
c454a46b | 15876 | goto err_free; |
838e9690 | 15877 | } |
d30d42e0 | 15878 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
15879 | verbose(env, |
15880 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 15881 | krecord[i].insn_off, prev_offset); |
c454a46b | 15882 | goto err_free; |
838e9690 YS |
15883 | } |
15884 | ||
838e9690 | 15885 | /* check type_id */ |
ba64e7d8 | 15886 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 15887 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 15888 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 15889 | krecord[i].type_id); |
c454a46b | 15890 | goto err_free; |
838e9690 | 15891 | } |
09b28d76 AS |
15892 | |
15893 | func_proto = btf_type_by_id(btf, type->type); | |
15894 | if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto))) | |
15895 | /* btf_func_check() already verified it during BTF load */ | |
15896 | goto err_free; | |
aaa619eb KKD |
15897 | |
15898 | prev_offset = krecord[i].insn_off; | |
15899 | bpfptr_add(&urecord, urec_size); | |
15900 | } | |
15901 | ||
15902 | prog->aux->func_info = krecord; | |
15903 | prog->aux->func_info_cnt = nfuncs; | |
15904 | return 0; | |
15905 | ||
15906 | err_free: | |
15907 | kvfree(krecord); | |
15908 | return ret; | |
15909 | } | |
15910 | ||
15911 | static int check_btf_func(struct bpf_verifier_env *env, | |
15912 | const union bpf_attr *attr, | |
15913 | bpfptr_t uattr) | |
15914 | { | |
15915 | const struct btf_type *type, *func_proto, *ret_type; | |
aec42f36 | 15916 | u32 i, nfuncs, urec_size; |
aaa619eb KKD |
15917 | struct bpf_func_info *krecord; |
15918 | struct bpf_func_info_aux *info_aux = NULL; | |
15919 | struct bpf_prog *prog; | |
15920 | const struct btf *btf; | |
15921 | bpfptr_t urecord; | |
aaa619eb KKD |
15922 | bool scalar_return; |
15923 | int ret = -ENOMEM; | |
15924 | ||
15925 | nfuncs = attr->func_info_cnt; | |
15926 | if (!nfuncs) { | |
15927 | if (check_abnormal_return(env)) | |
15928 | return -EINVAL; | |
15929 | return 0; | |
15930 | } | |
15931 | if (nfuncs != env->subprog_cnt) { | |
15932 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
15933 | return -EINVAL; | |
15934 | } | |
15935 | ||
15936 | urec_size = attr->func_info_rec_size; | |
15937 | ||
15938 | prog = env->prog; | |
15939 | btf = prog->aux->btf; | |
15940 | ||
15941 | urecord = make_bpfptr(attr->func_info, uattr.is_kernel); | |
aaa619eb KKD |
15942 | |
15943 | krecord = prog->aux->func_info; | |
15944 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); | |
15945 | if (!info_aux) | |
15946 | return -ENOMEM; | |
15947 | ||
15948 | for (i = 0; i < nfuncs; i++) { | |
15949 | /* check insn_off */ | |
15950 | ret = -EINVAL; | |
15951 | ||
15952 | if (env->subprog_info[i].start != krecord[i].insn_off) { | |
15953 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); | |
15954 | goto err_free; | |
15955 | } | |
15956 | ||
15957 | /* Already checked type_id */ | |
15958 | type = btf_type_by_id(btf, krecord[i].type_id); | |
15959 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); | |
15960 | /* Already checked func_proto */ | |
15961 | func_proto = btf_type_by_id(btf, type->type); | |
15962 | ||
09b28d76 AS |
15963 | ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL); |
15964 | scalar_return = | |
6089fb32 | 15965 | btf_type_is_small_int(ret_type) || btf_is_any_enum(ret_type); |
09b28d76 AS |
15966 | if (i && !scalar_return && env->subprog_info[i].has_ld_abs) { |
15967 | verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n"); | |
15968 | goto err_free; | |
15969 | } | |
15970 | if (i && !scalar_return && env->subprog_info[i].has_tail_call) { | |
15971 | verbose(env, "tail_call is only allowed in functions that return 'int'.\n"); | |
15972 | goto err_free; | |
15973 | } | |
15974 | ||
af2ac3e1 | 15975 | bpfptr_add(&urecord, urec_size); |
838e9690 YS |
15976 | } |
15977 | ||
8c1b6e69 | 15978 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
15979 | return 0; |
15980 | ||
c454a46b | 15981 | err_free: |
8c1b6e69 | 15982 | kfree(info_aux); |
838e9690 YS |
15983 | return ret; |
15984 | } | |
15985 | ||
ba64e7d8 YS |
15986 | static void adjust_btf_func(struct bpf_verifier_env *env) |
15987 | { | |
8c1b6e69 | 15988 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
15989 | int i; |
15990 | ||
8c1b6e69 | 15991 | if (!aux->func_info) |
ba64e7d8 YS |
15992 | return; |
15993 | ||
335d1c5b KKD |
15994 | /* func_info is not available for hidden subprogs */ |
15995 | for (i = 0; i < env->subprog_cnt - env->hidden_subprog_cnt; i++) | |
8c1b6e69 | 15996 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
15997 | } |
15998 | ||
1b773d00 | 15999 | #define MIN_BPF_LINEINFO_SIZE offsetofend(struct bpf_line_info, line_col) |
c454a46b MKL |
16000 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE |
16001 | ||
16002 | static int check_btf_line(struct bpf_verifier_env *env, | |
16003 | const union bpf_attr *attr, | |
af2ac3e1 | 16004 | bpfptr_t uattr) |
c454a46b MKL |
16005 | { |
16006 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
16007 | struct bpf_subprog_info *sub; | |
16008 | struct bpf_line_info *linfo; | |
16009 | struct bpf_prog *prog; | |
16010 | const struct btf *btf; | |
af2ac3e1 | 16011 | bpfptr_t ulinfo; |
c454a46b MKL |
16012 | int err; |
16013 | ||
16014 | nr_linfo = attr->line_info_cnt; | |
16015 | if (!nr_linfo) | |
16016 | return 0; | |
0e6491b5 BC |
16017 | if (nr_linfo > INT_MAX / sizeof(struct bpf_line_info)) |
16018 | return -EINVAL; | |
c454a46b MKL |
16019 | |
16020 | rec_size = attr->line_info_rec_size; | |
16021 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
16022 | rec_size > MAX_LINEINFO_REC_SIZE || | |
16023 | rec_size & (sizeof(u32) - 1)) | |
16024 | return -EINVAL; | |
16025 | ||
16026 | /* Need to zero it in case the userspace may | |
16027 | * pass in a smaller bpf_line_info object. | |
16028 | */ | |
16029 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
16030 | GFP_KERNEL | __GFP_NOWARN); | |
16031 | if (!linfo) | |
16032 | return -ENOMEM; | |
16033 | ||
16034 | prog = env->prog; | |
16035 | btf = prog->aux->btf; | |
16036 | ||
16037 | s = 0; | |
16038 | sub = env->subprog_info; | |
af2ac3e1 | 16039 | ulinfo = make_bpfptr(attr->line_info, uattr.is_kernel); |
c454a46b MKL |
16040 | expected_size = sizeof(struct bpf_line_info); |
16041 | ncopy = min_t(u32, expected_size, rec_size); | |
16042 | for (i = 0; i < nr_linfo; i++) { | |
16043 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
16044 | if (err) { | |
16045 | if (err == -E2BIG) { | |
16046 | verbose(env, "nonzero tailing record in line_info"); | |
af2ac3e1 AS |
16047 | if (copy_to_bpfptr_offset(uattr, |
16048 | offsetof(union bpf_attr, line_info_rec_size), | |
16049 | &expected_size, sizeof(expected_size))) | |
c454a46b MKL |
16050 | err = -EFAULT; |
16051 | } | |
16052 | goto err_free; | |
16053 | } | |
16054 | ||
af2ac3e1 | 16055 | if (copy_from_bpfptr(&linfo[i], ulinfo, ncopy)) { |
c454a46b MKL |
16056 | err = -EFAULT; |
16057 | goto err_free; | |
16058 | } | |
16059 | ||
16060 | /* | |
16061 | * Check insn_off to ensure | |
16062 | * 1) strictly increasing AND | |
16063 | * 2) bounded by prog->len | |
16064 | * | |
16065 | * The linfo[0].insn_off == 0 check logically falls into | |
16066 | * the later "missing bpf_line_info for func..." case | |
16067 | * because the first linfo[0].insn_off must be the | |
16068 | * first sub also and the first sub must have | |
16069 | * subprog_info[0].start == 0. | |
16070 | */ | |
16071 | if ((i && linfo[i].insn_off <= prev_offset) || | |
16072 | linfo[i].insn_off >= prog->len) { | |
16073 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
16074 | i, linfo[i].insn_off, prev_offset, | |
16075 | prog->len); | |
16076 | err = -EINVAL; | |
16077 | goto err_free; | |
16078 | } | |
16079 | ||
fdbaa0be MKL |
16080 | if (!prog->insnsi[linfo[i].insn_off].code) { |
16081 | verbose(env, | |
16082 | "Invalid insn code at line_info[%u].insn_off\n", | |
16083 | i); | |
16084 | err = -EINVAL; | |
16085 | goto err_free; | |
16086 | } | |
16087 | ||
23127b33 MKL |
16088 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
16089 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
16090 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
16091 | err = -EINVAL; | |
16092 | goto err_free; | |
16093 | } | |
16094 | ||
16095 | if (s != env->subprog_cnt) { | |
16096 | if (linfo[i].insn_off == sub[s].start) { | |
16097 | sub[s].linfo_idx = i; | |
16098 | s++; | |
16099 | } else if (sub[s].start < linfo[i].insn_off) { | |
16100 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
16101 | err = -EINVAL; | |
16102 | goto err_free; | |
16103 | } | |
16104 | } | |
16105 | ||
16106 | prev_offset = linfo[i].insn_off; | |
af2ac3e1 | 16107 | bpfptr_add(&ulinfo, rec_size); |
c454a46b MKL |
16108 | } |
16109 | ||
16110 | if (s != env->subprog_cnt) { | |
16111 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
16112 | env->subprog_cnt - s, s); | |
16113 | err = -EINVAL; | |
16114 | goto err_free; | |
16115 | } | |
16116 | ||
16117 | prog->aux->linfo = linfo; | |
16118 | prog->aux->nr_linfo = nr_linfo; | |
16119 | ||
16120 | return 0; | |
16121 | ||
16122 | err_free: | |
16123 | kvfree(linfo); | |
16124 | return err; | |
16125 | } | |
16126 | ||
fbd94c7a AS |
16127 | #define MIN_CORE_RELO_SIZE sizeof(struct bpf_core_relo) |
16128 | #define MAX_CORE_RELO_SIZE MAX_FUNCINFO_REC_SIZE | |
16129 | ||
16130 | static int check_core_relo(struct bpf_verifier_env *env, | |
16131 | const union bpf_attr *attr, | |
16132 | bpfptr_t uattr) | |
16133 | { | |
16134 | u32 i, nr_core_relo, ncopy, expected_size, rec_size; | |
16135 | struct bpf_core_relo core_relo = {}; | |
16136 | struct bpf_prog *prog = env->prog; | |
16137 | const struct btf *btf = prog->aux->btf; | |
16138 | struct bpf_core_ctx ctx = { | |
16139 | .log = &env->log, | |
16140 | .btf = btf, | |
16141 | }; | |
16142 | bpfptr_t u_core_relo; | |
16143 | int err; | |
16144 | ||
16145 | nr_core_relo = attr->core_relo_cnt; | |
16146 | if (!nr_core_relo) | |
16147 | return 0; | |
16148 | if (nr_core_relo > INT_MAX / sizeof(struct bpf_core_relo)) | |
16149 | return -EINVAL; | |
16150 | ||
16151 | rec_size = attr->core_relo_rec_size; | |
16152 | if (rec_size < MIN_CORE_RELO_SIZE || | |
16153 | rec_size > MAX_CORE_RELO_SIZE || | |
16154 | rec_size % sizeof(u32)) | |
16155 | return -EINVAL; | |
16156 | ||
16157 | u_core_relo = make_bpfptr(attr->core_relos, uattr.is_kernel); | |
16158 | expected_size = sizeof(struct bpf_core_relo); | |
16159 | ncopy = min_t(u32, expected_size, rec_size); | |
16160 | ||
16161 | /* Unlike func_info and line_info, copy and apply each CO-RE | |
16162 | * relocation record one at a time. | |
16163 | */ | |
16164 | for (i = 0; i < nr_core_relo; i++) { | |
16165 | /* future proofing when sizeof(bpf_core_relo) changes */ | |
16166 | err = bpf_check_uarg_tail_zero(u_core_relo, expected_size, rec_size); | |
16167 | if (err) { | |
16168 | if (err == -E2BIG) { | |
16169 | verbose(env, "nonzero tailing record in core_relo"); | |
16170 | if (copy_to_bpfptr_offset(uattr, | |
16171 | offsetof(union bpf_attr, core_relo_rec_size), | |
16172 | &expected_size, sizeof(expected_size))) | |
16173 | err = -EFAULT; | |
16174 | } | |
16175 | break; | |
16176 | } | |
16177 | ||
16178 | if (copy_from_bpfptr(&core_relo, u_core_relo, ncopy)) { | |
16179 | err = -EFAULT; | |
16180 | break; | |
16181 | } | |
16182 | ||
16183 | if (core_relo.insn_off % 8 || core_relo.insn_off / 8 >= prog->len) { | |
16184 | verbose(env, "Invalid core_relo[%u].insn_off:%u prog->len:%u\n", | |
16185 | i, core_relo.insn_off, prog->len); | |
16186 | err = -EINVAL; | |
16187 | break; | |
16188 | } | |
16189 | ||
16190 | err = bpf_core_apply(&ctx, &core_relo, i, | |
16191 | &prog->insnsi[core_relo.insn_off / 8]); | |
16192 | if (err) | |
16193 | break; | |
16194 | bpfptr_add(&u_core_relo, rec_size); | |
16195 | } | |
16196 | return err; | |
16197 | } | |
16198 | ||
aaa619eb KKD |
16199 | static int check_btf_info_early(struct bpf_verifier_env *env, |
16200 | const union bpf_attr *attr, | |
16201 | bpfptr_t uattr) | |
c454a46b MKL |
16202 | { |
16203 | struct btf *btf; | |
16204 | int err; | |
16205 | ||
09b28d76 AS |
16206 | if (!attr->func_info_cnt && !attr->line_info_cnt) { |
16207 | if (check_abnormal_return(env)) | |
16208 | return -EINVAL; | |
c454a46b | 16209 | return 0; |
09b28d76 | 16210 | } |
c454a46b MKL |
16211 | |
16212 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
16213 | if (IS_ERR(btf)) | |
16214 | return PTR_ERR(btf); | |
350a5c4d AS |
16215 | if (btf_is_kernel(btf)) { |
16216 | btf_put(btf); | |
16217 | return -EACCES; | |
16218 | } | |
c454a46b MKL |
16219 | env->prog->aux->btf = btf; |
16220 | ||
aaa619eb KKD |
16221 | err = check_btf_func_early(env, attr, uattr); |
16222 | if (err) | |
16223 | return err; | |
16224 | return 0; | |
16225 | } | |
16226 | ||
16227 | static int check_btf_info(struct bpf_verifier_env *env, | |
16228 | const union bpf_attr *attr, | |
16229 | bpfptr_t uattr) | |
16230 | { | |
16231 | int err; | |
16232 | ||
16233 | if (!attr->func_info_cnt && !attr->line_info_cnt) { | |
16234 | if (check_abnormal_return(env)) | |
16235 | return -EINVAL; | |
16236 | return 0; | |
16237 | } | |
16238 | ||
c454a46b MKL |
16239 | err = check_btf_func(env, attr, uattr); |
16240 | if (err) | |
16241 | return err; | |
16242 | ||
16243 | err = check_btf_line(env, attr, uattr); | |
16244 | if (err) | |
16245 | return err; | |
16246 | ||
fbd94c7a AS |
16247 | err = check_core_relo(env, attr, uattr); |
16248 | if (err) | |
16249 | return err; | |
16250 | ||
c454a46b | 16251 | return 0; |
ba64e7d8 YS |
16252 | } |
16253 | ||
f1174f77 EC |
16254 | /* check %cur's range satisfies %old's */ |
16255 | static bool range_within(struct bpf_reg_state *old, | |
16256 | struct bpf_reg_state *cur) | |
16257 | { | |
b03c9f9f EC |
16258 | return old->umin_value <= cur->umin_value && |
16259 | old->umax_value >= cur->umax_value && | |
16260 | old->smin_value <= cur->smin_value && | |
fd675184 DB |
16261 | old->smax_value >= cur->smax_value && |
16262 | old->u32_min_value <= cur->u32_min_value && | |
16263 | old->u32_max_value >= cur->u32_max_value && | |
16264 | old->s32_min_value <= cur->s32_min_value && | |
16265 | old->s32_max_value >= cur->s32_max_value; | |
f1174f77 EC |
16266 | } |
16267 | ||
f1174f77 EC |
16268 | /* If in the old state two registers had the same id, then they need to have |
16269 | * the same id in the new state as well. But that id could be different from | |
16270 | * the old state, so we need to track the mapping from old to new ids. | |
16271 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
16272 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
16273 | * regs with a different old id could still have new id 9, we don't care about | |
16274 | * that. | |
16275 | * So we look through our idmap to see if this old id has been seen before. If | |
16276 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 16277 | */ |
1ffc85d9 | 16278 | static bool check_ids(u32 old_id, u32 cur_id, struct bpf_idmap *idmap) |
969bf05e | 16279 | { |
1ffc85d9 | 16280 | struct bpf_id_pair *map = idmap->map; |
f1174f77 | 16281 | unsigned int i; |
969bf05e | 16282 | |
4633a006 AN |
16283 | /* either both IDs should be set or both should be zero */ |
16284 | if (!!old_id != !!cur_id) | |
16285 | return false; | |
16286 | ||
16287 | if (old_id == 0) /* cur_id == 0 as well */ | |
16288 | return true; | |
16289 | ||
c9e73e3d | 16290 | for (i = 0; i < BPF_ID_MAP_SIZE; i++) { |
1ffc85d9 | 16291 | if (!map[i].old) { |
f1174f77 | 16292 | /* Reached an empty slot; haven't seen this id before */ |
1ffc85d9 EZ |
16293 | map[i].old = old_id; |
16294 | map[i].cur = cur_id; | |
f1174f77 EC |
16295 | return true; |
16296 | } | |
1ffc85d9 EZ |
16297 | if (map[i].old == old_id) |
16298 | return map[i].cur == cur_id; | |
16299 | if (map[i].cur == cur_id) | |
16300 | return false; | |
f1174f77 EC |
16301 | } |
16302 | /* We ran out of idmap slots, which should be impossible */ | |
16303 | WARN_ON_ONCE(1); | |
16304 | return false; | |
16305 | } | |
16306 | ||
1ffc85d9 EZ |
16307 | /* Similar to check_ids(), but allocate a unique temporary ID |
16308 | * for 'old_id' or 'cur_id' of zero. | |
16309 | * This makes pairs like '0 vs unique ID', 'unique ID vs 0' valid. | |
16310 | */ | |
16311 | static bool check_scalar_ids(u32 old_id, u32 cur_id, struct bpf_idmap *idmap) | |
16312 | { | |
16313 | old_id = old_id ? old_id : ++idmap->tmp_id_gen; | |
16314 | cur_id = cur_id ? cur_id : ++idmap->tmp_id_gen; | |
16315 | ||
16316 | return check_ids(old_id, cur_id, idmap); | |
16317 | } | |
16318 | ||
9242b5f5 AS |
16319 | static void clean_func_state(struct bpf_verifier_env *env, |
16320 | struct bpf_func_state *st) | |
16321 | { | |
16322 | enum bpf_reg_liveness live; | |
16323 | int i, j; | |
16324 | ||
16325 | for (i = 0; i < BPF_REG_FP; i++) { | |
16326 | live = st->regs[i].live; | |
16327 | /* liveness must not touch this register anymore */ | |
16328 | st->regs[i].live |= REG_LIVE_DONE; | |
16329 | if (!(live & REG_LIVE_READ)) | |
16330 | /* since the register is unused, clear its state | |
16331 | * to make further comparison simpler | |
16332 | */ | |
f54c7898 | 16333 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
16334 | } |
16335 | ||
16336 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
16337 | live = st->stack[i].spilled_ptr.live; | |
16338 | /* liveness must not touch this stack slot anymore */ | |
16339 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
16340 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 16341 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
16342 | for (j = 0; j < BPF_REG_SIZE; j++) |
16343 | st->stack[i].slot_type[j] = STACK_INVALID; | |
16344 | } | |
16345 | } | |
16346 | } | |
16347 | ||
16348 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
16349 | struct bpf_verifier_state *st) | |
16350 | { | |
16351 | int i; | |
16352 | ||
16353 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
16354 | /* all regs in this state in all frames were already marked */ | |
16355 | return; | |
16356 | ||
16357 | for (i = 0; i <= st->curframe; i++) | |
16358 | clean_func_state(env, st->frame[i]); | |
16359 | } | |
16360 | ||
16361 | /* the parentage chains form a tree. | |
16362 | * the verifier states are added to state lists at given insn and | |
16363 | * pushed into state stack for future exploration. | |
16364 | * when the verifier reaches bpf_exit insn some of the verifer states | |
16365 | * stored in the state lists have their final liveness state already, | |
16366 | * but a lot of states will get revised from liveness point of view when | |
16367 | * the verifier explores other branches. | |
16368 | * Example: | |
16369 | * 1: r0 = 1 | |
16370 | * 2: if r1 == 100 goto pc+1 | |
16371 | * 3: r0 = 2 | |
16372 | * 4: exit | |
16373 | * when the verifier reaches exit insn the register r0 in the state list of | |
16374 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
16375 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
16376 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
16377 | * | |
16378 | * Since the verifier pushes the branch states as it sees them while exploring | |
16379 | * the program the condition of walking the branch instruction for the second | |
16380 | * time means that all states below this branch were already explored and | |
8fb33b60 | 16381 | * their final liveness marks are already propagated. |
9242b5f5 AS |
16382 | * Hence when the verifier completes the search of state list in is_state_visited() |
16383 | * we can call this clean_live_states() function to mark all liveness states | |
16384 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
16385 | * will not be used. | |
16386 | * This function also clears the registers and stack for states that !READ | |
16387 | * to simplify state merging. | |
16388 | * | |
16389 | * Important note here that walking the same branch instruction in the callee | |
16390 | * doesn't meant that the states are DONE. The verifier has to compare | |
16391 | * the callsites | |
16392 | */ | |
16393 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
16394 | struct bpf_verifier_state *cur) | |
16395 | { | |
16396 | struct bpf_verifier_state_list *sl; | |
9242b5f5 | 16397 | |
5d839021 | 16398 | sl = *explored_state(env, insn); |
a8f500af | 16399 | while (sl) { |
2589726d AS |
16400 | if (sl->state.branches) |
16401 | goto next; | |
dc2a4ebc | 16402 | if (sl->state.insn_idx != insn || |
4c97259a | 16403 | !same_callsites(&sl->state, cur)) |
9242b5f5 | 16404 | goto next; |
9242b5f5 AS |
16405 | clean_verifier_state(env, &sl->state); |
16406 | next: | |
16407 | sl = sl->next; | |
16408 | } | |
16409 | } | |
16410 | ||
4a95c85c | 16411 | static bool regs_exact(const struct bpf_reg_state *rold, |
4633a006 | 16412 | const struct bpf_reg_state *rcur, |
1ffc85d9 | 16413 | struct bpf_idmap *idmap) |
4a95c85c | 16414 | { |
d2dcc67d | 16415 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && |
4633a006 AN |
16416 | check_ids(rold->id, rcur->id, idmap) && |
16417 | check_ids(rold->ref_obj_id, rcur->ref_obj_id, idmap); | |
4a95c85c AN |
16418 | } |
16419 | ||
f1174f77 | 16420 | /* Returns true if (rold safe implies rcur safe) */ |
e042aa53 | 16421 | static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold, |
2793a8b0 | 16422 | struct bpf_reg_state *rcur, struct bpf_idmap *idmap, bool exact) |
f1174f77 | 16423 | { |
2793a8b0 EZ |
16424 | if (exact) |
16425 | return regs_exact(rold, rcur, idmap); | |
16426 | ||
dc503a8a EC |
16427 | if (!(rold->live & REG_LIVE_READ)) |
16428 | /* explored state didn't use this */ | |
16429 | return true; | |
f1174f77 EC |
16430 | if (rold->type == NOT_INIT) |
16431 | /* explored state can't have used this */ | |
969bf05e | 16432 | return true; |
f1174f77 EC |
16433 | if (rcur->type == NOT_INIT) |
16434 | return false; | |
7f4ce97c | 16435 | |
910f6999 AN |
16436 | /* Enforce that register types have to match exactly, including their |
16437 | * modifiers (like PTR_MAYBE_NULL, MEM_RDONLY, etc), as a general | |
16438 | * rule. | |
16439 | * | |
16440 | * One can make a point that using a pointer register as unbounded | |
16441 | * SCALAR would be technically acceptable, but this could lead to | |
16442 | * pointer leaks because scalars are allowed to leak while pointers | |
16443 | * are not. We could make this safe in special cases if root is | |
16444 | * calling us, but it's probably not worth the hassle. | |
16445 | * | |
16446 | * Also, register types that are *not* MAYBE_NULL could technically be | |
16447 | * safe to use as their MAYBE_NULL variants (e.g., PTR_TO_MAP_VALUE | |
16448 | * is safe to be used as PTR_TO_MAP_VALUE_OR_NULL, provided both point | |
16449 | * to the same map). | |
7f4ce97c AN |
16450 | * However, if the old MAYBE_NULL register then got NULL checked, |
16451 | * doing so could have affected others with the same id, and we can't | |
16452 | * check for that because we lost the id when we converted to | |
16453 | * a non-MAYBE_NULL variant. | |
16454 | * So, as a general rule we don't allow mixing MAYBE_NULL and | |
910f6999 | 16455 | * non-MAYBE_NULL registers as well. |
7f4ce97c | 16456 | */ |
910f6999 | 16457 | if (rold->type != rcur->type) |
7f4ce97c AN |
16458 | return false; |
16459 | ||
c25b2ae1 | 16460 | switch (base_type(rold->type)) { |
f1174f77 | 16461 | case SCALAR_VALUE: |
1ffc85d9 EZ |
16462 | if (env->explore_alu_limits) { |
16463 | /* explore_alu_limits disables tnum_in() and range_within() | |
16464 | * logic and requires everything to be strict | |
16465 | */ | |
16466 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
16467 | check_scalar_ids(rold->id, rcur->id, idmap); | |
16468 | } | |
910f6999 AN |
16469 | if (!rold->precise) |
16470 | return true; | |
1ffc85d9 EZ |
16471 | /* Why check_ids() for scalar registers? |
16472 | * | |
16473 | * Consider the following BPF code: | |
16474 | * 1: r6 = ... unbound scalar, ID=a ... | |
16475 | * 2: r7 = ... unbound scalar, ID=b ... | |
16476 | * 3: if (r6 > r7) goto +1 | |
16477 | * 4: r6 = r7 | |
16478 | * 5: if (r6 > X) goto ... | |
16479 | * 6: ... memory operation using r7 ... | |
16480 | * | |
16481 | * First verification path is [1-6]: | |
16482 | * - at (4) same bpf_reg_state::id (b) would be assigned to r6 and r7; | |
16483 | * - at (5) r6 would be marked <= X, find_equal_scalars() would also mark | |
16484 | * r7 <= X, because r6 and r7 share same id. | |
16485 | * Next verification path is [1-4, 6]. | |
16486 | * | |
16487 | * Instruction (6) would be reached in two states: | |
16488 | * I. r6{.id=b}, r7{.id=b} via path 1-6; | |
16489 | * II. r6{.id=a}, r7{.id=b} via path 1-4, 6. | |
16490 | * | |
16491 | * Use check_ids() to distinguish these states. | |
16492 | * --- | |
16493 | * Also verify that new value satisfies old value range knowledge. | |
16494 | */ | |
910f6999 | 16495 | return range_within(rold, rcur) && |
1ffc85d9 EZ |
16496 | tnum_in(rold->var_off, rcur->var_off) && |
16497 | check_scalar_ids(rold->id, rcur->id, idmap); | |
69c087ba | 16498 | case PTR_TO_MAP_KEY: |
f1174f77 | 16499 | case PTR_TO_MAP_VALUE: |
567da5d2 AN |
16500 | case PTR_TO_MEM: |
16501 | case PTR_TO_BUF: | |
16502 | case PTR_TO_TP_BUFFER: | |
1b688a19 EC |
16503 | /* If the new min/max/var_off satisfy the old ones and |
16504 | * everything else matches, we are OK. | |
1b688a19 | 16505 | */ |
a73bf9f2 | 16506 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, var_off)) == 0 && |
1b688a19 | 16507 | range_within(rold, rcur) && |
4ea2bb15 | 16508 | tnum_in(rold->var_off, rcur->var_off) && |
567da5d2 AN |
16509 | check_ids(rold->id, rcur->id, idmap) && |
16510 | check_ids(rold->ref_obj_id, rcur->ref_obj_id, idmap); | |
de8f3a83 | 16511 | case PTR_TO_PACKET_META: |
f1174f77 | 16512 | case PTR_TO_PACKET: |
f1174f77 EC |
16513 | /* We must have at least as much range as the old ptr |
16514 | * did, so that any accesses which were safe before are | |
16515 | * still safe. This is true even if old range < old off, | |
16516 | * since someone could have accessed through (ptr - k), or | |
16517 | * even done ptr -= k in a register, to get a safe access. | |
16518 | */ | |
16519 | if (rold->range > rcur->range) | |
16520 | return false; | |
16521 | /* If the offsets don't match, we can't trust our alignment; | |
16522 | * nor can we be sure that we won't fall out of range. | |
16523 | */ | |
16524 | if (rold->off != rcur->off) | |
16525 | return false; | |
16526 | /* id relations must be preserved */ | |
4633a006 | 16527 | if (!check_ids(rold->id, rcur->id, idmap)) |
f1174f77 EC |
16528 | return false; |
16529 | /* new val must satisfy old val knowledge */ | |
16530 | return range_within(rold, rcur) && | |
16531 | tnum_in(rold->var_off, rcur->var_off); | |
7c884339 EZ |
16532 | case PTR_TO_STACK: |
16533 | /* two stack pointers are equal only if they're pointing to | |
16534 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
f1174f77 | 16535 | */ |
4633a006 | 16536 | return regs_exact(rold, rcur, idmap) && rold->frameno == rcur->frameno; |
f1174f77 | 16537 | default: |
4633a006 | 16538 | return regs_exact(rold, rcur, idmap); |
f1174f77 | 16539 | } |
969bf05e AS |
16540 | } |
16541 | ||
e042aa53 | 16542 | static bool stacksafe(struct bpf_verifier_env *env, struct bpf_func_state *old, |
2793a8b0 | 16543 | struct bpf_func_state *cur, struct bpf_idmap *idmap, bool exact) |
638f5b90 AS |
16544 | { |
16545 | int i, spi; | |
16546 | ||
638f5b90 AS |
16547 | /* walk slots of the explored stack and ignore any additional |
16548 | * slots in the current stack, since explored(safe) state | |
16549 | * didn't use them | |
16550 | */ | |
16551 | for (i = 0; i < old->allocated_stack; i++) { | |
06accc87 AN |
16552 | struct bpf_reg_state *old_reg, *cur_reg; |
16553 | ||
638f5b90 AS |
16554 | spi = i / BPF_REG_SIZE; |
16555 | ||
2793a8b0 EZ |
16556 | if (exact && |
16557 | old->stack[spi].slot_type[i % BPF_REG_SIZE] != | |
16558 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
16559 | return false; | |
16560 | ||
16561 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ) && !exact) { | |
b233920c | 16562 | i += BPF_REG_SIZE - 1; |
cc2b14d5 | 16563 | /* explored state didn't use this */ |
fd05e57b | 16564 | continue; |
b233920c | 16565 | } |
cc2b14d5 | 16566 | |
638f5b90 AS |
16567 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
16568 | continue; | |
19e2dbb7 | 16569 | |
6715df8d EZ |
16570 | if (env->allow_uninit_stack && |
16571 | old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC) | |
16572 | continue; | |
16573 | ||
19e2dbb7 AS |
16574 | /* explored stack has more populated slots than current stack |
16575 | * and these slots were used | |
16576 | */ | |
16577 | if (i >= cur->allocated_stack) | |
16578 | return false; | |
16579 | ||
cc2b14d5 AS |
16580 | /* if old state was safe with misc data in the stack |
16581 | * it will be safe with zero-initialized stack. | |
16582 | * The opposite is not true | |
16583 | */ | |
16584 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
16585 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
16586 | continue; | |
638f5b90 AS |
16587 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
16588 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
16589 | /* Ex: old explored (safe) state has STACK_SPILL in | |
b8c1a309 | 16590 | * this stack slot, but current has STACK_MISC -> |
638f5b90 AS |
16591 | * this verifier states are not equivalent, |
16592 | * return false to continue verification of this path | |
16593 | */ | |
16594 | return false; | |
27113c59 | 16595 | if (i % BPF_REG_SIZE != BPF_REG_SIZE - 1) |
638f5b90 | 16596 | continue; |
d6fefa11 KKD |
16597 | /* Both old and cur are having same slot_type */ |
16598 | switch (old->stack[spi].slot_type[BPF_REG_SIZE - 1]) { | |
16599 | case STACK_SPILL: | |
638f5b90 AS |
16600 | /* when explored and current stack slot are both storing |
16601 | * spilled registers, check that stored pointers types | |
16602 | * are the same as well. | |
16603 | * Ex: explored safe path could have stored | |
16604 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
16605 | * but current path has stored: | |
16606 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
16607 | * such verifier states are not equivalent. | |
16608 | * return false to continue verification of this path | |
16609 | */ | |
d6fefa11 | 16610 | if (!regsafe(env, &old->stack[spi].spilled_ptr, |
2793a8b0 | 16611 | &cur->stack[spi].spilled_ptr, idmap, exact)) |
d6fefa11 KKD |
16612 | return false; |
16613 | break; | |
16614 | case STACK_DYNPTR: | |
d6fefa11 KKD |
16615 | old_reg = &old->stack[spi].spilled_ptr; |
16616 | cur_reg = &cur->stack[spi].spilled_ptr; | |
16617 | if (old_reg->dynptr.type != cur_reg->dynptr.type || | |
16618 | old_reg->dynptr.first_slot != cur_reg->dynptr.first_slot || | |
16619 | !check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap)) | |
16620 | return false; | |
16621 | break; | |
06accc87 AN |
16622 | case STACK_ITER: |
16623 | old_reg = &old->stack[spi].spilled_ptr; | |
16624 | cur_reg = &cur->stack[spi].spilled_ptr; | |
16625 | /* iter.depth is not compared between states as it | |
16626 | * doesn't matter for correctness and would otherwise | |
16627 | * prevent convergence; we maintain it only to prevent | |
16628 | * infinite loop check triggering, see | |
16629 | * iter_active_depths_differ() | |
16630 | */ | |
16631 | if (old_reg->iter.btf != cur_reg->iter.btf || | |
16632 | old_reg->iter.btf_id != cur_reg->iter.btf_id || | |
16633 | old_reg->iter.state != cur_reg->iter.state || | |
16634 | /* ignore {old_reg,cur_reg}->iter.depth, see above */ | |
16635 | !check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap)) | |
16636 | return false; | |
16637 | break; | |
d6fefa11 KKD |
16638 | case STACK_MISC: |
16639 | case STACK_ZERO: | |
16640 | case STACK_INVALID: | |
16641 | continue; | |
16642 | /* Ensure that new unhandled slot types return false by default */ | |
16643 | default: | |
638f5b90 | 16644 | return false; |
d6fefa11 | 16645 | } |
638f5b90 AS |
16646 | } |
16647 | return true; | |
16648 | } | |
16649 | ||
e8f55fcf | 16650 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur, |
1ffc85d9 | 16651 | struct bpf_idmap *idmap) |
fd978bf7 | 16652 | { |
e8f55fcf AN |
16653 | int i; |
16654 | ||
fd978bf7 JS |
16655 | if (old->acquired_refs != cur->acquired_refs) |
16656 | return false; | |
e8f55fcf AN |
16657 | |
16658 | for (i = 0; i < old->acquired_refs; i++) { | |
16659 | if (!check_ids(old->refs[i].id, cur->refs[i].id, idmap)) | |
16660 | return false; | |
16661 | } | |
16662 | ||
16663 | return true; | |
fd978bf7 JS |
16664 | } |
16665 | ||
f1bca824 AS |
16666 | /* compare two verifier states |
16667 | * | |
16668 | * all states stored in state_list are known to be valid, since | |
16669 | * verifier reached 'bpf_exit' instruction through them | |
16670 | * | |
16671 | * this function is called when verifier exploring different branches of | |
16672 | * execution popped from the state stack. If it sees an old state that has | |
16673 | * more strict register state and more strict stack state then this execution | |
16674 | * branch doesn't need to be explored further, since verifier already | |
16675 | * concluded that more strict state leads to valid finish. | |
16676 | * | |
16677 | * Therefore two states are equivalent if register state is more conservative | |
16678 | * and explored stack state is more conservative than the current one. | |
16679 | * Example: | |
16680 | * explored current | |
16681 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
16682 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
16683 | * | |
16684 | * In other words if current stack state (one being explored) has more | |
16685 | * valid slots than old one that already passed validation, it means | |
16686 | * the verifier can stop exploring and conclude that current state is valid too | |
16687 | * | |
16688 | * Similarly with registers. If explored state has register type as invalid | |
16689 | * whereas register type in current state is meaningful, it means that | |
16690 | * the current state will reach 'bpf_exit' instruction safely | |
16691 | */ | |
c9e73e3d | 16692 | static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_state *old, |
2793a8b0 | 16693 | struct bpf_func_state *cur, bool exact) |
f1bca824 AS |
16694 | { |
16695 | int i; | |
16696 | ||
c9e73e3d | 16697 | for (i = 0; i < MAX_BPF_REG; i++) |
e042aa53 | 16698 | if (!regsafe(env, &old->regs[i], &cur->regs[i], |
2793a8b0 | 16699 | &env->idmap_scratch, exact)) |
c9e73e3d | 16700 | return false; |
f1bca824 | 16701 | |
2793a8b0 | 16702 | if (!stacksafe(env, old, cur, &env->idmap_scratch, exact)) |
c9e73e3d | 16703 | return false; |
fd978bf7 | 16704 | |
1ffc85d9 | 16705 | if (!refsafe(old, cur, &env->idmap_scratch)) |
c9e73e3d LB |
16706 | return false; |
16707 | ||
16708 | return true; | |
f1bca824 AS |
16709 | } |
16710 | ||
2793a8b0 EZ |
16711 | static void reset_idmap_scratch(struct bpf_verifier_env *env) |
16712 | { | |
16713 | env->idmap_scratch.tmp_id_gen = env->id_gen; | |
16714 | memset(&env->idmap_scratch.map, 0, sizeof(env->idmap_scratch.map)); | |
16715 | } | |
16716 | ||
f4d7e40a AS |
16717 | static bool states_equal(struct bpf_verifier_env *env, |
16718 | struct bpf_verifier_state *old, | |
2793a8b0 EZ |
16719 | struct bpf_verifier_state *cur, |
16720 | bool exact) | |
f4d7e40a AS |
16721 | { |
16722 | int i; | |
16723 | ||
16724 | if (old->curframe != cur->curframe) | |
16725 | return false; | |
16726 | ||
2793a8b0 | 16727 | reset_idmap_scratch(env); |
5dd9cdbc | 16728 | |
979d63d5 DB |
16729 | /* Verification state from speculative execution simulation |
16730 | * must never prune a non-speculative execution one. | |
16731 | */ | |
16732 | if (old->speculative && !cur->speculative) | |
16733 | return false; | |
16734 | ||
4ea2bb15 EZ |
16735 | if (old->active_lock.ptr != cur->active_lock.ptr) |
16736 | return false; | |
16737 | ||
16738 | /* Old and cur active_lock's have to be either both present | |
16739 | * or both absent. | |
16740 | */ | |
16741 | if (!!old->active_lock.id != !!cur->active_lock.id) | |
16742 | return false; | |
16743 | ||
16744 | if (old->active_lock.id && | |
1ffc85d9 | 16745 | !check_ids(old->active_lock.id, cur->active_lock.id, &env->idmap_scratch)) |
d83525ca AS |
16746 | return false; |
16747 | ||
9bb00b28 | 16748 | if (old->active_rcu_lock != cur->active_rcu_lock) |
d83525ca AS |
16749 | return false; |
16750 | ||
f4d7e40a AS |
16751 | /* for states to be equal callsites have to be the same |
16752 | * and all frame states need to be equivalent | |
16753 | */ | |
16754 | for (i = 0; i <= old->curframe; i++) { | |
16755 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
16756 | return false; | |
2793a8b0 | 16757 | if (!func_states_equal(env, old->frame[i], cur->frame[i], exact)) |
f4d7e40a AS |
16758 | return false; |
16759 | } | |
16760 | return true; | |
16761 | } | |
16762 | ||
5327ed3d JW |
16763 | /* Return 0 if no propagation happened. Return negative error code if error |
16764 | * happened. Otherwise, return the propagated bit. | |
16765 | */ | |
55e7f3b5 JW |
16766 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
16767 | struct bpf_reg_state *reg, | |
16768 | struct bpf_reg_state *parent_reg) | |
16769 | { | |
5327ed3d JW |
16770 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
16771 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
16772 | int err; |
16773 | ||
5327ed3d JW |
16774 | /* When comes here, read flags of PARENT_REG or REG could be any of |
16775 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
16776 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
16777 | */ | |
16778 | if (parent_flag == REG_LIVE_READ64 || | |
16779 | /* Or if there is no read flag from REG. */ | |
16780 | !flag || | |
16781 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
16782 | parent_flag == flag) | |
55e7f3b5 JW |
16783 | return 0; |
16784 | ||
5327ed3d | 16785 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
16786 | if (err) |
16787 | return err; | |
16788 | ||
5327ed3d | 16789 | return flag; |
55e7f3b5 JW |
16790 | } |
16791 | ||
8e9cd9ce | 16792 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
16793 | * straight-line code between a state and its parent. When we arrive at an |
16794 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
16795 | * code, so read marks in the state must propagate to the parent regardless | |
16796 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 16797 | * in mark_reg_read() is for. |
8e9cd9ce | 16798 | */ |
f4d7e40a AS |
16799 | static int propagate_liveness(struct bpf_verifier_env *env, |
16800 | const struct bpf_verifier_state *vstate, | |
16801 | struct bpf_verifier_state *vparent) | |
dc503a8a | 16802 | { |
3f8cafa4 | 16803 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 16804 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 16805 | int i, frame, err = 0; |
dc503a8a | 16806 | |
f4d7e40a AS |
16807 | if (vparent->curframe != vstate->curframe) { |
16808 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
16809 | vparent->curframe, vstate->curframe); | |
16810 | return -EFAULT; | |
16811 | } | |
dc503a8a EC |
16812 | /* Propagate read liveness of registers... */ |
16813 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 16814 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
16815 | parent = vparent->frame[frame]; |
16816 | state = vstate->frame[frame]; | |
16817 | parent_reg = parent->regs; | |
16818 | state_reg = state->regs; | |
83d16312 JK |
16819 | /* We don't need to worry about FP liveness, it's read-only */ |
16820 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
16821 | err = propagate_liveness_reg(env, &state_reg[i], |
16822 | &parent_reg[i]); | |
5327ed3d | 16823 | if (err < 0) |
3f8cafa4 | 16824 | return err; |
5327ed3d JW |
16825 | if (err == REG_LIVE_READ64) |
16826 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 16827 | } |
f4d7e40a | 16828 | |
1b04aee7 | 16829 | /* Propagate stack slots. */ |
f4d7e40a AS |
16830 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
16831 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
16832 | parent_reg = &parent->stack[i].spilled_ptr; |
16833 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
16834 | err = propagate_liveness_reg(env, state_reg, |
16835 | parent_reg); | |
5327ed3d | 16836 | if (err < 0) |
3f8cafa4 | 16837 | return err; |
dc503a8a EC |
16838 | } |
16839 | } | |
5327ed3d | 16840 | return 0; |
dc503a8a EC |
16841 | } |
16842 | ||
a3ce685d AS |
16843 | /* find precise scalars in the previous equivalent state and |
16844 | * propagate them into the current state | |
16845 | */ | |
16846 | static int propagate_precision(struct bpf_verifier_env *env, | |
16847 | const struct bpf_verifier_state *old) | |
16848 | { | |
16849 | struct bpf_reg_state *state_reg; | |
16850 | struct bpf_func_state *state; | |
529409ea | 16851 | int i, err = 0, fr; |
f655badf | 16852 | bool first; |
a3ce685d | 16853 | |
529409ea AN |
16854 | for (fr = old->curframe; fr >= 0; fr--) { |
16855 | state = old->frame[fr]; | |
16856 | state_reg = state->regs; | |
f655badf | 16857 | first = true; |
529409ea AN |
16858 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { |
16859 | if (state_reg->type != SCALAR_VALUE || | |
52c2b005 AN |
16860 | !state_reg->precise || |
16861 | !(state_reg->live & REG_LIVE_READ)) | |
529409ea | 16862 | continue; |
f655badf AN |
16863 | if (env->log.level & BPF_LOG_LEVEL2) { |
16864 | if (first) | |
16865 | verbose(env, "frame %d: propagating r%d", fr, i); | |
16866 | else | |
16867 | verbose(env, ",r%d", i); | |
16868 | } | |
16869 | bt_set_frame_reg(&env->bt, fr, i); | |
16870 | first = false; | |
529409ea | 16871 | } |
a3ce685d | 16872 | |
529409ea AN |
16873 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
16874 | if (!is_spilled_reg(&state->stack[i])) | |
16875 | continue; | |
16876 | state_reg = &state->stack[i].spilled_ptr; | |
16877 | if (state_reg->type != SCALAR_VALUE || | |
52c2b005 AN |
16878 | !state_reg->precise || |
16879 | !(state_reg->live & REG_LIVE_READ)) | |
529409ea | 16880 | continue; |
f655badf AN |
16881 | if (env->log.level & BPF_LOG_LEVEL2) { |
16882 | if (first) | |
16883 | verbose(env, "frame %d: propagating fp%d", | |
16884 | fr, (-i - 1) * BPF_REG_SIZE); | |
16885 | else | |
16886 | verbose(env, ",fp%d", (-i - 1) * BPF_REG_SIZE); | |
16887 | } | |
16888 | bt_set_frame_slot(&env->bt, fr, i); | |
16889 | first = false; | |
529409ea | 16890 | } |
f655badf AN |
16891 | if (!first) |
16892 | verbose(env, "\n"); | |
a3ce685d | 16893 | } |
f655badf AN |
16894 | |
16895 | err = mark_chain_precision_batch(env); | |
16896 | if (err < 0) | |
16897 | return err; | |
16898 | ||
a3ce685d AS |
16899 | return 0; |
16900 | } | |
16901 | ||
2589726d AS |
16902 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
16903 | struct bpf_verifier_state *cur) | |
16904 | { | |
16905 | struct bpf_func_state *fold, *fcur; | |
16906 | int i, fr = cur->curframe; | |
16907 | ||
16908 | if (old->curframe != fr) | |
16909 | return false; | |
16910 | ||
16911 | fold = old->frame[fr]; | |
16912 | fcur = cur->frame[fr]; | |
16913 | for (i = 0; i < MAX_BPF_REG; i++) | |
16914 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
16915 | offsetof(struct bpf_reg_state, parent))) | |
16916 | return false; | |
16917 | return true; | |
16918 | } | |
16919 | ||
06accc87 AN |
16920 | static bool is_iter_next_insn(struct bpf_verifier_env *env, int insn_idx) |
16921 | { | |
16922 | return env->insn_aux_data[insn_idx].is_iter_next; | |
16923 | } | |
16924 | ||
16925 | /* is_state_visited() handles iter_next() (see process_iter_next_call() for | |
16926 | * terminology) calls specially: as opposed to bounded BPF loops, it *expects* | |
16927 | * states to match, which otherwise would look like an infinite loop. So while | |
16928 | * iter_next() calls are taken care of, we still need to be careful and | |
16929 | * prevent erroneous and too eager declaration of "ininite loop", when | |
16930 | * iterators are involved. | |
16931 | * | |
16932 | * Here's a situation in pseudo-BPF assembly form: | |
16933 | * | |
16934 | * 0: again: ; set up iter_next() call args | |
16935 | * 1: r1 = &it ; <CHECKPOINT HERE> | |
16936 | * 2: call bpf_iter_num_next ; this is iter_next() call | |
16937 | * 3: if r0 == 0 goto done | |
16938 | * 4: ... something useful here ... | |
16939 | * 5: goto again ; another iteration | |
16940 | * 6: done: | |
16941 | * 7: r1 = &it | |
16942 | * 8: call bpf_iter_num_destroy ; clean up iter state | |
16943 | * 9: exit | |
16944 | * | |
16945 | * This is a typical loop. Let's assume that we have a prune point at 1:, | |
16946 | * before we get to `call bpf_iter_num_next` (e.g., because of that `goto | |
16947 | * again`, assuming other heuristics don't get in a way). | |
16948 | * | |
16949 | * When we first time come to 1:, let's say we have some state X. We proceed | |
16950 | * to 2:, fork states, enqueue ACTIVE, validate NULL case successfully, exit. | |
16951 | * Now we come back to validate that forked ACTIVE state. We proceed through | |
16952 | * 3-5, come to goto, jump to 1:. Let's assume our state didn't change, so we | |
16953 | * are converging. But the problem is that we don't know that yet, as this | |
16954 | * convergence has to happen at iter_next() call site only. So if nothing is | |
16955 | * done, at 1: verifier will use bounded loop logic and declare infinite | |
16956 | * looping (and would be *technically* correct, if not for iterator's | |
16957 | * "eventual sticky NULL" contract, see process_iter_next_call()). But we | |
16958 | * don't want that. So what we do in process_iter_next_call() when we go on | |
16959 | * another ACTIVE iteration, we bump slot->iter.depth, to mark that it's | |
16960 | * a different iteration. So when we suspect an infinite loop, we additionally | |
16961 | * check if any of the *ACTIVE* iterator states depths differ. If yes, we | |
16962 | * pretend we are not looping and wait for next iter_next() call. | |
16963 | * | |
16964 | * This only applies to ACTIVE state. In DRAINED state we don't expect to | |
16965 | * loop, because that would actually mean infinite loop, as DRAINED state is | |
16966 | * "sticky", and so we'll keep returning into the same instruction with the | |
16967 | * same state (at least in one of possible code paths). | |
16968 | * | |
16969 | * This approach allows to keep infinite loop heuristic even in the face of | |
16970 | * active iterator. E.g., C snippet below is and will be detected as | |
16971 | * inifintely looping: | |
16972 | * | |
16973 | * struct bpf_iter_num it; | |
16974 | * int *p, x; | |
16975 | * | |
16976 | * bpf_iter_num_new(&it, 0, 10); | |
16977 | * while ((p = bpf_iter_num_next(&t))) { | |
16978 | * x = p; | |
16979 | * while (x--) {} // <<-- infinite loop here | |
16980 | * } | |
16981 | * | |
16982 | */ | |
16983 | static bool iter_active_depths_differ(struct bpf_verifier_state *old, struct bpf_verifier_state *cur) | |
16984 | { | |
16985 | struct bpf_reg_state *slot, *cur_slot; | |
16986 | struct bpf_func_state *state; | |
16987 | int i, fr; | |
16988 | ||
16989 | for (fr = old->curframe; fr >= 0; fr--) { | |
16990 | state = old->frame[fr]; | |
16991 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
16992 | if (state->stack[i].slot_type[0] != STACK_ITER) | |
16993 | continue; | |
16994 | ||
16995 | slot = &state->stack[i].spilled_ptr; | |
16996 | if (slot->iter.state != BPF_ITER_STATE_ACTIVE) | |
16997 | continue; | |
16998 | ||
16999 | cur_slot = &cur->frame[fr]->stack[i].spilled_ptr; | |
17000 | if (cur_slot->iter.depth != slot->iter.depth) | |
17001 | return true; | |
17002 | } | |
17003 | } | |
17004 | return false; | |
17005 | } | |
2589726d | 17006 | |
58e2af8b | 17007 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 17008 | { |
58e2af8b | 17009 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 17010 | struct bpf_verifier_state_list *sl, **pprev; |
2a099282 | 17011 | struct bpf_verifier_state *cur = env->cur_state, *new, *loop_entry; |
2793a8b0 | 17012 | int i, j, n, err, states_cnt = 0; |
4b5ce570 AN |
17013 | bool force_new_state = env->test_state_freq || is_force_checkpoint(env, insn_idx); |
17014 | bool add_new_state = force_new_state; | |
2a099282 | 17015 | bool force_exact; |
f1bca824 | 17016 | |
2589726d AS |
17017 | /* bpf progs typically have pruning point every 4 instructions |
17018 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
17019 | * Do not add new state for future pruning if the verifier hasn't seen | |
17020 | * at least 2 jumps and at least 8 instructions. | |
17021 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
17022 | * In tests that amounts to up to 50% reduction into total verifier | |
17023 | * memory consumption and 20% verifier time speedup. | |
17024 | */ | |
17025 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
17026 | env->insn_processed - env->prev_insn_processed >= 8) | |
17027 | add_new_state = true; | |
17028 | ||
a8f500af AS |
17029 | pprev = explored_state(env, insn_idx); |
17030 | sl = *pprev; | |
17031 | ||
9242b5f5 AS |
17032 | clean_live_states(env, insn_idx, cur); |
17033 | ||
a8f500af | 17034 | while (sl) { |
dc2a4ebc AS |
17035 | states_cnt++; |
17036 | if (sl->state.insn_idx != insn_idx) | |
17037 | goto next; | |
bfc6bb74 | 17038 | |
2589726d | 17039 | if (sl->state.branches) { |
bfc6bb74 AS |
17040 | struct bpf_func_state *frame = sl->state.frame[sl->state.curframe]; |
17041 | ||
17042 | if (frame->in_async_callback_fn && | |
17043 | frame->async_entry_cnt != cur->frame[cur->curframe]->async_entry_cnt) { | |
17044 | /* Different async_entry_cnt means that the verifier is | |
17045 | * processing another entry into async callback. | |
17046 | * Seeing the same state is not an indication of infinite | |
17047 | * loop or infinite recursion. | |
17048 | * But finding the same state doesn't mean that it's safe | |
17049 | * to stop processing the current state. The previous state | |
17050 | * hasn't yet reached bpf_exit, since state.branches > 0. | |
17051 | * Checking in_async_callback_fn alone is not enough either. | |
17052 | * Since the verifier still needs to catch infinite loops | |
17053 | * inside async callbacks. | |
17054 | */ | |
06accc87 AN |
17055 | goto skip_inf_loop_check; |
17056 | } | |
17057 | /* BPF open-coded iterators loop detection is special. | |
17058 | * states_maybe_looping() logic is too simplistic in detecting | |
17059 | * states that *might* be equivalent, because it doesn't know | |
17060 | * about ID remapping, so don't even perform it. | |
17061 | * See process_iter_next_call() and iter_active_depths_differ() | |
17062 | * for overview of the logic. When current and one of parent | |
17063 | * states are detected as equivalent, it's a good thing: we prove | |
17064 | * convergence and can stop simulating further iterations. | |
17065 | * It's safe to assume that iterator loop will finish, taking into | |
17066 | * account iter_next() contract of eventually returning | |
17067 | * sticky NULL result. | |
2793a8b0 EZ |
17068 | * |
17069 | * Note, that states have to be compared exactly in this case because | |
17070 | * read and precision marks might not be finalized inside the loop. | |
17071 | * E.g. as in the program below: | |
17072 | * | |
17073 | * 1. r7 = -16 | |
17074 | * 2. r6 = bpf_get_prandom_u32() | |
17075 | * 3. while (bpf_iter_num_next(&fp[-8])) { | |
17076 | * 4. if (r6 != 42) { | |
17077 | * 5. r7 = -32 | |
17078 | * 6. r6 = bpf_get_prandom_u32() | |
17079 | * 7. continue | |
17080 | * 8. } | |
17081 | * 9. r0 = r10 | |
17082 | * 10. r0 += r7 | |
17083 | * 11. r8 = *(u64 *)(r0 + 0) | |
17084 | * 12. r6 = bpf_get_prandom_u32() | |
17085 | * 13. } | |
17086 | * | |
17087 | * Here verifier would first visit path 1-3, create a checkpoint at 3 | |
17088 | * with r7=-16, continue to 4-7,3. Existing checkpoint at 3 does | |
17089 | * not have read or precision mark for r7 yet, thus inexact states | |
17090 | * comparison would discard current state with r7=-32 | |
17091 | * => unsafe memory access at 11 would not be caught. | |
06accc87 AN |
17092 | */ |
17093 | if (is_iter_next_insn(env, insn_idx)) { | |
2793a8b0 | 17094 | if (states_equal(env, &sl->state, cur, true)) { |
06accc87 AN |
17095 | struct bpf_func_state *cur_frame; |
17096 | struct bpf_reg_state *iter_state, *iter_reg; | |
17097 | int spi; | |
17098 | ||
17099 | cur_frame = cur->frame[cur->curframe]; | |
17100 | /* btf_check_iter_kfuncs() enforces that | |
17101 | * iter state pointer is always the first arg | |
17102 | */ | |
17103 | iter_reg = &cur_frame->regs[BPF_REG_1]; | |
17104 | /* current state is valid due to states_equal(), | |
17105 | * so we can assume valid iter and reg state, | |
17106 | * no need for extra (re-)validations | |
17107 | */ | |
17108 | spi = __get_spi(iter_reg->off + iter_reg->var_off.value); | |
17109 | iter_state = &func(env, iter_reg)->stack[spi].spilled_ptr; | |
2a099282 EZ |
17110 | if (iter_state->iter.state == BPF_ITER_STATE_ACTIVE) { |
17111 | update_loop_entry(cur, &sl->state); | |
06accc87 | 17112 | goto hit; |
2a099282 | 17113 | } |
06accc87 AN |
17114 | } |
17115 | goto skip_inf_loop_check; | |
17116 | } | |
ab5cfac1 EZ |
17117 | if (calls_callback(env, insn_idx)) { |
17118 | if (states_equal(env, &sl->state, cur, true)) | |
17119 | goto hit; | |
17120 | goto skip_inf_loop_check; | |
17121 | } | |
06accc87 AN |
17122 | /* attempt to detect infinite loop to avoid unnecessary doomed work */ |
17123 | if (states_maybe_looping(&sl->state, cur) && | |
2793a8b0 | 17124 | states_equal(env, &sl->state, cur, false) && |
ab5cfac1 EZ |
17125 | !iter_active_depths_differ(&sl->state, cur) && |
17126 | sl->state.callback_unroll_depth == cur->callback_unroll_depth) { | |
2589726d AS |
17127 | verbose_linfo(env, insn_idx, "; "); |
17128 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
b4d82395 EZ |
17129 | verbose(env, "cur state:"); |
17130 | print_verifier_state(env, cur->frame[cur->curframe], true); | |
17131 | verbose(env, "old state:"); | |
17132 | print_verifier_state(env, sl->state.frame[cur->curframe], true); | |
2589726d AS |
17133 | return -EINVAL; |
17134 | } | |
17135 | /* if the verifier is processing a loop, avoid adding new state | |
17136 | * too often, since different loop iterations have distinct | |
17137 | * states and may not help future pruning. | |
17138 | * This threshold shouldn't be too low to make sure that | |
17139 | * a loop with large bound will be rejected quickly. | |
17140 | * The most abusive loop will be: | |
17141 | * r1 += 1 | |
17142 | * if r1 < 1000000 goto pc-2 | |
17143 | * 1M insn_procssed limit / 100 == 10k peak states. | |
17144 | * This threshold shouldn't be too high either, since states | |
17145 | * at the end of the loop are likely to be useful in pruning. | |
17146 | */ | |
06accc87 | 17147 | skip_inf_loop_check: |
4b5ce570 | 17148 | if (!force_new_state && |
98ddcf38 | 17149 | env->jmps_processed - env->prev_jmps_processed < 20 && |
2589726d AS |
17150 | env->insn_processed - env->prev_insn_processed < 100) |
17151 | add_new_state = false; | |
17152 | goto miss; | |
17153 | } | |
2a099282 EZ |
17154 | /* If sl->state is a part of a loop and this loop's entry is a part of |
17155 | * current verification path then states have to be compared exactly. | |
17156 | * 'force_exact' is needed to catch the following case: | |
17157 | * | |
17158 | * initial Here state 'succ' was processed first, | |
17159 | * | it was eventually tracked to produce a | |
17160 | * V state identical to 'hdr'. | |
17161 | * .---------> hdr All branches from 'succ' had been explored | |
17162 | * | | and thus 'succ' has its .branches == 0. | |
17163 | * | V | |
17164 | * | .------... Suppose states 'cur' and 'succ' correspond | |
17165 | * | | | to the same instruction + callsites. | |
17166 | * | V V In such case it is necessary to check | |
17167 | * | ... ... if 'succ' and 'cur' are states_equal(). | |
17168 | * | | | If 'succ' and 'cur' are a part of the | |
17169 | * | V V same loop exact flag has to be set. | |
17170 | * | succ <- cur To check if that is the case, verify | |
17171 | * | | if loop entry of 'succ' is in current | |
17172 | * | V DFS path. | |
17173 | * | ... | |
17174 | * | | | |
17175 | * '----' | |
17176 | * | |
17177 | * Additional details are in the comment before get_loop_entry(). | |
17178 | */ | |
17179 | loop_entry = get_loop_entry(&sl->state); | |
17180 | force_exact = loop_entry && loop_entry->branches > 0; | |
17181 | if (states_equal(env, &sl->state, cur, force_exact)) { | |
17182 | if (force_exact) | |
17183 | update_loop_entry(cur, loop_entry); | |
06accc87 | 17184 | hit: |
9f4686c4 | 17185 | sl->hit_cnt++; |
f1bca824 | 17186 | /* reached equivalent register/stack state, |
dc503a8a EC |
17187 | * prune the search. |
17188 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
17189 | * If we have any write marks in env->cur_state, they |
17190 | * will prevent corresponding reads in the continuation | |
17191 | * from reaching our parent (an explored_state). Our | |
17192 | * own state will get the read marks recorded, but | |
17193 | * they'll be immediately forgotten as we're pruning | |
17194 | * this state and will pop a new one. | |
f1bca824 | 17195 | */ |
f4d7e40a | 17196 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
17197 | |
17198 | /* if previous state reached the exit with precision and | |
17199 | * current state is equivalent to it (except precsion marks) | |
17200 | * the precision needs to be propagated back in | |
17201 | * the current state. | |
17202 | */ | |
17203 | err = err ? : push_jmp_history(env, cur); | |
17204 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
17205 | if (err) |
17206 | return err; | |
f1bca824 | 17207 | return 1; |
dc503a8a | 17208 | } |
2589726d AS |
17209 | miss: |
17210 | /* when new state is not going to be added do not increase miss count. | |
17211 | * Otherwise several loop iterations will remove the state | |
17212 | * recorded earlier. The goal of these heuristics is to have | |
17213 | * states from some iterations of the loop (some in the beginning | |
17214 | * and some at the end) to help pruning. | |
17215 | */ | |
17216 | if (add_new_state) | |
17217 | sl->miss_cnt++; | |
9f4686c4 AS |
17218 | /* heuristic to determine whether this state is beneficial |
17219 | * to keep checking from state equivalence point of view. | |
17220 | * Higher numbers increase max_states_per_insn and verification time, | |
17221 | * but do not meaningfully decrease insn_processed. | |
2793a8b0 EZ |
17222 | * 'n' controls how many times state could miss before eviction. |
17223 | * Use bigger 'n' for checkpoints because evicting checkpoint states | |
17224 | * too early would hinder iterator convergence. | |
9f4686c4 | 17225 | */ |
2793a8b0 EZ |
17226 | n = is_force_checkpoint(env, insn_idx) && sl->state.branches > 0 ? 64 : 3; |
17227 | if (sl->miss_cnt > sl->hit_cnt * n + n) { | |
9f4686c4 AS |
17228 | /* the state is unlikely to be useful. Remove it to |
17229 | * speed up verification | |
17230 | */ | |
17231 | *pprev = sl->next; | |
2a099282 EZ |
17232 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE && |
17233 | !sl->state.used_as_loop_entry) { | |
2589726d AS |
17234 | u32 br = sl->state.branches; |
17235 | ||
17236 | WARN_ONCE(br, | |
17237 | "BUG live_done but branches_to_explore %d\n", | |
17238 | br); | |
9f4686c4 AS |
17239 | free_verifier_state(&sl->state, false); |
17240 | kfree(sl); | |
17241 | env->peak_states--; | |
17242 | } else { | |
17243 | /* cannot free this state, since parentage chain may | |
17244 | * walk it later. Add it for free_list instead to | |
17245 | * be freed at the end of verification | |
17246 | */ | |
17247 | sl->next = env->free_list; | |
17248 | env->free_list = sl; | |
17249 | } | |
17250 | sl = *pprev; | |
17251 | continue; | |
17252 | } | |
dc2a4ebc | 17253 | next: |
9f4686c4 AS |
17254 | pprev = &sl->next; |
17255 | sl = *pprev; | |
f1bca824 AS |
17256 | } |
17257 | ||
06ee7115 AS |
17258 | if (env->max_states_per_insn < states_cnt) |
17259 | env->max_states_per_insn = states_cnt; | |
17260 | ||
2c78ee89 | 17261 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
a095f421 | 17262 | return 0; |
ceefbc96 | 17263 | |
2589726d | 17264 | if (!add_new_state) |
a095f421 | 17265 | return 0; |
ceefbc96 | 17266 | |
2589726d AS |
17267 | /* There were no equivalent states, remember the current one. |
17268 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 17269 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 17270 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 17271 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
17272 | * again on the way to bpf_exit. |
17273 | * When looping the sl->state.branches will be > 0 and this state | |
17274 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 17275 | */ |
638f5b90 | 17276 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
17277 | if (!new_sl) |
17278 | return -ENOMEM; | |
06ee7115 AS |
17279 | env->total_states++; |
17280 | env->peak_states++; | |
2589726d AS |
17281 | env->prev_jmps_processed = env->jmps_processed; |
17282 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 | 17283 | |
7a830b53 AN |
17284 | /* forget precise markings we inherited, see __mark_chain_precision */ |
17285 | if (env->bpf_capable) | |
17286 | mark_all_scalars_imprecise(env, cur); | |
17287 | ||
f1bca824 | 17288 | /* add new state to the head of linked list */ |
679c782d EC |
17289 | new = &new_sl->state; |
17290 | err = copy_verifier_state(new, cur); | |
1969db47 | 17291 | if (err) { |
679c782d | 17292 | free_verifier_state(new, false); |
1969db47 AS |
17293 | kfree(new_sl); |
17294 | return err; | |
17295 | } | |
dc2a4ebc | 17296 | new->insn_idx = insn_idx; |
2589726d AS |
17297 | WARN_ONCE(new->branches != 1, |
17298 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 17299 | |
2589726d | 17300 | cur->parent = new; |
b5dc0163 | 17301 | cur->first_insn_idx = insn_idx; |
2793a8b0 | 17302 | cur->dfs_depth = new->dfs_depth + 1; |
b5dc0163 | 17303 | clear_jmp_history(cur); |
5d839021 AS |
17304 | new_sl->next = *explored_state(env, insn_idx); |
17305 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
17306 | /* connect new state to parentage chain. Current frame needs all |
17307 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
17308 | * to the stack implicitly by JITs) so in callers' frames connect just | |
17309 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
17310 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
17311 | * from callee with its full parentage chain, anyway. | |
17312 | */ | |
8e9cd9ce EC |
17313 | /* clear write marks in current state: the writes we did are not writes |
17314 | * our child did, so they don't screen off its reads from us. | |
17315 | * (There are no read marks in current state, because reads always mark | |
17316 | * their parent and current state never has children yet. Only | |
17317 | * explored_states can get read marks.) | |
17318 | */ | |
eea1c227 AS |
17319 | for (j = 0; j <= cur->curframe; j++) { |
17320 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
17321 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
17322 | for (i = 0; i < BPF_REG_FP; i++) | |
17323 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
17324 | } | |
f4d7e40a AS |
17325 | |
17326 | /* all stack frames are accessible from callee, clear them all */ | |
17327 | for (j = 0; j <= cur->curframe; j++) { | |
17328 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 17329 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 17330 | |
679c782d | 17331 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 17332 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
17333 | frame->stack[i].spilled_ptr.parent = |
17334 | &newframe->stack[i].spilled_ptr; | |
17335 | } | |
f4d7e40a | 17336 | } |
f1bca824 AS |
17337 | return 0; |
17338 | } | |
17339 | ||
c64b7983 JS |
17340 | /* Return true if it's OK to have the same insn return a different type. */ |
17341 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
17342 | { | |
c25b2ae1 | 17343 | switch (base_type(type)) { |
c64b7983 JS |
17344 | case PTR_TO_CTX: |
17345 | case PTR_TO_SOCKET: | |
46f8bc92 | 17346 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 17347 | case PTR_TO_TCP_SOCK: |
fada7fdc | 17348 | case PTR_TO_XDP_SOCK: |
2a02759e | 17349 | case PTR_TO_BTF_ID: |
c64b7983 JS |
17350 | return false; |
17351 | default: | |
17352 | return true; | |
17353 | } | |
17354 | } | |
17355 | ||
17356 | /* If an instruction was previously used with particular pointer types, then we | |
17357 | * need to be careful to avoid cases such as the below, where it may be ok | |
17358 | * for one branch accessing the pointer, but not ok for the other branch: | |
17359 | * | |
17360 | * R1 = sock_ptr | |
17361 | * goto X; | |
17362 | * ... | |
17363 | * R1 = some_other_valid_ptr; | |
17364 | * goto X; | |
17365 | * ... | |
17366 | * R2 = *(u32 *)(R1 + 0); | |
17367 | */ | |
17368 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
17369 | { | |
17370 | return src != prev && (!reg_type_mismatch_ok(src) || | |
17371 | !reg_type_mismatch_ok(prev)); | |
17372 | } | |
17373 | ||
0d80a619 EZ |
17374 | static int save_aux_ptr_type(struct bpf_verifier_env *env, enum bpf_reg_type type, |
17375 | bool allow_trust_missmatch) | |
17376 | { | |
17377 | enum bpf_reg_type *prev_type = &env->insn_aux_data[env->insn_idx].ptr_type; | |
17378 | ||
17379 | if (*prev_type == NOT_INIT) { | |
17380 | /* Saw a valid insn | |
17381 | * dst_reg = *(u32 *)(src_reg + off) | |
17382 | * save type to validate intersecting paths | |
17383 | */ | |
17384 | *prev_type = type; | |
17385 | } else if (reg_type_mismatch(type, *prev_type)) { | |
17386 | /* Abuser program is trying to use the same insn | |
17387 | * dst_reg = *(u32*) (src_reg + off) | |
17388 | * with different pointer types: | |
17389 | * src_reg == ctx in one branch and | |
17390 | * src_reg == stack|map in some other branch. | |
17391 | * Reject it. | |
17392 | */ | |
17393 | if (allow_trust_missmatch && | |
17394 | base_type(type) == PTR_TO_BTF_ID && | |
17395 | base_type(*prev_type) == PTR_TO_BTF_ID) { | |
17396 | /* | |
17397 | * Have to support a use case when one path through | |
17398 | * the program yields TRUSTED pointer while another | |
17399 | * is UNTRUSTED. Fallback to UNTRUSTED to generate | |
1f9a1ea8 | 17400 | * BPF_PROBE_MEM/BPF_PROBE_MEMSX. |
0d80a619 EZ |
17401 | */ |
17402 | *prev_type = PTR_TO_BTF_ID | PTR_UNTRUSTED; | |
17403 | } else { | |
17404 | verbose(env, "same insn cannot be used with different pointers\n"); | |
17405 | return -EINVAL; | |
17406 | } | |
17407 | } | |
17408 | ||
17409 | return 0; | |
17410 | } | |
17411 | ||
58e2af8b | 17412 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 17413 | { |
6f8a57cc | 17414 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 17415 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 17416 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 17417 | struct bpf_reg_state *regs; |
06ee7115 | 17418 | int insn_cnt = env->prog->len; |
17a52670 | 17419 | bool do_print_state = false; |
b5dc0163 | 17420 | int prev_insn_idx = -1; |
17a52670 | 17421 | |
17a52670 | 17422 | for (;;) { |
f18b03fa | 17423 | bool exception_exit = false; |
17a52670 AS |
17424 | struct bpf_insn *insn; |
17425 | u8 class; | |
17426 | int err; | |
17427 | ||
b5dc0163 | 17428 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 17429 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 17430 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 17431 | env->insn_idx, insn_cnt); |
17a52670 AS |
17432 | return -EFAULT; |
17433 | } | |
17434 | ||
c08435ec | 17435 | insn = &insns[env->insn_idx]; |
17a52670 AS |
17436 | class = BPF_CLASS(insn->code); |
17437 | ||
06ee7115 | 17438 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
17439 | verbose(env, |
17440 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 17441 | env->insn_processed); |
17a52670 AS |
17442 | return -E2BIG; |
17443 | } | |
17444 | ||
a095f421 AN |
17445 | state->last_insn_idx = env->prev_insn_idx; |
17446 | ||
17447 | if (is_prune_point(env, env->insn_idx)) { | |
17448 | err = is_state_visited(env, env->insn_idx); | |
17449 | if (err < 0) | |
17450 | return err; | |
17451 | if (err == 1) { | |
17452 | /* found equivalent state, can prune the search */ | |
17453 | if (env->log.level & BPF_LOG_LEVEL) { | |
17454 | if (do_print_state) | |
17455 | verbose(env, "\nfrom %d to %d%s: safe\n", | |
17456 | env->prev_insn_idx, env->insn_idx, | |
17457 | env->cur_state->speculative ? | |
17458 | " (speculative execution)" : ""); | |
17459 | else | |
17460 | verbose(env, "%d: safe\n", env->insn_idx); | |
17461 | } | |
17462 | goto process_bpf_exit; | |
f1bca824 | 17463 | } |
a095f421 AN |
17464 | } |
17465 | ||
17466 | if (is_jmp_point(env, env->insn_idx)) { | |
17467 | err = push_jmp_history(env, state); | |
17468 | if (err) | |
17469 | return err; | |
f1bca824 AS |
17470 | } |
17471 | ||
c3494801 AS |
17472 | if (signal_pending(current)) |
17473 | return -EAGAIN; | |
17474 | ||
3c2ce60b DB |
17475 | if (need_resched()) |
17476 | cond_resched(); | |
17477 | ||
2e576648 CL |
17478 | if (env->log.level & BPF_LOG_LEVEL2 && do_print_state) { |
17479 | verbose(env, "\nfrom %d to %d%s:", | |
17480 | env->prev_insn_idx, env->insn_idx, | |
17481 | env->cur_state->speculative ? | |
17482 | " (speculative execution)" : ""); | |
17483 | print_verifier_state(env, state->frame[state->curframe], true); | |
17a52670 AS |
17484 | do_print_state = false; |
17485 | } | |
17486 | ||
06ee7115 | 17487 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 | 17488 | const struct bpf_insn_cbs cbs = { |
e6ac2450 | 17489 | .cb_call = disasm_kfunc_name, |
7105e828 | 17490 | .cb_print = verbose, |
abe08840 | 17491 | .private_data = env, |
7105e828 DB |
17492 | }; |
17493 | ||
2e576648 CL |
17494 | if (verifier_state_scratched(env)) |
17495 | print_insn_state(env, state->frame[state->curframe]); | |
17496 | ||
c08435ec | 17497 | verbose_linfo(env, env->insn_idx, "; "); |
12166409 | 17498 | env->prev_log_pos = env->log.end_pos; |
c08435ec | 17499 | verbose(env, "%d: ", env->insn_idx); |
abe08840 | 17500 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
12166409 AN |
17501 | env->prev_insn_print_pos = env->log.end_pos - env->prev_log_pos; |
17502 | env->prev_log_pos = env->log.end_pos; | |
17a52670 AS |
17503 | } |
17504 | ||
9d03ebc7 | 17505 | if (bpf_prog_is_offloaded(env->prog->aux)) { |
c08435ec DB |
17506 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
17507 | env->prev_insn_idx); | |
cae1927c JK |
17508 | if (err) |
17509 | return err; | |
17510 | } | |
13a27dfc | 17511 | |
638f5b90 | 17512 | regs = cur_regs(env); |
fe9a5ca7 | 17513 | sanitize_mark_insn_seen(env); |
b5dc0163 | 17514 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 17515 | |
17a52670 | 17516 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 17517 | err = check_alu_op(env, insn); |
17a52670 AS |
17518 | if (err) |
17519 | return err; | |
17520 | ||
17521 | } else if (class == BPF_LDX) { | |
0d80a619 | 17522 | enum bpf_reg_type src_reg_type; |
9bac3d6d AS |
17523 | |
17524 | /* check for reserved fields is already done */ | |
17525 | ||
17a52670 | 17526 | /* check src operand */ |
dc503a8a | 17527 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
17528 | if (err) |
17529 | return err; | |
17530 | ||
dc503a8a | 17531 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
17532 | if (err) |
17533 | return err; | |
17534 | ||
725f9dcd AS |
17535 | src_reg_type = regs[insn->src_reg].type; |
17536 | ||
17a52670 AS |
17537 | /* check that memory (src_reg + off) is readable, |
17538 | * the state of dst_reg will be updated by this func | |
17539 | */ | |
c08435ec DB |
17540 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
17541 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 YS |
17542 | BPF_READ, insn->dst_reg, false, |
17543 | BPF_MODE(insn->code) == BPF_MEMSX); | |
17a52670 AS |
17544 | if (err) |
17545 | return err; | |
17546 | ||
0d80a619 EZ |
17547 | err = save_aux_ptr_type(env, src_reg_type, true); |
17548 | if (err) | |
17549 | return err; | |
17a52670 | 17550 | } else if (class == BPF_STX) { |
0d80a619 | 17551 | enum bpf_reg_type dst_reg_type; |
d691f9e8 | 17552 | |
91c960b0 BJ |
17553 | if (BPF_MODE(insn->code) == BPF_ATOMIC) { |
17554 | err = check_atomic(env, env->insn_idx, insn); | |
17a52670 AS |
17555 | if (err) |
17556 | return err; | |
c08435ec | 17557 | env->insn_idx++; |
17a52670 AS |
17558 | continue; |
17559 | } | |
17560 | ||
5ca419f2 BJ |
17561 | if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) { |
17562 | verbose(env, "BPF_STX uses reserved fields\n"); | |
17563 | return -EINVAL; | |
17564 | } | |
17565 | ||
17a52670 | 17566 | /* check src1 operand */ |
dc503a8a | 17567 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
17568 | if (err) |
17569 | return err; | |
17570 | /* check src2 operand */ | |
dc503a8a | 17571 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
17572 | if (err) |
17573 | return err; | |
17574 | ||
d691f9e8 AS |
17575 | dst_reg_type = regs[insn->dst_reg].type; |
17576 | ||
17a52670 | 17577 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
17578 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
17579 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 | 17580 | BPF_WRITE, insn->src_reg, false, false); |
17a52670 AS |
17581 | if (err) |
17582 | return err; | |
17583 | ||
0d80a619 EZ |
17584 | err = save_aux_ptr_type(env, dst_reg_type, false); |
17585 | if (err) | |
17586 | return err; | |
17a52670 | 17587 | } else if (class == BPF_ST) { |
0d80a619 EZ |
17588 | enum bpf_reg_type dst_reg_type; |
17589 | ||
17a52670 AS |
17590 | if (BPF_MODE(insn->code) != BPF_MEM || |
17591 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 17592 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
17593 | return -EINVAL; |
17594 | } | |
17595 | /* check src operand */ | |
dc503a8a | 17596 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
17597 | if (err) |
17598 | return err; | |
17599 | ||
0d80a619 | 17600 | dst_reg_type = regs[insn->dst_reg].type; |
f37a8cb8 | 17601 | |
17a52670 | 17602 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
17603 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
17604 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 | 17605 | BPF_WRITE, -1, false, false); |
17a52670 AS |
17606 | if (err) |
17607 | return err; | |
17608 | ||
0d80a619 EZ |
17609 | err = save_aux_ptr_type(env, dst_reg_type, false); |
17610 | if (err) | |
17611 | return err; | |
092ed096 | 17612 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
17613 | u8 opcode = BPF_OP(insn->code); |
17614 | ||
2589726d | 17615 | env->jmps_processed++; |
17a52670 AS |
17616 | if (opcode == BPF_CALL) { |
17617 | if (BPF_SRC(insn->code) != BPF_K || | |
2357672c KKD |
17618 | (insn->src_reg != BPF_PSEUDO_KFUNC_CALL |
17619 | && insn->off != 0) || | |
f4d7e40a | 17620 | (insn->src_reg != BPF_REG_0 && |
e6ac2450 MKL |
17621 | insn->src_reg != BPF_PSEUDO_CALL && |
17622 | insn->src_reg != BPF_PSEUDO_KFUNC_CALL) || | |
092ed096 JW |
17623 | insn->dst_reg != BPF_REG_0 || |
17624 | class == BPF_JMP32) { | |
61bd5218 | 17625 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
17626 | return -EINVAL; |
17627 | } | |
17628 | ||
8cab76ec KKD |
17629 | if (env->cur_state->active_lock.ptr) { |
17630 | if ((insn->src_reg == BPF_REG_0 && insn->imm != BPF_FUNC_spin_unlock) || | |
17631 | (insn->src_reg == BPF_PSEUDO_CALL) || | |
17632 | (insn->src_reg == BPF_PSEUDO_KFUNC_CALL && | |
cd6791b4 | 17633 | (insn->off != 0 || !is_bpf_graph_api_kfunc(insn->imm)))) { |
8cab76ec KKD |
17634 | verbose(env, "function calls are not allowed while holding a lock\n"); |
17635 | return -EINVAL; | |
17636 | } | |
d83525ca | 17637 | } |
f18b03fa | 17638 | if (insn->src_reg == BPF_PSEUDO_CALL) { |
c08435ec | 17639 | err = check_func_call(env, insn, &env->insn_idx); |
f18b03fa | 17640 | } else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
5c073f26 | 17641 | err = check_kfunc_call(env, insn, &env->insn_idx); |
f18b03fa KKD |
17642 | if (!err && is_bpf_throw_kfunc(insn)) { |
17643 | exception_exit = true; | |
17644 | goto process_bpf_exit_full; | |
17645 | } | |
17646 | } else { | |
69c087ba | 17647 | err = check_helper_call(env, insn, &env->insn_idx); |
f18b03fa | 17648 | } |
17a52670 AS |
17649 | if (err) |
17650 | return err; | |
553a64a8 AN |
17651 | |
17652 | mark_reg_scratched(env, BPF_REG_0); | |
17a52670 AS |
17653 | } else if (opcode == BPF_JA) { |
17654 | if (BPF_SRC(insn->code) != BPF_K || | |
17a52670 | 17655 | insn->src_reg != BPF_REG_0 || |
092ed096 | 17656 | insn->dst_reg != BPF_REG_0 || |
4cd58e9a YS |
17657 | (class == BPF_JMP && insn->imm != 0) || |
17658 | (class == BPF_JMP32 && insn->off != 0)) { | |
61bd5218 | 17659 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
17660 | return -EINVAL; |
17661 | } | |
17662 | ||
4cd58e9a YS |
17663 | if (class == BPF_JMP) |
17664 | env->insn_idx += insn->off + 1; | |
17665 | else | |
17666 | env->insn_idx += insn->imm + 1; | |
17a52670 AS |
17667 | continue; |
17668 | ||
17669 | } else if (opcode == BPF_EXIT) { | |
17670 | if (BPF_SRC(insn->code) != BPF_K || | |
17671 | insn->imm != 0 || | |
17672 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
17673 | insn->dst_reg != BPF_REG_0 || |
17674 | class == BPF_JMP32) { | |
61bd5218 | 17675 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
17676 | return -EINVAL; |
17677 | } | |
f18b03fa | 17678 | process_bpf_exit_full: |
5d92ddc3 DM |
17679 | if (env->cur_state->active_lock.ptr && |
17680 | !in_rbtree_lock_required_cb(env)) { | |
d83525ca AS |
17681 | verbose(env, "bpf_spin_unlock is missing\n"); |
17682 | return -EINVAL; | |
17683 | } | |
17684 | ||
0816b8c6 DM |
17685 | if (env->cur_state->active_rcu_lock && |
17686 | !in_rbtree_lock_required_cb(env)) { | |
9bb00b28 YS |
17687 | verbose(env, "bpf_rcu_read_unlock is missing\n"); |
17688 | return -EINVAL; | |
17689 | } | |
17690 | ||
9d9d00ac KKD |
17691 | /* We must do check_reference_leak here before |
17692 | * prepare_func_exit to handle the case when | |
17693 | * state->curframe > 0, it may be a callback | |
17694 | * function, for which reference_state must | |
17695 | * match caller reference state when it exits. | |
17696 | */ | |
f18b03fa | 17697 | err = check_reference_leak(env, exception_exit); |
9d9d00ac KKD |
17698 | if (err) |
17699 | return err; | |
17700 | ||
f18b03fa KKD |
17701 | /* The side effect of the prepare_func_exit |
17702 | * which is being skipped is that it frees | |
17703 | * bpf_func_state. Typically, process_bpf_exit | |
17704 | * will only be hit with outermost exit. | |
17705 | * copy_verifier_state in pop_stack will handle | |
17706 | * freeing of any extra bpf_func_state left over | |
17707 | * from not processing all nested function | |
17708 | * exits. We also skip return code checks as | |
17709 | * they are not needed for exceptional exits. | |
17710 | */ | |
17711 | if (exception_exit) | |
17712 | goto process_bpf_exit; | |
17713 | ||
f4d7e40a AS |
17714 | if (state->curframe) { |
17715 | /* exit from nested function */ | |
c08435ec | 17716 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
17717 | if (err) |
17718 | return err; | |
17719 | do_print_state = true; | |
17720 | continue; | |
17721 | } | |
17722 | ||
a923819f | 17723 | err = check_return_code(env, BPF_REG_0); |
390ee7e2 AS |
17724 | if (err) |
17725 | return err; | |
f1bca824 | 17726 | process_bpf_exit: |
0f55f9ed | 17727 | mark_verifier_state_scratched(env); |
2589726d | 17728 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 17729 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 17730 | &env->insn_idx, pop_log); |
638f5b90 AS |
17731 | if (err < 0) { |
17732 | if (err != -ENOENT) | |
17733 | return err; | |
17a52670 AS |
17734 | break; |
17735 | } else { | |
17736 | do_print_state = true; | |
17737 | continue; | |
17738 | } | |
17739 | } else { | |
c08435ec | 17740 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
17741 | if (err) |
17742 | return err; | |
17743 | } | |
17744 | } else if (class == BPF_LD) { | |
17745 | u8 mode = BPF_MODE(insn->code); | |
17746 | ||
17747 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
17748 | err = check_ld_abs(env, insn); |
17749 | if (err) | |
17750 | return err; | |
17751 | ||
17a52670 AS |
17752 | } else if (mode == BPF_IMM) { |
17753 | err = check_ld_imm(env, insn); | |
17754 | if (err) | |
17755 | return err; | |
17756 | ||
c08435ec | 17757 | env->insn_idx++; |
fe9a5ca7 | 17758 | sanitize_mark_insn_seen(env); |
17a52670 | 17759 | } else { |
61bd5218 | 17760 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
17761 | return -EINVAL; |
17762 | } | |
17763 | } else { | |
61bd5218 | 17764 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
17765 | return -EINVAL; |
17766 | } | |
17767 | ||
c08435ec | 17768 | env->insn_idx++; |
17a52670 AS |
17769 | } |
17770 | ||
17771 | return 0; | |
17772 | } | |
17773 | ||
541c3bad AN |
17774 | static int find_btf_percpu_datasec(struct btf *btf) |
17775 | { | |
17776 | const struct btf_type *t; | |
17777 | const char *tname; | |
17778 | int i, n; | |
17779 | ||
17780 | /* | |
17781 | * Both vmlinux and module each have their own ".data..percpu" | |
17782 | * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF | |
17783 | * types to look at only module's own BTF types. | |
17784 | */ | |
17785 | n = btf_nr_types(btf); | |
17786 | if (btf_is_module(btf)) | |
17787 | i = btf_nr_types(btf_vmlinux); | |
17788 | else | |
17789 | i = 1; | |
17790 | ||
17791 | for(; i < n; i++) { | |
17792 | t = btf_type_by_id(btf, i); | |
17793 | if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC) | |
17794 | continue; | |
17795 | ||
17796 | tname = btf_name_by_offset(btf, t->name_off); | |
17797 | if (!strcmp(tname, ".data..percpu")) | |
17798 | return i; | |
17799 | } | |
17800 | ||
17801 | return -ENOENT; | |
17802 | } | |
17803 | ||
4976b718 HL |
17804 | /* replace pseudo btf_id with kernel symbol address */ |
17805 | static int check_pseudo_btf_id(struct bpf_verifier_env *env, | |
17806 | struct bpf_insn *insn, | |
17807 | struct bpf_insn_aux_data *aux) | |
17808 | { | |
eaa6bcb7 HL |
17809 | const struct btf_var_secinfo *vsi; |
17810 | const struct btf_type *datasec; | |
541c3bad | 17811 | struct btf_mod_pair *btf_mod; |
4976b718 HL |
17812 | const struct btf_type *t; |
17813 | const char *sym_name; | |
eaa6bcb7 | 17814 | bool percpu = false; |
f16e6313 | 17815 | u32 type, id = insn->imm; |
541c3bad | 17816 | struct btf *btf; |
f16e6313 | 17817 | s32 datasec_id; |
4976b718 | 17818 | u64 addr; |
541c3bad | 17819 | int i, btf_fd, err; |
4976b718 | 17820 | |
541c3bad AN |
17821 | btf_fd = insn[1].imm; |
17822 | if (btf_fd) { | |
17823 | btf = btf_get_by_fd(btf_fd); | |
17824 | if (IS_ERR(btf)) { | |
17825 | verbose(env, "invalid module BTF object FD specified.\n"); | |
17826 | return -EINVAL; | |
17827 | } | |
17828 | } else { | |
17829 | if (!btf_vmlinux) { | |
17830 | verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n"); | |
17831 | return -EINVAL; | |
17832 | } | |
17833 | btf = btf_vmlinux; | |
17834 | btf_get(btf); | |
4976b718 HL |
17835 | } |
17836 | ||
541c3bad | 17837 | t = btf_type_by_id(btf, id); |
4976b718 HL |
17838 | if (!t) { |
17839 | verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id); | |
541c3bad AN |
17840 | err = -ENOENT; |
17841 | goto err_put; | |
4976b718 HL |
17842 | } |
17843 | ||
58aa2afb AS |
17844 | if (!btf_type_is_var(t) && !btf_type_is_func(t)) { |
17845 | verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR or KIND_FUNC\n", id); | |
541c3bad AN |
17846 | err = -EINVAL; |
17847 | goto err_put; | |
4976b718 HL |
17848 | } |
17849 | ||
541c3bad | 17850 | sym_name = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
17851 | addr = kallsyms_lookup_name(sym_name); |
17852 | if (!addr) { | |
17853 | verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n", | |
17854 | sym_name); | |
541c3bad AN |
17855 | err = -ENOENT; |
17856 | goto err_put; | |
4976b718 | 17857 | } |
58aa2afb AS |
17858 | insn[0].imm = (u32)addr; |
17859 | insn[1].imm = addr >> 32; | |
17860 | ||
17861 | if (btf_type_is_func(t)) { | |
17862 | aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY; | |
17863 | aux->btf_var.mem_size = 0; | |
17864 | goto check_btf; | |
17865 | } | |
4976b718 | 17866 | |
541c3bad | 17867 | datasec_id = find_btf_percpu_datasec(btf); |
eaa6bcb7 | 17868 | if (datasec_id > 0) { |
541c3bad | 17869 | datasec = btf_type_by_id(btf, datasec_id); |
eaa6bcb7 HL |
17870 | for_each_vsi(i, datasec, vsi) { |
17871 | if (vsi->type == id) { | |
17872 | percpu = true; | |
17873 | break; | |
17874 | } | |
17875 | } | |
17876 | } | |
17877 | ||
4976b718 | 17878 | type = t->type; |
541c3bad | 17879 | t = btf_type_skip_modifiers(btf, type, NULL); |
eaa6bcb7 | 17880 | if (percpu) { |
5844101a | 17881 | aux->btf_var.reg_type = PTR_TO_BTF_ID | MEM_PERCPU; |
541c3bad | 17882 | aux->btf_var.btf = btf; |
eaa6bcb7 HL |
17883 | aux->btf_var.btf_id = type; |
17884 | } else if (!btf_type_is_struct(t)) { | |
4976b718 HL |
17885 | const struct btf_type *ret; |
17886 | const char *tname; | |
17887 | u32 tsize; | |
17888 | ||
17889 | /* resolve the type size of ksym. */ | |
541c3bad | 17890 | ret = btf_resolve_size(btf, t, &tsize); |
4976b718 | 17891 | if (IS_ERR(ret)) { |
541c3bad | 17892 | tname = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
17893 | verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n", |
17894 | tname, PTR_ERR(ret)); | |
541c3bad AN |
17895 | err = -EINVAL; |
17896 | goto err_put; | |
4976b718 | 17897 | } |
34d3a78c | 17898 | aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY; |
4976b718 HL |
17899 | aux->btf_var.mem_size = tsize; |
17900 | } else { | |
17901 | aux->btf_var.reg_type = PTR_TO_BTF_ID; | |
541c3bad | 17902 | aux->btf_var.btf = btf; |
4976b718 HL |
17903 | aux->btf_var.btf_id = type; |
17904 | } | |
58aa2afb | 17905 | check_btf: |
541c3bad AN |
17906 | /* check whether we recorded this BTF (and maybe module) already */ |
17907 | for (i = 0; i < env->used_btf_cnt; i++) { | |
17908 | if (env->used_btfs[i].btf == btf) { | |
17909 | btf_put(btf); | |
17910 | return 0; | |
17911 | } | |
17912 | } | |
17913 | ||
17914 | if (env->used_btf_cnt >= MAX_USED_BTFS) { | |
17915 | err = -E2BIG; | |
17916 | goto err_put; | |
17917 | } | |
17918 | ||
17919 | btf_mod = &env->used_btfs[env->used_btf_cnt]; | |
17920 | btf_mod->btf = btf; | |
17921 | btf_mod->module = NULL; | |
17922 | ||
17923 | /* if we reference variables from kernel module, bump its refcount */ | |
17924 | if (btf_is_module(btf)) { | |
17925 | btf_mod->module = btf_try_get_module(btf); | |
17926 | if (!btf_mod->module) { | |
17927 | err = -ENXIO; | |
17928 | goto err_put; | |
17929 | } | |
17930 | } | |
17931 | ||
17932 | env->used_btf_cnt++; | |
17933 | ||
4976b718 | 17934 | return 0; |
541c3bad AN |
17935 | err_put: |
17936 | btf_put(btf); | |
17937 | return err; | |
4976b718 HL |
17938 | } |
17939 | ||
d83525ca AS |
17940 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
17941 | { | |
17942 | switch (type) { | |
17943 | case BPF_PROG_TYPE_KPROBE: | |
17944 | case BPF_PROG_TYPE_TRACEPOINT: | |
17945 | case BPF_PROG_TYPE_PERF_EVENT: | |
17946 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
5002615a | 17947 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: |
d83525ca AS |
17948 | return true; |
17949 | default: | |
17950 | return false; | |
17951 | } | |
17952 | } | |
17953 | ||
61bd5218 JK |
17954 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
17955 | struct bpf_map *map, | |
fdc15d38 AS |
17956 | struct bpf_prog *prog) |
17957 | ||
17958 | { | |
7e40781c | 17959 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
a3884572 | 17960 | |
9c395c1b DM |
17961 | if (btf_record_has_field(map->record, BPF_LIST_HEAD) || |
17962 | btf_record_has_field(map->record, BPF_RB_ROOT)) { | |
f0c5941f | 17963 | if (is_tracing_prog_type(prog_type)) { |
9c395c1b | 17964 | verbose(env, "tracing progs cannot use bpf_{list_head,rb_root} yet\n"); |
f0c5941f KKD |
17965 | return -EINVAL; |
17966 | } | |
17967 | } | |
17968 | ||
db559117 | 17969 | if (btf_record_has_field(map->record, BPF_SPIN_LOCK)) { |
9e7a4d98 KS |
17970 | if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) { |
17971 | verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n"); | |
17972 | return -EINVAL; | |
17973 | } | |
17974 | ||
17975 | if (is_tracing_prog_type(prog_type)) { | |
17976 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
17977 | return -EINVAL; | |
17978 | } | |
d83525ca AS |
17979 | } |
17980 | ||
db559117 | 17981 | if (btf_record_has_field(map->record, BPF_TIMER)) { |
5e0bc308 DB |
17982 | if (is_tracing_prog_type(prog_type)) { |
17983 | verbose(env, "tracing progs cannot use bpf_timer yet\n"); | |
17984 | return -EINVAL; | |
17985 | } | |
17986 | } | |
17987 | ||
9d03ebc7 | 17988 | if ((bpf_prog_is_offloaded(prog->aux) || bpf_map_is_offloaded(map)) && |
09728266 | 17989 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
17990 | verbose(env, "offload device mismatch between prog and map\n"); |
17991 | return -EINVAL; | |
17992 | } | |
17993 | ||
85d33df3 MKL |
17994 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
17995 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
17996 | return -EINVAL; | |
17997 | } | |
17998 | ||
1e6c62a8 AS |
17999 | if (prog->aux->sleepable) |
18000 | switch (map->map_type) { | |
18001 | case BPF_MAP_TYPE_HASH: | |
18002 | case BPF_MAP_TYPE_LRU_HASH: | |
18003 | case BPF_MAP_TYPE_ARRAY: | |
638e4b82 AS |
18004 | case BPF_MAP_TYPE_PERCPU_HASH: |
18005 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
18006 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
18007 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
18008 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
ba90c2cc | 18009 | case BPF_MAP_TYPE_RINGBUF: |
583c1f42 | 18010 | case BPF_MAP_TYPE_USER_RINGBUF: |
0fe4b381 KS |
18011 | case BPF_MAP_TYPE_INODE_STORAGE: |
18012 | case BPF_MAP_TYPE_SK_STORAGE: | |
18013 | case BPF_MAP_TYPE_TASK_STORAGE: | |
2c40d97d | 18014 | case BPF_MAP_TYPE_CGRP_STORAGE: |
ba90c2cc | 18015 | break; |
1e6c62a8 AS |
18016 | default: |
18017 | verbose(env, | |
2c40d97d | 18018 | "Sleepable programs can only use array, hash, ringbuf and local storage maps\n"); |
1e6c62a8 AS |
18019 | return -EINVAL; |
18020 | } | |
18021 | ||
fdc15d38 AS |
18022 | return 0; |
18023 | } | |
18024 | ||
b741f163 RG |
18025 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
18026 | { | |
18027 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
18028 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
18029 | } | |
18030 | ||
4976b718 HL |
18031 | /* find and rewrite pseudo imm in ld_imm64 instructions: |
18032 | * | |
18033 | * 1. if it accesses map FD, replace it with actual map pointer. | |
18034 | * 2. if it accesses btf_id of a VAR, replace it with pointer to the var. | |
18035 | * | |
18036 | * NOTE: btf_vmlinux is required for converting pseudo btf_id. | |
0246e64d | 18037 | */ |
4976b718 | 18038 | static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env) |
0246e64d AS |
18039 | { |
18040 | struct bpf_insn *insn = env->prog->insnsi; | |
18041 | int insn_cnt = env->prog->len; | |
fdc15d38 | 18042 | int i, j, err; |
0246e64d | 18043 | |
f1f7714e | 18044 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
18045 | if (err) |
18046 | return err; | |
18047 | ||
0246e64d | 18048 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 18049 | if (BPF_CLASS(insn->code) == BPF_LDX && |
1f9a1ea8 YS |
18050 | ((BPF_MODE(insn->code) != BPF_MEM && BPF_MODE(insn->code) != BPF_MEMSX) || |
18051 | insn->imm != 0)) { | |
61bd5218 | 18052 | verbose(env, "BPF_LDX uses reserved fields\n"); |
d691f9e8 AS |
18053 | return -EINVAL; |
18054 | } | |
18055 | ||
0246e64d | 18056 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 18057 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
18058 | struct bpf_map *map; |
18059 | struct fd f; | |
d8eca5bb | 18060 | u64 addr; |
387544bf | 18061 | u32 fd; |
0246e64d AS |
18062 | |
18063 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
18064 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
18065 | insn[1].off != 0) { | |
61bd5218 | 18066 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
18067 | return -EINVAL; |
18068 | } | |
18069 | ||
d8eca5bb | 18070 | if (insn[0].src_reg == 0) |
0246e64d AS |
18071 | /* valid generic load 64-bit imm */ |
18072 | goto next_insn; | |
18073 | ||
4976b718 HL |
18074 | if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) { |
18075 | aux = &env->insn_aux_data[i]; | |
18076 | err = check_pseudo_btf_id(env, insn, aux); | |
18077 | if (err) | |
18078 | return err; | |
18079 | goto next_insn; | |
18080 | } | |
18081 | ||
69c087ba YS |
18082 | if (insn[0].src_reg == BPF_PSEUDO_FUNC) { |
18083 | aux = &env->insn_aux_data[i]; | |
18084 | aux->ptr_type = PTR_TO_FUNC; | |
18085 | goto next_insn; | |
18086 | } | |
18087 | ||
d8eca5bb DB |
18088 | /* In final convert_pseudo_ld_imm64() step, this is |
18089 | * converted into regular 64-bit imm load insn. | |
18090 | */ | |
387544bf AS |
18091 | switch (insn[0].src_reg) { |
18092 | case BPF_PSEUDO_MAP_VALUE: | |
18093 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
18094 | break; | |
18095 | case BPF_PSEUDO_MAP_FD: | |
18096 | case BPF_PSEUDO_MAP_IDX: | |
18097 | if (insn[1].imm == 0) | |
18098 | break; | |
18099 | fallthrough; | |
18100 | default: | |
18101 | verbose(env, "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
18102 | return -EINVAL; |
18103 | } | |
18104 | ||
387544bf AS |
18105 | switch (insn[0].src_reg) { |
18106 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
18107 | case BPF_PSEUDO_MAP_IDX: | |
18108 | if (bpfptr_is_null(env->fd_array)) { | |
18109 | verbose(env, "fd_idx without fd_array is invalid\n"); | |
18110 | return -EPROTO; | |
18111 | } | |
18112 | if (copy_from_bpfptr_offset(&fd, env->fd_array, | |
18113 | insn[0].imm * sizeof(fd), | |
18114 | sizeof(fd))) | |
18115 | return -EFAULT; | |
18116 | break; | |
18117 | default: | |
18118 | fd = insn[0].imm; | |
18119 | break; | |
18120 | } | |
18121 | ||
18122 | f = fdget(fd); | |
c2101297 | 18123 | map = __bpf_map_get(f); |
0246e64d | 18124 | if (IS_ERR(map)) { |
61bd5218 | 18125 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 18126 | insn[0].imm); |
0246e64d AS |
18127 | return PTR_ERR(map); |
18128 | } | |
18129 | ||
61bd5218 | 18130 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
18131 | if (err) { |
18132 | fdput(f); | |
18133 | return err; | |
18134 | } | |
18135 | ||
d8eca5bb | 18136 | aux = &env->insn_aux_data[i]; |
387544bf AS |
18137 | if (insn[0].src_reg == BPF_PSEUDO_MAP_FD || |
18138 | insn[0].src_reg == BPF_PSEUDO_MAP_IDX) { | |
d8eca5bb DB |
18139 | addr = (unsigned long)map; |
18140 | } else { | |
18141 | u32 off = insn[1].imm; | |
18142 | ||
18143 | if (off >= BPF_MAX_VAR_OFF) { | |
18144 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
18145 | fdput(f); | |
18146 | return -EINVAL; | |
18147 | } | |
18148 | ||
18149 | if (!map->ops->map_direct_value_addr) { | |
18150 | verbose(env, "no direct value access support for this map type\n"); | |
18151 | fdput(f); | |
18152 | return -EINVAL; | |
18153 | } | |
18154 | ||
18155 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
18156 | if (err) { | |
18157 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
18158 | map->value_size, off); | |
18159 | fdput(f); | |
18160 | return err; | |
18161 | } | |
18162 | ||
18163 | aux->map_off = off; | |
18164 | addr += off; | |
18165 | } | |
18166 | ||
18167 | insn[0].imm = (u32)addr; | |
18168 | insn[1].imm = addr >> 32; | |
0246e64d AS |
18169 | |
18170 | /* check whether we recorded this map already */ | |
d8eca5bb | 18171 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 18172 | if (env->used_maps[j] == map) { |
d8eca5bb | 18173 | aux->map_index = j; |
0246e64d AS |
18174 | fdput(f); |
18175 | goto next_insn; | |
18176 | } | |
d8eca5bb | 18177 | } |
0246e64d AS |
18178 | |
18179 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
18180 | fdput(f); | |
18181 | return -E2BIG; | |
18182 | } | |
18183 | ||
0246e64d AS |
18184 | /* hold the map. If the program is rejected by verifier, |
18185 | * the map will be released by release_maps() or it | |
18186 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 18187 | * and all maps are released in free_used_maps() |
0246e64d | 18188 | */ |
1e0bd5a0 | 18189 | bpf_map_inc(map); |
d8eca5bb DB |
18190 | |
18191 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
18192 | env->used_maps[env->used_map_cnt++] = map; |
18193 | ||
b741f163 | 18194 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 18195 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 18196 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
18197 | fdput(f); |
18198 | return -EBUSY; | |
18199 | } | |
18200 | ||
0246e64d AS |
18201 | fdput(f); |
18202 | next_insn: | |
18203 | insn++; | |
18204 | i++; | |
5e581dad DB |
18205 | continue; |
18206 | } | |
18207 | ||
18208 | /* Basic sanity check before we invest more work here. */ | |
18209 | if (!bpf_opcode_in_insntable(insn->code)) { | |
18210 | verbose(env, "unknown opcode %02x\n", insn->code); | |
18211 | return -EINVAL; | |
0246e64d AS |
18212 | } |
18213 | } | |
18214 | ||
18215 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
18216 | * 'struct bpf_map *' into a register instead of user map_fd. | |
18217 | * These pointers will be used later by verifier to validate map access. | |
18218 | */ | |
18219 | return 0; | |
18220 | } | |
18221 | ||
18222 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 18223 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 18224 | { |
a2ea0746 DB |
18225 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
18226 | env->used_map_cnt); | |
0246e64d AS |
18227 | } |
18228 | ||
541c3bad AN |
18229 | /* drop refcnt of maps used by the rejected program */ |
18230 | static void release_btfs(struct bpf_verifier_env *env) | |
18231 | { | |
18232 | __bpf_free_used_btfs(env->prog->aux, env->used_btfs, | |
18233 | env->used_btf_cnt); | |
18234 | } | |
18235 | ||
0246e64d | 18236 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ |
58e2af8b | 18237 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
18238 | { |
18239 | struct bpf_insn *insn = env->prog->insnsi; | |
18240 | int insn_cnt = env->prog->len; | |
18241 | int i; | |
18242 | ||
69c087ba YS |
18243 | for (i = 0; i < insn_cnt; i++, insn++) { |
18244 | if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) | |
18245 | continue; | |
18246 | if (insn->src_reg == BPF_PSEUDO_FUNC) | |
18247 | continue; | |
18248 | insn->src_reg = 0; | |
18249 | } | |
0246e64d AS |
18250 | } |
18251 | ||
8041902d AS |
18252 | /* single env->prog->insni[off] instruction was replaced with the range |
18253 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
18254 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
18255 | */ | |
75f0fc7b HF |
18256 | static void adjust_insn_aux_data(struct bpf_verifier_env *env, |
18257 | struct bpf_insn_aux_data *new_data, | |
18258 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d | 18259 | { |
75f0fc7b | 18260 | struct bpf_insn_aux_data *old_data = env->insn_aux_data; |
b325fbca | 18261 | struct bpf_insn *insn = new_prog->insnsi; |
d203b0fd | 18262 | u32 old_seen = old_data[off].seen; |
b325fbca | 18263 | u32 prog_len; |
c131187d | 18264 | int i; |
8041902d | 18265 | |
b325fbca JW |
18266 | /* aux info at OFF always needs adjustment, no matter fast path |
18267 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
18268 | * original insn at old prog. | |
18269 | */ | |
18270 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
18271 | ||
8041902d | 18272 | if (cnt == 1) |
75f0fc7b | 18273 | return; |
b325fbca | 18274 | prog_len = new_prog->len; |
75f0fc7b | 18275 | |
8041902d AS |
18276 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); |
18277 | memcpy(new_data + off + cnt - 1, old_data + off, | |
18278 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 18279 | for (i = off; i < off + cnt - 1; i++) { |
d203b0fd DB |
18280 | /* Expand insni[off]'s seen count to the patched range. */ |
18281 | new_data[i].seen = old_seen; | |
b325fbca JW |
18282 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
18283 | } | |
8041902d AS |
18284 | env->insn_aux_data = new_data; |
18285 | vfree(old_data); | |
8041902d AS |
18286 | } |
18287 | ||
cc8b0b92 AS |
18288 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
18289 | { | |
18290 | int i; | |
18291 | ||
18292 | if (len == 1) | |
18293 | return; | |
4cb3d99c JW |
18294 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
18295 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 18296 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 18297 | continue; |
9c8105bd | 18298 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
18299 | } |
18300 | } | |
18301 | ||
7506d211 | 18302 | static void adjust_poke_descs(struct bpf_prog *prog, u32 off, u32 len) |
a748c697 MF |
18303 | { |
18304 | struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab; | |
18305 | int i, sz = prog->aux->size_poke_tab; | |
18306 | struct bpf_jit_poke_descriptor *desc; | |
18307 | ||
18308 | for (i = 0; i < sz; i++) { | |
18309 | desc = &tab[i]; | |
7506d211 JF |
18310 | if (desc->insn_idx <= off) |
18311 | continue; | |
a748c697 MF |
18312 | desc->insn_idx += len - 1; |
18313 | } | |
18314 | } | |
18315 | ||
8041902d AS |
18316 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
18317 | const struct bpf_insn *patch, u32 len) | |
18318 | { | |
18319 | struct bpf_prog *new_prog; | |
75f0fc7b HF |
18320 | struct bpf_insn_aux_data *new_data = NULL; |
18321 | ||
18322 | if (len > 1) { | |
18323 | new_data = vzalloc(array_size(env->prog->len + len - 1, | |
18324 | sizeof(struct bpf_insn_aux_data))); | |
18325 | if (!new_data) | |
18326 | return NULL; | |
18327 | } | |
8041902d AS |
18328 | |
18329 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
18330 | if (IS_ERR(new_prog)) { |
18331 | if (PTR_ERR(new_prog) == -ERANGE) | |
18332 | verbose(env, | |
18333 | "insn %d cannot be patched due to 16-bit range\n", | |
18334 | env->insn_aux_data[off].orig_idx); | |
75f0fc7b | 18335 | vfree(new_data); |
8041902d | 18336 | return NULL; |
4f73379e | 18337 | } |
75f0fc7b | 18338 | adjust_insn_aux_data(env, new_data, new_prog, off, len); |
cc8b0b92 | 18339 | adjust_subprog_starts(env, off, len); |
7506d211 | 18340 | adjust_poke_descs(new_prog, off, len); |
8041902d AS |
18341 | return new_prog; |
18342 | } | |
18343 | ||
52875a04 JK |
18344 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
18345 | u32 off, u32 cnt) | |
18346 | { | |
18347 | int i, j; | |
18348 | ||
18349 | /* find first prog starting at or after off (first to remove) */ | |
18350 | for (i = 0; i < env->subprog_cnt; i++) | |
18351 | if (env->subprog_info[i].start >= off) | |
18352 | break; | |
18353 | /* find first prog starting at or after off + cnt (first to stay) */ | |
18354 | for (j = i; j < env->subprog_cnt; j++) | |
18355 | if (env->subprog_info[j].start >= off + cnt) | |
18356 | break; | |
18357 | /* if j doesn't start exactly at off + cnt, we are just removing | |
18358 | * the front of previous prog | |
18359 | */ | |
18360 | if (env->subprog_info[j].start != off + cnt) | |
18361 | j--; | |
18362 | ||
18363 | if (j > i) { | |
18364 | struct bpf_prog_aux *aux = env->prog->aux; | |
18365 | int move; | |
18366 | ||
18367 | /* move fake 'exit' subprog as well */ | |
18368 | move = env->subprog_cnt + 1 - j; | |
18369 | ||
18370 | memmove(env->subprog_info + i, | |
18371 | env->subprog_info + j, | |
18372 | sizeof(*env->subprog_info) * move); | |
18373 | env->subprog_cnt -= j - i; | |
18374 | ||
18375 | /* remove func_info */ | |
18376 | if (aux->func_info) { | |
18377 | move = aux->func_info_cnt - j; | |
18378 | ||
18379 | memmove(aux->func_info + i, | |
18380 | aux->func_info + j, | |
18381 | sizeof(*aux->func_info) * move); | |
18382 | aux->func_info_cnt -= j - i; | |
18383 | /* func_info->insn_off is set after all code rewrites, | |
18384 | * in adjust_btf_func() - no need to adjust | |
18385 | */ | |
18386 | } | |
18387 | } else { | |
18388 | /* convert i from "first prog to remove" to "first to adjust" */ | |
18389 | if (env->subprog_info[i].start == off) | |
18390 | i++; | |
18391 | } | |
18392 | ||
18393 | /* update fake 'exit' subprog as well */ | |
18394 | for (; i <= env->subprog_cnt; i++) | |
18395 | env->subprog_info[i].start -= cnt; | |
18396 | ||
18397 | return 0; | |
18398 | } | |
18399 | ||
18400 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
18401 | u32 cnt) | |
18402 | { | |
18403 | struct bpf_prog *prog = env->prog; | |
18404 | u32 i, l_off, l_cnt, nr_linfo; | |
18405 | struct bpf_line_info *linfo; | |
18406 | ||
18407 | nr_linfo = prog->aux->nr_linfo; | |
18408 | if (!nr_linfo) | |
18409 | return 0; | |
18410 | ||
18411 | linfo = prog->aux->linfo; | |
18412 | ||
18413 | /* find first line info to remove, count lines to be removed */ | |
18414 | for (i = 0; i < nr_linfo; i++) | |
18415 | if (linfo[i].insn_off >= off) | |
18416 | break; | |
18417 | ||
18418 | l_off = i; | |
18419 | l_cnt = 0; | |
18420 | for (; i < nr_linfo; i++) | |
18421 | if (linfo[i].insn_off < off + cnt) | |
18422 | l_cnt++; | |
18423 | else | |
18424 | break; | |
18425 | ||
18426 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
18427 | * last removed linfo. prog is already modified, so prog->len == off | |
18428 | * means no live instructions after (tail of the program was removed). | |
18429 | */ | |
18430 | if (prog->len != off && l_cnt && | |
18431 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
18432 | l_cnt--; | |
18433 | linfo[--i].insn_off = off + cnt; | |
18434 | } | |
18435 | ||
18436 | /* remove the line info which refer to the removed instructions */ | |
18437 | if (l_cnt) { | |
18438 | memmove(linfo + l_off, linfo + i, | |
18439 | sizeof(*linfo) * (nr_linfo - i)); | |
18440 | ||
18441 | prog->aux->nr_linfo -= l_cnt; | |
18442 | nr_linfo = prog->aux->nr_linfo; | |
18443 | } | |
18444 | ||
18445 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
18446 | for (i = l_off; i < nr_linfo; i++) | |
18447 | linfo[i].insn_off -= cnt; | |
18448 | ||
18449 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
18450 | for (i = 0; i <= env->subprog_cnt; i++) | |
18451 | if (env->subprog_info[i].linfo_idx > l_off) { | |
18452 | /* program may have started in the removed region but | |
18453 | * may not be fully removed | |
18454 | */ | |
18455 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
18456 | env->subprog_info[i].linfo_idx -= l_cnt; | |
18457 | else | |
18458 | env->subprog_info[i].linfo_idx = l_off; | |
18459 | } | |
18460 | ||
18461 | return 0; | |
18462 | } | |
18463 | ||
18464 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
18465 | { | |
18466 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
18467 | unsigned int orig_prog_len = env->prog->len; | |
18468 | int err; | |
18469 | ||
9d03ebc7 | 18470 | if (bpf_prog_is_offloaded(env->prog->aux)) |
08ca90af JK |
18471 | bpf_prog_offload_remove_insns(env, off, cnt); |
18472 | ||
52875a04 JK |
18473 | err = bpf_remove_insns(env->prog, off, cnt); |
18474 | if (err) | |
18475 | return err; | |
18476 | ||
18477 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
18478 | if (err) | |
18479 | return err; | |
18480 | ||
18481 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
18482 | if (err) | |
18483 | return err; | |
18484 | ||
18485 | memmove(aux_data + off, aux_data + off + cnt, | |
18486 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
18487 | ||
18488 | return 0; | |
18489 | } | |
18490 | ||
2a5418a1 DB |
18491 | /* The verifier does more data flow analysis than llvm and will not |
18492 | * explore branches that are dead at run time. Malicious programs can | |
18493 | * have dead code too. Therefore replace all dead at-run-time code | |
18494 | * with 'ja -1'. | |
18495 | * | |
18496 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
18497 | * program and through another bug we would manage to jump there, then | |
18498 | * we'd execute beyond program memory otherwise. Returning exception | |
18499 | * code also wouldn't work since we can have subprogs where the dead | |
18500 | * code could be located. | |
c131187d AS |
18501 | */ |
18502 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
18503 | { | |
18504 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 18505 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
18506 | struct bpf_insn *insn = env->prog->insnsi; |
18507 | const int insn_cnt = env->prog->len; | |
18508 | int i; | |
18509 | ||
18510 | for (i = 0; i < insn_cnt; i++) { | |
18511 | if (aux_data[i].seen) | |
18512 | continue; | |
2a5418a1 | 18513 | memcpy(insn + i, &trap, sizeof(trap)); |
45c709f8 | 18514 | aux_data[i].zext_dst = false; |
c131187d AS |
18515 | } |
18516 | } | |
18517 | ||
e2ae4ca2 JK |
18518 | static bool insn_is_cond_jump(u8 code) |
18519 | { | |
18520 | u8 op; | |
18521 | ||
4cd58e9a | 18522 | op = BPF_OP(code); |
092ed096 | 18523 | if (BPF_CLASS(code) == BPF_JMP32) |
4cd58e9a | 18524 | return op != BPF_JA; |
092ed096 | 18525 | |
e2ae4ca2 JK |
18526 | if (BPF_CLASS(code) != BPF_JMP) |
18527 | return false; | |
18528 | ||
e2ae4ca2 JK |
18529 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; |
18530 | } | |
18531 | ||
18532 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
18533 | { | |
18534 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
18535 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
18536 | struct bpf_insn *insn = env->prog->insnsi; | |
18537 | const int insn_cnt = env->prog->len; | |
18538 | int i; | |
18539 | ||
18540 | for (i = 0; i < insn_cnt; i++, insn++) { | |
18541 | if (!insn_is_cond_jump(insn->code)) | |
18542 | continue; | |
18543 | ||
18544 | if (!aux_data[i + 1].seen) | |
18545 | ja.off = insn->off; | |
18546 | else if (!aux_data[i + 1 + insn->off].seen) | |
18547 | ja.off = 0; | |
18548 | else | |
18549 | continue; | |
18550 | ||
9d03ebc7 | 18551 | if (bpf_prog_is_offloaded(env->prog->aux)) |
08ca90af JK |
18552 | bpf_prog_offload_replace_insn(env, i, &ja); |
18553 | ||
e2ae4ca2 JK |
18554 | memcpy(insn, &ja, sizeof(ja)); |
18555 | } | |
18556 | } | |
18557 | ||
52875a04 JK |
18558 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
18559 | { | |
18560 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
18561 | int insn_cnt = env->prog->len; | |
18562 | int i, err; | |
18563 | ||
18564 | for (i = 0; i < insn_cnt; i++) { | |
18565 | int j; | |
18566 | ||
18567 | j = 0; | |
18568 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
18569 | j++; | |
18570 | if (!j) | |
18571 | continue; | |
18572 | ||
18573 | err = verifier_remove_insns(env, i, j); | |
18574 | if (err) | |
18575 | return err; | |
18576 | insn_cnt = env->prog->len; | |
18577 | } | |
18578 | ||
18579 | return 0; | |
18580 | } | |
18581 | ||
a1b14abc JK |
18582 | static int opt_remove_nops(struct bpf_verifier_env *env) |
18583 | { | |
18584 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
18585 | struct bpf_insn *insn = env->prog->insnsi; | |
18586 | int insn_cnt = env->prog->len; | |
18587 | int i, err; | |
18588 | ||
18589 | for (i = 0; i < insn_cnt; i++) { | |
18590 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
18591 | continue; | |
18592 | ||
18593 | err = verifier_remove_insns(env, i, 1); | |
18594 | if (err) | |
18595 | return err; | |
18596 | insn_cnt--; | |
18597 | i--; | |
18598 | } | |
18599 | ||
18600 | return 0; | |
18601 | } | |
18602 | ||
d6c2308c JW |
18603 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
18604 | const union bpf_attr *attr) | |
a4b1d3c1 | 18605 | { |
d6c2308c | 18606 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 18607 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 18608 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 18609 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 18610 | struct bpf_prog *new_prog; |
d6c2308c | 18611 | bool rnd_hi32; |
a4b1d3c1 | 18612 | |
d6c2308c | 18613 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 18614 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
18615 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
18616 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
18617 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
18618 | for (i = 0; i < len; i++) { |
18619 | int adj_idx = i + delta; | |
18620 | struct bpf_insn insn; | |
83a28819 | 18621 | int load_reg; |
a4b1d3c1 | 18622 | |
d6c2308c | 18623 | insn = insns[adj_idx]; |
83a28819 | 18624 | load_reg = insn_def_regno(&insn); |
d6c2308c JW |
18625 | if (!aux[adj_idx].zext_dst) { |
18626 | u8 code, class; | |
18627 | u32 imm_rnd; | |
18628 | ||
18629 | if (!rnd_hi32) | |
18630 | continue; | |
18631 | ||
18632 | code = insn.code; | |
18633 | class = BPF_CLASS(code); | |
83a28819 | 18634 | if (load_reg == -1) |
d6c2308c JW |
18635 | continue; |
18636 | ||
18637 | /* NOTE: arg "reg" (the fourth one) is only used for | |
83a28819 IL |
18638 | * BPF_STX + SRC_OP, so it is safe to pass NULL |
18639 | * here. | |
d6c2308c | 18640 | */ |
83a28819 | 18641 | if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) { |
d6c2308c JW |
18642 | if (class == BPF_LD && |
18643 | BPF_MODE(code) == BPF_IMM) | |
18644 | i++; | |
18645 | continue; | |
18646 | } | |
18647 | ||
18648 | /* ctx load could be transformed into wider load. */ | |
18649 | if (class == BPF_LDX && | |
18650 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
18651 | continue; | |
18652 | ||
a251c17a | 18653 | imm_rnd = get_random_u32(); |
d6c2308c JW |
18654 | rnd_hi32_patch[0] = insn; |
18655 | rnd_hi32_patch[1].imm = imm_rnd; | |
83a28819 | 18656 | rnd_hi32_patch[3].dst_reg = load_reg; |
d6c2308c JW |
18657 | patch = rnd_hi32_patch; |
18658 | patch_len = 4; | |
18659 | goto apply_patch_buffer; | |
18660 | } | |
18661 | ||
39491867 BJ |
18662 | /* Add in an zero-extend instruction if a) the JIT has requested |
18663 | * it or b) it's a CMPXCHG. | |
18664 | * | |
18665 | * The latter is because: BPF_CMPXCHG always loads a value into | |
18666 | * R0, therefore always zero-extends. However some archs' | |
18667 | * equivalent instruction only does this load when the | |
18668 | * comparison is successful. This detail of CMPXCHG is | |
18669 | * orthogonal to the general zero-extension behaviour of the | |
18670 | * CPU, so it's treated independently of bpf_jit_needs_zext. | |
18671 | */ | |
18672 | if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn)) | |
a4b1d3c1 JW |
18673 | continue; |
18674 | ||
d35af0a7 BT |
18675 | /* Zero-extension is done by the caller. */ |
18676 | if (bpf_pseudo_kfunc_call(&insn)) | |
18677 | continue; | |
18678 | ||
83a28819 IL |
18679 | if (WARN_ON(load_reg == -1)) { |
18680 | verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n"); | |
18681 | return -EFAULT; | |
b2e37a71 IL |
18682 | } |
18683 | ||
a4b1d3c1 | 18684 | zext_patch[0] = insn; |
b2e37a71 IL |
18685 | zext_patch[1].dst_reg = load_reg; |
18686 | zext_patch[1].src_reg = load_reg; | |
d6c2308c JW |
18687 | patch = zext_patch; |
18688 | patch_len = 2; | |
18689 | apply_patch_buffer: | |
18690 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
18691 | if (!new_prog) |
18692 | return -ENOMEM; | |
18693 | env->prog = new_prog; | |
18694 | insns = new_prog->insnsi; | |
18695 | aux = env->insn_aux_data; | |
d6c2308c | 18696 | delta += patch_len - 1; |
a4b1d3c1 JW |
18697 | } |
18698 | ||
18699 | return 0; | |
18700 | } | |
18701 | ||
c64b7983 JS |
18702 | /* convert load instructions that access fields of a context type into a |
18703 | * sequence of instructions that access fields of the underlying structure: | |
18704 | * struct __sk_buff -> struct sk_buff | |
18705 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 18706 | */ |
58e2af8b | 18707 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 18708 | { |
00176a34 | 18709 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 18710 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 18711 | const int insn_cnt = env->prog->len; |
36bbef52 | 18712 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 18713 | u32 target_size, size_default, off; |
9bac3d6d | 18714 | struct bpf_prog *new_prog; |
d691f9e8 | 18715 | enum bpf_access_type type; |
f96da094 | 18716 | bool is_narrower_load; |
9bac3d6d | 18717 | |
b09928b9 DB |
18718 | if (ops->gen_prologue || env->seen_direct_write) { |
18719 | if (!ops->gen_prologue) { | |
18720 | verbose(env, "bpf verifier is misconfigured\n"); | |
18721 | return -EINVAL; | |
18722 | } | |
36bbef52 DB |
18723 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
18724 | env->prog); | |
18725 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 18726 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
18727 | return -EINVAL; |
18728 | } else if (cnt) { | |
8041902d | 18729 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
18730 | if (!new_prog) |
18731 | return -ENOMEM; | |
8041902d | 18732 | |
36bbef52 | 18733 | env->prog = new_prog; |
3df126f3 | 18734 | delta += cnt - 1; |
36bbef52 DB |
18735 | } |
18736 | } | |
18737 | ||
9d03ebc7 | 18738 | if (bpf_prog_is_offloaded(env->prog->aux)) |
9bac3d6d AS |
18739 | return 0; |
18740 | ||
3df126f3 | 18741 | insn = env->prog->insnsi + delta; |
36bbef52 | 18742 | |
9bac3d6d | 18743 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 | 18744 | bpf_convert_ctx_access_t convert_ctx_access; |
1f1e864b | 18745 | u8 mode; |
c64b7983 | 18746 | |
62c7989b DB |
18747 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
18748 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
18749 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
1f9a1ea8 YS |
18750 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW) || |
18751 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_B) || | |
18752 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_H) || | |
18753 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_W)) { | |
d691f9e8 | 18754 | type = BPF_READ; |
2039f26f DB |
18755 | } else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
18756 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
18757 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
18758 | insn->code == (BPF_STX | BPF_MEM | BPF_DW) || | |
18759 | insn->code == (BPF_ST | BPF_MEM | BPF_B) || | |
18760 | insn->code == (BPF_ST | BPF_MEM | BPF_H) || | |
18761 | insn->code == (BPF_ST | BPF_MEM | BPF_W) || | |
18762 | insn->code == (BPF_ST | BPF_MEM | BPF_DW)) { | |
d691f9e8 | 18763 | type = BPF_WRITE; |
2039f26f | 18764 | } else { |
9bac3d6d | 18765 | continue; |
2039f26f | 18766 | } |
9bac3d6d | 18767 | |
af86ca4e | 18768 | if (type == BPF_WRITE && |
2039f26f | 18769 | env->insn_aux_data[i + delta].sanitize_stack_spill) { |
af86ca4e | 18770 | struct bpf_insn patch[] = { |
af86ca4e | 18771 | *insn, |
2039f26f | 18772 | BPF_ST_NOSPEC(), |
af86ca4e AS |
18773 | }; |
18774 | ||
18775 | cnt = ARRAY_SIZE(patch); | |
18776 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
18777 | if (!new_prog) | |
18778 | return -ENOMEM; | |
18779 | ||
18780 | delta += cnt - 1; | |
18781 | env->prog = new_prog; | |
18782 | insn = new_prog->insnsi + i + delta; | |
18783 | continue; | |
18784 | } | |
18785 | ||
6efe152d | 18786 | switch ((int)env->insn_aux_data[i + delta].ptr_type) { |
c64b7983 JS |
18787 | case PTR_TO_CTX: |
18788 | if (!ops->convert_ctx_access) | |
18789 | continue; | |
18790 | convert_ctx_access = ops->convert_ctx_access; | |
18791 | break; | |
18792 | case PTR_TO_SOCKET: | |
46f8bc92 | 18793 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
18794 | convert_ctx_access = bpf_sock_convert_ctx_access; |
18795 | break; | |
655a51e5 MKL |
18796 | case PTR_TO_TCP_SOCK: |
18797 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
18798 | break; | |
fada7fdc JL |
18799 | case PTR_TO_XDP_SOCK: |
18800 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
18801 | break; | |
2a02759e | 18802 | case PTR_TO_BTF_ID: |
6efe152d | 18803 | case PTR_TO_BTF_ID | PTR_UNTRUSTED: |
282de143 KKD |
18804 | /* PTR_TO_BTF_ID | MEM_ALLOC always has a valid lifetime, unlike |
18805 | * PTR_TO_BTF_ID, and an active ref_obj_id, but the same cannot | |
18806 | * be said once it is marked PTR_UNTRUSTED, hence we must handle | |
18807 | * any faults for loads into such types. BPF_WRITE is disallowed | |
18808 | * for this case. | |
18809 | */ | |
18810 | case PTR_TO_BTF_ID | MEM_ALLOC | PTR_UNTRUSTED: | |
27ae7997 | 18811 | if (type == BPF_READ) { |
1f9a1ea8 YS |
18812 | if (BPF_MODE(insn->code) == BPF_MEM) |
18813 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
18814 | BPF_SIZE((insn)->code); | |
18815 | else | |
18816 | insn->code = BPF_LDX | BPF_PROBE_MEMSX | | |
18817 | BPF_SIZE((insn)->code); | |
27ae7997 | 18818 | env->prog->aux->num_exentries++; |
2a02759e | 18819 | } |
2a02759e | 18820 | continue; |
c64b7983 | 18821 | default: |
9bac3d6d | 18822 | continue; |
c64b7983 | 18823 | } |
9bac3d6d | 18824 | |
31fd8581 | 18825 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 18826 | size = BPF_LDST_BYTES(insn); |
1f1e864b | 18827 | mode = BPF_MODE(insn->code); |
31fd8581 YS |
18828 | |
18829 | /* If the read access is a narrower load of the field, | |
18830 | * convert to a 4/8-byte load, to minimum program type specific | |
18831 | * convert_ctx_access changes. If conversion is successful, | |
18832 | * we will apply proper mask to the result. | |
18833 | */ | |
f96da094 | 18834 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
18835 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
18836 | off = insn->off; | |
31fd8581 | 18837 | if (is_narrower_load) { |
f96da094 DB |
18838 | u8 size_code; |
18839 | ||
18840 | if (type == BPF_WRITE) { | |
61bd5218 | 18841 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
18842 | return -EINVAL; |
18843 | } | |
31fd8581 | 18844 | |
f96da094 | 18845 | size_code = BPF_H; |
31fd8581 YS |
18846 | if (ctx_field_size == 4) |
18847 | size_code = BPF_W; | |
18848 | else if (ctx_field_size == 8) | |
18849 | size_code = BPF_DW; | |
f96da094 | 18850 | |
bc23105c | 18851 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
18852 | insn->code = BPF_LDX | BPF_MEM | size_code; |
18853 | } | |
f96da094 DB |
18854 | |
18855 | target_size = 0; | |
c64b7983 JS |
18856 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
18857 | &target_size); | |
f96da094 DB |
18858 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
18859 | (ctx_field_size && !target_size)) { | |
61bd5218 | 18860 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
18861 | return -EINVAL; |
18862 | } | |
f96da094 DB |
18863 | |
18864 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
18865 | u8 shift = bpf_ctx_narrow_access_offset( |
18866 | off, size, size_default) * 8; | |
d7af7e49 AI |
18867 | if (shift && cnt + 1 >= ARRAY_SIZE(insn_buf)) { |
18868 | verbose(env, "bpf verifier narrow ctx load misconfigured\n"); | |
18869 | return -EINVAL; | |
18870 | } | |
46f53a65 AI |
18871 | if (ctx_field_size <= 4) { |
18872 | if (shift) | |
18873 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
18874 | insn->dst_reg, | |
18875 | shift); | |
31fd8581 | 18876 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 18877 | (1 << size * 8) - 1); |
46f53a65 AI |
18878 | } else { |
18879 | if (shift) | |
18880 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
18881 | insn->dst_reg, | |
18882 | shift); | |
0613d8ca | 18883 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 18884 | (1ULL << size * 8) - 1); |
46f53a65 | 18885 | } |
31fd8581 | 18886 | } |
1f1e864b YS |
18887 | if (mode == BPF_MEMSX) |
18888 | insn_buf[cnt++] = BPF_RAW_INSN(BPF_ALU64 | BPF_MOV | BPF_X, | |
18889 | insn->dst_reg, insn->dst_reg, | |
18890 | size * 8, 0); | |
9bac3d6d | 18891 | |
8041902d | 18892 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
18893 | if (!new_prog) |
18894 | return -ENOMEM; | |
18895 | ||
3df126f3 | 18896 | delta += cnt - 1; |
9bac3d6d AS |
18897 | |
18898 | /* keep walking new program and skip insns we just inserted */ | |
18899 | env->prog = new_prog; | |
3df126f3 | 18900 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
18901 | } |
18902 | ||
18903 | return 0; | |
18904 | } | |
18905 | ||
1c2a088a AS |
18906 | static int jit_subprogs(struct bpf_verifier_env *env) |
18907 | { | |
18908 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
18909 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
a748c697 | 18910 | struct bpf_map *map_ptr; |
7105e828 | 18911 | struct bpf_insn *insn; |
1c2a088a | 18912 | void *old_bpf_func; |
c4c0bdc0 | 18913 | int err, num_exentries; |
1c2a088a | 18914 | |
f910cefa | 18915 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
18916 | return 0; |
18917 | ||
7105e828 | 18918 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
3990ed4c | 18919 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn)) |
69c087ba | 18920 | continue; |
69c087ba | 18921 | |
c7a89784 DB |
18922 | /* Upon error here we cannot fall back to interpreter but |
18923 | * need a hard reject of the program. Thus -EFAULT is | |
18924 | * propagated in any case. | |
18925 | */ | |
1c2a088a AS |
18926 | subprog = find_subprog(env, i + insn->imm + 1); |
18927 | if (subprog < 0) { | |
18928 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
18929 | i + insn->imm + 1); | |
18930 | return -EFAULT; | |
18931 | } | |
18932 | /* temporarily remember subprog id inside insn instead of | |
18933 | * aux_data, since next loop will split up all insns into funcs | |
18934 | */ | |
f910cefa | 18935 | insn->off = subprog; |
1c2a088a AS |
18936 | /* remember original imm in case JIT fails and fallback |
18937 | * to interpreter will be needed | |
18938 | */ | |
18939 | env->insn_aux_data[i].call_imm = insn->imm; | |
18940 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
18941 | insn->imm = 1; | |
3990ed4c MKL |
18942 | if (bpf_pseudo_func(insn)) |
18943 | /* jit (e.g. x86_64) may emit fewer instructions | |
18944 | * if it learns a u32 imm is the same as a u64 imm. | |
18945 | * Force a non zero here. | |
18946 | */ | |
18947 | insn[1].imm = 1; | |
1c2a088a AS |
18948 | } |
18949 | ||
c454a46b MKL |
18950 | err = bpf_prog_alloc_jited_linfo(prog); |
18951 | if (err) | |
18952 | goto out_undo_insn; | |
18953 | ||
18954 | err = -ENOMEM; | |
6396bb22 | 18955 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 18956 | if (!func) |
c7a89784 | 18957 | goto out_undo_insn; |
1c2a088a | 18958 | |
f910cefa | 18959 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 18960 | subprog_start = subprog_end; |
4cb3d99c | 18961 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
18962 | |
18963 | len = subprog_end - subprog_start; | |
fb7dd8bc | 18964 | /* bpf_prog_run() doesn't call subprogs directly, |
492ecee8 AS |
18965 | * hence main prog stats include the runtime of subprogs. |
18966 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
700d4796 | 18967 | * func[i]->stats will never be accessed and stays NULL |
492ecee8 AS |
18968 | */ |
18969 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
18970 | if (!func[i]) |
18971 | goto out_free; | |
18972 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
18973 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 18974 | func[i]->type = prog->type; |
1c2a088a | 18975 | func[i]->len = len; |
4f74d809 DB |
18976 | if (bpf_prog_calc_tag(func[i])) |
18977 | goto out_free; | |
1c2a088a | 18978 | func[i]->is_func = 1; |
ba64e7d8 | 18979 | func[i]->aux->func_idx = i; |
f263a814 | 18980 | /* Below members will be freed only at prog->aux */ |
ba64e7d8 YS |
18981 | func[i]->aux->btf = prog->aux->btf; |
18982 | func[i]->aux->func_info = prog->aux->func_info; | |
9c7c48d6 | 18983 | func[i]->aux->func_info_cnt = prog->aux->func_info_cnt; |
f263a814 JF |
18984 | func[i]->aux->poke_tab = prog->aux->poke_tab; |
18985 | func[i]->aux->size_poke_tab = prog->aux->size_poke_tab; | |
ba64e7d8 | 18986 | |
a748c697 | 18987 | for (j = 0; j < prog->aux->size_poke_tab; j++) { |
f263a814 | 18988 | struct bpf_jit_poke_descriptor *poke; |
a748c697 | 18989 | |
f263a814 JF |
18990 | poke = &prog->aux->poke_tab[j]; |
18991 | if (poke->insn_idx < subprog_end && | |
18992 | poke->insn_idx >= subprog_start) | |
18993 | poke->aux = func[i]->aux; | |
a748c697 MF |
18994 | } |
18995 | ||
1c2a088a | 18996 | func[i]->aux->name[0] = 'F'; |
9c8105bd | 18997 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 18998 | func[i]->jit_requested = 1; |
d2a3b7c5 | 18999 | func[i]->blinding_requested = prog->blinding_requested; |
e6ac2450 | 19000 | func[i]->aux->kfunc_tab = prog->aux->kfunc_tab; |
2357672c | 19001 | func[i]->aux->kfunc_btf_tab = prog->aux->kfunc_btf_tab; |
c454a46b MKL |
19002 | func[i]->aux->linfo = prog->aux->linfo; |
19003 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
19004 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
19005 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
c4c0bdc0 YS |
19006 | num_exentries = 0; |
19007 | insn = func[i]->insnsi; | |
19008 | for (j = 0; j < func[i]->len; j++, insn++) { | |
19009 | if (BPF_CLASS(insn->code) == BPF_LDX && | |
1f9a1ea8 YS |
19010 | (BPF_MODE(insn->code) == BPF_PROBE_MEM || |
19011 | BPF_MODE(insn->code) == BPF_PROBE_MEMSX)) | |
c4c0bdc0 YS |
19012 | num_exentries++; |
19013 | } | |
19014 | func[i]->aux->num_exentries = num_exentries; | |
ebf7d1f5 | 19015 | func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable; |
f18b03fa KKD |
19016 | func[i]->aux->exception_cb = env->subprog_info[i].is_exception_cb; |
19017 | if (!i) | |
19018 | func[i]->aux->exception_boundary = env->seen_exception; | |
1c2a088a AS |
19019 | func[i] = bpf_int_jit_compile(func[i]); |
19020 | if (!func[i]->jited) { | |
19021 | err = -ENOTSUPP; | |
19022 | goto out_free; | |
19023 | } | |
19024 | cond_resched(); | |
19025 | } | |
a748c697 | 19026 | |
1c2a088a AS |
19027 | /* at this point all bpf functions were successfully JITed |
19028 | * now populate all bpf_calls with correct addresses and | |
19029 | * run last pass of JIT | |
19030 | */ | |
f910cefa | 19031 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
19032 | insn = func[i]->insnsi; |
19033 | for (j = 0; j < func[i]->len; j++, insn++) { | |
69c087ba | 19034 | if (bpf_pseudo_func(insn)) { |
3990ed4c | 19035 | subprog = insn->off; |
69c087ba YS |
19036 | insn[0].imm = (u32)(long)func[subprog]->bpf_func; |
19037 | insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32; | |
19038 | continue; | |
19039 | } | |
23a2d70c | 19040 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
19041 | continue; |
19042 | subprog = insn->off; | |
3d717fad | 19043 | insn->imm = BPF_CALL_IMM(func[subprog]->bpf_func); |
1c2a088a | 19044 | } |
2162fed4 SD |
19045 | |
19046 | /* we use the aux data to keep a list of the start addresses | |
19047 | * of the JITed images for each function in the program | |
19048 | * | |
19049 | * for some architectures, such as powerpc64, the imm field | |
19050 | * might not be large enough to hold the offset of the start | |
19051 | * address of the callee's JITed image from __bpf_call_base | |
19052 | * | |
19053 | * in such cases, we can lookup the start address of a callee | |
19054 | * by using its subprog id, available from the off field of | |
19055 | * the call instruction, as an index for this list | |
19056 | */ | |
19057 | func[i]->aux->func = func; | |
335d1c5b KKD |
19058 | func[i]->aux->func_cnt = env->subprog_cnt - env->hidden_subprog_cnt; |
19059 | func[i]->aux->real_func_cnt = env->subprog_cnt; | |
1c2a088a | 19060 | } |
f910cefa | 19061 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
19062 | old_bpf_func = func[i]->bpf_func; |
19063 | tmp = bpf_int_jit_compile(func[i]); | |
19064 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
19065 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 19066 | err = -ENOTSUPP; |
1c2a088a AS |
19067 | goto out_free; |
19068 | } | |
19069 | cond_resched(); | |
19070 | } | |
19071 | ||
19072 | /* finally lock prog and jit images for all functions and | |
0108a4e9 KJ |
19073 | * populate kallsysm. Begin at the first subprogram, since |
19074 | * bpf_prog_load will add the kallsyms for the main program. | |
1c2a088a | 19075 | */ |
0108a4e9 | 19076 | for (i = 1; i < env->subprog_cnt; i++) { |
1c2a088a AS |
19077 | bpf_prog_lock_ro(func[i]); |
19078 | bpf_prog_kallsyms_add(func[i]); | |
19079 | } | |
7105e828 DB |
19080 | |
19081 | /* Last step: make now unused interpreter insns from main | |
19082 | * prog consistent for later dump requests, so they can | |
19083 | * later look the same as if they were interpreted only. | |
19084 | */ | |
19085 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
69c087ba YS |
19086 | if (bpf_pseudo_func(insn)) { |
19087 | insn[0].imm = env->insn_aux_data[i].call_imm; | |
3990ed4c MKL |
19088 | insn[1].imm = insn->off; |
19089 | insn->off = 0; | |
69c087ba YS |
19090 | continue; |
19091 | } | |
23a2d70c | 19092 | if (!bpf_pseudo_call(insn)) |
7105e828 DB |
19093 | continue; |
19094 | insn->off = env->insn_aux_data[i].call_imm; | |
19095 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 19096 | insn->imm = subprog; |
7105e828 DB |
19097 | } |
19098 | ||
1c2a088a AS |
19099 | prog->jited = 1; |
19100 | prog->bpf_func = func[0]->bpf_func; | |
d00c6473 | 19101 | prog->jited_len = func[0]->jited_len; |
0108a4e9 KJ |
19102 | prog->aux->extable = func[0]->aux->extable; |
19103 | prog->aux->num_exentries = func[0]->aux->num_exentries; | |
1c2a088a | 19104 | prog->aux->func = func; |
335d1c5b KKD |
19105 | prog->aux->func_cnt = env->subprog_cnt - env->hidden_subprog_cnt; |
19106 | prog->aux->real_func_cnt = env->subprog_cnt; | |
f18b03fa KKD |
19107 | prog->aux->bpf_exception_cb = (void *)func[env->exception_callback_subprog]->bpf_func; |
19108 | prog->aux->exception_boundary = func[0]->aux->exception_boundary; | |
e16301fb | 19109 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
19110 | return 0; |
19111 | out_free: | |
f263a814 JF |
19112 | /* We failed JIT'ing, so at this point we need to unregister poke |
19113 | * descriptors from subprogs, so that kernel is not attempting to | |
19114 | * patch it anymore as we're freeing the subprog JIT memory. | |
19115 | */ | |
19116 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
19117 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
19118 | map_ptr->ops->map_poke_untrack(map_ptr, prog->aux); | |
19119 | } | |
19120 | /* At this point we're guaranteed that poke descriptors are not | |
19121 | * live anymore. We can just unlink its descriptor table as it's | |
19122 | * released with the main prog. | |
19123 | */ | |
a748c697 MF |
19124 | for (i = 0; i < env->subprog_cnt; i++) { |
19125 | if (!func[i]) | |
19126 | continue; | |
f263a814 | 19127 | func[i]->aux->poke_tab = NULL; |
a748c697 MF |
19128 | bpf_jit_free(func[i]); |
19129 | } | |
1c2a088a | 19130 | kfree(func); |
c7a89784 | 19131 | out_undo_insn: |
1c2a088a AS |
19132 | /* cleanup main prog to be interpreted */ |
19133 | prog->jit_requested = 0; | |
d2a3b7c5 | 19134 | prog->blinding_requested = 0; |
1c2a088a | 19135 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
23a2d70c | 19136 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
19137 | continue; |
19138 | insn->off = 0; | |
19139 | insn->imm = env->insn_aux_data[i].call_imm; | |
19140 | } | |
e16301fb | 19141 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
19142 | return err; |
19143 | } | |
19144 | ||
1ea47e01 AS |
19145 | static int fixup_call_args(struct bpf_verifier_env *env) |
19146 | { | |
19d28fbd | 19147 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
19148 | struct bpf_prog *prog = env->prog; |
19149 | struct bpf_insn *insn = prog->insnsi; | |
e6ac2450 | 19150 | bool has_kfunc_call = bpf_prog_has_kfunc_call(prog); |
1ea47e01 | 19151 | int i, depth; |
19d28fbd | 19152 | #endif |
e4052d06 | 19153 | int err = 0; |
1ea47e01 | 19154 | |
e4052d06 | 19155 | if (env->prog->jit_requested && |
9d03ebc7 | 19156 | !bpf_prog_is_offloaded(env->prog->aux)) { |
19d28fbd DM |
19157 | err = jit_subprogs(env); |
19158 | if (err == 0) | |
1c2a088a | 19159 | return 0; |
c7a89784 DB |
19160 | if (err == -EFAULT) |
19161 | return err; | |
19d28fbd DM |
19162 | } |
19163 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
e6ac2450 MKL |
19164 | if (has_kfunc_call) { |
19165 | verbose(env, "calling kernel functions are not allowed in non-JITed programs\n"); | |
19166 | return -EINVAL; | |
19167 | } | |
e411901c MF |
19168 | if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) { |
19169 | /* When JIT fails the progs with bpf2bpf calls and tail_calls | |
19170 | * have to be rejected, since interpreter doesn't support them yet. | |
19171 | */ | |
19172 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
19173 | return -EINVAL; | |
19174 | } | |
1ea47e01 | 19175 | for (i = 0; i < prog->len; i++, insn++) { |
69c087ba YS |
19176 | if (bpf_pseudo_func(insn)) { |
19177 | /* When JIT fails the progs with callback calls | |
19178 | * have to be rejected, since interpreter doesn't support them yet. | |
19179 | */ | |
19180 | verbose(env, "callbacks are not allowed in non-JITed programs\n"); | |
19181 | return -EINVAL; | |
19182 | } | |
19183 | ||
23a2d70c | 19184 | if (!bpf_pseudo_call(insn)) |
1ea47e01 AS |
19185 | continue; |
19186 | depth = get_callee_stack_depth(env, insn, i); | |
19187 | if (depth < 0) | |
19188 | return depth; | |
19189 | bpf_patch_call_args(insn, depth); | |
19190 | } | |
19d28fbd DM |
19191 | err = 0; |
19192 | #endif | |
19193 | return err; | |
1ea47e01 AS |
19194 | } |
19195 | ||
1cf3bfc6 IL |
19196 | /* replace a generic kfunc with a specialized version if necessary */ |
19197 | static void specialize_kfunc(struct bpf_verifier_env *env, | |
19198 | u32 func_id, u16 offset, unsigned long *addr) | |
19199 | { | |
19200 | struct bpf_prog *prog = env->prog; | |
19201 | bool seen_direct_write; | |
19202 | void *xdp_kfunc; | |
19203 | bool is_rdonly; | |
19204 | ||
19205 | if (bpf_dev_bound_kfunc_id(func_id)) { | |
19206 | xdp_kfunc = bpf_dev_bound_resolve_kfunc(prog, func_id); | |
19207 | if (xdp_kfunc) { | |
19208 | *addr = (unsigned long)xdp_kfunc; | |
19209 | return; | |
19210 | } | |
19211 | /* fallback to default kfunc when not supported by netdev */ | |
19212 | } | |
19213 | ||
19214 | if (offset) | |
19215 | return; | |
19216 | ||
19217 | if (func_id == special_kfunc_list[KF_bpf_dynptr_from_skb]) { | |
19218 | seen_direct_write = env->seen_direct_write; | |
19219 | is_rdonly = !may_access_direct_pkt_data(env, NULL, BPF_WRITE); | |
19220 | ||
19221 | if (is_rdonly) | |
19222 | *addr = (unsigned long)bpf_dynptr_from_skb_rdonly; | |
19223 | ||
19224 | /* restore env->seen_direct_write to its original value, since | |
19225 | * may_access_direct_pkt_data mutates it | |
19226 | */ | |
19227 | env->seen_direct_write = seen_direct_write; | |
19228 | } | |
19229 | } | |
19230 | ||
d2dcc67d DM |
19231 | static void __fixup_collection_insert_kfunc(struct bpf_insn_aux_data *insn_aux, |
19232 | u16 struct_meta_reg, | |
19233 | u16 node_offset_reg, | |
19234 | struct bpf_insn *insn, | |
19235 | struct bpf_insn *insn_buf, | |
19236 | int *cnt) | |
19237 | { | |
19238 | struct btf_struct_meta *kptr_struct_meta = insn_aux->kptr_struct_meta; | |
19239 | struct bpf_insn addr[2] = { BPF_LD_IMM64(struct_meta_reg, (long)kptr_struct_meta) }; | |
19240 | ||
19241 | insn_buf[0] = addr[0]; | |
19242 | insn_buf[1] = addr[1]; | |
19243 | insn_buf[2] = BPF_MOV64_IMM(node_offset_reg, insn_aux->insert_off); | |
19244 | insn_buf[3] = *insn; | |
19245 | *cnt = 4; | |
19246 | } | |
19247 | ||
958cf2e2 KKD |
19248 | static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
19249 | struct bpf_insn *insn_buf, int insn_idx, int *cnt) | |
e6ac2450 MKL |
19250 | { |
19251 | const struct bpf_kfunc_desc *desc; | |
19252 | ||
a5d82727 KKD |
19253 | if (!insn->imm) { |
19254 | verbose(env, "invalid kernel function call not eliminated in verifier pass\n"); | |
19255 | return -EINVAL; | |
19256 | } | |
19257 | ||
3d76a4d3 SF |
19258 | *cnt = 0; |
19259 | ||
1cf3bfc6 IL |
19260 | /* insn->imm has the btf func_id. Replace it with an offset relative to |
19261 | * __bpf_call_base, unless the JIT needs to call functions that are | |
19262 | * further than 32 bits away (bpf_jit_supports_far_kfunc_call()). | |
e6ac2450 | 19263 | */ |
2357672c | 19264 | desc = find_kfunc_desc(env->prog, insn->imm, insn->off); |
e6ac2450 MKL |
19265 | if (!desc) { |
19266 | verbose(env, "verifier internal error: kernel function descriptor not found for func_id %u\n", | |
19267 | insn->imm); | |
19268 | return -EFAULT; | |
19269 | } | |
19270 | ||
1cf3bfc6 IL |
19271 | if (!bpf_jit_supports_far_kfunc_call()) |
19272 | insn->imm = BPF_CALL_IMM(desc->addr); | |
958cf2e2 KKD |
19273 | if (insn->off) |
19274 | return 0; | |
36d8bdf7 YS |
19275 | if (desc->func_id == special_kfunc_list[KF_bpf_obj_new_impl] || |
19276 | desc->func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl]) { | |
958cf2e2 KKD |
19277 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; |
19278 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
19279 | u64 obj_new_size = env->insn_aux_data[insn_idx].obj_new_size; | |
e6ac2450 | 19280 | |
36d8bdf7 YS |
19281 | if (desc->func_id == special_kfunc_list[KF_bpf_percpu_obj_new_impl] && kptr_struct_meta) { |
19282 | verbose(env, "verifier internal error: NULL kptr_struct_meta expected at insn_idx %d\n", | |
19283 | insn_idx); | |
19284 | return -EFAULT; | |
19285 | } | |
19286 | ||
958cf2e2 KKD |
19287 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_1, obj_new_size); |
19288 | insn_buf[1] = addr[0]; | |
19289 | insn_buf[2] = addr[1]; | |
19290 | insn_buf[3] = *insn; | |
19291 | *cnt = 4; | |
7c50b1cb | 19292 | } else if (desc->func_id == special_kfunc_list[KF_bpf_obj_drop_impl] || |
36d8bdf7 | 19293 | desc->func_id == special_kfunc_list[KF_bpf_percpu_obj_drop_impl] || |
7c50b1cb | 19294 | desc->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]) { |
ac9f0605 KKD |
19295 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; |
19296 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
19297 | ||
36d8bdf7 YS |
19298 | if (desc->func_id == special_kfunc_list[KF_bpf_percpu_obj_drop_impl] && kptr_struct_meta) { |
19299 | verbose(env, "verifier internal error: NULL kptr_struct_meta expected at insn_idx %d\n", | |
19300 | insn_idx); | |
19301 | return -EFAULT; | |
19302 | } | |
19303 | ||
f0d991a0 DM |
19304 | if (desc->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl] && |
19305 | !kptr_struct_meta) { | |
19306 | verbose(env, "verifier internal error: kptr_struct_meta expected at insn_idx %d\n", | |
19307 | insn_idx); | |
19308 | return -EFAULT; | |
19309 | } | |
19310 | ||
ac9f0605 KKD |
19311 | insn_buf[0] = addr[0]; |
19312 | insn_buf[1] = addr[1]; | |
19313 | insn_buf[2] = *insn; | |
19314 | *cnt = 3; | |
d2dcc67d DM |
19315 | } else if (desc->func_id == special_kfunc_list[KF_bpf_list_push_back_impl] || |
19316 | desc->func_id == special_kfunc_list[KF_bpf_list_push_front_impl] || | |
19317 | desc->func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
f0d991a0 | 19318 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; |
d2dcc67d DM |
19319 | int struct_meta_reg = BPF_REG_3; |
19320 | int node_offset_reg = BPF_REG_4; | |
19321 | ||
19322 | /* rbtree_add has extra 'less' arg, so args-to-fixup are in diff regs */ | |
19323 | if (desc->func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
19324 | struct_meta_reg = BPF_REG_4; | |
19325 | node_offset_reg = BPF_REG_5; | |
19326 | } | |
19327 | ||
f0d991a0 DM |
19328 | if (!kptr_struct_meta) { |
19329 | verbose(env, "verifier internal error: kptr_struct_meta expected at insn_idx %d\n", | |
19330 | insn_idx); | |
19331 | return -EFAULT; | |
19332 | } | |
19333 | ||
d2dcc67d DM |
19334 | __fixup_collection_insert_kfunc(&env->insn_aux_data[insn_idx], struct_meta_reg, |
19335 | node_offset_reg, insn, insn_buf, cnt); | |
a35b9af4 YS |
19336 | } else if (desc->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx] || |
19337 | desc->func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { | |
fd264ca0 YS |
19338 | insn_buf[0] = BPF_MOV64_REG(BPF_REG_0, BPF_REG_1); |
19339 | *cnt = 1; | |
958cf2e2 | 19340 | } |
e6ac2450 MKL |
19341 | return 0; |
19342 | } | |
19343 | ||
335d1c5b | 19344 | /* The function requires that first instruction in 'patch' is insnsi[prog->len - 1] */ |
f18b03fa | 19345 | static int add_hidden_subprog(struct bpf_verifier_env *env, struct bpf_insn *patch, int len) |
335d1c5b KKD |
19346 | { |
19347 | struct bpf_subprog_info *info = env->subprog_info; | |
19348 | int cnt = env->subprog_cnt; | |
19349 | struct bpf_prog *prog; | |
19350 | ||
19351 | /* We only reserve one slot for hidden subprogs in subprog_info. */ | |
19352 | if (env->hidden_subprog_cnt) { | |
19353 | verbose(env, "verifier internal error: only one hidden subprog supported\n"); | |
19354 | return -EFAULT; | |
19355 | } | |
19356 | /* We're not patching any existing instruction, just appending the new | |
19357 | * ones for the hidden subprog. Hence all of the adjustment operations | |
19358 | * in bpf_patch_insn_data are no-ops. | |
19359 | */ | |
19360 | prog = bpf_patch_insn_data(env, env->prog->len - 1, patch, len); | |
19361 | if (!prog) | |
19362 | return -ENOMEM; | |
19363 | env->prog = prog; | |
19364 | info[cnt + 1].start = info[cnt].start; | |
19365 | info[cnt].start = prog->len - len + 1; | |
19366 | env->subprog_cnt++; | |
19367 | env->hidden_subprog_cnt++; | |
19368 | return 0; | |
19369 | } | |
19370 | ||
e6ac5933 BJ |
19371 | /* Do various post-verification rewrites in a single program pass. |
19372 | * These rewrites simplify JIT and interpreter implementations. | |
e245c5c6 | 19373 | */ |
e6ac5933 | 19374 | static int do_misc_fixups(struct bpf_verifier_env *env) |
e245c5c6 | 19375 | { |
79741b3b | 19376 | struct bpf_prog *prog = env->prog; |
f92c1e18 | 19377 | enum bpf_attach_type eatype = prog->expected_attach_type; |
9b99edca | 19378 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
79741b3b | 19379 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 19380 | const struct bpf_func_proto *fn; |
79741b3b | 19381 | const int insn_cnt = prog->len; |
09772d92 | 19382 | const struct bpf_map_ops *ops; |
c93552c4 | 19383 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
19384 | struct bpf_insn insn_buf[16]; |
19385 | struct bpf_prog *new_prog; | |
19386 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 19387 | int i, ret, cnt, delta = 0; |
e245c5c6 | 19388 | |
f18b03fa KKD |
19389 | if (env->seen_exception && !env->exception_callback_subprog) { |
19390 | struct bpf_insn patch[] = { | |
19391 | env->prog->insnsi[insn_cnt - 1], | |
a923819f | 19392 | BPF_MOV64_REG(BPF_REG_0, BPF_REG_1), |
f18b03fa KKD |
19393 | BPF_EXIT_INSN(), |
19394 | }; | |
19395 | ||
19396 | ret = add_hidden_subprog(env, patch, ARRAY_SIZE(patch)); | |
19397 | if (ret < 0) | |
19398 | return ret; | |
19399 | prog = env->prog; | |
19400 | insn = prog->insnsi; | |
19401 | ||
19402 | env->exception_callback_subprog = env->subprog_cnt - 1; | |
19403 | /* Don't update insn_cnt, as add_hidden_subprog always appends insns */ | |
19404 | env->subprog_info[env->exception_callback_subprog].is_cb = true; | |
19405 | env->subprog_info[env->exception_callback_subprog].is_async_cb = true; | |
19406 | env->subprog_info[env->exception_callback_subprog].is_exception_cb = true; | |
19407 | } | |
19408 | ||
79741b3b | 19409 | for (i = 0; i < insn_cnt; i++, insn++) { |
e6ac5933 | 19410 | /* Make divide-by-zero exceptions impossible. */ |
f6b1b3bf DB |
19411 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
19412 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
19413 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 19414 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf | 19415 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
e88b2c6e DB |
19416 | bool isdiv = BPF_OP(insn->code) == BPF_DIV; |
19417 | struct bpf_insn *patchlet; | |
19418 | struct bpf_insn chk_and_div[] = { | |
9b00f1b7 | 19419 | /* [R,W]x div 0 -> 0 */ |
e88b2c6e DB |
19420 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
19421 | BPF_JNE | BPF_K, insn->src_reg, | |
19422 | 0, 2, 0), | |
f6b1b3bf DB |
19423 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), |
19424 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
19425 | *insn, | |
19426 | }; | |
e88b2c6e | 19427 | struct bpf_insn chk_and_mod[] = { |
9b00f1b7 | 19428 | /* [R,W]x mod 0 -> [R,W]x */ |
e88b2c6e DB |
19429 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
19430 | BPF_JEQ | BPF_K, insn->src_reg, | |
9b00f1b7 | 19431 | 0, 1 + (is64 ? 0 : 1), 0), |
f6b1b3bf | 19432 | *insn, |
9b00f1b7 DB |
19433 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), |
19434 | BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), | |
f6b1b3bf | 19435 | }; |
f6b1b3bf | 19436 | |
e88b2c6e DB |
19437 | patchlet = isdiv ? chk_and_div : chk_and_mod; |
19438 | cnt = isdiv ? ARRAY_SIZE(chk_and_div) : | |
9b00f1b7 | 19439 | ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0); |
f6b1b3bf DB |
19440 | |
19441 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
19442 | if (!new_prog) |
19443 | return -ENOMEM; | |
19444 | ||
19445 | delta += cnt - 1; | |
19446 | env->prog = prog = new_prog; | |
19447 | insn = new_prog->insnsi + i + delta; | |
19448 | continue; | |
19449 | } | |
19450 | ||
e6ac5933 | 19451 | /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */ |
e0cea7ce DB |
19452 | if (BPF_CLASS(insn->code) == BPF_LD && |
19453 | (BPF_MODE(insn->code) == BPF_ABS || | |
19454 | BPF_MODE(insn->code) == BPF_IND)) { | |
19455 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
19456 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
19457 | verbose(env, "bpf verifier is misconfigured\n"); | |
19458 | return -EINVAL; | |
19459 | } | |
19460 | ||
19461 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19462 | if (!new_prog) | |
19463 | return -ENOMEM; | |
19464 | ||
19465 | delta += cnt - 1; | |
19466 | env->prog = prog = new_prog; | |
19467 | insn = new_prog->insnsi + i + delta; | |
19468 | continue; | |
19469 | } | |
19470 | ||
e6ac5933 | 19471 | /* Rewrite pointer arithmetic to mitigate speculation attacks. */ |
979d63d5 DB |
19472 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
19473 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
19474 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
19475 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
979d63d5 | 19476 | struct bpf_insn *patch = &insn_buf[0]; |
801c6058 | 19477 | bool issrc, isneg, isimm; |
979d63d5 DB |
19478 | u32 off_reg; |
19479 | ||
19480 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
19481 | if (!aux->alu_state || |
19482 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
19483 | continue; |
19484 | ||
19485 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
19486 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
19487 | BPF_ALU_SANITIZE_SRC; | |
801c6058 | 19488 | isimm = aux->alu_state & BPF_ALU_IMMEDIATE; |
979d63d5 DB |
19489 | |
19490 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
801c6058 DB |
19491 | if (isimm) { |
19492 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
19493 | } else { | |
19494 | if (isneg) | |
19495 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
19496 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
19497 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
19498 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
19499 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
19500 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
19501 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, off_reg); | |
19502 | } | |
b9b34ddb DB |
19503 | if (!issrc) |
19504 | *patch++ = BPF_MOV64_REG(insn->dst_reg, insn->src_reg); | |
19505 | insn->src_reg = BPF_REG_AX; | |
979d63d5 DB |
19506 | if (isneg) |
19507 | insn->code = insn->code == code_add ? | |
19508 | code_sub : code_add; | |
19509 | *patch++ = *insn; | |
801c6058 | 19510 | if (issrc && isneg && !isimm) |
979d63d5 DB |
19511 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); |
19512 | cnt = patch - insn_buf; | |
19513 | ||
19514 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19515 | if (!new_prog) | |
19516 | return -ENOMEM; | |
19517 | ||
19518 | delta += cnt - 1; | |
19519 | env->prog = prog = new_prog; | |
19520 | insn = new_prog->insnsi + i + delta; | |
19521 | continue; | |
19522 | } | |
19523 | ||
79741b3b AS |
19524 | if (insn->code != (BPF_JMP | BPF_CALL)) |
19525 | continue; | |
cc8b0b92 AS |
19526 | if (insn->src_reg == BPF_PSEUDO_CALL) |
19527 | continue; | |
e6ac2450 | 19528 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
958cf2e2 | 19529 | ret = fixup_kfunc_call(env, insn, insn_buf, i + delta, &cnt); |
e6ac2450 MKL |
19530 | if (ret) |
19531 | return ret; | |
958cf2e2 KKD |
19532 | if (cnt == 0) |
19533 | continue; | |
19534 | ||
19535 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19536 | if (!new_prog) | |
19537 | return -ENOMEM; | |
19538 | ||
19539 | delta += cnt - 1; | |
19540 | env->prog = prog = new_prog; | |
19541 | insn = new_prog->insnsi + i + delta; | |
e6ac2450 MKL |
19542 | continue; |
19543 | } | |
e245c5c6 | 19544 | |
79741b3b AS |
19545 | if (insn->imm == BPF_FUNC_get_route_realm) |
19546 | prog->dst_needed = 1; | |
19547 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
19548 | bpf_user_rnd_init_once(); | |
9802d865 JB |
19549 | if (insn->imm == BPF_FUNC_override_return) |
19550 | prog->kprobe_override = 1; | |
79741b3b | 19551 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
19552 | /* If we tail call into other programs, we |
19553 | * cannot make any assumptions since they can | |
19554 | * be replaced dynamically during runtime in | |
19555 | * the program array. | |
19556 | */ | |
19557 | prog->cb_access = 1; | |
e411901c MF |
19558 | if (!allow_tail_call_in_subprogs(env)) |
19559 | prog->aux->stack_depth = MAX_BPF_STACK; | |
19560 | prog->aux->max_pkt_offset = MAX_PACKET_OFF; | |
7b9f6da1 | 19561 | |
79741b3b | 19562 | /* mark bpf_tail_call as different opcode to avoid |
8fb33b60 | 19563 | * conditional branch in the interpreter for every normal |
79741b3b AS |
19564 | * call and to prevent accidental JITing by JIT compiler |
19565 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 19566 | */ |
79741b3b | 19567 | insn->imm = 0; |
71189fa9 | 19568 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 19569 | |
c93552c4 | 19570 | aux = &env->insn_aux_data[i + delta]; |
d2a3b7c5 | 19571 | if (env->bpf_capable && !prog->blinding_requested && |
cc52d914 | 19572 | prog->jit_requested && |
d2e4c1e6 DB |
19573 | !bpf_map_key_poisoned(aux) && |
19574 | !bpf_map_ptr_poisoned(aux) && | |
19575 | !bpf_map_ptr_unpriv(aux)) { | |
19576 | struct bpf_jit_poke_descriptor desc = { | |
19577 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
19578 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
19579 | .tail_call.key = bpf_map_key_immediate(aux), | |
a748c697 | 19580 | .insn_idx = i + delta, |
d2e4c1e6 DB |
19581 | }; |
19582 | ||
19583 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
19584 | if (ret < 0) { | |
19585 | verbose(env, "adding tail call poke descriptor failed\n"); | |
19586 | return ret; | |
19587 | } | |
19588 | ||
19589 | insn->imm = ret + 1; | |
19590 | continue; | |
19591 | } | |
19592 | ||
c93552c4 DB |
19593 | if (!bpf_map_ptr_unpriv(aux)) |
19594 | continue; | |
19595 | ||
b2157399 AS |
19596 | /* instead of changing every JIT dealing with tail_call |
19597 | * emit two extra insns: | |
19598 | * if (index >= max_entries) goto out; | |
19599 | * index &= array->index_mask; | |
19600 | * to avoid out-of-bounds cpu speculation | |
19601 | */ | |
c93552c4 | 19602 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 19603 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
19604 | return -EINVAL; |
19605 | } | |
c93552c4 | 19606 | |
d2e4c1e6 | 19607 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
19608 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
19609 | map_ptr->max_entries, 2); | |
19610 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
19611 | container_of(map_ptr, | |
19612 | struct bpf_array, | |
19613 | map)->index_mask); | |
19614 | insn_buf[2] = *insn; | |
19615 | cnt = 3; | |
19616 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19617 | if (!new_prog) | |
19618 | return -ENOMEM; | |
19619 | ||
19620 | delta += cnt - 1; | |
19621 | env->prog = prog = new_prog; | |
19622 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
19623 | continue; |
19624 | } | |
e245c5c6 | 19625 | |
b00628b1 AS |
19626 | if (insn->imm == BPF_FUNC_timer_set_callback) { |
19627 | /* The verifier will process callback_fn as many times as necessary | |
19628 | * with different maps and the register states prepared by | |
19629 | * set_timer_callback_state will be accurate. | |
19630 | * | |
19631 | * The following use case is valid: | |
19632 | * map1 is shared by prog1, prog2, prog3. | |
19633 | * prog1 calls bpf_timer_init for some map1 elements | |
19634 | * prog2 calls bpf_timer_set_callback for some map1 elements. | |
19635 | * Those that were not bpf_timer_init-ed will return -EINVAL. | |
19636 | * prog3 calls bpf_timer_start for some map1 elements. | |
19637 | * Those that were not both bpf_timer_init-ed and | |
19638 | * bpf_timer_set_callback-ed will return -EINVAL. | |
19639 | */ | |
19640 | struct bpf_insn ld_addrs[2] = { | |
19641 | BPF_LD_IMM64(BPF_REG_3, (long)prog->aux), | |
19642 | }; | |
19643 | ||
19644 | insn_buf[0] = ld_addrs[0]; | |
19645 | insn_buf[1] = ld_addrs[1]; | |
19646 | insn_buf[2] = *insn; | |
19647 | cnt = 3; | |
19648 | ||
19649 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19650 | if (!new_prog) | |
19651 | return -ENOMEM; | |
19652 | ||
19653 | delta += cnt - 1; | |
19654 | env->prog = prog = new_prog; | |
19655 | insn = new_prog->insnsi + i + delta; | |
19656 | goto patch_call_imm; | |
19657 | } | |
19658 | ||
9bb00b28 YS |
19659 | if (is_storage_get_function(insn->imm)) { |
19660 | if (!env->prog->aux->sleepable || | |
19661 | env->insn_aux_data[i + delta].storage_get_func_atomic) | |
d56c9fe6 | 19662 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_ATOMIC); |
9bb00b28 YS |
19663 | else |
19664 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_KERNEL); | |
b00fa38a JK |
19665 | insn_buf[1] = *insn; |
19666 | cnt = 2; | |
19667 | ||
19668 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19669 | if (!new_prog) | |
19670 | return -ENOMEM; | |
19671 | ||
19672 | delta += cnt - 1; | |
19673 | env->prog = prog = new_prog; | |
19674 | insn = new_prog->insnsi + i + delta; | |
19675 | goto patch_call_imm; | |
19676 | } | |
19677 | ||
01cc55af YS |
19678 | /* bpf_per_cpu_ptr() and bpf_this_cpu_ptr() */ |
19679 | if (env->insn_aux_data[i + delta].call_with_percpu_alloc_ptr) { | |
19680 | /* patch with 'r1 = *(u64 *)(r1 + 0)' since for percpu data, | |
19681 | * bpf_mem_alloc() returns a ptr to the percpu data ptr. | |
19682 | */ | |
19683 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_1, BPF_REG_1, 0); | |
19684 | insn_buf[1] = *insn; | |
19685 | cnt = 2; | |
19686 | ||
19687 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19688 | if (!new_prog) | |
19689 | return -ENOMEM; | |
19690 | ||
19691 | delta += cnt - 1; | |
19692 | env->prog = prog = new_prog; | |
19693 | insn = new_prog->insnsi + i + delta; | |
19694 | goto patch_call_imm; | |
19695 | } | |
19696 | ||
89c63074 | 19697 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
19698 | * and other inlining handlers are currently limited to 64 bit |
19699 | * only. | |
89c63074 | 19700 | */ |
60b58afc | 19701 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
19702 | (insn->imm == BPF_FUNC_map_lookup_elem || |
19703 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
19704 | insn->imm == BPF_FUNC_map_delete_elem || |
19705 | insn->imm == BPF_FUNC_map_push_elem || | |
19706 | insn->imm == BPF_FUNC_map_pop_elem || | |
e6a4750f | 19707 | insn->imm == BPF_FUNC_map_peek_elem || |
0640c77c | 19708 | insn->imm == BPF_FUNC_redirect_map || |
07343110 FZ |
19709 | insn->imm == BPF_FUNC_for_each_map_elem || |
19710 | insn->imm == BPF_FUNC_map_lookup_percpu_elem)) { | |
c93552c4 DB |
19711 | aux = &env->insn_aux_data[i + delta]; |
19712 | if (bpf_map_ptr_poisoned(aux)) | |
19713 | goto patch_call_imm; | |
19714 | ||
d2e4c1e6 | 19715 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
19716 | ops = map_ptr->ops; |
19717 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
19718 | ops->map_gen_lookup) { | |
19719 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
4a8f87e6 DB |
19720 | if (cnt == -EOPNOTSUPP) |
19721 | goto patch_map_ops_generic; | |
19722 | if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
09772d92 DB |
19723 | verbose(env, "bpf verifier is misconfigured\n"); |
19724 | return -EINVAL; | |
19725 | } | |
81ed18ab | 19726 | |
09772d92 DB |
19727 | new_prog = bpf_patch_insn_data(env, i + delta, |
19728 | insn_buf, cnt); | |
19729 | if (!new_prog) | |
19730 | return -ENOMEM; | |
81ed18ab | 19731 | |
09772d92 DB |
19732 | delta += cnt - 1; |
19733 | env->prog = prog = new_prog; | |
19734 | insn = new_prog->insnsi + i + delta; | |
19735 | continue; | |
19736 | } | |
81ed18ab | 19737 | |
09772d92 DB |
19738 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
19739 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
19740 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
d7ba4cc9 | 19741 | (long (*)(struct bpf_map *map, void *key))NULL)); |
09772d92 | 19742 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, |
d7ba4cc9 | 19743 | (long (*)(struct bpf_map *map, void *key, void *value, |
09772d92 | 19744 | u64 flags))NULL)); |
84430d42 | 19745 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
d7ba4cc9 | 19746 | (long (*)(struct bpf_map *map, void *value, |
84430d42 DB |
19747 | u64 flags))NULL)); |
19748 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
d7ba4cc9 | 19749 | (long (*)(struct bpf_map *map, void *value))NULL)); |
84430d42 | 19750 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, |
d7ba4cc9 | 19751 | (long (*)(struct bpf_map *map, void *value))NULL)); |
e6a4750f | 19752 | BUILD_BUG_ON(!__same_type(ops->map_redirect, |
d7ba4cc9 | 19753 | (long (*)(struct bpf_map *map, u64 index, u64 flags))NULL)); |
0640c77c | 19754 | BUILD_BUG_ON(!__same_type(ops->map_for_each_callback, |
d7ba4cc9 | 19755 | (long (*)(struct bpf_map *map, |
0640c77c AI |
19756 | bpf_callback_t callback_fn, |
19757 | void *callback_ctx, | |
19758 | u64 flags))NULL)); | |
07343110 FZ |
19759 | BUILD_BUG_ON(!__same_type(ops->map_lookup_percpu_elem, |
19760 | (void *(*)(struct bpf_map *map, void *key, u32 cpu))NULL)); | |
e6a4750f | 19761 | |
4a8f87e6 | 19762 | patch_map_ops_generic: |
09772d92 DB |
19763 | switch (insn->imm) { |
19764 | case BPF_FUNC_map_lookup_elem: | |
3d717fad | 19765 | insn->imm = BPF_CALL_IMM(ops->map_lookup_elem); |
09772d92 DB |
19766 | continue; |
19767 | case BPF_FUNC_map_update_elem: | |
3d717fad | 19768 | insn->imm = BPF_CALL_IMM(ops->map_update_elem); |
09772d92 DB |
19769 | continue; |
19770 | case BPF_FUNC_map_delete_elem: | |
3d717fad | 19771 | insn->imm = BPF_CALL_IMM(ops->map_delete_elem); |
09772d92 | 19772 | continue; |
84430d42 | 19773 | case BPF_FUNC_map_push_elem: |
3d717fad | 19774 | insn->imm = BPF_CALL_IMM(ops->map_push_elem); |
84430d42 DB |
19775 | continue; |
19776 | case BPF_FUNC_map_pop_elem: | |
3d717fad | 19777 | insn->imm = BPF_CALL_IMM(ops->map_pop_elem); |
84430d42 DB |
19778 | continue; |
19779 | case BPF_FUNC_map_peek_elem: | |
3d717fad | 19780 | insn->imm = BPF_CALL_IMM(ops->map_peek_elem); |
84430d42 | 19781 | continue; |
e6a4750f | 19782 | case BPF_FUNC_redirect_map: |
3d717fad | 19783 | insn->imm = BPF_CALL_IMM(ops->map_redirect); |
e6a4750f | 19784 | continue; |
0640c77c AI |
19785 | case BPF_FUNC_for_each_map_elem: |
19786 | insn->imm = BPF_CALL_IMM(ops->map_for_each_callback); | |
e6a4750f | 19787 | continue; |
07343110 FZ |
19788 | case BPF_FUNC_map_lookup_percpu_elem: |
19789 | insn->imm = BPF_CALL_IMM(ops->map_lookup_percpu_elem); | |
19790 | continue; | |
09772d92 | 19791 | } |
81ed18ab | 19792 | |
09772d92 | 19793 | goto patch_call_imm; |
81ed18ab AS |
19794 | } |
19795 | ||
e6ac5933 | 19796 | /* Implement bpf_jiffies64 inline. */ |
5576b991 MKL |
19797 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
19798 | insn->imm == BPF_FUNC_jiffies64) { | |
19799 | struct bpf_insn ld_jiffies_addr[2] = { | |
19800 | BPF_LD_IMM64(BPF_REG_0, | |
19801 | (unsigned long)&jiffies), | |
19802 | }; | |
19803 | ||
19804 | insn_buf[0] = ld_jiffies_addr[0]; | |
19805 | insn_buf[1] = ld_jiffies_addr[1]; | |
19806 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
19807 | BPF_REG_0, 0); | |
19808 | cnt = 3; | |
19809 | ||
19810 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
19811 | cnt); | |
19812 | if (!new_prog) | |
19813 | return -ENOMEM; | |
19814 | ||
19815 | delta += cnt - 1; | |
19816 | env->prog = prog = new_prog; | |
19817 | insn = new_prog->insnsi + i + delta; | |
19818 | continue; | |
19819 | } | |
19820 | ||
f92c1e18 JO |
19821 | /* Implement bpf_get_func_arg inline. */ |
19822 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
19823 | insn->imm == BPF_FUNC_get_func_arg) { | |
19824 | /* Load nr_args from ctx - 8 */ | |
19825 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
19826 | insn_buf[1] = BPF_JMP32_REG(BPF_JGE, BPF_REG_2, BPF_REG_0, 6); | |
19827 | insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 3); | |
19828 | insn_buf[3] = BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1); | |
19829 | insn_buf[4] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_2, 0); | |
19830 | insn_buf[5] = BPF_STX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
19831 | insn_buf[6] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
19832 | insn_buf[7] = BPF_JMP_A(1); | |
19833 | insn_buf[8] = BPF_MOV64_IMM(BPF_REG_0, -EINVAL); | |
19834 | cnt = 9; | |
19835 | ||
19836 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19837 | if (!new_prog) | |
19838 | return -ENOMEM; | |
19839 | ||
19840 | delta += cnt - 1; | |
19841 | env->prog = prog = new_prog; | |
19842 | insn = new_prog->insnsi + i + delta; | |
19843 | continue; | |
19844 | } | |
19845 | ||
19846 | /* Implement bpf_get_func_ret inline. */ | |
19847 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
19848 | insn->imm == BPF_FUNC_get_func_ret) { | |
19849 | if (eatype == BPF_TRACE_FEXIT || | |
19850 | eatype == BPF_MODIFY_RETURN) { | |
19851 | /* Load nr_args from ctx - 8 */ | |
19852 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
19853 | insn_buf[1] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3); | |
19854 | insn_buf[2] = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1); | |
19855 | insn_buf[3] = BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
19856 | insn_buf[4] = BPF_STX_MEM(BPF_DW, BPF_REG_2, BPF_REG_3, 0); | |
19857 | insn_buf[5] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
19858 | cnt = 6; | |
19859 | } else { | |
19860 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, -EOPNOTSUPP); | |
19861 | cnt = 1; | |
19862 | } | |
19863 | ||
19864 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
19865 | if (!new_prog) | |
19866 | return -ENOMEM; | |
19867 | ||
19868 | delta += cnt - 1; | |
19869 | env->prog = prog = new_prog; | |
19870 | insn = new_prog->insnsi + i + delta; | |
19871 | continue; | |
19872 | } | |
19873 | ||
19874 | /* Implement get_func_arg_cnt inline. */ | |
19875 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
19876 | insn->imm == BPF_FUNC_get_func_arg_cnt) { | |
19877 | /* Load nr_args from ctx - 8 */ | |
19878 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
19879 | ||
19880 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
19881 | if (!new_prog) | |
19882 | return -ENOMEM; | |
19883 | ||
19884 | env->prog = prog = new_prog; | |
19885 | insn = new_prog->insnsi + i + delta; | |
19886 | continue; | |
19887 | } | |
19888 | ||
f705ec76 | 19889 | /* Implement bpf_get_func_ip inline. */ |
9b99edca JO |
19890 | if (prog_type == BPF_PROG_TYPE_TRACING && |
19891 | insn->imm == BPF_FUNC_get_func_ip) { | |
f92c1e18 JO |
19892 | /* Load IP address from ctx - 16 */ |
19893 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -16); | |
9b99edca JO |
19894 | |
19895 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
19896 | if (!new_prog) | |
19897 | return -ENOMEM; | |
19898 | ||
19899 | env->prog = prog = new_prog; | |
19900 | insn = new_prog->insnsi + i + delta; | |
19901 | continue; | |
19902 | } | |
19903 | ||
81ed18ab | 19904 | patch_call_imm: |
5e43f899 | 19905 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
19906 | /* all functions that have prototype and verifier allowed |
19907 | * programs to call them, must be real in-kernel functions | |
19908 | */ | |
19909 | if (!fn->func) { | |
61bd5218 JK |
19910 | verbose(env, |
19911 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
19912 | func_id_name(insn->imm), insn->imm); |
19913 | return -EFAULT; | |
e245c5c6 | 19914 | } |
79741b3b | 19915 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 19916 | } |
e245c5c6 | 19917 | |
d2e4c1e6 DB |
19918 | /* Since poke tab is now finalized, publish aux to tracker. */ |
19919 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
19920 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
19921 | if (!map_ptr->ops->map_poke_track || | |
19922 | !map_ptr->ops->map_poke_untrack || | |
19923 | !map_ptr->ops->map_poke_run) { | |
19924 | verbose(env, "bpf verifier is misconfigured\n"); | |
19925 | return -EINVAL; | |
19926 | } | |
19927 | ||
19928 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
19929 | if (ret < 0) { | |
19930 | verbose(env, "tracking tail call prog failed\n"); | |
19931 | return ret; | |
19932 | } | |
19933 | } | |
19934 | ||
1cf3bfc6 | 19935 | sort_kfunc_descs_by_imm_off(env->prog); |
e6ac2450 | 19936 | |
79741b3b AS |
19937 | return 0; |
19938 | } | |
e245c5c6 | 19939 | |
1ade2371 EZ |
19940 | static struct bpf_prog *inline_bpf_loop(struct bpf_verifier_env *env, |
19941 | int position, | |
19942 | s32 stack_base, | |
19943 | u32 callback_subprogno, | |
19944 | u32 *cnt) | |
19945 | { | |
19946 | s32 r6_offset = stack_base + 0 * BPF_REG_SIZE; | |
19947 | s32 r7_offset = stack_base + 1 * BPF_REG_SIZE; | |
19948 | s32 r8_offset = stack_base + 2 * BPF_REG_SIZE; | |
19949 | int reg_loop_max = BPF_REG_6; | |
19950 | int reg_loop_cnt = BPF_REG_7; | |
19951 | int reg_loop_ctx = BPF_REG_8; | |
19952 | ||
19953 | struct bpf_prog *new_prog; | |
19954 | u32 callback_start; | |
19955 | u32 call_insn_offset; | |
19956 | s32 callback_offset; | |
19957 | ||
19958 | /* This represents an inlined version of bpf_iter.c:bpf_loop, | |
19959 | * be careful to modify this code in sync. | |
19960 | */ | |
19961 | struct bpf_insn insn_buf[] = { | |
19962 | /* Return error and jump to the end of the patch if | |
19963 | * expected number of iterations is too big. | |
19964 | */ | |
19965 | BPF_JMP_IMM(BPF_JLE, BPF_REG_1, BPF_MAX_LOOPS, 2), | |
19966 | BPF_MOV32_IMM(BPF_REG_0, -E2BIG), | |
19967 | BPF_JMP_IMM(BPF_JA, 0, 0, 16), | |
19968 | /* spill R6, R7, R8 to use these as loop vars */ | |
19969 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_6, r6_offset), | |
19970 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_7, r7_offset), | |
19971 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_8, r8_offset), | |
19972 | /* initialize loop vars */ | |
19973 | BPF_MOV64_REG(reg_loop_max, BPF_REG_1), | |
19974 | BPF_MOV32_IMM(reg_loop_cnt, 0), | |
19975 | BPF_MOV64_REG(reg_loop_ctx, BPF_REG_3), | |
19976 | /* loop header, | |
19977 | * if reg_loop_cnt >= reg_loop_max skip the loop body | |
19978 | */ | |
19979 | BPF_JMP_REG(BPF_JGE, reg_loop_cnt, reg_loop_max, 5), | |
19980 | /* callback call, | |
19981 | * correct callback offset would be set after patching | |
19982 | */ | |
19983 | BPF_MOV64_REG(BPF_REG_1, reg_loop_cnt), | |
19984 | BPF_MOV64_REG(BPF_REG_2, reg_loop_ctx), | |
19985 | BPF_CALL_REL(0), | |
19986 | /* increment loop counter */ | |
19987 | BPF_ALU64_IMM(BPF_ADD, reg_loop_cnt, 1), | |
19988 | /* jump to loop header if callback returned 0 */ | |
19989 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, -6), | |
19990 | /* return value of bpf_loop, | |
19991 | * set R0 to the number of iterations | |
19992 | */ | |
19993 | BPF_MOV64_REG(BPF_REG_0, reg_loop_cnt), | |
19994 | /* restore original values of R6, R7, R8 */ | |
19995 | BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_10, r6_offset), | |
19996 | BPF_LDX_MEM(BPF_DW, BPF_REG_7, BPF_REG_10, r7_offset), | |
19997 | BPF_LDX_MEM(BPF_DW, BPF_REG_8, BPF_REG_10, r8_offset), | |
19998 | }; | |
19999 | ||
20000 | *cnt = ARRAY_SIZE(insn_buf); | |
20001 | new_prog = bpf_patch_insn_data(env, position, insn_buf, *cnt); | |
20002 | if (!new_prog) | |
20003 | return new_prog; | |
20004 | ||
20005 | /* callback start is known only after patching */ | |
20006 | callback_start = env->subprog_info[callback_subprogno].start; | |
20007 | /* Note: insn_buf[12] is an offset of BPF_CALL_REL instruction */ | |
20008 | call_insn_offset = position + 12; | |
20009 | callback_offset = callback_start - call_insn_offset - 1; | |
fb4e3b33 | 20010 | new_prog->insnsi[call_insn_offset].imm = callback_offset; |
1ade2371 EZ |
20011 | |
20012 | return new_prog; | |
20013 | } | |
20014 | ||
20015 | static bool is_bpf_loop_call(struct bpf_insn *insn) | |
20016 | { | |
20017 | return insn->code == (BPF_JMP | BPF_CALL) && | |
20018 | insn->src_reg == 0 && | |
20019 | insn->imm == BPF_FUNC_loop; | |
20020 | } | |
20021 | ||
20022 | /* For all sub-programs in the program (including main) check | |
20023 | * insn_aux_data to see if there are bpf_loop calls that require | |
20024 | * inlining. If such calls are found the calls are replaced with a | |
20025 | * sequence of instructions produced by `inline_bpf_loop` function and | |
20026 | * subprog stack_depth is increased by the size of 3 registers. | |
20027 | * This stack space is used to spill values of the R6, R7, R8. These | |
20028 | * registers are used to store the loop bound, counter and context | |
20029 | * variables. | |
20030 | */ | |
20031 | static int optimize_bpf_loop(struct bpf_verifier_env *env) | |
20032 | { | |
20033 | struct bpf_subprog_info *subprogs = env->subprog_info; | |
20034 | int i, cur_subprog = 0, cnt, delta = 0; | |
20035 | struct bpf_insn *insn = env->prog->insnsi; | |
20036 | int insn_cnt = env->prog->len; | |
20037 | u16 stack_depth = subprogs[cur_subprog].stack_depth; | |
20038 | u16 stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
20039 | u16 stack_depth_extra = 0; | |
20040 | ||
20041 | for (i = 0; i < insn_cnt; i++, insn++) { | |
20042 | struct bpf_loop_inline_state *inline_state = | |
20043 | &env->insn_aux_data[i + delta].loop_inline_state; | |
20044 | ||
20045 | if (is_bpf_loop_call(insn) && inline_state->fit_for_inline) { | |
20046 | struct bpf_prog *new_prog; | |
20047 | ||
20048 | stack_depth_extra = BPF_REG_SIZE * 3 + stack_depth_roundup; | |
20049 | new_prog = inline_bpf_loop(env, | |
20050 | i + delta, | |
20051 | -(stack_depth + stack_depth_extra), | |
20052 | inline_state->callback_subprogno, | |
20053 | &cnt); | |
20054 | if (!new_prog) | |
20055 | return -ENOMEM; | |
20056 | ||
20057 | delta += cnt - 1; | |
20058 | env->prog = new_prog; | |
20059 | insn = new_prog->insnsi + i + delta; | |
20060 | } | |
20061 | ||
20062 | if (subprogs[cur_subprog + 1].start == i + delta + 1) { | |
20063 | subprogs[cur_subprog].stack_depth += stack_depth_extra; | |
20064 | cur_subprog++; | |
20065 | stack_depth = subprogs[cur_subprog].stack_depth; | |
20066 | stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
20067 | stack_depth_extra = 0; | |
20068 | } | |
20069 | } | |
20070 | ||
20071 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
20072 | ||
20073 | return 0; | |
20074 | } | |
20075 | ||
58e2af8b | 20076 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 20077 | { |
58e2af8b | 20078 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
20079 | int i; |
20080 | ||
9f4686c4 AS |
20081 | sl = env->free_list; |
20082 | while (sl) { | |
20083 | sln = sl->next; | |
20084 | free_verifier_state(&sl->state, false); | |
20085 | kfree(sl); | |
20086 | sl = sln; | |
20087 | } | |
51c39bb1 | 20088 | env->free_list = NULL; |
9f4686c4 | 20089 | |
f1bca824 AS |
20090 | if (!env->explored_states) |
20091 | return; | |
20092 | ||
dc2a4ebc | 20093 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
20094 | sl = env->explored_states[i]; |
20095 | ||
a8f500af AS |
20096 | while (sl) { |
20097 | sln = sl->next; | |
20098 | free_verifier_state(&sl->state, false); | |
20099 | kfree(sl); | |
20100 | sl = sln; | |
20101 | } | |
51c39bb1 | 20102 | env->explored_states[i] = NULL; |
f1bca824 | 20103 | } |
51c39bb1 | 20104 | } |
f1bca824 | 20105 | |
b9ae0c9d | 20106 | static int do_check_common(struct bpf_verifier_env *env, int subprog, bool is_ex_cb) |
51c39bb1 | 20107 | { |
6f8a57cc | 20108 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
20109 | struct bpf_verifier_state *state; |
20110 | struct bpf_reg_state *regs; | |
20111 | int ret, i; | |
20112 | ||
20113 | env->prev_linfo = NULL; | |
20114 | env->pass_cnt++; | |
20115 | ||
20116 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
20117 | if (!state) | |
20118 | return -ENOMEM; | |
20119 | state->curframe = 0; | |
20120 | state->speculative = false; | |
20121 | state->branches = 1; | |
20122 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
20123 | if (!state->frame[0]) { | |
20124 | kfree(state); | |
20125 | return -ENOMEM; | |
20126 | } | |
20127 | env->cur_state = state; | |
20128 | init_func_state(env, state->frame[0], | |
20129 | BPF_MAIN_FUNC /* callsite */, | |
20130 | 0 /* frameno */, | |
20131 | subprog); | |
be2ef816 AN |
20132 | state->first_insn_idx = env->subprog_info[subprog].start; |
20133 | state->last_insn_idx = -1; | |
51c39bb1 AS |
20134 | |
20135 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 20136 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
b9ae0c9d | 20137 | ret = btf_prepare_func_args(env, subprog, regs, is_ex_cb); |
51c39bb1 AS |
20138 | if (ret) |
20139 | goto out; | |
20140 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
20141 | if (regs[i].type == PTR_TO_CTX) | |
20142 | mark_reg_known_zero(env, regs, i); | |
20143 | else if (regs[i].type == SCALAR_VALUE) | |
20144 | mark_reg_unknown(env, regs, i); | |
cf9f2f8d | 20145 | else if (base_type(regs[i].type) == PTR_TO_MEM) { |
e5069b9c DB |
20146 | const u32 mem_size = regs[i].mem_size; |
20147 | ||
20148 | mark_reg_known_zero(env, regs, i); | |
20149 | regs[i].mem_size = mem_size; | |
20150 | regs[i].id = ++env->id_gen; | |
20151 | } | |
51c39bb1 | 20152 | } |
b9ae0c9d KKD |
20153 | if (is_ex_cb) { |
20154 | state->frame[0]->in_exception_callback_fn = true; | |
20155 | env->subprog_info[subprog].is_cb = true; | |
20156 | env->subprog_info[subprog].is_async_cb = true; | |
20157 | env->subprog_info[subprog].is_exception_cb = true; | |
20158 | } | |
51c39bb1 AS |
20159 | } else { |
20160 | /* 1st arg to a function */ | |
20161 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
20162 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
34747c41 | 20163 | ret = btf_check_subprog_arg_match(env, subprog, regs); |
51c39bb1 AS |
20164 | if (ret == -EFAULT) |
20165 | /* unlikely verifier bug. abort. | |
20166 | * ret == 0 and ret < 0 are sadly acceptable for | |
20167 | * main() function due to backward compatibility. | |
20168 | * Like socket filter program may be written as: | |
20169 | * int bpf_prog(struct pt_regs *ctx) | |
20170 | * and never dereference that ctx in the program. | |
20171 | * 'struct pt_regs' is a type mismatch for socket | |
20172 | * filter that should be using 'struct __sk_buff'. | |
20173 | */ | |
20174 | goto out; | |
20175 | } | |
20176 | ||
20177 | ret = do_check(env); | |
20178 | out: | |
f59bbfc2 AS |
20179 | /* check for NULL is necessary, since cur_state can be freed inside |
20180 | * do_check() under memory pressure. | |
20181 | */ | |
20182 | if (env->cur_state) { | |
20183 | free_verifier_state(env->cur_state, true); | |
20184 | env->cur_state = NULL; | |
20185 | } | |
6f8a57cc AN |
20186 | while (!pop_stack(env, NULL, NULL, false)); |
20187 | if (!ret && pop_log) | |
20188 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 | 20189 | free_states(env); |
51c39bb1 AS |
20190 | return ret; |
20191 | } | |
20192 | ||
20193 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
20194 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
20195 | * Consider: | |
20196 | * int bar(int); | |
20197 | * int foo(int f) | |
20198 | * { | |
20199 | * return bar(f); | |
20200 | * } | |
20201 | * int bar(int b) | |
20202 | * { | |
20203 | * ... | |
20204 | * } | |
20205 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
20206 | * will be assumed that bar() already verified successfully and call to bar() | |
20207 | * from foo() will be checked for type match only. Later bar() will be verified | |
20208 | * independently to check that it's safe for R1=any_scalar_value. | |
20209 | */ | |
20210 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
20211 | { | |
20212 | struct bpf_prog_aux *aux = env->prog->aux; | |
20213 | int i, ret; | |
20214 | ||
20215 | if (!aux->func_info) | |
20216 | return 0; | |
20217 | ||
20218 | for (i = 1; i < env->subprog_cnt; i++) { | |
20219 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
20220 | continue; | |
20221 | env->insn_idx = env->subprog_info[i].start; | |
20222 | WARN_ON_ONCE(env->insn_idx == 0); | |
b9ae0c9d | 20223 | ret = do_check_common(env, i, env->exception_callback_subprog == i); |
51c39bb1 AS |
20224 | if (ret) { |
20225 | return ret; | |
20226 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
20227 | verbose(env, | |
20228 | "Func#%d is safe for any args that match its prototype\n", | |
20229 | i); | |
20230 | } | |
20231 | } | |
20232 | return 0; | |
20233 | } | |
20234 | ||
20235 | static int do_check_main(struct bpf_verifier_env *env) | |
20236 | { | |
20237 | int ret; | |
20238 | ||
20239 | env->insn_idx = 0; | |
b9ae0c9d | 20240 | ret = do_check_common(env, 0, false); |
51c39bb1 AS |
20241 | if (!ret) |
20242 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
20243 | return ret; | |
20244 | } | |
20245 | ||
20246 | ||
06ee7115 AS |
20247 | static void print_verification_stats(struct bpf_verifier_env *env) |
20248 | { | |
20249 | int i; | |
20250 | ||
20251 | if (env->log.level & BPF_LOG_STATS) { | |
20252 | verbose(env, "verification time %lld usec\n", | |
20253 | div_u64(env->verification_time, 1000)); | |
20254 | verbose(env, "stack depth "); | |
20255 | for (i = 0; i < env->subprog_cnt; i++) { | |
20256 | u32 depth = env->subprog_info[i].stack_depth; | |
20257 | ||
20258 | verbose(env, "%d", depth); | |
20259 | if (i + 1 < env->subprog_cnt) | |
20260 | verbose(env, "+"); | |
20261 | } | |
20262 | verbose(env, "\n"); | |
20263 | } | |
20264 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
20265 | "total_states %d peak_states %d mark_read %d\n", | |
20266 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
20267 | env->max_states_per_insn, env->total_states, | |
20268 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
20269 | } |
20270 | ||
27ae7997 MKL |
20271 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
20272 | { | |
20273 | const struct btf_type *t, *func_proto; | |
20274 | const struct bpf_struct_ops *st_ops; | |
20275 | const struct btf_member *member; | |
20276 | struct bpf_prog *prog = env->prog; | |
20277 | u32 btf_id, member_idx; | |
20278 | const char *mname; | |
20279 | ||
12aa8a94 THJ |
20280 | if (!prog->gpl_compatible) { |
20281 | verbose(env, "struct ops programs must have a GPL compatible license\n"); | |
20282 | return -EINVAL; | |
20283 | } | |
20284 | ||
27ae7997 MKL |
20285 | btf_id = prog->aux->attach_btf_id; |
20286 | st_ops = bpf_struct_ops_find(btf_id); | |
20287 | if (!st_ops) { | |
20288 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
20289 | btf_id); | |
20290 | return -ENOTSUPP; | |
20291 | } | |
20292 | ||
20293 | t = st_ops->type; | |
20294 | member_idx = prog->expected_attach_type; | |
20295 | if (member_idx >= btf_type_vlen(t)) { | |
20296 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
20297 | member_idx, st_ops->name); | |
20298 | return -EINVAL; | |
20299 | } | |
20300 | ||
20301 | member = &btf_type_member(t)[member_idx]; | |
20302 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
20303 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
20304 | NULL); | |
20305 | if (!func_proto) { | |
20306 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
20307 | mname, member_idx, st_ops->name); | |
20308 | return -EINVAL; | |
20309 | } | |
20310 | ||
20311 | if (st_ops->check_member) { | |
51a52a29 | 20312 | int err = st_ops->check_member(t, member, prog); |
27ae7997 MKL |
20313 | |
20314 | if (err) { | |
20315 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
20316 | mname, st_ops->name); | |
20317 | return err; | |
20318 | } | |
20319 | } | |
20320 | ||
20321 | prog->aux->attach_func_proto = func_proto; | |
20322 | prog->aux->attach_func_name = mname; | |
20323 | env->ops = st_ops->verifier_ops; | |
20324 | ||
20325 | return 0; | |
20326 | } | |
6ba43b76 KS |
20327 | #define SECURITY_PREFIX "security_" |
20328 | ||
f7b12b6f | 20329 | static int check_attach_modify_return(unsigned long addr, const char *func_name) |
6ba43b76 | 20330 | { |
69191754 | 20331 | if (within_error_injection_list(addr) || |
f7b12b6f | 20332 | !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1)) |
6ba43b76 | 20333 | return 0; |
6ba43b76 | 20334 | |
6ba43b76 KS |
20335 | return -EINVAL; |
20336 | } | |
27ae7997 | 20337 | |
1e6c62a8 AS |
20338 | /* list of non-sleepable functions that are otherwise on |
20339 | * ALLOW_ERROR_INJECTION list | |
20340 | */ | |
20341 | BTF_SET_START(btf_non_sleepable_error_inject) | |
20342 | /* Three functions below can be called from sleepable and non-sleepable context. | |
20343 | * Assume non-sleepable from bpf safety point of view. | |
20344 | */ | |
9dd3d069 | 20345 | BTF_ID(func, __filemap_add_folio) |
1e6c62a8 AS |
20346 | BTF_ID(func, should_fail_alloc_page) |
20347 | BTF_ID(func, should_failslab) | |
20348 | BTF_SET_END(btf_non_sleepable_error_inject) | |
20349 | ||
20350 | static int check_non_sleepable_error_inject(u32 btf_id) | |
20351 | { | |
20352 | return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id); | |
20353 | } | |
20354 | ||
f7b12b6f THJ |
20355 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
20356 | const struct bpf_prog *prog, | |
20357 | const struct bpf_prog *tgt_prog, | |
20358 | u32 btf_id, | |
20359 | struct bpf_attach_target_info *tgt_info) | |
38207291 | 20360 | { |
be8704ff | 20361 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
f1b9509c | 20362 | const char prefix[] = "btf_trace_"; |
5b92a28a | 20363 | int ret = 0, subprog = -1, i; |
38207291 | 20364 | const struct btf_type *t; |
5b92a28a | 20365 | bool conservative = true; |
38207291 | 20366 | const char *tname; |
5b92a28a | 20367 | struct btf *btf; |
f7b12b6f | 20368 | long addr = 0; |
31bf1dbc | 20369 | struct module *mod = NULL; |
38207291 | 20370 | |
f1b9509c | 20371 | if (!btf_id) { |
efc68158 | 20372 | bpf_log(log, "Tracing programs must provide btf_id\n"); |
f1b9509c AS |
20373 | return -EINVAL; |
20374 | } | |
22dc4a0f | 20375 | btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf; |
5b92a28a | 20376 | if (!btf) { |
efc68158 | 20377 | bpf_log(log, |
5b92a28a AS |
20378 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); |
20379 | return -EINVAL; | |
20380 | } | |
20381 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c | 20382 | if (!t) { |
efc68158 | 20383 | bpf_log(log, "attach_btf_id %u is invalid\n", btf_id); |
f1b9509c AS |
20384 | return -EINVAL; |
20385 | } | |
5b92a28a | 20386 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c | 20387 | if (!tname) { |
efc68158 | 20388 | bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id); |
f1b9509c AS |
20389 | return -EINVAL; |
20390 | } | |
5b92a28a AS |
20391 | if (tgt_prog) { |
20392 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
20393 | ||
fd7c211d THJ |
20394 | if (bpf_prog_is_dev_bound(prog->aux) && |
20395 | !bpf_prog_dev_bound_match(prog, tgt_prog)) { | |
20396 | bpf_log(log, "Target program bound device mismatch"); | |
3d76a4d3 SF |
20397 | return -EINVAL; |
20398 | } | |
20399 | ||
5b92a28a AS |
20400 | for (i = 0; i < aux->func_info_cnt; i++) |
20401 | if (aux->func_info[i].type_id == btf_id) { | |
20402 | subprog = i; | |
20403 | break; | |
20404 | } | |
20405 | if (subprog == -1) { | |
efc68158 | 20406 | bpf_log(log, "Subprog %s doesn't exist\n", tname); |
5b92a28a AS |
20407 | return -EINVAL; |
20408 | } | |
fd548e1a KKD |
20409 | if (aux->func && aux->func[subprog]->aux->exception_cb) { |
20410 | bpf_log(log, | |
20411 | "%s programs cannot attach to exception callback\n", | |
20412 | prog_extension ? "Extension" : "FENTRY/FEXIT"); | |
20413 | return -EINVAL; | |
20414 | } | |
5b92a28a | 20415 | conservative = aux->func_info_aux[subprog].unreliable; |
be8704ff AS |
20416 | if (prog_extension) { |
20417 | if (conservative) { | |
efc68158 | 20418 | bpf_log(log, |
be8704ff AS |
20419 | "Cannot replace static functions\n"); |
20420 | return -EINVAL; | |
20421 | } | |
20422 | if (!prog->jit_requested) { | |
efc68158 | 20423 | bpf_log(log, |
be8704ff AS |
20424 | "Extension programs should be JITed\n"); |
20425 | return -EINVAL; | |
20426 | } | |
be8704ff AS |
20427 | } |
20428 | if (!tgt_prog->jited) { | |
efc68158 | 20429 | bpf_log(log, "Can attach to only JITed progs\n"); |
be8704ff AS |
20430 | return -EINVAL; |
20431 | } | |
20432 | if (tgt_prog->type == prog->type) { | |
20433 | /* Cannot fentry/fexit another fentry/fexit program. | |
20434 | * Cannot attach program extension to another extension. | |
20435 | * It's ok to attach fentry/fexit to extension program. | |
20436 | */ | |
efc68158 | 20437 | bpf_log(log, "Cannot recursively attach\n"); |
be8704ff AS |
20438 | return -EINVAL; |
20439 | } | |
20440 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
20441 | prog_extension && | |
20442 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
20443 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
20444 | /* Program extensions can extend all program types | |
20445 | * except fentry/fexit. The reason is the following. | |
20446 | * The fentry/fexit programs are used for performance | |
20447 | * analysis, stats and can be attached to any program | |
20448 | * type except themselves. When extension program is | |
20449 | * replacing XDP function it is necessary to allow | |
20450 | * performance analysis of all functions. Both original | |
20451 | * XDP program and its program extension. Hence | |
20452 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
20453 | * allowed. If extending of fentry/fexit was allowed it | |
20454 | * would be possible to create long call chain | |
20455 | * fentry->extension->fentry->extension beyond | |
20456 | * reasonable stack size. Hence extending fentry is not | |
20457 | * allowed. | |
20458 | */ | |
efc68158 | 20459 | bpf_log(log, "Cannot extend fentry/fexit\n"); |
be8704ff AS |
20460 | return -EINVAL; |
20461 | } | |
5b92a28a | 20462 | } else { |
be8704ff | 20463 | if (prog_extension) { |
efc68158 | 20464 | bpf_log(log, "Cannot replace kernel functions\n"); |
be8704ff AS |
20465 | return -EINVAL; |
20466 | } | |
5b92a28a | 20467 | } |
f1b9509c AS |
20468 | |
20469 | switch (prog->expected_attach_type) { | |
20470 | case BPF_TRACE_RAW_TP: | |
5b92a28a | 20471 | if (tgt_prog) { |
efc68158 | 20472 | bpf_log(log, |
5b92a28a AS |
20473 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); |
20474 | return -EINVAL; | |
20475 | } | |
38207291 | 20476 | if (!btf_type_is_typedef(t)) { |
efc68158 | 20477 | bpf_log(log, "attach_btf_id %u is not a typedef\n", |
38207291 MKL |
20478 | btf_id); |
20479 | return -EINVAL; | |
20480 | } | |
f1b9509c | 20481 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
efc68158 | 20482 | bpf_log(log, "attach_btf_id %u points to wrong type name %s\n", |
38207291 MKL |
20483 | btf_id, tname); |
20484 | return -EINVAL; | |
20485 | } | |
20486 | tname += sizeof(prefix) - 1; | |
5b92a28a | 20487 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
20488 | if (!btf_type_is_ptr(t)) |
20489 | /* should never happen in valid vmlinux build */ | |
20490 | return -EINVAL; | |
5b92a28a | 20491 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
20492 | if (!btf_type_is_func_proto(t)) |
20493 | /* should never happen in valid vmlinux build */ | |
20494 | return -EINVAL; | |
20495 | ||
f7b12b6f | 20496 | break; |
15d83c4d YS |
20497 | case BPF_TRACE_ITER: |
20498 | if (!btf_type_is_func(t)) { | |
efc68158 | 20499 | bpf_log(log, "attach_btf_id %u is not a function\n", |
15d83c4d YS |
20500 | btf_id); |
20501 | return -EINVAL; | |
20502 | } | |
20503 | t = btf_type_by_id(btf, t->type); | |
20504 | if (!btf_type_is_func_proto(t)) | |
20505 | return -EINVAL; | |
f7b12b6f THJ |
20506 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); |
20507 | if (ret) | |
20508 | return ret; | |
20509 | break; | |
be8704ff AS |
20510 | default: |
20511 | if (!prog_extension) | |
20512 | return -EINVAL; | |
df561f66 | 20513 | fallthrough; |
ae240823 | 20514 | case BPF_MODIFY_RETURN: |
9e4e01df | 20515 | case BPF_LSM_MAC: |
69fd337a | 20516 | case BPF_LSM_CGROUP: |
fec56f58 AS |
20517 | case BPF_TRACE_FENTRY: |
20518 | case BPF_TRACE_FEXIT: | |
20519 | if (!btf_type_is_func(t)) { | |
efc68158 | 20520 | bpf_log(log, "attach_btf_id %u is not a function\n", |
fec56f58 AS |
20521 | btf_id); |
20522 | return -EINVAL; | |
20523 | } | |
be8704ff | 20524 | if (prog_extension && |
efc68158 | 20525 | btf_check_type_match(log, prog, btf, t)) |
be8704ff | 20526 | return -EINVAL; |
5b92a28a | 20527 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
20528 | if (!btf_type_is_func_proto(t)) |
20529 | return -EINVAL; | |
f7b12b6f | 20530 | |
4a1e7c0c THJ |
20531 | if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) && |
20532 | (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type || | |
20533 | prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type)) | |
20534 | return -EINVAL; | |
20535 | ||
f7b12b6f | 20536 | if (tgt_prog && conservative) |
5b92a28a | 20537 | t = NULL; |
f7b12b6f THJ |
20538 | |
20539 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); | |
fec56f58 | 20540 | if (ret < 0) |
f7b12b6f THJ |
20541 | return ret; |
20542 | ||
5b92a28a | 20543 | if (tgt_prog) { |
e9eeec58 YS |
20544 | if (subprog == 0) |
20545 | addr = (long) tgt_prog->bpf_func; | |
20546 | else | |
20547 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a | 20548 | } else { |
31bf1dbc VM |
20549 | if (btf_is_module(btf)) { |
20550 | mod = btf_try_get_module(btf); | |
20551 | if (mod) | |
20552 | addr = find_kallsyms_symbol_value(mod, tname); | |
20553 | else | |
20554 | addr = 0; | |
20555 | } else { | |
20556 | addr = kallsyms_lookup_name(tname); | |
20557 | } | |
5b92a28a | 20558 | if (!addr) { |
31bf1dbc | 20559 | module_put(mod); |
efc68158 | 20560 | bpf_log(log, |
5b92a28a AS |
20561 | "The address of function %s cannot be found\n", |
20562 | tname); | |
f7b12b6f | 20563 | return -ENOENT; |
5b92a28a | 20564 | } |
fec56f58 | 20565 | } |
18644cec | 20566 | |
1e6c62a8 AS |
20567 | if (prog->aux->sleepable) { |
20568 | ret = -EINVAL; | |
20569 | switch (prog->type) { | |
20570 | case BPF_PROG_TYPE_TRACING: | |
5b481aca BT |
20571 | |
20572 | /* fentry/fexit/fmod_ret progs can be sleepable if they are | |
1e6c62a8 AS |
20573 | * attached to ALLOW_ERROR_INJECTION and are not in denylist. |
20574 | */ | |
20575 | if (!check_non_sleepable_error_inject(btf_id) && | |
20576 | within_error_injection_list(addr)) | |
20577 | ret = 0; | |
5b481aca BT |
20578 | /* fentry/fexit/fmod_ret progs can also be sleepable if they are |
20579 | * in the fmodret id set with the KF_SLEEPABLE flag. | |
20580 | */ | |
20581 | else { | |
e924e80e AG |
20582 | u32 *flags = btf_kfunc_is_modify_return(btf, btf_id, |
20583 | prog); | |
5b481aca BT |
20584 | |
20585 | if (flags && (*flags & KF_SLEEPABLE)) | |
20586 | ret = 0; | |
20587 | } | |
1e6c62a8 AS |
20588 | break; |
20589 | case BPF_PROG_TYPE_LSM: | |
20590 | /* LSM progs check that they are attached to bpf_lsm_*() funcs. | |
20591 | * Only some of them are sleepable. | |
20592 | */ | |
423f1610 | 20593 | if (bpf_lsm_is_sleepable_hook(btf_id)) |
1e6c62a8 AS |
20594 | ret = 0; |
20595 | break; | |
20596 | default: | |
20597 | break; | |
20598 | } | |
f7b12b6f | 20599 | if (ret) { |
31bf1dbc | 20600 | module_put(mod); |
f7b12b6f THJ |
20601 | bpf_log(log, "%s is not sleepable\n", tname); |
20602 | return ret; | |
20603 | } | |
1e6c62a8 | 20604 | } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) { |
1af9270e | 20605 | if (tgt_prog) { |
31bf1dbc | 20606 | module_put(mod); |
efc68158 | 20607 | bpf_log(log, "can't modify return codes of BPF programs\n"); |
f7b12b6f THJ |
20608 | return -EINVAL; |
20609 | } | |
5b481aca | 20610 | ret = -EINVAL; |
e924e80e | 20611 | if (btf_kfunc_is_modify_return(btf, btf_id, prog) || |
5b481aca BT |
20612 | !check_attach_modify_return(addr, tname)) |
20613 | ret = 0; | |
f7b12b6f | 20614 | if (ret) { |
31bf1dbc | 20615 | module_put(mod); |
f7b12b6f THJ |
20616 | bpf_log(log, "%s() is not modifiable\n", tname); |
20617 | return ret; | |
1af9270e | 20618 | } |
18644cec | 20619 | } |
f7b12b6f THJ |
20620 | |
20621 | break; | |
20622 | } | |
20623 | tgt_info->tgt_addr = addr; | |
20624 | tgt_info->tgt_name = tname; | |
20625 | tgt_info->tgt_type = t; | |
31bf1dbc | 20626 | tgt_info->tgt_mod = mod; |
f7b12b6f THJ |
20627 | return 0; |
20628 | } | |
20629 | ||
35e3815f JO |
20630 | BTF_SET_START(btf_id_deny) |
20631 | BTF_ID_UNUSED | |
20632 | #ifdef CONFIG_SMP | |
20633 | BTF_ID(func, migrate_disable) | |
20634 | BTF_ID(func, migrate_enable) | |
20635 | #endif | |
20636 | #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU | |
20637 | BTF_ID(func, rcu_read_unlock_strict) | |
20638 | #endif | |
c11bd046 Y |
20639 | #if defined(CONFIG_DEBUG_PREEMPT) || defined(CONFIG_TRACE_PREEMPT_TOGGLE) |
20640 | BTF_ID(func, preempt_count_add) | |
20641 | BTF_ID(func, preempt_count_sub) | |
20642 | #endif | |
a0c109dc YS |
20643 | #ifdef CONFIG_PREEMPT_RCU |
20644 | BTF_ID(func, __rcu_read_lock) | |
20645 | BTF_ID(func, __rcu_read_unlock) | |
20646 | #endif | |
35e3815f JO |
20647 | BTF_SET_END(btf_id_deny) |
20648 | ||
700e6f85 JO |
20649 | static bool can_be_sleepable(struct bpf_prog *prog) |
20650 | { | |
20651 | if (prog->type == BPF_PROG_TYPE_TRACING) { | |
20652 | switch (prog->expected_attach_type) { | |
20653 | case BPF_TRACE_FENTRY: | |
20654 | case BPF_TRACE_FEXIT: | |
20655 | case BPF_MODIFY_RETURN: | |
20656 | case BPF_TRACE_ITER: | |
20657 | return true; | |
20658 | default: | |
20659 | return false; | |
20660 | } | |
20661 | } | |
20662 | return prog->type == BPF_PROG_TYPE_LSM || | |
1e12d3ef DV |
20663 | prog->type == BPF_PROG_TYPE_KPROBE /* only for uprobes */ || |
20664 | prog->type == BPF_PROG_TYPE_STRUCT_OPS; | |
700e6f85 JO |
20665 | } |
20666 | ||
f7b12b6f THJ |
20667 | static int check_attach_btf_id(struct bpf_verifier_env *env) |
20668 | { | |
20669 | struct bpf_prog *prog = env->prog; | |
3aac1ead | 20670 | struct bpf_prog *tgt_prog = prog->aux->dst_prog; |
f7b12b6f THJ |
20671 | struct bpf_attach_target_info tgt_info = {}; |
20672 | u32 btf_id = prog->aux->attach_btf_id; | |
20673 | struct bpf_trampoline *tr; | |
20674 | int ret; | |
20675 | u64 key; | |
20676 | ||
79a7f8bd AS |
20677 | if (prog->type == BPF_PROG_TYPE_SYSCALL) { |
20678 | if (prog->aux->sleepable) | |
20679 | /* attach_btf_id checked to be zero already */ | |
20680 | return 0; | |
20681 | verbose(env, "Syscall programs can only be sleepable\n"); | |
20682 | return -EINVAL; | |
20683 | } | |
20684 | ||
700e6f85 | 20685 | if (prog->aux->sleepable && !can_be_sleepable(prog)) { |
1e12d3ef | 20686 | verbose(env, "Only fentry/fexit/fmod_ret, lsm, iter, uprobe, and struct_ops programs can be sleepable\n"); |
f7b12b6f THJ |
20687 | return -EINVAL; |
20688 | } | |
20689 | ||
20690 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) | |
20691 | return check_struct_ops_btf_id(env); | |
20692 | ||
20693 | if (prog->type != BPF_PROG_TYPE_TRACING && | |
20694 | prog->type != BPF_PROG_TYPE_LSM && | |
20695 | prog->type != BPF_PROG_TYPE_EXT) | |
20696 | return 0; | |
20697 | ||
20698 | ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info); | |
20699 | if (ret) | |
fec56f58 | 20700 | return ret; |
f7b12b6f THJ |
20701 | |
20702 | if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) { | |
3aac1ead THJ |
20703 | /* to make freplace equivalent to their targets, they need to |
20704 | * inherit env->ops and expected_attach_type for the rest of the | |
20705 | * verification | |
20706 | */ | |
f7b12b6f THJ |
20707 | env->ops = bpf_verifier_ops[tgt_prog->type]; |
20708 | prog->expected_attach_type = tgt_prog->expected_attach_type; | |
20709 | } | |
20710 | ||
20711 | /* store info about the attachment target that will be used later */ | |
20712 | prog->aux->attach_func_proto = tgt_info.tgt_type; | |
20713 | prog->aux->attach_func_name = tgt_info.tgt_name; | |
31bf1dbc | 20714 | prog->aux->mod = tgt_info.tgt_mod; |
f7b12b6f | 20715 | |
4a1e7c0c THJ |
20716 | if (tgt_prog) { |
20717 | prog->aux->saved_dst_prog_type = tgt_prog->type; | |
20718 | prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type; | |
20719 | } | |
20720 | ||
f7b12b6f THJ |
20721 | if (prog->expected_attach_type == BPF_TRACE_RAW_TP) { |
20722 | prog->aux->attach_btf_trace = true; | |
20723 | return 0; | |
20724 | } else if (prog->expected_attach_type == BPF_TRACE_ITER) { | |
20725 | if (!bpf_iter_prog_supported(prog)) | |
20726 | return -EINVAL; | |
20727 | return 0; | |
20728 | } | |
20729 | ||
20730 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
20731 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
20732 | if (ret < 0) | |
20733 | return ret; | |
35e3815f JO |
20734 | } else if (prog->type == BPF_PROG_TYPE_TRACING && |
20735 | btf_id_set_contains(&btf_id_deny, btf_id)) { | |
20736 | return -EINVAL; | |
38207291 | 20737 | } |
f7b12b6f | 20738 | |
22dc4a0f | 20739 | key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id); |
f7b12b6f THJ |
20740 | tr = bpf_trampoline_get(key, &tgt_info); |
20741 | if (!tr) | |
20742 | return -ENOMEM; | |
20743 | ||
2b5dcb31 LH |
20744 | if (tgt_prog && tgt_prog->aux->tail_call_reachable) |
20745 | tr->flags = BPF_TRAMP_F_TAIL_CALL_CTX; | |
20746 | ||
3aac1ead | 20747 | prog->aux->dst_trampoline = tr; |
f7b12b6f | 20748 | return 0; |
38207291 MKL |
20749 | } |
20750 | ||
76654e67 AM |
20751 | struct btf *bpf_get_btf_vmlinux(void) |
20752 | { | |
20753 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { | |
20754 | mutex_lock(&bpf_verifier_lock); | |
20755 | if (!btf_vmlinux) | |
20756 | btf_vmlinux = btf_parse_vmlinux(); | |
20757 | mutex_unlock(&bpf_verifier_lock); | |
20758 | } | |
20759 | return btf_vmlinux; | |
20760 | } | |
20761 | ||
47a71c1f | 20762 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size) |
51580e79 | 20763 | { |
06ee7115 | 20764 | u64 start_time = ktime_get_ns(); |
58e2af8b | 20765 | struct bpf_verifier_env *env; |
bdcab414 AN |
20766 | int i, len, ret = -EINVAL, err; |
20767 | u32 log_true_size; | |
e2ae4ca2 | 20768 | bool is_priv; |
51580e79 | 20769 | |
eba0c929 AB |
20770 | /* no program is valid */ |
20771 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
20772 | return -EINVAL; | |
20773 | ||
58e2af8b | 20774 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
20775 | * allocate/free it every time bpf_check() is called |
20776 | */ | |
58e2af8b | 20777 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
20778 | if (!env) |
20779 | return -ENOMEM; | |
20780 | ||
407958a0 AN |
20781 | env->bt.env = env; |
20782 | ||
9e4c24e7 | 20783 | len = (*prog)->len; |
fad953ce | 20784 | env->insn_aux_data = |
9e4c24e7 | 20785 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
20786 | ret = -ENOMEM; |
20787 | if (!env->insn_aux_data) | |
20788 | goto err_free_env; | |
9e4c24e7 JK |
20789 | for (i = 0; i < len; i++) |
20790 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 20791 | env->prog = *prog; |
00176a34 | 20792 | env->ops = bpf_verifier_ops[env->prog->type]; |
387544bf | 20793 | env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel); |
2c78ee89 | 20794 | is_priv = bpf_capable(); |
0246e64d | 20795 | |
76654e67 | 20796 | bpf_get_btf_vmlinux(); |
8580ac94 | 20797 | |
cbd35700 | 20798 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
20799 | if (!is_priv) |
20800 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 | 20801 | |
bdcab414 AN |
20802 | /* user could have requested verbose verifier output |
20803 | * and supplied buffer to store the verification trace | |
20804 | */ | |
20805 | ret = bpf_vlog_init(&env->log, attr->log_level, | |
20806 | (char __user *) (unsigned long) attr->log_buf, | |
20807 | attr->log_size); | |
20808 | if (ret) | |
20809 | goto err_unlock; | |
1ad2f583 | 20810 | |
0f55f9ed CL |
20811 | mark_verifier_state_clean(env); |
20812 | ||
8580ac94 AS |
20813 | if (IS_ERR(btf_vmlinux)) { |
20814 | /* Either gcc or pahole or kernel are broken. */ | |
20815 | verbose(env, "in-kernel BTF is malformed\n"); | |
20816 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 20817 | goto skip_full_check; |
8580ac94 AS |
20818 | } |
20819 | ||
1ad2f583 DB |
20820 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
20821 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 20822 | env->strict_alignment = true; |
e9ee9efc DM |
20823 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
20824 | env->strict_alignment = false; | |
cbd35700 | 20825 | |
2c78ee89 | 20826 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
01f810ac | 20827 | env->allow_uninit_stack = bpf_allow_uninit_stack(); |
2c78ee89 AS |
20828 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); |
20829 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
20830 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 20831 | |
10d274e8 AS |
20832 | if (is_priv) |
20833 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
20834 | ||
dc2a4ebc | 20835 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 20836 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
20837 | GFP_USER); |
20838 | ret = -ENOMEM; | |
20839 | if (!env->explored_states) | |
20840 | goto skip_full_check; | |
20841 | ||
aaa619eb KKD |
20842 | ret = check_btf_info_early(env, attr, uattr); |
20843 | if (ret < 0) | |
20844 | goto skip_full_check; | |
20845 | ||
e6ac2450 MKL |
20846 | ret = add_subprog_and_kfunc(env); |
20847 | if (ret < 0) | |
20848 | goto skip_full_check; | |
20849 | ||
d9762e84 | 20850 | ret = check_subprogs(env); |
475fb78f AS |
20851 | if (ret < 0) |
20852 | goto skip_full_check; | |
20853 | ||
c454a46b | 20854 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
20855 | if (ret < 0) |
20856 | goto skip_full_check; | |
20857 | ||
be8704ff AS |
20858 | ret = check_attach_btf_id(env); |
20859 | if (ret) | |
20860 | goto skip_full_check; | |
20861 | ||
4976b718 HL |
20862 | ret = resolve_pseudo_ldimm64(env); |
20863 | if (ret < 0) | |
20864 | goto skip_full_check; | |
20865 | ||
9d03ebc7 | 20866 | if (bpf_prog_is_offloaded(env->prog->aux)) { |
ceb11679 YZ |
20867 | ret = bpf_prog_offload_verifier_prep(env->prog); |
20868 | if (ret) | |
20869 | goto skip_full_check; | |
20870 | } | |
20871 | ||
d9762e84 MKL |
20872 | ret = check_cfg(env); |
20873 | if (ret < 0) | |
20874 | goto skip_full_check; | |
20875 | ||
51c39bb1 AS |
20876 | ret = do_check_subprogs(env); |
20877 | ret = ret ?: do_check_main(env); | |
cbd35700 | 20878 | |
9d03ebc7 | 20879 | if (ret == 0 && bpf_prog_is_offloaded(env->prog->aux)) |
c941ce9c QM |
20880 | ret = bpf_prog_offload_finalize(env); |
20881 | ||
0246e64d | 20882 | skip_full_check: |
51c39bb1 | 20883 | kvfree(env->explored_states); |
0246e64d | 20884 | |
c131187d | 20885 | if (ret == 0) |
9b38c405 | 20886 | ret = check_max_stack_depth(env); |
c131187d | 20887 | |
9b38c405 | 20888 | /* instruction rewrites happen after this point */ |
1ade2371 EZ |
20889 | if (ret == 0) |
20890 | ret = optimize_bpf_loop(env); | |
20891 | ||
e2ae4ca2 JK |
20892 | if (is_priv) { |
20893 | if (ret == 0) | |
20894 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
20895 | if (ret == 0) |
20896 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
20897 | if (ret == 0) |
20898 | ret = opt_remove_nops(env); | |
52875a04 JK |
20899 | } else { |
20900 | if (ret == 0) | |
20901 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
20902 | } |
20903 | ||
9bac3d6d AS |
20904 | if (ret == 0) |
20905 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
20906 | ret = convert_ctx_accesses(env); | |
20907 | ||
e245c5c6 | 20908 | if (ret == 0) |
e6ac5933 | 20909 | ret = do_misc_fixups(env); |
e245c5c6 | 20910 | |
a4b1d3c1 JW |
20911 | /* do 32-bit optimization after insn patching has done so those patched |
20912 | * insns could be handled correctly. | |
20913 | */ | |
9d03ebc7 | 20914 | if (ret == 0 && !bpf_prog_is_offloaded(env->prog->aux)) { |
d6c2308c JW |
20915 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); |
20916 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
20917 | : false; | |
a4b1d3c1 JW |
20918 | } |
20919 | ||
1ea47e01 AS |
20920 | if (ret == 0) |
20921 | ret = fixup_call_args(env); | |
20922 | ||
06ee7115 AS |
20923 | env->verification_time = ktime_get_ns() - start_time; |
20924 | print_verification_stats(env); | |
aba64c7d | 20925 | env->prog->aux->verified_insns = env->insn_processed; |
06ee7115 | 20926 | |
bdcab414 AN |
20927 | /* preserve original error even if log finalization is successful */ |
20928 | err = bpf_vlog_finalize(&env->log, &log_true_size); | |
20929 | if (err) | |
20930 | ret = err; | |
20931 | ||
47a71c1f AN |
20932 | if (uattr_size >= offsetofend(union bpf_attr, log_true_size) && |
20933 | copy_to_bpfptr_offset(uattr, offsetof(union bpf_attr, log_true_size), | |
bdcab414 | 20934 | &log_true_size, sizeof(log_true_size))) { |
47a71c1f AN |
20935 | ret = -EFAULT; |
20936 | goto err_release_maps; | |
20937 | } | |
cbd35700 | 20938 | |
541c3bad AN |
20939 | if (ret) |
20940 | goto err_release_maps; | |
20941 | ||
20942 | if (env->used_map_cnt) { | |
0246e64d | 20943 | /* if program passed verifier, update used_maps in bpf_prog_info */ |
9bac3d6d AS |
20944 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
20945 | sizeof(env->used_maps[0]), | |
20946 | GFP_KERNEL); | |
0246e64d | 20947 | |
9bac3d6d | 20948 | if (!env->prog->aux->used_maps) { |
0246e64d | 20949 | ret = -ENOMEM; |
a2a7d570 | 20950 | goto err_release_maps; |
0246e64d AS |
20951 | } |
20952 | ||
9bac3d6d | 20953 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 20954 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 20955 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
541c3bad AN |
20956 | } |
20957 | if (env->used_btf_cnt) { | |
20958 | /* if program passed verifier, update used_btfs in bpf_prog_aux */ | |
20959 | env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt, | |
20960 | sizeof(env->used_btfs[0]), | |
20961 | GFP_KERNEL); | |
20962 | if (!env->prog->aux->used_btfs) { | |
20963 | ret = -ENOMEM; | |
20964 | goto err_release_maps; | |
20965 | } | |
0246e64d | 20966 | |
541c3bad AN |
20967 | memcpy(env->prog->aux->used_btfs, env->used_btfs, |
20968 | sizeof(env->used_btfs[0]) * env->used_btf_cnt); | |
20969 | env->prog->aux->used_btf_cnt = env->used_btf_cnt; | |
20970 | } | |
20971 | if (env->used_map_cnt || env->used_btf_cnt) { | |
0246e64d AS |
20972 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic |
20973 | * bpf_ld_imm64 instructions | |
20974 | */ | |
20975 | convert_pseudo_ld_imm64(env); | |
20976 | } | |
cbd35700 | 20977 | |
541c3bad | 20978 | adjust_btf_func(env); |
ba64e7d8 | 20979 | |
a2a7d570 | 20980 | err_release_maps: |
9bac3d6d | 20981 | if (!env->prog->aux->used_maps) |
0246e64d | 20982 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 20983 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
20984 | */ |
20985 | release_maps(env); | |
541c3bad AN |
20986 | if (!env->prog->aux->used_btfs) |
20987 | release_btfs(env); | |
03f87c0b THJ |
20988 | |
20989 | /* extension progs temporarily inherit the attach_type of their targets | |
20990 | for verification purposes, so set it back to zero before returning | |
20991 | */ | |
20992 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
20993 | env->prog->expected_attach_type = 0; | |
20994 | ||
9bac3d6d | 20995 | *prog = env->prog; |
3df126f3 | 20996 | err_unlock: |
45a73c17 AS |
20997 | if (!is_priv) |
20998 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
20999 | vfree(env->insn_aux_data); |
21000 | err_free_env: | |
21001 | kfree(env); | |
51580e79 AS |
21002 | return ret; |
21003 | } |