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> |
680ee045 | 29 | #include <net/xdp.h> |
51580e79 | 30 | |
f4ac7e0b JK |
31 | #include "disasm.h" |
32 | ||
00176a34 | 33 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
91cc1a99 | 34 | #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ |
00176a34 JK |
35 | [_id] = & _name ## _verifier_ops, |
36 | #define BPF_MAP_TYPE(_id, _ops) | |
f2e10bff | 37 | #define BPF_LINK_TYPE(_id, _name) |
00176a34 JK |
38 | #include <linux/bpf_types.h> |
39 | #undef BPF_PROG_TYPE | |
40 | #undef BPF_MAP_TYPE | |
f2e10bff | 41 | #undef BPF_LINK_TYPE |
00176a34 JK |
42 | }; |
43 | ||
51580e79 AS |
44 | /* bpf_check() is a static code analyzer that walks eBPF program |
45 | * instruction by instruction and updates register/stack state. | |
46 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
47 | * | |
48 | * The first pass is depth-first-search to check that the program is a DAG. | |
49 | * It rejects the following programs: | |
50 | * - larger than BPF_MAXINSNS insns | |
51 | * - if loop is present (detected via back-edge) | |
52 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
53 | * - out of bounds or malformed jumps | |
54 | * The second pass is all possible path descent from the 1st insn. | |
8fb33b60 | 55 | * Since it's analyzing all paths through the program, the length of the |
eba38a96 | 56 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
57 | * insn is less then 4K, but there are too many branches that change stack/regs. |
58 | * Number of 'branches to be analyzed' is limited to 1k | |
59 | * | |
60 | * On entry to each instruction, each register has a type, and the instruction | |
61 | * changes the types of the registers depending on instruction semantics. | |
62 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
63 | * copied to R1. | |
64 | * | |
65 | * All registers are 64-bit. | |
66 | * R0 - return register | |
67 | * R1-R5 argument passing registers | |
68 | * R6-R9 callee saved registers | |
69 | * R10 - frame pointer read-only | |
70 | * | |
71 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
72 | * and has type PTR_TO_CTX. | |
73 | * | |
74 | * Verifier tracks arithmetic operations on pointers in case: | |
75 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
76 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
77 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
78 | * and 2nd arithmetic instruction is pattern matched to recognize | |
79 | * that it wants to construct a pointer to some element within stack. | |
80 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
81 | * (and -20 constant is saved for further stack bounds checking). | |
82 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
83 | * | |
f1174f77 | 84 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 85 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 86 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
87 | * |
88 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
89 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
90 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
91 | * |
92 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
93 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
94 | * | |
95 | * registers used to pass values to function calls are checked against | |
96 | * function argument constraints. | |
97 | * | |
98 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
99 | * It means that the register type passed to this function must be | |
100 | * PTR_TO_STACK and it will be used inside the function as | |
101 | * 'pointer to map element key' | |
102 | * | |
103 | * For example the argument constraints for bpf_map_lookup_elem(): | |
104 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
105 | * .arg1_type = ARG_CONST_MAP_PTR, | |
106 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
107 | * | |
108 | * ret_type says that this function returns 'pointer to map elem value or null' | |
109 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
110 | * 2nd argument should be a pointer to stack, which will be used inside | |
111 | * the helper function as a pointer to map element key. | |
112 | * | |
113 | * On the kernel side the helper function looks like: | |
114 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
115 | * { | |
116 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
117 | * void *key = (void *) (unsigned long) r2; | |
118 | * void *value; | |
119 | * | |
120 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
121 | * [key, key + map->key_size) bytes are valid and were initialized on | |
122 | * the stack of eBPF program. | |
123 | * } | |
124 | * | |
125 | * Corresponding eBPF program may look like: | |
126 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
127 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
128 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
129 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
130 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
131 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
132 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
133 | * | |
134 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
135 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
136 | * and were initialized prior to this call. | |
137 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
138 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
139 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
8fb33b60 | 140 | * returns either pointer to map value or NULL. |
51580e79 AS |
141 | * |
142 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
143 | * insn, the register holding that pointer in the true branch changes state to | |
144 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
145 | * branch. See check_cond_jmp_op(). | |
146 | * | |
147 | * After the call R0 is set to return type of the function and registers R1-R5 | |
148 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
149 | * |
150 | * The following reference types represent a potential reference to a kernel | |
151 | * resource which, after first being allocated, must be checked and freed by | |
152 | * the BPF program: | |
153 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
154 | * | |
155 | * When the verifier sees a helper call return a reference type, it allocates a | |
156 | * pointer id for the reference and stores it in the current function state. | |
157 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
158 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
159 | * passes through a NULL-check conditional. For the branch wherein the state is | |
160 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
161 | * |
162 | * For each helper function that allocates a reference, such as | |
163 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
164 | * bpf_sk_release(). When a reference type passes into the release function, | |
165 | * the verifier also releases the reference. If any unchecked or unreleased | |
166 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
167 | */ |
168 | ||
17a52670 | 169 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 170 | struct bpf_verifier_stack_elem { |
17a52670 AS |
171 | /* verifer state is 'st' |
172 | * before processing instruction 'insn_idx' | |
173 | * and after processing instruction 'prev_insn_idx' | |
174 | */ | |
58e2af8b | 175 | struct bpf_verifier_state st; |
17a52670 AS |
176 | int insn_idx; |
177 | int prev_insn_idx; | |
58e2af8b | 178 | struct bpf_verifier_stack_elem *next; |
6f8a57cc AN |
179 | /* length of verifier log at the time this state was pushed on stack */ |
180 | u32 log_pos; | |
cbd35700 AS |
181 | }; |
182 | ||
b285fcb7 | 183 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 184 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 185 | |
d2e4c1e6 DB |
186 | #define BPF_MAP_KEY_POISON (1ULL << 63) |
187 | #define BPF_MAP_KEY_SEEN (1ULL << 62) | |
188 | ||
c93552c4 DB |
189 | #define BPF_MAP_PTR_UNPRIV 1UL |
190 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
191 | POISON_POINTER_DELTA)) | |
192 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
193 | ||
bc34dee6 JK |
194 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx); |
195 | static int release_reference(struct bpf_verifier_env *env, int ref_obj_id); | |
6a3cd331 | 196 | static void invalidate_non_owning_refs(struct bpf_verifier_env *env); |
5d92ddc3 | 197 | static bool in_rbtree_lock_required_cb(struct bpf_verifier_env *env); |
6a3cd331 DM |
198 | static int ref_set_non_owning(struct bpf_verifier_env *env, |
199 | struct bpf_reg_state *reg); | |
1cf3bfc6 IL |
200 | static void specialize_kfunc(struct bpf_verifier_env *env, |
201 | u32 func_id, u16 offset, unsigned long *addr); | |
51302c95 | 202 | static bool is_trusted_reg(const struct bpf_reg_state *reg); |
bc34dee6 | 203 | |
c93552c4 DB |
204 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) |
205 | { | |
d2e4c1e6 | 206 | return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON; |
c93552c4 DB |
207 | } |
208 | ||
209 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
210 | { | |
d2e4c1e6 | 211 | return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV; |
c93552c4 DB |
212 | } |
213 | ||
214 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
215 | const struct bpf_map *map, bool unpriv) | |
216 | { | |
217 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
218 | unpriv |= bpf_map_ptr_unpriv(aux); | |
d2e4c1e6 DB |
219 | aux->map_ptr_state = (unsigned long)map | |
220 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
221 | } | |
222 | ||
223 | static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux) | |
224 | { | |
225 | return aux->map_key_state & BPF_MAP_KEY_POISON; | |
226 | } | |
227 | ||
228 | static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux) | |
229 | { | |
230 | return !(aux->map_key_state & BPF_MAP_KEY_SEEN); | |
231 | } | |
232 | ||
233 | static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux) | |
234 | { | |
235 | return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON); | |
236 | } | |
237 | ||
238 | static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state) | |
239 | { | |
240 | bool poisoned = bpf_map_key_poisoned(aux); | |
241 | ||
242 | aux->map_key_state = state | BPF_MAP_KEY_SEEN | | |
243 | (poisoned ? BPF_MAP_KEY_POISON : 0ULL); | |
c93552c4 | 244 | } |
fad73a1a | 245 | |
fde2a388 AN |
246 | static bool bpf_helper_call(const struct bpf_insn *insn) |
247 | { | |
248 | return insn->code == (BPF_JMP | BPF_CALL) && | |
249 | insn->src_reg == 0; | |
250 | } | |
251 | ||
23a2d70c YS |
252 | static bool bpf_pseudo_call(const struct bpf_insn *insn) |
253 | { | |
254 | return insn->code == (BPF_JMP | BPF_CALL) && | |
255 | insn->src_reg == BPF_PSEUDO_CALL; | |
256 | } | |
257 | ||
e6ac2450 MKL |
258 | static bool bpf_pseudo_kfunc_call(const struct bpf_insn *insn) |
259 | { | |
260 | return insn->code == (BPF_JMP | BPF_CALL) && | |
261 | insn->src_reg == BPF_PSEUDO_KFUNC_CALL; | |
262 | } | |
263 | ||
33ff9823 DB |
264 | struct bpf_call_arg_meta { |
265 | struct bpf_map *map_ptr; | |
435faee1 | 266 | bool raw_mode; |
36bbef52 | 267 | bool pkt_access; |
8f14852e | 268 | u8 release_regno; |
435faee1 DB |
269 | int regno; |
270 | int access_size; | |
457f4436 | 271 | int mem_size; |
10060503 | 272 | u64 msize_max_value; |
1b986589 | 273 | int ref_obj_id; |
f8064ab9 | 274 | int dynptr_id; |
3e8ce298 | 275 | int map_uid; |
d83525ca | 276 | int func_id; |
22dc4a0f | 277 | struct btf *btf; |
eaa6bcb7 | 278 | u32 btf_id; |
22dc4a0f | 279 | struct btf *ret_btf; |
eaa6bcb7 | 280 | u32 ret_btf_id; |
69c087ba | 281 | u32 subprogno; |
aa3496ac | 282 | struct btf_field *kptr_field; |
33ff9823 DB |
283 | }; |
284 | ||
d0e1ac22 AN |
285 | struct bpf_kfunc_call_arg_meta { |
286 | /* In parameters */ | |
287 | struct btf *btf; | |
288 | u32 func_id; | |
289 | u32 kfunc_flags; | |
290 | const struct btf_type *func_proto; | |
291 | const char *func_name; | |
292 | /* Out parameters */ | |
293 | u32 ref_obj_id; | |
294 | u8 release_regno; | |
295 | bool r0_rdonly; | |
296 | u32 ret_btf_id; | |
297 | u64 r0_size; | |
298 | u32 subprogno; | |
299 | struct { | |
300 | u64 value; | |
301 | bool found; | |
302 | } arg_constant; | |
4d585f48 | 303 | |
7793fc3b | 304 | /* arg_{btf,btf_id,owning_ref} are used by kfunc-specific handling, |
4d585f48 DM |
305 | * generally to pass info about user-defined local kptr types to later |
306 | * verification logic | |
307 | * bpf_obj_drop | |
308 | * Record the local kptr type to be drop'd | |
309 | * bpf_refcount_acquire (via KF_ARG_PTR_TO_REFCOUNTED_KPTR arg type) | |
7793fc3b DM |
310 | * Record the local kptr type to be refcount_incr'd and use |
311 | * arg_owning_ref to determine whether refcount_acquire should be | |
312 | * fallible | |
4d585f48 DM |
313 | */ |
314 | struct btf *arg_btf; | |
315 | u32 arg_btf_id; | |
7793fc3b | 316 | bool arg_owning_ref; |
4d585f48 | 317 | |
d0e1ac22 AN |
318 | struct { |
319 | struct btf_field *field; | |
320 | } arg_list_head; | |
321 | struct { | |
322 | struct btf_field *field; | |
323 | } arg_rbtree_root; | |
324 | struct { | |
325 | enum bpf_dynptr_type type; | |
326 | u32 id; | |
361f129f | 327 | u32 ref_obj_id; |
d0e1ac22 | 328 | } initialized_dynptr; |
06accc87 AN |
329 | struct { |
330 | u8 spi; | |
331 | u8 frameno; | |
332 | } iter; | |
d0e1ac22 AN |
333 | u64 mem_size; |
334 | }; | |
335 | ||
8580ac94 AS |
336 | struct btf *btf_vmlinux; |
337 | ||
cbd35700 AS |
338 | static DEFINE_MUTEX(bpf_verifier_lock); |
339 | ||
d9762e84 MKL |
340 | static const struct bpf_line_info * |
341 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
342 | { | |
343 | const struct bpf_line_info *linfo; | |
344 | const struct bpf_prog *prog; | |
345 | u32 i, nr_linfo; | |
346 | ||
347 | prog = env->prog; | |
348 | nr_linfo = prog->aux->nr_linfo; | |
349 | ||
350 | if (!nr_linfo || insn_off >= prog->len) | |
351 | return NULL; | |
352 | ||
353 | linfo = prog->aux->linfo; | |
354 | for (i = 1; i < nr_linfo; i++) | |
355 | if (insn_off < linfo[i].insn_off) | |
356 | break; | |
357 | ||
358 | return &linfo[i - 1]; | |
359 | } | |
360 | ||
abe08840 JO |
361 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) |
362 | { | |
77d2e05a | 363 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
364 | va_list args; |
365 | ||
77d2e05a MKL |
366 | if (!bpf_verifier_log_needed(&env->log)) |
367 | return; | |
368 | ||
abe08840 | 369 | va_start(args, fmt); |
77d2e05a | 370 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
371 | va_end(args); |
372 | } | |
cbd35700 | 373 | |
d9762e84 MKL |
374 | static const char *ltrim(const char *s) |
375 | { | |
376 | while (isspace(*s)) | |
377 | s++; | |
378 | ||
379 | return s; | |
380 | } | |
381 | ||
382 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
383 | u32 insn_off, | |
384 | const char *prefix_fmt, ...) | |
385 | { | |
386 | const struct bpf_line_info *linfo; | |
387 | ||
388 | if (!bpf_verifier_log_needed(&env->log)) | |
389 | return; | |
390 | ||
391 | linfo = find_linfo(env, insn_off); | |
392 | if (!linfo || linfo == env->prev_linfo) | |
393 | return; | |
394 | ||
395 | if (prefix_fmt) { | |
396 | va_list args; | |
397 | ||
398 | va_start(args, prefix_fmt); | |
399 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
400 | va_end(args); | |
401 | } | |
402 | ||
403 | verbose(env, "%s\n", | |
404 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
405 | linfo->line_off))); | |
406 | ||
407 | env->prev_linfo = linfo; | |
408 | } | |
409 | ||
bc2591d6 YS |
410 | static void verbose_invalid_scalar(struct bpf_verifier_env *env, |
411 | struct bpf_reg_state *reg, | |
412 | struct tnum *range, const char *ctx, | |
413 | const char *reg_name) | |
414 | { | |
415 | char tn_buf[48]; | |
416 | ||
417 | verbose(env, "At %s the register %s ", ctx, reg_name); | |
418 | if (!tnum_is_unknown(reg->var_off)) { | |
419 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
420 | verbose(env, "has value %s", tn_buf); | |
421 | } else { | |
422 | verbose(env, "has unknown scalar value"); | |
423 | } | |
424 | tnum_strn(tn_buf, sizeof(tn_buf), *range); | |
425 | verbose(env, " should have been in %s\n", tn_buf); | |
426 | } | |
427 | ||
de8f3a83 DB |
428 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
429 | { | |
0c9a7a7e | 430 | type = base_type(type); |
de8f3a83 DB |
431 | return type == PTR_TO_PACKET || |
432 | type == PTR_TO_PACKET_META; | |
433 | } | |
434 | ||
46f8bc92 MKL |
435 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
436 | { | |
437 | return type == PTR_TO_SOCKET || | |
655a51e5 | 438 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
439 | type == PTR_TO_TCP_SOCK || |
440 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
441 | } |
442 | ||
1057d299 AS |
443 | static bool type_may_be_null(u32 type) |
444 | { | |
445 | return type & PTR_MAYBE_NULL; | |
446 | } | |
447 | ||
51302c95 | 448 | static bool reg_not_null(const struct bpf_reg_state *reg) |
cac616db | 449 | { |
51302c95 DV |
450 | enum bpf_reg_type type; |
451 | ||
452 | type = reg->type; | |
1057d299 AS |
453 | if (type_may_be_null(type)) |
454 | return false; | |
455 | ||
456 | type = base_type(type); | |
cac616db JF |
457 | return type == PTR_TO_SOCKET || |
458 | type == PTR_TO_TCP_SOCK || | |
459 | type == PTR_TO_MAP_VALUE || | |
69c087ba | 460 | type == PTR_TO_MAP_KEY || |
d5271c5b | 461 | type == PTR_TO_SOCK_COMMON || |
51302c95 | 462 | (type == PTR_TO_BTF_ID && is_trusted_reg(reg)) || |
d5271c5b | 463 | type == PTR_TO_MEM; |
cac616db JF |
464 | } |
465 | ||
d8939cb0 DM |
466 | static bool type_is_ptr_alloc_obj(u32 type) |
467 | { | |
468 | return base_type(type) == PTR_TO_BTF_ID && type_flag(type) & MEM_ALLOC; | |
469 | } | |
470 | ||
6a3cd331 DM |
471 | static bool type_is_non_owning_ref(u32 type) |
472 | { | |
473 | return type_is_ptr_alloc_obj(type) && type_flag(type) & NON_OWN_REF; | |
474 | } | |
475 | ||
4e814da0 KKD |
476 | static struct btf_record *reg_btf_record(const struct bpf_reg_state *reg) |
477 | { | |
478 | struct btf_record *rec = NULL; | |
479 | struct btf_struct_meta *meta; | |
480 | ||
481 | if (reg->type == PTR_TO_MAP_VALUE) { | |
482 | rec = reg->map_ptr->record; | |
d8939cb0 | 483 | } else if (type_is_ptr_alloc_obj(reg->type)) { |
4e814da0 KKD |
484 | meta = btf_find_struct_meta(reg->btf, reg->btf_id); |
485 | if (meta) | |
486 | rec = meta->record; | |
487 | } | |
488 | return rec; | |
489 | } | |
490 | ||
fde2a388 AN |
491 | static bool subprog_is_global(const struct bpf_verifier_env *env, int subprog) |
492 | { | |
493 | struct bpf_func_info_aux *aux = env->prog->aux->func_info_aux; | |
494 | ||
495 | return aux && aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
496 | } | |
497 | ||
d83525ca AS |
498 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
499 | { | |
4e814da0 | 500 | return btf_record_has_field(reg_btf_record(reg), BPF_SPIN_LOCK); |
cba368c1 MKL |
501 | } |
502 | ||
20b2aff4 HL |
503 | static bool type_is_rdonly_mem(u32 type) |
504 | { | |
505 | return type & MEM_RDONLY; | |
cba368c1 MKL |
506 | } |
507 | ||
64d85290 JS |
508 | static bool is_acquire_function(enum bpf_func_id func_id, |
509 | const struct bpf_map *map) | |
510 | { | |
511 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
512 | ||
513 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
514 | func_id == BPF_FUNC_sk_lookup_udp || | |
457f4436 | 515 | func_id == BPF_FUNC_skc_lookup_tcp || |
c0a5a21c KKD |
516 | func_id == BPF_FUNC_ringbuf_reserve || |
517 | func_id == BPF_FUNC_kptr_xchg) | |
64d85290 JS |
518 | return true; |
519 | ||
520 | if (func_id == BPF_FUNC_map_lookup_elem && | |
521 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
522 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
523 | return true; | |
524 | ||
525 | return false; | |
46f8bc92 MKL |
526 | } |
527 | ||
1b986589 MKL |
528 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
529 | { | |
530 | return func_id == BPF_FUNC_tcp_sock || | |
1df8f55a MKL |
531 | func_id == BPF_FUNC_sk_fullsock || |
532 | func_id == BPF_FUNC_skc_to_tcp_sock || | |
533 | func_id == BPF_FUNC_skc_to_tcp6_sock || | |
534 | func_id == BPF_FUNC_skc_to_udp6_sock || | |
3bc253c2 | 535 | func_id == BPF_FUNC_skc_to_mptcp_sock || |
1df8f55a MKL |
536 | func_id == BPF_FUNC_skc_to_tcp_timewait_sock || |
537 | func_id == BPF_FUNC_skc_to_tcp_request_sock; | |
1b986589 MKL |
538 | } |
539 | ||
88374342 | 540 | static bool is_dynptr_ref_function(enum bpf_func_id func_id) |
b2d8ef19 DM |
541 | { |
542 | return func_id == BPF_FUNC_dynptr_data; | |
543 | } | |
544 | ||
fde2a388 AN |
545 | static bool is_callback_calling_kfunc(u32 btf_id); |
546 | ||
be2ef816 AN |
547 | static bool is_callback_calling_function(enum bpf_func_id func_id) |
548 | { | |
549 | return func_id == BPF_FUNC_for_each_map_elem || | |
550 | func_id == BPF_FUNC_timer_set_callback || | |
551 | func_id == BPF_FUNC_find_vma || | |
552 | func_id == BPF_FUNC_loop || | |
553 | func_id == BPF_FUNC_user_ringbuf_drain; | |
554 | } | |
555 | ||
fde2a388 AN |
556 | static bool is_async_callback_calling_function(enum bpf_func_id func_id) |
557 | { | |
558 | return func_id == BPF_FUNC_timer_set_callback; | |
559 | } | |
560 | ||
9bb00b28 YS |
561 | static bool is_storage_get_function(enum bpf_func_id func_id) |
562 | { | |
563 | return func_id == BPF_FUNC_sk_storage_get || | |
564 | func_id == BPF_FUNC_inode_storage_get || | |
565 | func_id == BPF_FUNC_task_storage_get || | |
566 | func_id == BPF_FUNC_cgrp_storage_get; | |
567 | } | |
568 | ||
b2d8ef19 DM |
569 | static bool helper_multiple_ref_obj_use(enum bpf_func_id func_id, |
570 | const struct bpf_map *map) | |
571 | { | |
572 | int ref_obj_uses = 0; | |
573 | ||
574 | if (is_ptr_cast_function(func_id)) | |
575 | ref_obj_uses++; | |
576 | if (is_acquire_function(func_id, map)) | |
577 | ref_obj_uses++; | |
88374342 | 578 | if (is_dynptr_ref_function(func_id)) |
b2d8ef19 DM |
579 | ref_obj_uses++; |
580 | ||
581 | return ref_obj_uses > 1; | |
582 | } | |
583 | ||
39491867 BJ |
584 | static bool is_cmpxchg_insn(const struct bpf_insn *insn) |
585 | { | |
586 | return BPF_CLASS(insn->code) == BPF_STX && | |
587 | BPF_MODE(insn->code) == BPF_ATOMIC && | |
588 | insn->imm == BPF_CMPXCHG; | |
589 | } | |
590 | ||
c25b2ae1 HL |
591 | /* string representation of 'enum bpf_reg_type' |
592 | * | |
593 | * Note that reg_type_str() can not appear more than once in a single verbose() | |
594 | * statement. | |
595 | */ | |
596 | static const char *reg_type_str(struct bpf_verifier_env *env, | |
597 | enum bpf_reg_type type) | |
598 | { | |
ef66c547 | 599 | char postfix[16] = {0}, prefix[64] = {0}; |
c25b2ae1 HL |
600 | static const char * const str[] = { |
601 | [NOT_INIT] = "?", | |
7df5072c | 602 | [SCALAR_VALUE] = "scalar", |
c25b2ae1 HL |
603 | [PTR_TO_CTX] = "ctx", |
604 | [CONST_PTR_TO_MAP] = "map_ptr", | |
605 | [PTR_TO_MAP_VALUE] = "map_value", | |
606 | [PTR_TO_STACK] = "fp", | |
607 | [PTR_TO_PACKET] = "pkt", | |
608 | [PTR_TO_PACKET_META] = "pkt_meta", | |
609 | [PTR_TO_PACKET_END] = "pkt_end", | |
610 | [PTR_TO_FLOW_KEYS] = "flow_keys", | |
611 | [PTR_TO_SOCKET] = "sock", | |
612 | [PTR_TO_SOCK_COMMON] = "sock_common", | |
613 | [PTR_TO_TCP_SOCK] = "tcp_sock", | |
614 | [PTR_TO_TP_BUFFER] = "tp_buffer", | |
615 | [PTR_TO_XDP_SOCK] = "xdp_sock", | |
616 | [PTR_TO_BTF_ID] = "ptr_", | |
c25b2ae1 | 617 | [PTR_TO_MEM] = "mem", |
20b2aff4 | 618 | [PTR_TO_BUF] = "buf", |
c25b2ae1 HL |
619 | [PTR_TO_FUNC] = "func", |
620 | [PTR_TO_MAP_KEY] = "map_key", | |
27060531 | 621 | [CONST_PTR_TO_DYNPTR] = "dynptr_ptr", |
c25b2ae1 HL |
622 | }; |
623 | ||
624 | if (type & PTR_MAYBE_NULL) { | |
5844101a | 625 | if (base_type(type) == PTR_TO_BTF_ID) |
c25b2ae1 HL |
626 | strncpy(postfix, "or_null_", 16); |
627 | else | |
628 | strncpy(postfix, "_or_null", 16); | |
629 | } | |
630 | ||
9bb00b28 | 631 | snprintf(prefix, sizeof(prefix), "%s%s%s%s%s%s%s", |
ef66c547 DV |
632 | type & MEM_RDONLY ? "rdonly_" : "", |
633 | type & MEM_RINGBUF ? "ringbuf_" : "", | |
634 | type & MEM_USER ? "user_" : "", | |
635 | type & MEM_PERCPU ? "percpu_" : "", | |
9bb00b28 | 636 | type & MEM_RCU ? "rcu_" : "", |
3f00c523 DV |
637 | type & PTR_UNTRUSTED ? "untrusted_" : "", |
638 | type & PTR_TRUSTED ? "trusted_" : "" | |
ef66c547 | 639 | ); |
20b2aff4 | 640 | |
d9439c21 | 641 | snprintf(env->tmp_str_buf, TMP_STR_BUF_LEN, "%s%s%s", |
20b2aff4 | 642 | prefix, str[base_type(type)], postfix); |
d9439c21 | 643 | return env->tmp_str_buf; |
c25b2ae1 | 644 | } |
17a52670 | 645 | |
8efea21d EC |
646 | static char slot_type_char[] = { |
647 | [STACK_INVALID] = '?', | |
648 | [STACK_SPILL] = 'r', | |
649 | [STACK_MISC] = 'm', | |
650 | [STACK_ZERO] = '0', | |
97e03f52 | 651 | [STACK_DYNPTR] = 'd', |
06accc87 | 652 | [STACK_ITER] = 'i', |
8efea21d EC |
653 | }; |
654 | ||
4e92024a AS |
655 | static void print_liveness(struct bpf_verifier_env *env, |
656 | enum bpf_reg_liveness live) | |
657 | { | |
9242b5f5 | 658 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
659 | verbose(env, "_"); |
660 | if (live & REG_LIVE_READ) | |
661 | verbose(env, "r"); | |
662 | if (live & REG_LIVE_WRITTEN) | |
663 | verbose(env, "w"); | |
9242b5f5 AS |
664 | if (live & REG_LIVE_DONE) |
665 | verbose(env, "D"); | |
4e92024a AS |
666 | } |
667 | ||
79168a66 | 668 | static int __get_spi(s32 off) |
97e03f52 JK |
669 | { |
670 | return (-off - 1) / BPF_REG_SIZE; | |
671 | } | |
672 | ||
f5b625e5 KKD |
673 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
674 | const struct bpf_reg_state *reg) | |
675 | { | |
676 | struct bpf_verifier_state *cur = env->cur_state; | |
677 | ||
678 | return cur->frame[reg->frameno]; | |
679 | } | |
680 | ||
97e03f52 JK |
681 | static bool is_spi_bounds_valid(struct bpf_func_state *state, int spi, int nr_slots) |
682 | { | |
f5b625e5 | 683 | int allocated_slots = state->allocated_stack / BPF_REG_SIZE; |
97e03f52 | 684 | |
f5b625e5 KKD |
685 | /* We need to check that slots between [spi - nr_slots + 1, spi] are |
686 | * within [0, allocated_stack). | |
687 | * | |
688 | * Please note that the spi grows downwards. For example, a dynptr | |
689 | * takes the size of two stack slots; the first slot will be at | |
690 | * spi and the second slot will be at spi - 1. | |
691 | */ | |
692 | return spi - nr_slots + 1 >= 0 && spi < allocated_slots; | |
97e03f52 JK |
693 | } |
694 | ||
a461f5ad AN |
695 | static int stack_slot_obj_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
696 | const char *obj_kind, int nr_slots) | |
f4d7e40a | 697 | { |
79168a66 | 698 | int off, spi; |
f4d7e40a | 699 | |
79168a66 | 700 | if (!tnum_is_const(reg->var_off)) { |
a461f5ad | 701 | verbose(env, "%s has to be at a constant offset\n", obj_kind); |
79168a66 KKD |
702 | return -EINVAL; |
703 | } | |
704 | ||
705 | off = reg->off + reg->var_off.value; | |
706 | if (off % BPF_REG_SIZE) { | |
a461f5ad | 707 | verbose(env, "cannot pass in %s at an offset=%d\n", obj_kind, off); |
79168a66 KKD |
708 | return -EINVAL; |
709 | } | |
710 | ||
711 | spi = __get_spi(off); | |
a461f5ad AN |
712 | if (spi + 1 < nr_slots) { |
713 | verbose(env, "cannot pass in %s at an offset=%d\n", obj_kind, off); | |
79168a66 KKD |
714 | return -EINVAL; |
715 | } | |
97e03f52 | 716 | |
a461f5ad | 717 | if (!is_spi_bounds_valid(func(env, reg), spi, nr_slots)) |
f5b625e5 KKD |
718 | return -ERANGE; |
719 | return spi; | |
f4d7e40a AS |
720 | } |
721 | ||
a461f5ad AN |
722 | static int dynptr_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
723 | { | |
724 | return stack_slot_obj_get_spi(env, reg, "dynptr", BPF_DYNPTR_NR_SLOTS); | |
725 | } | |
726 | ||
06accc87 AN |
727 | static int iter_get_spi(struct bpf_verifier_env *env, struct bpf_reg_state *reg, int nr_slots) |
728 | { | |
729 | return stack_slot_obj_get_spi(env, reg, "iter", nr_slots); | |
730 | } | |
731 | ||
b32a5dae | 732 | static const char *btf_type_name(const struct btf *btf, u32 id) |
9e15db66 | 733 | { |
22dc4a0f | 734 | return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off); |
9e15db66 AS |
735 | } |
736 | ||
d54e0f6c AN |
737 | static const char *dynptr_type_str(enum bpf_dynptr_type type) |
738 | { | |
739 | switch (type) { | |
740 | case BPF_DYNPTR_TYPE_LOCAL: | |
741 | return "local"; | |
742 | case BPF_DYNPTR_TYPE_RINGBUF: | |
743 | return "ringbuf"; | |
744 | case BPF_DYNPTR_TYPE_SKB: | |
745 | return "skb"; | |
746 | case BPF_DYNPTR_TYPE_XDP: | |
747 | return "xdp"; | |
748 | case BPF_DYNPTR_TYPE_INVALID: | |
749 | return "<invalid>"; | |
750 | default: | |
751 | WARN_ONCE(1, "unknown dynptr type %d\n", type); | |
752 | return "<unknown>"; | |
753 | } | |
754 | } | |
755 | ||
06accc87 AN |
756 | static const char *iter_type_str(const struct btf *btf, u32 btf_id) |
757 | { | |
758 | if (!btf || btf_id == 0) | |
759 | return "<invalid>"; | |
760 | ||
761 | /* we already validated that type is valid and has conforming name */ | |
b32a5dae | 762 | return btf_type_name(btf, btf_id) + sizeof(ITER_PREFIX) - 1; |
06accc87 AN |
763 | } |
764 | ||
765 | static const char *iter_state_str(enum bpf_iter_state state) | |
766 | { | |
767 | switch (state) { | |
768 | case BPF_ITER_STATE_ACTIVE: | |
769 | return "active"; | |
770 | case BPF_ITER_STATE_DRAINED: | |
771 | return "drained"; | |
772 | case BPF_ITER_STATE_INVALID: | |
773 | return "<invalid>"; | |
774 | default: | |
775 | WARN_ONCE(1, "unknown iter state %d\n", state); | |
776 | return "<unknown>"; | |
777 | } | |
778 | } | |
779 | ||
0f55f9ed CL |
780 | static void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno) |
781 | { | |
782 | env->scratched_regs |= 1U << regno; | |
783 | } | |
784 | ||
785 | static void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi) | |
786 | { | |
343e5375 | 787 | env->scratched_stack_slots |= 1ULL << spi; |
0f55f9ed CL |
788 | } |
789 | ||
790 | static bool reg_scratched(const struct bpf_verifier_env *env, u32 regno) | |
791 | { | |
792 | return (env->scratched_regs >> regno) & 1; | |
793 | } | |
794 | ||
795 | static bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno) | |
796 | { | |
797 | return (env->scratched_stack_slots >> regno) & 1; | |
798 | } | |
799 | ||
800 | static bool verifier_state_scratched(const struct bpf_verifier_env *env) | |
801 | { | |
802 | return env->scratched_regs || env->scratched_stack_slots; | |
803 | } | |
804 | ||
805 | static void mark_verifier_state_clean(struct bpf_verifier_env *env) | |
806 | { | |
807 | env->scratched_regs = 0U; | |
343e5375 | 808 | env->scratched_stack_slots = 0ULL; |
0f55f9ed CL |
809 | } |
810 | ||
811 | /* Used for printing the entire verifier state. */ | |
812 | static void mark_verifier_state_scratched(struct bpf_verifier_env *env) | |
813 | { | |
814 | env->scratched_regs = ~0U; | |
343e5375 | 815 | env->scratched_stack_slots = ~0ULL; |
0f55f9ed CL |
816 | } |
817 | ||
97e03f52 JK |
818 | static enum bpf_dynptr_type arg_to_dynptr_type(enum bpf_arg_type arg_type) |
819 | { | |
820 | switch (arg_type & DYNPTR_TYPE_FLAG_MASK) { | |
821 | case DYNPTR_TYPE_LOCAL: | |
822 | return BPF_DYNPTR_TYPE_LOCAL; | |
bc34dee6 JK |
823 | case DYNPTR_TYPE_RINGBUF: |
824 | return BPF_DYNPTR_TYPE_RINGBUF; | |
b5964b96 JK |
825 | case DYNPTR_TYPE_SKB: |
826 | return BPF_DYNPTR_TYPE_SKB; | |
05421aec JK |
827 | case DYNPTR_TYPE_XDP: |
828 | return BPF_DYNPTR_TYPE_XDP; | |
97e03f52 JK |
829 | default: |
830 | return BPF_DYNPTR_TYPE_INVALID; | |
831 | } | |
832 | } | |
833 | ||
66e3a13e JK |
834 | static enum bpf_type_flag get_dynptr_type_flag(enum bpf_dynptr_type type) |
835 | { | |
836 | switch (type) { | |
837 | case BPF_DYNPTR_TYPE_LOCAL: | |
838 | return DYNPTR_TYPE_LOCAL; | |
839 | case BPF_DYNPTR_TYPE_RINGBUF: | |
840 | return DYNPTR_TYPE_RINGBUF; | |
841 | case BPF_DYNPTR_TYPE_SKB: | |
842 | return DYNPTR_TYPE_SKB; | |
843 | case BPF_DYNPTR_TYPE_XDP: | |
844 | return DYNPTR_TYPE_XDP; | |
845 | default: | |
846 | return 0; | |
847 | } | |
848 | } | |
849 | ||
bc34dee6 JK |
850 | static bool dynptr_type_refcounted(enum bpf_dynptr_type type) |
851 | { | |
852 | return type == BPF_DYNPTR_TYPE_RINGBUF; | |
853 | } | |
854 | ||
27060531 KKD |
855 | static void __mark_dynptr_reg(struct bpf_reg_state *reg, |
856 | enum bpf_dynptr_type type, | |
f8064ab9 | 857 | bool first_slot, int dynptr_id); |
27060531 KKD |
858 | |
859 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, | |
860 | struct bpf_reg_state *reg); | |
861 | ||
f8064ab9 KKD |
862 | static void mark_dynptr_stack_regs(struct bpf_verifier_env *env, |
863 | struct bpf_reg_state *sreg1, | |
27060531 KKD |
864 | struct bpf_reg_state *sreg2, |
865 | enum bpf_dynptr_type type) | |
866 | { | |
f8064ab9 KKD |
867 | int id = ++env->id_gen; |
868 | ||
869 | __mark_dynptr_reg(sreg1, type, true, id); | |
870 | __mark_dynptr_reg(sreg2, type, false, id); | |
27060531 KKD |
871 | } |
872 | ||
f8064ab9 KKD |
873 | static void mark_dynptr_cb_reg(struct bpf_verifier_env *env, |
874 | struct bpf_reg_state *reg, | |
27060531 KKD |
875 | enum bpf_dynptr_type type) |
876 | { | |
f8064ab9 | 877 | __mark_dynptr_reg(reg, type, true, ++env->id_gen); |
27060531 KKD |
878 | } |
879 | ||
ef8fc7a0 KKD |
880 | static int destroy_if_dynptr_stack_slot(struct bpf_verifier_env *env, |
881 | struct bpf_func_state *state, int spi); | |
27060531 | 882 | |
97e03f52 | 883 | static int mark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
361f129f | 884 | enum bpf_arg_type arg_type, int insn_idx, int clone_ref_obj_id) |
97e03f52 JK |
885 | { |
886 | struct bpf_func_state *state = func(env, reg); | |
887 | enum bpf_dynptr_type type; | |
361f129f | 888 | int spi, i, err; |
97e03f52 | 889 | |
79168a66 KKD |
890 | spi = dynptr_get_spi(env, reg); |
891 | if (spi < 0) | |
892 | return spi; | |
97e03f52 | 893 | |
379d4ba8 KKD |
894 | /* We cannot assume both spi and spi - 1 belong to the same dynptr, |
895 | * hence we need to call destroy_if_dynptr_stack_slot twice for both, | |
896 | * to ensure that for the following example: | |
897 | * [d1][d1][d2][d2] | |
898 | * spi 3 2 1 0 | |
899 | * So marking spi = 2 should lead to destruction of both d1 and d2. In | |
900 | * case they do belong to same dynptr, second call won't see slot_type | |
901 | * as STACK_DYNPTR and will simply skip destruction. | |
902 | */ | |
903 | err = destroy_if_dynptr_stack_slot(env, state, spi); | |
904 | if (err) | |
905 | return err; | |
906 | err = destroy_if_dynptr_stack_slot(env, state, spi - 1); | |
907 | if (err) | |
908 | return err; | |
97e03f52 JK |
909 | |
910 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
911 | state->stack[spi].slot_type[i] = STACK_DYNPTR; | |
912 | state->stack[spi - 1].slot_type[i] = STACK_DYNPTR; | |
913 | } | |
914 | ||
915 | type = arg_to_dynptr_type(arg_type); | |
916 | if (type == BPF_DYNPTR_TYPE_INVALID) | |
917 | return -EINVAL; | |
918 | ||
f8064ab9 | 919 | mark_dynptr_stack_regs(env, &state->stack[spi].spilled_ptr, |
27060531 | 920 | &state->stack[spi - 1].spilled_ptr, type); |
97e03f52 | 921 | |
bc34dee6 JK |
922 | if (dynptr_type_refcounted(type)) { |
923 | /* The id is used to track proper releasing */ | |
361f129f JK |
924 | int id; |
925 | ||
926 | if (clone_ref_obj_id) | |
927 | id = clone_ref_obj_id; | |
928 | else | |
929 | id = acquire_reference_state(env, insn_idx); | |
930 | ||
bc34dee6 JK |
931 | if (id < 0) |
932 | return id; | |
933 | ||
27060531 KKD |
934 | state->stack[spi].spilled_ptr.ref_obj_id = id; |
935 | state->stack[spi - 1].spilled_ptr.ref_obj_id = id; | |
bc34dee6 JK |
936 | } |
937 | ||
d6fefa11 KKD |
938 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; |
939 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
940 | ||
97e03f52 JK |
941 | return 0; |
942 | } | |
943 | ||
361f129f | 944 | static void invalidate_dynptr(struct bpf_verifier_env *env, struct bpf_func_state *state, int spi) |
97e03f52 | 945 | { |
361f129f | 946 | int i; |
97e03f52 JK |
947 | |
948 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
949 | state->stack[spi].slot_type[i] = STACK_INVALID; | |
950 | state->stack[spi - 1].slot_type[i] = STACK_INVALID; | |
951 | } | |
952 | ||
27060531 KKD |
953 | __mark_reg_not_init(env, &state->stack[spi].spilled_ptr); |
954 | __mark_reg_not_init(env, &state->stack[spi - 1].spilled_ptr); | |
d6fefa11 KKD |
955 | |
956 | /* Why do we need to set REG_LIVE_WRITTEN for STACK_INVALID slot? | |
957 | * | |
958 | * While we don't allow reading STACK_INVALID, it is still possible to | |
959 | * do <8 byte writes marking some but not all slots as STACK_MISC. Then, | |
960 | * helpers or insns can do partial read of that part without failing, | |
961 | * but check_stack_range_initialized, check_stack_read_var_off, and | |
962 | * check_stack_read_fixed_off will do mark_reg_read for all 8-bytes of | |
963 | * the slot conservatively. Hence we need to prevent those liveness | |
964 | * marking walks. | |
965 | * | |
966 | * This was not a problem before because STACK_INVALID is only set by | |
967 | * default (where the default reg state has its reg->parent as NULL), or | |
968 | * in clean_live_states after REG_LIVE_DONE (at which point | |
969 | * mark_reg_read won't walk reg->parent chain), but not randomly during | |
970 | * verifier state exploration (like we did above). Hence, for our case | |
971 | * parentage chain will still be live (i.e. reg->parent may be | |
972 | * non-NULL), while earlier reg->parent was NULL, so we need | |
973 | * REG_LIVE_WRITTEN to screen off read marker propagation when it is | |
974 | * done later on reads or by mark_dynptr_read as well to unnecessary | |
975 | * mark registers in verifier state. | |
976 | */ | |
977 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
978 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
361f129f JK |
979 | } |
980 | ||
981 | static int unmark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
982 | { | |
983 | struct bpf_func_state *state = func(env, reg); | |
984 | int spi, ref_obj_id, i; | |
985 | ||
986 | spi = dynptr_get_spi(env, reg); | |
987 | if (spi < 0) | |
988 | return spi; | |
989 | ||
990 | if (!dynptr_type_refcounted(state->stack[spi].spilled_ptr.dynptr.type)) { | |
991 | invalidate_dynptr(env, state, spi); | |
992 | return 0; | |
993 | } | |
994 | ||
995 | ref_obj_id = state->stack[spi].spilled_ptr.ref_obj_id; | |
996 | ||
997 | /* If the dynptr has a ref_obj_id, then we need to invalidate | |
998 | * two things: | |
999 | * | |
1000 | * 1) Any dynptrs with a matching ref_obj_id (clones) | |
1001 | * 2) Any slices derived from this dynptr. | |
1002 | */ | |
1003 | ||
1004 | /* Invalidate any slices associated with this dynptr */ | |
1005 | WARN_ON_ONCE(release_reference(env, ref_obj_id)); | |
1006 | ||
1007 | /* Invalidate any dynptr clones */ | |
1008 | for (i = 1; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
1009 | if (state->stack[i].spilled_ptr.ref_obj_id != ref_obj_id) | |
1010 | continue; | |
1011 | ||
1012 | /* it should always be the case that if the ref obj id | |
1013 | * matches then the stack slot also belongs to a | |
1014 | * dynptr | |
1015 | */ | |
1016 | if (state->stack[i].slot_type[0] != STACK_DYNPTR) { | |
1017 | verbose(env, "verifier internal error: misconfigured ref_obj_id\n"); | |
1018 | return -EFAULT; | |
1019 | } | |
1020 | if (state->stack[i].spilled_ptr.dynptr.first_slot) | |
1021 | invalidate_dynptr(env, state, i); | |
1022 | } | |
d6fefa11 | 1023 | |
97e03f52 JK |
1024 | return 0; |
1025 | } | |
1026 | ||
ef8fc7a0 KKD |
1027 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1028 | struct bpf_reg_state *reg); | |
1029 | ||
dbd8d228 KKD |
1030 | static void mark_reg_invalid(const struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
1031 | { | |
1032 | if (!env->allow_ptr_leaks) | |
1033 | __mark_reg_not_init(env, reg); | |
1034 | else | |
1035 | __mark_reg_unknown(env, reg); | |
1036 | } | |
1037 | ||
ef8fc7a0 KKD |
1038 | static int destroy_if_dynptr_stack_slot(struct bpf_verifier_env *env, |
1039 | struct bpf_func_state *state, int spi) | |
97e03f52 | 1040 | { |
f8064ab9 KKD |
1041 | struct bpf_func_state *fstate; |
1042 | struct bpf_reg_state *dreg; | |
1043 | int i, dynptr_id; | |
27060531 | 1044 | |
ef8fc7a0 KKD |
1045 | /* We always ensure that STACK_DYNPTR is never set partially, |
1046 | * hence just checking for slot_type[0] is enough. This is | |
1047 | * different for STACK_SPILL, where it may be only set for | |
1048 | * 1 byte, so code has to use is_spilled_reg. | |
1049 | */ | |
1050 | if (state->stack[spi].slot_type[0] != STACK_DYNPTR) | |
1051 | return 0; | |
97e03f52 | 1052 | |
ef8fc7a0 KKD |
1053 | /* Reposition spi to first slot */ |
1054 | if (!state->stack[spi].spilled_ptr.dynptr.first_slot) | |
1055 | spi = spi + 1; | |
1056 | ||
1057 | if (dynptr_type_refcounted(state->stack[spi].spilled_ptr.dynptr.type)) { | |
1058 | verbose(env, "cannot overwrite referenced dynptr\n"); | |
1059 | return -EINVAL; | |
1060 | } | |
1061 | ||
1062 | mark_stack_slot_scratched(env, spi); | |
1063 | mark_stack_slot_scratched(env, spi - 1); | |
97e03f52 | 1064 | |
ef8fc7a0 | 1065 | /* Writing partially to one dynptr stack slot destroys both. */ |
97e03f52 | 1066 | for (i = 0; i < BPF_REG_SIZE; i++) { |
ef8fc7a0 KKD |
1067 | state->stack[spi].slot_type[i] = STACK_INVALID; |
1068 | state->stack[spi - 1].slot_type[i] = STACK_INVALID; | |
97e03f52 JK |
1069 | } |
1070 | ||
f8064ab9 KKD |
1071 | dynptr_id = state->stack[spi].spilled_ptr.id; |
1072 | /* Invalidate any slices associated with this dynptr */ | |
1073 | bpf_for_each_reg_in_vstate(env->cur_state, fstate, dreg, ({ | |
1074 | /* Dynptr slices are only PTR_TO_MEM_OR_NULL and PTR_TO_MEM */ | |
1075 | if (dreg->type != (PTR_TO_MEM | PTR_MAYBE_NULL) && dreg->type != PTR_TO_MEM) | |
1076 | continue; | |
dbd8d228 KKD |
1077 | if (dreg->dynptr_id == dynptr_id) |
1078 | mark_reg_invalid(env, dreg); | |
f8064ab9 | 1079 | })); |
ef8fc7a0 KKD |
1080 | |
1081 | /* Do not release reference state, we are destroying dynptr on stack, | |
1082 | * not using some helper to release it. Just reset register. | |
1083 | */ | |
1084 | __mark_reg_not_init(env, &state->stack[spi].spilled_ptr); | |
1085 | __mark_reg_not_init(env, &state->stack[spi - 1].spilled_ptr); | |
1086 | ||
1087 | /* Same reason as unmark_stack_slots_dynptr above */ | |
1088 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1089 | state->stack[spi - 1].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1090 | ||
1091 | return 0; | |
1092 | } | |
1093 | ||
7e0dac28 | 1094 | static bool is_dynptr_reg_valid_uninit(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
97e03f52 | 1095 | { |
7e0dac28 JK |
1096 | int spi; |
1097 | ||
27060531 KKD |
1098 | if (reg->type == CONST_PTR_TO_DYNPTR) |
1099 | return false; | |
97e03f52 | 1100 | |
7e0dac28 JK |
1101 | spi = dynptr_get_spi(env, reg); |
1102 | ||
1103 | /* -ERANGE (i.e. spi not falling into allocated stack slots) isn't an | |
1104 | * error because this just means the stack state hasn't been updated yet. | |
1105 | * We will do check_mem_access to check and update stack bounds later. | |
f5b625e5 | 1106 | */ |
7e0dac28 JK |
1107 | if (spi < 0 && spi != -ERANGE) |
1108 | return false; | |
1109 | ||
1110 | /* We don't need to check if the stack slots are marked by previous | |
1111 | * dynptr initializations because we allow overwriting existing unreferenced | |
1112 | * STACK_DYNPTR slots, see mark_stack_slots_dynptr which calls | |
1113 | * destroy_if_dynptr_stack_slot to ensure dynptr objects at the slots we are | |
1114 | * touching are completely destructed before we reinitialize them for a new | |
1115 | * one. For referenced ones, destroy_if_dynptr_stack_slot returns an error early | |
1116 | * instead of delaying it until the end where the user will get "Unreleased | |
379d4ba8 KKD |
1117 | * reference" error. |
1118 | */ | |
97e03f52 JK |
1119 | return true; |
1120 | } | |
1121 | ||
7e0dac28 | 1122 | static bool is_dynptr_reg_valid_init(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
97e03f52 JK |
1123 | { |
1124 | struct bpf_func_state *state = func(env, reg); | |
7e0dac28 | 1125 | int i, spi; |
97e03f52 | 1126 | |
7e0dac28 JK |
1127 | /* This already represents first slot of initialized bpf_dynptr. |
1128 | * | |
1129 | * CONST_PTR_TO_DYNPTR already has fixed and var_off as 0 due to | |
1130 | * check_func_arg_reg_off's logic, so we don't need to check its | |
1131 | * offset and alignment. | |
1132 | */ | |
27060531 KKD |
1133 | if (reg->type == CONST_PTR_TO_DYNPTR) |
1134 | return true; | |
1135 | ||
7e0dac28 | 1136 | spi = dynptr_get_spi(env, reg); |
79168a66 KKD |
1137 | if (spi < 0) |
1138 | return false; | |
f5b625e5 | 1139 | if (!state->stack[spi].spilled_ptr.dynptr.first_slot) |
97e03f52 JK |
1140 | return false; |
1141 | ||
1142 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
1143 | if (state->stack[spi].slot_type[i] != STACK_DYNPTR || | |
1144 | state->stack[spi - 1].slot_type[i] != STACK_DYNPTR) | |
1145 | return false; | |
1146 | } | |
1147 | ||
e9e315b4 RS |
1148 | return true; |
1149 | } | |
1150 | ||
6b75bd3d KKD |
1151 | static bool is_dynptr_type_expected(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
1152 | enum bpf_arg_type arg_type) | |
e9e315b4 RS |
1153 | { |
1154 | struct bpf_func_state *state = func(env, reg); | |
1155 | enum bpf_dynptr_type dynptr_type; | |
27060531 | 1156 | int spi; |
e9e315b4 | 1157 | |
97e03f52 JK |
1158 | /* ARG_PTR_TO_DYNPTR takes any type of dynptr */ |
1159 | if (arg_type == ARG_PTR_TO_DYNPTR) | |
1160 | return true; | |
1161 | ||
e9e315b4 | 1162 | dynptr_type = arg_to_dynptr_type(arg_type); |
27060531 KKD |
1163 | if (reg->type == CONST_PTR_TO_DYNPTR) { |
1164 | return reg->dynptr.type == dynptr_type; | |
1165 | } else { | |
79168a66 KKD |
1166 | spi = dynptr_get_spi(env, reg); |
1167 | if (spi < 0) | |
1168 | return false; | |
27060531 KKD |
1169 | return state->stack[spi].spilled_ptr.dynptr.type == dynptr_type; |
1170 | } | |
97e03f52 JK |
1171 | } |
1172 | ||
06accc87 AN |
1173 | static void __mark_reg_known_zero(struct bpf_reg_state *reg); |
1174 | ||
1175 | static int mark_stack_slots_iter(struct bpf_verifier_env *env, | |
1176 | struct bpf_reg_state *reg, int insn_idx, | |
1177 | struct btf *btf, u32 btf_id, int nr_slots) | |
1178 | { | |
1179 | struct bpf_func_state *state = func(env, reg); | |
1180 | int spi, i, j, id; | |
1181 | ||
1182 | spi = iter_get_spi(env, reg, nr_slots); | |
1183 | if (spi < 0) | |
1184 | return spi; | |
1185 | ||
1186 | id = acquire_reference_state(env, insn_idx); | |
1187 | if (id < 0) | |
1188 | return id; | |
1189 | ||
1190 | for (i = 0; i < nr_slots; i++) { | |
1191 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1192 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1193 | ||
1194 | __mark_reg_known_zero(st); | |
1195 | st->type = PTR_TO_STACK; /* we don't have dedicated reg type */ | |
1196 | st->live |= REG_LIVE_WRITTEN; | |
1197 | st->ref_obj_id = i == 0 ? id : 0; | |
1198 | st->iter.btf = btf; | |
1199 | st->iter.btf_id = btf_id; | |
1200 | st->iter.state = BPF_ITER_STATE_ACTIVE; | |
1201 | st->iter.depth = 0; | |
1202 | ||
1203 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1204 | slot->slot_type[j] = STACK_ITER; | |
1205 | ||
1206 | mark_stack_slot_scratched(env, spi - i); | |
1207 | } | |
1208 | ||
1209 | return 0; | |
1210 | } | |
1211 | ||
1212 | static int unmark_stack_slots_iter(struct bpf_verifier_env *env, | |
1213 | struct bpf_reg_state *reg, int nr_slots) | |
1214 | { | |
1215 | struct bpf_func_state *state = func(env, reg); | |
1216 | int spi, i, j; | |
1217 | ||
1218 | spi = iter_get_spi(env, reg, nr_slots); | |
1219 | if (spi < 0) | |
1220 | return spi; | |
1221 | ||
1222 | for (i = 0; i < nr_slots; i++) { | |
1223 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1224 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1225 | ||
1226 | if (i == 0) | |
1227 | WARN_ON_ONCE(release_reference(env, st->ref_obj_id)); | |
1228 | ||
1229 | __mark_reg_not_init(env, st); | |
1230 | ||
1231 | /* see unmark_stack_slots_dynptr() for why we need to set REG_LIVE_WRITTEN */ | |
1232 | st->live |= REG_LIVE_WRITTEN; | |
1233 | ||
1234 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1235 | slot->slot_type[j] = STACK_INVALID; | |
1236 | ||
1237 | mark_stack_slot_scratched(env, spi - i); | |
1238 | } | |
1239 | ||
1240 | return 0; | |
1241 | } | |
1242 | ||
1243 | static bool is_iter_reg_valid_uninit(struct bpf_verifier_env *env, | |
1244 | struct bpf_reg_state *reg, int nr_slots) | |
1245 | { | |
1246 | struct bpf_func_state *state = func(env, reg); | |
1247 | int spi, i, j; | |
1248 | ||
1249 | /* For -ERANGE (i.e. spi not falling into allocated stack slots), we | |
1250 | * will do check_mem_access to check and update stack bounds later, so | |
1251 | * return true for that case. | |
1252 | */ | |
1253 | spi = iter_get_spi(env, reg, nr_slots); | |
1254 | if (spi == -ERANGE) | |
1255 | return true; | |
1256 | if (spi < 0) | |
1257 | return false; | |
1258 | ||
1259 | for (i = 0; i < nr_slots; i++) { | |
1260 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1261 | ||
1262 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1263 | if (slot->slot_type[j] == STACK_ITER) | |
1264 | return false; | |
1265 | } | |
1266 | ||
1267 | return true; | |
1268 | } | |
1269 | ||
1270 | static bool is_iter_reg_valid_init(struct bpf_verifier_env *env, struct bpf_reg_state *reg, | |
1271 | struct btf *btf, u32 btf_id, int nr_slots) | |
1272 | { | |
1273 | struct bpf_func_state *state = func(env, reg); | |
1274 | int spi, i, j; | |
1275 | ||
1276 | spi = iter_get_spi(env, reg, nr_slots); | |
1277 | if (spi < 0) | |
1278 | return false; | |
1279 | ||
1280 | for (i = 0; i < nr_slots; i++) { | |
1281 | struct bpf_stack_state *slot = &state->stack[spi - i]; | |
1282 | struct bpf_reg_state *st = &slot->spilled_ptr; | |
1283 | ||
1284 | /* only main (first) slot has ref_obj_id set */ | |
1285 | if (i == 0 && !st->ref_obj_id) | |
1286 | return false; | |
1287 | if (i != 0 && st->ref_obj_id) | |
1288 | return false; | |
1289 | if (st->iter.btf != btf || st->iter.btf_id != btf_id) | |
1290 | return false; | |
1291 | ||
1292 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1293 | if (slot->slot_type[j] != STACK_ITER) | |
1294 | return false; | |
1295 | } | |
1296 | ||
1297 | return true; | |
1298 | } | |
1299 | ||
1300 | /* Check if given stack slot is "special": | |
1301 | * - spilled register state (STACK_SPILL); | |
1302 | * - dynptr state (STACK_DYNPTR); | |
1303 | * - iter state (STACK_ITER). | |
1304 | */ | |
1305 | static bool is_stack_slot_special(const struct bpf_stack_state *stack) | |
1306 | { | |
1307 | enum bpf_stack_slot_type type = stack->slot_type[BPF_REG_SIZE - 1]; | |
1308 | ||
1309 | switch (type) { | |
1310 | case STACK_SPILL: | |
1311 | case STACK_DYNPTR: | |
1312 | case STACK_ITER: | |
1313 | return true; | |
1314 | case STACK_INVALID: | |
1315 | case STACK_MISC: | |
1316 | case STACK_ZERO: | |
1317 | return false; | |
1318 | default: | |
1319 | WARN_ONCE(1, "unknown stack slot type %d\n", type); | |
1320 | return true; | |
1321 | } | |
1322 | } | |
1323 | ||
27113c59 MKL |
1324 | /* The reg state of a pointer or a bounded scalar was saved when |
1325 | * it was spilled to the stack. | |
1326 | */ | |
1327 | static bool is_spilled_reg(const struct bpf_stack_state *stack) | |
1328 | { | |
1329 | return stack->slot_type[BPF_REG_SIZE - 1] == STACK_SPILL; | |
1330 | } | |
1331 | ||
407958a0 AN |
1332 | static bool is_spilled_scalar_reg(const struct bpf_stack_state *stack) |
1333 | { | |
1334 | return stack->slot_type[BPF_REG_SIZE - 1] == STACK_SPILL && | |
1335 | stack->spilled_ptr.type == SCALAR_VALUE; | |
1336 | } | |
1337 | ||
354e8f19 MKL |
1338 | static void scrub_spilled_slot(u8 *stype) |
1339 | { | |
1340 | if (*stype != STACK_INVALID) | |
1341 | *stype = STACK_MISC; | |
1342 | } | |
1343 | ||
61bd5218 | 1344 | static void print_verifier_state(struct bpf_verifier_env *env, |
0f55f9ed CL |
1345 | const struct bpf_func_state *state, |
1346 | bool print_all) | |
17a52670 | 1347 | { |
f4d7e40a | 1348 | const struct bpf_reg_state *reg; |
17a52670 AS |
1349 | enum bpf_reg_type t; |
1350 | int i; | |
1351 | ||
f4d7e40a AS |
1352 | if (state->frameno) |
1353 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 1354 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
1355 | reg = &state->regs[i]; |
1356 | t = reg->type; | |
17a52670 AS |
1357 | if (t == NOT_INIT) |
1358 | continue; | |
0f55f9ed CL |
1359 | if (!print_all && !reg_scratched(env, i)) |
1360 | continue; | |
4e92024a AS |
1361 | verbose(env, " R%d", i); |
1362 | print_liveness(env, reg->live); | |
7df5072c | 1363 | verbose(env, "="); |
b5dc0163 AS |
1364 | if (t == SCALAR_VALUE && reg->precise) |
1365 | verbose(env, "P"); | |
f1174f77 EC |
1366 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
1367 | tnum_is_const(reg->var_off)) { | |
1368 | /* reg->off should be 0 for SCALAR_VALUE */ | |
7df5072c | 1369 | verbose(env, "%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
61bd5218 | 1370 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 1371 | } else { |
7df5072c ML |
1372 | const char *sep = ""; |
1373 | ||
1374 | verbose(env, "%s", reg_type_str(env, t)); | |
5844101a | 1375 | if (base_type(t) == PTR_TO_BTF_ID) |
b32a5dae | 1376 | verbose(env, "%s", btf_type_name(reg->btf, reg->btf_id)); |
7df5072c ML |
1377 | verbose(env, "("); |
1378 | /* | |
1379 | * _a stands for append, was shortened to avoid multiline statements below. | |
1380 | * This macro is used to output a comma separated list of attributes. | |
1381 | */ | |
1382 | #define verbose_a(fmt, ...) ({ verbose(env, "%s" fmt, sep, __VA_ARGS__); sep = ","; }) | |
1383 | ||
1384 | if (reg->id) | |
1385 | verbose_a("id=%d", reg->id); | |
a28ace78 | 1386 | if (reg->ref_obj_id) |
7df5072c | 1387 | verbose_a("ref_obj_id=%d", reg->ref_obj_id); |
6a3cd331 DM |
1388 | if (type_is_non_owning_ref(reg->type)) |
1389 | verbose_a("%s", "non_own_ref"); | |
f1174f77 | 1390 | if (t != SCALAR_VALUE) |
7df5072c | 1391 | verbose_a("off=%d", reg->off); |
de8f3a83 | 1392 | if (type_is_pkt_pointer(t)) |
7df5072c | 1393 | verbose_a("r=%d", reg->range); |
c25b2ae1 HL |
1394 | else if (base_type(t) == CONST_PTR_TO_MAP || |
1395 | base_type(t) == PTR_TO_MAP_KEY || | |
1396 | base_type(t) == PTR_TO_MAP_VALUE) | |
7df5072c ML |
1397 | verbose_a("ks=%d,vs=%d", |
1398 | reg->map_ptr->key_size, | |
1399 | reg->map_ptr->value_size); | |
7d1238f2 EC |
1400 | if (tnum_is_const(reg->var_off)) { |
1401 | /* Typically an immediate SCALAR_VALUE, but | |
1402 | * could be a pointer whose offset is too big | |
1403 | * for reg->off | |
1404 | */ | |
7df5072c | 1405 | verbose_a("imm=%llx", reg->var_off.value); |
7d1238f2 EC |
1406 | } else { |
1407 | if (reg->smin_value != reg->umin_value && | |
1408 | reg->smin_value != S64_MIN) | |
7df5072c | 1409 | verbose_a("smin=%lld", (long long)reg->smin_value); |
7d1238f2 EC |
1410 | if (reg->smax_value != reg->umax_value && |
1411 | reg->smax_value != S64_MAX) | |
7df5072c | 1412 | verbose_a("smax=%lld", (long long)reg->smax_value); |
7d1238f2 | 1413 | if (reg->umin_value != 0) |
7df5072c | 1414 | verbose_a("umin=%llu", (unsigned long long)reg->umin_value); |
7d1238f2 | 1415 | if (reg->umax_value != U64_MAX) |
7df5072c | 1416 | verbose_a("umax=%llu", (unsigned long long)reg->umax_value); |
7d1238f2 EC |
1417 | if (!tnum_is_unknown(reg->var_off)) { |
1418 | char tn_buf[48]; | |
f1174f77 | 1419 | |
7d1238f2 | 1420 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
7df5072c | 1421 | verbose_a("var_off=%s", tn_buf); |
7d1238f2 | 1422 | } |
3f50f132 JF |
1423 | if (reg->s32_min_value != reg->smin_value && |
1424 | reg->s32_min_value != S32_MIN) | |
7df5072c | 1425 | verbose_a("s32_min=%d", (int)(reg->s32_min_value)); |
3f50f132 JF |
1426 | if (reg->s32_max_value != reg->smax_value && |
1427 | reg->s32_max_value != S32_MAX) | |
7df5072c | 1428 | verbose_a("s32_max=%d", (int)(reg->s32_max_value)); |
3f50f132 JF |
1429 | if (reg->u32_min_value != reg->umin_value && |
1430 | reg->u32_min_value != U32_MIN) | |
7df5072c | 1431 | verbose_a("u32_min=%d", (int)(reg->u32_min_value)); |
3f50f132 JF |
1432 | if (reg->u32_max_value != reg->umax_value && |
1433 | reg->u32_max_value != U32_MAX) | |
7df5072c | 1434 | verbose_a("u32_max=%d", (int)(reg->u32_max_value)); |
f1174f77 | 1435 | } |
7df5072c ML |
1436 | #undef verbose_a |
1437 | ||
61bd5218 | 1438 | verbose(env, ")"); |
f1174f77 | 1439 | } |
17a52670 | 1440 | } |
638f5b90 | 1441 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
1442 | char types_buf[BPF_REG_SIZE + 1]; |
1443 | bool valid = false; | |
1444 | int j; | |
1445 | ||
1446 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
1447 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
1448 | valid = true; | |
d54e0f6c | 1449 | types_buf[j] = slot_type_char[state->stack[i].slot_type[j]]; |
8efea21d EC |
1450 | } |
1451 | types_buf[BPF_REG_SIZE] = 0; | |
1452 | if (!valid) | |
1453 | continue; | |
0f55f9ed CL |
1454 | if (!print_all && !stack_slot_scratched(env, i)) |
1455 | continue; | |
d54e0f6c AN |
1456 | switch (state->stack[i].slot_type[BPF_REG_SIZE - 1]) { |
1457 | case STACK_SPILL: | |
b5dc0163 AS |
1458 | reg = &state->stack[i].spilled_ptr; |
1459 | t = reg->type; | |
d54e0f6c AN |
1460 | |
1461 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1462 | print_liveness(env, reg->live); | |
7df5072c | 1463 | verbose(env, "=%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
b5dc0163 AS |
1464 | if (t == SCALAR_VALUE && reg->precise) |
1465 | verbose(env, "P"); | |
1466 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
1467 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
d54e0f6c AN |
1468 | break; |
1469 | case STACK_DYNPTR: | |
1470 | i += BPF_DYNPTR_NR_SLOTS - 1; | |
1471 | reg = &state->stack[i].spilled_ptr; | |
1472 | ||
1473 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1474 | print_liveness(env, reg->live); | |
1475 | verbose(env, "=dynptr_%s", dynptr_type_str(reg->dynptr.type)); | |
1476 | if (reg->ref_obj_id) | |
1477 | verbose(env, "(ref_id=%d)", reg->ref_obj_id); | |
1478 | break; | |
06accc87 AN |
1479 | case STACK_ITER: |
1480 | /* only main slot has ref_obj_id set; skip others */ | |
1481 | reg = &state->stack[i].spilled_ptr; | |
1482 | if (!reg->ref_obj_id) | |
1483 | continue; | |
1484 | ||
1485 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1486 | print_liveness(env, reg->live); | |
1487 | verbose(env, "=iter_%s(ref_id=%d,state=%s,depth=%u)", | |
1488 | iter_type_str(reg->iter.btf, reg->iter.btf_id), | |
1489 | reg->ref_obj_id, iter_state_str(reg->iter.state), | |
1490 | reg->iter.depth); | |
1491 | break; | |
d54e0f6c AN |
1492 | case STACK_MISC: |
1493 | case STACK_ZERO: | |
1494 | default: | |
1495 | reg = &state->stack[i].spilled_ptr; | |
1496 | ||
1497 | for (j = 0; j < BPF_REG_SIZE; j++) | |
1498 | types_buf[j] = slot_type_char[state->stack[i].slot_type[j]]; | |
1499 | types_buf[BPF_REG_SIZE] = 0; | |
1500 | ||
1501 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
1502 | print_liveness(env, reg->live); | |
8efea21d | 1503 | verbose(env, "=%s", types_buf); |
d54e0f6c | 1504 | break; |
b5dc0163 | 1505 | } |
17a52670 | 1506 | } |
fd978bf7 JS |
1507 | if (state->acquired_refs && state->refs[0].id) { |
1508 | verbose(env, " refs=%d", state->refs[0].id); | |
1509 | for (i = 1; i < state->acquired_refs; i++) | |
1510 | if (state->refs[i].id) | |
1511 | verbose(env, ",%d", state->refs[i].id); | |
1512 | } | |
bfc6bb74 AS |
1513 | if (state->in_callback_fn) |
1514 | verbose(env, " cb"); | |
1515 | if (state->in_async_callback_fn) | |
1516 | verbose(env, " async_cb"); | |
61bd5218 | 1517 | verbose(env, "\n"); |
0f55f9ed | 1518 | mark_verifier_state_clean(env); |
17a52670 AS |
1519 | } |
1520 | ||
2e576648 CL |
1521 | static inline u32 vlog_alignment(u32 pos) |
1522 | { | |
1523 | return round_up(max(pos + BPF_LOG_MIN_ALIGNMENT / 2, BPF_LOG_ALIGNMENT), | |
1524 | BPF_LOG_MIN_ALIGNMENT) - pos - 1; | |
1525 | } | |
1526 | ||
1527 | static void print_insn_state(struct bpf_verifier_env *env, | |
1528 | const struct bpf_func_state *state) | |
1529 | { | |
12166409 | 1530 | if (env->prev_log_pos && env->prev_log_pos == env->log.end_pos) { |
2e576648 | 1531 | /* remove new line character */ |
12166409 AN |
1532 | bpf_vlog_reset(&env->log, env->prev_log_pos - 1); |
1533 | verbose(env, "%*c;", vlog_alignment(env->prev_insn_print_pos), ' '); | |
2e576648 CL |
1534 | } else { |
1535 | verbose(env, "%d:", env->insn_idx); | |
1536 | } | |
1537 | print_verifier_state(env, state, false); | |
17a52670 AS |
1538 | } |
1539 | ||
c69431aa LB |
1540 | /* copy array src of length n * size bytes to dst. dst is reallocated if it's too |
1541 | * small to hold src. This is different from krealloc since we don't want to preserve | |
1542 | * the contents of dst. | |
1543 | * | |
1544 | * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could | |
1545 | * not be allocated. | |
638f5b90 | 1546 | */ |
c69431aa | 1547 | static void *copy_array(void *dst, const void *src, size_t n, size_t size, gfp_t flags) |
638f5b90 | 1548 | { |
45435d8d KC |
1549 | size_t alloc_bytes; |
1550 | void *orig = dst; | |
c69431aa LB |
1551 | size_t bytes; |
1552 | ||
1553 | if (ZERO_OR_NULL_PTR(src)) | |
1554 | goto out; | |
1555 | ||
1556 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
1557 | return NULL; | |
1558 | ||
45435d8d KC |
1559 | alloc_bytes = max(ksize(orig), kmalloc_size_roundup(bytes)); |
1560 | dst = krealloc(orig, alloc_bytes, flags); | |
1561 | if (!dst) { | |
1562 | kfree(orig); | |
1563 | return NULL; | |
c69431aa LB |
1564 | } |
1565 | ||
1566 | memcpy(dst, src, bytes); | |
1567 | out: | |
1568 | return dst ? dst : ZERO_SIZE_PTR; | |
1569 | } | |
1570 | ||
1571 | /* resize an array from old_n items to new_n items. the array is reallocated if it's too | |
1572 | * small to hold new_n items. new items are zeroed out if the array grows. | |
1573 | * | |
1574 | * Contrary to krealloc_array, does not free arr if new_n is zero. | |
1575 | */ | |
1576 | static void *realloc_array(void *arr, size_t old_n, size_t new_n, size_t size) | |
1577 | { | |
ceb35b66 | 1578 | size_t alloc_size; |
42378a9c KC |
1579 | void *new_arr; |
1580 | ||
c69431aa LB |
1581 | if (!new_n || old_n == new_n) |
1582 | goto out; | |
1583 | ||
ceb35b66 KC |
1584 | alloc_size = kmalloc_size_roundup(size_mul(new_n, size)); |
1585 | new_arr = krealloc(arr, alloc_size, GFP_KERNEL); | |
42378a9c KC |
1586 | if (!new_arr) { |
1587 | kfree(arr); | |
c69431aa | 1588 | return NULL; |
42378a9c KC |
1589 | } |
1590 | arr = new_arr; | |
c69431aa LB |
1591 | |
1592 | if (new_n > old_n) | |
1593 | memset(arr + old_n * size, 0, (new_n - old_n) * size); | |
1594 | ||
1595 | out: | |
1596 | return arr ? arr : ZERO_SIZE_PTR; | |
1597 | } | |
1598 | ||
1599 | static int copy_reference_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1600 | { | |
1601 | dst->refs = copy_array(dst->refs, src->refs, src->acquired_refs, | |
1602 | sizeof(struct bpf_reference_state), GFP_KERNEL); | |
1603 | if (!dst->refs) | |
1604 | return -ENOMEM; | |
1605 | ||
1606 | dst->acquired_refs = src->acquired_refs; | |
1607 | return 0; | |
1608 | } | |
1609 | ||
1610 | static int copy_stack_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1611 | { | |
1612 | size_t n = src->allocated_stack / BPF_REG_SIZE; | |
1613 | ||
1614 | dst->stack = copy_array(dst->stack, src->stack, n, sizeof(struct bpf_stack_state), | |
1615 | GFP_KERNEL); | |
1616 | if (!dst->stack) | |
1617 | return -ENOMEM; | |
1618 | ||
1619 | dst->allocated_stack = src->allocated_stack; | |
1620 | return 0; | |
1621 | } | |
1622 | ||
1623 | static int resize_reference_state(struct bpf_func_state *state, size_t n) | |
1624 | { | |
1625 | state->refs = realloc_array(state->refs, state->acquired_refs, n, | |
1626 | sizeof(struct bpf_reference_state)); | |
1627 | if (!state->refs) | |
1628 | return -ENOMEM; | |
1629 | ||
1630 | state->acquired_refs = n; | |
1631 | return 0; | |
1632 | } | |
1633 | ||
1634 | static int grow_stack_state(struct bpf_func_state *state, int size) | |
1635 | { | |
1636 | size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE; | |
1637 | ||
1638 | if (old_n >= n) | |
1639 | return 0; | |
1640 | ||
1641 | state->stack = realloc_array(state->stack, old_n, n, sizeof(struct bpf_stack_state)); | |
1642 | if (!state->stack) | |
1643 | return -ENOMEM; | |
1644 | ||
1645 | state->allocated_stack = size; | |
1646 | return 0; | |
fd978bf7 JS |
1647 | } |
1648 | ||
1649 | /* Acquire a pointer id from the env and update the state->refs to include | |
1650 | * this new pointer reference. | |
1651 | * On success, returns a valid pointer id to associate with the register | |
1652 | * On failure, returns a negative errno. | |
638f5b90 | 1653 | */ |
fd978bf7 | 1654 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 1655 | { |
fd978bf7 JS |
1656 | struct bpf_func_state *state = cur_func(env); |
1657 | int new_ofs = state->acquired_refs; | |
1658 | int id, err; | |
1659 | ||
c69431aa | 1660 | err = resize_reference_state(state, state->acquired_refs + 1); |
fd978bf7 JS |
1661 | if (err) |
1662 | return err; | |
1663 | id = ++env->id_gen; | |
1664 | state->refs[new_ofs].id = id; | |
1665 | state->refs[new_ofs].insn_idx = insn_idx; | |
9d9d00ac | 1666 | state->refs[new_ofs].callback_ref = state->in_callback_fn ? state->frameno : 0; |
638f5b90 | 1667 | |
fd978bf7 JS |
1668 | return id; |
1669 | } | |
1670 | ||
1671 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 1672 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
1673 | { |
1674 | int i, last_idx; | |
1675 | ||
fd978bf7 JS |
1676 | last_idx = state->acquired_refs - 1; |
1677 | for (i = 0; i < state->acquired_refs; i++) { | |
1678 | if (state->refs[i].id == ptr_id) { | |
9d9d00ac KKD |
1679 | /* Cannot release caller references in callbacks */ |
1680 | if (state->in_callback_fn && state->refs[i].callback_ref != state->frameno) | |
1681 | return -EINVAL; | |
fd978bf7 JS |
1682 | if (last_idx && i != last_idx) |
1683 | memcpy(&state->refs[i], &state->refs[last_idx], | |
1684 | sizeof(*state->refs)); | |
1685 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
1686 | state->acquired_refs--; | |
638f5b90 | 1687 | return 0; |
638f5b90 | 1688 | } |
638f5b90 | 1689 | } |
46f8bc92 | 1690 | return -EINVAL; |
fd978bf7 JS |
1691 | } |
1692 | ||
f4d7e40a AS |
1693 | static void free_func_state(struct bpf_func_state *state) |
1694 | { | |
5896351e AS |
1695 | if (!state) |
1696 | return; | |
fd978bf7 | 1697 | kfree(state->refs); |
f4d7e40a AS |
1698 | kfree(state->stack); |
1699 | kfree(state); | |
1700 | } | |
1701 | ||
b5dc0163 AS |
1702 | static void clear_jmp_history(struct bpf_verifier_state *state) |
1703 | { | |
1704 | kfree(state->jmp_history); | |
1705 | state->jmp_history = NULL; | |
1706 | state->jmp_history_cnt = 0; | |
1707 | } | |
1708 | ||
1969db47 AS |
1709 | static void free_verifier_state(struct bpf_verifier_state *state, |
1710 | bool free_self) | |
638f5b90 | 1711 | { |
f4d7e40a AS |
1712 | int i; |
1713 | ||
1714 | for (i = 0; i <= state->curframe; i++) { | |
1715 | free_func_state(state->frame[i]); | |
1716 | state->frame[i] = NULL; | |
1717 | } | |
b5dc0163 | 1718 | clear_jmp_history(state); |
1969db47 AS |
1719 | if (free_self) |
1720 | kfree(state); | |
638f5b90 AS |
1721 | } |
1722 | ||
1723 | /* copy verifier state from src to dst growing dst stack space | |
1724 | * when necessary to accommodate larger src stack | |
1725 | */ | |
f4d7e40a AS |
1726 | static int copy_func_state(struct bpf_func_state *dst, |
1727 | const struct bpf_func_state *src) | |
638f5b90 AS |
1728 | { |
1729 | int err; | |
1730 | ||
fd978bf7 JS |
1731 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); |
1732 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
1733 | if (err) |
1734 | return err; | |
638f5b90 AS |
1735 | return copy_stack_state(dst, src); |
1736 | } | |
1737 | ||
f4d7e40a AS |
1738 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
1739 | const struct bpf_verifier_state *src) | |
1740 | { | |
1741 | struct bpf_func_state *dst; | |
1742 | int i, err; | |
1743 | ||
06ab6a50 LB |
1744 | dst_state->jmp_history = copy_array(dst_state->jmp_history, src->jmp_history, |
1745 | src->jmp_history_cnt, sizeof(struct bpf_idx_pair), | |
1746 | GFP_USER); | |
1747 | if (!dst_state->jmp_history) | |
1748 | return -ENOMEM; | |
b5dc0163 AS |
1749 | dst_state->jmp_history_cnt = src->jmp_history_cnt; |
1750 | ||
f4d7e40a AS |
1751 | /* if dst has more stack frames then src frame, free them */ |
1752 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
1753 | free_func_state(dst_state->frame[i]); | |
1754 | dst_state->frame[i] = NULL; | |
1755 | } | |
979d63d5 | 1756 | dst_state->speculative = src->speculative; |
9bb00b28 | 1757 | dst_state->active_rcu_lock = src->active_rcu_lock; |
f4d7e40a | 1758 | dst_state->curframe = src->curframe; |
d0d78c1d KKD |
1759 | dst_state->active_lock.ptr = src->active_lock.ptr; |
1760 | dst_state->active_lock.id = src->active_lock.id; | |
2589726d AS |
1761 | dst_state->branches = src->branches; |
1762 | dst_state->parent = src->parent; | |
b5dc0163 AS |
1763 | dst_state->first_insn_idx = src->first_insn_idx; |
1764 | dst_state->last_insn_idx = src->last_insn_idx; | |
f4d7e40a AS |
1765 | for (i = 0; i <= src->curframe; i++) { |
1766 | dst = dst_state->frame[i]; | |
1767 | if (!dst) { | |
1768 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
1769 | if (!dst) | |
1770 | return -ENOMEM; | |
1771 | dst_state->frame[i] = dst; | |
1772 | } | |
1773 | err = copy_func_state(dst, src->frame[i]); | |
1774 | if (err) | |
1775 | return err; | |
1776 | } | |
1777 | return 0; | |
1778 | } | |
1779 | ||
2589726d AS |
1780 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
1781 | { | |
1782 | while (st) { | |
1783 | u32 br = --st->branches; | |
1784 | ||
1785 | /* WARN_ON(br > 1) technically makes sense here, | |
1786 | * but see comment in push_stack(), hence: | |
1787 | */ | |
1788 | WARN_ONCE((int)br < 0, | |
1789 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
1790 | br); | |
1791 | if (br) | |
1792 | break; | |
1793 | st = st->parent; | |
1794 | } | |
1795 | } | |
1796 | ||
638f5b90 | 1797 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 1798 | int *insn_idx, bool pop_log) |
638f5b90 AS |
1799 | { |
1800 | struct bpf_verifier_state *cur = env->cur_state; | |
1801 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
1802 | int err; | |
17a52670 AS |
1803 | |
1804 | if (env->head == NULL) | |
638f5b90 | 1805 | return -ENOENT; |
17a52670 | 1806 | |
638f5b90 AS |
1807 | if (cur) { |
1808 | err = copy_verifier_state(cur, &head->st); | |
1809 | if (err) | |
1810 | return err; | |
1811 | } | |
6f8a57cc AN |
1812 | if (pop_log) |
1813 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
1814 | if (insn_idx) |
1815 | *insn_idx = head->insn_idx; | |
17a52670 | 1816 | if (prev_insn_idx) |
638f5b90 AS |
1817 | *prev_insn_idx = head->prev_insn_idx; |
1818 | elem = head->next; | |
1969db47 | 1819 | free_verifier_state(&head->st, false); |
638f5b90 | 1820 | kfree(head); |
17a52670 AS |
1821 | env->head = elem; |
1822 | env->stack_size--; | |
638f5b90 | 1823 | return 0; |
17a52670 AS |
1824 | } |
1825 | ||
58e2af8b | 1826 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
1827 | int insn_idx, int prev_insn_idx, |
1828 | bool speculative) | |
17a52670 | 1829 | { |
638f5b90 | 1830 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 1831 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 1832 | int err; |
17a52670 | 1833 | |
638f5b90 | 1834 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
1835 | if (!elem) |
1836 | goto err; | |
1837 | ||
17a52670 AS |
1838 | elem->insn_idx = insn_idx; |
1839 | elem->prev_insn_idx = prev_insn_idx; | |
1840 | elem->next = env->head; | |
12166409 | 1841 | elem->log_pos = env->log.end_pos; |
17a52670 AS |
1842 | env->head = elem; |
1843 | env->stack_size++; | |
1969db47 AS |
1844 | err = copy_verifier_state(&elem->st, cur); |
1845 | if (err) | |
1846 | goto err; | |
979d63d5 | 1847 | elem->st.speculative |= speculative; |
b285fcb7 AS |
1848 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
1849 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
1850 | env->stack_size); | |
17a52670 AS |
1851 | goto err; |
1852 | } | |
2589726d AS |
1853 | if (elem->st.parent) { |
1854 | ++elem->st.parent->branches; | |
1855 | /* WARN_ON(branches > 2) technically makes sense here, | |
1856 | * but | |
1857 | * 1. speculative states will bump 'branches' for non-branch | |
1858 | * instructions | |
1859 | * 2. is_state_visited() heuristics may decide not to create | |
1860 | * a new state for a sequence of branches and all such current | |
1861 | * and cloned states will be pointing to a single parent state | |
1862 | * which might have large 'branches' count. | |
1863 | */ | |
1864 | } | |
17a52670 AS |
1865 | return &elem->st; |
1866 | err: | |
5896351e AS |
1867 | free_verifier_state(env->cur_state, true); |
1868 | env->cur_state = NULL; | |
17a52670 | 1869 | /* pop all elements and return */ |
6f8a57cc | 1870 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
1871 | return NULL; |
1872 | } | |
1873 | ||
1874 | #define CALLER_SAVED_REGS 6 | |
1875 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
1876 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
1877 | }; | |
1878 | ||
e688c3db AS |
1879 | /* This helper doesn't clear reg->id */ |
1880 | static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
b03c9f9f | 1881 | { |
b03c9f9f EC |
1882 | reg->var_off = tnum_const(imm); |
1883 | reg->smin_value = (s64)imm; | |
1884 | reg->smax_value = (s64)imm; | |
1885 | reg->umin_value = imm; | |
1886 | reg->umax_value = imm; | |
3f50f132 JF |
1887 | |
1888 | reg->s32_min_value = (s32)imm; | |
1889 | reg->s32_max_value = (s32)imm; | |
1890 | reg->u32_min_value = (u32)imm; | |
1891 | reg->u32_max_value = (u32)imm; | |
1892 | } | |
1893 | ||
e688c3db AS |
1894 | /* Mark the unknown part of a register (variable offset or scalar value) as |
1895 | * known to have the value @imm. | |
1896 | */ | |
1897 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
1898 | { | |
a73bf9f2 | 1899 | /* Clear off and union(map_ptr, range) */ |
e688c3db AS |
1900 | memset(((u8 *)reg) + sizeof(reg->type), 0, |
1901 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
a73bf9f2 AN |
1902 | reg->id = 0; |
1903 | reg->ref_obj_id = 0; | |
e688c3db AS |
1904 | ___mark_reg_known(reg, imm); |
1905 | } | |
1906 | ||
3f50f132 JF |
1907 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) |
1908 | { | |
1909 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
1910 | reg->s32_min_value = (s32)imm; | |
1911 | reg->s32_max_value = (s32)imm; | |
1912 | reg->u32_min_value = (u32)imm; | |
1913 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
1914 | } |
1915 | ||
f1174f77 EC |
1916 | /* Mark the 'variable offset' part of a register as zero. This should be |
1917 | * used only on registers holding a pointer type. | |
1918 | */ | |
1919 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 1920 | { |
b03c9f9f | 1921 | __mark_reg_known(reg, 0); |
f1174f77 | 1922 | } |
a9789ef9 | 1923 | |
cc2b14d5 AS |
1924 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
1925 | { | |
1926 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
1927 | reg->type = SCALAR_VALUE; |
1928 | } | |
1929 | ||
61bd5218 JK |
1930 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
1931 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1932 | { |
1933 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1934 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
1935 | /* Something bad happened, let's kill all regs */ |
1936 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 1937 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1938 | return; |
1939 | } | |
1940 | __mark_reg_known_zero(regs + regno); | |
1941 | } | |
1942 | ||
27060531 | 1943 | static void __mark_dynptr_reg(struct bpf_reg_state *reg, enum bpf_dynptr_type type, |
f8064ab9 | 1944 | bool first_slot, int dynptr_id) |
27060531 KKD |
1945 | { |
1946 | /* reg->type has no meaning for STACK_DYNPTR, but when we set reg for | |
1947 | * callback arguments, it does need to be CONST_PTR_TO_DYNPTR, so simply | |
1948 | * set it unconditionally as it is ignored for STACK_DYNPTR anyway. | |
1949 | */ | |
1950 | __mark_reg_known_zero(reg); | |
1951 | reg->type = CONST_PTR_TO_DYNPTR; | |
f8064ab9 KKD |
1952 | /* Give each dynptr a unique id to uniquely associate slices to it. */ |
1953 | reg->id = dynptr_id; | |
27060531 KKD |
1954 | reg->dynptr.type = type; |
1955 | reg->dynptr.first_slot = first_slot; | |
1956 | } | |
1957 | ||
4ddb7416 DB |
1958 | static void mark_ptr_not_null_reg(struct bpf_reg_state *reg) |
1959 | { | |
c25b2ae1 | 1960 | if (base_type(reg->type) == PTR_TO_MAP_VALUE) { |
4ddb7416 DB |
1961 | const struct bpf_map *map = reg->map_ptr; |
1962 | ||
1963 | if (map->inner_map_meta) { | |
1964 | reg->type = CONST_PTR_TO_MAP; | |
1965 | reg->map_ptr = map->inner_map_meta; | |
3e8ce298 AS |
1966 | /* transfer reg's id which is unique for every map_lookup_elem |
1967 | * as UID of the inner map. | |
1968 | */ | |
db559117 | 1969 | if (btf_record_has_field(map->inner_map_meta->record, BPF_TIMER)) |
34d11a44 | 1970 | reg->map_uid = reg->id; |
4ddb7416 DB |
1971 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { |
1972 | reg->type = PTR_TO_XDP_SOCK; | |
1973 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || | |
1974 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
1975 | reg->type = PTR_TO_SOCKET; | |
1976 | } else { | |
1977 | reg->type = PTR_TO_MAP_VALUE; | |
1978 | } | |
c25b2ae1 | 1979 | return; |
4ddb7416 | 1980 | } |
c25b2ae1 HL |
1981 | |
1982 | reg->type &= ~PTR_MAYBE_NULL; | |
4ddb7416 DB |
1983 | } |
1984 | ||
5d92ddc3 DM |
1985 | static void mark_reg_graph_node(struct bpf_reg_state *regs, u32 regno, |
1986 | struct btf_field_graph_root *ds_head) | |
1987 | { | |
1988 | __mark_reg_known_zero(®s[regno]); | |
1989 | regs[regno].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
1990 | regs[regno].btf = ds_head->btf; | |
1991 | regs[regno].btf_id = ds_head->value_btf_id; | |
1992 | regs[regno].off = ds_head->node_offset; | |
1993 | } | |
1994 | ||
de8f3a83 DB |
1995 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
1996 | { | |
1997 | return type_is_pkt_pointer(reg->type); | |
1998 | } | |
1999 | ||
2000 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
2001 | { | |
2002 | return reg_is_pkt_pointer(reg) || | |
2003 | reg->type == PTR_TO_PACKET_END; | |
2004 | } | |
2005 | ||
66e3a13e JK |
2006 | static bool reg_is_dynptr_slice_pkt(const struct bpf_reg_state *reg) |
2007 | { | |
2008 | return base_type(reg->type) == PTR_TO_MEM && | |
2009 | (reg->type & DYNPTR_TYPE_SKB || reg->type & DYNPTR_TYPE_XDP); | |
2010 | } | |
2011 | ||
de8f3a83 DB |
2012 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ |
2013 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
2014 | enum bpf_reg_type which) | |
2015 | { | |
2016 | /* The register can already have a range from prior markings. | |
2017 | * This is fine as long as it hasn't been advanced from its | |
2018 | * origin. | |
2019 | */ | |
2020 | return reg->type == which && | |
2021 | reg->id == 0 && | |
2022 | reg->off == 0 && | |
2023 | tnum_equals_const(reg->var_off, 0); | |
2024 | } | |
2025 | ||
3f50f132 JF |
2026 | /* Reset the min/max bounds of a register */ |
2027 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
2028 | { | |
2029 | reg->smin_value = S64_MIN; | |
2030 | reg->smax_value = S64_MAX; | |
2031 | reg->umin_value = 0; | |
2032 | reg->umax_value = U64_MAX; | |
2033 | ||
2034 | reg->s32_min_value = S32_MIN; | |
2035 | reg->s32_max_value = S32_MAX; | |
2036 | reg->u32_min_value = 0; | |
2037 | reg->u32_max_value = U32_MAX; | |
2038 | } | |
2039 | ||
2040 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
2041 | { | |
2042 | reg->smin_value = S64_MIN; | |
2043 | reg->smax_value = S64_MAX; | |
2044 | reg->umin_value = 0; | |
2045 | reg->umax_value = U64_MAX; | |
2046 | } | |
2047 | ||
2048 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
2049 | { | |
2050 | reg->s32_min_value = S32_MIN; | |
2051 | reg->s32_max_value = S32_MAX; | |
2052 | reg->u32_min_value = 0; | |
2053 | reg->u32_max_value = U32_MAX; | |
2054 | } | |
2055 | ||
2056 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
2057 | { | |
2058 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
2059 | ||
2060 | /* min signed is max(sign bit) | min(other bits) */ | |
2061 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
2062 | var32_off.value | (var32_off.mask & S32_MIN)); | |
2063 | /* max signed is min(sign bit) | max(other bits) */ | |
2064 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
2065 | var32_off.value | (var32_off.mask & S32_MAX)); | |
2066 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
2067 | reg->u32_max_value = min(reg->u32_max_value, | |
2068 | (u32)(var32_off.value | var32_off.mask)); | |
2069 | } | |
2070 | ||
2071 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
2072 | { |
2073 | /* min signed is max(sign bit) | min(other bits) */ | |
2074 | reg->smin_value = max_t(s64, reg->smin_value, | |
2075 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
2076 | /* max signed is min(sign bit) | max(other bits) */ | |
2077 | reg->smax_value = min_t(s64, reg->smax_value, | |
2078 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
2079 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
2080 | reg->umax_value = min(reg->umax_value, | |
2081 | reg->var_off.value | reg->var_off.mask); | |
2082 | } | |
2083 | ||
3f50f132 JF |
2084 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
2085 | { | |
2086 | __update_reg32_bounds(reg); | |
2087 | __update_reg64_bounds(reg); | |
2088 | } | |
2089 | ||
b03c9f9f | 2090 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
2091 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
2092 | { | |
2093 | /* Learn sign from signed bounds. | |
2094 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
2095 | * are the same, so combine. This works even in the negative case, e.g. | |
2096 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
2097 | */ | |
2098 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
2099 | reg->s32_min_value = reg->u32_min_value = | |
2100 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
2101 | reg->s32_max_value = reg->u32_max_value = | |
2102 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
2103 | return; | |
2104 | } | |
2105 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
2106 | * boundary, so we must be careful. | |
2107 | */ | |
2108 | if ((s32)reg->u32_max_value >= 0) { | |
2109 | /* Positive. We can't learn anything from the smin, but smax | |
2110 | * is positive, hence safe. | |
2111 | */ | |
2112 | reg->s32_min_value = reg->u32_min_value; | |
2113 | reg->s32_max_value = reg->u32_max_value = | |
2114 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
2115 | } else if ((s32)reg->u32_min_value < 0) { | |
2116 | /* Negative. We can't learn anything from the smax, but smin | |
2117 | * is negative, hence safe. | |
2118 | */ | |
2119 | reg->s32_min_value = reg->u32_min_value = | |
2120 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
2121 | reg->s32_max_value = reg->u32_max_value; | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
2126 | { |
2127 | /* Learn sign from signed bounds. | |
2128 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
2129 | * are the same, so combine. This works even in the negative case, e.g. | |
2130 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
2131 | */ | |
2132 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
2133 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
2134 | reg->umin_value); | |
2135 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
2136 | reg->umax_value); | |
2137 | return; | |
2138 | } | |
2139 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
2140 | * boundary, so we must be careful. | |
2141 | */ | |
2142 | if ((s64)reg->umax_value >= 0) { | |
2143 | /* Positive. We can't learn anything from the smin, but smax | |
2144 | * is positive, hence safe. | |
2145 | */ | |
2146 | reg->smin_value = reg->umin_value; | |
2147 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
2148 | reg->umax_value); | |
2149 | } else if ((s64)reg->umin_value < 0) { | |
2150 | /* Negative. We can't learn anything from the smax, but smin | |
2151 | * is negative, hence safe. | |
2152 | */ | |
2153 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
2154 | reg->umin_value); | |
2155 | reg->smax_value = reg->umax_value; | |
2156 | } | |
2157 | } | |
2158 | ||
3f50f132 JF |
2159 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
2160 | { | |
2161 | __reg32_deduce_bounds(reg); | |
2162 | __reg64_deduce_bounds(reg); | |
2163 | } | |
2164 | ||
b03c9f9f EC |
2165 | /* Attempts to improve var_off based on unsigned min/max information */ |
2166 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
2167 | { | |
3f50f132 JF |
2168 | struct tnum var64_off = tnum_intersect(reg->var_off, |
2169 | tnum_range(reg->umin_value, | |
2170 | reg->umax_value)); | |
7be14c1c DB |
2171 | struct tnum var32_off = tnum_intersect(tnum_subreg(var64_off), |
2172 | tnum_range(reg->u32_min_value, | |
2173 | reg->u32_max_value)); | |
3f50f132 JF |
2174 | |
2175 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
2176 | } |
2177 | ||
3844d153 DB |
2178 | static void reg_bounds_sync(struct bpf_reg_state *reg) |
2179 | { | |
2180 | /* We might have learned new bounds from the var_off. */ | |
2181 | __update_reg_bounds(reg); | |
2182 | /* We might have learned something about the sign bit. */ | |
2183 | __reg_deduce_bounds(reg); | |
2184 | /* We might have learned some bits from the bounds. */ | |
2185 | __reg_bound_offset(reg); | |
2186 | /* Intersecting with the old var_off might have improved our bounds | |
2187 | * slightly, e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2188 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2189 | */ | |
2190 | __update_reg_bounds(reg); | |
2191 | } | |
2192 | ||
e572ff80 DB |
2193 | static bool __reg32_bound_s64(s32 a) |
2194 | { | |
2195 | return a >= 0 && a <= S32_MAX; | |
2196 | } | |
2197 | ||
3f50f132 | 2198 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 2199 | { |
3f50f132 JF |
2200 | reg->umin_value = reg->u32_min_value; |
2201 | reg->umax_value = reg->u32_max_value; | |
e572ff80 DB |
2202 | |
2203 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds but must | |
2204 | * be positive otherwise set to worse case bounds and refine later | |
2205 | * from tnum. | |
3f50f132 | 2206 | */ |
e572ff80 DB |
2207 | if (__reg32_bound_s64(reg->s32_min_value) && |
2208 | __reg32_bound_s64(reg->s32_max_value)) { | |
3a71dc36 | 2209 | reg->smin_value = reg->s32_min_value; |
e572ff80 DB |
2210 | reg->smax_value = reg->s32_max_value; |
2211 | } else { | |
3a71dc36 | 2212 | reg->smin_value = 0; |
e572ff80 DB |
2213 | reg->smax_value = U32_MAX; |
2214 | } | |
3f50f132 JF |
2215 | } |
2216 | ||
2217 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
2218 | { | |
2219 | /* special case when 64-bit register has upper 32-bit register | |
2220 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
2221 | * allowing us to use 32-bit bounds directly, | |
2222 | */ | |
2223 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
2224 | __reg_assign_32_into_64(reg); | |
2225 | } else { | |
2226 | /* Otherwise the best we can do is push lower 32bit known and | |
2227 | * unknown bits into register (var_off set from jmp logic) | |
2228 | * then learn as much as possible from the 64-bit tnum | |
2229 | * known and unknown bits. The previous smin/smax bounds are | |
2230 | * invalid here because of jmp32 compare so mark them unknown | |
2231 | * so they do not impact tnum bounds calculation. | |
2232 | */ | |
2233 | __mark_reg64_unbounded(reg); | |
3f50f132 | 2234 | } |
3844d153 | 2235 | reg_bounds_sync(reg); |
3f50f132 JF |
2236 | } |
2237 | ||
2238 | static bool __reg64_bound_s32(s64 a) | |
2239 | { | |
388e2c0b | 2240 | return a >= S32_MIN && a <= S32_MAX; |
3f50f132 JF |
2241 | } |
2242 | ||
2243 | static bool __reg64_bound_u32(u64 a) | |
2244 | { | |
b9979db8 | 2245 | return a >= U32_MIN && a <= U32_MAX; |
3f50f132 JF |
2246 | } |
2247 | ||
2248 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
2249 | { | |
2250 | __mark_reg32_unbounded(reg); | |
b0270958 | 2251 | if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) { |
3f50f132 | 2252 | reg->s32_min_value = (s32)reg->smin_value; |
3f50f132 | 2253 | reg->s32_max_value = (s32)reg->smax_value; |
b0270958 | 2254 | } |
10bf4e83 | 2255 | if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value)) { |
3f50f132 | 2256 | reg->u32_min_value = (u32)reg->umin_value; |
3f50f132 | 2257 | reg->u32_max_value = (u32)reg->umax_value; |
10bf4e83 | 2258 | } |
3844d153 | 2259 | reg_bounds_sync(reg); |
b03c9f9f EC |
2260 | } |
2261 | ||
f1174f77 | 2262 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
2263 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
2264 | struct bpf_reg_state *reg) | |
f1174f77 | 2265 | { |
a9c676bc | 2266 | /* |
a73bf9f2 | 2267 | * Clear type, off, and union(map_ptr, range) and |
a9c676bc AS |
2268 | * padding between 'type' and union |
2269 | */ | |
2270 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 2271 | reg->type = SCALAR_VALUE; |
a73bf9f2 AN |
2272 | reg->id = 0; |
2273 | reg->ref_obj_id = 0; | |
f1174f77 | 2274 | reg->var_off = tnum_unknown; |
f4d7e40a | 2275 | reg->frameno = 0; |
be2ef816 | 2276 | reg->precise = !env->bpf_capable; |
b03c9f9f | 2277 | __mark_reg_unbounded(reg); |
f1174f77 EC |
2278 | } |
2279 | ||
61bd5218 JK |
2280 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
2281 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
2282 | { |
2283 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 2284 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
2285 | /* Something bad happened, let's kill all regs except FP */ |
2286 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 2287 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
2288 | return; |
2289 | } | |
f54c7898 | 2290 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
2291 | } |
2292 | ||
f54c7898 DB |
2293 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
2294 | struct bpf_reg_state *reg) | |
f1174f77 | 2295 | { |
f54c7898 | 2296 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
2297 | reg->type = NOT_INIT; |
2298 | } | |
2299 | ||
61bd5218 JK |
2300 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
2301 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
2302 | { |
2303 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 2304 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
2305 | /* Something bad happened, let's kill all regs except FP */ |
2306 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 2307 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
2308 | return; |
2309 | } | |
f54c7898 | 2310 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
2311 | } |
2312 | ||
41c48f3a AI |
2313 | static void mark_btf_ld_reg(struct bpf_verifier_env *env, |
2314 | struct bpf_reg_state *regs, u32 regno, | |
22dc4a0f | 2315 | enum bpf_reg_type reg_type, |
c6f1bfe8 YS |
2316 | struct btf *btf, u32 btf_id, |
2317 | enum bpf_type_flag flag) | |
41c48f3a AI |
2318 | { |
2319 | if (reg_type == SCALAR_VALUE) { | |
2320 | mark_reg_unknown(env, regs, regno); | |
2321 | return; | |
2322 | } | |
2323 | mark_reg_known_zero(env, regs, regno); | |
c6f1bfe8 | 2324 | regs[regno].type = PTR_TO_BTF_ID | flag; |
22dc4a0f | 2325 | regs[regno].btf = btf; |
41c48f3a AI |
2326 | regs[regno].btf_id = btf_id; |
2327 | } | |
2328 | ||
5327ed3d | 2329 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 2330 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 2331 | struct bpf_func_state *state) |
17a52670 | 2332 | { |
f4d7e40a | 2333 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
2334 | int i; |
2335 | ||
dc503a8a | 2336 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 2337 | mark_reg_not_init(env, regs, i); |
dc503a8a | 2338 | regs[i].live = REG_LIVE_NONE; |
679c782d | 2339 | regs[i].parent = NULL; |
5327ed3d | 2340 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 2341 | } |
17a52670 AS |
2342 | |
2343 | /* frame pointer */ | |
f1174f77 | 2344 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 2345 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 2346 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
2347 | } |
2348 | ||
f4d7e40a AS |
2349 | #define BPF_MAIN_FUNC (-1) |
2350 | static void init_func_state(struct bpf_verifier_env *env, | |
2351 | struct bpf_func_state *state, | |
2352 | int callsite, int frameno, int subprogno) | |
2353 | { | |
2354 | state->callsite = callsite; | |
2355 | state->frameno = frameno; | |
2356 | state->subprogno = subprogno; | |
1bfe26fb | 2357 | state->callback_ret_range = tnum_range(0, 0); |
f4d7e40a | 2358 | init_reg_state(env, state); |
0f55f9ed | 2359 | mark_verifier_state_scratched(env); |
f4d7e40a AS |
2360 | } |
2361 | ||
bfc6bb74 AS |
2362 | /* Similar to push_stack(), but for async callbacks */ |
2363 | static struct bpf_verifier_state *push_async_cb(struct bpf_verifier_env *env, | |
2364 | int insn_idx, int prev_insn_idx, | |
2365 | int subprog) | |
2366 | { | |
2367 | struct bpf_verifier_stack_elem *elem; | |
2368 | struct bpf_func_state *frame; | |
2369 | ||
2370 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); | |
2371 | if (!elem) | |
2372 | goto err; | |
2373 | ||
2374 | elem->insn_idx = insn_idx; | |
2375 | elem->prev_insn_idx = prev_insn_idx; | |
2376 | elem->next = env->head; | |
12166409 | 2377 | elem->log_pos = env->log.end_pos; |
bfc6bb74 AS |
2378 | env->head = elem; |
2379 | env->stack_size++; | |
2380 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { | |
2381 | verbose(env, | |
2382 | "The sequence of %d jumps is too complex for async cb.\n", | |
2383 | env->stack_size); | |
2384 | goto err; | |
2385 | } | |
2386 | /* Unlike push_stack() do not copy_verifier_state(). | |
2387 | * The caller state doesn't matter. | |
2388 | * This is async callback. It starts in a fresh stack. | |
2389 | * Initialize it similar to do_check_common(). | |
2390 | */ | |
2391 | elem->st.branches = 1; | |
2392 | frame = kzalloc(sizeof(*frame), GFP_KERNEL); | |
2393 | if (!frame) | |
2394 | goto err; | |
2395 | init_func_state(env, frame, | |
2396 | BPF_MAIN_FUNC /* callsite */, | |
2397 | 0 /* frameno within this callchain */, | |
2398 | subprog /* subprog number within this prog */); | |
2399 | elem->st.frame[0] = frame; | |
2400 | return &elem->st; | |
2401 | err: | |
2402 | free_verifier_state(env->cur_state, true); | |
2403 | env->cur_state = NULL; | |
2404 | /* pop all elements and return */ | |
2405 | while (!pop_stack(env, NULL, NULL, false)); | |
2406 | return NULL; | |
2407 | } | |
2408 | ||
2409 | ||
17a52670 AS |
2410 | enum reg_arg_type { |
2411 | SRC_OP, /* register is used as source operand */ | |
2412 | DST_OP, /* register is used as destination operand */ | |
2413 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
2414 | }; | |
2415 | ||
cc8b0b92 AS |
2416 | static int cmp_subprogs(const void *a, const void *b) |
2417 | { | |
9c8105bd JW |
2418 | return ((struct bpf_subprog_info *)a)->start - |
2419 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
2420 | } |
2421 | ||
2422 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
2423 | { | |
9c8105bd | 2424 | struct bpf_subprog_info *p; |
cc8b0b92 | 2425 | |
9c8105bd JW |
2426 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
2427 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
2428 | if (!p) |
2429 | return -ENOENT; | |
9c8105bd | 2430 | return p - env->subprog_info; |
cc8b0b92 AS |
2431 | |
2432 | } | |
2433 | ||
2434 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
2435 | { | |
2436 | int insn_cnt = env->prog->len; | |
2437 | int ret; | |
2438 | ||
2439 | if (off >= insn_cnt || off < 0) { | |
2440 | verbose(env, "call to invalid destination\n"); | |
2441 | return -EINVAL; | |
2442 | } | |
2443 | ret = find_subprog(env, off); | |
2444 | if (ret >= 0) | |
282a0f46 | 2445 | return ret; |
4cb3d99c | 2446 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
2447 | verbose(env, "too many subprograms\n"); |
2448 | return -E2BIG; | |
2449 | } | |
e6ac2450 | 2450 | /* determine subprog starts. The end is one before the next starts */ |
9c8105bd JW |
2451 | env->subprog_info[env->subprog_cnt++].start = off; |
2452 | sort(env->subprog_info, env->subprog_cnt, | |
2453 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
282a0f46 | 2454 | return env->subprog_cnt - 1; |
cc8b0b92 AS |
2455 | } |
2456 | ||
2357672c KKD |
2457 | #define MAX_KFUNC_DESCS 256 |
2458 | #define MAX_KFUNC_BTFS 256 | |
2459 | ||
e6ac2450 MKL |
2460 | struct bpf_kfunc_desc { |
2461 | struct btf_func_model func_model; | |
2462 | u32 func_id; | |
2463 | s32 imm; | |
2357672c | 2464 | u16 offset; |
1cf3bfc6 | 2465 | unsigned long addr; |
2357672c KKD |
2466 | }; |
2467 | ||
2468 | struct bpf_kfunc_btf { | |
2469 | struct btf *btf; | |
2470 | struct module *module; | |
2471 | u16 offset; | |
e6ac2450 MKL |
2472 | }; |
2473 | ||
e6ac2450 | 2474 | struct bpf_kfunc_desc_tab { |
1cf3bfc6 IL |
2475 | /* Sorted by func_id (BTF ID) and offset (fd_array offset) during |
2476 | * verification. JITs do lookups by bpf_insn, where func_id may not be | |
2477 | * available, therefore at the end of verification do_misc_fixups() | |
2478 | * sorts this by imm and offset. | |
2479 | */ | |
e6ac2450 MKL |
2480 | struct bpf_kfunc_desc descs[MAX_KFUNC_DESCS]; |
2481 | u32 nr_descs; | |
2482 | }; | |
2483 | ||
2357672c KKD |
2484 | struct bpf_kfunc_btf_tab { |
2485 | struct bpf_kfunc_btf descs[MAX_KFUNC_BTFS]; | |
2486 | u32 nr_descs; | |
2487 | }; | |
2488 | ||
2489 | static int kfunc_desc_cmp_by_id_off(const void *a, const void *b) | |
e6ac2450 MKL |
2490 | { |
2491 | const struct bpf_kfunc_desc *d0 = a; | |
2492 | const struct bpf_kfunc_desc *d1 = b; | |
2493 | ||
2494 | /* func_id is not greater than BTF_MAX_TYPE */ | |
2357672c KKD |
2495 | return d0->func_id - d1->func_id ?: d0->offset - d1->offset; |
2496 | } | |
2497 | ||
2498 | static int kfunc_btf_cmp_by_off(const void *a, const void *b) | |
2499 | { | |
2500 | const struct bpf_kfunc_btf *d0 = a; | |
2501 | const struct bpf_kfunc_btf *d1 = b; | |
2502 | ||
2503 | return d0->offset - d1->offset; | |
e6ac2450 MKL |
2504 | } |
2505 | ||
2506 | static const struct bpf_kfunc_desc * | |
2357672c | 2507 | find_kfunc_desc(const struct bpf_prog *prog, u32 func_id, u16 offset) |
e6ac2450 MKL |
2508 | { |
2509 | struct bpf_kfunc_desc desc = { | |
2510 | .func_id = func_id, | |
2357672c | 2511 | .offset = offset, |
e6ac2450 MKL |
2512 | }; |
2513 | struct bpf_kfunc_desc_tab *tab; | |
2514 | ||
2515 | tab = prog->aux->kfunc_tab; | |
2516 | return bsearch(&desc, tab->descs, tab->nr_descs, | |
2357672c KKD |
2517 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_id_off); |
2518 | } | |
2519 | ||
1cf3bfc6 IL |
2520 | int bpf_get_kfunc_addr(const struct bpf_prog *prog, u32 func_id, |
2521 | u16 btf_fd_idx, u8 **func_addr) | |
2522 | { | |
2523 | const struct bpf_kfunc_desc *desc; | |
2524 | ||
2525 | desc = find_kfunc_desc(prog, func_id, btf_fd_idx); | |
2526 | if (!desc) | |
2527 | return -EFAULT; | |
2528 | ||
2529 | *func_addr = (u8 *)desc->addr; | |
2530 | return 0; | |
2531 | } | |
2532 | ||
2357672c | 2533 | static struct btf *__find_kfunc_desc_btf(struct bpf_verifier_env *env, |
b202d844 | 2534 | s16 offset) |
2357672c KKD |
2535 | { |
2536 | struct bpf_kfunc_btf kf_btf = { .offset = offset }; | |
2537 | struct bpf_kfunc_btf_tab *tab; | |
2538 | struct bpf_kfunc_btf *b; | |
2539 | struct module *mod; | |
2540 | struct btf *btf; | |
2541 | int btf_fd; | |
2542 | ||
2543 | tab = env->prog->aux->kfunc_btf_tab; | |
2544 | b = bsearch(&kf_btf, tab->descs, tab->nr_descs, | |
2545 | sizeof(tab->descs[0]), kfunc_btf_cmp_by_off); | |
2546 | if (!b) { | |
2547 | if (tab->nr_descs == MAX_KFUNC_BTFS) { | |
2548 | verbose(env, "too many different module BTFs\n"); | |
2549 | return ERR_PTR(-E2BIG); | |
2550 | } | |
2551 | ||
2552 | if (bpfptr_is_null(env->fd_array)) { | |
2553 | verbose(env, "kfunc offset > 0 without fd_array is invalid\n"); | |
2554 | return ERR_PTR(-EPROTO); | |
2555 | } | |
2556 | ||
2557 | if (copy_from_bpfptr_offset(&btf_fd, env->fd_array, | |
2558 | offset * sizeof(btf_fd), | |
2559 | sizeof(btf_fd))) | |
2560 | return ERR_PTR(-EFAULT); | |
2561 | ||
2562 | btf = btf_get_by_fd(btf_fd); | |
588cd7ef KKD |
2563 | if (IS_ERR(btf)) { |
2564 | verbose(env, "invalid module BTF fd specified\n"); | |
2357672c | 2565 | return btf; |
588cd7ef | 2566 | } |
2357672c KKD |
2567 | |
2568 | if (!btf_is_module(btf)) { | |
2569 | verbose(env, "BTF fd for kfunc is not a module BTF\n"); | |
2570 | btf_put(btf); | |
2571 | return ERR_PTR(-EINVAL); | |
2572 | } | |
2573 | ||
2574 | mod = btf_try_get_module(btf); | |
2575 | if (!mod) { | |
2576 | btf_put(btf); | |
2577 | return ERR_PTR(-ENXIO); | |
2578 | } | |
2579 | ||
2580 | b = &tab->descs[tab->nr_descs++]; | |
2581 | b->btf = btf; | |
2582 | b->module = mod; | |
2583 | b->offset = offset; | |
2584 | ||
2585 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2586 | kfunc_btf_cmp_by_off, NULL); | |
2587 | } | |
2357672c | 2588 | return b->btf; |
e6ac2450 MKL |
2589 | } |
2590 | ||
2357672c KKD |
2591 | void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab) |
2592 | { | |
2593 | if (!tab) | |
2594 | return; | |
2595 | ||
2596 | while (tab->nr_descs--) { | |
2597 | module_put(tab->descs[tab->nr_descs].module); | |
2598 | btf_put(tab->descs[tab->nr_descs].btf); | |
2599 | } | |
2600 | kfree(tab); | |
2601 | } | |
2602 | ||
43bf0878 | 2603 | static struct btf *find_kfunc_desc_btf(struct bpf_verifier_env *env, s16 offset) |
2357672c | 2604 | { |
2357672c KKD |
2605 | if (offset) { |
2606 | if (offset < 0) { | |
2607 | /* In the future, this can be allowed to increase limit | |
2608 | * of fd index into fd_array, interpreted as u16. | |
2609 | */ | |
2610 | verbose(env, "negative offset disallowed for kernel module function call\n"); | |
2611 | return ERR_PTR(-EINVAL); | |
2612 | } | |
2613 | ||
b202d844 | 2614 | return __find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2615 | } |
2616 | return btf_vmlinux ?: ERR_PTR(-ENOENT); | |
e6ac2450 MKL |
2617 | } |
2618 | ||
2357672c | 2619 | static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset) |
e6ac2450 MKL |
2620 | { |
2621 | const struct btf_type *func, *func_proto; | |
2357672c | 2622 | struct bpf_kfunc_btf_tab *btf_tab; |
e6ac2450 MKL |
2623 | struct bpf_kfunc_desc_tab *tab; |
2624 | struct bpf_prog_aux *prog_aux; | |
2625 | struct bpf_kfunc_desc *desc; | |
2626 | const char *func_name; | |
2357672c | 2627 | struct btf *desc_btf; |
8cbf062a | 2628 | unsigned long call_imm; |
e6ac2450 MKL |
2629 | unsigned long addr; |
2630 | int err; | |
2631 | ||
2632 | prog_aux = env->prog->aux; | |
2633 | tab = prog_aux->kfunc_tab; | |
2357672c | 2634 | btf_tab = prog_aux->kfunc_btf_tab; |
e6ac2450 MKL |
2635 | if (!tab) { |
2636 | if (!btf_vmlinux) { | |
2637 | verbose(env, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
2638 | return -ENOTSUPP; | |
2639 | } | |
2640 | ||
2641 | if (!env->prog->jit_requested) { | |
2642 | verbose(env, "JIT is required for calling kernel function\n"); | |
2643 | return -ENOTSUPP; | |
2644 | } | |
2645 | ||
2646 | if (!bpf_jit_supports_kfunc_call()) { | |
2647 | verbose(env, "JIT does not support calling kernel function\n"); | |
2648 | return -ENOTSUPP; | |
2649 | } | |
2650 | ||
2651 | if (!env->prog->gpl_compatible) { | |
2652 | verbose(env, "cannot call kernel function from non-GPL compatible program\n"); | |
2653 | return -EINVAL; | |
2654 | } | |
2655 | ||
2656 | tab = kzalloc(sizeof(*tab), GFP_KERNEL); | |
2657 | if (!tab) | |
2658 | return -ENOMEM; | |
2659 | prog_aux->kfunc_tab = tab; | |
2660 | } | |
2661 | ||
a5d82727 KKD |
2662 | /* func_id == 0 is always invalid, but instead of returning an error, be |
2663 | * conservative and wait until the code elimination pass before returning | |
2664 | * error, so that invalid calls that get pruned out can be in BPF programs | |
2665 | * loaded from userspace. It is also required that offset be untouched | |
2666 | * for such calls. | |
2667 | */ | |
2668 | if (!func_id && !offset) | |
2669 | return 0; | |
2670 | ||
2357672c KKD |
2671 | if (!btf_tab && offset) { |
2672 | btf_tab = kzalloc(sizeof(*btf_tab), GFP_KERNEL); | |
2673 | if (!btf_tab) | |
2674 | return -ENOMEM; | |
2675 | prog_aux->kfunc_btf_tab = btf_tab; | |
2676 | } | |
2677 | ||
43bf0878 | 2678 | desc_btf = find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2679 | if (IS_ERR(desc_btf)) { |
2680 | verbose(env, "failed to find BTF for kernel function\n"); | |
2681 | return PTR_ERR(desc_btf); | |
2682 | } | |
2683 | ||
2684 | if (find_kfunc_desc(env->prog, func_id, offset)) | |
e6ac2450 MKL |
2685 | return 0; |
2686 | ||
2687 | if (tab->nr_descs == MAX_KFUNC_DESCS) { | |
2688 | verbose(env, "too many different kernel function calls\n"); | |
2689 | return -E2BIG; | |
2690 | } | |
2691 | ||
2357672c | 2692 | func = btf_type_by_id(desc_btf, func_id); |
e6ac2450 MKL |
2693 | if (!func || !btf_type_is_func(func)) { |
2694 | verbose(env, "kernel btf_id %u is not a function\n", | |
2695 | func_id); | |
2696 | return -EINVAL; | |
2697 | } | |
2357672c | 2698 | func_proto = btf_type_by_id(desc_btf, func->type); |
e6ac2450 MKL |
2699 | if (!func_proto || !btf_type_is_func_proto(func_proto)) { |
2700 | verbose(env, "kernel function btf_id %u does not have a valid func_proto\n", | |
2701 | func_id); | |
2702 | return -EINVAL; | |
2703 | } | |
2704 | ||
2357672c | 2705 | func_name = btf_name_by_offset(desc_btf, func->name_off); |
e6ac2450 MKL |
2706 | addr = kallsyms_lookup_name(func_name); |
2707 | if (!addr) { | |
2708 | verbose(env, "cannot find address for kernel function %s\n", | |
2709 | func_name); | |
2710 | return -EINVAL; | |
2711 | } | |
1cf3bfc6 | 2712 | specialize_kfunc(env, func_id, offset, &addr); |
e6ac2450 | 2713 | |
1cf3bfc6 IL |
2714 | if (bpf_jit_supports_far_kfunc_call()) { |
2715 | call_imm = func_id; | |
2716 | } else { | |
2717 | call_imm = BPF_CALL_IMM(addr); | |
2718 | /* Check whether the relative offset overflows desc->imm */ | |
2719 | if ((unsigned long)(s32)call_imm != call_imm) { | |
2720 | verbose(env, "address of kernel function %s is out of range\n", | |
2721 | func_name); | |
2722 | return -EINVAL; | |
2723 | } | |
8cbf062a HT |
2724 | } |
2725 | ||
3d76a4d3 SF |
2726 | if (bpf_dev_bound_kfunc_id(func_id)) { |
2727 | err = bpf_dev_bound_kfunc_check(&env->log, prog_aux); | |
2728 | if (err) | |
2729 | return err; | |
2730 | } | |
2731 | ||
e6ac2450 MKL |
2732 | desc = &tab->descs[tab->nr_descs++]; |
2733 | desc->func_id = func_id; | |
8cbf062a | 2734 | desc->imm = call_imm; |
2357672c | 2735 | desc->offset = offset; |
1cf3bfc6 | 2736 | desc->addr = addr; |
2357672c | 2737 | err = btf_distill_func_proto(&env->log, desc_btf, |
e6ac2450 MKL |
2738 | func_proto, func_name, |
2739 | &desc->func_model); | |
2740 | if (!err) | |
2741 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2357672c | 2742 | kfunc_desc_cmp_by_id_off, NULL); |
e6ac2450 MKL |
2743 | return err; |
2744 | } | |
2745 | ||
1cf3bfc6 | 2746 | static int kfunc_desc_cmp_by_imm_off(const void *a, const void *b) |
e6ac2450 MKL |
2747 | { |
2748 | const struct bpf_kfunc_desc *d0 = a; | |
2749 | const struct bpf_kfunc_desc *d1 = b; | |
2750 | ||
1cf3bfc6 IL |
2751 | if (d0->imm != d1->imm) |
2752 | return d0->imm < d1->imm ? -1 : 1; | |
2753 | if (d0->offset != d1->offset) | |
2754 | return d0->offset < d1->offset ? -1 : 1; | |
e6ac2450 MKL |
2755 | return 0; |
2756 | } | |
2757 | ||
1cf3bfc6 | 2758 | static void sort_kfunc_descs_by_imm_off(struct bpf_prog *prog) |
e6ac2450 MKL |
2759 | { |
2760 | struct bpf_kfunc_desc_tab *tab; | |
2761 | ||
2762 | tab = prog->aux->kfunc_tab; | |
2763 | if (!tab) | |
2764 | return; | |
2765 | ||
2766 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
1cf3bfc6 | 2767 | kfunc_desc_cmp_by_imm_off, NULL); |
e6ac2450 MKL |
2768 | } |
2769 | ||
2770 | bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) | |
2771 | { | |
2772 | return !!prog->aux->kfunc_tab; | |
2773 | } | |
2774 | ||
2775 | const struct btf_func_model * | |
2776 | bpf_jit_find_kfunc_model(const struct bpf_prog *prog, | |
2777 | const struct bpf_insn *insn) | |
2778 | { | |
2779 | const struct bpf_kfunc_desc desc = { | |
2780 | .imm = insn->imm, | |
1cf3bfc6 | 2781 | .offset = insn->off, |
e6ac2450 MKL |
2782 | }; |
2783 | const struct bpf_kfunc_desc *res; | |
2784 | struct bpf_kfunc_desc_tab *tab; | |
2785 | ||
2786 | tab = prog->aux->kfunc_tab; | |
2787 | res = bsearch(&desc, tab->descs, tab->nr_descs, | |
1cf3bfc6 | 2788 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm_off); |
e6ac2450 MKL |
2789 | |
2790 | return res ? &res->func_model : NULL; | |
2791 | } | |
2792 | ||
2793 | static int add_subprog_and_kfunc(struct bpf_verifier_env *env) | |
cc8b0b92 | 2794 | { |
9c8105bd | 2795 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 | 2796 | struct bpf_insn *insn = env->prog->insnsi; |
e6ac2450 | 2797 | int i, ret, insn_cnt = env->prog->len; |
cc8b0b92 | 2798 | |
f910cefa JW |
2799 | /* Add entry function. */ |
2800 | ret = add_subprog(env, 0); | |
e6ac2450 | 2801 | if (ret) |
f910cefa JW |
2802 | return ret; |
2803 | ||
e6ac2450 MKL |
2804 | for (i = 0; i < insn_cnt; i++, insn++) { |
2805 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn) && | |
2806 | !bpf_pseudo_kfunc_call(insn)) | |
cc8b0b92 | 2807 | continue; |
e6ac2450 | 2808 | |
2c78ee89 | 2809 | if (!env->bpf_capable) { |
e6ac2450 | 2810 | verbose(env, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); |
cc8b0b92 AS |
2811 | return -EPERM; |
2812 | } | |
e6ac2450 | 2813 | |
3990ed4c | 2814 | if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn)) |
e6ac2450 | 2815 | ret = add_subprog(env, i + insn->imm + 1); |
3990ed4c | 2816 | else |
2357672c | 2817 | ret = add_kfunc_call(env, insn->imm, insn->off); |
e6ac2450 | 2818 | |
cc8b0b92 AS |
2819 | if (ret < 0) |
2820 | return ret; | |
2821 | } | |
2822 | ||
4cb3d99c JW |
2823 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
2824 | * logic. 'subprog_cnt' should not be increased. | |
2825 | */ | |
2826 | subprog[env->subprog_cnt].start = insn_cnt; | |
2827 | ||
06ee7115 | 2828 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 2829 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 2830 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 | 2831 | |
e6ac2450 MKL |
2832 | return 0; |
2833 | } | |
2834 | ||
2835 | static int check_subprogs(struct bpf_verifier_env *env) | |
2836 | { | |
2837 | int i, subprog_start, subprog_end, off, cur_subprog = 0; | |
2838 | struct bpf_subprog_info *subprog = env->subprog_info; | |
2839 | struct bpf_insn *insn = env->prog->insnsi; | |
2840 | int insn_cnt = env->prog->len; | |
2841 | ||
cc8b0b92 | 2842 | /* now check that all jumps are within the same subprog */ |
4cb3d99c JW |
2843 | subprog_start = subprog[cur_subprog].start; |
2844 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
2845 | for (i = 0; i < insn_cnt; i++) { |
2846 | u8 code = insn[i].code; | |
2847 | ||
7f6e4312 | 2848 | if (code == (BPF_JMP | BPF_CALL) && |
df2ccc18 IL |
2849 | insn[i].src_reg == 0 && |
2850 | insn[i].imm == BPF_FUNC_tail_call) | |
7f6e4312 | 2851 | subprog[cur_subprog].has_tail_call = true; |
09b28d76 AS |
2852 | if (BPF_CLASS(code) == BPF_LD && |
2853 | (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND)) | |
2854 | subprog[cur_subprog].has_ld_abs = true; | |
092ed096 | 2855 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
2856 | goto next; |
2857 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
2858 | goto next; | |
4cd58e9a YS |
2859 | if (code == (BPF_JMP32 | BPF_JA)) |
2860 | off = i + insn[i].imm + 1; | |
2861 | else | |
2862 | off = i + insn[i].off + 1; | |
cc8b0b92 AS |
2863 | if (off < subprog_start || off >= subprog_end) { |
2864 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
2865 | return -EINVAL; | |
2866 | } | |
2867 | next: | |
2868 | if (i == subprog_end - 1) { | |
2869 | /* to avoid fall-through from one subprog into another | |
2870 | * the last insn of the subprog should be either exit | |
2871 | * or unconditional jump back | |
2872 | */ | |
2873 | if (code != (BPF_JMP | BPF_EXIT) && | |
4cd58e9a | 2874 | code != (BPF_JMP32 | BPF_JA) && |
cc8b0b92 AS |
2875 | code != (BPF_JMP | BPF_JA)) { |
2876 | verbose(env, "last insn is not an exit or jmp\n"); | |
2877 | return -EINVAL; | |
2878 | } | |
2879 | subprog_start = subprog_end; | |
4cb3d99c JW |
2880 | cur_subprog++; |
2881 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 2882 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
2883 | } |
2884 | } | |
2885 | return 0; | |
2886 | } | |
2887 | ||
679c782d EC |
2888 | /* Parentage chain of this register (or stack slot) should take care of all |
2889 | * issues like callee-saved registers, stack slot allocation time, etc. | |
2890 | */ | |
f4d7e40a | 2891 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 2892 | const struct bpf_reg_state *state, |
5327ed3d | 2893 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
2894 | { |
2895 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 2896 | int cnt = 0; |
dc503a8a EC |
2897 | |
2898 | while (parent) { | |
2899 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 2900 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 2901 | break; |
9242b5f5 AS |
2902 | if (parent->live & REG_LIVE_DONE) { |
2903 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
c25b2ae1 | 2904 | reg_type_str(env, parent->type), |
9242b5f5 AS |
2905 | parent->var_off.value, parent->off); |
2906 | return -EFAULT; | |
2907 | } | |
5327ed3d JW |
2908 | /* The first condition is more likely to be true than the |
2909 | * second, checked it first. | |
2910 | */ | |
2911 | if ((parent->live & REG_LIVE_READ) == flag || | |
2912 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
2913 | /* The parentage chain never changes and |
2914 | * this parent was already marked as LIVE_READ. | |
2915 | * There is no need to keep walking the chain again and | |
2916 | * keep re-marking all parents as LIVE_READ. | |
2917 | * This case happens when the same register is read | |
2918 | * multiple times without writes into it in-between. | |
5327ed3d JW |
2919 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
2920 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
2921 | */ |
2922 | break; | |
dc503a8a | 2923 | /* ... then we depend on parent's value */ |
5327ed3d JW |
2924 | parent->live |= flag; |
2925 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
2926 | if (flag == REG_LIVE_READ64) | |
2927 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
2928 | state = parent; |
2929 | parent = state->parent; | |
f4d7e40a | 2930 | writes = true; |
06ee7115 | 2931 | cnt++; |
dc503a8a | 2932 | } |
06ee7115 AS |
2933 | |
2934 | if (env->longest_mark_read_walk < cnt) | |
2935 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 2936 | return 0; |
dc503a8a EC |
2937 | } |
2938 | ||
d6fefa11 KKD |
2939 | static int mark_dynptr_read(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
2940 | { | |
2941 | struct bpf_func_state *state = func(env, reg); | |
2942 | int spi, ret; | |
2943 | ||
2944 | /* For CONST_PTR_TO_DYNPTR, it must have already been done by | |
2945 | * check_reg_arg in check_helper_call and mark_btf_func_reg_size in | |
2946 | * check_kfunc_call. | |
2947 | */ | |
2948 | if (reg->type == CONST_PTR_TO_DYNPTR) | |
2949 | return 0; | |
79168a66 KKD |
2950 | spi = dynptr_get_spi(env, reg); |
2951 | if (spi < 0) | |
2952 | return spi; | |
d6fefa11 KKD |
2953 | /* Caller ensures dynptr is valid and initialized, which means spi is in |
2954 | * bounds and spi is the first dynptr slot. Simply mark stack slot as | |
2955 | * read. | |
2956 | */ | |
2957 | ret = mark_reg_read(env, &state->stack[spi].spilled_ptr, | |
2958 | state->stack[spi].spilled_ptr.parent, REG_LIVE_READ64); | |
2959 | if (ret) | |
2960 | return ret; | |
2961 | return mark_reg_read(env, &state->stack[spi - 1].spilled_ptr, | |
2962 | state->stack[spi - 1].spilled_ptr.parent, REG_LIVE_READ64); | |
2963 | } | |
2964 | ||
06accc87 AN |
2965 | static int mark_iter_read(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
2966 | int spi, int nr_slots) | |
2967 | { | |
2968 | struct bpf_func_state *state = func(env, reg); | |
2969 | int err, i; | |
2970 | ||
2971 | for (i = 0; i < nr_slots; i++) { | |
2972 | struct bpf_reg_state *st = &state->stack[spi - i].spilled_ptr; | |
2973 | ||
2974 | err = mark_reg_read(env, st, st->parent, REG_LIVE_READ64); | |
2975 | if (err) | |
2976 | return err; | |
2977 | ||
2978 | mark_stack_slot_scratched(env, spi - i); | |
2979 | } | |
2980 | ||
2981 | return 0; | |
2982 | } | |
2983 | ||
5327ed3d JW |
2984 | /* This function is supposed to be used by the following 32-bit optimization |
2985 | * code only. It returns TRUE if the source or destination register operates | |
2986 | * on 64-bit, otherwise return FALSE. | |
2987 | */ | |
2988 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
2989 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
2990 | { | |
2991 | u8 code, class, op; | |
2992 | ||
2993 | code = insn->code; | |
2994 | class = BPF_CLASS(code); | |
2995 | op = BPF_OP(code); | |
2996 | if (class == BPF_JMP) { | |
2997 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
2998 | * conservatively. | |
2999 | */ | |
3000 | if (op == BPF_EXIT) | |
3001 | return true; | |
3002 | if (op == BPF_CALL) { | |
3003 | /* BPF to BPF call will reach here because of marking | |
3004 | * caller saved clobber with DST_OP_NO_MARK for which we | |
3005 | * don't care the register def because they are anyway | |
3006 | * marked as NOT_INIT already. | |
3007 | */ | |
3008 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
3009 | return false; | |
3010 | /* Helper call will reach here because of arg type | |
3011 | * check, conservatively return TRUE. | |
3012 | */ | |
3013 | if (t == SRC_OP) | |
3014 | return true; | |
3015 | ||
3016 | return false; | |
3017 | } | |
3018 | } | |
3019 | ||
0845c3db YS |
3020 | if (class == BPF_ALU64 && op == BPF_END && (insn->imm == 16 || insn->imm == 32)) |
3021 | return false; | |
3022 | ||
5327ed3d | 3023 | if (class == BPF_ALU64 || class == BPF_JMP || |
5327ed3d JW |
3024 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) |
3025 | return true; | |
3026 | ||
3027 | if (class == BPF_ALU || class == BPF_JMP32) | |
3028 | return false; | |
3029 | ||
3030 | if (class == BPF_LDX) { | |
3031 | if (t != SRC_OP) | |
3032 | return BPF_SIZE(code) == BPF_DW; | |
3033 | /* LDX source must be ptr. */ | |
3034 | return true; | |
3035 | } | |
3036 | ||
3037 | if (class == BPF_STX) { | |
83a28819 IL |
3038 | /* BPF_STX (including atomic variants) has multiple source |
3039 | * operands, one of which is a ptr. Check whether the caller is | |
3040 | * asking about it. | |
3041 | */ | |
3042 | if (t == SRC_OP && reg->type != SCALAR_VALUE) | |
5327ed3d JW |
3043 | return true; |
3044 | return BPF_SIZE(code) == BPF_DW; | |
3045 | } | |
3046 | ||
3047 | if (class == BPF_LD) { | |
3048 | u8 mode = BPF_MODE(code); | |
3049 | ||
3050 | /* LD_IMM64 */ | |
3051 | if (mode == BPF_IMM) | |
3052 | return true; | |
3053 | ||
3054 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
3055 | if (t != SRC_OP) | |
3056 | return false; | |
3057 | ||
3058 | /* Implicit ctx ptr. */ | |
3059 | if (regno == BPF_REG_6) | |
3060 | return true; | |
3061 | ||
3062 | /* Explicit source could be any width. */ | |
3063 | return true; | |
3064 | } | |
3065 | ||
3066 | if (class == BPF_ST) | |
3067 | /* The only source register for BPF_ST is a ptr. */ | |
3068 | return true; | |
3069 | ||
3070 | /* Conservatively return true at default. */ | |
3071 | return true; | |
3072 | } | |
3073 | ||
83a28819 IL |
3074 | /* Return the regno defined by the insn, or -1. */ |
3075 | static int insn_def_regno(const struct bpf_insn *insn) | |
b325fbca | 3076 | { |
83a28819 IL |
3077 | switch (BPF_CLASS(insn->code)) { |
3078 | case BPF_JMP: | |
3079 | case BPF_JMP32: | |
3080 | case BPF_ST: | |
3081 | return -1; | |
3082 | case BPF_STX: | |
3083 | if (BPF_MODE(insn->code) == BPF_ATOMIC && | |
3084 | (insn->imm & BPF_FETCH)) { | |
3085 | if (insn->imm == BPF_CMPXCHG) | |
3086 | return BPF_REG_0; | |
3087 | else | |
3088 | return insn->src_reg; | |
3089 | } else { | |
3090 | return -1; | |
3091 | } | |
3092 | default: | |
3093 | return insn->dst_reg; | |
3094 | } | |
b325fbca JW |
3095 | } |
3096 | ||
3097 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
3098 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
3099 | { | |
83a28819 IL |
3100 | int dst_reg = insn_def_regno(insn); |
3101 | ||
3102 | if (dst_reg == -1) | |
b325fbca JW |
3103 | return false; |
3104 | ||
83a28819 | 3105 | return !is_reg64(env, insn, dst_reg, NULL, DST_OP); |
b325fbca JW |
3106 | } |
3107 | ||
5327ed3d JW |
3108 | static void mark_insn_zext(struct bpf_verifier_env *env, |
3109 | struct bpf_reg_state *reg) | |
3110 | { | |
3111 | s32 def_idx = reg->subreg_def; | |
3112 | ||
3113 | if (def_idx == DEF_NOT_SUBREG) | |
3114 | return; | |
3115 | ||
3116 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
3117 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
3118 | reg->subreg_def = DEF_NOT_SUBREG; | |
3119 | } | |
3120 | ||
dc503a8a | 3121 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
3122 | enum reg_arg_type t) |
3123 | { | |
f4d7e40a AS |
3124 | struct bpf_verifier_state *vstate = env->cur_state; |
3125 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 3126 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 3127 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 3128 | bool rw64; |
dc503a8a | 3129 | |
17a52670 | 3130 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 3131 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
3132 | return -EINVAL; |
3133 | } | |
3134 | ||
0f55f9ed CL |
3135 | mark_reg_scratched(env, regno); |
3136 | ||
c342dc10 | 3137 | reg = ®s[regno]; |
5327ed3d | 3138 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
3139 | if (t == SRC_OP) { |
3140 | /* check whether register used as source operand can be read */ | |
c342dc10 | 3141 | if (reg->type == NOT_INIT) { |
61bd5218 | 3142 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
3143 | return -EACCES; |
3144 | } | |
679c782d | 3145 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
3146 | if (regno == BPF_REG_FP) |
3147 | return 0; | |
3148 | ||
5327ed3d JW |
3149 | if (rw64) |
3150 | mark_insn_zext(env, reg); | |
3151 | ||
3152 | return mark_reg_read(env, reg, reg->parent, | |
3153 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
3154 | } else { |
3155 | /* check whether register used as dest operand can be written to */ | |
3156 | if (regno == BPF_REG_FP) { | |
61bd5218 | 3157 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
3158 | return -EACCES; |
3159 | } | |
c342dc10 | 3160 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 3161 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 3162 | if (t == DST_OP) |
61bd5218 | 3163 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
3164 | } |
3165 | return 0; | |
3166 | } | |
3167 | ||
bffdeaa8 AN |
3168 | static void mark_jmp_point(struct bpf_verifier_env *env, int idx) |
3169 | { | |
3170 | env->insn_aux_data[idx].jmp_point = true; | |
3171 | } | |
3172 | ||
3173 | static bool is_jmp_point(struct bpf_verifier_env *env, int insn_idx) | |
3174 | { | |
3175 | return env->insn_aux_data[insn_idx].jmp_point; | |
3176 | } | |
3177 | ||
b5dc0163 AS |
3178 | /* for any branch, call, exit record the history of jmps in the given state */ |
3179 | static int push_jmp_history(struct bpf_verifier_env *env, | |
3180 | struct bpf_verifier_state *cur) | |
3181 | { | |
3182 | u32 cnt = cur->jmp_history_cnt; | |
3183 | struct bpf_idx_pair *p; | |
ceb35b66 | 3184 | size_t alloc_size; |
b5dc0163 | 3185 | |
bffdeaa8 AN |
3186 | if (!is_jmp_point(env, env->insn_idx)) |
3187 | return 0; | |
3188 | ||
b5dc0163 | 3189 | cnt++; |
ceb35b66 KC |
3190 | alloc_size = kmalloc_size_roundup(size_mul(cnt, sizeof(*p))); |
3191 | p = krealloc(cur->jmp_history, alloc_size, GFP_USER); | |
b5dc0163 AS |
3192 | if (!p) |
3193 | return -ENOMEM; | |
3194 | p[cnt - 1].idx = env->insn_idx; | |
3195 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
3196 | cur->jmp_history = p; | |
3197 | cur->jmp_history_cnt = cnt; | |
3198 | return 0; | |
3199 | } | |
3200 | ||
3201 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
3202 | * history then previous instruction came from straight line execution. | |
3203 | */ | |
3204 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
3205 | u32 *history) | |
3206 | { | |
3207 | u32 cnt = *history; | |
3208 | ||
3209 | if (cnt && st->jmp_history[cnt - 1].idx == i) { | |
3210 | i = st->jmp_history[cnt - 1].prev_idx; | |
3211 | (*history)--; | |
3212 | } else { | |
3213 | i--; | |
3214 | } | |
3215 | return i; | |
3216 | } | |
3217 | ||
e6ac2450 MKL |
3218 | static const char *disasm_kfunc_name(void *data, const struct bpf_insn *insn) |
3219 | { | |
3220 | const struct btf_type *func; | |
2357672c | 3221 | struct btf *desc_btf; |
e6ac2450 MKL |
3222 | |
3223 | if (insn->src_reg != BPF_PSEUDO_KFUNC_CALL) | |
3224 | return NULL; | |
3225 | ||
43bf0878 | 3226 | desc_btf = find_kfunc_desc_btf(data, insn->off); |
2357672c KKD |
3227 | if (IS_ERR(desc_btf)) |
3228 | return "<error>"; | |
3229 | ||
3230 | func = btf_type_by_id(desc_btf, insn->imm); | |
3231 | return btf_name_by_offset(desc_btf, func->name_off); | |
e6ac2450 MKL |
3232 | } |
3233 | ||
407958a0 AN |
3234 | static inline void bt_init(struct backtrack_state *bt, u32 frame) |
3235 | { | |
3236 | bt->frame = frame; | |
3237 | } | |
3238 | ||
3239 | static inline void bt_reset(struct backtrack_state *bt) | |
3240 | { | |
3241 | struct bpf_verifier_env *env = bt->env; | |
3242 | ||
3243 | memset(bt, 0, sizeof(*bt)); | |
3244 | bt->env = env; | |
3245 | } | |
3246 | ||
3247 | static inline u32 bt_empty(struct backtrack_state *bt) | |
3248 | { | |
3249 | u64 mask = 0; | |
3250 | int i; | |
3251 | ||
3252 | for (i = 0; i <= bt->frame; i++) | |
3253 | mask |= bt->reg_masks[i] | bt->stack_masks[i]; | |
3254 | ||
3255 | return mask == 0; | |
3256 | } | |
3257 | ||
3258 | static inline int bt_subprog_enter(struct backtrack_state *bt) | |
3259 | { | |
3260 | if (bt->frame == MAX_CALL_FRAMES - 1) { | |
3261 | verbose(bt->env, "BUG subprog enter from frame %d\n", bt->frame); | |
3262 | WARN_ONCE(1, "verifier backtracking bug"); | |
3263 | return -EFAULT; | |
3264 | } | |
3265 | bt->frame++; | |
3266 | return 0; | |
3267 | } | |
3268 | ||
3269 | static inline int bt_subprog_exit(struct backtrack_state *bt) | |
3270 | { | |
3271 | if (bt->frame == 0) { | |
3272 | verbose(bt->env, "BUG subprog exit from frame 0\n"); | |
3273 | WARN_ONCE(1, "verifier backtracking bug"); | |
3274 | return -EFAULT; | |
3275 | } | |
3276 | bt->frame--; | |
3277 | return 0; | |
3278 | } | |
3279 | ||
3280 | static inline void bt_set_frame_reg(struct backtrack_state *bt, u32 frame, u32 reg) | |
3281 | { | |
3282 | bt->reg_masks[frame] |= 1 << reg; | |
3283 | } | |
3284 | ||
3285 | static inline void bt_clear_frame_reg(struct backtrack_state *bt, u32 frame, u32 reg) | |
3286 | { | |
3287 | bt->reg_masks[frame] &= ~(1 << reg); | |
3288 | } | |
3289 | ||
3290 | static inline void bt_set_reg(struct backtrack_state *bt, u32 reg) | |
3291 | { | |
3292 | bt_set_frame_reg(bt, bt->frame, reg); | |
3293 | } | |
3294 | ||
3295 | static inline void bt_clear_reg(struct backtrack_state *bt, u32 reg) | |
3296 | { | |
3297 | bt_clear_frame_reg(bt, bt->frame, reg); | |
3298 | } | |
3299 | ||
3300 | static inline void bt_set_frame_slot(struct backtrack_state *bt, u32 frame, u32 slot) | |
3301 | { | |
3302 | bt->stack_masks[frame] |= 1ull << slot; | |
3303 | } | |
3304 | ||
3305 | static inline void bt_clear_frame_slot(struct backtrack_state *bt, u32 frame, u32 slot) | |
3306 | { | |
3307 | bt->stack_masks[frame] &= ~(1ull << slot); | |
3308 | } | |
3309 | ||
3310 | static inline void bt_set_slot(struct backtrack_state *bt, u32 slot) | |
3311 | { | |
3312 | bt_set_frame_slot(bt, bt->frame, slot); | |
3313 | } | |
3314 | ||
3315 | static inline void bt_clear_slot(struct backtrack_state *bt, u32 slot) | |
3316 | { | |
3317 | bt_clear_frame_slot(bt, bt->frame, slot); | |
3318 | } | |
3319 | ||
3320 | static inline u32 bt_frame_reg_mask(struct backtrack_state *bt, u32 frame) | |
3321 | { | |
3322 | return bt->reg_masks[frame]; | |
3323 | } | |
3324 | ||
3325 | static inline u32 bt_reg_mask(struct backtrack_state *bt) | |
3326 | { | |
3327 | return bt->reg_masks[bt->frame]; | |
3328 | } | |
3329 | ||
3330 | static inline u64 bt_frame_stack_mask(struct backtrack_state *bt, u32 frame) | |
3331 | { | |
3332 | return bt->stack_masks[frame]; | |
3333 | } | |
3334 | ||
3335 | static inline u64 bt_stack_mask(struct backtrack_state *bt) | |
3336 | { | |
3337 | return bt->stack_masks[bt->frame]; | |
3338 | } | |
3339 | ||
3340 | static inline bool bt_is_reg_set(struct backtrack_state *bt, u32 reg) | |
3341 | { | |
3342 | return bt->reg_masks[bt->frame] & (1 << reg); | |
3343 | } | |
3344 | ||
3345 | static inline bool bt_is_slot_set(struct backtrack_state *bt, u32 slot) | |
3346 | { | |
3347 | return bt->stack_masks[bt->frame] & (1ull << slot); | |
3348 | } | |
3349 | ||
d9439c21 AN |
3350 | /* format registers bitmask, e.g., "r0,r2,r4" for 0x15 mask */ |
3351 | static void fmt_reg_mask(char *buf, ssize_t buf_sz, u32 reg_mask) | |
3352 | { | |
3353 | DECLARE_BITMAP(mask, 64); | |
3354 | bool first = true; | |
3355 | int i, n; | |
3356 | ||
3357 | buf[0] = '\0'; | |
3358 | ||
3359 | bitmap_from_u64(mask, reg_mask); | |
3360 | for_each_set_bit(i, mask, 32) { | |
3361 | n = snprintf(buf, buf_sz, "%sr%d", first ? "" : ",", i); | |
3362 | first = false; | |
3363 | buf += n; | |
3364 | buf_sz -= n; | |
3365 | if (buf_sz < 0) | |
3366 | break; | |
3367 | } | |
3368 | } | |
3369 | /* format stack slots bitmask, e.g., "-8,-24,-40" for 0x15 mask */ | |
3370 | static void fmt_stack_mask(char *buf, ssize_t buf_sz, u64 stack_mask) | |
3371 | { | |
3372 | DECLARE_BITMAP(mask, 64); | |
3373 | bool first = true; | |
3374 | int i, n; | |
3375 | ||
3376 | buf[0] = '\0'; | |
3377 | ||
3378 | bitmap_from_u64(mask, stack_mask); | |
3379 | for_each_set_bit(i, mask, 64) { | |
3380 | n = snprintf(buf, buf_sz, "%s%d", first ? "" : ",", -(i + 1) * 8); | |
3381 | first = false; | |
3382 | buf += n; | |
3383 | buf_sz -= n; | |
3384 | if (buf_sz < 0) | |
3385 | break; | |
3386 | } | |
3387 | } | |
3388 | ||
b5dc0163 AS |
3389 | /* For given verifier state backtrack_insn() is called from the last insn to |
3390 | * the first insn. Its purpose is to compute a bitmask of registers and | |
3391 | * stack slots that needs precision in the parent verifier state. | |
fde2a388 AN |
3392 | * |
3393 | * @idx is an index of the instruction we are currently processing; | |
3394 | * @subseq_idx is an index of the subsequent instruction that: | |
3395 | * - *would be* executed next, if jump history is viewed in forward order; | |
3396 | * - *was* processed previously during backtracking. | |
b5dc0163 | 3397 | */ |
fde2a388 | 3398 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, int subseq_idx, |
407958a0 | 3399 | struct backtrack_state *bt) |
b5dc0163 AS |
3400 | { |
3401 | const struct bpf_insn_cbs cbs = { | |
e6ac2450 | 3402 | .cb_call = disasm_kfunc_name, |
b5dc0163 AS |
3403 | .cb_print = verbose, |
3404 | .private_data = env, | |
3405 | }; | |
3406 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
3407 | u8 class = BPF_CLASS(insn->code); | |
3408 | u8 opcode = BPF_OP(insn->code); | |
3409 | u8 mode = BPF_MODE(insn->code); | |
407958a0 AN |
3410 | u32 dreg = insn->dst_reg; |
3411 | u32 sreg = insn->src_reg; | |
fde2a388 | 3412 | u32 spi, i; |
b5dc0163 AS |
3413 | |
3414 | if (insn->code == 0) | |
3415 | return 0; | |
496f3324 | 3416 | if (env->log.level & BPF_LOG_LEVEL2) { |
d9439c21 AN |
3417 | fmt_reg_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, bt_reg_mask(bt)); |
3418 | verbose(env, "mark_precise: frame%d: regs=%s ", | |
3419 | bt->frame, env->tmp_str_buf); | |
3420 | fmt_stack_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, bt_stack_mask(bt)); | |
3421 | verbose(env, "stack=%s before ", env->tmp_str_buf); | |
b5dc0163 AS |
3422 | verbose(env, "%d: ", idx); |
3423 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
3424 | } | |
3425 | ||
3426 | if (class == BPF_ALU || class == BPF_ALU64) { | |
407958a0 | 3427 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 AS |
3428 | return 0; |
3429 | if (opcode == BPF_MOV) { | |
3430 | if (BPF_SRC(insn->code) == BPF_X) { | |
8100928c | 3431 | /* dreg = sreg or dreg = (s8, s16, s32)sreg |
b5dc0163 AS |
3432 | * dreg needs precision after this insn |
3433 | * sreg needs precision before this insn | |
3434 | */ | |
407958a0 AN |
3435 | bt_clear_reg(bt, dreg); |
3436 | bt_set_reg(bt, sreg); | |
b5dc0163 AS |
3437 | } else { |
3438 | /* dreg = K | |
3439 | * dreg needs precision after this insn. | |
3440 | * Corresponding register is already marked | |
3441 | * as precise=true in this verifier state. | |
3442 | * No further markings in parent are necessary | |
3443 | */ | |
407958a0 | 3444 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
3445 | } |
3446 | } else { | |
3447 | if (BPF_SRC(insn->code) == BPF_X) { | |
3448 | /* dreg += sreg | |
3449 | * both dreg and sreg need precision | |
3450 | * before this insn | |
3451 | */ | |
407958a0 | 3452 | bt_set_reg(bt, sreg); |
b5dc0163 AS |
3453 | } /* else dreg += K |
3454 | * dreg still needs precision before this insn | |
3455 | */ | |
3456 | } | |
3457 | } else if (class == BPF_LDX) { | |
407958a0 | 3458 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 | 3459 | return 0; |
407958a0 | 3460 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
3461 | |
3462 | /* scalars can only be spilled into stack w/o losing precision. | |
3463 | * Load from any other memory can be zero extended. | |
3464 | * The desire to keep that precision is already indicated | |
3465 | * by 'precise' mark in corresponding register of this state. | |
3466 | * No further tracking necessary. | |
3467 | */ | |
3468 | if (insn->src_reg != BPF_REG_FP) | |
3469 | return 0; | |
b5dc0163 AS |
3470 | |
3471 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
3472 | * that [fp - off] slot contains scalar that needs to be | |
3473 | * tracked with precision | |
3474 | */ | |
3475 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
3476 | if (spi >= 64) { | |
3477 | verbose(env, "BUG spi %d\n", spi); | |
3478 | WARN_ONCE(1, "verifier backtracking bug"); | |
3479 | return -EFAULT; | |
3480 | } | |
407958a0 | 3481 | bt_set_slot(bt, spi); |
b3b50f05 | 3482 | } else if (class == BPF_STX || class == BPF_ST) { |
407958a0 | 3483 | if (bt_is_reg_set(bt, dreg)) |
b3b50f05 | 3484 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
3485 | * to access memory. It means backtracking |
3486 | * encountered a case of pointer subtraction. | |
3487 | */ | |
3488 | return -ENOTSUPP; | |
3489 | /* scalars can only be spilled into stack */ | |
3490 | if (insn->dst_reg != BPF_REG_FP) | |
3491 | return 0; | |
b5dc0163 AS |
3492 | spi = (-insn->off - 1) / BPF_REG_SIZE; |
3493 | if (spi >= 64) { | |
3494 | verbose(env, "BUG spi %d\n", spi); | |
3495 | WARN_ONCE(1, "verifier backtracking bug"); | |
3496 | return -EFAULT; | |
3497 | } | |
407958a0 | 3498 | if (!bt_is_slot_set(bt, spi)) |
b5dc0163 | 3499 | return 0; |
407958a0 | 3500 | bt_clear_slot(bt, spi); |
b3b50f05 | 3501 | if (class == BPF_STX) |
407958a0 | 3502 | bt_set_reg(bt, sreg); |
b5dc0163 | 3503 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
fde2a388 AN |
3504 | if (bpf_pseudo_call(insn)) { |
3505 | int subprog_insn_idx, subprog; | |
3506 | ||
3507 | subprog_insn_idx = idx + insn->imm + 1; | |
3508 | subprog = find_subprog(env, subprog_insn_idx); | |
3509 | if (subprog < 0) | |
3510 | return -EFAULT; | |
3511 | ||
3512 | if (subprog_is_global(env, subprog)) { | |
3513 | /* check that jump history doesn't have any | |
3514 | * extra instructions from subprog; the next | |
3515 | * instruction after call to global subprog | |
3516 | * should be literally next instruction in | |
3517 | * caller program | |
3518 | */ | |
3519 | WARN_ONCE(idx + 1 != subseq_idx, "verifier backtracking bug"); | |
3520 | /* r1-r5 are invalidated after subprog call, | |
3521 | * so for global func call it shouldn't be set | |
3522 | * anymore | |
3523 | */ | |
3524 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { | |
3525 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3526 | WARN_ONCE(1, "verifier backtracking bug"); | |
3527 | return -EFAULT; | |
3528 | } | |
3529 | /* global subprog always sets R0 */ | |
3530 | bt_clear_reg(bt, BPF_REG_0); | |
3531 | return 0; | |
3532 | } else { | |
3533 | /* static subprog call instruction, which | |
3534 | * means that we are exiting current subprog, | |
3535 | * so only r1-r5 could be still requested as | |
3536 | * precise, r0 and r6-r10 or any stack slot in | |
3537 | * the current frame should be zero by now | |
3538 | */ | |
3539 | if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) { | |
3540 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3541 | WARN_ONCE(1, "verifier backtracking bug"); | |
3542 | return -EFAULT; | |
3543 | } | |
3544 | /* we don't track register spills perfectly, | |
3545 | * so fallback to force-precise instead of failing */ | |
3546 | if (bt_stack_mask(bt) != 0) | |
3547 | return -ENOTSUPP; | |
3548 | /* propagate r1-r5 to the caller */ | |
3549 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
3550 | if (bt_is_reg_set(bt, i)) { | |
3551 | bt_clear_reg(bt, i); | |
3552 | bt_set_frame_reg(bt, bt->frame - 1, i); | |
3553 | } | |
3554 | } | |
3555 | if (bt_subprog_exit(bt)) | |
3556 | return -EFAULT; | |
3557 | return 0; | |
3558 | } | |
3559 | } else if ((bpf_helper_call(insn) && | |
3560 | is_callback_calling_function(insn->imm) && | |
3561 | !is_async_callback_calling_function(insn->imm)) || | |
3562 | (bpf_pseudo_kfunc_call(insn) && is_callback_calling_kfunc(insn->imm))) { | |
3563 | /* callback-calling helper or kfunc call, which means | |
3564 | * we are exiting from subprog, but unlike the subprog | |
3565 | * call handling above, we shouldn't propagate | |
3566 | * precision of r1-r5 (if any requested), as they are | |
3567 | * not actually arguments passed directly to callback | |
3568 | * subprogs | |
be2ef816 | 3569 | */ |
fde2a388 AN |
3570 | if (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) { |
3571 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3572 | WARN_ONCE(1, "verifier backtracking bug"); | |
3573 | return -EFAULT; | |
3574 | } | |
3575 | if (bt_stack_mask(bt) != 0) | |
be2ef816 | 3576 | return -ENOTSUPP; |
fde2a388 AN |
3577 | /* clear r1-r5 in callback subprog's mask */ |
3578 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
3579 | bt_clear_reg(bt, i); | |
3580 | if (bt_subprog_exit(bt)) | |
3581 | return -EFAULT; | |
3582 | return 0; | |
3583 | } else if (opcode == BPF_CALL) { | |
d3178e8a HS |
3584 | /* kfunc with imm==0 is invalid and fixup_kfunc_call will |
3585 | * catch this error later. Make backtracking conservative | |
3586 | * with ENOTSUPP. | |
3587 | */ | |
3588 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL && insn->imm == 0) | |
3589 | return -ENOTSUPP; | |
b5dc0163 | 3590 | /* regular helper call sets R0 */ |
407958a0 AN |
3591 | bt_clear_reg(bt, BPF_REG_0); |
3592 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { | |
b5dc0163 AS |
3593 | /* if backtracing was looking for registers R1-R5 |
3594 | * they should have been found already. | |
3595 | */ | |
407958a0 | 3596 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); |
b5dc0163 AS |
3597 | WARN_ONCE(1, "verifier backtracking bug"); |
3598 | return -EFAULT; | |
3599 | } | |
3600 | } else if (opcode == BPF_EXIT) { | |
fde2a388 AN |
3601 | bool r0_precise; |
3602 | ||
3603 | if (bt_reg_mask(bt) & BPF_REGMASK_ARGS) { | |
3604 | /* if backtracing was looking for registers R1-R5 | |
3605 | * they should have been found already. | |
3606 | */ | |
3607 | verbose(env, "BUG regs %x\n", bt_reg_mask(bt)); | |
3608 | WARN_ONCE(1, "verifier backtracking bug"); | |
3609 | return -EFAULT; | |
3610 | } | |
3611 | ||
3612 | /* BPF_EXIT in subprog or callback always returns | |
3613 | * right after the call instruction, so by checking | |
3614 | * whether the instruction at subseq_idx-1 is subprog | |
3615 | * call or not we can distinguish actual exit from | |
3616 | * *subprog* from exit from *callback*. In the former | |
3617 | * case, we need to propagate r0 precision, if | |
3618 | * necessary. In the former we never do that. | |
3619 | */ | |
3620 | r0_precise = subseq_idx - 1 >= 0 && | |
3621 | bpf_pseudo_call(&env->prog->insnsi[subseq_idx - 1]) && | |
3622 | bt_is_reg_set(bt, BPF_REG_0); | |
3623 | ||
3624 | bt_clear_reg(bt, BPF_REG_0); | |
3625 | if (bt_subprog_enter(bt)) | |
3626 | return -EFAULT; | |
3627 | ||
3628 | if (r0_precise) | |
3629 | bt_set_reg(bt, BPF_REG_0); | |
3630 | /* r6-r9 and stack slots will stay set in caller frame | |
3631 | * bitmasks until we return back from callee(s) | |
3632 | */ | |
3633 | return 0; | |
71b547f5 | 3634 | } else if (BPF_SRC(insn->code) == BPF_X) { |
407958a0 | 3635 | if (!bt_is_reg_set(bt, dreg) && !bt_is_reg_set(bt, sreg)) |
71b547f5 DB |
3636 | return 0; |
3637 | /* dreg <cond> sreg | |
3638 | * Both dreg and sreg need precision before | |
3639 | * this insn. If only sreg was marked precise | |
3640 | * before it would be equally necessary to | |
3641 | * propagate it to dreg. | |
3642 | */ | |
407958a0 AN |
3643 | bt_set_reg(bt, dreg); |
3644 | bt_set_reg(bt, sreg); | |
71b547f5 DB |
3645 | /* else dreg <cond> K |
3646 | * Only dreg still needs precision before | |
3647 | * this insn, so for the K-based conditional | |
3648 | * there is nothing new to be marked. | |
3649 | */ | |
b5dc0163 AS |
3650 | } |
3651 | } else if (class == BPF_LD) { | |
407958a0 | 3652 | if (!bt_is_reg_set(bt, dreg)) |
b5dc0163 | 3653 | return 0; |
407958a0 | 3654 | bt_clear_reg(bt, dreg); |
b5dc0163 AS |
3655 | /* It's ld_imm64 or ld_abs or ld_ind. |
3656 | * For ld_imm64 no further tracking of precision | |
3657 | * into parent is necessary | |
3658 | */ | |
3659 | if (mode == BPF_IND || mode == BPF_ABS) | |
3660 | /* to be analyzed */ | |
3661 | return -ENOTSUPP; | |
b5dc0163 AS |
3662 | } |
3663 | return 0; | |
3664 | } | |
3665 | ||
3666 | /* the scalar precision tracking algorithm: | |
3667 | * . at the start all registers have precise=false. | |
3668 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
3669 | * . once precise value of the scalar register is used in: | |
3670 | * . ptr + scalar alu | |
3671 | * . if (scalar cond K|scalar) | |
3672 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
3673 | * backtrack through the verifier states and mark all registers and | |
3674 | * stack slots with spilled constants that these scalar regisers | |
3675 | * should be precise. | |
3676 | * . during state pruning two registers (or spilled stack slots) | |
3677 | * are equivalent if both are not precise. | |
3678 | * | |
3679 | * Note the verifier cannot simply walk register parentage chain, | |
3680 | * since many different registers and stack slots could have been | |
3681 | * used to compute single precise scalar. | |
3682 | * | |
3683 | * The approach of starting with precise=true for all registers and then | |
3684 | * backtrack to mark a register as not precise when the verifier detects | |
3685 | * that program doesn't care about specific value (e.g., when helper | |
3686 | * takes register as ARG_ANYTHING parameter) is not safe. | |
3687 | * | |
3688 | * It's ok to walk single parentage chain of the verifier states. | |
3689 | * It's possible that this backtracking will go all the way till 1st insn. | |
3690 | * All other branches will be explored for needing precision later. | |
3691 | * | |
3692 | * The backtracking needs to deal with cases like: | |
3693 | * 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) | |
3694 | * r9 -= r8 | |
3695 | * r5 = r9 | |
3696 | * if r5 > 0x79f goto pc+7 | |
3697 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
3698 | * r5 += 1 | |
3699 | * ... | |
3700 | * call bpf_perf_event_output#25 | |
3701 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
3702 | * | |
3703 | * and this case: | |
3704 | * r6 = 1 | |
3705 | * call foo // uses callee's r6 inside to compute r0 | |
3706 | * r0 += r6 | |
3707 | * if r0 == 0 goto | |
3708 | * | |
3709 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
3710 | * | |
3711 | * Also if parent's curframe > frame where backtracking started, | |
3712 | * the verifier need to mark registers in both frames, otherwise callees | |
3713 | * may incorrectly prune callers. This is similar to | |
3714 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
3715 | * | |
3716 | * For now backtracking falls back into conservative marking. | |
3717 | */ | |
3718 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
3719 | struct bpf_verifier_state *st) | |
3720 | { | |
3721 | struct bpf_func_state *func; | |
3722 | struct bpf_reg_state *reg; | |
3723 | int i, j; | |
3724 | ||
d9439c21 AN |
3725 | if (env->log.level & BPF_LOG_LEVEL2) { |
3726 | verbose(env, "mark_precise: frame%d: falling back to forcing all scalars precise\n", | |
3727 | st->curframe); | |
3728 | } | |
3729 | ||
b5dc0163 AS |
3730 | /* big hammer: mark all scalars precise in this path. |
3731 | * pop_stack may still get !precise scalars. | |
f63181b6 AN |
3732 | * We also skip current state and go straight to first parent state, |
3733 | * because precision markings in current non-checkpointed state are | |
3734 | * not needed. See why in the comment in __mark_chain_precision below. | |
b5dc0163 | 3735 | */ |
f63181b6 | 3736 | for (st = st->parent; st; st = st->parent) { |
b5dc0163 AS |
3737 | for (i = 0; i <= st->curframe; i++) { |
3738 | func = st->frame[i]; | |
3739 | for (j = 0; j < BPF_REG_FP; j++) { | |
3740 | reg = &func->regs[j]; | |
d9439c21 | 3741 | if (reg->type != SCALAR_VALUE || reg->precise) |
b5dc0163 AS |
3742 | continue; |
3743 | reg->precise = true; | |
d9439c21 AN |
3744 | if (env->log.level & BPF_LOG_LEVEL2) { |
3745 | verbose(env, "force_precise: frame%d: forcing r%d to be precise\n", | |
3746 | i, j); | |
3747 | } | |
b5dc0163 AS |
3748 | } |
3749 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
27113c59 | 3750 | if (!is_spilled_reg(&func->stack[j])) |
b5dc0163 AS |
3751 | continue; |
3752 | reg = &func->stack[j].spilled_ptr; | |
d9439c21 | 3753 | if (reg->type != SCALAR_VALUE || reg->precise) |
b5dc0163 AS |
3754 | continue; |
3755 | reg->precise = true; | |
d9439c21 AN |
3756 | if (env->log.level & BPF_LOG_LEVEL2) { |
3757 | verbose(env, "force_precise: frame%d: forcing fp%d to be precise\n", | |
3758 | i, -(j + 1) * 8); | |
3759 | } | |
b5dc0163 AS |
3760 | } |
3761 | } | |
f63181b6 | 3762 | } |
b5dc0163 AS |
3763 | } |
3764 | ||
7a830b53 AN |
3765 | static void mark_all_scalars_imprecise(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
3766 | { | |
3767 | struct bpf_func_state *func; | |
3768 | struct bpf_reg_state *reg; | |
3769 | int i, j; | |
3770 | ||
3771 | for (i = 0; i <= st->curframe; i++) { | |
3772 | func = st->frame[i]; | |
3773 | for (j = 0; j < BPF_REG_FP; j++) { | |
3774 | reg = &func->regs[j]; | |
3775 | if (reg->type != SCALAR_VALUE) | |
3776 | continue; | |
3777 | reg->precise = false; | |
3778 | } | |
3779 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
3780 | if (!is_spilled_reg(&func->stack[j])) | |
3781 | continue; | |
3782 | reg = &func->stack[j].spilled_ptr; | |
3783 | if (reg->type != SCALAR_VALUE) | |
3784 | continue; | |
3785 | reg->precise = false; | |
3786 | } | |
3787 | } | |
3788 | } | |
3789 | ||
904e6ddf EZ |
3790 | static bool idset_contains(struct bpf_idset *s, u32 id) |
3791 | { | |
3792 | u32 i; | |
3793 | ||
3794 | for (i = 0; i < s->count; ++i) | |
3795 | if (s->ids[i] == id) | |
3796 | return true; | |
3797 | ||
3798 | return false; | |
3799 | } | |
3800 | ||
3801 | static int idset_push(struct bpf_idset *s, u32 id) | |
3802 | { | |
3803 | if (WARN_ON_ONCE(s->count >= ARRAY_SIZE(s->ids))) | |
3804 | return -EFAULT; | |
3805 | s->ids[s->count++] = id; | |
3806 | return 0; | |
3807 | } | |
3808 | ||
3809 | static void idset_reset(struct bpf_idset *s) | |
3810 | { | |
3811 | s->count = 0; | |
3812 | } | |
3813 | ||
3814 | /* Collect a set of IDs for all registers currently marked as precise in env->bt. | |
3815 | * Mark all registers with these IDs as precise. | |
3816 | */ | |
3817 | static int mark_precise_scalar_ids(struct bpf_verifier_env *env, struct bpf_verifier_state *st) | |
3818 | { | |
3819 | struct bpf_idset *precise_ids = &env->idset_scratch; | |
3820 | struct backtrack_state *bt = &env->bt; | |
3821 | struct bpf_func_state *func; | |
3822 | struct bpf_reg_state *reg; | |
3823 | DECLARE_BITMAP(mask, 64); | |
3824 | int i, fr; | |
3825 | ||
3826 | idset_reset(precise_ids); | |
3827 | ||
3828 | for (fr = bt->frame; fr >= 0; fr--) { | |
3829 | func = st->frame[fr]; | |
3830 | ||
3831 | bitmap_from_u64(mask, bt_frame_reg_mask(bt, fr)); | |
3832 | for_each_set_bit(i, mask, 32) { | |
3833 | reg = &func->regs[i]; | |
3834 | if (!reg->id || reg->type != SCALAR_VALUE) | |
3835 | continue; | |
3836 | if (idset_push(precise_ids, reg->id)) | |
3837 | return -EFAULT; | |
3838 | } | |
3839 | ||
3840 | bitmap_from_u64(mask, bt_frame_stack_mask(bt, fr)); | |
3841 | for_each_set_bit(i, mask, 64) { | |
3842 | if (i >= func->allocated_stack / BPF_REG_SIZE) | |
3843 | break; | |
3844 | if (!is_spilled_scalar_reg(&func->stack[i])) | |
3845 | continue; | |
3846 | reg = &func->stack[i].spilled_ptr; | |
3847 | if (!reg->id) | |
3848 | continue; | |
3849 | if (idset_push(precise_ids, reg->id)) | |
3850 | return -EFAULT; | |
3851 | } | |
3852 | } | |
3853 | ||
3854 | for (fr = 0; fr <= st->curframe; ++fr) { | |
3855 | func = st->frame[fr]; | |
3856 | ||
3857 | for (i = BPF_REG_0; i < BPF_REG_10; ++i) { | |
3858 | reg = &func->regs[i]; | |
3859 | if (!reg->id) | |
3860 | continue; | |
3861 | if (!idset_contains(precise_ids, reg->id)) | |
3862 | continue; | |
3863 | bt_set_frame_reg(bt, fr, i); | |
3864 | } | |
3865 | for (i = 0; i < func->allocated_stack / BPF_REG_SIZE; ++i) { | |
3866 | if (!is_spilled_scalar_reg(&func->stack[i])) | |
3867 | continue; | |
3868 | reg = &func->stack[i].spilled_ptr; | |
3869 | if (!reg->id) | |
3870 | continue; | |
3871 | if (!idset_contains(precise_ids, reg->id)) | |
3872 | continue; | |
3873 | bt_set_frame_slot(bt, fr, i); | |
3874 | } | |
3875 | } | |
3876 | ||
3877 | return 0; | |
3878 | } | |
3879 | ||
f63181b6 AN |
3880 | /* |
3881 | * __mark_chain_precision() backtracks BPF program instruction sequence and | |
3882 | * chain of verifier states making sure that register *regno* (if regno >= 0) | |
3883 | * and/or stack slot *spi* (if spi >= 0) are marked as precisely tracked | |
3884 | * SCALARS, as well as any other registers and slots that contribute to | |
3885 | * a tracked state of given registers/stack slots, depending on specific BPF | |
3886 | * assembly instructions (see backtrack_insns() for exact instruction handling | |
3887 | * logic). This backtracking relies on recorded jmp_history and is able to | |
3888 | * traverse entire chain of parent states. This process ends only when all the | |
3889 | * necessary registers/slots and their transitive dependencies are marked as | |
3890 | * precise. | |
3891 | * | |
3892 | * One important and subtle aspect is that precise marks *do not matter* in | |
3893 | * the currently verified state (current state). It is important to understand | |
3894 | * why this is the case. | |
3895 | * | |
3896 | * First, note that current state is the state that is not yet "checkpointed", | |
3897 | * i.e., it is not yet put into env->explored_states, and it has no children | |
3898 | * states as well. It's ephemeral, and can end up either a) being discarded if | |
3899 | * compatible explored state is found at some point or BPF_EXIT instruction is | |
3900 | * reached or b) checkpointed and put into env->explored_states, branching out | |
3901 | * into one or more children states. | |
3902 | * | |
3903 | * In the former case, precise markings in current state are completely | |
3904 | * ignored by state comparison code (see regsafe() for details). Only | |
3905 | * checkpointed ("old") state precise markings are important, and if old | |
3906 | * state's register/slot is precise, regsafe() assumes current state's | |
3907 | * register/slot as precise and checks value ranges exactly and precisely. If | |
3908 | * states turn out to be compatible, current state's necessary precise | |
3909 | * markings and any required parent states' precise markings are enforced | |
3910 | * after the fact with propagate_precision() logic, after the fact. But it's | |
3911 | * important to realize that in this case, even after marking current state | |
3912 | * registers/slots as precise, we immediately discard current state. So what | |
3913 | * actually matters is any of the precise markings propagated into current | |
3914 | * state's parent states, which are always checkpointed (due to b) case above). | |
3915 | * As such, for scenario a) it doesn't matter if current state has precise | |
3916 | * markings set or not. | |
3917 | * | |
3918 | * Now, for the scenario b), checkpointing and forking into child(ren) | |
3919 | * state(s). Note that before current state gets to checkpointing step, any | |
3920 | * processed instruction always assumes precise SCALAR register/slot | |
3921 | * knowledge: if precise value or range is useful to prune jump branch, BPF | |
3922 | * verifier takes this opportunity enthusiastically. Similarly, when | |
3923 | * register's value is used to calculate offset or memory address, exact | |
3924 | * knowledge of SCALAR range is assumed, checked, and enforced. So, similar to | |
3925 | * what we mentioned above about state comparison ignoring precise markings | |
3926 | * during state comparison, BPF verifier ignores and also assumes precise | |
3927 | * markings *at will* during instruction verification process. But as verifier | |
3928 | * assumes precision, it also propagates any precision dependencies across | |
3929 | * parent states, which are not yet finalized, so can be further restricted | |
3930 | * based on new knowledge gained from restrictions enforced by their children | |
3931 | * states. This is so that once those parent states are finalized, i.e., when | |
3932 | * they have no more active children state, state comparison logic in | |
3933 | * is_state_visited() would enforce strict and precise SCALAR ranges, if | |
3934 | * required for correctness. | |
3935 | * | |
3936 | * To build a bit more intuition, note also that once a state is checkpointed, | |
3937 | * the path we took to get to that state is not important. This is crucial | |
3938 | * property for state pruning. When state is checkpointed and finalized at | |
3939 | * some instruction index, it can be correctly and safely used to "short | |
3940 | * circuit" any *compatible* state that reaches exactly the same instruction | |
3941 | * index. I.e., if we jumped to that instruction from a completely different | |
3942 | * code path than original finalized state was derived from, it doesn't | |
3943 | * matter, current state can be discarded because from that instruction | |
3944 | * forward having a compatible state will ensure we will safely reach the | |
3945 | * exit. States describe preconditions for further exploration, but completely | |
3946 | * forget the history of how we got here. | |
3947 | * | |
3948 | * This also means that even if we needed precise SCALAR range to get to | |
3949 | * finalized state, but from that point forward *that same* SCALAR register is | |
3950 | * never used in a precise context (i.e., it's precise value is not needed for | |
3951 | * correctness), it's correct and safe to mark such register as "imprecise" | |
3952 | * (i.e., precise marking set to false). This is what we rely on when we do | |
3953 | * not set precise marking in current state. If no child state requires | |
3954 | * precision for any given SCALAR register, it's safe to dictate that it can | |
3955 | * be imprecise. If any child state does require this register to be precise, | |
3956 | * we'll mark it precise later retroactively during precise markings | |
3957 | * propagation from child state to parent states. | |
7a830b53 AN |
3958 | * |
3959 | * Skipping precise marking setting in current state is a mild version of | |
3960 | * relying on the above observation. But we can utilize this property even | |
3961 | * more aggressively by proactively forgetting any precise marking in the | |
3962 | * current state (which we inherited from the parent state), right before we | |
3963 | * checkpoint it and branch off into new child state. This is done by | |
3964 | * mark_all_scalars_imprecise() to hopefully get more permissive and generic | |
3965 | * finalized states which help in short circuiting more future states. | |
f63181b6 | 3966 | */ |
f655badf | 3967 | static int __mark_chain_precision(struct bpf_verifier_env *env, int regno) |
b5dc0163 | 3968 | { |
407958a0 | 3969 | struct backtrack_state *bt = &env->bt; |
b5dc0163 AS |
3970 | struct bpf_verifier_state *st = env->cur_state; |
3971 | int first_idx = st->first_insn_idx; | |
3972 | int last_idx = env->insn_idx; | |
d84b1a67 | 3973 | int subseq_idx = -1; |
b5dc0163 AS |
3974 | struct bpf_func_state *func; |
3975 | struct bpf_reg_state *reg; | |
b5dc0163 | 3976 | bool skip_first = true; |
d84b1a67 | 3977 | int i, fr, err; |
b5dc0163 | 3978 | |
2c78ee89 | 3979 | if (!env->bpf_capable) |
b5dc0163 AS |
3980 | return 0; |
3981 | ||
407958a0 | 3982 | /* set frame number from which we are starting to backtrack */ |
f655badf | 3983 | bt_init(bt, env->cur_state->curframe); |
407958a0 | 3984 | |
f63181b6 AN |
3985 | /* Do sanity checks against current state of register and/or stack |
3986 | * slot, but don't set precise flag in current state, as precision | |
3987 | * tracking in the current state is unnecessary. | |
3988 | */ | |
f655badf | 3989 | func = st->frame[bt->frame]; |
a3ce685d AS |
3990 | if (regno >= 0) { |
3991 | reg = &func->regs[regno]; | |
3992 | if (reg->type != SCALAR_VALUE) { | |
3993 | WARN_ONCE(1, "backtracing misuse"); | |
3994 | return -EFAULT; | |
3995 | } | |
407958a0 | 3996 | bt_set_reg(bt, regno); |
b5dc0163 | 3997 | } |
b5dc0163 | 3998 | |
407958a0 | 3999 | if (bt_empty(bt)) |
a3ce685d | 4000 | return 0; |
be2ef816 | 4001 | |
b5dc0163 AS |
4002 | for (;;) { |
4003 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
4004 | u32 history = st->jmp_history_cnt; |
4005 | ||
d9439c21 | 4006 | if (env->log.level & BPF_LOG_LEVEL2) { |
d84b1a67 AN |
4007 | verbose(env, "mark_precise: frame%d: last_idx %d first_idx %d subseq_idx %d \n", |
4008 | bt->frame, last_idx, first_idx, subseq_idx); | |
d9439c21 | 4009 | } |
be2ef816 | 4010 | |
904e6ddf EZ |
4011 | /* If some register with scalar ID is marked as precise, |
4012 | * make sure that all registers sharing this ID are also precise. | |
4013 | * This is needed to estimate effect of find_equal_scalars(). | |
4014 | * Do this at the last instruction of each state, | |
4015 | * bpf_reg_state::id fields are valid for these instructions. | |
4016 | * | |
4017 | * Allows to track precision in situation like below: | |
4018 | * | |
4019 | * r2 = unknown value | |
4020 | * ... | |
4021 | * --- state #0 --- | |
4022 | * ... | |
4023 | * r1 = r2 // r1 and r2 now share the same ID | |
4024 | * ... | |
4025 | * --- state #1 {r1.id = A, r2.id = A} --- | |
4026 | * ... | |
4027 | * if (r2 > 10) goto exit; // find_equal_scalars() assigns range to r1 | |
4028 | * ... | |
4029 | * --- state #2 {r1.id = A, r2.id = A} --- | |
4030 | * r3 = r10 | |
4031 | * r3 += r1 // need to mark both r1 and r2 | |
4032 | */ | |
4033 | if (mark_precise_scalar_ids(env, st)) | |
4034 | return -EFAULT; | |
4035 | ||
be2ef816 AN |
4036 | if (last_idx < 0) { |
4037 | /* we are at the entry into subprog, which | |
4038 | * is expected for global funcs, but only if | |
4039 | * requested precise registers are R1-R5 | |
4040 | * (which are global func's input arguments) | |
4041 | */ | |
4042 | if (st->curframe == 0 && | |
4043 | st->frame[0]->subprogno > 0 && | |
4044 | st->frame[0]->callsite == BPF_MAIN_FUNC && | |
407958a0 AN |
4045 | bt_stack_mask(bt) == 0 && |
4046 | (bt_reg_mask(bt) & ~BPF_REGMASK_ARGS) == 0) { | |
4047 | bitmap_from_u64(mask, bt_reg_mask(bt)); | |
be2ef816 AN |
4048 | for_each_set_bit(i, mask, 32) { |
4049 | reg = &st->frame[0]->regs[i]; | |
4050 | if (reg->type != SCALAR_VALUE) { | |
407958a0 | 4051 | bt_clear_reg(bt, i); |
be2ef816 AN |
4052 | continue; |
4053 | } | |
4054 | reg->precise = true; | |
4055 | } | |
4056 | return 0; | |
4057 | } | |
4058 | ||
407958a0 AN |
4059 | verbose(env, "BUG backtracking func entry subprog %d reg_mask %x stack_mask %llx\n", |
4060 | st->frame[0]->subprogno, bt_reg_mask(bt), bt_stack_mask(bt)); | |
be2ef816 AN |
4061 | WARN_ONCE(1, "verifier backtracking bug"); |
4062 | return -EFAULT; | |
4063 | } | |
4064 | ||
d84b1a67 | 4065 | for (i = last_idx;;) { |
b5dc0163 AS |
4066 | if (skip_first) { |
4067 | err = 0; | |
4068 | skip_first = false; | |
4069 | } else { | |
d84b1a67 | 4070 | err = backtrack_insn(env, i, subseq_idx, bt); |
b5dc0163 AS |
4071 | } |
4072 | if (err == -ENOTSUPP) { | |
c50c0b57 | 4073 | mark_all_scalars_precise(env, env->cur_state); |
407958a0 | 4074 | bt_reset(bt); |
b5dc0163 AS |
4075 | return 0; |
4076 | } else if (err) { | |
4077 | return err; | |
4078 | } | |
407958a0 | 4079 | if (bt_empty(bt)) |
b5dc0163 AS |
4080 | /* Found assignment(s) into tracked register in this state. |
4081 | * Since this state is already marked, just return. | |
4082 | * Nothing to be tracked further in the parent state. | |
4083 | */ | |
4084 | return 0; | |
4085 | if (i == first_idx) | |
4086 | break; | |
d84b1a67 | 4087 | subseq_idx = i; |
b5dc0163 AS |
4088 | i = get_prev_insn_idx(st, i, &history); |
4089 | if (i >= env->prog->len) { | |
4090 | /* This can happen if backtracking reached insn 0 | |
4091 | * and there are still reg_mask or stack_mask | |
4092 | * to backtrack. | |
4093 | * It means the backtracking missed the spot where | |
4094 | * particular register was initialized with a constant. | |
4095 | */ | |
4096 | verbose(env, "BUG backtracking idx %d\n", i); | |
4097 | WARN_ONCE(1, "verifier backtracking bug"); | |
4098 | return -EFAULT; | |
4099 | } | |
4100 | } | |
4101 | st = st->parent; | |
4102 | if (!st) | |
4103 | break; | |
4104 | ||
1ef22b68 AN |
4105 | for (fr = bt->frame; fr >= 0; fr--) { |
4106 | func = st->frame[fr]; | |
4107 | bitmap_from_u64(mask, bt_frame_reg_mask(bt, fr)); | |
4108 | for_each_set_bit(i, mask, 32) { | |
4109 | reg = &func->regs[i]; | |
4110 | if (reg->type != SCALAR_VALUE) { | |
4111 | bt_clear_frame_reg(bt, fr, i); | |
4112 | continue; | |
4113 | } | |
4114 | if (reg->precise) | |
4115 | bt_clear_frame_reg(bt, fr, i); | |
4116 | else | |
4117 | reg->precise = true; | |
a3ce685d | 4118 | } |
b5dc0163 | 4119 | |
1ef22b68 AN |
4120 | bitmap_from_u64(mask, bt_frame_stack_mask(bt, fr)); |
4121 | for_each_set_bit(i, mask, 64) { | |
4122 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
4123 | /* the sequence of instructions: | |
4124 | * 2: (bf) r3 = r10 | |
4125 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
4126 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
4127 | * doesn't contain jmps. It's backtracked | |
4128 | * as a single block. | |
4129 | * During backtracking insn 3 is not recognized as | |
4130 | * stack access, so at the end of backtracking | |
4131 | * stack slot fp-8 is still marked in stack_mask. | |
4132 | * However the parent state may not have accessed | |
4133 | * fp-8 and it's "unallocated" stack space. | |
4134 | * In such case fallback to conservative. | |
4135 | */ | |
c50c0b57 | 4136 | mark_all_scalars_precise(env, env->cur_state); |
1ef22b68 AN |
4137 | bt_reset(bt); |
4138 | return 0; | |
4139 | } | |
b5dc0163 | 4140 | |
1ef22b68 AN |
4141 | if (!is_spilled_scalar_reg(&func->stack[i])) { |
4142 | bt_clear_frame_slot(bt, fr, i); | |
4143 | continue; | |
4144 | } | |
4145 | reg = &func->stack[i].spilled_ptr; | |
4146 | if (reg->precise) | |
4147 | bt_clear_frame_slot(bt, fr, i); | |
4148 | else | |
4149 | reg->precise = true; | |
4150 | } | |
4151 | if (env->log.level & BPF_LOG_LEVEL2) { | |
4152 | fmt_reg_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, | |
4153 | bt_frame_reg_mask(bt, fr)); | |
4154 | verbose(env, "mark_precise: frame%d: parent state regs=%s ", | |
4155 | fr, env->tmp_str_buf); | |
4156 | fmt_stack_mask(env->tmp_str_buf, TMP_STR_BUF_LEN, | |
4157 | bt_frame_stack_mask(bt, fr)); | |
4158 | verbose(env, "stack=%s: ", env->tmp_str_buf); | |
4159 | print_verifier_state(env, func, true); | |
a3ce685d | 4160 | } |
b5dc0163 AS |
4161 | } |
4162 | ||
407958a0 | 4163 | if (bt_empty(bt)) |
c50c0b57 | 4164 | return 0; |
b5dc0163 | 4165 | |
d84b1a67 | 4166 | subseq_idx = first_idx; |
b5dc0163 AS |
4167 | last_idx = st->last_insn_idx; |
4168 | first_idx = st->first_insn_idx; | |
4169 | } | |
c50c0b57 AN |
4170 | |
4171 | /* if we still have requested precise regs or slots, we missed | |
4172 | * something (e.g., stack access through non-r10 register), so | |
4173 | * fallback to marking all precise | |
4174 | */ | |
4175 | if (!bt_empty(bt)) { | |
4176 | mark_all_scalars_precise(env, env->cur_state); | |
4177 | bt_reset(bt); | |
4178 | } | |
4179 | ||
b5dc0163 AS |
4180 | return 0; |
4181 | } | |
4182 | ||
eb1f7f71 | 4183 | int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
a3ce685d | 4184 | { |
f655badf | 4185 | return __mark_chain_precision(env, regno); |
a3ce685d AS |
4186 | } |
4187 | ||
f655badf AN |
4188 | /* mark_chain_precision_batch() assumes that env->bt is set in the caller to |
4189 | * desired reg and stack masks across all relevant frames | |
4190 | */ | |
4191 | static int mark_chain_precision_batch(struct bpf_verifier_env *env) | |
a3ce685d | 4192 | { |
f655badf | 4193 | return __mark_chain_precision(env, -1); |
a3ce685d | 4194 | } |
b5dc0163 | 4195 | |
1be7f75d AS |
4196 | static bool is_spillable_regtype(enum bpf_reg_type type) |
4197 | { | |
c25b2ae1 | 4198 | switch (base_type(type)) { |
1be7f75d | 4199 | case PTR_TO_MAP_VALUE: |
1be7f75d AS |
4200 | case PTR_TO_STACK: |
4201 | case PTR_TO_CTX: | |
969bf05e | 4202 | case PTR_TO_PACKET: |
de8f3a83 | 4203 | case PTR_TO_PACKET_META: |
969bf05e | 4204 | case PTR_TO_PACKET_END: |
d58e468b | 4205 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 4206 | case CONST_PTR_TO_MAP: |
c64b7983 | 4207 | case PTR_TO_SOCKET: |
46f8bc92 | 4208 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 4209 | case PTR_TO_TCP_SOCK: |
fada7fdc | 4210 | case PTR_TO_XDP_SOCK: |
65726b5b | 4211 | case PTR_TO_BTF_ID: |
20b2aff4 | 4212 | case PTR_TO_BUF: |
744ea4e3 | 4213 | case PTR_TO_MEM: |
69c087ba YS |
4214 | case PTR_TO_FUNC: |
4215 | case PTR_TO_MAP_KEY: | |
1be7f75d AS |
4216 | return true; |
4217 | default: | |
4218 | return false; | |
4219 | } | |
4220 | } | |
4221 | ||
cc2b14d5 AS |
4222 | /* Does this register contain a constant zero? */ |
4223 | static bool register_is_null(struct bpf_reg_state *reg) | |
4224 | { | |
4225 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
4226 | } | |
4227 | ||
f7cf25b2 AS |
4228 | static bool register_is_const(struct bpf_reg_state *reg) |
4229 | { | |
4230 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
4231 | } | |
4232 | ||
5689d49b YS |
4233 | static bool __is_scalar_unbounded(struct bpf_reg_state *reg) |
4234 | { | |
4235 | return tnum_is_unknown(reg->var_off) && | |
4236 | reg->smin_value == S64_MIN && reg->smax_value == S64_MAX && | |
4237 | reg->umin_value == 0 && reg->umax_value == U64_MAX && | |
4238 | reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX && | |
4239 | reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX; | |
4240 | } | |
4241 | ||
4242 | static bool register_is_bounded(struct bpf_reg_state *reg) | |
4243 | { | |
4244 | return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg); | |
4245 | } | |
4246 | ||
6e7e63cb JH |
4247 | static bool __is_pointer_value(bool allow_ptr_leaks, |
4248 | const struct bpf_reg_state *reg) | |
4249 | { | |
4250 | if (allow_ptr_leaks) | |
4251 | return false; | |
4252 | ||
4253 | return reg->type != SCALAR_VALUE; | |
4254 | } | |
4255 | ||
71f656a5 EZ |
4256 | /* Copy src state preserving dst->parent and dst->live fields */ |
4257 | static void copy_register_state(struct bpf_reg_state *dst, const struct bpf_reg_state *src) | |
4258 | { | |
4259 | struct bpf_reg_state *parent = dst->parent; | |
4260 | enum bpf_reg_liveness live = dst->live; | |
4261 | ||
4262 | *dst = *src; | |
4263 | dst->parent = parent; | |
4264 | dst->live = live; | |
4265 | } | |
4266 | ||
f7cf25b2 | 4267 | static void save_register_state(struct bpf_func_state *state, |
354e8f19 MKL |
4268 | int spi, struct bpf_reg_state *reg, |
4269 | int size) | |
f7cf25b2 AS |
4270 | { |
4271 | int i; | |
4272 | ||
71f656a5 | 4273 | copy_register_state(&state->stack[spi].spilled_ptr, reg); |
354e8f19 MKL |
4274 | if (size == BPF_REG_SIZE) |
4275 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
f7cf25b2 | 4276 | |
354e8f19 MKL |
4277 | for (i = BPF_REG_SIZE; i > BPF_REG_SIZE - size; i--) |
4278 | state->stack[spi].slot_type[i - 1] = STACK_SPILL; | |
f7cf25b2 | 4279 | |
354e8f19 MKL |
4280 | /* size < 8 bytes spill */ |
4281 | for (; i; i--) | |
4282 | scrub_spilled_slot(&state->stack[spi].slot_type[i - 1]); | |
f7cf25b2 AS |
4283 | } |
4284 | ||
ecdf985d EZ |
4285 | static bool is_bpf_st_mem(struct bpf_insn *insn) |
4286 | { | |
4287 | return BPF_CLASS(insn->code) == BPF_ST && BPF_MODE(insn->code) == BPF_MEM; | |
4288 | } | |
4289 | ||
01f810ac | 4290 | /* check_stack_{read,write}_fixed_off functions track spill/fill of registers, |
17a52670 AS |
4291 | * stack boundary and alignment are checked in check_mem_access() |
4292 | */ | |
01f810ac AM |
4293 | static int check_stack_write_fixed_off(struct bpf_verifier_env *env, |
4294 | /* stack frame we're writing to */ | |
4295 | struct bpf_func_state *state, | |
4296 | int off, int size, int value_regno, | |
4297 | int insn_idx) | |
17a52670 | 4298 | { |
f4d7e40a | 4299 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 4300 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
ecdf985d | 4301 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; |
f7cf25b2 | 4302 | struct bpf_reg_state *reg = NULL; |
ecdf985d | 4303 | u32 dst_reg = insn->dst_reg; |
638f5b90 | 4304 | |
c69431aa | 4305 | err = grow_stack_state(state, round_up(slot + 1, BPF_REG_SIZE)); |
638f5b90 AS |
4306 | if (err) |
4307 | return err; | |
9c399760 AS |
4308 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
4309 | * so it's aligned access and [off, off + size) are within stack limits | |
4310 | */ | |
638f5b90 AS |
4311 | if (!env->allow_ptr_leaks && |
4312 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
4313 | size != BPF_REG_SIZE) { | |
4314 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
4315 | return -EACCES; | |
4316 | } | |
17a52670 | 4317 | |
f4d7e40a | 4318 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
4319 | if (value_regno >= 0) |
4320 | reg = &cur->regs[value_regno]; | |
2039f26f DB |
4321 | if (!env->bypass_spec_v4) { |
4322 | bool sanitize = reg && is_spillable_regtype(reg->type); | |
4323 | ||
4324 | for (i = 0; i < size; i++) { | |
e4f4db47 LG |
4325 | u8 type = state->stack[spi].slot_type[i]; |
4326 | ||
4327 | if (type != STACK_MISC && type != STACK_ZERO) { | |
2039f26f DB |
4328 | sanitize = true; |
4329 | break; | |
4330 | } | |
4331 | } | |
4332 | ||
4333 | if (sanitize) | |
4334 | env->insn_aux_data[insn_idx].sanitize_stack_spill = true; | |
4335 | } | |
17a52670 | 4336 | |
ef8fc7a0 KKD |
4337 | err = destroy_if_dynptr_stack_slot(env, state, spi); |
4338 | if (err) | |
4339 | return err; | |
4340 | ||
0f55f9ed | 4341 | mark_stack_slot_scratched(env, spi); |
354e8f19 | 4342 | if (reg && !(off % BPF_REG_SIZE) && register_is_bounded(reg) && |
2c78ee89 | 4343 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
4344 | if (dst_reg != BPF_REG_FP) { |
4345 | /* The backtracking logic can only recognize explicit | |
4346 | * stack slot address like [fp - 8]. Other spill of | |
8fb33b60 | 4347 | * scalar via different register has to be conservative. |
b5dc0163 AS |
4348 | * Backtrack from here and mark all registers as precise |
4349 | * that contributed into 'reg' being a constant. | |
4350 | */ | |
4351 | err = mark_chain_precision(env, value_regno); | |
4352 | if (err) | |
4353 | return err; | |
4354 | } | |
354e8f19 | 4355 | save_register_state(state, spi, reg, size); |
713274f1 MM |
4356 | /* Break the relation on a narrowing spill. */ |
4357 | if (fls64(reg->umax_value) > BITS_PER_BYTE * size) | |
4358 | state->stack[spi].spilled_ptr.id = 0; | |
ecdf985d EZ |
4359 | } else if (!reg && !(off % BPF_REG_SIZE) && is_bpf_st_mem(insn) && |
4360 | insn->imm != 0 && env->bpf_capable) { | |
4361 | struct bpf_reg_state fake_reg = {}; | |
4362 | ||
4363 | __mark_reg_known(&fake_reg, (u32)insn->imm); | |
4364 | fake_reg.type = SCALAR_VALUE; | |
4365 | save_register_state(state, spi, &fake_reg, size); | |
f7cf25b2 | 4366 | } else if (reg && is_spillable_regtype(reg->type)) { |
17a52670 | 4367 | /* register containing pointer is being spilled into stack */ |
9c399760 | 4368 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 4369 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 4370 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
4371 | return -EACCES; |
4372 | } | |
f7cf25b2 | 4373 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
4374 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
4375 | return -EINVAL; | |
4376 | } | |
354e8f19 | 4377 | save_register_state(state, spi, reg, size); |
9c399760 | 4378 | } else { |
cc2b14d5 AS |
4379 | u8 type = STACK_MISC; |
4380 | ||
679c782d EC |
4381 | /* regular write of data into stack destroys any spilled ptr */ |
4382 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
06accc87 AN |
4383 | /* Mark slots as STACK_MISC if they belonged to spilled ptr/dynptr/iter. */ |
4384 | if (is_stack_slot_special(&state->stack[spi])) | |
0bae2d4d | 4385 | for (i = 0; i < BPF_REG_SIZE; i++) |
354e8f19 | 4386 | scrub_spilled_slot(&state->stack[spi].slot_type[i]); |
9c399760 | 4387 | |
cc2b14d5 AS |
4388 | /* only mark the slot as written if all 8 bytes were written |
4389 | * otherwise read propagation may incorrectly stop too soon | |
4390 | * when stack slots are partially written. | |
4391 | * This heuristic means that read propagation will be | |
4392 | * conservative, since it will add reg_live_read marks | |
4393 | * to stack slots all the way to first state when programs | |
4394 | * writes+reads less than 8 bytes | |
4395 | */ | |
4396 | if (size == BPF_REG_SIZE) | |
4397 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
4398 | ||
4399 | /* when we zero initialize stack slots mark them as such */ | |
ecdf985d EZ |
4400 | if ((reg && register_is_null(reg)) || |
4401 | (!reg && is_bpf_st_mem(insn) && insn->imm == 0)) { | |
b5dc0163 AS |
4402 | /* backtracking doesn't work for STACK_ZERO yet. */ |
4403 | err = mark_chain_precision(env, value_regno); | |
4404 | if (err) | |
4405 | return err; | |
cc2b14d5 | 4406 | type = STACK_ZERO; |
b5dc0163 | 4407 | } |
cc2b14d5 | 4408 | |
0bae2d4d | 4409 | /* Mark slots affected by this stack write. */ |
9c399760 | 4410 | for (i = 0; i < size; i++) |
638f5b90 | 4411 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 4412 | type; |
17a52670 AS |
4413 | } |
4414 | return 0; | |
4415 | } | |
4416 | ||
01f810ac AM |
4417 | /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is |
4418 | * known to contain a variable offset. | |
4419 | * This function checks whether the write is permitted and conservatively | |
4420 | * tracks the effects of the write, considering that each stack slot in the | |
4421 | * dynamic range is potentially written to. | |
4422 | * | |
4423 | * 'off' includes 'regno->off'. | |
4424 | * 'value_regno' can be -1, meaning that an unknown value is being written to | |
4425 | * the stack. | |
4426 | * | |
4427 | * Spilled pointers in range are not marked as written because we don't know | |
4428 | * what's going to be actually written. This means that read propagation for | |
4429 | * future reads cannot be terminated by this write. | |
4430 | * | |
4431 | * For privileged programs, uninitialized stack slots are considered | |
4432 | * initialized by this write (even though we don't know exactly what offsets | |
4433 | * are going to be written to). The idea is that we don't want the verifier to | |
4434 | * reject future reads that access slots written to through variable offsets. | |
4435 | */ | |
4436 | static int check_stack_write_var_off(struct bpf_verifier_env *env, | |
4437 | /* func where register points to */ | |
4438 | struct bpf_func_state *state, | |
4439 | int ptr_regno, int off, int size, | |
4440 | int value_regno, int insn_idx) | |
4441 | { | |
4442 | struct bpf_func_state *cur; /* state of the current function */ | |
4443 | int min_off, max_off; | |
4444 | int i, err; | |
4445 | struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL; | |
31ff2135 | 4446 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; |
01f810ac AM |
4447 | bool writing_zero = false; |
4448 | /* set if the fact that we're writing a zero is used to let any | |
4449 | * stack slots remain STACK_ZERO | |
4450 | */ | |
4451 | bool zero_used = false; | |
4452 | ||
4453 | cur = env->cur_state->frame[env->cur_state->curframe]; | |
4454 | ptr_reg = &cur->regs[ptr_regno]; | |
4455 | min_off = ptr_reg->smin_value + off; | |
4456 | max_off = ptr_reg->smax_value + off + size; | |
4457 | if (value_regno >= 0) | |
4458 | value_reg = &cur->regs[value_regno]; | |
31ff2135 EZ |
4459 | if ((value_reg && register_is_null(value_reg)) || |
4460 | (!value_reg && is_bpf_st_mem(insn) && insn->imm == 0)) | |
01f810ac AM |
4461 | writing_zero = true; |
4462 | ||
c69431aa | 4463 | err = grow_stack_state(state, round_up(-min_off, BPF_REG_SIZE)); |
01f810ac AM |
4464 | if (err) |
4465 | return err; | |
4466 | ||
ef8fc7a0 KKD |
4467 | for (i = min_off; i < max_off; i++) { |
4468 | int spi; | |
4469 | ||
4470 | spi = __get_spi(i); | |
4471 | err = destroy_if_dynptr_stack_slot(env, state, spi); | |
4472 | if (err) | |
4473 | return err; | |
4474 | } | |
01f810ac AM |
4475 | |
4476 | /* Variable offset writes destroy any spilled pointers in range. */ | |
4477 | for (i = min_off; i < max_off; i++) { | |
4478 | u8 new_type, *stype; | |
4479 | int slot, spi; | |
4480 | ||
4481 | slot = -i - 1; | |
4482 | spi = slot / BPF_REG_SIZE; | |
4483 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
0f55f9ed | 4484 | mark_stack_slot_scratched(env, spi); |
01f810ac | 4485 | |
f5e477a8 KKD |
4486 | if (!env->allow_ptr_leaks && *stype != STACK_MISC && *stype != STACK_ZERO) { |
4487 | /* Reject the write if range we may write to has not | |
4488 | * been initialized beforehand. If we didn't reject | |
4489 | * here, the ptr status would be erased below (even | |
4490 | * though not all slots are actually overwritten), | |
4491 | * possibly opening the door to leaks. | |
4492 | * | |
4493 | * We do however catch STACK_INVALID case below, and | |
4494 | * only allow reading possibly uninitialized memory | |
4495 | * later for CAP_PERFMON, as the write may not happen to | |
4496 | * that slot. | |
01f810ac AM |
4497 | */ |
4498 | verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d", | |
4499 | insn_idx, i); | |
4500 | return -EINVAL; | |
4501 | } | |
4502 | ||
4503 | /* Erase all spilled pointers. */ | |
4504 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
4505 | ||
4506 | /* Update the slot type. */ | |
4507 | new_type = STACK_MISC; | |
4508 | if (writing_zero && *stype == STACK_ZERO) { | |
4509 | new_type = STACK_ZERO; | |
4510 | zero_used = true; | |
4511 | } | |
4512 | /* If the slot is STACK_INVALID, we check whether it's OK to | |
4513 | * pretend that it will be initialized by this write. The slot | |
4514 | * might not actually be written to, and so if we mark it as | |
4515 | * initialized future reads might leak uninitialized memory. | |
4516 | * For privileged programs, we will accept such reads to slots | |
4517 | * that may or may not be written because, if we're reject | |
4518 | * them, the error would be too confusing. | |
4519 | */ | |
4520 | if (*stype == STACK_INVALID && !env->allow_uninit_stack) { | |
4521 | verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d", | |
4522 | insn_idx, i); | |
4523 | return -EINVAL; | |
4524 | } | |
4525 | *stype = new_type; | |
4526 | } | |
4527 | if (zero_used) { | |
4528 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
4529 | err = mark_chain_precision(env, value_regno); | |
4530 | if (err) | |
4531 | return err; | |
4532 | } | |
4533 | return 0; | |
4534 | } | |
4535 | ||
4536 | /* When register 'dst_regno' is assigned some values from stack[min_off, | |
4537 | * max_off), we set the register's type according to the types of the | |
4538 | * respective stack slots. If all the stack values are known to be zeros, then | |
4539 | * so is the destination reg. Otherwise, the register is considered to be | |
4540 | * SCALAR. This function does not deal with register filling; the caller must | |
4541 | * ensure that all spilled registers in the stack range have been marked as | |
4542 | * read. | |
4543 | */ | |
4544 | static void mark_reg_stack_read(struct bpf_verifier_env *env, | |
4545 | /* func where src register points to */ | |
4546 | struct bpf_func_state *ptr_state, | |
4547 | int min_off, int max_off, int dst_regno) | |
4548 | { | |
4549 | struct bpf_verifier_state *vstate = env->cur_state; | |
4550 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
4551 | int i, slot, spi; | |
4552 | u8 *stype; | |
4553 | int zeros = 0; | |
4554 | ||
4555 | for (i = min_off; i < max_off; i++) { | |
4556 | slot = -i - 1; | |
4557 | spi = slot / BPF_REG_SIZE; | |
e0bf4622 | 4558 | mark_stack_slot_scratched(env, spi); |
01f810ac AM |
4559 | stype = ptr_state->stack[spi].slot_type; |
4560 | if (stype[slot % BPF_REG_SIZE] != STACK_ZERO) | |
4561 | break; | |
4562 | zeros++; | |
4563 | } | |
4564 | if (zeros == max_off - min_off) { | |
4565 | /* any access_size read into register is zero extended, | |
4566 | * so the whole register == const_zero | |
4567 | */ | |
4568 | __mark_reg_const_zero(&state->regs[dst_regno]); | |
4569 | /* backtracking doesn't support STACK_ZERO yet, | |
4570 | * so mark it precise here, so that later | |
4571 | * backtracking can stop here. | |
4572 | * Backtracking may not need this if this register | |
4573 | * doesn't participate in pointer adjustment. | |
4574 | * Forward propagation of precise flag is not | |
4575 | * necessary either. This mark is only to stop | |
4576 | * backtracking. Any register that contributed | |
4577 | * to const 0 was marked precise before spill. | |
4578 | */ | |
4579 | state->regs[dst_regno].precise = true; | |
4580 | } else { | |
4581 | /* have read misc data from the stack */ | |
4582 | mark_reg_unknown(env, state->regs, dst_regno); | |
4583 | } | |
4584 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
4585 | } | |
4586 | ||
4587 | /* Read the stack at 'off' and put the results into the register indicated by | |
4588 | * 'dst_regno'. It handles reg filling if the addressed stack slot is a | |
4589 | * spilled reg. | |
4590 | * | |
4591 | * 'dst_regno' can be -1, meaning that the read value is not going to a | |
4592 | * register. | |
4593 | * | |
4594 | * The access is assumed to be within the current stack bounds. | |
4595 | */ | |
4596 | static int check_stack_read_fixed_off(struct bpf_verifier_env *env, | |
4597 | /* func where src register points to */ | |
4598 | struct bpf_func_state *reg_state, | |
4599 | int off, int size, int dst_regno) | |
17a52670 | 4600 | { |
f4d7e40a AS |
4601 | struct bpf_verifier_state *vstate = env->cur_state; |
4602 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 4603 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 4604 | struct bpf_reg_state *reg; |
354e8f19 | 4605 | u8 *stype, type; |
17a52670 | 4606 | |
f4d7e40a | 4607 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 4608 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 4609 | |
e0bf4622 AN |
4610 | mark_stack_slot_scratched(env, spi); |
4611 | ||
27113c59 | 4612 | if (is_spilled_reg(®_state->stack[spi])) { |
f30d4968 MKL |
4613 | u8 spill_size = 1; |
4614 | ||
4615 | for (i = BPF_REG_SIZE - 1; i > 0 && stype[i - 1] == STACK_SPILL; i--) | |
4616 | spill_size++; | |
354e8f19 | 4617 | |
f30d4968 | 4618 | if (size != BPF_REG_SIZE || spill_size != BPF_REG_SIZE) { |
f7cf25b2 AS |
4619 | if (reg->type != SCALAR_VALUE) { |
4620 | verbose_linfo(env, env->insn_idx, "; "); | |
4621 | verbose(env, "invalid size of register fill\n"); | |
4622 | return -EACCES; | |
4623 | } | |
354e8f19 MKL |
4624 | |
4625 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
4626 | if (dst_regno < 0) | |
4627 | return 0; | |
4628 | ||
f30d4968 | 4629 | if (!(off % BPF_REG_SIZE) && size == spill_size) { |
354e8f19 MKL |
4630 | /* The earlier check_reg_arg() has decided the |
4631 | * subreg_def for this insn. Save it first. | |
4632 | */ | |
4633 | s32 subreg_def = state->regs[dst_regno].subreg_def; | |
4634 | ||
71f656a5 | 4635 | copy_register_state(&state->regs[dst_regno], reg); |
354e8f19 MKL |
4636 | state->regs[dst_regno].subreg_def = subreg_def; |
4637 | } else { | |
4638 | for (i = 0; i < size; i++) { | |
4639 | type = stype[(slot - i) % BPF_REG_SIZE]; | |
4640 | if (type == STACK_SPILL) | |
4641 | continue; | |
4642 | if (type == STACK_MISC) | |
4643 | continue; | |
6715df8d EZ |
4644 | if (type == STACK_INVALID && env->allow_uninit_stack) |
4645 | continue; | |
354e8f19 MKL |
4646 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
4647 | off, i, size); | |
4648 | return -EACCES; | |
4649 | } | |
01f810ac | 4650 | mark_reg_unknown(env, state->regs, dst_regno); |
f7cf25b2 | 4651 | } |
354e8f19 | 4652 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
f7cf25b2 | 4653 | return 0; |
17a52670 | 4654 | } |
17a52670 | 4655 | |
01f810ac | 4656 | if (dst_regno >= 0) { |
17a52670 | 4657 | /* restore register state from stack */ |
71f656a5 | 4658 | copy_register_state(&state->regs[dst_regno], reg); |
2f18f62e AS |
4659 | /* mark reg as written since spilled pointer state likely |
4660 | * has its liveness marks cleared by is_state_visited() | |
4661 | * which resets stack/reg liveness for state transitions | |
4662 | */ | |
01f810ac | 4663 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
6e7e63cb | 4664 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
01f810ac | 4665 | /* If dst_regno==-1, the caller is asking us whether |
6e7e63cb JH |
4666 | * it is acceptable to use this value as a SCALAR_VALUE |
4667 | * (e.g. for XADD). | |
4668 | * We must not allow unprivileged callers to do that | |
4669 | * with spilled pointers. | |
4670 | */ | |
4671 | verbose(env, "leaking pointer from stack off %d\n", | |
4672 | off); | |
4673 | return -EACCES; | |
dc503a8a | 4674 | } |
f7cf25b2 | 4675 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 AS |
4676 | } else { |
4677 | for (i = 0; i < size; i++) { | |
01f810ac AM |
4678 | type = stype[(slot - i) % BPF_REG_SIZE]; |
4679 | if (type == STACK_MISC) | |
cc2b14d5 | 4680 | continue; |
01f810ac | 4681 | if (type == STACK_ZERO) |
cc2b14d5 | 4682 | continue; |
6715df8d EZ |
4683 | if (type == STACK_INVALID && env->allow_uninit_stack) |
4684 | continue; | |
cc2b14d5 AS |
4685 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
4686 | off, i, size); | |
4687 | return -EACCES; | |
4688 | } | |
f7cf25b2 | 4689 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
01f810ac AM |
4690 | if (dst_regno >= 0) |
4691 | mark_reg_stack_read(env, reg_state, off, off + size, dst_regno); | |
17a52670 | 4692 | } |
f7cf25b2 | 4693 | return 0; |
17a52670 AS |
4694 | } |
4695 | ||
61df10c7 | 4696 | enum bpf_access_src { |
01f810ac AM |
4697 | ACCESS_DIRECT = 1, /* the access is performed by an instruction */ |
4698 | ACCESS_HELPER = 2, /* the access is performed by a helper */ | |
4699 | }; | |
4700 | ||
4701 | static int check_stack_range_initialized(struct bpf_verifier_env *env, | |
4702 | int regno, int off, int access_size, | |
4703 | bool zero_size_allowed, | |
61df10c7 | 4704 | enum bpf_access_src type, |
01f810ac AM |
4705 | struct bpf_call_arg_meta *meta); |
4706 | ||
4707 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) | |
4708 | { | |
4709 | return cur_regs(env) + regno; | |
4710 | } | |
4711 | ||
4712 | /* Read the stack at 'ptr_regno + off' and put the result into the register | |
4713 | * 'dst_regno'. | |
4714 | * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'), | |
4715 | * but not its variable offset. | |
4716 | * 'size' is assumed to be <= reg size and the access is assumed to be aligned. | |
4717 | * | |
4718 | * As opposed to check_stack_read_fixed_off, this function doesn't deal with | |
4719 | * filling registers (i.e. reads of spilled register cannot be detected when | |
4720 | * the offset is not fixed). We conservatively mark 'dst_regno' as containing | |
4721 | * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable | |
4722 | * offset; for a fixed offset check_stack_read_fixed_off should be used | |
4723 | * instead. | |
4724 | */ | |
4725 | static int check_stack_read_var_off(struct bpf_verifier_env *env, | |
4726 | int ptr_regno, int off, int size, int dst_regno) | |
e4298d25 | 4727 | { |
01f810ac AM |
4728 | /* The state of the source register. */ |
4729 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
4730 | struct bpf_func_state *ptr_state = func(env, reg); | |
4731 | int err; | |
4732 | int min_off, max_off; | |
4733 | ||
4734 | /* Note that we pass a NULL meta, so raw access will not be permitted. | |
e4298d25 | 4735 | */ |
01f810ac AM |
4736 | err = check_stack_range_initialized(env, ptr_regno, off, size, |
4737 | false, ACCESS_DIRECT, NULL); | |
4738 | if (err) | |
4739 | return err; | |
4740 | ||
4741 | min_off = reg->smin_value + off; | |
4742 | max_off = reg->smax_value + off; | |
4743 | mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno); | |
4744 | return 0; | |
4745 | } | |
4746 | ||
4747 | /* check_stack_read dispatches to check_stack_read_fixed_off or | |
4748 | * check_stack_read_var_off. | |
4749 | * | |
4750 | * The caller must ensure that the offset falls within the allocated stack | |
4751 | * bounds. | |
4752 | * | |
4753 | * 'dst_regno' is a register which will receive the value from the stack. It | |
4754 | * can be -1, meaning that the read value is not going to a register. | |
4755 | */ | |
4756 | static int check_stack_read(struct bpf_verifier_env *env, | |
4757 | int ptr_regno, int off, int size, | |
4758 | int dst_regno) | |
4759 | { | |
4760 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
4761 | struct bpf_func_state *state = func(env, reg); | |
4762 | int err; | |
4763 | /* Some accesses are only permitted with a static offset. */ | |
4764 | bool var_off = !tnum_is_const(reg->var_off); | |
4765 | ||
4766 | /* The offset is required to be static when reads don't go to a | |
4767 | * register, in order to not leak pointers (see | |
4768 | * check_stack_read_fixed_off). | |
4769 | */ | |
4770 | if (dst_regno < 0 && var_off) { | |
e4298d25 DB |
4771 | char tn_buf[48]; |
4772 | ||
4773 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac | 4774 | verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n", |
e4298d25 DB |
4775 | tn_buf, off, size); |
4776 | return -EACCES; | |
4777 | } | |
01f810ac AM |
4778 | /* Variable offset is prohibited for unprivileged mode for simplicity |
4779 | * since it requires corresponding support in Spectre masking for stack | |
082cdc69 LG |
4780 | * ALU. See also retrieve_ptr_limit(). The check in |
4781 | * check_stack_access_for_ptr_arithmetic() called by | |
4782 | * adjust_ptr_min_max_vals() prevents users from creating stack pointers | |
4783 | * with variable offsets, therefore no check is required here. Further, | |
4784 | * just checking it here would be insufficient as speculative stack | |
4785 | * writes could still lead to unsafe speculative behaviour. | |
01f810ac | 4786 | */ |
01f810ac AM |
4787 | if (!var_off) { |
4788 | off += reg->var_off.value; | |
4789 | err = check_stack_read_fixed_off(env, state, off, size, | |
4790 | dst_regno); | |
4791 | } else { | |
4792 | /* Variable offset stack reads need more conservative handling | |
4793 | * than fixed offset ones. Note that dst_regno >= 0 on this | |
4794 | * branch. | |
4795 | */ | |
4796 | err = check_stack_read_var_off(env, ptr_regno, off, size, | |
4797 | dst_regno); | |
4798 | } | |
4799 | return err; | |
4800 | } | |
4801 | ||
4802 | ||
4803 | /* check_stack_write dispatches to check_stack_write_fixed_off or | |
4804 | * check_stack_write_var_off. | |
4805 | * | |
4806 | * 'ptr_regno' is the register used as a pointer into the stack. | |
4807 | * 'off' includes 'ptr_regno->off', but not its variable offset (if any). | |
4808 | * 'value_regno' is the register whose value we're writing to the stack. It can | |
4809 | * be -1, meaning that we're not writing from a register. | |
4810 | * | |
4811 | * The caller must ensure that the offset falls within the maximum stack size. | |
4812 | */ | |
4813 | static int check_stack_write(struct bpf_verifier_env *env, | |
4814 | int ptr_regno, int off, int size, | |
4815 | int value_regno, int insn_idx) | |
4816 | { | |
4817 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
4818 | struct bpf_func_state *state = func(env, reg); | |
4819 | int err; | |
4820 | ||
4821 | if (tnum_is_const(reg->var_off)) { | |
4822 | off += reg->var_off.value; | |
4823 | err = check_stack_write_fixed_off(env, state, off, size, | |
4824 | value_regno, insn_idx); | |
4825 | } else { | |
4826 | /* Variable offset stack reads need more conservative handling | |
4827 | * than fixed offset ones. | |
4828 | */ | |
4829 | err = check_stack_write_var_off(env, state, | |
4830 | ptr_regno, off, size, | |
4831 | value_regno, insn_idx); | |
4832 | } | |
4833 | return err; | |
e4298d25 DB |
4834 | } |
4835 | ||
591fe988 DB |
4836 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
4837 | int off, int size, enum bpf_access_type type) | |
4838 | { | |
4839 | struct bpf_reg_state *regs = cur_regs(env); | |
4840 | struct bpf_map *map = regs[regno].map_ptr; | |
4841 | u32 cap = bpf_map_flags_to_cap(map); | |
4842 | ||
4843 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
4844 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
4845 | map->value_size, off, size); | |
4846 | return -EACCES; | |
4847 | } | |
4848 | ||
4849 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
4850 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
4851 | map->value_size, off, size); | |
4852 | return -EACCES; | |
4853 | } | |
4854 | ||
4855 | return 0; | |
4856 | } | |
4857 | ||
457f4436 AN |
4858 | /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */ |
4859 | static int __check_mem_access(struct bpf_verifier_env *env, int regno, | |
4860 | int off, int size, u32 mem_size, | |
4861 | bool zero_size_allowed) | |
17a52670 | 4862 | { |
457f4436 AN |
4863 | bool size_ok = size > 0 || (size == 0 && zero_size_allowed); |
4864 | struct bpf_reg_state *reg; | |
4865 | ||
4866 | if (off >= 0 && size_ok && (u64)off + size <= mem_size) | |
4867 | return 0; | |
17a52670 | 4868 | |
457f4436 AN |
4869 | reg = &cur_regs(env)[regno]; |
4870 | switch (reg->type) { | |
69c087ba YS |
4871 | case PTR_TO_MAP_KEY: |
4872 | verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n", | |
4873 | mem_size, off, size); | |
4874 | break; | |
457f4436 | 4875 | case PTR_TO_MAP_VALUE: |
61bd5218 | 4876 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
457f4436 AN |
4877 | mem_size, off, size); |
4878 | break; | |
4879 | case PTR_TO_PACKET: | |
4880 | case PTR_TO_PACKET_META: | |
4881 | case PTR_TO_PACKET_END: | |
4882 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", | |
4883 | off, size, regno, reg->id, off, mem_size); | |
4884 | break; | |
4885 | case PTR_TO_MEM: | |
4886 | default: | |
4887 | verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n", | |
4888 | mem_size, off, size); | |
17a52670 | 4889 | } |
457f4436 AN |
4890 | |
4891 | return -EACCES; | |
17a52670 AS |
4892 | } |
4893 | ||
457f4436 AN |
4894 | /* check read/write into a memory region with possible variable offset */ |
4895 | static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno, | |
4896 | int off, int size, u32 mem_size, | |
4897 | bool zero_size_allowed) | |
dbcfe5f7 | 4898 | { |
f4d7e40a AS |
4899 | struct bpf_verifier_state *vstate = env->cur_state; |
4900 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
4901 | struct bpf_reg_state *reg = &state->regs[regno]; |
4902 | int err; | |
4903 | ||
457f4436 | 4904 | /* We may have adjusted the register pointing to memory region, so we |
f1174f77 EC |
4905 | * need to try adding each of min_value and max_value to off |
4906 | * to make sure our theoretical access will be safe. | |
2e576648 CL |
4907 | * |
4908 | * The minimum value is only important with signed | |
dbcfe5f7 GB |
4909 | * comparisons where we can't assume the floor of a |
4910 | * value is 0. If we are using signed variables for our | |
4911 | * index'es we need to make sure that whatever we use | |
4912 | * will have a set floor within our range. | |
4913 | */ | |
b7137c4e DB |
4914 | if (reg->smin_value < 0 && |
4915 | (reg->smin_value == S64_MIN || | |
4916 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
4917 | reg->smin_value + off < 0)) { | |
61bd5218 | 4918 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
4919 | regno); |
4920 | return -EACCES; | |
4921 | } | |
457f4436 AN |
4922 | err = __check_mem_access(env, regno, reg->smin_value + off, size, |
4923 | mem_size, zero_size_allowed); | |
dbcfe5f7 | 4924 | if (err) { |
457f4436 | 4925 | verbose(env, "R%d min value is outside of the allowed memory range\n", |
61bd5218 | 4926 | regno); |
dbcfe5f7 GB |
4927 | return err; |
4928 | } | |
4929 | ||
b03c9f9f EC |
4930 | /* If we haven't set a max value then we need to bail since we can't be |
4931 | * sure we won't do bad things. | |
4932 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 4933 | */ |
b03c9f9f | 4934 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
457f4436 | 4935 | verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n", |
dbcfe5f7 GB |
4936 | regno); |
4937 | return -EACCES; | |
4938 | } | |
457f4436 AN |
4939 | err = __check_mem_access(env, regno, reg->umax_value + off, size, |
4940 | mem_size, zero_size_allowed); | |
4941 | if (err) { | |
4942 | verbose(env, "R%d max value is outside of the allowed memory range\n", | |
61bd5218 | 4943 | regno); |
457f4436 AN |
4944 | return err; |
4945 | } | |
4946 | ||
4947 | return 0; | |
4948 | } | |
d83525ca | 4949 | |
e9147b44 KKD |
4950 | static int __check_ptr_off_reg(struct bpf_verifier_env *env, |
4951 | const struct bpf_reg_state *reg, int regno, | |
4952 | bool fixed_off_ok) | |
4953 | { | |
4954 | /* Access to this pointer-typed register or passing it to a helper | |
4955 | * is only allowed in its original, unmodified form. | |
4956 | */ | |
4957 | ||
4958 | if (reg->off < 0) { | |
4959 | verbose(env, "negative offset %s ptr R%d off=%d disallowed\n", | |
4960 | reg_type_str(env, reg->type), regno, reg->off); | |
4961 | return -EACCES; | |
4962 | } | |
4963 | ||
4964 | if (!fixed_off_ok && reg->off) { | |
4965 | verbose(env, "dereference of modified %s ptr R%d off=%d disallowed\n", | |
4966 | reg_type_str(env, reg->type), regno, reg->off); | |
4967 | return -EACCES; | |
4968 | } | |
4969 | ||
4970 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
4971 | char tn_buf[48]; | |
4972 | ||
4973 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
4974 | verbose(env, "variable %s access var_off=%s disallowed\n", | |
4975 | reg_type_str(env, reg->type), tn_buf); | |
4976 | return -EACCES; | |
4977 | } | |
4978 | ||
4979 | return 0; | |
4980 | } | |
4981 | ||
4982 | int check_ptr_off_reg(struct bpf_verifier_env *env, | |
4983 | const struct bpf_reg_state *reg, int regno) | |
4984 | { | |
4985 | return __check_ptr_off_reg(env, reg, regno, false); | |
4986 | } | |
4987 | ||
61df10c7 | 4988 | static int map_kptr_match_type(struct bpf_verifier_env *env, |
aa3496ac | 4989 | struct btf_field *kptr_field, |
61df10c7 KKD |
4990 | struct bpf_reg_state *reg, u32 regno) |
4991 | { | |
b32a5dae | 4992 | const char *targ_name = btf_type_name(kptr_field->kptr.btf, kptr_field->kptr.btf_id); |
ab6c637a | 4993 | int perm_flags; |
61df10c7 KKD |
4994 | const char *reg_name = ""; |
4995 | ||
ab6c637a YS |
4996 | if (btf_is_kernel(reg->btf)) { |
4997 | perm_flags = PTR_MAYBE_NULL | PTR_TRUSTED | MEM_RCU; | |
4998 | ||
4999 | /* Only unreferenced case accepts untrusted pointers */ | |
5000 | if (kptr_field->type == BPF_KPTR_UNREF) | |
5001 | perm_flags |= PTR_UNTRUSTED; | |
5002 | } else { | |
5003 | perm_flags = PTR_MAYBE_NULL | MEM_ALLOC; | |
5004 | } | |
6efe152d KKD |
5005 | |
5006 | if (base_type(reg->type) != PTR_TO_BTF_ID || (type_flag(reg->type) & ~perm_flags)) | |
61df10c7 KKD |
5007 | goto bad_type; |
5008 | ||
61df10c7 | 5009 | /* We need to verify reg->type and reg->btf, before accessing reg->btf */ |
b32a5dae | 5010 | reg_name = btf_type_name(reg->btf, reg->btf_id); |
61df10c7 | 5011 | |
c0a5a21c KKD |
5012 | /* For ref_ptr case, release function check should ensure we get one |
5013 | * referenced PTR_TO_BTF_ID, and that its fixed offset is 0. For the | |
5014 | * normal store of unreferenced kptr, we must ensure var_off is zero. | |
5015 | * Since ref_ptr cannot be accessed directly by BPF insns, checks for | |
5016 | * reg->off and reg->ref_obj_id are not needed here. | |
5017 | */ | |
61df10c7 KKD |
5018 | if (__check_ptr_off_reg(env, reg, regno, true)) |
5019 | return -EACCES; | |
5020 | ||
ab6c637a | 5021 | /* A full type match is needed, as BTF can be vmlinux, module or prog BTF, and |
61df10c7 KKD |
5022 | * we also need to take into account the reg->off. |
5023 | * | |
5024 | * We want to support cases like: | |
5025 | * | |
5026 | * struct foo { | |
5027 | * struct bar br; | |
5028 | * struct baz bz; | |
5029 | * }; | |
5030 | * | |
5031 | * struct foo *v; | |
5032 | * v = func(); // PTR_TO_BTF_ID | |
5033 | * val->foo = v; // reg->off is zero, btf and btf_id match type | |
5034 | * val->bar = &v->br; // reg->off is still zero, but we need to retry with | |
5035 | * // first member type of struct after comparison fails | |
5036 | * val->baz = &v->bz; // reg->off is non-zero, so struct needs to be walked | |
5037 | * // to match type | |
5038 | * | |
5039 | * In the kptr_ref case, check_func_arg_reg_off already ensures reg->off | |
2ab3b380 KKD |
5040 | * is zero. We must also ensure that btf_struct_ids_match does not walk |
5041 | * the struct to match type against first member of struct, i.e. reject | |
5042 | * second case from above. Hence, when type is BPF_KPTR_REF, we set | |
5043 | * strict mode to true for type match. | |
61df10c7 KKD |
5044 | */ |
5045 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
aa3496ac KKD |
5046 | kptr_field->kptr.btf, kptr_field->kptr.btf_id, |
5047 | kptr_field->type == BPF_KPTR_REF)) | |
61df10c7 KKD |
5048 | goto bad_type; |
5049 | return 0; | |
5050 | bad_type: | |
5051 | verbose(env, "invalid kptr access, R%d type=%s%s ", regno, | |
5052 | reg_type_str(env, reg->type), reg_name); | |
6efe152d | 5053 | verbose(env, "expected=%s%s", reg_type_str(env, PTR_TO_BTF_ID), targ_name); |
aa3496ac | 5054 | if (kptr_field->type == BPF_KPTR_UNREF) |
6efe152d KKD |
5055 | verbose(env, " or %s%s\n", reg_type_str(env, PTR_TO_BTF_ID | PTR_UNTRUSTED), |
5056 | targ_name); | |
5057 | else | |
5058 | verbose(env, "\n"); | |
61df10c7 KKD |
5059 | return -EINVAL; |
5060 | } | |
5061 | ||
20c09d92 AS |
5062 | /* The non-sleepable programs and sleepable programs with explicit bpf_rcu_read_lock() |
5063 | * can dereference RCU protected pointers and result is PTR_TRUSTED. | |
5064 | */ | |
5065 | static bool in_rcu_cs(struct bpf_verifier_env *env) | |
5066 | { | |
5067 | return env->cur_state->active_rcu_lock || !env->prog->aux->sleepable; | |
5068 | } | |
5069 | ||
5070 | /* Once GCC supports btf_type_tag the following mechanism will be replaced with tag check */ | |
5071 | BTF_SET_START(rcu_protected_types) | |
5072 | BTF_ID(struct, prog_test_ref_kfunc) | |
5073 | BTF_ID(struct, cgroup) | |
63d2d83d | 5074 | BTF_ID(struct, bpf_cpumask) |
d02c48fa | 5075 | BTF_ID(struct, task_struct) |
20c09d92 AS |
5076 | BTF_SET_END(rcu_protected_types) |
5077 | ||
5078 | static bool rcu_protected_object(const struct btf *btf, u32 btf_id) | |
5079 | { | |
5080 | if (!btf_is_kernel(btf)) | |
5081 | return false; | |
5082 | return btf_id_set_contains(&rcu_protected_types, btf_id); | |
5083 | } | |
5084 | ||
5085 | static bool rcu_safe_kptr(const struct btf_field *field) | |
5086 | { | |
5087 | const struct btf_field_kptr *kptr = &field->kptr; | |
5088 | ||
5089 | return field->type == BPF_KPTR_REF && rcu_protected_object(kptr->btf, kptr->btf_id); | |
5090 | } | |
5091 | ||
61df10c7 KKD |
5092 | static int check_map_kptr_access(struct bpf_verifier_env *env, u32 regno, |
5093 | int value_regno, int insn_idx, | |
aa3496ac | 5094 | struct btf_field *kptr_field) |
61df10c7 KKD |
5095 | { |
5096 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; | |
5097 | int class = BPF_CLASS(insn->code); | |
5098 | struct bpf_reg_state *val_reg; | |
5099 | ||
5100 | /* Things we already checked for in check_map_access and caller: | |
5101 | * - Reject cases where variable offset may touch kptr | |
5102 | * - size of access (must be BPF_DW) | |
5103 | * - tnum_is_const(reg->var_off) | |
aa3496ac | 5104 | * - kptr_field->offset == off + reg->var_off.value |
61df10c7 KKD |
5105 | */ |
5106 | /* Only BPF_[LDX,STX,ST] | BPF_MEM | BPF_DW is supported */ | |
5107 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
5108 | verbose(env, "kptr in map can only be accessed using BPF_MEM instruction mode\n"); | |
5109 | return -EACCES; | |
5110 | } | |
5111 | ||
6efe152d KKD |
5112 | /* We only allow loading referenced kptr, since it will be marked as |
5113 | * untrusted, similar to unreferenced kptr. | |
5114 | */ | |
aa3496ac | 5115 | if (class != BPF_LDX && kptr_field->type == BPF_KPTR_REF) { |
6efe152d | 5116 | verbose(env, "store to referenced kptr disallowed\n"); |
c0a5a21c KKD |
5117 | return -EACCES; |
5118 | } | |
5119 | ||
61df10c7 KKD |
5120 | if (class == BPF_LDX) { |
5121 | val_reg = reg_state(env, value_regno); | |
5122 | /* We can simply mark the value_regno receiving the pointer | |
5123 | * value from map as PTR_TO_BTF_ID, with the correct type. | |
5124 | */ | |
aa3496ac | 5125 | mark_btf_ld_reg(env, cur_regs(env), value_regno, PTR_TO_BTF_ID, kptr_field->kptr.btf, |
20c09d92 AS |
5126 | kptr_field->kptr.btf_id, |
5127 | rcu_safe_kptr(kptr_field) && in_rcu_cs(env) ? | |
5128 | PTR_MAYBE_NULL | MEM_RCU : | |
5129 | PTR_MAYBE_NULL | PTR_UNTRUSTED); | |
61df10c7 KKD |
5130 | /* For mark_ptr_or_null_reg */ |
5131 | val_reg->id = ++env->id_gen; | |
5132 | } else if (class == BPF_STX) { | |
5133 | val_reg = reg_state(env, value_regno); | |
5134 | if (!register_is_null(val_reg) && | |
aa3496ac | 5135 | map_kptr_match_type(env, kptr_field, val_reg, value_regno)) |
61df10c7 KKD |
5136 | return -EACCES; |
5137 | } else if (class == BPF_ST) { | |
5138 | if (insn->imm) { | |
5139 | verbose(env, "BPF_ST imm must be 0 when storing to kptr at off=%u\n", | |
aa3496ac | 5140 | kptr_field->offset); |
61df10c7 KKD |
5141 | return -EACCES; |
5142 | } | |
5143 | } else { | |
5144 | verbose(env, "kptr in map can only be accessed using BPF_LDX/BPF_STX/BPF_ST\n"); | |
5145 | return -EACCES; | |
5146 | } | |
5147 | return 0; | |
5148 | } | |
5149 | ||
457f4436 AN |
5150 | /* check read/write into a map element with possible variable offset */ |
5151 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
61df10c7 KKD |
5152 | int off, int size, bool zero_size_allowed, |
5153 | enum bpf_access_src src) | |
457f4436 AN |
5154 | { |
5155 | struct bpf_verifier_state *vstate = env->cur_state; | |
5156 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5157 | struct bpf_reg_state *reg = &state->regs[regno]; | |
5158 | struct bpf_map *map = reg->map_ptr; | |
aa3496ac KKD |
5159 | struct btf_record *rec; |
5160 | int err, i; | |
457f4436 AN |
5161 | |
5162 | err = check_mem_region_access(env, regno, off, size, map->value_size, | |
5163 | zero_size_allowed); | |
5164 | if (err) | |
5165 | return err; | |
5166 | ||
aa3496ac KKD |
5167 | if (IS_ERR_OR_NULL(map->record)) |
5168 | return 0; | |
5169 | rec = map->record; | |
5170 | for (i = 0; i < rec->cnt; i++) { | |
5171 | struct btf_field *field = &rec->fields[i]; | |
5172 | u32 p = field->offset; | |
d83525ca | 5173 | |
db559117 KKD |
5174 | /* If any part of a field can be touched by load/store, reject |
5175 | * this program. To check that [x1, x2) overlaps with [y1, y2), | |
d83525ca AS |
5176 | * it is sufficient to check x1 < y2 && y1 < x2. |
5177 | */ | |
aa3496ac KKD |
5178 | if (reg->smin_value + off < p + btf_field_type_size(field->type) && |
5179 | p < reg->umax_value + off + size) { | |
5180 | switch (field->type) { | |
5181 | case BPF_KPTR_UNREF: | |
5182 | case BPF_KPTR_REF: | |
61df10c7 KKD |
5183 | if (src != ACCESS_DIRECT) { |
5184 | verbose(env, "kptr cannot be accessed indirectly by helper\n"); | |
5185 | return -EACCES; | |
5186 | } | |
5187 | if (!tnum_is_const(reg->var_off)) { | |
5188 | verbose(env, "kptr access cannot have variable offset\n"); | |
5189 | return -EACCES; | |
5190 | } | |
5191 | if (p != off + reg->var_off.value) { | |
5192 | verbose(env, "kptr access misaligned expected=%u off=%llu\n", | |
5193 | p, off + reg->var_off.value); | |
5194 | return -EACCES; | |
5195 | } | |
5196 | if (size != bpf_size_to_bytes(BPF_DW)) { | |
5197 | verbose(env, "kptr access size must be BPF_DW\n"); | |
5198 | return -EACCES; | |
5199 | } | |
5200 | break; | |
aa3496ac | 5201 | default: |
db559117 KKD |
5202 | verbose(env, "%s cannot be accessed directly by load/store\n", |
5203 | btf_field_type_name(field->type)); | |
aa3496ac | 5204 | return -EACCES; |
61df10c7 KKD |
5205 | } |
5206 | } | |
5207 | } | |
aa3496ac | 5208 | return 0; |
dbcfe5f7 GB |
5209 | } |
5210 | ||
969bf05e AS |
5211 | #define MAX_PACKET_OFF 0xffff |
5212 | ||
58e2af8b | 5213 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
5214 | const struct bpf_call_arg_meta *meta, |
5215 | enum bpf_access_type t) | |
4acf6c0b | 5216 | { |
7e40781c UP |
5217 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
5218 | ||
5219 | switch (prog_type) { | |
5d66fa7d | 5220 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
5221 | case BPF_PROG_TYPE_LWT_IN: |
5222 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 5223 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 5224 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 5225 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 5226 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
5227 | if (t == BPF_WRITE) |
5228 | return false; | |
8731745e | 5229 | fallthrough; |
5d66fa7d DB |
5230 | |
5231 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
5232 | case BPF_PROG_TYPE_SCHED_CLS: |
5233 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 5234 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 5235 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 5236 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 5237 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
5238 | if (meta) |
5239 | return meta->pkt_access; | |
5240 | ||
5241 | env->seen_direct_write = true; | |
4acf6c0b | 5242 | return true; |
0d01da6a SF |
5243 | |
5244 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
5245 | if (t == BPF_WRITE) | |
5246 | env->seen_direct_write = true; | |
5247 | ||
5248 | return true; | |
5249 | ||
4acf6c0b BB |
5250 | default: |
5251 | return false; | |
5252 | } | |
5253 | } | |
5254 | ||
f1174f77 | 5255 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 5256 | int size, bool zero_size_allowed) |
f1174f77 | 5257 | { |
638f5b90 | 5258 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
5259 | struct bpf_reg_state *reg = ®s[regno]; |
5260 | int err; | |
5261 | ||
5262 | /* We may have added a variable offset to the packet pointer; but any | |
5263 | * reg->range we have comes after that. We are only checking the fixed | |
5264 | * offset. | |
5265 | */ | |
5266 | ||
5267 | /* We don't allow negative numbers, because we aren't tracking enough | |
5268 | * detail to prove they're safe. | |
5269 | */ | |
b03c9f9f | 5270 | if (reg->smin_value < 0) { |
61bd5218 | 5271 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
5272 | regno); |
5273 | return -EACCES; | |
5274 | } | |
6d94e741 AS |
5275 | |
5276 | err = reg->range < 0 ? -EINVAL : | |
5277 | __check_mem_access(env, regno, off, size, reg->range, | |
457f4436 | 5278 | zero_size_allowed); |
f1174f77 | 5279 | if (err) { |
61bd5218 | 5280 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
5281 | return err; |
5282 | } | |
e647815a | 5283 | |
457f4436 | 5284 | /* __check_mem_access has made sure "off + size - 1" is within u16. |
e647815a JW |
5285 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, |
5286 | * otherwise find_good_pkt_pointers would have refused to set range info | |
457f4436 | 5287 | * that __check_mem_access would have rejected this pkt access. |
e647815a JW |
5288 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. |
5289 | */ | |
5290 | env->prog->aux->max_pkt_offset = | |
5291 | max_t(u32, env->prog->aux->max_pkt_offset, | |
5292 | off + reg->umax_value + size - 1); | |
5293 | ||
f1174f77 EC |
5294 | return err; |
5295 | } | |
5296 | ||
5297 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 5298 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 | 5299 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
22dc4a0f | 5300 | struct btf **btf, u32 *btf_id) |
17a52670 | 5301 | { |
f96da094 DB |
5302 | struct bpf_insn_access_aux info = { |
5303 | .reg_type = *reg_type, | |
9e15db66 | 5304 | .log = &env->log, |
f96da094 | 5305 | }; |
31fd8581 | 5306 | |
4f9218aa | 5307 | if (env->ops->is_valid_access && |
5e43f899 | 5308 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
5309 | /* A non zero info.ctx_field_size indicates that this field is a |
5310 | * candidate for later verifier transformation to load the whole | |
5311 | * field and then apply a mask when accessed with a narrower | |
5312 | * access than actual ctx access size. A zero info.ctx_field_size | |
5313 | * will only allow for whole field access and rejects any other | |
5314 | * type of narrower access. | |
31fd8581 | 5315 | */ |
23994631 | 5316 | *reg_type = info.reg_type; |
31fd8581 | 5317 | |
c25b2ae1 | 5318 | if (base_type(*reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 5319 | *btf = info.btf; |
9e15db66 | 5320 | *btf_id = info.btf_id; |
22dc4a0f | 5321 | } else { |
9e15db66 | 5322 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
22dc4a0f | 5323 | } |
32bbe007 AS |
5324 | /* remember the offset of last byte accessed in ctx */ |
5325 | if (env->prog->aux->max_ctx_offset < off + size) | |
5326 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 5327 | return 0; |
32bbe007 | 5328 | } |
17a52670 | 5329 | |
61bd5218 | 5330 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
5331 | return -EACCES; |
5332 | } | |
5333 | ||
d58e468b PP |
5334 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
5335 | int size) | |
5336 | { | |
5337 | if (size < 0 || off < 0 || | |
5338 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
5339 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
5340 | off, size); | |
5341 | return -EACCES; | |
5342 | } | |
5343 | return 0; | |
5344 | } | |
5345 | ||
5f456649 MKL |
5346 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
5347 | u32 regno, int off, int size, | |
5348 | enum bpf_access_type t) | |
c64b7983 JS |
5349 | { |
5350 | struct bpf_reg_state *regs = cur_regs(env); | |
5351 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 5352 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 5353 | bool valid; |
c64b7983 JS |
5354 | |
5355 | if (reg->smin_value < 0) { | |
5356 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
5357 | regno); | |
5358 | return -EACCES; | |
5359 | } | |
5360 | ||
46f8bc92 MKL |
5361 | switch (reg->type) { |
5362 | case PTR_TO_SOCK_COMMON: | |
5363 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
5364 | break; | |
5365 | case PTR_TO_SOCKET: | |
5366 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
5367 | break; | |
655a51e5 MKL |
5368 | case PTR_TO_TCP_SOCK: |
5369 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
5370 | break; | |
fada7fdc JL |
5371 | case PTR_TO_XDP_SOCK: |
5372 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
5373 | break; | |
46f8bc92 MKL |
5374 | default: |
5375 | valid = false; | |
c64b7983 JS |
5376 | } |
5377 | ||
5f456649 | 5378 | |
46f8bc92 MKL |
5379 | if (valid) { |
5380 | env->insn_aux_data[insn_idx].ctx_field_size = | |
5381 | info.ctx_field_size; | |
5382 | return 0; | |
5383 | } | |
5384 | ||
5385 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
c25b2ae1 | 5386 | regno, reg_type_str(env, reg->type), off, size); |
46f8bc92 MKL |
5387 | |
5388 | return -EACCES; | |
c64b7983 JS |
5389 | } |
5390 | ||
4cabc5b1 DB |
5391 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
5392 | { | |
2a159c6f | 5393 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
5394 | } |
5395 | ||
f37a8cb8 DB |
5396 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
5397 | { | |
2a159c6f | 5398 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 5399 | |
46f8bc92 MKL |
5400 | return reg->type == PTR_TO_CTX; |
5401 | } | |
5402 | ||
5403 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
5404 | { | |
5405 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
5406 | ||
5407 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
5408 | } |
5409 | ||
ca369602 DB |
5410 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
5411 | { | |
2a159c6f | 5412 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
5413 | |
5414 | return type_is_pkt_pointer(reg->type); | |
5415 | } | |
5416 | ||
4b5defde DB |
5417 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
5418 | { | |
5419 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
5420 | ||
5421 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
5422 | return reg->type == PTR_TO_FLOW_KEYS; | |
5423 | } | |
5424 | ||
831deb29 AP |
5425 | static u32 *reg2btf_ids[__BPF_REG_TYPE_MAX] = { |
5426 | #ifdef CONFIG_NET | |
5427 | [PTR_TO_SOCKET] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK], | |
5428 | [PTR_TO_SOCK_COMMON] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
5429 | [PTR_TO_TCP_SOCK] = &btf_sock_ids[BTF_SOCK_TYPE_TCP], | |
5430 | #endif | |
5ba190c2 | 5431 | [CONST_PTR_TO_MAP] = btf_bpf_map_id, |
831deb29 AP |
5432 | }; |
5433 | ||
9bb00b28 YS |
5434 | static bool is_trusted_reg(const struct bpf_reg_state *reg) |
5435 | { | |
5436 | /* A referenced register is always trusted. */ | |
5437 | if (reg->ref_obj_id) | |
5438 | return true; | |
5439 | ||
831deb29 AP |
5440 | /* Types listed in the reg2btf_ids are always trusted */ |
5441 | if (reg2btf_ids[base_type(reg->type)]) | |
5442 | return true; | |
5443 | ||
9bb00b28 | 5444 | /* If a register is not referenced, it is trusted if it has the |
fca1aa75 | 5445 | * MEM_ALLOC or PTR_TRUSTED type modifiers, and no others. Some of the |
9bb00b28 YS |
5446 | * other type modifiers may be safe, but we elect to take an opt-in |
5447 | * approach here as some (e.g. PTR_UNTRUSTED and PTR_MAYBE_NULL) are | |
5448 | * not. | |
5449 | * | |
5450 | * Eventually, we should make PTR_TRUSTED the single source of truth | |
5451 | * for whether a register is trusted. | |
5452 | */ | |
5453 | return type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS && | |
5454 | !bpf_type_has_unsafe_modifiers(reg->type); | |
5455 | } | |
5456 | ||
fca1aa75 YS |
5457 | static bool is_rcu_reg(const struct bpf_reg_state *reg) |
5458 | { | |
5459 | return reg->type & MEM_RCU; | |
5460 | } | |
5461 | ||
afeebf9f AS |
5462 | static void clear_trusted_flags(enum bpf_type_flag *flag) |
5463 | { | |
5464 | *flag &= ~(BPF_REG_TRUSTED_MODIFIERS | MEM_RCU); | |
5465 | } | |
5466 | ||
61bd5218 JK |
5467 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
5468 | const struct bpf_reg_state *reg, | |
d1174416 | 5469 | int off, int size, bool strict) |
969bf05e | 5470 | { |
f1174f77 | 5471 | struct tnum reg_off; |
e07b98d9 | 5472 | int ip_align; |
d1174416 DM |
5473 | |
5474 | /* Byte size accesses are always allowed. */ | |
5475 | if (!strict || size == 1) | |
5476 | return 0; | |
5477 | ||
e4eda884 DM |
5478 | /* For platforms that do not have a Kconfig enabling |
5479 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
5480 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
5481 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
5482 | * to this code only in strict mode where we want to emulate | |
5483 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
5484 | * unconditional IP align value of '2'. | |
e07b98d9 | 5485 | */ |
e4eda884 | 5486 | ip_align = 2; |
f1174f77 EC |
5487 | |
5488 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
5489 | if (!tnum_is_aligned(reg_off, size)) { | |
5490 | char tn_buf[48]; | |
5491 | ||
5492 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
5493 | verbose(env, |
5494 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 5495 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
5496 | return -EACCES; |
5497 | } | |
79adffcd | 5498 | |
969bf05e AS |
5499 | return 0; |
5500 | } | |
5501 | ||
61bd5218 JK |
5502 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
5503 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
5504 | const char *pointer_desc, |
5505 | int off, int size, bool strict) | |
79adffcd | 5506 | { |
f1174f77 EC |
5507 | struct tnum reg_off; |
5508 | ||
5509 | /* Byte size accesses are always allowed. */ | |
5510 | if (!strict || size == 1) | |
5511 | return 0; | |
5512 | ||
5513 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
5514 | if (!tnum_is_aligned(reg_off, size)) { | |
5515 | char tn_buf[48]; | |
5516 | ||
5517 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 5518 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 5519 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
5520 | return -EACCES; |
5521 | } | |
5522 | ||
969bf05e AS |
5523 | return 0; |
5524 | } | |
5525 | ||
e07b98d9 | 5526 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
5527 | const struct bpf_reg_state *reg, int off, |
5528 | int size, bool strict_alignment_once) | |
79adffcd | 5529 | { |
ca369602 | 5530 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 5531 | const char *pointer_desc = ""; |
d1174416 | 5532 | |
79adffcd DB |
5533 | switch (reg->type) { |
5534 | case PTR_TO_PACKET: | |
de8f3a83 DB |
5535 | case PTR_TO_PACKET_META: |
5536 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
5537 | * right in front, treat it the very same way. | |
5538 | */ | |
61bd5218 | 5539 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
5540 | case PTR_TO_FLOW_KEYS: |
5541 | pointer_desc = "flow keys "; | |
5542 | break; | |
69c087ba YS |
5543 | case PTR_TO_MAP_KEY: |
5544 | pointer_desc = "key "; | |
5545 | break; | |
f1174f77 EC |
5546 | case PTR_TO_MAP_VALUE: |
5547 | pointer_desc = "value "; | |
5548 | break; | |
5549 | case PTR_TO_CTX: | |
5550 | pointer_desc = "context "; | |
5551 | break; | |
5552 | case PTR_TO_STACK: | |
5553 | pointer_desc = "stack "; | |
01f810ac AM |
5554 | /* The stack spill tracking logic in check_stack_write_fixed_off() |
5555 | * and check_stack_read_fixed_off() relies on stack accesses being | |
a5ec6ae1 JH |
5556 | * aligned. |
5557 | */ | |
5558 | strict = true; | |
f1174f77 | 5559 | break; |
c64b7983 JS |
5560 | case PTR_TO_SOCKET: |
5561 | pointer_desc = "sock "; | |
5562 | break; | |
46f8bc92 MKL |
5563 | case PTR_TO_SOCK_COMMON: |
5564 | pointer_desc = "sock_common "; | |
5565 | break; | |
655a51e5 MKL |
5566 | case PTR_TO_TCP_SOCK: |
5567 | pointer_desc = "tcp_sock "; | |
5568 | break; | |
fada7fdc JL |
5569 | case PTR_TO_XDP_SOCK: |
5570 | pointer_desc = "xdp_sock "; | |
5571 | break; | |
79adffcd | 5572 | default: |
f1174f77 | 5573 | break; |
79adffcd | 5574 | } |
61bd5218 JK |
5575 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
5576 | strict); | |
79adffcd DB |
5577 | } |
5578 | ||
f4d7e40a AS |
5579 | static int update_stack_depth(struct bpf_verifier_env *env, |
5580 | const struct bpf_func_state *func, | |
5581 | int off) | |
5582 | { | |
9c8105bd | 5583 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
5584 | |
5585 | if (stack >= -off) | |
5586 | return 0; | |
5587 | ||
5588 | /* update known max for given subprogram */ | |
9c8105bd | 5589 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
5590 | return 0; |
5591 | } | |
f4d7e40a | 5592 | |
70a87ffe AS |
5593 | /* starting from main bpf function walk all instructions of the function |
5594 | * and recursively walk all callees that given function can call. | |
5595 | * Ignore jump and exit insns. | |
5596 | * Since recursion is prevented by check_cfg() this algorithm | |
5597 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
5598 | */ | |
b5e9ad52 | 5599 | static int check_max_stack_depth_subprog(struct bpf_verifier_env *env, int idx) |
70a87ffe | 5600 | { |
9c8105bd | 5601 | struct bpf_subprog_info *subprog = env->subprog_info; |
70a87ffe | 5602 | struct bpf_insn *insn = env->prog->insnsi; |
b5e9ad52 | 5603 | int depth = 0, frame = 0, i, subprog_end; |
ebf7d1f5 | 5604 | bool tail_call_reachable = false; |
70a87ffe AS |
5605 | int ret_insn[MAX_CALL_FRAMES]; |
5606 | int ret_prog[MAX_CALL_FRAMES]; | |
ebf7d1f5 | 5607 | int j; |
f4d7e40a | 5608 | |
b5e9ad52 | 5609 | i = subprog[idx].start; |
70a87ffe | 5610 | process_func: |
7f6e4312 MF |
5611 | /* protect against potential stack overflow that might happen when |
5612 | * bpf2bpf calls get combined with tailcalls. Limit the caller's stack | |
5613 | * depth for such case down to 256 so that the worst case scenario | |
5614 | * would result in 8k stack size (32 which is tailcall limit * 256 = | |
5615 | * 8k). | |
5616 | * | |
5617 | * To get the idea what might happen, see an example: | |
5618 | * func1 -> sub rsp, 128 | |
5619 | * subfunc1 -> sub rsp, 256 | |
5620 | * tailcall1 -> add rsp, 256 | |
5621 | * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320) | |
5622 | * subfunc2 -> sub rsp, 64 | |
5623 | * subfunc22 -> sub rsp, 128 | |
5624 | * tailcall2 -> add rsp, 128 | |
5625 | * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416) | |
5626 | * | |
5627 | * tailcall will unwind the current stack frame but it will not get rid | |
5628 | * of caller's stack as shown on the example above. | |
5629 | */ | |
5630 | if (idx && subprog[idx].has_tail_call && depth >= 256) { | |
5631 | verbose(env, | |
5632 | "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n", | |
5633 | depth); | |
5634 | return -EACCES; | |
5635 | } | |
70a87ffe AS |
5636 | /* round up to 32-bytes, since this is granularity |
5637 | * of interpreter stack size | |
5638 | */ | |
9c8105bd | 5639 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 5640 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 5641 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 5642 | frame + 1, depth); |
f4d7e40a AS |
5643 | return -EACCES; |
5644 | } | |
70a87ffe | 5645 | continue_func: |
4cb3d99c | 5646 | subprog_end = subprog[idx + 1].start; |
70a87ffe | 5647 | for (; i < subprog_end; i++) { |
ba7b3e7d | 5648 | int next_insn, sidx; |
7ddc80a4 | 5649 | |
69c087ba | 5650 | if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i)) |
70a87ffe AS |
5651 | continue; |
5652 | /* remember insn and function to return to */ | |
5653 | ret_insn[frame] = i + 1; | |
9c8105bd | 5654 | ret_prog[frame] = idx; |
70a87ffe AS |
5655 | |
5656 | /* find the callee */ | |
7ddc80a4 | 5657 | next_insn = i + insn[i].imm + 1; |
ba7b3e7d KKD |
5658 | sidx = find_subprog(env, next_insn); |
5659 | if (sidx < 0) { | |
70a87ffe | 5660 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
7ddc80a4 | 5661 | next_insn); |
70a87ffe AS |
5662 | return -EFAULT; |
5663 | } | |
ba7b3e7d KKD |
5664 | if (subprog[sidx].is_async_cb) { |
5665 | if (subprog[sidx].has_tail_call) { | |
7ddc80a4 AS |
5666 | verbose(env, "verifier bug. subprog has tail_call and async cb\n"); |
5667 | return -EFAULT; | |
5668 | } | |
5415ccd5 KKD |
5669 | /* async callbacks don't increase bpf prog stack size unless called directly */ |
5670 | if (!bpf_pseudo_call(insn + i)) | |
5671 | continue; | |
7ddc80a4 AS |
5672 | } |
5673 | i = next_insn; | |
ba7b3e7d | 5674 | idx = sidx; |
ebf7d1f5 MF |
5675 | |
5676 | if (subprog[idx].has_tail_call) | |
5677 | tail_call_reachable = true; | |
5678 | ||
70a87ffe AS |
5679 | frame++; |
5680 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
5681 | verbose(env, "the call stack of %d frames is too deep !\n", |
5682 | frame); | |
5683 | return -E2BIG; | |
70a87ffe AS |
5684 | } |
5685 | goto process_func; | |
5686 | } | |
ebf7d1f5 MF |
5687 | /* if tail call got detected across bpf2bpf calls then mark each of the |
5688 | * currently present subprog frames as tail call reachable subprogs; | |
5689 | * this info will be utilized by JIT so that we will be preserving the | |
5690 | * tail call counter throughout bpf2bpf calls combined with tailcalls | |
5691 | */ | |
5692 | if (tail_call_reachable) | |
5693 | for (j = 0; j < frame; j++) | |
5694 | subprog[ret_prog[j]].tail_call_reachable = true; | |
5dd0a6b8 DB |
5695 | if (subprog[0].tail_call_reachable) |
5696 | env->prog->aux->tail_call_reachable = true; | |
ebf7d1f5 | 5697 | |
70a87ffe AS |
5698 | /* end of for() loop means the last insn of the 'subprog' |
5699 | * was reached. Doesn't matter whether it was JA or EXIT | |
5700 | */ | |
5701 | if (frame == 0) | |
5702 | return 0; | |
9c8105bd | 5703 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
5704 | frame--; |
5705 | i = ret_insn[frame]; | |
9c8105bd | 5706 | idx = ret_prog[frame]; |
70a87ffe | 5707 | goto continue_func; |
f4d7e40a AS |
5708 | } |
5709 | ||
b5e9ad52 KKD |
5710 | static int check_max_stack_depth(struct bpf_verifier_env *env) |
5711 | { | |
5712 | struct bpf_subprog_info *si = env->subprog_info; | |
5713 | int ret; | |
5714 | ||
5715 | for (int i = 0; i < env->subprog_cnt; i++) { | |
5716 | if (!i || si[i].is_async_cb) { | |
5717 | ret = check_max_stack_depth_subprog(env, i); | |
5718 | if (ret < 0) | |
5719 | return ret; | |
5720 | } | |
5721 | continue; | |
5722 | } | |
5723 | return 0; | |
5724 | } | |
5725 | ||
19d28fbd | 5726 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
5727 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
5728 | const struct bpf_insn *insn, int idx) | |
5729 | { | |
5730 | int start = idx + insn->imm + 1, subprog; | |
5731 | ||
5732 | subprog = find_subprog(env, start); | |
5733 | if (subprog < 0) { | |
5734 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
5735 | start); | |
5736 | return -EFAULT; | |
5737 | } | |
9c8105bd | 5738 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 5739 | } |
19d28fbd | 5740 | #endif |
1ea47e01 | 5741 | |
afbf21dc YS |
5742 | static int __check_buffer_access(struct bpf_verifier_env *env, |
5743 | const char *buf_info, | |
5744 | const struct bpf_reg_state *reg, | |
5745 | int regno, int off, int size) | |
9df1c28b MM |
5746 | { |
5747 | if (off < 0) { | |
5748 | verbose(env, | |
4fc00b79 | 5749 | "R%d invalid %s buffer access: off=%d, size=%d\n", |
afbf21dc | 5750 | regno, buf_info, off, size); |
9df1c28b MM |
5751 | return -EACCES; |
5752 | } | |
5753 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
5754 | char tn_buf[48]; | |
5755 | ||
5756 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
5757 | verbose(env, | |
4fc00b79 | 5758 | "R%d invalid variable buffer offset: off=%d, var_off=%s\n", |
9df1c28b MM |
5759 | regno, off, tn_buf); |
5760 | return -EACCES; | |
5761 | } | |
afbf21dc YS |
5762 | |
5763 | return 0; | |
5764 | } | |
5765 | ||
5766 | static int check_tp_buffer_access(struct bpf_verifier_env *env, | |
5767 | const struct bpf_reg_state *reg, | |
5768 | int regno, int off, int size) | |
5769 | { | |
5770 | int err; | |
5771 | ||
5772 | err = __check_buffer_access(env, "tracepoint", reg, regno, off, size); | |
5773 | if (err) | |
5774 | return err; | |
5775 | ||
9df1c28b MM |
5776 | if (off + size > env->prog->aux->max_tp_access) |
5777 | env->prog->aux->max_tp_access = off + size; | |
5778 | ||
5779 | return 0; | |
5780 | } | |
5781 | ||
afbf21dc YS |
5782 | static int check_buffer_access(struct bpf_verifier_env *env, |
5783 | const struct bpf_reg_state *reg, | |
5784 | int regno, int off, int size, | |
5785 | bool zero_size_allowed, | |
afbf21dc YS |
5786 | u32 *max_access) |
5787 | { | |
44e9a741 | 5788 | const char *buf_info = type_is_rdonly_mem(reg->type) ? "rdonly" : "rdwr"; |
afbf21dc YS |
5789 | int err; |
5790 | ||
5791 | err = __check_buffer_access(env, buf_info, reg, regno, off, size); | |
5792 | if (err) | |
5793 | return err; | |
5794 | ||
5795 | if (off + size > *max_access) | |
5796 | *max_access = off + size; | |
5797 | ||
5798 | return 0; | |
5799 | } | |
5800 | ||
3f50f132 JF |
5801 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
5802 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
5803 | { | |
5804 | reg->var_off = tnum_subreg(reg->var_off); | |
5805 | __reg_assign_32_into_64(reg); | |
5806 | } | |
9df1c28b | 5807 | |
0c17d1d2 JH |
5808 | /* truncate register to smaller size (in bytes) |
5809 | * must be called with size < BPF_REG_SIZE | |
5810 | */ | |
5811 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
5812 | { | |
5813 | u64 mask; | |
5814 | ||
5815 | /* clear high bits in bit representation */ | |
5816 | reg->var_off = tnum_cast(reg->var_off, size); | |
5817 | ||
5818 | /* fix arithmetic bounds */ | |
5819 | mask = ((u64)1 << (size * 8)) - 1; | |
5820 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
5821 | reg->umin_value &= mask; | |
5822 | reg->umax_value &= mask; | |
5823 | } else { | |
5824 | reg->umin_value = 0; | |
5825 | reg->umax_value = mask; | |
5826 | } | |
5827 | reg->smin_value = reg->umin_value; | |
5828 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
5829 | |
5830 | /* If size is smaller than 32bit register the 32bit register | |
5831 | * values are also truncated so we push 64-bit bounds into | |
5832 | * 32-bit bounds. Above were truncated < 32-bits already. | |
5833 | */ | |
5834 | if (size >= 4) | |
5835 | return; | |
5836 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
5837 | } |
5838 | ||
1f9a1ea8 YS |
5839 | static void set_sext64_default_val(struct bpf_reg_state *reg, int size) |
5840 | { | |
5841 | if (size == 1) { | |
5842 | reg->smin_value = reg->s32_min_value = S8_MIN; | |
5843 | reg->smax_value = reg->s32_max_value = S8_MAX; | |
5844 | } else if (size == 2) { | |
5845 | reg->smin_value = reg->s32_min_value = S16_MIN; | |
5846 | reg->smax_value = reg->s32_max_value = S16_MAX; | |
5847 | } else { | |
5848 | /* size == 4 */ | |
5849 | reg->smin_value = reg->s32_min_value = S32_MIN; | |
5850 | reg->smax_value = reg->s32_max_value = S32_MAX; | |
5851 | } | |
5852 | reg->umin_value = reg->u32_min_value = 0; | |
5853 | reg->umax_value = U64_MAX; | |
5854 | reg->u32_max_value = U32_MAX; | |
5855 | reg->var_off = tnum_unknown; | |
5856 | } | |
5857 | ||
5858 | static void coerce_reg_to_size_sx(struct bpf_reg_state *reg, int size) | |
5859 | { | |
5860 | s64 init_s64_max, init_s64_min, s64_max, s64_min, u64_cval; | |
5861 | u64 top_smax_value, top_smin_value; | |
5862 | u64 num_bits = size * 8; | |
5863 | ||
5864 | if (tnum_is_const(reg->var_off)) { | |
5865 | u64_cval = reg->var_off.value; | |
5866 | if (size == 1) | |
5867 | reg->var_off = tnum_const((s8)u64_cval); | |
5868 | else if (size == 2) | |
5869 | reg->var_off = tnum_const((s16)u64_cval); | |
5870 | else | |
5871 | /* size == 4 */ | |
5872 | reg->var_off = tnum_const((s32)u64_cval); | |
5873 | ||
5874 | u64_cval = reg->var_off.value; | |
5875 | reg->smax_value = reg->smin_value = u64_cval; | |
5876 | reg->umax_value = reg->umin_value = u64_cval; | |
5877 | reg->s32_max_value = reg->s32_min_value = u64_cval; | |
5878 | reg->u32_max_value = reg->u32_min_value = u64_cval; | |
5879 | return; | |
5880 | } | |
5881 | ||
5882 | top_smax_value = ((u64)reg->smax_value >> num_bits) << num_bits; | |
5883 | top_smin_value = ((u64)reg->smin_value >> num_bits) << num_bits; | |
5884 | ||
5885 | if (top_smax_value != top_smin_value) | |
5886 | goto out; | |
5887 | ||
5888 | /* find the s64_min and s64_min after sign extension */ | |
5889 | if (size == 1) { | |
5890 | init_s64_max = (s8)reg->smax_value; | |
5891 | init_s64_min = (s8)reg->smin_value; | |
5892 | } else if (size == 2) { | |
5893 | init_s64_max = (s16)reg->smax_value; | |
5894 | init_s64_min = (s16)reg->smin_value; | |
5895 | } else { | |
5896 | init_s64_max = (s32)reg->smax_value; | |
5897 | init_s64_min = (s32)reg->smin_value; | |
5898 | } | |
5899 | ||
5900 | s64_max = max(init_s64_max, init_s64_min); | |
5901 | s64_min = min(init_s64_max, init_s64_min); | |
5902 | ||
5903 | /* both of s64_max/s64_min positive or negative */ | |
09fedc73 | 5904 | if ((s64_max >= 0) == (s64_min >= 0)) { |
1f9a1ea8 YS |
5905 | reg->smin_value = reg->s32_min_value = s64_min; |
5906 | reg->smax_value = reg->s32_max_value = s64_max; | |
5907 | reg->umin_value = reg->u32_min_value = s64_min; | |
5908 | reg->umax_value = reg->u32_max_value = s64_max; | |
5909 | reg->var_off = tnum_range(s64_min, s64_max); | |
5910 | return; | |
5911 | } | |
5912 | ||
5913 | out: | |
5914 | set_sext64_default_val(reg, size); | |
5915 | } | |
5916 | ||
8100928c YS |
5917 | static void set_sext32_default_val(struct bpf_reg_state *reg, int size) |
5918 | { | |
5919 | if (size == 1) { | |
5920 | reg->s32_min_value = S8_MIN; | |
5921 | reg->s32_max_value = S8_MAX; | |
5922 | } else { | |
5923 | /* size == 2 */ | |
5924 | reg->s32_min_value = S16_MIN; | |
5925 | reg->s32_max_value = S16_MAX; | |
5926 | } | |
5927 | reg->u32_min_value = 0; | |
5928 | reg->u32_max_value = U32_MAX; | |
5929 | } | |
5930 | ||
5931 | static void coerce_subreg_to_size_sx(struct bpf_reg_state *reg, int size) | |
5932 | { | |
5933 | s32 init_s32_max, init_s32_min, s32_max, s32_min, u32_val; | |
5934 | u32 top_smax_value, top_smin_value; | |
5935 | u32 num_bits = size * 8; | |
5936 | ||
5937 | if (tnum_is_const(reg->var_off)) { | |
5938 | u32_val = reg->var_off.value; | |
5939 | if (size == 1) | |
5940 | reg->var_off = tnum_const((s8)u32_val); | |
5941 | else | |
5942 | reg->var_off = tnum_const((s16)u32_val); | |
5943 | ||
5944 | u32_val = reg->var_off.value; | |
5945 | reg->s32_min_value = reg->s32_max_value = u32_val; | |
5946 | reg->u32_min_value = reg->u32_max_value = u32_val; | |
5947 | return; | |
5948 | } | |
5949 | ||
5950 | top_smax_value = ((u32)reg->s32_max_value >> num_bits) << num_bits; | |
5951 | top_smin_value = ((u32)reg->s32_min_value >> num_bits) << num_bits; | |
5952 | ||
5953 | if (top_smax_value != top_smin_value) | |
5954 | goto out; | |
5955 | ||
5956 | /* find the s32_min and s32_min after sign extension */ | |
5957 | if (size == 1) { | |
5958 | init_s32_max = (s8)reg->s32_max_value; | |
5959 | init_s32_min = (s8)reg->s32_min_value; | |
5960 | } else { | |
5961 | /* size == 2 */ | |
5962 | init_s32_max = (s16)reg->s32_max_value; | |
5963 | init_s32_min = (s16)reg->s32_min_value; | |
5964 | } | |
5965 | s32_max = max(init_s32_max, init_s32_min); | |
5966 | s32_min = min(init_s32_max, init_s32_min); | |
5967 | ||
09fedc73 | 5968 | if ((s32_min >= 0) == (s32_max >= 0)) { |
8100928c YS |
5969 | reg->s32_min_value = s32_min; |
5970 | reg->s32_max_value = s32_max; | |
5971 | reg->u32_min_value = (u32)s32_min; | |
5972 | reg->u32_max_value = (u32)s32_max; | |
5973 | return; | |
5974 | } | |
5975 | ||
5976 | out: | |
5977 | set_sext32_default_val(reg, size); | |
5978 | } | |
5979 | ||
a23740ec AN |
5980 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
5981 | { | |
353050be DB |
5982 | /* A map is considered read-only if the following condition are true: |
5983 | * | |
5984 | * 1) BPF program side cannot change any of the map content. The | |
5985 | * BPF_F_RDONLY_PROG flag is throughout the lifetime of a map | |
5986 | * and was set at map creation time. | |
5987 | * 2) The map value(s) have been initialized from user space by a | |
5988 | * loader and then "frozen", such that no new map update/delete | |
5989 | * operations from syscall side are possible for the rest of | |
5990 | * the map's lifetime from that point onwards. | |
5991 | * 3) Any parallel/pending map update/delete operations from syscall | |
5992 | * side have been completed. Only after that point, it's safe to | |
5993 | * assume that map value(s) are immutable. | |
5994 | */ | |
5995 | return (map->map_flags & BPF_F_RDONLY_PROG) && | |
5996 | READ_ONCE(map->frozen) && | |
5997 | !bpf_map_write_active(map); | |
a23740ec AN |
5998 | } |
5999 | ||
1f9a1ea8 YS |
6000 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val, |
6001 | bool is_ldsx) | |
a23740ec AN |
6002 | { |
6003 | void *ptr; | |
6004 | u64 addr; | |
6005 | int err; | |
6006 | ||
6007 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
6008 | if (err) | |
6009 | return err; | |
2dedd7d2 | 6010 | ptr = (void *)(long)addr + off; |
a23740ec AN |
6011 | |
6012 | switch (size) { | |
6013 | case sizeof(u8): | |
1f9a1ea8 | 6014 | *val = is_ldsx ? (s64)*(s8 *)ptr : (u64)*(u8 *)ptr; |
a23740ec AN |
6015 | break; |
6016 | case sizeof(u16): | |
1f9a1ea8 | 6017 | *val = is_ldsx ? (s64)*(s16 *)ptr : (u64)*(u16 *)ptr; |
a23740ec AN |
6018 | break; |
6019 | case sizeof(u32): | |
1f9a1ea8 | 6020 | *val = is_ldsx ? (s64)*(s32 *)ptr : (u64)*(u32 *)ptr; |
a23740ec AN |
6021 | break; |
6022 | case sizeof(u64): | |
6023 | *val = *(u64 *)ptr; | |
6024 | break; | |
6025 | default: | |
6026 | return -EINVAL; | |
6027 | } | |
6028 | return 0; | |
6029 | } | |
6030 | ||
6fcd486b | 6031 | #define BTF_TYPE_SAFE_RCU(__type) __PASTE(__type, __safe_rcu) |
30ee9821 | 6032 | #define BTF_TYPE_SAFE_RCU_OR_NULL(__type) __PASTE(__type, __safe_rcu_or_null) |
6fcd486b | 6033 | #define BTF_TYPE_SAFE_TRUSTED(__type) __PASTE(__type, __safe_trusted) |
57539b1c | 6034 | |
6fcd486b AS |
6035 | /* |
6036 | * Allow list few fields as RCU trusted or full trusted. | |
6037 | * This logic doesn't allow mix tagging and will be removed once GCC supports | |
6038 | * btf_type_tag. | |
6039 | */ | |
6040 | ||
6041 | /* RCU trusted: these fields are trusted in RCU CS and never NULL */ | |
6042 | BTF_TYPE_SAFE_RCU(struct task_struct) { | |
57539b1c | 6043 | const cpumask_t *cpus_ptr; |
8d093b4e | 6044 | struct css_set __rcu *cgroups; |
6fcd486b AS |
6045 | struct task_struct __rcu *real_parent; |
6046 | struct task_struct *group_leader; | |
8d093b4e AS |
6047 | }; |
6048 | ||
30ee9821 AS |
6049 | BTF_TYPE_SAFE_RCU(struct cgroup) { |
6050 | /* cgrp->kn is always accessible as documented in kernel/cgroup/cgroup.c */ | |
6051 | struct kernfs_node *kn; | |
6052 | }; | |
6053 | ||
6fcd486b | 6054 | BTF_TYPE_SAFE_RCU(struct css_set) { |
8d093b4e | 6055 | struct cgroup *dfl_cgrp; |
57539b1c DV |
6056 | }; |
6057 | ||
30ee9821 AS |
6058 | /* RCU trusted: these fields are trusted in RCU CS and can be NULL */ |
6059 | BTF_TYPE_SAFE_RCU_OR_NULL(struct mm_struct) { | |
6060 | struct file __rcu *exe_file; | |
6061 | }; | |
6062 | ||
6063 | /* skb->sk, req->sk are not RCU protected, but we mark them as such | |
6064 | * because bpf prog accessible sockets are SOCK_RCU_FREE. | |
6065 | */ | |
6066 | BTF_TYPE_SAFE_RCU_OR_NULL(struct sk_buff) { | |
6067 | struct sock *sk; | |
6068 | }; | |
6069 | ||
6070 | BTF_TYPE_SAFE_RCU_OR_NULL(struct request_sock) { | |
6071 | struct sock *sk; | |
6072 | }; | |
6073 | ||
6fcd486b AS |
6074 | /* full trusted: these fields are trusted even outside of RCU CS and never NULL */ |
6075 | BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta) { | |
63260df1 | 6076 | struct seq_file *seq; |
6fcd486b AS |
6077 | }; |
6078 | ||
6079 | BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task) { | |
63260df1 AS |
6080 | struct bpf_iter_meta *meta; |
6081 | struct task_struct *task; | |
6fcd486b AS |
6082 | }; |
6083 | ||
6084 | BTF_TYPE_SAFE_TRUSTED(struct linux_binprm) { | |
6085 | struct file *file; | |
6086 | }; | |
6087 | ||
6088 | BTF_TYPE_SAFE_TRUSTED(struct file) { | |
6089 | struct inode *f_inode; | |
6090 | }; | |
6091 | ||
6092 | BTF_TYPE_SAFE_TRUSTED(struct dentry) { | |
6093 | /* no negative dentry-s in places where bpf can see it */ | |
6094 | struct inode *d_inode; | |
6095 | }; | |
6096 | ||
6097 | BTF_TYPE_SAFE_TRUSTED(struct socket) { | |
6098 | struct sock *sk; | |
6099 | }; | |
6100 | ||
6101 | static bool type_is_rcu(struct bpf_verifier_env *env, | |
6102 | struct bpf_reg_state *reg, | |
63260df1 | 6103 | const char *field_name, u32 btf_id) |
57539b1c | 6104 | { |
6fcd486b | 6105 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct task_struct)); |
30ee9821 | 6106 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct cgroup)); |
6fcd486b | 6107 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU(struct css_set)); |
57539b1c | 6108 | |
63260df1 | 6109 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_rcu"); |
6fcd486b | 6110 | } |
57539b1c | 6111 | |
30ee9821 AS |
6112 | static bool type_is_rcu_or_null(struct bpf_verifier_env *env, |
6113 | struct bpf_reg_state *reg, | |
6114 | const char *field_name, u32 btf_id) | |
6115 | { | |
6116 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct mm_struct)); | |
6117 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct sk_buff)); | |
6118 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_RCU_OR_NULL(struct request_sock)); | |
6119 | ||
6120 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_rcu_or_null"); | |
6121 | } | |
6122 | ||
6fcd486b AS |
6123 | static bool type_is_trusted(struct bpf_verifier_env *env, |
6124 | struct bpf_reg_state *reg, | |
63260df1 | 6125 | const char *field_name, u32 btf_id) |
6fcd486b AS |
6126 | { |
6127 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter_meta)); | |
6128 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct bpf_iter__task)); | |
6129 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct linux_binprm)); | |
6130 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct file)); | |
6131 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct dentry)); | |
6132 | BTF_TYPE_EMIT(BTF_TYPE_SAFE_TRUSTED(struct socket)); | |
6133 | ||
63260df1 | 6134 | return btf_nested_type_is_trusted(&env->log, reg, field_name, btf_id, "__safe_trusted"); |
57539b1c DV |
6135 | } |
6136 | ||
9e15db66 AS |
6137 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
6138 | struct bpf_reg_state *regs, | |
6139 | int regno, int off, int size, | |
6140 | enum bpf_access_type atype, | |
6141 | int value_regno) | |
6142 | { | |
6143 | struct bpf_reg_state *reg = regs + regno; | |
22dc4a0f AN |
6144 | const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id); |
6145 | const char *tname = btf_name_by_offset(reg->btf, t->name_off); | |
63260df1 | 6146 | const char *field_name = NULL; |
c6f1bfe8 | 6147 | enum bpf_type_flag flag = 0; |
b7e852a9 | 6148 | u32 btf_id = 0; |
9e15db66 AS |
6149 | int ret; |
6150 | ||
c67cae55 AS |
6151 | if (!env->allow_ptr_leaks) { |
6152 | verbose(env, | |
6153 | "'struct %s' access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", | |
6154 | tname); | |
6155 | return -EPERM; | |
6156 | } | |
6157 | if (!env->prog->gpl_compatible && btf_is_kernel(reg->btf)) { | |
6158 | verbose(env, | |
6159 | "Cannot access kernel 'struct %s' from non-GPL compatible program\n", | |
6160 | tname); | |
6161 | return -EINVAL; | |
6162 | } | |
9e15db66 AS |
6163 | if (off < 0) { |
6164 | verbose(env, | |
6165 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
6166 | regno, tname, off); | |
6167 | return -EACCES; | |
6168 | } | |
6169 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
6170 | char tn_buf[48]; | |
6171 | ||
6172 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6173 | verbose(env, | |
6174 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
6175 | regno, tname, off, tn_buf); | |
6176 | return -EACCES; | |
6177 | } | |
6178 | ||
c6f1bfe8 YS |
6179 | if (reg->type & MEM_USER) { |
6180 | verbose(env, | |
6181 | "R%d is ptr_%s access user memory: off=%d\n", | |
6182 | regno, tname, off); | |
6183 | return -EACCES; | |
6184 | } | |
6185 | ||
5844101a HL |
6186 | if (reg->type & MEM_PERCPU) { |
6187 | verbose(env, | |
6188 | "R%d is ptr_%s access percpu memory: off=%d\n", | |
6189 | regno, tname, off); | |
6190 | return -EACCES; | |
6191 | } | |
6192 | ||
7d64c513 | 6193 | if (env->ops->btf_struct_access && !type_is_alloc(reg->type) && atype == BPF_WRITE) { |
282de143 KKD |
6194 | if (!btf_is_kernel(reg->btf)) { |
6195 | verbose(env, "verifier internal error: reg->btf must be kernel btf\n"); | |
6196 | return -EFAULT; | |
6197 | } | |
b7e852a9 | 6198 | ret = env->ops->btf_struct_access(&env->log, reg, off, size); |
27ae7997 | 6199 | } else { |
282de143 KKD |
6200 | /* Writes are permitted with default btf_struct_access for |
6201 | * program allocated objects (which always have ref_obj_id > 0), | |
6202 | * but not for untrusted PTR_TO_BTF_ID | MEM_ALLOC. | |
6203 | */ | |
503e4def | 6204 | if (atype != BPF_READ && !type_is_ptr_alloc_obj(reg->type)) { |
27ae7997 MKL |
6205 | verbose(env, "only read is supported\n"); |
6206 | return -EACCES; | |
6207 | } | |
6208 | ||
6a3cd331 DM |
6209 | if (type_is_alloc(reg->type) && !type_is_non_owning_ref(reg->type) && |
6210 | !reg->ref_obj_id) { | |
282de143 KKD |
6211 | verbose(env, "verifier internal error: ref_obj_id for allocated object must be non-zero\n"); |
6212 | return -EFAULT; | |
6213 | } | |
6214 | ||
63260df1 | 6215 | ret = btf_struct_access(&env->log, reg, off, size, atype, &btf_id, &flag, &field_name); |
27ae7997 MKL |
6216 | } |
6217 | ||
9e15db66 AS |
6218 | if (ret < 0) |
6219 | return ret; | |
6220 | ||
6fcd486b AS |
6221 | if (ret != PTR_TO_BTF_ID) { |
6222 | /* just mark; */ | |
6efe152d | 6223 | |
6fcd486b AS |
6224 | } else if (type_flag(reg->type) & PTR_UNTRUSTED) { |
6225 | /* If this is an untrusted pointer, all pointers formed by walking it | |
6226 | * also inherit the untrusted flag. | |
6227 | */ | |
6228 | flag = PTR_UNTRUSTED; | |
6229 | ||
6230 | } else if (is_trusted_reg(reg) || is_rcu_reg(reg)) { | |
6231 | /* By default any pointer obtained from walking a trusted pointer is no | |
6232 | * longer trusted, unless the field being accessed has explicitly been | |
6233 | * marked as inheriting its parent's state of trust (either full or RCU). | |
6234 | * For example: | |
6235 | * 'cgroups' pointer is untrusted if task->cgroups dereference | |
6236 | * happened in a sleepable program outside of bpf_rcu_read_lock() | |
6237 | * section. In a non-sleepable program it's trusted while in RCU CS (aka MEM_RCU). | |
6238 | * Note bpf_rcu_read_unlock() converts MEM_RCU pointers to PTR_UNTRUSTED. | |
6239 | * | |
6240 | * A regular RCU-protected pointer with __rcu tag can also be deemed | |
6241 | * trusted if we are in an RCU CS. Such pointer can be NULL. | |
20c09d92 | 6242 | */ |
63260df1 | 6243 | if (type_is_trusted(env, reg, field_name, btf_id)) { |
6fcd486b AS |
6244 | flag |= PTR_TRUSTED; |
6245 | } else if (in_rcu_cs(env) && !type_may_be_null(reg->type)) { | |
63260df1 | 6246 | if (type_is_rcu(env, reg, field_name, btf_id)) { |
6fcd486b AS |
6247 | /* ignore __rcu tag and mark it MEM_RCU */ |
6248 | flag |= MEM_RCU; | |
30ee9821 AS |
6249 | } else if (flag & MEM_RCU || |
6250 | type_is_rcu_or_null(env, reg, field_name, btf_id)) { | |
6fcd486b | 6251 | /* __rcu tagged pointers can be NULL */ |
30ee9821 | 6252 | flag |= MEM_RCU | PTR_MAYBE_NULL; |
7ce4dc3e YS |
6253 | |
6254 | /* We always trust them */ | |
6255 | if (type_is_rcu_or_null(env, reg, field_name, btf_id) && | |
6256 | flag & PTR_UNTRUSTED) | |
6257 | flag &= ~PTR_UNTRUSTED; | |
6fcd486b AS |
6258 | } else if (flag & (MEM_PERCPU | MEM_USER)) { |
6259 | /* keep as-is */ | |
6260 | } else { | |
afeebf9f AS |
6261 | /* walking unknown pointers yields old deprecated PTR_TO_BTF_ID */ |
6262 | clear_trusted_flags(&flag); | |
6fcd486b AS |
6263 | } |
6264 | } else { | |
6265 | /* | |
6266 | * If not in RCU CS or MEM_RCU pointer can be NULL then | |
6267 | * aggressively mark as untrusted otherwise such | |
6268 | * pointers will be plain PTR_TO_BTF_ID without flags | |
6269 | * and will be allowed to be passed into helpers for | |
6270 | * compat reasons. | |
6271 | */ | |
6272 | flag = PTR_UNTRUSTED; | |
6273 | } | |
20c09d92 | 6274 | } else { |
6fcd486b | 6275 | /* Old compat. Deprecated */ |
afeebf9f | 6276 | clear_trusted_flags(&flag); |
20c09d92 | 6277 | } |
3f00c523 | 6278 | |
41c48f3a | 6279 | if (atype == BPF_READ && value_regno >= 0) |
c6f1bfe8 | 6280 | mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id, flag); |
41c48f3a AI |
6281 | |
6282 | return 0; | |
6283 | } | |
6284 | ||
6285 | static int check_ptr_to_map_access(struct bpf_verifier_env *env, | |
6286 | struct bpf_reg_state *regs, | |
6287 | int regno, int off, int size, | |
6288 | enum bpf_access_type atype, | |
6289 | int value_regno) | |
6290 | { | |
6291 | struct bpf_reg_state *reg = regs + regno; | |
6292 | struct bpf_map *map = reg->map_ptr; | |
6728aea7 | 6293 | struct bpf_reg_state map_reg; |
c6f1bfe8 | 6294 | enum bpf_type_flag flag = 0; |
41c48f3a AI |
6295 | const struct btf_type *t; |
6296 | const char *tname; | |
6297 | u32 btf_id; | |
6298 | int ret; | |
6299 | ||
6300 | if (!btf_vmlinux) { | |
6301 | verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
6302 | return -ENOTSUPP; | |
6303 | } | |
6304 | ||
6305 | if (!map->ops->map_btf_id || !*map->ops->map_btf_id) { | |
6306 | verbose(env, "map_ptr access not supported for map type %d\n", | |
6307 | map->map_type); | |
6308 | return -ENOTSUPP; | |
6309 | } | |
6310 | ||
6311 | t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id); | |
6312 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
6313 | ||
c67cae55 | 6314 | if (!env->allow_ptr_leaks) { |
41c48f3a | 6315 | verbose(env, |
c67cae55 | 6316 | "'struct %s' access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", |
41c48f3a AI |
6317 | tname); |
6318 | return -EPERM; | |
9e15db66 | 6319 | } |
27ae7997 | 6320 | |
41c48f3a AI |
6321 | if (off < 0) { |
6322 | verbose(env, "R%d is %s invalid negative access: off=%d\n", | |
6323 | regno, tname, off); | |
6324 | return -EACCES; | |
6325 | } | |
6326 | ||
6327 | if (atype != BPF_READ) { | |
6328 | verbose(env, "only read from %s is supported\n", tname); | |
6329 | return -EACCES; | |
6330 | } | |
6331 | ||
6728aea7 KKD |
6332 | /* Simulate access to a PTR_TO_BTF_ID */ |
6333 | memset(&map_reg, 0, sizeof(map_reg)); | |
6334 | mark_btf_ld_reg(env, &map_reg, 0, PTR_TO_BTF_ID, btf_vmlinux, *map->ops->map_btf_id, 0); | |
63260df1 | 6335 | ret = btf_struct_access(&env->log, &map_reg, off, size, atype, &btf_id, &flag, NULL); |
41c48f3a AI |
6336 | if (ret < 0) |
6337 | return ret; | |
6338 | ||
6339 | if (value_regno >= 0) | |
c6f1bfe8 | 6340 | mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id, flag); |
41c48f3a | 6341 | |
9e15db66 AS |
6342 | return 0; |
6343 | } | |
6344 | ||
01f810ac AM |
6345 | /* Check that the stack access at the given offset is within bounds. The |
6346 | * maximum valid offset is -1. | |
6347 | * | |
6348 | * The minimum valid offset is -MAX_BPF_STACK for writes, and | |
6349 | * -state->allocated_stack for reads. | |
6350 | */ | |
6351 | static int check_stack_slot_within_bounds(int off, | |
6352 | struct bpf_func_state *state, | |
6353 | enum bpf_access_type t) | |
6354 | { | |
6355 | int min_valid_off; | |
6356 | ||
6357 | if (t == BPF_WRITE) | |
6358 | min_valid_off = -MAX_BPF_STACK; | |
6359 | else | |
6360 | min_valid_off = -state->allocated_stack; | |
6361 | ||
6362 | if (off < min_valid_off || off > -1) | |
6363 | return -EACCES; | |
6364 | return 0; | |
6365 | } | |
6366 | ||
6367 | /* Check that the stack access at 'regno + off' falls within the maximum stack | |
6368 | * bounds. | |
6369 | * | |
6370 | * 'off' includes `regno->offset`, but not its dynamic part (if any). | |
6371 | */ | |
6372 | static int check_stack_access_within_bounds( | |
6373 | struct bpf_verifier_env *env, | |
6374 | int regno, int off, int access_size, | |
61df10c7 | 6375 | enum bpf_access_src src, enum bpf_access_type type) |
01f810ac AM |
6376 | { |
6377 | struct bpf_reg_state *regs = cur_regs(env); | |
6378 | struct bpf_reg_state *reg = regs + regno; | |
6379 | struct bpf_func_state *state = func(env, reg); | |
6380 | int min_off, max_off; | |
6381 | int err; | |
6382 | char *err_extra; | |
6383 | ||
6384 | if (src == ACCESS_HELPER) | |
6385 | /* We don't know if helpers are reading or writing (or both). */ | |
6386 | err_extra = " indirect access to"; | |
6387 | else if (type == BPF_READ) | |
6388 | err_extra = " read from"; | |
6389 | else | |
6390 | err_extra = " write to"; | |
6391 | ||
6392 | if (tnum_is_const(reg->var_off)) { | |
6393 | min_off = reg->var_off.value + off; | |
6394 | if (access_size > 0) | |
6395 | max_off = min_off + access_size - 1; | |
6396 | else | |
6397 | max_off = min_off; | |
6398 | } else { | |
6399 | if (reg->smax_value >= BPF_MAX_VAR_OFF || | |
6400 | reg->smin_value <= -BPF_MAX_VAR_OFF) { | |
6401 | verbose(env, "invalid unbounded variable-offset%s stack R%d\n", | |
6402 | err_extra, regno); | |
6403 | return -EACCES; | |
6404 | } | |
6405 | min_off = reg->smin_value + off; | |
6406 | if (access_size > 0) | |
6407 | max_off = reg->smax_value + off + access_size - 1; | |
6408 | else | |
6409 | max_off = min_off; | |
6410 | } | |
6411 | ||
6412 | err = check_stack_slot_within_bounds(min_off, state, type); | |
6413 | if (!err) | |
6414 | err = check_stack_slot_within_bounds(max_off, state, type); | |
6415 | ||
6416 | if (err) { | |
6417 | if (tnum_is_const(reg->var_off)) { | |
6418 | verbose(env, "invalid%s stack R%d off=%d size=%d\n", | |
6419 | err_extra, regno, off, access_size); | |
6420 | } else { | |
6421 | char tn_buf[48]; | |
6422 | ||
6423 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6424 | verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n", | |
6425 | err_extra, regno, tn_buf, access_size); | |
6426 | } | |
6427 | } | |
6428 | return err; | |
6429 | } | |
41c48f3a | 6430 | |
17a52670 AS |
6431 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
6432 | * if t==write, value_regno is a register which value is stored into memory | |
6433 | * if t==read, value_regno is a register which will receive the value from memory | |
6434 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
6435 | * if t==read && value_regno==-1, don't care what we read from memory | |
6436 | */ | |
ca369602 DB |
6437 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
6438 | int off, int bpf_size, enum bpf_access_type t, | |
1f9a1ea8 | 6439 | int value_regno, bool strict_alignment_once, bool is_ldsx) |
17a52670 | 6440 | { |
638f5b90 AS |
6441 | struct bpf_reg_state *regs = cur_regs(env); |
6442 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 6443 | struct bpf_func_state *state; |
17a52670 AS |
6444 | int size, err = 0; |
6445 | ||
6446 | size = bpf_size_to_bytes(bpf_size); | |
6447 | if (size < 0) | |
6448 | return size; | |
6449 | ||
f1174f77 | 6450 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 6451 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
6452 | if (err) |
6453 | return err; | |
17a52670 | 6454 | |
f1174f77 EC |
6455 | /* for access checks, reg->off is just part of off */ |
6456 | off += reg->off; | |
6457 | ||
69c087ba YS |
6458 | if (reg->type == PTR_TO_MAP_KEY) { |
6459 | if (t == BPF_WRITE) { | |
6460 | verbose(env, "write to change key R%d not allowed\n", regno); | |
6461 | return -EACCES; | |
6462 | } | |
6463 | ||
6464 | err = check_mem_region_access(env, regno, off, size, | |
6465 | reg->map_ptr->key_size, false); | |
6466 | if (err) | |
6467 | return err; | |
6468 | if (value_regno >= 0) | |
6469 | mark_reg_unknown(env, regs, value_regno); | |
6470 | } else if (reg->type == PTR_TO_MAP_VALUE) { | |
aa3496ac | 6471 | struct btf_field *kptr_field = NULL; |
61df10c7 | 6472 | |
1be7f75d AS |
6473 | if (t == BPF_WRITE && value_regno >= 0 && |
6474 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 6475 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
6476 | return -EACCES; |
6477 | } | |
591fe988 DB |
6478 | err = check_map_access_type(env, regno, off, size, t); |
6479 | if (err) | |
6480 | return err; | |
61df10c7 KKD |
6481 | err = check_map_access(env, regno, off, size, false, ACCESS_DIRECT); |
6482 | if (err) | |
6483 | return err; | |
6484 | if (tnum_is_const(reg->var_off)) | |
aa3496ac KKD |
6485 | kptr_field = btf_record_find(reg->map_ptr->record, |
6486 | off + reg->var_off.value, BPF_KPTR); | |
6487 | if (kptr_field) { | |
6488 | err = check_map_kptr_access(env, regno, value_regno, insn_idx, kptr_field); | |
61df10c7 | 6489 | } else if (t == BPF_READ && value_regno >= 0) { |
a23740ec AN |
6490 | struct bpf_map *map = reg->map_ptr; |
6491 | ||
6492 | /* if map is read-only, track its contents as scalars */ | |
6493 | if (tnum_is_const(reg->var_off) && | |
6494 | bpf_map_is_rdonly(map) && | |
6495 | map->ops->map_direct_value_addr) { | |
6496 | int map_off = off + reg->var_off.value; | |
6497 | u64 val = 0; | |
6498 | ||
6499 | err = bpf_map_direct_read(map, map_off, size, | |
1f9a1ea8 | 6500 | &val, is_ldsx); |
a23740ec AN |
6501 | if (err) |
6502 | return err; | |
6503 | ||
6504 | regs[value_regno].type = SCALAR_VALUE; | |
6505 | __mark_reg_known(®s[value_regno], val); | |
6506 | } else { | |
6507 | mark_reg_unknown(env, regs, value_regno); | |
6508 | } | |
6509 | } | |
34d3a78c HL |
6510 | } else if (base_type(reg->type) == PTR_TO_MEM) { |
6511 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
6512 | ||
6513 | if (type_may_be_null(reg->type)) { | |
6514 | verbose(env, "R%d invalid mem access '%s'\n", regno, | |
6515 | reg_type_str(env, reg->type)); | |
6516 | return -EACCES; | |
6517 | } | |
6518 | ||
6519 | if (t == BPF_WRITE && rdonly_mem) { | |
6520 | verbose(env, "R%d cannot write into %s\n", | |
6521 | regno, reg_type_str(env, reg->type)); | |
6522 | return -EACCES; | |
6523 | } | |
6524 | ||
457f4436 AN |
6525 | if (t == BPF_WRITE && value_regno >= 0 && |
6526 | is_pointer_value(env, value_regno)) { | |
6527 | verbose(env, "R%d leaks addr into mem\n", value_regno); | |
6528 | return -EACCES; | |
6529 | } | |
34d3a78c | 6530 | |
457f4436 AN |
6531 | err = check_mem_region_access(env, regno, off, size, |
6532 | reg->mem_size, false); | |
34d3a78c | 6533 | if (!err && value_regno >= 0 && (t == BPF_READ || rdonly_mem)) |
457f4436 | 6534 | mark_reg_unknown(env, regs, value_regno); |
1a0dc1ac | 6535 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 6536 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
22dc4a0f | 6537 | struct btf *btf = NULL; |
9e15db66 | 6538 | u32 btf_id = 0; |
19de99f7 | 6539 | |
1be7f75d AS |
6540 | if (t == BPF_WRITE && value_regno >= 0 && |
6541 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 6542 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
6543 | return -EACCES; |
6544 | } | |
f1174f77 | 6545 | |
be80a1d3 | 6546 | err = check_ptr_off_reg(env, reg, regno); |
58990d1f DB |
6547 | if (err < 0) |
6548 | return err; | |
6549 | ||
c6f1bfe8 YS |
6550 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf, |
6551 | &btf_id); | |
9e15db66 AS |
6552 | if (err) |
6553 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 6554 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 6555 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
6556 | * PTR_TO_PACKET[_META,_END]. In the latter |
6557 | * case, we know the offset is zero. | |
f1174f77 | 6558 | */ |
46f8bc92 | 6559 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 6560 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 6561 | } else { |
638f5b90 | 6562 | mark_reg_known_zero(env, regs, |
61bd5218 | 6563 | value_regno); |
c25b2ae1 | 6564 | if (type_may_be_null(reg_type)) |
46f8bc92 | 6565 | regs[value_regno].id = ++env->id_gen; |
5327ed3d JW |
6566 | /* A load of ctx field could have different |
6567 | * actual load size with the one encoded in the | |
6568 | * insn. When the dst is PTR, it is for sure not | |
6569 | * a sub-register. | |
6570 | */ | |
6571 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
c25b2ae1 | 6572 | if (base_type(reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 6573 | regs[value_regno].btf = btf; |
9e15db66 | 6574 | regs[value_regno].btf_id = btf_id; |
22dc4a0f | 6575 | } |
46f8bc92 | 6576 | } |
638f5b90 | 6577 | regs[value_regno].type = reg_type; |
969bf05e | 6578 | } |
17a52670 | 6579 | |
f1174f77 | 6580 | } else if (reg->type == PTR_TO_STACK) { |
01f810ac AM |
6581 | /* Basic bounds checks. */ |
6582 | err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t); | |
e4298d25 DB |
6583 | if (err) |
6584 | return err; | |
8726679a | 6585 | |
f4d7e40a AS |
6586 | state = func(env, reg); |
6587 | err = update_stack_depth(env, state, off); | |
6588 | if (err) | |
6589 | return err; | |
8726679a | 6590 | |
01f810ac AM |
6591 | if (t == BPF_READ) |
6592 | err = check_stack_read(env, regno, off, size, | |
61bd5218 | 6593 | value_regno); |
01f810ac AM |
6594 | else |
6595 | err = check_stack_write(env, regno, off, size, | |
6596 | value_regno, insn_idx); | |
de8f3a83 | 6597 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 6598 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 6599 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
6600 | return -EACCES; |
6601 | } | |
4acf6c0b BB |
6602 | if (t == BPF_WRITE && value_regno >= 0 && |
6603 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
6604 | verbose(env, "R%d leaks addr into packet\n", |
6605 | value_regno); | |
4acf6c0b BB |
6606 | return -EACCES; |
6607 | } | |
9fd29c08 | 6608 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 6609 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 6610 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
6611 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
6612 | if (t == BPF_WRITE && value_regno >= 0 && | |
6613 | is_pointer_value(env, value_regno)) { | |
6614 | verbose(env, "R%d leaks addr into flow keys\n", | |
6615 | value_regno); | |
6616 | return -EACCES; | |
6617 | } | |
6618 | ||
6619 | err = check_flow_keys_access(env, off, size); | |
6620 | if (!err && t == BPF_READ && value_regno >= 0) | |
6621 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 6622 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 6623 | if (t == BPF_WRITE) { |
46f8bc92 | 6624 | verbose(env, "R%d cannot write into %s\n", |
c25b2ae1 | 6625 | regno, reg_type_str(env, reg->type)); |
c64b7983 JS |
6626 | return -EACCES; |
6627 | } | |
5f456649 | 6628 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
6629 | if (!err && value_regno >= 0) |
6630 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
6631 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
6632 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
6633 | if (!err && t == BPF_READ && value_regno >= 0) | |
6634 | mark_reg_unknown(env, regs, value_regno); | |
bff61f6f HL |
6635 | } else if (base_type(reg->type) == PTR_TO_BTF_ID && |
6636 | !type_may_be_null(reg->type)) { | |
9e15db66 AS |
6637 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, |
6638 | value_regno); | |
41c48f3a AI |
6639 | } else if (reg->type == CONST_PTR_TO_MAP) { |
6640 | err = check_ptr_to_map_access(env, regs, regno, off, size, t, | |
6641 | value_regno); | |
20b2aff4 HL |
6642 | } else if (base_type(reg->type) == PTR_TO_BUF) { |
6643 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
20b2aff4 HL |
6644 | u32 *max_access; |
6645 | ||
6646 | if (rdonly_mem) { | |
6647 | if (t == BPF_WRITE) { | |
6648 | verbose(env, "R%d cannot write into %s\n", | |
6649 | regno, reg_type_str(env, reg->type)); | |
6650 | return -EACCES; | |
6651 | } | |
20b2aff4 HL |
6652 | max_access = &env->prog->aux->max_rdonly_access; |
6653 | } else { | |
20b2aff4 | 6654 | max_access = &env->prog->aux->max_rdwr_access; |
afbf21dc | 6655 | } |
20b2aff4 | 6656 | |
f6dfbe31 | 6657 | err = check_buffer_access(env, reg, regno, off, size, false, |
44e9a741 | 6658 | max_access); |
20b2aff4 HL |
6659 | |
6660 | if (!err && value_regno >= 0 && (rdonly_mem || t == BPF_READ)) | |
afbf21dc | 6661 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 6662 | } else { |
61bd5218 | 6663 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
c25b2ae1 | 6664 | reg_type_str(env, reg->type)); |
17a52670 AS |
6665 | return -EACCES; |
6666 | } | |
969bf05e | 6667 | |
f1174f77 | 6668 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 6669 | regs[value_regno].type == SCALAR_VALUE) { |
1f9a1ea8 YS |
6670 | if (!is_ldsx) |
6671 | /* b/h/w load zero-extends, mark upper bits as known 0 */ | |
6672 | coerce_reg_to_size(®s[value_regno], size); | |
6673 | else | |
6674 | coerce_reg_to_size_sx(®s[value_regno], size); | |
969bf05e | 6675 | } |
17a52670 AS |
6676 | return err; |
6677 | } | |
6678 | ||
91c960b0 | 6679 | static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 6680 | { |
5ffa2550 | 6681 | int load_reg; |
17a52670 AS |
6682 | int err; |
6683 | ||
5ca419f2 BJ |
6684 | switch (insn->imm) { |
6685 | case BPF_ADD: | |
6686 | case BPF_ADD | BPF_FETCH: | |
981f94c3 BJ |
6687 | case BPF_AND: |
6688 | case BPF_AND | BPF_FETCH: | |
6689 | case BPF_OR: | |
6690 | case BPF_OR | BPF_FETCH: | |
6691 | case BPF_XOR: | |
6692 | case BPF_XOR | BPF_FETCH: | |
5ffa2550 BJ |
6693 | case BPF_XCHG: |
6694 | case BPF_CMPXCHG: | |
5ca419f2 BJ |
6695 | break; |
6696 | default: | |
91c960b0 BJ |
6697 | verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm); |
6698 | return -EINVAL; | |
6699 | } | |
6700 | ||
6701 | if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) { | |
6702 | verbose(env, "invalid atomic operand size\n"); | |
17a52670 AS |
6703 | return -EINVAL; |
6704 | } | |
6705 | ||
6706 | /* check src1 operand */ | |
dc503a8a | 6707 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6708 | if (err) |
6709 | return err; | |
6710 | ||
6711 | /* check src2 operand */ | |
dc503a8a | 6712 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6713 | if (err) |
6714 | return err; | |
6715 | ||
5ffa2550 BJ |
6716 | if (insn->imm == BPF_CMPXCHG) { |
6717 | /* Check comparison of R0 with memory location */ | |
a82fe085 DB |
6718 | const u32 aux_reg = BPF_REG_0; |
6719 | ||
6720 | err = check_reg_arg(env, aux_reg, SRC_OP); | |
5ffa2550 BJ |
6721 | if (err) |
6722 | return err; | |
a82fe085 DB |
6723 | |
6724 | if (is_pointer_value(env, aux_reg)) { | |
6725 | verbose(env, "R%d leaks addr into mem\n", aux_reg); | |
6726 | return -EACCES; | |
6727 | } | |
5ffa2550 BJ |
6728 | } |
6729 | ||
6bdf6abc | 6730 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 6731 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
6732 | return -EACCES; |
6733 | } | |
6734 | ||
ca369602 | 6735 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 6736 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
6737 | is_flow_key_reg(env, insn->dst_reg) || |
6738 | is_sk_reg(env, insn->dst_reg)) { | |
91c960b0 | 6739 | verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n", |
2a159c6f | 6740 | insn->dst_reg, |
c25b2ae1 | 6741 | reg_type_str(env, reg_state(env, insn->dst_reg)->type)); |
f37a8cb8 DB |
6742 | return -EACCES; |
6743 | } | |
6744 | ||
37086bfd BJ |
6745 | if (insn->imm & BPF_FETCH) { |
6746 | if (insn->imm == BPF_CMPXCHG) | |
6747 | load_reg = BPF_REG_0; | |
6748 | else | |
6749 | load_reg = insn->src_reg; | |
6750 | ||
6751 | /* check and record load of old value */ | |
6752 | err = check_reg_arg(env, load_reg, DST_OP); | |
6753 | if (err) | |
6754 | return err; | |
6755 | } else { | |
6756 | /* This instruction accesses a memory location but doesn't | |
6757 | * actually load it into a register. | |
6758 | */ | |
6759 | load_reg = -1; | |
6760 | } | |
6761 | ||
7d3baf0a DB |
6762 | /* Check whether we can read the memory, with second call for fetch |
6763 | * case to simulate the register fill. | |
6764 | */ | |
31fd8581 | 6765 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
1f9a1ea8 | 6766 | BPF_SIZE(insn->code), BPF_READ, -1, true, false); |
7d3baf0a DB |
6767 | if (!err && load_reg >= 0) |
6768 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, | |
6769 | BPF_SIZE(insn->code), BPF_READ, load_reg, | |
1f9a1ea8 | 6770 | true, false); |
17a52670 AS |
6771 | if (err) |
6772 | return err; | |
6773 | ||
7d3baf0a | 6774 | /* Check whether we can write into the same memory. */ |
5ca419f2 | 6775 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
1f9a1ea8 | 6776 | BPF_SIZE(insn->code), BPF_WRITE, -1, true, false); |
5ca419f2 BJ |
6777 | if (err) |
6778 | return err; | |
6779 | ||
5ca419f2 | 6780 | return 0; |
17a52670 AS |
6781 | } |
6782 | ||
01f810ac AM |
6783 | /* When register 'regno' is used to read the stack (either directly or through |
6784 | * a helper function) make sure that it's within stack boundary and, depending | |
6785 | * on the access type, that all elements of the stack are initialized. | |
6786 | * | |
6787 | * 'off' includes 'regno->off', but not its dynamic part (if any). | |
6788 | * | |
6789 | * All registers that have been spilled on the stack in the slots within the | |
6790 | * read offsets are marked as read. | |
6791 | */ | |
6792 | static int check_stack_range_initialized( | |
6793 | struct bpf_verifier_env *env, int regno, int off, | |
6794 | int access_size, bool zero_size_allowed, | |
61df10c7 | 6795 | enum bpf_access_src type, struct bpf_call_arg_meta *meta) |
2011fccf AI |
6796 | { |
6797 | struct bpf_reg_state *reg = reg_state(env, regno); | |
01f810ac AM |
6798 | struct bpf_func_state *state = func(env, reg); |
6799 | int err, min_off, max_off, i, j, slot, spi; | |
6800 | char *err_extra = type == ACCESS_HELPER ? " indirect" : ""; | |
6801 | enum bpf_access_type bounds_check_type; | |
6802 | /* Some accesses can write anything into the stack, others are | |
6803 | * read-only. | |
6804 | */ | |
6805 | bool clobber = false; | |
2011fccf | 6806 | |
01f810ac AM |
6807 | if (access_size == 0 && !zero_size_allowed) { |
6808 | verbose(env, "invalid zero-sized read\n"); | |
2011fccf AI |
6809 | return -EACCES; |
6810 | } | |
2011fccf | 6811 | |
01f810ac AM |
6812 | if (type == ACCESS_HELPER) { |
6813 | /* The bounds checks for writes are more permissive than for | |
6814 | * reads. However, if raw_mode is not set, we'll do extra | |
6815 | * checks below. | |
6816 | */ | |
6817 | bounds_check_type = BPF_WRITE; | |
6818 | clobber = true; | |
6819 | } else { | |
6820 | bounds_check_type = BPF_READ; | |
6821 | } | |
6822 | err = check_stack_access_within_bounds(env, regno, off, access_size, | |
6823 | type, bounds_check_type); | |
6824 | if (err) | |
6825 | return err; | |
6826 | ||
17a52670 | 6827 | |
2011fccf | 6828 | if (tnum_is_const(reg->var_off)) { |
01f810ac | 6829 | min_off = max_off = reg->var_off.value + off; |
2011fccf | 6830 | } else { |
088ec26d AI |
6831 | /* Variable offset is prohibited for unprivileged mode for |
6832 | * simplicity since it requires corresponding support in | |
6833 | * Spectre masking for stack ALU. | |
6834 | * See also retrieve_ptr_limit(). | |
6835 | */ | |
2c78ee89 | 6836 | if (!env->bypass_spec_v1) { |
088ec26d | 6837 | char tn_buf[48]; |
f1174f77 | 6838 | |
088ec26d | 6839 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
01f810ac AM |
6840 | verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n", |
6841 | regno, err_extra, tn_buf); | |
088ec26d AI |
6842 | return -EACCES; |
6843 | } | |
f2bcd05e AI |
6844 | /* Only initialized buffer on stack is allowed to be accessed |
6845 | * with variable offset. With uninitialized buffer it's hard to | |
6846 | * guarantee that whole memory is marked as initialized on | |
6847 | * helper return since specific bounds are unknown what may | |
6848 | * cause uninitialized stack leaking. | |
6849 | */ | |
6850 | if (meta && meta->raw_mode) | |
6851 | meta = NULL; | |
6852 | ||
01f810ac AM |
6853 | min_off = reg->smin_value + off; |
6854 | max_off = reg->smax_value + off; | |
17a52670 AS |
6855 | } |
6856 | ||
435faee1 | 6857 | if (meta && meta->raw_mode) { |
ef8fc7a0 KKD |
6858 | /* Ensure we won't be overwriting dynptrs when simulating byte |
6859 | * by byte access in check_helper_call using meta.access_size. | |
6860 | * This would be a problem if we have a helper in the future | |
6861 | * which takes: | |
6862 | * | |
6863 | * helper(uninit_mem, len, dynptr) | |
6864 | * | |
6865 | * Now, uninint_mem may overlap with dynptr pointer. Hence, it | |
6866 | * may end up writing to dynptr itself when touching memory from | |
6867 | * arg 1. This can be relaxed on a case by case basis for known | |
6868 | * safe cases, but reject due to the possibilitiy of aliasing by | |
6869 | * default. | |
6870 | */ | |
6871 | for (i = min_off; i < max_off + access_size; i++) { | |
6872 | int stack_off = -i - 1; | |
6873 | ||
6874 | spi = __get_spi(i); | |
6875 | /* raw_mode may write past allocated_stack */ | |
6876 | if (state->allocated_stack <= stack_off) | |
6877 | continue; | |
6878 | if (state->stack[spi].slot_type[stack_off % BPF_REG_SIZE] == STACK_DYNPTR) { | |
6879 | verbose(env, "potential write to dynptr at off=%d disallowed\n", i); | |
6880 | return -EACCES; | |
6881 | } | |
6882 | } | |
435faee1 DB |
6883 | meta->access_size = access_size; |
6884 | meta->regno = regno; | |
6885 | return 0; | |
6886 | } | |
6887 | ||
2011fccf | 6888 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
6889 | u8 *stype; |
6890 | ||
2011fccf | 6891 | slot = -i - 1; |
638f5b90 | 6892 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
6893 | if (state->allocated_stack <= slot) |
6894 | goto err; | |
6895 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
6896 | if (*stype == STACK_MISC) | |
6897 | goto mark; | |
6715df8d EZ |
6898 | if ((*stype == STACK_ZERO) || |
6899 | (*stype == STACK_INVALID && env->allow_uninit_stack)) { | |
01f810ac AM |
6900 | if (clobber) { |
6901 | /* helper can write anything into the stack */ | |
6902 | *stype = STACK_MISC; | |
6903 | } | |
cc2b14d5 | 6904 | goto mark; |
17a52670 | 6905 | } |
1d68f22b | 6906 | |
27113c59 | 6907 | if (is_spilled_reg(&state->stack[spi]) && |
cd17d38f YS |
6908 | (state->stack[spi].spilled_ptr.type == SCALAR_VALUE || |
6909 | env->allow_ptr_leaks)) { | |
01f810ac AM |
6910 | if (clobber) { |
6911 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); | |
6912 | for (j = 0; j < BPF_REG_SIZE; j++) | |
354e8f19 | 6913 | scrub_spilled_slot(&state->stack[spi].slot_type[j]); |
01f810ac | 6914 | } |
f7cf25b2 AS |
6915 | goto mark; |
6916 | } | |
6917 | ||
cc2b14d5 | 6918 | err: |
2011fccf | 6919 | if (tnum_is_const(reg->var_off)) { |
01f810ac AM |
6920 | verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n", |
6921 | err_extra, regno, min_off, i - min_off, access_size); | |
2011fccf AI |
6922 | } else { |
6923 | char tn_buf[48]; | |
6924 | ||
6925 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac AM |
6926 | verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n", |
6927 | err_extra, regno, tn_buf, i - min_off, access_size); | |
2011fccf | 6928 | } |
cc2b14d5 AS |
6929 | return -EACCES; |
6930 | mark: | |
6931 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
6932 | * the whole slot to be marked as 'read' | |
6933 | */ | |
679c782d | 6934 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
6935 | state->stack[spi].spilled_ptr.parent, |
6936 | REG_LIVE_READ64); | |
261f4664 KKD |
6937 | /* We do not set REG_LIVE_WRITTEN for stack slot, as we can not |
6938 | * be sure that whether stack slot is written to or not. Hence, | |
6939 | * we must still conservatively propagate reads upwards even if | |
6940 | * helper may write to the entire memory range. | |
6941 | */ | |
17a52670 | 6942 | } |
2011fccf | 6943 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
6944 | } |
6945 | ||
06c1c049 GB |
6946 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
6947 | int access_size, bool zero_size_allowed, | |
6948 | struct bpf_call_arg_meta *meta) | |
6949 | { | |
638f5b90 | 6950 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
20b2aff4 | 6951 | u32 *max_access; |
06c1c049 | 6952 | |
20b2aff4 | 6953 | switch (base_type(reg->type)) { |
06c1c049 | 6954 | case PTR_TO_PACKET: |
de8f3a83 | 6955 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
6956 | return check_packet_access(env, regno, reg->off, access_size, |
6957 | zero_size_allowed); | |
69c087ba | 6958 | case PTR_TO_MAP_KEY: |
7b3552d3 KKD |
6959 | if (meta && meta->raw_mode) { |
6960 | verbose(env, "R%d cannot write into %s\n", regno, | |
6961 | reg_type_str(env, reg->type)); | |
6962 | return -EACCES; | |
6963 | } | |
69c087ba YS |
6964 | return check_mem_region_access(env, regno, reg->off, access_size, |
6965 | reg->map_ptr->key_size, false); | |
06c1c049 | 6966 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
6967 | if (check_map_access_type(env, regno, reg->off, access_size, |
6968 | meta && meta->raw_mode ? BPF_WRITE : | |
6969 | BPF_READ)) | |
6970 | return -EACCES; | |
9fd29c08 | 6971 | return check_map_access(env, regno, reg->off, access_size, |
61df10c7 | 6972 | zero_size_allowed, ACCESS_HELPER); |
457f4436 | 6973 | case PTR_TO_MEM: |
97e6d7da KKD |
6974 | if (type_is_rdonly_mem(reg->type)) { |
6975 | if (meta && meta->raw_mode) { | |
6976 | verbose(env, "R%d cannot write into %s\n", regno, | |
6977 | reg_type_str(env, reg->type)); | |
6978 | return -EACCES; | |
6979 | } | |
6980 | } | |
457f4436 AN |
6981 | return check_mem_region_access(env, regno, reg->off, |
6982 | access_size, reg->mem_size, | |
6983 | zero_size_allowed); | |
20b2aff4 HL |
6984 | case PTR_TO_BUF: |
6985 | if (type_is_rdonly_mem(reg->type)) { | |
97e6d7da KKD |
6986 | if (meta && meta->raw_mode) { |
6987 | verbose(env, "R%d cannot write into %s\n", regno, | |
6988 | reg_type_str(env, reg->type)); | |
20b2aff4 | 6989 | return -EACCES; |
97e6d7da | 6990 | } |
20b2aff4 | 6991 | |
20b2aff4 HL |
6992 | max_access = &env->prog->aux->max_rdonly_access; |
6993 | } else { | |
20b2aff4 HL |
6994 | max_access = &env->prog->aux->max_rdwr_access; |
6995 | } | |
afbf21dc YS |
6996 | return check_buffer_access(env, reg, regno, reg->off, |
6997 | access_size, zero_size_allowed, | |
44e9a741 | 6998 | max_access); |
0d004c02 | 6999 | case PTR_TO_STACK: |
01f810ac AM |
7000 | return check_stack_range_initialized( |
7001 | env, | |
7002 | regno, reg->off, access_size, | |
7003 | zero_size_allowed, ACCESS_HELPER, meta); | |
3e30be42 AS |
7004 | case PTR_TO_BTF_ID: |
7005 | return check_ptr_to_btf_access(env, regs, regno, reg->off, | |
7006 | access_size, BPF_READ, -1); | |
15baa55f BT |
7007 | case PTR_TO_CTX: |
7008 | /* in case the function doesn't know how to access the context, | |
7009 | * (because we are in a program of type SYSCALL for example), we | |
7010 | * can not statically check its size. | |
7011 | * Dynamically check it now. | |
7012 | */ | |
7013 | if (!env->ops->convert_ctx_access) { | |
7014 | enum bpf_access_type atype = meta && meta->raw_mode ? BPF_WRITE : BPF_READ; | |
7015 | int offset = access_size - 1; | |
7016 | ||
7017 | /* Allow zero-byte read from PTR_TO_CTX */ | |
7018 | if (access_size == 0) | |
7019 | return zero_size_allowed ? 0 : -EACCES; | |
7020 | ||
7021 | return check_mem_access(env, env->insn_idx, regno, offset, BPF_B, | |
1f9a1ea8 | 7022 | atype, -1, false, false); |
15baa55f BT |
7023 | } |
7024 | ||
7025 | fallthrough; | |
0d004c02 LB |
7026 | default: /* scalar_value or invalid ptr */ |
7027 | /* Allow zero-byte read from NULL, regardless of pointer type */ | |
7028 | if (zero_size_allowed && access_size == 0 && | |
7029 | register_is_null(reg)) | |
7030 | return 0; | |
7031 | ||
c25b2ae1 HL |
7032 | verbose(env, "R%d type=%s ", regno, |
7033 | reg_type_str(env, reg->type)); | |
7034 | verbose(env, "expected=%s\n", reg_type_str(env, PTR_TO_STACK)); | |
0d004c02 | 7035 | return -EACCES; |
06c1c049 GB |
7036 | } |
7037 | } | |
7038 | ||
d583691c KKD |
7039 | static int check_mem_size_reg(struct bpf_verifier_env *env, |
7040 | struct bpf_reg_state *reg, u32 regno, | |
7041 | bool zero_size_allowed, | |
7042 | struct bpf_call_arg_meta *meta) | |
7043 | { | |
7044 | int err; | |
7045 | ||
7046 | /* This is used to refine r0 return value bounds for helpers | |
7047 | * that enforce this value as an upper bound on return values. | |
7048 | * See do_refine_retval_range() for helpers that can refine | |
7049 | * the return value. C type of helper is u32 so we pull register | |
7050 | * bound from umax_value however, if negative verifier errors | |
7051 | * out. Only upper bounds can be learned because retval is an | |
7052 | * int type and negative retvals are allowed. | |
7053 | */ | |
be77354a | 7054 | meta->msize_max_value = reg->umax_value; |
d583691c KKD |
7055 | |
7056 | /* The register is SCALAR_VALUE; the access check | |
7057 | * happens using its boundaries. | |
7058 | */ | |
7059 | if (!tnum_is_const(reg->var_off)) | |
7060 | /* For unprivileged variable accesses, disable raw | |
7061 | * mode so that the program is required to | |
7062 | * initialize all the memory that the helper could | |
7063 | * just partially fill up. | |
7064 | */ | |
7065 | meta = NULL; | |
7066 | ||
7067 | if (reg->smin_value < 0) { | |
7068 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", | |
7069 | regno); | |
7070 | return -EACCES; | |
7071 | } | |
7072 | ||
7073 | if (reg->umin_value == 0) { | |
7074 | err = check_helper_mem_access(env, regno - 1, 0, | |
7075 | zero_size_allowed, | |
7076 | meta); | |
7077 | if (err) | |
7078 | return err; | |
7079 | } | |
7080 | ||
7081 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { | |
7082 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", | |
7083 | regno); | |
7084 | return -EACCES; | |
7085 | } | |
7086 | err = check_helper_mem_access(env, regno - 1, | |
7087 | reg->umax_value, | |
7088 | zero_size_allowed, meta); | |
7089 | if (!err) | |
7090 | err = mark_chain_precision(env, regno); | |
7091 | return err; | |
7092 | } | |
7093 | ||
e5069b9c DB |
7094 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
7095 | u32 regno, u32 mem_size) | |
7096 | { | |
be77354a KKD |
7097 | bool may_be_null = type_may_be_null(reg->type); |
7098 | struct bpf_reg_state saved_reg; | |
7099 | struct bpf_call_arg_meta meta; | |
7100 | int err; | |
7101 | ||
e5069b9c DB |
7102 | if (register_is_null(reg)) |
7103 | return 0; | |
7104 | ||
be77354a KKD |
7105 | memset(&meta, 0, sizeof(meta)); |
7106 | /* Assuming that the register contains a value check if the memory | |
7107 | * access is safe. Temporarily save and restore the register's state as | |
7108 | * the conversion shouldn't be visible to a caller. | |
7109 | */ | |
7110 | if (may_be_null) { | |
7111 | saved_reg = *reg; | |
e5069b9c | 7112 | mark_ptr_not_null_reg(reg); |
e5069b9c DB |
7113 | } |
7114 | ||
be77354a KKD |
7115 | err = check_helper_mem_access(env, regno, mem_size, true, &meta); |
7116 | /* Check access for BPF_WRITE */ | |
7117 | meta.raw_mode = true; | |
7118 | err = err ?: check_helper_mem_access(env, regno, mem_size, true, &meta); | |
7119 | ||
7120 | if (may_be_null) | |
7121 | *reg = saved_reg; | |
7122 | ||
7123 | return err; | |
e5069b9c DB |
7124 | } |
7125 | ||
00b85860 KKD |
7126 | static int check_kfunc_mem_size_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
7127 | u32 regno) | |
d583691c KKD |
7128 | { |
7129 | struct bpf_reg_state *mem_reg = &cur_regs(env)[regno - 1]; | |
7130 | bool may_be_null = type_may_be_null(mem_reg->type); | |
7131 | struct bpf_reg_state saved_reg; | |
be77354a | 7132 | struct bpf_call_arg_meta meta; |
d583691c KKD |
7133 | int err; |
7134 | ||
7135 | WARN_ON_ONCE(regno < BPF_REG_2 || regno > BPF_REG_5); | |
7136 | ||
be77354a KKD |
7137 | memset(&meta, 0, sizeof(meta)); |
7138 | ||
d583691c KKD |
7139 | if (may_be_null) { |
7140 | saved_reg = *mem_reg; | |
7141 | mark_ptr_not_null_reg(mem_reg); | |
7142 | } | |
7143 | ||
be77354a KKD |
7144 | err = check_mem_size_reg(env, reg, regno, true, &meta); |
7145 | /* Check access for BPF_WRITE */ | |
7146 | meta.raw_mode = true; | |
7147 | err = err ?: check_mem_size_reg(env, reg, regno, true, &meta); | |
d583691c KKD |
7148 | |
7149 | if (may_be_null) | |
7150 | *mem_reg = saved_reg; | |
7151 | return err; | |
7152 | } | |
7153 | ||
d83525ca | 7154 | /* Implementation details: |
4e814da0 KKD |
7155 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL. |
7156 | * bpf_obj_new returns PTR_TO_BTF_ID | MEM_ALLOC | PTR_MAYBE_NULL. | |
d83525ca | 7157 | * Two bpf_map_lookups (even with the same key) will have different reg->id. |
4e814da0 KKD |
7158 | * Two separate bpf_obj_new will also have different reg->id. |
7159 | * For traditional PTR_TO_MAP_VALUE or PTR_TO_BTF_ID | MEM_ALLOC, the verifier | |
7160 | * clears reg->id after value_or_null->value transition, since the verifier only | |
7161 | * cares about the range of access to valid map value pointer and doesn't care | |
7162 | * about actual address of the map element. | |
d83525ca AS |
7163 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps |
7164 | * reg->id > 0 after value_or_null->value transition. By doing so | |
7165 | * two bpf_map_lookups will be considered two different pointers that | |
4e814da0 KKD |
7166 | * point to different bpf_spin_locks. Likewise for pointers to allocated objects |
7167 | * returned from bpf_obj_new. | |
d83525ca AS |
7168 | * The verifier allows taking only one bpf_spin_lock at a time to avoid |
7169 | * dead-locks. | |
7170 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
7171 | * reg_is_refcounted() logic. The verifier needs to remember only | |
7172 | * one spin_lock instead of array of acquired_refs. | |
d0d78c1d | 7173 | * cur_state->active_lock remembers which map value element or allocated |
4e814da0 | 7174 | * object got locked and clears it after bpf_spin_unlock. |
d83525ca AS |
7175 | */ |
7176 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
7177 | bool is_lock) | |
7178 | { | |
7179 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7180 | struct bpf_verifier_state *cur = env->cur_state; | |
7181 | bool is_const = tnum_is_const(reg->var_off); | |
d83525ca | 7182 | u64 val = reg->var_off.value; |
4e814da0 KKD |
7183 | struct bpf_map *map = NULL; |
7184 | struct btf *btf = NULL; | |
7185 | struct btf_record *rec; | |
d83525ca | 7186 | |
d83525ca AS |
7187 | if (!is_const) { |
7188 | verbose(env, | |
7189 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
7190 | regno); | |
7191 | return -EINVAL; | |
7192 | } | |
4e814da0 KKD |
7193 | if (reg->type == PTR_TO_MAP_VALUE) { |
7194 | map = reg->map_ptr; | |
7195 | if (!map->btf) { | |
7196 | verbose(env, | |
7197 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
7198 | map->name); | |
7199 | return -EINVAL; | |
7200 | } | |
7201 | } else { | |
7202 | btf = reg->btf; | |
d83525ca | 7203 | } |
4e814da0 KKD |
7204 | |
7205 | rec = reg_btf_record(reg); | |
7206 | if (!btf_record_has_field(rec, BPF_SPIN_LOCK)) { | |
7207 | verbose(env, "%s '%s' has no valid bpf_spin_lock\n", map ? "map" : "local", | |
7208 | map ? map->name : "kptr"); | |
d83525ca AS |
7209 | return -EINVAL; |
7210 | } | |
4e814da0 | 7211 | if (rec->spin_lock_off != val + reg->off) { |
db559117 | 7212 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock' that is at %d\n", |
4e814da0 | 7213 | val + reg->off, rec->spin_lock_off); |
d83525ca AS |
7214 | return -EINVAL; |
7215 | } | |
7216 | if (is_lock) { | |
d0d78c1d | 7217 | if (cur->active_lock.ptr) { |
d83525ca AS |
7218 | verbose(env, |
7219 | "Locking two bpf_spin_locks are not allowed\n"); | |
7220 | return -EINVAL; | |
7221 | } | |
d0d78c1d KKD |
7222 | if (map) |
7223 | cur->active_lock.ptr = map; | |
7224 | else | |
7225 | cur->active_lock.ptr = btf; | |
7226 | cur->active_lock.id = reg->id; | |
d83525ca | 7227 | } else { |
d0d78c1d KKD |
7228 | void *ptr; |
7229 | ||
7230 | if (map) | |
7231 | ptr = map; | |
7232 | else | |
7233 | ptr = btf; | |
7234 | ||
7235 | if (!cur->active_lock.ptr) { | |
d83525ca AS |
7236 | verbose(env, "bpf_spin_unlock without taking a lock\n"); |
7237 | return -EINVAL; | |
7238 | } | |
d0d78c1d KKD |
7239 | if (cur->active_lock.ptr != ptr || |
7240 | cur->active_lock.id != reg->id) { | |
d83525ca AS |
7241 | verbose(env, "bpf_spin_unlock of different lock\n"); |
7242 | return -EINVAL; | |
7243 | } | |
534e86bc | 7244 | |
6a3cd331 | 7245 | invalidate_non_owning_refs(env); |
534e86bc | 7246 | |
6a3cd331 DM |
7247 | cur->active_lock.ptr = NULL; |
7248 | cur->active_lock.id = 0; | |
d83525ca AS |
7249 | } |
7250 | return 0; | |
7251 | } | |
7252 | ||
b00628b1 AS |
7253 | static int process_timer_func(struct bpf_verifier_env *env, int regno, |
7254 | struct bpf_call_arg_meta *meta) | |
7255 | { | |
7256 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7257 | bool is_const = tnum_is_const(reg->var_off); | |
7258 | struct bpf_map *map = reg->map_ptr; | |
7259 | u64 val = reg->var_off.value; | |
7260 | ||
7261 | if (!is_const) { | |
7262 | verbose(env, | |
7263 | "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n", | |
7264 | regno); | |
7265 | return -EINVAL; | |
7266 | } | |
7267 | if (!map->btf) { | |
7268 | verbose(env, "map '%s' has to have BTF in order to use bpf_timer\n", | |
7269 | map->name); | |
7270 | return -EINVAL; | |
7271 | } | |
db559117 KKD |
7272 | if (!btf_record_has_field(map->record, BPF_TIMER)) { |
7273 | verbose(env, "map '%s' has no valid bpf_timer\n", map->name); | |
68134668 AS |
7274 | return -EINVAL; |
7275 | } | |
db559117 | 7276 | if (map->record->timer_off != val + reg->off) { |
68134668 | 7277 | verbose(env, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n", |
db559117 | 7278 | val + reg->off, map->record->timer_off); |
b00628b1 AS |
7279 | return -EINVAL; |
7280 | } | |
7281 | if (meta->map_ptr) { | |
7282 | verbose(env, "verifier bug. Two map pointers in a timer helper\n"); | |
7283 | return -EFAULT; | |
7284 | } | |
3e8ce298 | 7285 | meta->map_uid = reg->map_uid; |
b00628b1 AS |
7286 | meta->map_ptr = map; |
7287 | return 0; | |
7288 | } | |
7289 | ||
c0a5a21c KKD |
7290 | static int process_kptr_func(struct bpf_verifier_env *env, int regno, |
7291 | struct bpf_call_arg_meta *meta) | |
7292 | { | |
7293 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
c0a5a21c | 7294 | struct bpf_map *map_ptr = reg->map_ptr; |
aa3496ac | 7295 | struct btf_field *kptr_field; |
c0a5a21c | 7296 | u32 kptr_off; |
c0a5a21c KKD |
7297 | |
7298 | if (!tnum_is_const(reg->var_off)) { | |
7299 | verbose(env, | |
7300 | "R%d doesn't have constant offset. kptr has to be at the constant offset\n", | |
7301 | regno); | |
7302 | return -EINVAL; | |
7303 | } | |
7304 | if (!map_ptr->btf) { | |
7305 | verbose(env, "map '%s' has to have BTF in order to use bpf_kptr_xchg\n", | |
7306 | map_ptr->name); | |
7307 | return -EINVAL; | |
7308 | } | |
aa3496ac KKD |
7309 | if (!btf_record_has_field(map_ptr->record, BPF_KPTR)) { |
7310 | verbose(env, "map '%s' has no valid kptr\n", map_ptr->name); | |
c0a5a21c KKD |
7311 | return -EINVAL; |
7312 | } | |
7313 | ||
7314 | meta->map_ptr = map_ptr; | |
7315 | kptr_off = reg->off + reg->var_off.value; | |
aa3496ac KKD |
7316 | kptr_field = btf_record_find(map_ptr->record, kptr_off, BPF_KPTR); |
7317 | if (!kptr_field) { | |
c0a5a21c KKD |
7318 | verbose(env, "off=%d doesn't point to kptr\n", kptr_off); |
7319 | return -EACCES; | |
7320 | } | |
aa3496ac | 7321 | if (kptr_field->type != BPF_KPTR_REF) { |
c0a5a21c KKD |
7322 | verbose(env, "off=%d kptr isn't referenced kptr\n", kptr_off); |
7323 | return -EACCES; | |
7324 | } | |
aa3496ac | 7325 | meta->kptr_field = kptr_field; |
c0a5a21c KKD |
7326 | return 0; |
7327 | } | |
7328 | ||
27060531 KKD |
7329 | /* There are two register types representing a bpf_dynptr, one is PTR_TO_STACK |
7330 | * which points to a stack slot, and the other is CONST_PTR_TO_DYNPTR. | |
7331 | * | |
7332 | * In both cases we deal with the first 8 bytes, but need to mark the next 8 | |
7333 | * bytes as STACK_DYNPTR in case of PTR_TO_STACK. In case of | |
7334 | * CONST_PTR_TO_DYNPTR, we are guaranteed to get the beginning of the object. | |
7335 | * | |
7336 | * Mutability of bpf_dynptr is at two levels, one is at the level of struct | |
7337 | * bpf_dynptr itself, i.e. whether the helper is receiving a pointer to struct | |
7338 | * bpf_dynptr or pointer to const struct bpf_dynptr. In the former case, it can | |
7339 | * mutate the view of the dynptr and also possibly destroy it. In the latter | |
7340 | * case, it cannot mutate the bpf_dynptr itself but it can still mutate the | |
7341 | * memory that dynptr points to. | |
7342 | * | |
7343 | * The verifier will keep track both levels of mutation (bpf_dynptr's in | |
7344 | * reg->type and the memory's in reg->dynptr.type), but there is no support for | |
7345 | * readonly dynptr view yet, hence only the first case is tracked and checked. | |
7346 | * | |
7347 | * This is consistent with how C applies the const modifier to a struct object, | |
7348 | * where the pointer itself inside bpf_dynptr becomes const but not what it | |
7349 | * points to. | |
7350 | * | |
7351 | * Helpers which do not mutate the bpf_dynptr set MEM_RDONLY in their argument | |
7352 | * type, and declare it as 'const struct bpf_dynptr *' in their prototype. | |
7353 | */ | |
1d18feb2 | 7354 | static int process_dynptr_func(struct bpf_verifier_env *env, int regno, int insn_idx, |
361f129f | 7355 | enum bpf_arg_type arg_type, int clone_ref_obj_id) |
6b75bd3d KKD |
7356 | { |
7357 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
1d18feb2 | 7358 | int err; |
6b75bd3d | 7359 | |
27060531 KKD |
7360 | /* MEM_UNINIT and MEM_RDONLY are exclusive, when applied to an |
7361 | * ARG_PTR_TO_DYNPTR (or ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_*): | |
7362 | */ | |
7363 | if ((arg_type & (MEM_UNINIT | MEM_RDONLY)) == (MEM_UNINIT | MEM_RDONLY)) { | |
7364 | verbose(env, "verifier internal error: misconfigured dynptr helper type flags\n"); | |
7365 | return -EFAULT; | |
7366 | } | |
79168a66 | 7367 | |
27060531 KKD |
7368 | /* MEM_UNINIT - Points to memory that is an appropriate candidate for |
7369 | * constructing a mutable bpf_dynptr object. | |
7370 | * | |
7371 | * Currently, this is only possible with PTR_TO_STACK | |
7372 | * pointing to a region of at least 16 bytes which doesn't | |
7373 | * contain an existing bpf_dynptr. | |
7374 | * | |
7375 | * MEM_RDONLY - Points to a initialized bpf_dynptr that will not be | |
7376 | * mutated or destroyed. However, the memory it points to | |
7377 | * may be mutated. | |
7378 | * | |
7379 | * None - Points to a initialized dynptr that can be mutated and | |
7380 | * destroyed, including mutation of the memory it points | |
7381 | * to. | |
6b75bd3d | 7382 | */ |
6b75bd3d | 7383 | if (arg_type & MEM_UNINIT) { |
1d18feb2 JK |
7384 | int i; |
7385 | ||
7e0dac28 | 7386 | if (!is_dynptr_reg_valid_uninit(env, reg)) { |
6b75bd3d KKD |
7387 | verbose(env, "Dynptr has to be an uninitialized dynptr\n"); |
7388 | return -EINVAL; | |
7389 | } | |
7390 | ||
1d18feb2 JK |
7391 | /* we write BPF_DW bits (8 bytes) at a time */ |
7392 | for (i = 0; i < BPF_DYNPTR_SIZE; i += 8) { | |
7393 | err = check_mem_access(env, insn_idx, regno, | |
1f9a1ea8 | 7394 | i, BPF_DW, BPF_WRITE, -1, false, false); |
1d18feb2 JK |
7395 | if (err) |
7396 | return err; | |
6b75bd3d KKD |
7397 | } |
7398 | ||
361f129f | 7399 | err = mark_stack_slots_dynptr(env, reg, arg_type, insn_idx, clone_ref_obj_id); |
27060531 KKD |
7400 | } else /* MEM_RDONLY and None case from above */ { |
7401 | /* For the reg->type == PTR_TO_STACK case, bpf_dynptr is never const */ | |
7402 | if (reg->type == CONST_PTR_TO_DYNPTR && !(arg_type & MEM_RDONLY)) { | |
7403 | verbose(env, "cannot pass pointer to const bpf_dynptr, the helper mutates it\n"); | |
7404 | return -EINVAL; | |
7405 | } | |
7406 | ||
7e0dac28 | 7407 | if (!is_dynptr_reg_valid_init(env, reg)) { |
6b75bd3d KKD |
7408 | verbose(env, |
7409 | "Expected an initialized dynptr as arg #%d\n", | |
7410 | regno); | |
7411 | return -EINVAL; | |
7412 | } | |
7413 | ||
27060531 KKD |
7414 | /* Fold modifiers (in this case, MEM_RDONLY) when checking expected type */ |
7415 | if (!is_dynptr_type_expected(env, reg, arg_type & ~MEM_RDONLY)) { | |
6b75bd3d KKD |
7416 | verbose(env, |
7417 | "Expected a dynptr of type %s as arg #%d\n", | |
d54e0f6c | 7418 | dynptr_type_str(arg_to_dynptr_type(arg_type)), regno); |
6b75bd3d KKD |
7419 | return -EINVAL; |
7420 | } | |
d6fefa11 KKD |
7421 | |
7422 | err = mark_dynptr_read(env, reg); | |
6b75bd3d | 7423 | } |
1d18feb2 | 7424 | return err; |
6b75bd3d KKD |
7425 | } |
7426 | ||
06accc87 AN |
7427 | static u32 iter_ref_obj_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg, int spi) |
7428 | { | |
7429 | struct bpf_func_state *state = func(env, reg); | |
7430 | ||
7431 | return state->stack[spi].spilled_ptr.ref_obj_id; | |
7432 | } | |
7433 | ||
7434 | static bool is_iter_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7435 | { | |
7436 | return meta->kfunc_flags & (KF_ITER_NEW | KF_ITER_NEXT | KF_ITER_DESTROY); | |
7437 | } | |
7438 | ||
7439 | static bool is_iter_new_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7440 | { | |
7441 | return meta->kfunc_flags & KF_ITER_NEW; | |
7442 | } | |
7443 | ||
7444 | static bool is_iter_next_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7445 | { | |
7446 | return meta->kfunc_flags & KF_ITER_NEXT; | |
7447 | } | |
7448 | ||
7449 | static bool is_iter_destroy_kfunc(struct bpf_kfunc_call_arg_meta *meta) | |
7450 | { | |
7451 | return meta->kfunc_flags & KF_ITER_DESTROY; | |
7452 | } | |
7453 | ||
7454 | static bool is_kfunc_arg_iter(struct bpf_kfunc_call_arg_meta *meta, int arg) | |
7455 | { | |
7456 | /* btf_check_iter_kfuncs() guarantees that first argument of any iter | |
7457 | * kfunc is iter state pointer | |
7458 | */ | |
7459 | return arg == 0 && is_iter_kfunc(meta); | |
7460 | } | |
7461 | ||
7462 | static int process_iter_arg(struct bpf_verifier_env *env, int regno, int insn_idx, | |
7463 | struct bpf_kfunc_call_arg_meta *meta) | |
7464 | { | |
7465 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7466 | const struct btf_type *t; | |
7467 | const struct btf_param *arg; | |
7468 | int spi, err, i, nr_slots; | |
7469 | u32 btf_id; | |
7470 | ||
7471 | /* btf_check_iter_kfuncs() ensures we don't need to validate anything here */ | |
7472 | arg = &btf_params(meta->func_proto)[0]; | |
7473 | t = btf_type_skip_modifiers(meta->btf, arg->type, NULL); /* PTR */ | |
7474 | t = btf_type_skip_modifiers(meta->btf, t->type, &btf_id); /* STRUCT */ | |
7475 | nr_slots = t->size / BPF_REG_SIZE; | |
7476 | ||
06accc87 AN |
7477 | if (is_iter_new_kfunc(meta)) { |
7478 | /* bpf_iter_<type>_new() expects pointer to uninit iter state */ | |
7479 | if (!is_iter_reg_valid_uninit(env, reg, nr_slots)) { | |
7480 | verbose(env, "expected uninitialized iter_%s as arg #%d\n", | |
7481 | iter_type_str(meta->btf, btf_id), regno); | |
7482 | return -EINVAL; | |
7483 | } | |
7484 | ||
7485 | for (i = 0; i < nr_slots * 8; i += BPF_REG_SIZE) { | |
7486 | err = check_mem_access(env, insn_idx, regno, | |
1f9a1ea8 | 7487 | i, BPF_DW, BPF_WRITE, -1, false, false); |
06accc87 AN |
7488 | if (err) |
7489 | return err; | |
7490 | } | |
7491 | ||
7492 | err = mark_stack_slots_iter(env, reg, insn_idx, meta->btf, btf_id, nr_slots); | |
7493 | if (err) | |
7494 | return err; | |
7495 | } else { | |
7496 | /* iter_next() or iter_destroy() expect initialized iter state*/ | |
7497 | if (!is_iter_reg_valid_init(env, reg, meta->btf, btf_id, nr_slots)) { | |
7498 | verbose(env, "expected an initialized iter_%s as arg #%d\n", | |
7499 | iter_type_str(meta->btf, btf_id), regno); | |
7500 | return -EINVAL; | |
7501 | } | |
7502 | ||
b63cbc49 AN |
7503 | spi = iter_get_spi(env, reg, nr_slots); |
7504 | if (spi < 0) | |
7505 | return spi; | |
7506 | ||
06accc87 AN |
7507 | err = mark_iter_read(env, reg, spi, nr_slots); |
7508 | if (err) | |
7509 | return err; | |
7510 | ||
b63cbc49 AN |
7511 | /* remember meta->iter info for process_iter_next_call() */ |
7512 | meta->iter.spi = spi; | |
7513 | meta->iter.frameno = reg->frameno; | |
06accc87 AN |
7514 | meta->ref_obj_id = iter_ref_obj_id(env, reg, spi); |
7515 | ||
7516 | if (is_iter_destroy_kfunc(meta)) { | |
7517 | err = unmark_stack_slots_iter(env, reg, nr_slots); | |
7518 | if (err) | |
7519 | return err; | |
7520 | } | |
7521 | } | |
7522 | ||
7523 | return 0; | |
7524 | } | |
7525 | ||
7526 | /* process_iter_next_call() is called when verifier gets to iterator's next | |
7527 | * "method" (e.g., bpf_iter_num_next() for numbers iterator) call. We'll refer | |
7528 | * to it as just "iter_next()" in comments below. | |
7529 | * | |
7530 | * BPF verifier relies on a crucial contract for any iter_next() | |
7531 | * implementation: it should *eventually* return NULL, and once that happens | |
7532 | * it should keep returning NULL. That is, once iterator exhausts elements to | |
7533 | * iterate, it should never reset or spuriously return new elements. | |
7534 | * | |
7535 | * With the assumption of such contract, process_iter_next_call() simulates | |
7536 | * a fork in the verifier state to validate loop logic correctness and safety | |
7537 | * without having to simulate infinite amount of iterations. | |
7538 | * | |
7539 | * In current state, we first assume that iter_next() returned NULL and | |
7540 | * iterator state is set to DRAINED (BPF_ITER_STATE_DRAINED). In such | |
7541 | * conditions we should not form an infinite loop and should eventually reach | |
7542 | * exit. | |
7543 | * | |
7544 | * Besides that, we also fork current state and enqueue it for later | |
7545 | * verification. In a forked state we keep iterator state as ACTIVE | |
7546 | * (BPF_ITER_STATE_ACTIVE) and assume non-NULL return from iter_next(). We | |
7547 | * also bump iteration depth to prevent erroneous infinite loop detection | |
7548 | * later on (see iter_active_depths_differ() comment for details). In this | |
7549 | * state we assume that we'll eventually loop back to another iter_next() | |
7550 | * calls (it could be in exactly same location or in some other instruction, | |
7551 | * it doesn't matter, we don't make any unnecessary assumptions about this, | |
7552 | * everything revolves around iterator state in a stack slot, not which | |
7553 | * instruction is calling iter_next()). When that happens, we either will come | |
7554 | * to iter_next() with equivalent state and can conclude that next iteration | |
7555 | * will proceed in exactly the same way as we just verified, so it's safe to | |
7556 | * assume that loop converges. If not, we'll go on another iteration | |
7557 | * simulation with a different input state, until all possible starting states | |
7558 | * are validated or we reach maximum number of instructions limit. | |
7559 | * | |
7560 | * This way, we will either exhaustively discover all possible input states | |
7561 | * that iterator loop can start with and eventually will converge, or we'll | |
7562 | * effectively regress into bounded loop simulation logic and either reach | |
7563 | * maximum number of instructions if loop is not provably convergent, or there | |
7564 | * is some statically known limit on number of iterations (e.g., if there is | |
7565 | * an explicit `if n > 100 then break;` statement somewhere in the loop). | |
7566 | * | |
7567 | * One very subtle but very important aspect is that we *always* simulate NULL | |
7568 | * condition first (as the current state) before we simulate non-NULL case. | |
7569 | * This has to do with intricacies of scalar precision tracking. By simulating | |
7570 | * "exit condition" of iter_next() returning NULL first, we make sure all the | |
7571 | * relevant precision marks *that will be set **after** we exit iterator loop* | |
7572 | * are propagated backwards to common parent state of NULL and non-NULL | |
7573 | * branches. Thanks to that, state equivalence checks done later in forked | |
7574 | * state, when reaching iter_next() for ACTIVE iterator, can assume that | |
7575 | * precision marks are finalized and won't change. Because simulating another | |
7576 | * ACTIVE iterator iteration won't change them (because given same input | |
7577 | * states we'll end up with exactly same output states which we are currently | |
7578 | * comparing; and verification after the loop already propagated back what | |
7579 | * needs to be **additionally** tracked as precise). It's subtle, grok | |
7580 | * precision tracking for more intuitive understanding. | |
7581 | */ | |
7582 | static int process_iter_next_call(struct bpf_verifier_env *env, int insn_idx, | |
7583 | struct bpf_kfunc_call_arg_meta *meta) | |
7584 | { | |
7585 | struct bpf_verifier_state *cur_st = env->cur_state, *queued_st; | |
7586 | struct bpf_func_state *cur_fr = cur_st->frame[cur_st->curframe], *queued_fr; | |
7587 | struct bpf_reg_state *cur_iter, *queued_iter; | |
7588 | int iter_frameno = meta->iter.frameno; | |
7589 | int iter_spi = meta->iter.spi; | |
7590 | ||
7591 | BTF_TYPE_EMIT(struct bpf_iter); | |
7592 | ||
7593 | cur_iter = &env->cur_state->frame[iter_frameno]->stack[iter_spi].spilled_ptr; | |
7594 | ||
7595 | if (cur_iter->iter.state != BPF_ITER_STATE_ACTIVE && | |
7596 | cur_iter->iter.state != BPF_ITER_STATE_DRAINED) { | |
7597 | verbose(env, "verifier internal error: unexpected iterator state %d (%s)\n", | |
7598 | cur_iter->iter.state, iter_state_str(cur_iter->iter.state)); | |
7599 | return -EFAULT; | |
7600 | } | |
7601 | ||
7602 | if (cur_iter->iter.state == BPF_ITER_STATE_ACTIVE) { | |
7603 | /* branch out active iter state */ | |
7604 | queued_st = push_stack(env, insn_idx + 1, insn_idx, false); | |
7605 | if (!queued_st) | |
7606 | return -ENOMEM; | |
7607 | ||
7608 | queued_iter = &queued_st->frame[iter_frameno]->stack[iter_spi].spilled_ptr; | |
7609 | queued_iter->iter.state = BPF_ITER_STATE_ACTIVE; | |
7610 | queued_iter->iter.depth++; | |
7611 | ||
7612 | queued_fr = queued_st->frame[queued_st->curframe]; | |
7613 | mark_ptr_not_null_reg(&queued_fr->regs[BPF_REG_0]); | |
7614 | } | |
7615 | ||
7616 | /* switch to DRAINED state, but keep the depth unchanged */ | |
7617 | /* mark current iter state as drained and assume returned NULL */ | |
7618 | cur_iter->iter.state = BPF_ITER_STATE_DRAINED; | |
7619 | __mark_reg_const_zero(&cur_fr->regs[BPF_REG_0]); | |
7620 | ||
7621 | return 0; | |
7622 | } | |
7623 | ||
90133415 DB |
7624 | static bool arg_type_is_mem_size(enum bpf_arg_type type) |
7625 | { | |
7626 | return type == ARG_CONST_SIZE || | |
7627 | type == ARG_CONST_SIZE_OR_ZERO; | |
7628 | } | |
7629 | ||
8f14852e KKD |
7630 | static bool arg_type_is_release(enum bpf_arg_type type) |
7631 | { | |
7632 | return type & OBJ_RELEASE; | |
7633 | } | |
7634 | ||
97e03f52 JK |
7635 | static bool arg_type_is_dynptr(enum bpf_arg_type type) |
7636 | { | |
7637 | return base_type(type) == ARG_PTR_TO_DYNPTR; | |
7638 | } | |
7639 | ||
57c3bb72 AI |
7640 | static int int_ptr_type_to_size(enum bpf_arg_type type) |
7641 | { | |
7642 | if (type == ARG_PTR_TO_INT) | |
7643 | return sizeof(u32); | |
7644 | else if (type == ARG_PTR_TO_LONG) | |
7645 | return sizeof(u64); | |
7646 | ||
7647 | return -EINVAL; | |
7648 | } | |
7649 | ||
912f442c LB |
7650 | static int resolve_map_arg_type(struct bpf_verifier_env *env, |
7651 | const struct bpf_call_arg_meta *meta, | |
7652 | enum bpf_arg_type *arg_type) | |
7653 | { | |
7654 | if (!meta->map_ptr) { | |
7655 | /* kernel subsystem misconfigured verifier */ | |
7656 | verbose(env, "invalid map_ptr to access map->type\n"); | |
7657 | return -EACCES; | |
7658 | } | |
7659 | ||
7660 | switch (meta->map_ptr->map_type) { | |
7661 | case BPF_MAP_TYPE_SOCKMAP: | |
7662 | case BPF_MAP_TYPE_SOCKHASH: | |
7663 | if (*arg_type == ARG_PTR_TO_MAP_VALUE) { | |
6550f2dd | 7664 | *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON; |
912f442c LB |
7665 | } else { |
7666 | verbose(env, "invalid arg_type for sockmap/sockhash\n"); | |
7667 | return -EINVAL; | |
7668 | } | |
7669 | break; | |
9330986c JK |
7670 | case BPF_MAP_TYPE_BLOOM_FILTER: |
7671 | if (meta->func_id == BPF_FUNC_map_peek_elem) | |
7672 | *arg_type = ARG_PTR_TO_MAP_VALUE; | |
7673 | break; | |
912f442c LB |
7674 | default: |
7675 | break; | |
7676 | } | |
7677 | return 0; | |
7678 | } | |
7679 | ||
f79e7ea5 LB |
7680 | struct bpf_reg_types { |
7681 | const enum bpf_reg_type types[10]; | |
1df8f55a | 7682 | u32 *btf_id; |
f79e7ea5 LB |
7683 | }; |
7684 | ||
f79e7ea5 LB |
7685 | static const struct bpf_reg_types sock_types = { |
7686 | .types = { | |
7687 | PTR_TO_SOCK_COMMON, | |
7688 | PTR_TO_SOCKET, | |
7689 | PTR_TO_TCP_SOCK, | |
7690 | PTR_TO_XDP_SOCK, | |
7691 | }, | |
7692 | }; | |
7693 | ||
49a2a4d4 | 7694 | #ifdef CONFIG_NET |
1df8f55a MKL |
7695 | static const struct bpf_reg_types btf_id_sock_common_types = { |
7696 | .types = { | |
7697 | PTR_TO_SOCK_COMMON, | |
7698 | PTR_TO_SOCKET, | |
7699 | PTR_TO_TCP_SOCK, | |
7700 | PTR_TO_XDP_SOCK, | |
7701 | PTR_TO_BTF_ID, | |
3f00c523 | 7702 | PTR_TO_BTF_ID | PTR_TRUSTED, |
1df8f55a MKL |
7703 | }, |
7704 | .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
7705 | }; | |
49a2a4d4 | 7706 | #endif |
1df8f55a | 7707 | |
f79e7ea5 LB |
7708 | static const struct bpf_reg_types mem_types = { |
7709 | .types = { | |
7710 | PTR_TO_STACK, | |
7711 | PTR_TO_PACKET, | |
7712 | PTR_TO_PACKET_META, | |
69c087ba | 7713 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
7714 | PTR_TO_MAP_VALUE, |
7715 | PTR_TO_MEM, | |
894f2a8b | 7716 | PTR_TO_MEM | MEM_RINGBUF, |
20b2aff4 | 7717 | PTR_TO_BUF, |
3e30be42 | 7718 | PTR_TO_BTF_ID | PTR_TRUSTED, |
f79e7ea5 LB |
7719 | }, |
7720 | }; | |
7721 | ||
7722 | static const struct bpf_reg_types int_ptr_types = { | |
7723 | .types = { | |
7724 | PTR_TO_STACK, | |
7725 | PTR_TO_PACKET, | |
7726 | PTR_TO_PACKET_META, | |
69c087ba | 7727 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
7728 | PTR_TO_MAP_VALUE, |
7729 | }, | |
7730 | }; | |
7731 | ||
4e814da0 KKD |
7732 | static const struct bpf_reg_types spin_lock_types = { |
7733 | .types = { | |
7734 | PTR_TO_MAP_VALUE, | |
7735 | PTR_TO_BTF_ID | MEM_ALLOC, | |
7736 | } | |
7737 | }; | |
7738 | ||
f79e7ea5 LB |
7739 | static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } }; |
7740 | static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } }; | |
7741 | static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } }; | |
894f2a8b | 7742 | static const struct bpf_reg_types ringbuf_mem_types = { .types = { PTR_TO_MEM | MEM_RINGBUF } }; |
f79e7ea5 | 7743 | static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } }; |
3f00c523 DV |
7744 | static const struct bpf_reg_types btf_ptr_types = { |
7745 | .types = { | |
7746 | PTR_TO_BTF_ID, | |
7747 | PTR_TO_BTF_ID | PTR_TRUSTED, | |
fca1aa75 | 7748 | PTR_TO_BTF_ID | MEM_RCU, |
3f00c523 DV |
7749 | }, |
7750 | }; | |
7751 | static const struct bpf_reg_types percpu_btf_ptr_types = { | |
7752 | .types = { | |
7753 | PTR_TO_BTF_ID | MEM_PERCPU, | |
7754 | PTR_TO_BTF_ID | MEM_PERCPU | PTR_TRUSTED, | |
7755 | } | |
7756 | }; | |
69c087ba YS |
7757 | static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } }; |
7758 | static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } }; | |
fff13c4b | 7759 | static const struct bpf_reg_types const_str_ptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
b00628b1 | 7760 | static const struct bpf_reg_types timer_types = { .types = { PTR_TO_MAP_VALUE } }; |
c0a5a21c | 7761 | static const struct bpf_reg_types kptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
20571567 DV |
7762 | static const struct bpf_reg_types dynptr_types = { |
7763 | .types = { | |
7764 | PTR_TO_STACK, | |
27060531 | 7765 | CONST_PTR_TO_DYNPTR, |
20571567 DV |
7766 | } |
7767 | }; | |
f79e7ea5 | 7768 | |
0789e13b | 7769 | static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = { |
d1673304 DM |
7770 | [ARG_PTR_TO_MAP_KEY] = &mem_types, |
7771 | [ARG_PTR_TO_MAP_VALUE] = &mem_types, | |
f79e7ea5 LB |
7772 | [ARG_CONST_SIZE] = &scalar_types, |
7773 | [ARG_CONST_SIZE_OR_ZERO] = &scalar_types, | |
7774 | [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types, | |
7775 | [ARG_CONST_MAP_PTR] = &const_map_ptr_types, | |
7776 | [ARG_PTR_TO_CTX] = &context_types, | |
f79e7ea5 | 7777 | [ARG_PTR_TO_SOCK_COMMON] = &sock_types, |
49a2a4d4 | 7778 | #ifdef CONFIG_NET |
1df8f55a | 7779 | [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types, |
49a2a4d4 | 7780 | #endif |
f79e7ea5 | 7781 | [ARG_PTR_TO_SOCKET] = &fullsock_types, |
f79e7ea5 LB |
7782 | [ARG_PTR_TO_BTF_ID] = &btf_ptr_types, |
7783 | [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types, | |
7784 | [ARG_PTR_TO_MEM] = &mem_types, | |
894f2a8b | 7785 | [ARG_PTR_TO_RINGBUF_MEM] = &ringbuf_mem_types, |
f79e7ea5 LB |
7786 | [ARG_PTR_TO_INT] = &int_ptr_types, |
7787 | [ARG_PTR_TO_LONG] = &int_ptr_types, | |
eaa6bcb7 | 7788 | [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types, |
69c087ba | 7789 | [ARG_PTR_TO_FUNC] = &func_ptr_types, |
48946bd6 | 7790 | [ARG_PTR_TO_STACK] = &stack_ptr_types, |
fff13c4b | 7791 | [ARG_PTR_TO_CONST_STR] = &const_str_ptr_types, |
b00628b1 | 7792 | [ARG_PTR_TO_TIMER] = &timer_types, |
c0a5a21c | 7793 | [ARG_PTR_TO_KPTR] = &kptr_types, |
20571567 | 7794 | [ARG_PTR_TO_DYNPTR] = &dynptr_types, |
f79e7ea5 LB |
7795 | }; |
7796 | ||
7797 | static int check_reg_type(struct bpf_verifier_env *env, u32 regno, | |
a968d5e2 | 7798 | enum bpf_arg_type arg_type, |
c0a5a21c KKD |
7799 | const u32 *arg_btf_id, |
7800 | struct bpf_call_arg_meta *meta) | |
f79e7ea5 LB |
7801 | { |
7802 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
7803 | enum bpf_reg_type expected, type = reg->type; | |
a968d5e2 | 7804 | const struct bpf_reg_types *compatible; |
f79e7ea5 LB |
7805 | int i, j; |
7806 | ||
48946bd6 | 7807 | compatible = compatible_reg_types[base_type(arg_type)]; |
a968d5e2 MKL |
7808 | if (!compatible) { |
7809 | verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type); | |
7810 | return -EFAULT; | |
7811 | } | |
7812 | ||
216e3cd2 HL |
7813 | /* ARG_PTR_TO_MEM + RDONLY is compatible with PTR_TO_MEM and PTR_TO_MEM + RDONLY, |
7814 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM and NOT with PTR_TO_MEM + RDONLY | |
7815 | * | |
7816 | * Same for MAYBE_NULL: | |
7817 | * | |
7818 | * ARG_PTR_TO_MEM + MAYBE_NULL is compatible with PTR_TO_MEM and PTR_TO_MEM + MAYBE_NULL, | |
7819 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM but NOT with PTR_TO_MEM + MAYBE_NULL | |
7820 | * | |
2012c867 DR |
7821 | * ARG_PTR_TO_MEM is compatible with PTR_TO_MEM that is tagged with a dynptr type. |
7822 | * | |
216e3cd2 HL |
7823 | * Therefore we fold these flags depending on the arg_type before comparison. |
7824 | */ | |
7825 | if (arg_type & MEM_RDONLY) | |
7826 | type &= ~MEM_RDONLY; | |
7827 | if (arg_type & PTR_MAYBE_NULL) | |
7828 | type &= ~PTR_MAYBE_NULL; | |
2012c867 DR |
7829 | if (base_type(arg_type) == ARG_PTR_TO_MEM) |
7830 | type &= ~DYNPTR_TYPE_FLAG_MASK; | |
216e3cd2 | 7831 | |
503e4def | 7832 | if (meta->func_id == BPF_FUNC_kptr_xchg && type_is_alloc(type)) |
738c96d5 DM |
7833 | type &= ~MEM_ALLOC; |
7834 | ||
f79e7ea5 LB |
7835 | for (i = 0; i < ARRAY_SIZE(compatible->types); i++) { |
7836 | expected = compatible->types[i]; | |
7837 | if (expected == NOT_INIT) | |
7838 | break; | |
7839 | ||
7840 | if (type == expected) | |
a968d5e2 | 7841 | goto found; |
f79e7ea5 LB |
7842 | } |
7843 | ||
216e3cd2 | 7844 | verbose(env, "R%d type=%s expected=", regno, reg_type_str(env, reg->type)); |
f79e7ea5 | 7845 | for (j = 0; j + 1 < i; j++) |
c25b2ae1 HL |
7846 | verbose(env, "%s, ", reg_type_str(env, compatible->types[j])); |
7847 | verbose(env, "%s\n", reg_type_str(env, compatible->types[j])); | |
f79e7ea5 | 7848 | return -EACCES; |
a968d5e2 MKL |
7849 | |
7850 | found: | |
da03e43a KKD |
7851 | if (base_type(reg->type) != PTR_TO_BTF_ID) |
7852 | return 0; | |
7853 | ||
3e30be42 AS |
7854 | if (compatible == &mem_types) { |
7855 | if (!(arg_type & MEM_RDONLY)) { | |
7856 | verbose(env, | |
7857 | "%s() may write into memory pointed by R%d type=%s\n", | |
7858 | func_id_name(meta->func_id), | |
7859 | regno, reg_type_str(env, reg->type)); | |
7860 | return -EACCES; | |
7861 | } | |
7862 | return 0; | |
7863 | } | |
7864 | ||
da03e43a KKD |
7865 | switch ((int)reg->type) { |
7866 | case PTR_TO_BTF_ID: | |
7867 | case PTR_TO_BTF_ID | PTR_TRUSTED: | |
7868 | case PTR_TO_BTF_ID | MEM_RCU: | |
add68b84 AS |
7869 | case PTR_TO_BTF_ID | PTR_MAYBE_NULL: |
7870 | case PTR_TO_BTF_ID | PTR_MAYBE_NULL | MEM_RCU: | |
da03e43a | 7871 | { |
2ab3b380 KKD |
7872 | /* For bpf_sk_release, it needs to match against first member |
7873 | * 'struct sock_common', hence make an exception for it. This | |
7874 | * allows bpf_sk_release to work for multiple socket types. | |
7875 | */ | |
7876 | bool strict_type_match = arg_type_is_release(arg_type) && | |
7877 | meta->func_id != BPF_FUNC_sk_release; | |
7878 | ||
add68b84 AS |
7879 | if (type_may_be_null(reg->type) && |
7880 | (!type_may_be_null(arg_type) || arg_type_is_release(arg_type))) { | |
7881 | verbose(env, "Possibly NULL pointer passed to helper arg%d\n", regno); | |
7882 | return -EACCES; | |
7883 | } | |
7884 | ||
1df8f55a MKL |
7885 | if (!arg_btf_id) { |
7886 | if (!compatible->btf_id) { | |
7887 | verbose(env, "verifier internal error: missing arg compatible BTF ID\n"); | |
7888 | return -EFAULT; | |
7889 | } | |
7890 | arg_btf_id = compatible->btf_id; | |
7891 | } | |
7892 | ||
c0a5a21c | 7893 | if (meta->func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac | 7894 | if (map_kptr_match_type(env, meta->kptr_field, reg, regno)) |
c0a5a21c | 7895 | return -EACCES; |
47e34cb7 DM |
7896 | } else { |
7897 | if (arg_btf_id == BPF_PTR_POISON) { | |
7898 | verbose(env, "verifier internal error:"); | |
7899 | verbose(env, "R%d has non-overwritten BPF_PTR_POISON type\n", | |
7900 | regno); | |
7901 | return -EACCES; | |
7902 | } | |
7903 | ||
7904 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
7905 | btf_vmlinux, *arg_btf_id, | |
7906 | strict_type_match)) { | |
7907 | verbose(env, "R%d is of type %s but %s is expected\n", | |
b32a5dae DM |
7908 | regno, btf_type_name(reg->btf, reg->btf_id), |
7909 | btf_type_name(btf_vmlinux, *arg_btf_id)); | |
47e34cb7 DM |
7910 | return -EACCES; |
7911 | } | |
a968d5e2 | 7912 | } |
da03e43a KKD |
7913 | break; |
7914 | } | |
7915 | case PTR_TO_BTF_ID | MEM_ALLOC: | |
738c96d5 DM |
7916 | if (meta->func_id != BPF_FUNC_spin_lock && meta->func_id != BPF_FUNC_spin_unlock && |
7917 | meta->func_id != BPF_FUNC_kptr_xchg) { | |
4e814da0 KKD |
7918 | verbose(env, "verifier internal error: unimplemented handling of MEM_ALLOC\n"); |
7919 | return -EFAULT; | |
7920 | } | |
ab6c637a YS |
7921 | if (meta->func_id == BPF_FUNC_kptr_xchg) { |
7922 | if (map_kptr_match_type(env, meta->kptr_field, reg, regno)) | |
7923 | return -EACCES; | |
7924 | } | |
da03e43a KKD |
7925 | break; |
7926 | case PTR_TO_BTF_ID | MEM_PERCPU: | |
7927 | case PTR_TO_BTF_ID | MEM_PERCPU | PTR_TRUSTED: | |
7928 | /* Handled by helper specific checks */ | |
7929 | break; | |
7930 | default: | |
7931 | verbose(env, "verifier internal error: invalid PTR_TO_BTF_ID register for type match\n"); | |
7932 | return -EFAULT; | |
a968d5e2 | 7933 | } |
a968d5e2 | 7934 | return 0; |
f79e7ea5 LB |
7935 | } |
7936 | ||
6a3cd331 DM |
7937 | static struct btf_field * |
7938 | reg_find_field_offset(const struct bpf_reg_state *reg, s32 off, u32 fields) | |
7939 | { | |
7940 | struct btf_field *field; | |
7941 | struct btf_record *rec; | |
7942 | ||
7943 | rec = reg_btf_record(reg); | |
7944 | if (!rec) | |
7945 | return NULL; | |
7946 | ||
7947 | field = btf_record_find(rec, off, fields); | |
7948 | if (!field) | |
7949 | return NULL; | |
7950 | ||
7951 | return field; | |
7952 | } | |
7953 | ||
25b35dd2 KKD |
7954 | int check_func_arg_reg_off(struct bpf_verifier_env *env, |
7955 | const struct bpf_reg_state *reg, int regno, | |
8f14852e | 7956 | enum bpf_arg_type arg_type) |
25b35dd2 | 7957 | { |
184c9bdb | 7958 | u32 type = reg->type; |
25b35dd2 | 7959 | |
184c9bdb KKD |
7960 | /* When referenced register is passed to release function, its fixed |
7961 | * offset must be 0. | |
7962 | * | |
7963 | * We will check arg_type_is_release reg has ref_obj_id when storing | |
7964 | * meta->release_regno. | |
7965 | */ | |
7966 | if (arg_type_is_release(arg_type)) { | |
7967 | /* ARG_PTR_TO_DYNPTR with OBJ_RELEASE is a bit special, as it | |
7968 | * may not directly point to the object being released, but to | |
7969 | * dynptr pointing to such object, which might be at some offset | |
7970 | * on the stack. In that case, we simply to fallback to the | |
7971 | * default handling. | |
7972 | */ | |
7973 | if (arg_type_is_dynptr(arg_type) && type == PTR_TO_STACK) | |
7974 | return 0; | |
6a3cd331 | 7975 | |
184c9bdb KKD |
7976 | /* Doing check_ptr_off_reg check for the offset will catch this |
7977 | * because fixed_off_ok is false, but checking here allows us | |
7978 | * to give the user a better error message. | |
7979 | */ | |
7980 | if (reg->off) { | |
7981 | verbose(env, "R%d must have zero offset when passed to release func or trusted arg to kfunc\n", | |
7982 | regno); | |
7983 | return -EINVAL; | |
7984 | } | |
7985 | return __check_ptr_off_reg(env, reg, regno, false); | |
7986 | } | |
7987 | ||
7988 | switch (type) { | |
7989 | /* Pointer types where both fixed and variable offset is explicitly allowed: */ | |
97e03f52 | 7990 | case PTR_TO_STACK: |
25b35dd2 KKD |
7991 | case PTR_TO_PACKET: |
7992 | case PTR_TO_PACKET_META: | |
7993 | case PTR_TO_MAP_KEY: | |
7994 | case PTR_TO_MAP_VALUE: | |
7995 | case PTR_TO_MEM: | |
7996 | case PTR_TO_MEM | MEM_RDONLY: | |
894f2a8b | 7997 | case PTR_TO_MEM | MEM_RINGBUF: |
25b35dd2 KKD |
7998 | case PTR_TO_BUF: |
7999 | case PTR_TO_BUF | MEM_RDONLY: | |
97e03f52 | 8000 | case SCALAR_VALUE: |
184c9bdb | 8001 | return 0; |
25b35dd2 KKD |
8002 | /* All the rest must be rejected, except PTR_TO_BTF_ID which allows |
8003 | * fixed offset. | |
8004 | */ | |
8005 | case PTR_TO_BTF_ID: | |
282de143 | 8006 | case PTR_TO_BTF_ID | MEM_ALLOC: |
3f00c523 | 8007 | case PTR_TO_BTF_ID | PTR_TRUSTED: |
fca1aa75 | 8008 | case PTR_TO_BTF_ID | MEM_RCU: |
6a3cd331 | 8009 | case PTR_TO_BTF_ID | MEM_ALLOC | NON_OWN_REF: |
0816b8c6 | 8010 | case PTR_TO_BTF_ID | MEM_ALLOC | NON_OWN_REF | MEM_RCU: |
24d5bb80 | 8011 | /* When referenced PTR_TO_BTF_ID is passed to release function, |
184c9bdb KKD |
8012 | * its fixed offset must be 0. In the other cases, fixed offset |
8013 | * can be non-zero. This was already checked above. So pass | |
8014 | * fixed_off_ok as true to allow fixed offset for all other | |
8015 | * cases. var_off always must be 0 for PTR_TO_BTF_ID, hence we | |
8016 | * still need to do checks instead of returning. | |
24d5bb80 | 8017 | */ |
184c9bdb | 8018 | return __check_ptr_off_reg(env, reg, regno, true); |
25b35dd2 | 8019 | default: |
184c9bdb | 8020 | return __check_ptr_off_reg(env, reg, regno, false); |
25b35dd2 | 8021 | } |
25b35dd2 KKD |
8022 | } |
8023 | ||
485ec51e JK |
8024 | static struct bpf_reg_state *get_dynptr_arg_reg(struct bpf_verifier_env *env, |
8025 | const struct bpf_func_proto *fn, | |
8026 | struct bpf_reg_state *regs) | |
8027 | { | |
8028 | struct bpf_reg_state *state = NULL; | |
8029 | int i; | |
8030 | ||
8031 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) | |
8032 | if (arg_type_is_dynptr(fn->arg_type[i])) { | |
8033 | if (state) { | |
8034 | verbose(env, "verifier internal error: multiple dynptr args\n"); | |
8035 | return NULL; | |
8036 | } | |
8037 | state = ®s[BPF_REG_1 + i]; | |
8038 | } | |
8039 | ||
8040 | if (!state) | |
8041 | verbose(env, "verifier internal error: no dynptr arg found\n"); | |
8042 | ||
8043 | return state; | |
8044 | } | |
8045 | ||
f8064ab9 | 8046 | static int dynptr_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
34d4ef57 JK |
8047 | { |
8048 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8049 | int spi; |
34d4ef57 | 8050 | |
27060531 | 8051 | if (reg->type == CONST_PTR_TO_DYNPTR) |
f8064ab9 KKD |
8052 | return reg->id; |
8053 | spi = dynptr_get_spi(env, reg); | |
8054 | if (spi < 0) | |
8055 | return spi; | |
8056 | return state->stack[spi].spilled_ptr.id; | |
8057 | } | |
8058 | ||
79168a66 | 8059 | static int dynptr_ref_obj_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
34d4ef57 JK |
8060 | { |
8061 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8062 | int spi; |
27060531 | 8063 | |
27060531 KKD |
8064 | if (reg->type == CONST_PTR_TO_DYNPTR) |
8065 | return reg->ref_obj_id; | |
79168a66 KKD |
8066 | spi = dynptr_get_spi(env, reg); |
8067 | if (spi < 0) | |
8068 | return spi; | |
27060531 | 8069 | return state->stack[spi].spilled_ptr.ref_obj_id; |
34d4ef57 JK |
8070 | } |
8071 | ||
b5964b96 JK |
8072 | static enum bpf_dynptr_type dynptr_get_type(struct bpf_verifier_env *env, |
8073 | struct bpf_reg_state *reg) | |
8074 | { | |
8075 | struct bpf_func_state *state = func(env, reg); | |
8076 | int spi; | |
8077 | ||
8078 | if (reg->type == CONST_PTR_TO_DYNPTR) | |
8079 | return reg->dynptr.type; | |
8080 | ||
8081 | spi = __get_spi(reg->off); | |
8082 | if (spi < 0) { | |
8083 | verbose(env, "verifier internal error: invalid spi when querying dynptr type\n"); | |
8084 | return BPF_DYNPTR_TYPE_INVALID; | |
8085 | } | |
8086 | ||
8087 | return state->stack[spi].spilled_ptr.dynptr.type; | |
8088 | } | |
8089 | ||
af7ec138 YS |
8090 | static int check_func_arg(struct bpf_verifier_env *env, u32 arg, |
8091 | struct bpf_call_arg_meta *meta, | |
1d18feb2 JK |
8092 | const struct bpf_func_proto *fn, |
8093 | int insn_idx) | |
17a52670 | 8094 | { |
af7ec138 | 8095 | u32 regno = BPF_REG_1 + arg; |
638f5b90 | 8096 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
af7ec138 | 8097 | enum bpf_arg_type arg_type = fn->arg_type[arg]; |
f79e7ea5 | 8098 | enum bpf_reg_type type = reg->type; |
508362ac | 8099 | u32 *arg_btf_id = NULL; |
17a52670 AS |
8100 | int err = 0; |
8101 | ||
80f1d68c | 8102 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
8103 | return 0; |
8104 | ||
dc503a8a EC |
8105 | err = check_reg_arg(env, regno, SRC_OP); |
8106 | if (err) | |
8107 | return err; | |
17a52670 | 8108 | |
1be7f75d AS |
8109 | if (arg_type == ARG_ANYTHING) { |
8110 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
8111 | verbose(env, "R%d leaks addr into helper function\n", |
8112 | regno); | |
1be7f75d AS |
8113 | return -EACCES; |
8114 | } | |
80f1d68c | 8115 | return 0; |
1be7f75d | 8116 | } |
80f1d68c | 8117 | |
de8f3a83 | 8118 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 8119 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 8120 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
8121 | return -EACCES; |
8122 | } | |
8123 | ||
16d1e00c | 8124 | if (base_type(arg_type) == ARG_PTR_TO_MAP_VALUE) { |
912f442c LB |
8125 | err = resolve_map_arg_type(env, meta, &arg_type); |
8126 | if (err) | |
8127 | return err; | |
8128 | } | |
8129 | ||
48946bd6 | 8130 | if (register_is_null(reg) && type_may_be_null(arg_type)) |
fd1b0d60 LB |
8131 | /* A NULL register has a SCALAR_VALUE type, so skip |
8132 | * type checking. | |
8133 | */ | |
8134 | goto skip_type_check; | |
8135 | ||
508362ac | 8136 | /* arg_btf_id and arg_size are in a union. */ |
4e814da0 KKD |
8137 | if (base_type(arg_type) == ARG_PTR_TO_BTF_ID || |
8138 | base_type(arg_type) == ARG_PTR_TO_SPIN_LOCK) | |
508362ac MM |
8139 | arg_btf_id = fn->arg_btf_id[arg]; |
8140 | ||
8141 | err = check_reg_type(env, regno, arg_type, arg_btf_id, meta); | |
f79e7ea5 LB |
8142 | if (err) |
8143 | return err; | |
8144 | ||
8f14852e | 8145 | err = check_func_arg_reg_off(env, reg, regno, arg_type); |
25b35dd2 KKD |
8146 | if (err) |
8147 | return err; | |
d7b9454a | 8148 | |
fd1b0d60 | 8149 | skip_type_check: |
8f14852e | 8150 | if (arg_type_is_release(arg_type)) { |
bc34dee6 JK |
8151 | if (arg_type_is_dynptr(arg_type)) { |
8152 | struct bpf_func_state *state = func(env, reg); | |
27060531 | 8153 | int spi; |
bc34dee6 | 8154 | |
27060531 KKD |
8155 | /* Only dynptr created on stack can be released, thus |
8156 | * the get_spi and stack state checks for spilled_ptr | |
8157 | * should only be done before process_dynptr_func for | |
8158 | * PTR_TO_STACK. | |
8159 | */ | |
8160 | if (reg->type == PTR_TO_STACK) { | |
79168a66 | 8161 | spi = dynptr_get_spi(env, reg); |
f5b625e5 | 8162 | if (spi < 0 || !state->stack[spi].spilled_ptr.ref_obj_id) { |
27060531 KKD |
8163 | verbose(env, "arg %d is an unacquired reference\n", regno); |
8164 | return -EINVAL; | |
8165 | } | |
8166 | } else { | |
8167 | verbose(env, "cannot release unowned const bpf_dynptr\n"); | |
bc34dee6 JK |
8168 | return -EINVAL; |
8169 | } | |
8170 | } else if (!reg->ref_obj_id && !register_is_null(reg)) { | |
8f14852e KKD |
8171 | verbose(env, "R%d must be referenced when passed to release function\n", |
8172 | regno); | |
8173 | return -EINVAL; | |
8174 | } | |
8175 | if (meta->release_regno) { | |
8176 | verbose(env, "verifier internal error: more than one release argument\n"); | |
8177 | return -EFAULT; | |
8178 | } | |
8179 | meta->release_regno = regno; | |
8180 | } | |
8181 | ||
02f7c958 | 8182 | if (reg->ref_obj_id) { |
457f4436 AN |
8183 | if (meta->ref_obj_id) { |
8184 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
8185 | regno, reg->ref_obj_id, | |
8186 | meta->ref_obj_id); | |
8187 | return -EFAULT; | |
8188 | } | |
8189 | meta->ref_obj_id = reg->ref_obj_id; | |
17a52670 AS |
8190 | } |
8191 | ||
8ab4cdcf JK |
8192 | switch (base_type(arg_type)) { |
8193 | case ARG_CONST_MAP_PTR: | |
17a52670 | 8194 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ |
3e8ce298 AS |
8195 | if (meta->map_ptr) { |
8196 | /* Use map_uid (which is unique id of inner map) to reject: | |
8197 | * inner_map1 = bpf_map_lookup_elem(outer_map, key1) | |
8198 | * inner_map2 = bpf_map_lookup_elem(outer_map, key2) | |
8199 | * if (inner_map1 && inner_map2) { | |
8200 | * timer = bpf_map_lookup_elem(inner_map1); | |
8201 | * if (timer) | |
8202 | * // mismatch would have been allowed | |
8203 | * bpf_timer_init(timer, inner_map2); | |
8204 | * } | |
8205 | * | |
8206 | * Comparing map_ptr is enough to distinguish normal and outer maps. | |
8207 | */ | |
8208 | if (meta->map_ptr != reg->map_ptr || | |
8209 | meta->map_uid != reg->map_uid) { | |
8210 | verbose(env, | |
8211 | "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n", | |
8212 | meta->map_uid, reg->map_uid); | |
8213 | return -EINVAL; | |
8214 | } | |
b00628b1 | 8215 | } |
33ff9823 | 8216 | meta->map_ptr = reg->map_ptr; |
3e8ce298 | 8217 | meta->map_uid = reg->map_uid; |
8ab4cdcf JK |
8218 | break; |
8219 | case ARG_PTR_TO_MAP_KEY: | |
17a52670 AS |
8220 | /* bpf_map_xxx(..., map_ptr, ..., key) call: |
8221 | * check that [key, key + map->key_size) are within | |
8222 | * stack limits and initialized | |
8223 | */ | |
33ff9823 | 8224 | if (!meta->map_ptr) { |
17a52670 AS |
8225 | /* in function declaration map_ptr must come before |
8226 | * map_key, so that it's verified and known before | |
8227 | * we have to check map_key here. Otherwise it means | |
8228 | * that kernel subsystem misconfigured verifier | |
8229 | */ | |
61bd5218 | 8230 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
8231 | return -EACCES; |
8232 | } | |
d71962f3 PC |
8233 | err = check_helper_mem_access(env, regno, |
8234 | meta->map_ptr->key_size, false, | |
8235 | NULL); | |
8ab4cdcf JK |
8236 | break; |
8237 | case ARG_PTR_TO_MAP_VALUE: | |
48946bd6 HL |
8238 | if (type_may_be_null(arg_type) && register_is_null(reg)) |
8239 | return 0; | |
8240 | ||
17a52670 AS |
8241 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
8242 | * check [value, value + map->value_size) validity | |
8243 | */ | |
33ff9823 | 8244 | if (!meta->map_ptr) { |
17a52670 | 8245 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 8246 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
8247 | return -EACCES; |
8248 | } | |
16d1e00c | 8249 | meta->raw_mode = arg_type & MEM_UNINIT; |
d71962f3 PC |
8250 | err = check_helper_mem_access(env, regno, |
8251 | meta->map_ptr->value_size, false, | |
2ea864c5 | 8252 | meta); |
8ab4cdcf JK |
8253 | break; |
8254 | case ARG_PTR_TO_PERCPU_BTF_ID: | |
eaa6bcb7 HL |
8255 | if (!reg->btf_id) { |
8256 | verbose(env, "Helper has invalid btf_id in R%d\n", regno); | |
8257 | return -EACCES; | |
8258 | } | |
22dc4a0f | 8259 | meta->ret_btf = reg->btf; |
eaa6bcb7 | 8260 | meta->ret_btf_id = reg->btf_id; |
8ab4cdcf JK |
8261 | break; |
8262 | case ARG_PTR_TO_SPIN_LOCK: | |
5d92ddc3 DM |
8263 | if (in_rbtree_lock_required_cb(env)) { |
8264 | verbose(env, "can't spin_{lock,unlock} in rbtree cb\n"); | |
8265 | return -EACCES; | |
8266 | } | |
c18f0b6a | 8267 | if (meta->func_id == BPF_FUNC_spin_lock) { |
ac50fe51 KKD |
8268 | err = process_spin_lock(env, regno, true); |
8269 | if (err) | |
8270 | return err; | |
c18f0b6a | 8271 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { |
ac50fe51 KKD |
8272 | err = process_spin_lock(env, regno, false); |
8273 | if (err) | |
8274 | return err; | |
c18f0b6a LB |
8275 | } else { |
8276 | verbose(env, "verifier internal error\n"); | |
8277 | return -EFAULT; | |
8278 | } | |
8ab4cdcf JK |
8279 | break; |
8280 | case ARG_PTR_TO_TIMER: | |
ac50fe51 KKD |
8281 | err = process_timer_func(env, regno, meta); |
8282 | if (err) | |
8283 | return err; | |
8ab4cdcf JK |
8284 | break; |
8285 | case ARG_PTR_TO_FUNC: | |
69c087ba | 8286 | meta->subprogno = reg->subprogno; |
8ab4cdcf JK |
8287 | break; |
8288 | case ARG_PTR_TO_MEM: | |
a2bbe7cc LB |
8289 | /* The access to this pointer is only checked when we hit the |
8290 | * next is_mem_size argument below. | |
8291 | */ | |
16d1e00c | 8292 | meta->raw_mode = arg_type & MEM_UNINIT; |
508362ac MM |
8293 | if (arg_type & MEM_FIXED_SIZE) { |
8294 | err = check_helper_mem_access(env, regno, | |
8295 | fn->arg_size[arg], false, | |
8296 | meta); | |
8297 | } | |
8ab4cdcf JK |
8298 | break; |
8299 | case ARG_CONST_SIZE: | |
8300 | err = check_mem_size_reg(env, reg, regno, false, meta); | |
8301 | break; | |
8302 | case ARG_CONST_SIZE_OR_ZERO: | |
8303 | err = check_mem_size_reg(env, reg, regno, true, meta); | |
8304 | break; | |
8305 | case ARG_PTR_TO_DYNPTR: | |
361f129f | 8306 | err = process_dynptr_func(env, regno, insn_idx, arg_type, 0); |
ac50fe51 KKD |
8307 | if (err) |
8308 | return err; | |
8ab4cdcf JK |
8309 | break; |
8310 | case ARG_CONST_ALLOC_SIZE_OR_ZERO: | |
457f4436 | 8311 | if (!tnum_is_const(reg->var_off)) { |
28a8add6 | 8312 | verbose(env, "R%d is not a known constant'\n", |
457f4436 AN |
8313 | regno); |
8314 | return -EACCES; | |
8315 | } | |
8316 | meta->mem_size = reg->var_off.value; | |
2fc31465 KKD |
8317 | err = mark_chain_precision(env, regno); |
8318 | if (err) | |
8319 | return err; | |
8ab4cdcf JK |
8320 | break; |
8321 | case ARG_PTR_TO_INT: | |
8322 | case ARG_PTR_TO_LONG: | |
8323 | { | |
57c3bb72 AI |
8324 | int size = int_ptr_type_to_size(arg_type); |
8325 | ||
8326 | err = check_helper_mem_access(env, regno, size, false, meta); | |
8327 | if (err) | |
8328 | return err; | |
8329 | err = check_ptr_alignment(env, reg, 0, size, true); | |
8ab4cdcf JK |
8330 | break; |
8331 | } | |
8332 | case ARG_PTR_TO_CONST_STR: | |
8333 | { | |
fff13c4b FR |
8334 | struct bpf_map *map = reg->map_ptr; |
8335 | int map_off; | |
8336 | u64 map_addr; | |
8337 | char *str_ptr; | |
8338 | ||
a8fad73e | 8339 | if (!bpf_map_is_rdonly(map)) { |
fff13c4b FR |
8340 | verbose(env, "R%d does not point to a readonly map'\n", regno); |
8341 | return -EACCES; | |
8342 | } | |
8343 | ||
8344 | if (!tnum_is_const(reg->var_off)) { | |
8345 | verbose(env, "R%d is not a constant address'\n", regno); | |
8346 | return -EACCES; | |
8347 | } | |
8348 | ||
8349 | if (!map->ops->map_direct_value_addr) { | |
8350 | verbose(env, "no direct value access support for this map type\n"); | |
8351 | return -EACCES; | |
8352 | } | |
8353 | ||
8354 | err = check_map_access(env, regno, reg->off, | |
61df10c7 KKD |
8355 | map->value_size - reg->off, false, |
8356 | ACCESS_HELPER); | |
fff13c4b FR |
8357 | if (err) |
8358 | return err; | |
8359 | ||
8360 | map_off = reg->off + reg->var_off.value; | |
8361 | err = map->ops->map_direct_value_addr(map, &map_addr, map_off); | |
8362 | if (err) { | |
8363 | verbose(env, "direct value access on string failed\n"); | |
8364 | return err; | |
8365 | } | |
8366 | ||
8367 | str_ptr = (char *)(long)(map_addr); | |
8368 | if (!strnchr(str_ptr + map_off, map->value_size - map_off, 0)) { | |
8369 | verbose(env, "string is not zero-terminated\n"); | |
8370 | return -EINVAL; | |
8371 | } | |
8ab4cdcf JK |
8372 | break; |
8373 | } | |
8374 | case ARG_PTR_TO_KPTR: | |
ac50fe51 KKD |
8375 | err = process_kptr_func(env, regno, meta); |
8376 | if (err) | |
8377 | return err; | |
8ab4cdcf | 8378 | break; |
17a52670 AS |
8379 | } |
8380 | ||
8381 | return err; | |
8382 | } | |
8383 | ||
0126240f LB |
8384 | static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) |
8385 | { | |
8386 | enum bpf_attach_type eatype = env->prog->expected_attach_type; | |
7e40781c | 8387 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
0126240f LB |
8388 | |
8389 | if (func_id != BPF_FUNC_map_update_elem) | |
8390 | return false; | |
8391 | ||
8392 | /* It's not possible to get access to a locked struct sock in these | |
8393 | * contexts, so updating is safe. | |
8394 | */ | |
8395 | switch (type) { | |
8396 | case BPF_PROG_TYPE_TRACING: | |
8397 | if (eatype == BPF_TRACE_ITER) | |
8398 | return true; | |
8399 | break; | |
8400 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
8401 | case BPF_PROG_TYPE_SCHED_CLS: | |
8402 | case BPF_PROG_TYPE_SCHED_ACT: | |
8403 | case BPF_PROG_TYPE_XDP: | |
8404 | case BPF_PROG_TYPE_SK_REUSEPORT: | |
8405 | case BPF_PROG_TYPE_FLOW_DISSECTOR: | |
8406 | case BPF_PROG_TYPE_SK_LOOKUP: | |
8407 | return true; | |
8408 | default: | |
8409 | break; | |
8410 | } | |
8411 | ||
8412 | verbose(env, "cannot update sockmap in this context\n"); | |
8413 | return false; | |
8414 | } | |
8415 | ||
e411901c MF |
8416 | static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env) |
8417 | { | |
95acd881 TA |
8418 | return env->prog->jit_requested && |
8419 | bpf_jit_supports_subprog_tailcalls(); | |
e411901c MF |
8420 | } |
8421 | ||
61bd5218 JK |
8422 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
8423 | struct bpf_map *map, int func_id) | |
35578d79 | 8424 | { |
35578d79 KX |
8425 | if (!map) |
8426 | return 0; | |
8427 | ||
6aff67c8 AS |
8428 | /* We need a two way check, first is from map perspective ... */ |
8429 | switch (map->map_type) { | |
8430 | case BPF_MAP_TYPE_PROG_ARRAY: | |
8431 | if (func_id != BPF_FUNC_tail_call) | |
8432 | goto error; | |
8433 | break; | |
8434 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
8435 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 8436 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 8437 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
8438 | func_id != BPF_FUNC_perf_event_read_value && |
8439 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
8440 | goto error; |
8441 | break; | |
457f4436 AN |
8442 | case BPF_MAP_TYPE_RINGBUF: |
8443 | if (func_id != BPF_FUNC_ringbuf_output && | |
8444 | func_id != BPF_FUNC_ringbuf_reserve && | |
bc34dee6 JK |
8445 | func_id != BPF_FUNC_ringbuf_query && |
8446 | func_id != BPF_FUNC_ringbuf_reserve_dynptr && | |
8447 | func_id != BPF_FUNC_ringbuf_submit_dynptr && | |
8448 | func_id != BPF_FUNC_ringbuf_discard_dynptr) | |
457f4436 AN |
8449 | goto error; |
8450 | break; | |
583c1f42 | 8451 | case BPF_MAP_TYPE_USER_RINGBUF: |
20571567 DV |
8452 | if (func_id != BPF_FUNC_user_ringbuf_drain) |
8453 | goto error; | |
8454 | break; | |
6aff67c8 AS |
8455 | case BPF_MAP_TYPE_STACK_TRACE: |
8456 | if (func_id != BPF_FUNC_get_stackid) | |
8457 | goto error; | |
8458 | break; | |
4ed8ec52 | 8459 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 8460 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 8461 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
8462 | goto error; |
8463 | break; | |
cd339431 | 8464 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 8465 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
8466 | if (func_id != BPF_FUNC_get_local_storage) |
8467 | goto error; | |
8468 | break; | |
546ac1ff | 8469 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 8470 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
8471 | if (func_id != BPF_FUNC_redirect_map && |
8472 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
8473 | goto error; |
8474 | break; | |
fbfc504a BT |
8475 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
8476 | * appear. | |
8477 | */ | |
6710e112 JDB |
8478 | case BPF_MAP_TYPE_CPUMAP: |
8479 | if (func_id != BPF_FUNC_redirect_map) | |
8480 | goto error; | |
8481 | break; | |
fada7fdc JL |
8482 | case BPF_MAP_TYPE_XSKMAP: |
8483 | if (func_id != BPF_FUNC_redirect_map && | |
8484 | func_id != BPF_FUNC_map_lookup_elem) | |
8485 | goto error; | |
8486 | break; | |
56f668df | 8487 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 8488 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
8489 | if (func_id != BPF_FUNC_map_lookup_elem) |
8490 | goto error; | |
16a43625 | 8491 | break; |
174a79ff JF |
8492 | case BPF_MAP_TYPE_SOCKMAP: |
8493 | if (func_id != BPF_FUNC_sk_redirect_map && | |
8494 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 8495 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 8496 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 | 8497 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
8498 | func_id != BPF_FUNC_map_lookup_elem && |
8499 | !may_update_sockmap(env, func_id)) | |
174a79ff JF |
8500 | goto error; |
8501 | break; | |
81110384 JF |
8502 | case BPF_MAP_TYPE_SOCKHASH: |
8503 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
8504 | func_id != BPF_FUNC_sock_hash_update && | |
8505 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 8506 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 | 8507 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
8508 | func_id != BPF_FUNC_map_lookup_elem && |
8509 | !may_update_sockmap(env, func_id)) | |
81110384 JF |
8510 | goto error; |
8511 | break; | |
2dbb9b9e MKL |
8512 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
8513 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
8514 | goto error; | |
8515 | break; | |
f1a2e44a MV |
8516 | case BPF_MAP_TYPE_QUEUE: |
8517 | case BPF_MAP_TYPE_STACK: | |
8518 | if (func_id != BPF_FUNC_map_peek_elem && | |
8519 | func_id != BPF_FUNC_map_pop_elem && | |
8520 | func_id != BPF_FUNC_map_push_elem) | |
8521 | goto error; | |
8522 | break; | |
6ac99e8f MKL |
8523 | case BPF_MAP_TYPE_SK_STORAGE: |
8524 | if (func_id != BPF_FUNC_sk_storage_get && | |
9db44fdd KKD |
8525 | func_id != BPF_FUNC_sk_storage_delete && |
8526 | func_id != BPF_FUNC_kptr_xchg) | |
6ac99e8f MKL |
8527 | goto error; |
8528 | break; | |
8ea63684 KS |
8529 | case BPF_MAP_TYPE_INODE_STORAGE: |
8530 | if (func_id != BPF_FUNC_inode_storage_get && | |
9db44fdd KKD |
8531 | func_id != BPF_FUNC_inode_storage_delete && |
8532 | func_id != BPF_FUNC_kptr_xchg) | |
8ea63684 KS |
8533 | goto error; |
8534 | break; | |
4cf1bc1f KS |
8535 | case BPF_MAP_TYPE_TASK_STORAGE: |
8536 | if (func_id != BPF_FUNC_task_storage_get && | |
9db44fdd KKD |
8537 | func_id != BPF_FUNC_task_storage_delete && |
8538 | func_id != BPF_FUNC_kptr_xchg) | |
4cf1bc1f KS |
8539 | goto error; |
8540 | break; | |
c4bcfb38 YS |
8541 | case BPF_MAP_TYPE_CGRP_STORAGE: |
8542 | if (func_id != BPF_FUNC_cgrp_storage_get && | |
9db44fdd KKD |
8543 | func_id != BPF_FUNC_cgrp_storage_delete && |
8544 | func_id != BPF_FUNC_kptr_xchg) | |
c4bcfb38 YS |
8545 | goto error; |
8546 | break; | |
9330986c JK |
8547 | case BPF_MAP_TYPE_BLOOM_FILTER: |
8548 | if (func_id != BPF_FUNC_map_peek_elem && | |
8549 | func_id != BPF_FUNC_map_push_elem) | |
8550 | goto error; | |
8551 | break; | |
6aff67c8 AS |
8552 | default: |
8553 | break; | |
8554 | } | |
8555 | ||
8556 | /* ... and second from the function itself. */ | |
8557 | switch (func_id) { | |
8558 | case BPF_FUNC_tail_call: | |
8559 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
8560 | goto error; | |
e411901c MF |
8561 | if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) { |
8562 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
f4d7e40a AS |
8563 | return -EINVAL; |
8564 | } | |
6aff67c8 AS |
8565 | break; |
8566 | case BPF_FUNC_perf_event_read: | |
8567 | case BPF_FUNC_perf_event_output: | |
908432ca | 8568 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 8569 | case BPF_FUNC_skb_output: |
d831ee84 | 8570 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
8571 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
8572 | goto error; | |
8573 | break; | |
5b029a32 DB |
8574 | case BPF_FUNC_ringbuf_output: |
8575 | case BPF_FUNC_ringbuf_reserve: | |
8576 | case BPF_FUNC_ringbuf_query: | |
bc34dee6 JK |
8577 | case BPF_FUNC_ringbuf_reserve_dynptr: |
8578 | case BPF_FUNC_ringbuf_submit_dynptr: | |
8579 | case BPF_FUNC_ringbuf_discard_dynptr: | |
5b029a32 DB |
8580 | if (map->map_type != BPF_MAP_TYPE_RINGBUF) |
8581 | goto error; | |
8582 | break; | |
20571567 DV |
8583 | case BPF_FUNC_user_ringbuf_drain: |
8584 | if (map->map_type != BPF_MAP_TYPE_USER_RINGBUF) | |
8585 | goto error; | |
8586 | break; | |
6aff67c8 AS |
8587 | case BPF_FUNC_get_stackid: |
8588 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
8589 | goto error; | |
8590 | break; | |
60d20f91 | 8591 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 8592 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
8593 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
8594 | goto error; | |
8595 | break; | |
97f91a7c | 8596 | case BPF_FUNC_redirect_map: |
9c270af3 | 8597 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 8598 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
8599 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
8600 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
8601 | goto error; |
8602 | break; | |
174a79ff | 8603 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 8604 | case BPF_FUNC_msg_redirect_map: |
81110384 | 8605 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
8606 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
8607 | goto error; | |
8608 | break; | |
81110384 JF |
8609 | case BPF_FUNC_sk_redirect_hash: |
8610 | case BPF_FUNC_msg_redirect_hash: | |
8611 | case BPF_FUNC_sock_hash_update: | |
8612 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
8613 | goto error; |
8614 | break; | |
cd339431 | 8615 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
8616 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
8617 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
8618 | goto error; |
8619 | break; | |
2dbb9b9e | 8620 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
8621 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
8622 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
8623 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
8624 | goto error; |
8625 | break; | |
f1a2e44a | 8626 | case BPF_FUNC_map_pop_elem: |
f1a2e44a MV |
8627 | if (map->map_type != BPF_MAP_TYPE_QUEUE && |
8628 | map->map_type != BPF_MAP_TYPE_STACK) | |
8629 | goto error; | |
8630 | break; | |
9330986c JK |
8631 | case BPF_FUNC_map_peek_elem: |
8632 | case BPF_FUNC_map_push_elem: | |
8633 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
8634 | map->map_type != BPF_MAP_TYPE_STACK && | |
8635 | map->map_type != BPF_MAP_TYPE_BLOOM_FILTER) | |
8636 | goto error; | |
8637 | break; | |
07343110 FZ |
8638 | case BPF_FUNC_map_lookup_percpu_elem: |
8639 | if (map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY && | |
8640 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && | |
8641 | map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH) | |
8642 | goto error; | |
8643 | break; | |
6ac99e8f MKL |
8644 | case BPF_FUNC_sk_storage_get: |
8645 | case BPF_FUNC_sk_storage_delete: | |
8646 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
8647 | goto error; | |
8648 | break; | |
8ea63684 KS |
8649 | case BPF_FUNC_inode_storage_get: |
8650 | case BPF_FUNC_inode_storage_delete: | |
8651 | if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE) | |
8652 | goto error; | |
8653 | break; | |
4cf1bc1f KS |
8654 | case BPF_FUNC_task_storage_get: |
8655 | case BPF_FUNC_task_storage_delete: | |
8656 | if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) | |
8657 | goto error; | |
8658 | break; | |
c4bcfb38 YS |
8659 | case BPF_FUNC_cgrp_storage_get: |
8660 | case BPF_FUNC_cgrp_storage_delete: | |
8661 | if (map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) | |
8662 | goto error; | |
8663 | break; | |
6aff67c8 AS |
8664 | default: |
8665 | break; | |
35578d79 KX |
8666 | } |
8667 | ||
8668 | return 0; | |
6aff67c8 | 8669 | error: |
61bd5218 | 8670 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 8671 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 8672 | return -EINVAL; |
35578d79 KX |
8673 | } |
8674 | ||
90133415 | 8675 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
8676 | { |
8677 | int count = 0; | |
8678 | ||
39f19ebb | 8679 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 8680 | count++; |
39f19ebb | 8681 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 8682 | count++; |
39f19ebb | 8683 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 8684 | count++; |
39f19ebb | 8685 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 8686 | count++; |
39f19ebb | 8687 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
8688 | count++; |
8689 | ||
90133415 DB |
8690 | /* We only support one arg being in raw mode at the moment, |
8691 | * which is sufficient for the helper functions we have | |
8692 | * right now. | |
8693 | */ | |
8694 | return count <= 1; | |
8695 | } | |
8696 | ||
508362ac | 8697 | static bool check_args_pair_invalid(const struct bpf_func_proto *fn, int arg) |
90133415 | 8698 | { |
508362ac MM |
8699 | bool is_fixed = fn->arg_type[arg] & MEM_FIXED_SIZE; |
8700 | bool has_size = fn->arg_size[arg] != 0; | |
8701 | bool is_next_size = false; | |
8702 | ||
8703 | if (arg + 1 < ARRAY_SIZE(fn->arg_type)) | |
8704 | is_next_size = arg_type_is_mem_size(fn->arg_type[arg + 1]); | |
8705 | ||
8706 | if (base_type(fn->arg_type[arg]) != ARG_PTR_TO_MEM) | |
8707 | return is_next_size; | |
8708 | ||
8709 | return has_size == is_next_size || is_next_size == is_fixed; | |
90133415 DB |
8710 | } |
8711 | ||
8712 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
8713 | { | |
8714 | /* bpf_xxx(..., buf, len) call will access 'len' | |
8715 | * bytes from memory 'buf'. Both arg types need | |
8716 | * to be paired, so make sure there's no buggy | |
8717 | * helper function specification. | |
8718 | */ | |
8719 | if (arg_type_is_mem_size(fn->arg1_type) || | |
508362ac MM |
8720 | check_args_pair_invalid(fn, 0) || |
8721 | check_args_pair_invalid(fn, 1) || | |
8722 | check_args_pair_invalid(fn, 2) || | |
8723 | check_args_pair_invalid(fn, 3) || | |
8724 | check_args_pair_invalid(fn, 4)) | |
90133415 DB |
8725 | return false; |
8726 | ||
8727 | return true; | |
8728 | } | |
8729 | ||
9436ef6e LB |
8730 | static bool check_btf_id_ok(const struct bpf_func_proto *fn) |
8731 | { | |
8732 | int i; | |
8733 | ||
1df8f55a | 8734 | for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) { |
4e814da0 KKD |
8735 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_BTF_ID) |
8736 | return !!fn->arg_btf_id[i]; | |
8737 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_SPIN_LOCK) | |
8738 | return fn->arg_btf_id[i] == BPF_PTR_POISON; | |
508362ac MM |
8739 | if (base_type(fn->arg_type[i]) != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i] && |
8740 | /* arg_btf_id and arg_size are in a union. */ | |
8741 | (base_type(fn->arg_type[i]) != ARG_PTR_TO_MEM || | |
8742 | !(fn->arg_type[i] & MEM_FIXED_SIZE))) | |
1df8f55a MKL |
8743 | return false; |
8744 | } | |
8745 | ||
9436ef6e LB |
8746 | return true; |
8747 | } | |
8748 | ||
0c9a7a7e | 8749 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
8750 | { |
8751 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 8752 | check_arg_pair_ok(fn) && |
b2d8ef19 | 8753 | check_btf_id_ok(fn) ? 0 : -EINVAL; |
435faee1 DB |
8754 | } |
8755 | ||
de8f3a83 DB |
8756 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
8757 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
66e3a13e JK |
8758 | * |
8759 | * This also applies to dynptr slices belonging to skb and xdp dynptrs, | |
8760 | * since these slices point to packet data. | |
f1174f77 | 8761 | */ |
b239da34 | 8762 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 8763 | { |
b239da34 KKD |
8764 | struct bpf_func_state *state; |
8765 | struct bpf_reg_state *reg; | |
969bf05e | 8766 | |
b239da34 | 8767 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
66e3a13e | 8768 | if (reg_is_pkt_pointer_any(reg) || reg_is_dynptr_slice_pkt(reg)) |
dbd8d228 | 8769 | mark_reg_invalid(env, reg); |
b239da34 | 8770 | })); |
f4d7e40a AS |
8771 | } |
8772 | ||
6d94e741 AS |
8773 | enum { |
8774 | AT_PKT_END = -1, | |
8775 | BEYOND_PKT_END = -2, | |
8776 | }; | |
8777 | ||
8778 | static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open) | |
8779 | { | |
8780 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
8781 | struct bpf_reg_state *reg = &state->regs[regn]; | |
8782 | ||
8783 | if (reg->type != PTR_TO_PACKET) | |
8784 | /* PTR_TO_PACKET_META is not supported yet */ | |
8785 | return; | |
8786 | ||
8787 | /* The 'reg' is pkt > pkt_end or pkt >= pkt_end. | |
8788 | * How far beyond pkt_end it goes is unknown. | |
8789 | * if (!range_open) it's the case of pkt >= pkt_end | |
8790 | * if (range_open) it's the case of pkt > pkt_end | |
8791 | * hence this pointer is at least 1 byte bigger than pkt_end | |
8792 | */ | |
8793 | if (range_open) | |
8794 | reg->range = BEYOND_PKT_END; | |
8795 | else | |
8796 | reg->range = AT_PKT_END; | |
8797 | } | |
8798 | ||
fd978bf7 JS |
8799 | /* The pointer with the specified id has released its reference to kernel |
8800 | * resources. Identify all copies of the same pointer and clear the reference. | |
8801 | */ | |
8802 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 8803 | int ref_obj_id) |
fd978bf7 | 8804 | { |
b239da34 KKD |
8805 | struct bpf_func_state *state; |
8806 | struct bpf_reg_state *reg; | |
1b986589 | 8807 | int err; |
fd978bf7 | 8808 | |
1b986589 MKL |
8809 | err = release_reference_state(cur_func(env), ref_obj_id); |
8810 | if (err) | |
8811 | return err; | |
8812 | ||
b239da34 | 8813 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
dbd8d228 KKD |
8814 | if (reg->ref_obj_id == ref_obj_id) |
8815 | mark_reg_invalid(env, reg); | |
b239da34 | 8816 | })); |
fd978bf7 | 8817 | |
1b986589 | 8818 | return 0; |
fd978bf7 JS |
8819 | } |
8820 | ||
6a3cd331 DM |
8821 | static void invalidate_non_owning_refs(struct bpf_verifier_env *env) |
8822 | { | |
8823 | struct bpf_func_state *unused; | |
8824 | struct bpf_reg_state *reg; | |
8825 | ||
8826 | bpf_for_each_reg_in_vstate(env->cur_state, unused, reg, ({ | |
8827 | if (type_is_non_owning_ref(reg->type)) | |
dbd8d228 | 8828 | mark_reg_invalid(env, reg); |
6a3cd331 DM |
8829 | })); |
8830 | } | |
8831 | ||
51c39bb1 AS |
8832 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
8833 | struct bpf_reg_state *regs) | |
8834 | { | |
8835 | int i; | |
8836 | ||
8837 | /* after the call registers r0 - r5 were scratched */ | |
8838 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
8839 | mark_reg_not_init(env, regs, caller_saved[i]); | |
8840 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
8841 | } | |
8842 | } | |
8843 | ||
14351375 YS |
8844 | typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env, |
8845 | struct bpf_func_state *caller, | |
8846 | struct bpf_func_state *callee, | |
8847 | int insn_idx); | |
8848 | ||
be2ef816 AN |
8849 | static int set_callee_state(struct bpf_verifier_env *env, |
8850 | struct bpf_func_state *caller, | |
8851 | struct bpf_func_state *callee, int insn_idx); | |
8852 | ||
14351375 YS |
8853 | static int __check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
8854 | int *insn_idx, int subprog, | |
8855 | set_callee_state_fn set_callee_state_cb) | |
f4d7e40a AS |
8856 | { |
8857 | struct bpf_verifier_state *state = env->cur_state; | |
8858 | struct bpf_func_state *caller, *callee; | |
14351375 | 8859 | int err; |
f4d7e40a | 8860 | |
aada9ce6 | 8861 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 8862 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 8863 | state->curframe + 2); |
f4d7e40a AS |
8864 | return -E2BIG; |
8865 | } | |
8866 | ||
f4d7e40a AS |
8867 | caller = state->frame[state->curframe]; |
8868 | if (state->frame[state->curframe + 1]) { | |
8869 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
8870 | state->curframe + 1); | |
8871 | return -EFAULT; | |
8872 | } | |
8873 | ||
95f2f26f | 8874 | err = btf_check_subprog_call(env, subprog, caller->regs); |
51c39bb1 AS |
8875 | if (err == -EFAULT) |
8876 | return err; | |
fde2a388 | 8877 | if (subprog_is_global(env, subprog)) { |
51c39bb1 AS |
8878 | if (err) { |
8879 | verbose(env, "Caller passes invalid args into func#%d\n", | |
8880 | subprog); | |
8881 | return err; | |
8882 | } else { | |
8883 | if (env->log.level & BPF_LOG_LEVEL) | |
8884 | verbose(env, | |
8885 | "Func#%d is global and valid. Skipping.\n", | |
8886 | subprog); | |
8887 | clear_caller_saved_regs(env, caller->regs); | |
8888 | ||
45159b27 | 8889 | /* All global functions return a 64-bit SCALAR_VALUE */ |
51c39bb1 | 8890 | mark_reg_unknown(env, caller->regs, BPF_REG_0); |
45159b27 | 8891 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; |
51c39bb1 AS |
8892 | |
8893 | /* continue with next insn after call */ | |
8894 | return 0; | |
8895 | } | |
8896 | } | |
8897 | ||
be2ef816 AN |
8898 | /* set_callee_state is used for direct subprog calls, but we are |
8899 | * interested in validating only BPF helpers that can call subprogs as | |
8900 | * callbacks | |
8901 | */ | |
5d92ddc3 DM |
8902 | if (set_callee_state_cb != set_callee_state) { |
8903 | if (bpf_pseudo_kfunc_call(insn) && | |
8904 | !is_callback_calling_kfunc(insn->imm)) { | |
8905 | verbose(env, "verifier bug: kfunc %s#%d not marked as callback-calling\n", | |
8906 | func_id_name(insn->imm), insn->imm); | |
8907 | return -EFAULT; | |
8908 | } else if (!bpf_pseudo_kfunc_call(insn) && | |
8909 | !is_callback_calling_function(insn->imm)) { /* helper */ | |
8910 | verbose(env, "verifier bug: helper %s#%d not marked as callback-calling\n", | |
8911 | func_id_name(insn->imm), insn->imm); | |
8912 | return -EFAULT; | |
8913 | } | |
be2ef816 AN |
8914 | } |
8915 | ||
bfc6bb74 | 8916 | if (insn->code == (BPF_JMP | BPF_CALL) && |
a5bebc4f | 8917 | insn->src_reg == 0 && |
bfc6bb74 AS |
8918 | insn->imm == BPF_FUNC_timer_set_callback) { |
8919 | struct bpf_verifier_state *async_cb; | |
8920 | ||
8921 | /* there is no real recursion here. timer callbacks are async */ | |
7ddc80a4 | 8922 | env->subprog_info[subprog].is_async_cb = true; |
bfc6bb74 AS |
8923 | async_cb = push_async_cb(env, env->subprog_info[subprog].start, |
8924 | *insn_idx, subprog); | |
8925 | if (!async_cb) | |
8926 | return -EFAULT; | |
8927 | callee = async_cb->frame[0]; | |
8928 | callee->async_entry_cnt = caller->async_entry_cnt + 1; | |
8929 | ||
8930 | /* Convert bpf_timer_set_callback() args into timer callback args */ | |
8931 | err = set_callee_state_cb(env, caller, callee, *insn_idx); | |
8932 | if (err) | |
8933 | return err; | |
8934 | ||
8935 | clear_caller_saved_regs(env, caller->regs); | |
8936 | mark_reg_unknown(env, caller->regs, BPF_REG_0); | |
8937 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
8938 | /* continue with next insn after call */ | |
8939 | return 0; | |
8940 | } | |
8941 | ||
f4d7e40a AS |
8942 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); |
8943 | if (!callee) | |
8944 | return -ENOMEM; | |
8945 | state->frame[state->curframe + 1] = callee; | |
8946 | ||
8947 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
8948 | * into its own stack before reading from it. | |
8949 | * callee can read/write into caller's stack | |
8950 | */ | |
8951 | init_func_state(env, callee, | |
8952 | /* remember the callsite, it will be used by bpf_exit */ | |
8953 | *insn_idx /* callsite */, | |
8954 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 8955 | subprog /* subprog number within this prog */); |
f4d7e40a | 8956 | |
fd978bf7 | 8957 | /* Transfer references to the callee */ |
c69431aa | 8958 | err = copy_reference_state(callee, caller); |
fd978bf7 | 8959 | if (err) |
eb86559a | 8960 | goto err_out; |
fd978bf7 | 8961 | |
14351375 YS |
8962 | err = set_callee_state_cb(env, caller, callee, *insn_idx); |
8963 | if (err) | |
eb86559a | 8964 | goto err_out; |
f4d7e40a | 8965 | |
51c39bb1 | 8966 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a AS |
8967 | |
8968 | /* only increment it after check_reg_arg() finished */ | |
8969 | state->curframe++; | |
8970 | ||
8971 | /* and go analyze first insn of the callee */ | |
14351375 | 8972 | *insn_idx = env->subprog_info[subprog].start - 1; |
f4d7e40a | 8973 | |
06ee7115 | 8974 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 8975 | verbose(env, "caller:\n"); |
0f55f9ed | 8976 | print_verifier_state(env, caller, true); |
f4d7e40a | 8977 | verbose(env, "callee:\n"); |
0f55f9ed | 8978 | print_verifier_state(env, callee, true); |
f4d7e40a AS |
8979 | } |
8980 | return 0; | |
eb86559a WY |
8981 | |
8982 | err_out: | |
8983 | free_func_state(callee); | |
8984 | state->frame[state->curframe + 1] = NULL; | |
8985 | return err; | |
f4d7e40a AS |
8986 | } |
8987 | ||
314ee05e YS |
8988 | int map_set_for_each_callback_args(struct bpf_verifier_env *env, |
8989 | struct bpf_func_state *caller, | |
8990 | struct bpf_func_state *callee) | |
8991 | { | |
8992 | /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn, | |
8993 | * void *callback_ctx, u64 flags); | |
8994 | * callback_fn(struct bpf_map *map, void *key, void *value, | |
8995 | * void *callback_ctx); | |
8996 | */ | |
8997 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
8998 | ||
8999 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
9000 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9001 | callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9002 | ||
9003 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
9004 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
9005 | callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9006 | ||
9007 | /* pointer to stack or null */ | |
9008 | callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3]; | |
9009 | ||
9010 | /* unused */ | |
9011 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9012 | return 0; | |
9013 | } | |
9014 | ||
14351375 YS |
9015 | static int set_callee_state(struct bpf_verifier_env *env, |
9016 | struct bpf_func_state *caller, | |
9017 | struct bpf_func_state *callee, int insn_idx) | |
9018 | { | |
9019 | int i; | |
9020 | ||
9021 | /* copy r1 - r5 args that callee can access. The copy includes parent | |
9022 | * pointers, which connects us up to the liveness chain | |
9023 | */ | |
9024 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
9025 | callee->regs[i] = caller->regs[i]; | |
9026 | return 0; | |
9027 | } | |
9028 | ||
9029 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
9030 | int *insn_idx) | |
9031 | { | |
9032 | int subprog, target_insn; | |
9033 | ||
9034 | target_insn = *insn_idx + insn->imm + 1; | |
9035 | subprog = find_subprog(env, target_insn); | |
9036 | if (subprog < 0) { | |
9037 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
9038 | target_insn); | |
9039 | return -EFAULT; | |
9040 | } | |
9041 | ||
9042 | return __check_func_call(env, insn, insn_idx, subprog, set_callee_state); | |
9043 | } | |
9044 | ||
69c087ba YS |
9045 | static int set_map_elem_callback_state(struct bpf_verifier_env *env, |
9046 | struct bpf_func_state *caller, | |
9047 | struct bpf_func_state *callee, | |
9048 | int insn_idx) | |
9049 | { | |
9050 | struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx]; | |
9051 | struct bpf_map *map; | |
9052 | int err; | |
9053 | ||
9054 | if (bpf_map_ptr_poisoned(insn_aux)) { | |
9055 | verbose(env, "tail_call abusing map_ptr\n"); | |
9056 | return -EINVAL; | |
9057 | } | |
9058 | ||
9059 | map = BPF_MAP_PTR(insn_aux->map_ptr_state); | |
9060 | if (!map->ops->map_set_for_each_callback_args || | |
9061 | !map->ops->map_for_each_callback) { | |
9062 | verbose(env, "callback function not allowed for map\n"); | |
9063 | return -ENOTSUPP; | |
9064 | } | |
9065 | ||
9066 | err = map->ops->map_set_for_each_callback_args(env, caller, callee); | |
9067 | if (err) | |
9068 | return err; | |
9069 | ||
9070 | callee->in_callback_fn = true; | |
1bfe26fb | 9071 | callee->callback_ret_range = tnum_range(0, 1); |
69c087ba YS |
9072 | return 0; |
9073 | } | |
9074 | ||
e6f2dd0f JK |
9075 | static int set_loop_callback_state(struct bpf_verifier_env *env, |
9076 | struct bpf_func_state *caller, | |
9077 | struct bpf_func_state *callee, | |
9078 | int insn_idx) | |
9079 | { | |
9080 | /* bpf_loop(u32 nr_loops, void *callback_fn, void *callback_ctx, | |
9081 | * u64 flags); | |
9082 | * callback_fn(u32 index, void *callback_ctx); | |
9083 | */ | |
9084 | callee->regs[BPF_REG_1].type = SCALAR_VALUE; | |
9085 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; | |
9086 | ||
9087 | /* unused */ | |
9088 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9089 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9090 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9091 | ||
9092 | callee->in_callback_fn = true; | |
1bfe26fb | 9093 | callee->callback_ret_range = tnum_range(0, 1); |
e6f2dd0f JK |
9094 | return 0; |
9095 | } | |
9096 | ||
b00628b1 AS |
9097 | static int set_timer_callback_state(struct bpf_verifier_env *env, |
9098 | struct bpf_func_state *caller, | |
9099 | struct bpf_func_state *callee, | |
9100 | int insn_idx) | |
9101 | { | |
9102 | struct bpf_map *map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
9103 | ||
9104 | /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn); | |
9105 | * callback_fn(struct bpf_map *map, void *key, void *value); | |
9106 | */ | |
9107 | callee->regs[BPF_REG_1].type = CONST_PTR_TO_MAP; | |
9108 | __mark_reg_known_zero(&callee->regs[BPF_REG_1]); | |
9109 | callee->regs[BPF_REG_1].map_ptr = map_ptr; | |
9110 | ||
9111 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
9112 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9113 | callee->regs[BPF_REG_2].map_ptr = map_ptr; | |
9114 | ||
9115 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
9116 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
9117 | callee->regs[BPF_REG_3].map_ptr = map_ptr; | |
9118 | ||
9119 | /* unused */ | |
9120 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9121 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
bfc6bb74 | 9122 | callee->in_async_callback_fn = true; |
1bfe26fb | 9123 | callee->callback_ret_range = tnum_range(0, 1); |
b00628b1 AS |
9124 | return 0; |
9125 | } | |
9126 | ||
7c7e3d31 SL |
9127 | static int set_find_vma_callback_state(struct bpf_verifier_env *env, |
9128 | struct bpf_func_state *caller, | |
9129 | struct bpf_func_state *callee, | |
9130 | int insn_idx) | |
9131 | { | |
9132 | /* bpf_find_vma(struct task_struct *task, u64 addr, | |
9133 | * void *callback_fn, void *callback_ctx, u64 flags) | |
9134 | * (callback_fn)(struct task_struct *task, | |
9135 | * struct vm_area_struct *vma, void *callback_ctx); | |
9136 | */ | |
9137 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
9138 | ||
9139 | callee->regs[BPF_REG_2].type = PTR_TO_BTF_ID; | |
9140 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
9141 | callee->regs[BPF_REG_2].btf = btf_vmlinux; | |
d19ddb47 | 9142 | callee->regs[BPF_REG_2].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA], |
7c7e3d31 SL |
9143 | |
9144 | /* pointer to stack or null */ | |
9145 | callee->regs[BPF_REG_3] = caller->regs[BPF_REG_4]; | |
9146 | ||
9147 | /* unused */ | |
9148 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9149 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9150 | callee->in_callback_fn = true; | |
1bfe26fb | 9151 | callee->callback_ret_range = tnum_range(0, 1); |
7c7e3d31 SL |
9152 | return 0; |
9153 | } | |
9154 | ||
20571567 DV |
9155 | static int set_user_ringbuf_callback_state(struct bpf_verifier_env *env, |
9156 | struct bpf_func_state *caller, | |
9157 | struct bpf_func_state *callee, | |
9158 | int insn_idx) | |
9159 | { | |
9160 | /* bpf_user_ringbuf_drain(struct bpf_map *map, void *callback_fn, void | |
9161 | * callback_ctx, u64 flags); | |
27060531 | 9162 | * callback_fn(const struct bpf_dynptr_t* dynptr, void *callback_ctx); |
20571567 DV |
9163 | */ |
9164 | __mark_reg_not_init(env, &callee->regs[BPF_REG_0]); | |
f8064ab9 | 9165 | mark_dynptr_cb_reg(env, &callee->regs[BPF_REG_1], BPF_DYNPTR_TYPE_LOCAL); |
20571567 DV |
9166 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; |
9167 | ||
9168 | /* unused */ | |
9169 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9170 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9171 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9172 | ||
9173 | callee->in_callback_fn = true; | |
c92a7a52 | 9174 | callee->callback_ret_range = tnum_range(0, 1); |
20571567 DV |
9175 | return 0; |
9176 | } | |
9177 | ||
5d92ddc3 DM |
9178 | static int set_rbtree_add_callback_state(struct bpf_verifier_env *env, |
9179 | struct bpf_func_state *caller, | |
9180 | struct bpf_func_state *callee, | |
9181 | int insn_idx) | |
9182 | { | |
d2dcc67d | 9183 | /* void bpf_rbtree_add_impl(struct bpf_rb_root *root, struct bpf_rb_node *node, |
5d92ddc3 DM |
9184 | * bool (less)(struct bpf_rb_node *a, const struct bpf_rb_node *b)); |
9185 | * | |
d2dcc67d | 9186 | * 'struct bpf_rb_node *node' arg to bpf_rbtree_add_impl is the same PTR_TO_BTF_ID w/ offset |
5d92ddc3 DM |
9187 | * that 'less' callback args will be receiving. However, 'node' arg was release_reference'd |
9188 | * by this point, so look at 'root' | |
9189 | */ | |
9190 | struct btf_field *field; | |
9191 | ||
9192 | field = reg_find_field_offset(&caller->regs[BPF_REG_1], caller->regs[BPF_REG_1].off, | |
9193 | BPF_RB_ROOT); | |
9194 | if (!field || !field->graph_root.value_btf_id) | |
9195 | return -EFAULT; | |
9196 | ||
9197 | mark_reg_graph_node(callee->regs, BPF_REG_1, &field->graph_root); | |
9198 | ref_set_non_owning(env, &callee->regs[BPF_REG_1]); | |
9199 | mark_reg_graph_node(callee->regs, BPF_REG_2, &field->graph_root); | |
9200 | ref_set_non_owning(env, &callee->regs[BPF_REG_2]); | |
9201 | ||
9202 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
9203 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
9204 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
9205 | callee->in_callback_fn = true; | |
9206 | callee->callback_ret_range = tnum_range(0, 1); | |
9207 | return 0; | |
9208 | } | |
9209 | ||
9210 | static bool is_rbtree_lock_required_kfunc(u32 btf_id); | |
9211 | ||
9212 | /* Are we currently verifying the callback for a rbtree helper that must | |
9213 | * be called with lock held? If so, no need to complain about unreleased | |
9214 | * lock | |
9215 | */ | |
9216 | static bool in_rbtree_lock_required_cb(struct bpf_verifier_env *env) | |
9217 | { | |
9218 | struct bpf_verifier_state *state = env->cur_state; | |
9219 | struct bpf_insn *insn = env->prog->insnsi; | |
9220 | struct bpf_func_state *callee; | |
9221 | int kfunc_btf_id; | |
9222 | ||
9223 | if (!state->curframe) | |
9224 | return false; | |
9225 | ||
9226 | callee = state->frame[state->curframe]; | |
9227 | ||
9228 | if (!callee->in_callback_fn) | |
9229 | return false; | |
9230 | ||
9231 | kfunc_btf_id = insn[callee->callsite].imm; | |
9232 | return is_rbtree_lock_required_kfunc(kfunc_btf_id); | |
9233 | } | |
9234 | ||
f4d7e40a AS |
9235 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) |
9236 | { | |
9237 | struct bpf_verifier_state *state = env->cur_state; | |
9238 | struct bpf_func_state *caller, *callee; | |
9239 | struct bpf_reg_state *r0; | |
fd978bf7 | 9240 | int err; |
f4d7e40a AS |
9241 | |
9242 | callee = state->frame[state->curframe]; | |
9243 | r0 = &callee->regs[BPF_REG_0]; | |
9244 | if (r0->type == PTR_TO_STACK) { | |
9245 | /* technically it's ok to return caller's stack pointer | |
9246 | * (or caller's caller's pointer) back to the caller, | |
9247 | * since these pointers are valid. Only current stack | |
9248 | * pointer will be invalid as soon as function exits, | |
9249 | * but let's be conservative | |
9250 | */ | |
9251 | verbose(env, "cannot return stack pointer to the caller\n"); | |
9252 | return -EINVAL; | |
9253 | } | |
9254 | ||
eb86559a | 9255 | caller = state->frame[state->curframe - 1]; |
69c087ba YS |
9256 | if (callee->in_callback_fn) { |
9257 | /* enforce R0 return value range [0, 1]. */ | |
1bfe26fb | 9258 | struct tnum range = callee->callback_ret_range; |
69c087ba YS |
9259 | |
9260 | if (r0->type != SCALAR_VALUE) { | |
9261 | verbose(env, "R0 not a scalar value\n"); | |
9262 | return -EACCES; | |
9263 | } | |
9264 | if (!tnum_in(range, r0->var_off)) { | |
9265 | verbose_invalid_scalar(env, r0, &range, "callback return", "R0"); | |
9266 | return -EINVAL; | |
9267 | } | |
9268 | } else { | |
9269 | /* return to the caller whatever r0 had in the callee */ | |
9270 | caller->regs[BPF_REG_0] = *r0; | |
9271 | } | |
f4d7e40a | 9272 | |
9d9d00ac KKD |
9273 | /* callback_fn frame should have released its own additions to parent's |
9274 | * reference state at this point, or check_reference_leak would | |
9275 | * complain, hence it must be the same as the caller. There is no need | |
9276 | * to copy it back. | |
9277 | */ | |
9278 | if (!callee->in_callback_fn) { | |
9279 | /* Transfer references to the caller */ | |
9280 | err = copy_reference_state(caller, callee); | |
9281 | if (err) | |
9282 | return err; | |
9283 | } | |
fd978bf7 | 9284 | |
f4d7e40a | 9285 | *insn_idx = callee->callsite + 1; |
06ee7115 | 9286 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 9287 | verbose(env, "returning from callee:\n"); |
0f55f9ed | 9288 | print_verifier_state(env, callee, true); |
f4d7e40a | 9289 | verbose(env, "to caller at %d:\n", *insn_idx); |
0f55f9ed | 9290 | print_verifier_state(env, caller, true); |
f4d7e40a AS |
9291 | } |
9292 | /* clear everything in the callee */ | |
9293 | free_func_state(callee); | |
eb86559a | 9294 | state->frame[state->curframe--] = NULL; |
f4d7e40a AS |
9295 | return 0; |
9296 | } | |
9297 | ||
849fa506 YS |
9298 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
9299 | int func_id, | |
9300 | struct bpf_call_arg_meta *meta) | |
9301 | { | |
9302 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
9303 | ||
f42bcd16 | 9304 | if (ret_type != RET_INTEGER) |
849fa506 YS |
9305 | return; |
9306 | ||
f42bcd16 AN |
9307 | switch (func_id) { |
9308 | case BPF_FUNC_get_stack: | |
9309 | case BPF_FUNC_get_task_stack: | |
9310 | case BPF_FUNC_probe_read_str: | |
9311 | case BPF_FUNC_probe_read_kernel_str: | |
9312 | case BPF_FUNC_probe_read_user_str: | |
9313 | ret_reg->smax_value = meta->msize_max_value; | |
9314 | ret_reg->s32_max_value = meta->msize_max_value; | |
9315 | ret_reg->smin_value = -MAX_ERRNO; | |
9316 | ret_reg->s32_min_value = -MAX_ERRNO; | |
9317 | reg_bounds_sync(ret_reg); | |
9318 | break; | |
9319 | case BPF_FUNC_get_smp_processor_id: | |
9320 | ret_reg->umax_value = nr_cpu_ids - 1; | |
9321 | ret_reg->u32_max_value = nr_cpu_ids - 1; | |
9322 | ret_reg->smax_value = nr_cpu_ids - 1; | |
9323 | ret_reg->s32_max_value = nr_cpu_ids - 1; | |
9324 | ret_reg->umin_value = 0; | |
9325 | ret_reg->u32_min_value = 0; | |
9326 | ret_reg->smin_value = 0; | |
9327 | ret_reg->s32_min_value = 0; | |
9328 | reg_bounds_sync(ret_reg); | |
9329 | break; | |
9330 | } | |
849fa506 YS |
9331 | } |
9332 | ||
c93552c4 DB |
9333 | static int |
9334 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
9335 | int func_id, int insn_idx) | |
9336 | { | |
9337 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 9338 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
9339 | |
9340 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
9341 | func_id != BPF_FUNC_map_lookup_elem && |
9342 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
9343 | func_id != BPF_FUNC_map_delete_elem && |
9344 | func_id != BPF_FUNC_map_push_elem && | |
9345 | func_id != BPF_FUNC_map_pop_elem && | |
69c087ba | 9346 | func_id != BPF_FUNC_map_peek_elem && |
e6a4750f | 9347 | func_id != BPF_FUNC_for_each_map_elem && |
07343110 FZ |
9348 | func_id != BPF_FUNC_redirect_map && |
9349 | func_id != BPF_FUNC_map_lookup_percpu_elem) | |
c93552c4 | 9350 | return 0; |
09772d92 | 9351 | |
591fe988 | 9352 | if (map == NULL) { |
c93552c4 DB |
9353 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
9354 | return -EINVAL; | |
9355 | } | |
9356 | ||
591fe988 DB |
9357 | /* In case of read-only, some additional restrictions |
9358 | * need to be applied in order to prevent altering the | |
9359 | * state of the map from program side. | |
9360 | */ | |
9361 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
9362 | (func_id == BPF_FUNC_map_delete_elem || | |
9363 | func_id == BPF_FUNC_map_update_elem || | |
9364 | func_id == BPF_FUNC_map_push_elem || | |
9365 | func_id == BPF_FUNC_map_pop_elem)) { | |
9366 | verbose(env, "write into map forbidden\n"); | |
9367 | return -EACCES; | |
9368 | } | |
9369 | ||
d2e4c1e6 | 9370 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 9371 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 9372 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 9373 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 9374 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 9375 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
9376 | return 0; |
9377 | } | |
9378 | ||
d2e4c1e6 DB |
9379 | static int |
9380 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
9381 | int func_id, int insn_idx) | |
9382 | { | |
9383 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
9384 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
9385 | struct bpf_map *map = meta->map_ptr; | |
a657182a | 9386 | u64 val, max; |
cc52d914 | 9387 | int err; |
d2e4c1e6 DB |
9388 | |
9389 | if (func_id != BPF_FUNC_tail_call) | |
9390 | return 0; | |
9391 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
9392 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
9393 | return -EINVAL; | |
9394 | } | |
9395 | ||
d2e4c1e6 | 9396 | reg = ®s[BPF_REG_3]; |
a657182a DB |
9397 | val = reg->var_off.value; |
9398 | max = map->max_entries; | |
d2e4c1e6 | 9399 | |
a657182a | 9400 | if (!(register_is_const(reg) && val < max)) { |
d2e4c1e6 DB |
9401 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); |
9402 | return 0; | |
9403 | } | |
9404 | ||
cc52d914 DB |
9405 | err = mark_chain_precision(env, BPF_REG_3); |
9406 | if (err) | |
9407 | return err; | |
d2e4c1e6 DB |
9408 | if (bpf_map_key_unseen(aux)) |
9409 | bpf_map_key_store(aux, val); | |
9410 | else if (!bpf_map_key_poisoned(aux) && | |
9411 | bpf_map_key_immediate(aux) != val) | |
9412 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
9413 | return 0; | |
9414 | } | |
9415 | ||
fd978bf7 JS |
9416 | static int check_reference_leak(struct bpf_verifier_env *env) |
9417 | { | |
9418 | struct bpf_func_state *state = cur_func(env); | |
9d9d00ac | 9419 | bool refs_lingering = false; |
fd978bf7 JS |
9420 | int i; |
9421 | ||
9d9d00ac KKD |
9422 | if (state->frameno && !state->in_callback_fn) |
9423 | return 0; | |
9424 | ||
fd978bf7 | 9425 | for (i = 0; i < state->acquired_refs; i++) { |
9d9d00ac KKD |
9426 | if (state->in_callback_fn && state->refs[i].callback_ref != state->frameno) |
9427 | continue; | |
fd978bf7 JS |
9428 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", |
9429 | state->refs[i].id, state->refs[i].insn_idx); | |
9d9d00ac | 9430 | refs_lingering = true; |
fd978bf7 | 9431 | } |
9d9d00ac | 9432 | return refs_lingering ? -EINVAL : 0; |
fd978bf7 JS |
9433 | } |
9434 | ||
7b15523a FR |
9435 | static int check_bpf_snprintf_call(struct bpf_verifier_env *env, |
9436 | struct bpf_reg_state *regs) | |
9437 | { | |
9438 | struct bpf_reg_state *fmt_reg = ®s[BPF_REG_3]; | |
9439 | struct bpf_reg_state *data_len_reg = ®s[BPF_REG_5]; | |
9440 | struct bpf_map *fmt_map = fmt_reg->map_ptr; | |
78aa1cc9 | 9441 | struct bpf_bprintf_data data = {}; |
7b15523a FR |
9442 | int err, fmt_map_off, num_args; |
9443 | u64 fmt_addr; | |
9444 | char *fmt; | |
9445 | ||
9446 | /* data must be an array of u64 */ | |
9447 | if (data_len_reg->var_off.value % 8) | |
9448 | return -EINVAL; | |
9449 | num_args = data_len_reg->var_off.value / 8; | |
9450 | ||
9451 | /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const | |
9452 | * and map_direct_value_addr is set. | |
9453 | */ | |
9454 | fmt_map_off = fmt_reg->off + fmt_reg->var_off.value; | |
9455 | err = fmt_map->ops->map_direct_value_addr(fmt_map, &fmt_addr, | |
9456 | fmt_map_off); | |
8e8ee109 FR |
9457 | if (err) { |
9458 | verbose(env, "verifier bug\n"); | |
9459 | return -EFAULT; | |
9460 | } | |
7b15523a FR |
9461 | fmt = (char *)(long)fmt_addr + fmt_map_off; |
9462 | ||
9463 | /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we | |
9464 | * can focus on validating the format specifiers. | |
9465 | */ | |
78aa1cc9 | 9466 | err = bpf_bprintf_prepare(fmt, UINT_MAX, NULL, num_args, &data); |
7b15523a FR |
9467 | if (err < 0) |
9468 | verbose(env, "Invalid format string\n"); | |
9469 | ||
9470 | return err; | |
9471 | } | |
9472 | ||
9b99edca JO |
9473 | static int check_get_func_ip(struct bpf_verifier_env *env) |
9474 | { | |
9b99edca JO |
9475 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
9476 | int func_id = BPF_FUNC_get_func_ip; | |
9477 | ||
9478 | if (type == BPF_PROG_TYPE_TRACING) { | |
f92c1e18 | 9479 | if (!bpf_prog_has_trampoline(env->prog)) { |
9b99edca JO |
9480 | verbose(env, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n", |
9481 | func_id_name(func_id), func_id); | |
9482 | return -ENOTSUPP; | |
9483 | } | |
9484 | return 0; | |
9ffd9f3f JO |
9485 | } else if (type == BPF_PROG_TYPE_KPROBE) { |
9486 | return 0; | |
9b99edca JO |
9487 | } |
9488 | ||
9489 | verbose(env, "func %s#%d not supported for program type %d\n", | |
9490 | func_id_name(func_id), func_id, type); | |
9491 | return -ENOTSUPP; | |
9492 | } | |
9493 | ||
1ade2371 EZ |
9494 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
9495 | { | |
9496 | return &env->insn_aux_data[env->insn_idx]; | |
9497 | } | |
9498 | ||
9499 | static bool loop_flag_is_zero(struct bpf_verifier_env *env) | |
9500 | { | |
9501 | struct bpf_reg_state *regs = cur_regs(env); | |
9502 | struct bpf_reg_state *reg = ®s[BPF_REG_4]; | |
9503 | bool reg_is_null = register_is_null(reg); | |
9504 | ||
9505 | if (reg_is_null) | |
9506 | mark_chain_precision(env, BPF_REG_4); | |
9507 | ||
9508 | return reg_is_null; | |
9509 | } | |
9510 | ||
9511 | static void update_loop_inline_state(struct bpf_verifier_env *env, u32 subprogno) | |
9512 | { | |
9513 | struct bpf_loop_inline_state *state = &cur_aux(env)->loop_inline_state; | |
9514 | ||
9515 | if (!state->initialized) { | |
9516 | state->initialized = 1; | |
9517 | state->fit_for_inline = loop_flag_is_zero(env); | |
9518 | state->callback_subprogno = subprogno; | |
9519 | return; | |
9520 | } | |
9521 | ||
9522 | if (!state->fit_for_inline) | |
9523 | return; | |
9524 | ||
9525 | state->fit_for_inline = (loop_flag_is_zero(env) && | |
9526 | state->callback_subprogno == subprogno); | |
9527 | } | |
9528 | ||
69c087ba YS |
9529 | static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
9530 | int *insn_idx_p) | |
17a52670 | 9531 | { |
aef9d4a3 | 9532 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
17a52670 | 9533 | const struct bpf_func_proto *fn = NULL; |
3c480732 | 9534 | enum bpf_return_type ret_type; |
c25b2ae1 | 9535 | enum bpf_type_flag ret_flag; |
638f5b90 | 9536 | struct bpf_reg_state *regs; |
33ff9823 | 9537 | struct bpf_call_arg_meta meta; |
69c087ba | 9538 | int insn_idx = *insn_idx_p; |
969bf05e | 9539 | bool changes_data; |
69c087ba | 9540 | int i, err, func_id; |
17a52670 AS |
9541 | |
9542 | /* find function prototype */ | |
69c087ba | 9543 | func_id = insn->imm; |
17a52670 | 9544 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { |
61bd5218 JK |
9545 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
9546 | func_id); | |
17a52670 AS |
9547 | return -EINVAL; |
9548 | } | |
9549 | ||
00176a34 | 9550 | if (env->ops->get_func_proto) |
5e43f899 | 9551 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 9552 | if (!fn) { |
61bd5218 JK |
9553 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
9554 | func_id); | |
17a52670 AS |
9555 | return -EINVAL; |
9556 | } | |
9557 | ||
9558 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 9559 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 9560 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
9561 | return -EINVAL; |
9562 | } | |
9563 | ||
eae2e83e JO |
9564 | if (fn->allowed && !fn->allowed(env->prog)) { |
9565 | verbose(env, "helper call is not allowed in probe\n"); | |
9566 | return -EINVAL; | |
9567 | } | |
9568 | ||
01685c5b YS |
9569 | if (!env->prog->aux->sleepable && fn->might_sleep) { |
9570 | verbose(env, "helper call might sleep in a non-sleepable prog\n"); | |
9571 | return -EINVAL; | |
9572 | } | |
9573 | ||
04514d13 | 9574 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 9575 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
9576 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
9577 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
9578 | func_id_name(func_id), func_id); | |
9579 | return -EINVAL; | |
9580 | } | |
969bf05e | 9581 | |
33ff9823 | 9582 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 9583 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 9584 | |
0c9a7a7e | 9585 | err = check_func_proto(fn, func_id); |
435faee1 | 9586 | if (err) { |
61bd5218 | 9587 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 9588 | func_id_name(func_id), func_id); |
435faee1 DB |
9589 | return err; |
9590 | } | |
9591 | ||
9bb00b28 YS |
9592 | if (env->cur_state->active_rcu_lock) { |
9593 | if (fn->might_sleep) { | |
9594 | verbose(env, "sleepable helper %s#%d in rcu_read_lock region\n", | |
9595 | func_id_name(func_id), func_id); | |
9596 | return -EINVAL; | |
9597 | } | |
9598 | ||
9599 | if (env->prog->aux->sleepable && is_storage_get_function(func_id)) | |
9600 | env->insn_aux_data[insn_idx].storage_get_func_atomic = true; | |
9601 | } | |
9602 | ||
d83525ca | 9603 | meta.func_id = func_id; |
17a52670 | 9604 | /* check args */ |
523a4cf4 | 9605 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { |
1d18feb2 | 9606 | err = check_func_arg(env, i, &meta, fn, insn_idx); |
a7658e1a AS |
9607 | if (err) |
9608 | return err; | |
9609 | } | |
17a52670 | 9610 | |
c93552c4 DB |
9611 | err = record_func_map(env, &meta, func_id, insn_idx); |
9612 | if (err) | |
9613 | return err; | |
9614 | ||
d2e4c1e6 DB |
9615 | err = record_func_key(env, &meta, func_id, insn_idx); |
9616 | if (err) | |
9617 | return err; | |
9618 | ||
435faee1 DB |
9619 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
9620 | * is inferred from register state. | |
9621 | */ | |
9622 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 | 9623 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
1f9a1ea8 | 9624 | BPF_WRITE, -1, false, false); |
435faee1 DB |
9625 | if (err) |
9626 | return err; | |
9627 | } | |
9628 | ||
8f14852e KKD |
9629 | regs = cur_regs(env); |
9630 | ||
9631 | if (meta.release_regno) { | |
9632 | err = -EINVAL; | |
27060531 KKD |
9633 | /* This can only be set for PTR_TO_STACK, as CONST_PTR_TO_DYNPTR cannot |
9634 | * be released by any dynptr helper. Hence, unmark_stack_slots_dynptr | |
9635 | * is safe to do directly. | |
9636 | */ | |
9637 | if (arg_type_is_dynptr(fn->arg_type[meta.release_regno - BPF_REG_1])) { | |
9638 | if (regs[meta.release_regno].type == CONST_PTR_TO_DYNPTR) { | |
9639 | verbose(env, "verifier internal error: CONST_PTR_TO_DYNPTR cannot be released\n"); | |
9640 | return -EFAULT; | |
9641 | } | |
97e03f52 | 9642 | err = unmark_stack_slots_dynptr(env, ®s[meta.release_regno]); |
27060531 | 9643 | } else if (meta.ref_obj_id) { |
8f14852e | 9644 | err = release_reference(env, meta.ref_obj_id); |
27060531 KKD |
9645 | } else if (register_is_null(®s[meta.release_regno])) { |
9646 | /* meta.ref_obj_id can only be 0 if register that is meant to be | |
9647 | * released is NULL, which must be > R0. | |
9648 | */ | |
8f14852e | 9649 | err = 0; |
27060531 | 9650 | } |
46f8bc92 MKL |
9651 | if (err) { |
9652 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
9653 | func_id_name(func_id), func_id); | |
fd978bf7 | 9654 | return err; |
46f8bc92 | 9655 | } |
fd978bf7 JS |
9656 | } |
9657 | ||
e6f2dd0f JK |
9658 | switch (func_id) { |
9659 | case BPF_FUNC_tail_call: | |
9660 | err = check_reference_leak(env); | |
9661 | if (err) { | |
9662 | verbose(env, "tail_call would lead to reference leak\n"); | |
9663 | return err; | |
9664 | } | |
9665 | break; | |
9666 | case BPF_FUNC_get_local_storage: | |
9667 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
9668 | * this is required because get_local_storage() can't return an error. | |
9669 | */ | |
9670 | if (!register_is_null(®s[BPF_REG_2])) { | |
9671 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
9672 | return -EINVAL; | |
9673 | } | |
9674 | break; | |
9675 | case BPF_FUNC_for_each_map_elem: | |
69c087ba YS |
9676 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
9677 | set_map_elem_callback_state); | |
e6f2dd0f JK |
9678 | break; |
9679 | case BPF_FUNC_timer_set_callback: | |
b00628b1 AS |
9680 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
9681 | set_timer_callback_state); | |
e6f2dd0f JK |
9682 | break; |
9683 | case BPF_FUNC_find_vma: | |
7c7e3d31 SL |
9684 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
9685 | set_find_vma_callback_state); | |
e6f2dd0f JK |
9686 | break; |
9687 | case BPF_FUNC_snprintf: | |
7b15523a | 9688 | err = check_bpf_snprintf_call(env, regs); |
e6f2dd0f JK |
9689 | break; |
9690 | case BPF_FUNC_loop: | |
1ade2371 | 9691 | update_loop_inline_state(env, meta.subprogno); |
e6f2dd0f JK |
9692 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
9693 | set_loop_callback_state); | |
9694 | break; | |
263ae152 JK |
9695 | case BPF_FUNC_dynptr_from_mem: |
9696 | if (regs[BPF_REG_1].type != PTR_TO_MAP_VALUE) { | |
9697 | verbose(env, "Unsupported reg type %s for bpf_dynptr_from_mem data\n", | |
9698 | reg_type_str(env, regs[BPF_REG_1].type)); | |
9699 | return -EACCES; | |
9700 | } | |
69fd337a SF |
9701 | break; |
9702 | case BPF_FUNC_set_retval: | |
aef9d4a3 SF |
9703 | if (prog_type == BPF_PROG_TYPE_LSM && |
9704 | env->prog->expected_attach_type == BPF_LSM_CGROUP) { | |
69fd337a SF |
9705 | if (!env->prog->aux->attach_func_proto->type) { |
9706 | /* Make sure programs that attach to void | |
9707 | * hooks don't try to modify return value. | |
9708 | */ | |
9709 | verbose(env, "BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
9710 | return -EINVAL; | |
9711 | } | |
9712 | } | |
9713 | break; | |
88374342 | 9714 | case BPF_FUNC_dynptr_data: |
485ec51e JK |
9715 | { |
9716 | struct bpf_reg_state *reg; | |
9717 | int id, ref_obj_id; | |
20571567 | 9718 | |
485ec51e JK |
9719 | reg = get_dynptr_arg_reg(env, fn, regs); |
9720 | if (!reg) | |
9721 | return -EFAULT; | |
f8064ab9 | 9722 | |
f8064ab9 | 9723 | |
485ec51e JK |
9724 | if (meta.dynptr_id) { |
9725 | verbose(env, "verifier internal error: meta.dynptr_id already set\n"); | |
9726 | return -EFAULT; | |
88374342 | 9727 | } |
485ec51e JK |
9728 | if (meta.ref_obj_id) { |
9729 | verbose(env, "verifier internal error: meta.ref_obj_id already set\n"); | |
88374342 JK |
9730 | return -EFAULT; |
9731 | } | |
485ec51e JK |
9732 | |
9733 | id = dynptr_id(env, reg); | |
9734 | if (id < 0) { | |
9735 | verbose(env, "verifier internal error: failed to obtain dynptr id\n"); | |
9736 | return id; | |
9737 | } | |
9738 | ||
9739 | ref_obj_id = dynptr_ref_obj_id(env, reg); | |
9740 | if (ref_obj_id < 0) { | |
9741 | verbose(env, "verifier internal error: failed to obtain dynptr ref_obj_id\n"); | |
9742 | return ref_obj_id; | |
9743 | } | |
9744 | ||
9745 | meta.dynptr_id = id; | |
9746 | meta.ref_obj_id = ref_obj_id; | |
9747 | ||
88374342 | 9748 | break; |
485ec51e | 9749 | } |
b5964b96 JK |
9750 | case BPF_FUNC_dynptr_write: |
9751 | { | |
9752 | enum bpf_dynptr_type dynptr_type; | |
9753 | struct bpf_reg_state *reg; | |
9754 | ||
9755 | reg = get_dynptr_arg_reg(env, fn, regs); | |
9756 | if (!reg) | |
9757 | return -EFAULT; | |
9758 | ||
9759 | dynptr_type = dynptr_get_type(env, reg); | |
9760 | if (dynptr_type == BPF_DYNPTR_TYPE_INVALID) | |
9761 | return -EFAULT; | |
9762 | ||
9763 | if (dynptr_type == BPF_DYNPTR_TYPE_SKB) | |
9764 | /* this will trigger clear_all_pkt_pointers(), which will | |
9765 | * invalidate all dynptr slices associated with the skb | |
9766 | */ | |
9767 | changes_data = true; | |
9768 | ||
9769 | break; | |
9770 | } | |
20571567 DV |
9771 | case BPF_FUNC_user_ringbuf_drain: |
9772 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, | |
9773 | set_user_ringbuf_callback_state); | |
9774 | break; | |
7b15523a FR |
9775 | } |
9776 | ||
e6f2dd0f JK |
9777 | if (err) |
9778 | return err; | |
9779 | ||
17a52670 | 9780 | /* reset caller saved regs */ |
dc503a8a | 9781 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 9782 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
9783 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
9784 | } | |
17a52670 | 9785 | |
5327ed3d JW |
9786 | /* helper call returns 64-bit value. */ |
9787 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
9788 | ||
dc503a8a | 9789 | /* update return register (already marked as written above) */ |
3c480732 | 9790 | ret_type = fn->ret_type; |
0c9a7a7e JK |
9791 | ret_flag = type_flag(ret_type); |
9792 | ||
9793 | switch (base_type(ret_type)) { | |
9794 | case RET_INTEGER: | |
f1174f77 | 9795 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 9796 | mark_reg_unknown(env, regs, BPF_REG_0); |
0c9a7a7e JK |
9797 | break; |
9798 | case RET_VOID: | |
17a52670 | 9799 | regs[BPF_REG_0].type = NOT_INIT; |
0c9a7a7e JK |
9800 | break; |
9801 | case RET_PTR_TO_MAP_VALUE: | |
f1174f77 | 9802 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 9803 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
9804 | /* remember map_ptr, so that check_map_access() |
9805 | * can check 'value_size' boundary of memory access | |
9806 | * to map element returned from bpf_map_lookup_elem() | |
9807 | */ | |
33ff9823 | 9808 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
9809 | verbose(env, |
9810 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
9811 | return -EINVAL; |
9812 | } | |
33ff9823 | 9813 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
3e8ce298 | 9814 | regs[BPF_REG_0].map_uid = meta.map_uid; |
c25b2ae1 HL |
9815 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE | ret_flag; |
9816 | if (!type_may_be_null(ret_type) && | |
db559117 | 9817 | btf_record_has_field(meta.map_ptr->record, BPF_SPIN_LOCK)) { |
c25b2ae1 | 9818 | regs[BPF_REG_0].id = ++env->id_gen; |
4d31f301 | 9819 | } |
0c9a7a7e JK |
9820 | break; |
9821 | case RET_PTR_TO_SOCKET: | |
c64b7983 | 9822 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 9823 | regs[BPF_REG_0].type = PTR_TO_SOCKET | ret_flag; |
0c9a7a7e JK |
9824 | break; |
9825 | case RET_PTR_TO_SOCK_COMMON: | |
85a51f8c | 9826 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 9827 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON | ret_flag; |
0c9a7a7e JK |
9828 | break; |
9829 | case RET_PTR_TO_TCP_SOCK: | |
655a51e5 | 9830 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 9831 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK | ret_flag; |
0c9a7a7e | 9832 | break; |
2de2669b | 9833 | case RET_PTR_TO_MEM: |
457f4436 | 9834 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 9835 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
457f4436 | 9836 | regs[BPF_REG_0].mem_size = meta.mem_size; |
0c9a7a7e JK |
9837 | break; |
9838 | case RET_PTR_TO_MEM_OR_BTF_ID: | |
9839 | { | |
eaa6bcb7 HL |
9840 | const struct btf_type *t; |
9841 | ||
9842 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
22dc4a0f | 9843 | t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL); |
eaa6bcb7 HL |
9844 | if (!btf_type_is_struct(t)) { |
9845 | u32 tsize; | |
9846 | const struct btf_type *ret; | |
9847 | const char *tname; | |
9848 | ||
9849 | /* resolve the type size of ksym. */ | |
22dc4a0f | 9850 | ret = btf_resolve_size(meta.ret_btf, t, &tsize); |
eaa6bcb7 | 9851 | if (IS_ERR(ret)) { |
22dc4a0f | 9852 | tname = btf_name_by_offset(meta.ret_btf, t->name_off); |
eaa6bcb7 HL |
9853 | verbose(env, "unable to resolve the size of type '%s': %ld\n", |
9854 | tname, PTR_ERR(ret)); | |
9855 | return -EINVAL; | |
9856 | } | |
c25b2ae1 | 9857 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
eaa6bcb7 HL |
9858 | regs[BPF_REG_0].mem_size = tsize; |
9859 | } else { | |
34d3a78c HL |
9860 | /* MEM_RDONLY may be carried from ret_flag, but it |
9861 | * doesn't apply on PTR_TO_BTF_ID. Fold it, otherwise | |
9862 | * it will confuse the check of PTR_TO_BTF_ID in | |
9863 | * check_mem_access(). | |
9864 | */ | |
9865 | ret_flag &= ~MEM_RDONLY; | |
9866 | ||
c25b2ae1 | 9867 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; |
22dc4a0f | 9868 | regs[BPF_REG_0].btf = meta.ret_btf; |
eaa6bcb7 HL |
9869 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; |
9870 | } | |
0c9a7a7e JK |
9871 | break; |
9872 | } | |
9873 | case RET_PTR_TO_BTF_ID: | |
9874 | { | |
c0a5a21c | 9875 | struct btf *ret_btf; |
af7ec138 YS |
9876 | int ret_btf_id; |
9877 | ||
9878 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
c25b2ae1 | 9879 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; |
c0a5a21c | 9880 | if (func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac KKD |
9881 | ret_btf = meta.kptr_field->kptr.btf; |
9882 | ret_btf_id = meta.kptr_field->kptr.btf_id; | |
738c96d5 DM |
9883 | if (!btf_is_kernel(ret_btf)) |
9884 | regs[BPF_REG_0].type |= MEM_ALLOC; | |
c0a5a21c | 9885 | } else { |
47e34cb7 DM |
9886 | if (fn->ret_btf_id == BPF_PTR_POISON) { |
9887 | verbose(env, "verifier internal error:"); | |
9888 | verbose(env, "func %s has non-overwritten BPF_PTR_POISON return type\n", | |
9889 | func_id_name(func_id)); | |
9890 | return -EINVAL; | |
9891 | } | |
c0a5a21c KKD |
9892 | ret_btf = btf_vmlinux; |
9893 | ret_btf_id = *fn->ret_btf_id; | |
9894 | } | |
af7ec138 | 9895 | if (ret_btf_id == 0) { |
3c480732 HL |
9896 | verbose(env, "invalid return type %u of func %s#%d\n", |
9897 | base_type(ret_type), func_id_name(func_id), | |
9898 | func_id); | |
af7ec138 YS |
9899 | return -EINVAL; |
9900 | } | |
c0a5a21c | 9901 | regs[BPF_REG_0].btf = ret_btf; |
af7ec138 | 9902 | regs[BPF_REG_0].btf_id = ret_btf_id; |
0c9a7a7e JK |
9903 | break; |
9904 | } | |
9905 | default: | |
3c480732 HL |
9906 | verbose(env, "unknown return type %u of func %s#%d\n", |
9907 | base_type(ret_type), func_id_name(func_id), func_id); | |
17a52670 AS |
9908 | return -EINVAL; |
9909 | } | |
04fd61ab | 9910 | |
c25b2ae1 | 9911 | if (type_may_be_null(regs[BPF_REG_0].type)) |
93c230e3 MKL |
9912 | regs[BPF_REG_0].id = ++env->id_gen; |
9913 | ||
b2d8ef19 DM |
9914 | if (helper_multiple_ref_obj_use(func_id, meta.map_ptr)) { |
9915 | verbose(env, "verifier internal error: func %s#%d sets ref_obj_id more than once\n", | |
9916 | func_id_name(func_id), func_id); | |
9917 | return -EFAULT; | |
9918 | } | |
9919 | ||
f8064ab9 KKD |
9920 | if (is_dynptr_ref_function(func_id)) |
9921 | regs[BPF_REG_0].dynptr_id = meta.dynptr_id; | |
9922 | ||
88374342 | 9923 | if (is_ptr_cast_function(func_id) || is_dynptr_ref_function(func_id)) { |
1b986589 MKL |
9924 | /* For release_reference() */ |
9925 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 9926 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
9927 | int id = acquire_reference_state(env, insn_idx); |
9928 | ||
9929 | if (id < 0) | |
9930 | return id; | |
9931 | /* For mark_ptr_or_null_reg() */ | |
9932 | regs[BPF_REG_0].id = id; | |
9933 | /* For release_reference() */ | |
9934 | regs[BPF_REG_0].ref_obj_id = id; | |
9935 | } | |
1b986589 | 9936 | |
849fa506 YS |
9937 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
9938 | ||
61bd5218 | 9939 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
9940 | if (err) |
9941 | return err; | |
04fd61ab | 9942 | |
fa28dcb8 SL |
9943 | if ((func_id == BPF_FUNC_get_stack || |
9944 | func_id == BPF_FUNC_get_task_stack) && | |
9945 | !env->prog->has_callchain_buf) { | |
c195651e YS |
9946 | const char *err_str; |
9947 | ||
9948 | #ifdef CONFIG_PERF_EVENTS | |
9949 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
9950 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
9951 | #else | |
9952 | err = -ENOTSUPP; | |
9953 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
9954 | #endif | |
9955 | if (err) { | |
9956 | verbose(env, err_str, func_id_name(func_id), func_id); | |
9957 | return err; | |
9958 | } | |
9959 | ||
9960 | env->prog->has_callchain_buf = true; | |
9961 | } | |
9962 | ||
5d99cb2c SL |
9963 | if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack) |
9964 | env->prog->call_get_stack = true; | |
9965 | ||
9b99edca JO |
9966 | if (func_id == BPF_FUNC_get_func_ip) { |
9967 | if (check_get_func_ip(env)) | |
9968 | return -ENOTSUPP; | |
9969 | env->prog->call_get_func_ip = true; | |
9970 | } | |
9971 | ||
969bf05e AS |
9972 | if (changes_data) |
9973 | clear_all_pkt_pointers(env); | |
9974 | return 0; | |
9975 | } | |
9976 | ||
e6ac2450 MKL |
9977 | /* mark_btf_func_reg_size() is used when the reg size is determined by |
9978 | * the BTF func_proto's return value size and argument. | |
9979 | */ | |
9980 | static void mark_btf_func_reg_size(struct bpf_verifier_env *env, u32 regno, | |
9981 | size_t reg_size) | |
9982 | { | |
9983 | struct bpf_reg_state *reg = &cur_regs(env)[regno]; | |
9984 | ||
9985 | if (regno == BPF_REG_0) { | |
9986 | /* Function return value */ | |
9987 | reg->live |= REG_LIVE_WRITTEN; | |
9988 | reg->subreg_def = reg_size == sizeof(u64) ? | |
9989 | DEF_NOT_SUBREG : env->insn_idx + 1; | |
9990 | } else { | |
9991 | /* Function argument */ | |
9992 | if (reg_size == sizeof(u64)) { | |
9993 | mark_insn_zext(env, reg); | |
9994 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
9995 | } else { | |
9996 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ32); | |
9997 | } | |
9998 | } | |
9999 | } | |
10000 | ||
00b85860 KKD |
10001 | static bool is_kfunc_acquire(struct bpf_kfunc_call_arg_meta *meta) |
10002 | { | |
10003 | return meta->kfunc_flags & KF_ACQUIRE; | |
10004 | } | |
a5d82727 | 10005 | |
00b85860 KKD |
10006 | static bool is_kfunc_release(struct bpf_kfunc_call_arg_meta *meta) |
10007 | { | |
10008 | return meta->kfunc_flags & KF_RELEASE; | |
10009 | } | |
e6ac2450 | 10010 | |
00b85860 KKD |
10011 | static bool is_kfunc_trusted_args(struct bpf_kfunc_call_arg_meta *meta) |
10012 | { | |
6c831c46 | 10013 | return (meta->kfunc_flags & KF_TRUSTED_ARGS) || is_kfunc_release(meta); |
00b85860 | 10014 | } |
4dd48c6f | 10015 | |
00b85860 KKD |
10016 | static bool is_kfunc_sleepable(struct bpf_kfunc_call_arg_meta *meta) |
10017 | { | |
10018 | return meta->kfunc_flags & KF_SLEEPABLE; | |
10019 | } | |
5c073f26 | 10020 | |
00b85860 KKD |
10021 | static bool is_kfunc_destructive(struct bpf_kfunc_call_arg_meta *meta) |
10022 | { | |
10023 | return meta->kfunc_flags & KF_DESTRUCTIVE; | |
10024 | } | |
eb1f7f71 | 10025 | |
fca1aa75 YS |
10026 | static bool is_kfunc_rcu(struct bpf_kfunc_call_arg_meta *meta) |
10027 | { | |
10028 | return meta->kfunc_flags & KF_RCU; | |
10029 | } | |
10030 | ||
a50388db KKD |
10031 | static bool __kfunc_param_match_suffix(const struct btf *btf, |
10032 | const struct btf_param *arg, | |
10033 | const char *suffix) | |
00b85860 | 10034 | { |
a50388db | 10035 | int suffix_len = strlen(suffix), len; |
00b85860 | 10036 | const char *param_name; |
e6ac2450 | 10037 | |
00b85860 KKD |
10038 | /* In the future, this can be ported to use BTF tagging */ |
10039 | param_name = btf_name_by_offset(btf, arg->name_off); | |
10040 | if (str_is_empty(param_name)) | |
10041 | return false; | |
10042 | len = strlen(param_name); | |
a50388db | 10043 | if (len < suffix_len) |
00b85860 | 10044 | return false; |
a50388db KKD |
10045 | param_name += len - suffix_len; |
10046 | return !strncmp(param_name, suffix, suffix_len); | |
10047 | } | |
5c073f26 | 10048 | |
a50388db KKD |
10049 | static bool is_kfunc_arg_mem_size(const struct btf *btf, |
10050 | const struct btf_param *arg, | |
10051 | const struct bpf_reg_state *reg) | |
10052 | { | |
10053 | const struct btf_type *t; | |
5c073f26 | 10054 | |
a50388db KKD |
10055 | t = btf_type_skip_modifiers(btf, arg->type, NULL); |
10056 | if (!btf_type_is_scalar(t) || reg->type != SCALAR_VALUE) | |
00b85860 | 10057 | return false; |
eb1f7f71 | 10058 | |
a50388db KKD |
10059 | return __kfunc_param_match_suffix(btf, arg, "__sz"); |
10060 | } | |
eb1f7f71 | 10061 | |
66e3a13e JK |
10062 | static bool is_kfunc_arg_const_mem_size(const struct btf *btf, |
10063 | const struct btf_param *arg, | |
10064 | const struct bpf_reg_state *reg) | |
10065 | { | |
10066 | const struct btf_type *t; | |
10067 | ||
10068 | t = btf_type_skip_modifiers(btf, arg->type, NULL); | |
10069 | if (!btf_type_is_scalar(t) || reg->type != SCALAR_VALUE) | |
10070 | return false; | |
10071 | ||
10072 | return __kfunc_param_match_suffix(btf, arg, "__szk"); | |
10073 | } | |
10074 | ||
3bda08b6 DR |
10075 | static bool is_kfunc_arg_optional(const struct btf *btf, const struct btf_param *arg) |
10076 | { | |
10077 | return __kfunc_param_match_suffix(btf, arg, "__opt"); | |
10078 | } | |
10079 | ||
a50388db KKD |
10080 | static bool is_kfunc_arg_constant(const struct btf *btf, const struct btf_param *arg) |
10081 | { | |
10082 | return __kfunc_param_match_suffix(btf, arg, "__k"); | |
00b85860 | 10083 | } |
eb1f7f71 | 10084 | |
958cf2e2 KKD |
10085 | static bool is_kfunc_arg_ignore(const struct btf *btf, const struct btf_param *arg) |
10086 | { | |
10087 | return __kfunc_param_match_suffix(btf, arg, "__ign"); | |
10088 | } | |
5c073f26 | 10089 | |
ac9f0605 KKD |
10090 | static bool is_kfunc_arg_alloc_obj(const struct btf *btf, const struct btf_param *arg) |
10091 | { | |
10092 | return __kfunc_param_match_suffix(btf, arg, "__alloc"); | |
10093 | } | |
e6ac2450 | 10094 | |
d96d937d JK |
10095 | static bool is_kfunc_arg_uninit(const struct btf *btf, const struct btf_param *arg) |
10096 | { | |
10097 | return __kfunc_param_match_suffix(btf, arg, "__uninit"); | |
10098 | } | |
10099 | ||
7c50b1cb DM |
10100 | static bool is_kfunc_arg_refcounted_kptr(const struct btf *btf, const struct btf_param *arg) |
10101 | { | |
10102 | return __kfunc_param_match_suffix(btf, arg, "__refcounted_kptr"); | |
10103 | } | |
10104 | ||
00b85860 KKD |
10105 | static bool is_kfunc_arg_scalar_with_name(const struct btf *btf, |
10106 | const struct btf_param *arg, | |
10107 | const char *name) | |
10108 | { | |
10109 | int len, target_len = strlen(name); | |
10110 | const char *param_name; | |
e6ac2450 | 10111 | |
00b85860 KKD |
10112 | param_name = btf_name_by_offset(btf, arg->name_off); |
10113 | if (str_is_empty(param_name)) | |
10114 | return false; | |
10115 | len = strlen(param_name); | |
10116 | if (len != target_len) | |
10117 | return false; | |
10118 | if (strcmp(param_name, name)) | |
10119 | return false; | |
e6ac2450 | 10120 | |
00b85860 | 10121 | return true; |
e6ac2450 MKL |
10122 | } |
10123 | ||
00b85860 KKD |
10124 | enum { |
10125 | KF_ARG_DYNPTR_ID, | |
8cab76ec KKD |
10126 | KF_ARG_LIST_HEAD_ID, |
10127 | KF_ARG_LIST_NODE_ID, | |
cd6791b4 DM |
10128 | KF_ARG_RB_ROOT_ID, |
10129 | KF_ARG_RB_NODE_ID, | |
00b85860 | 10130 | }; |
b03c9f9f | 10131 | |
00b85860 KKD |
10132 | BTF_ID_LIST(kf_arg_btf_ids) |
10133 | BTF_ID(struct, bpf_dynptr_kern) | |
8cab76ec KKD |
10134 | BTF_ID(struct, bpf_list_head) |
10135 | BTF_ID(struct, bpf_list_node) | |
bd1279ae DM |
10136 | BTF_ID(struct, bpf_rb_root) |
10137 | BTF_ID(struct, bpf_rb_node) | |
b03c9f9f | 10138 | |
8cab76ec KKD |
10139 | static bool __is_kfunc_ptr_arg_type(const struct btf *btf, |
10140 | const struct btf_param *arg, int type) | |
3f50f132 | 10141 | { |
00b85860 KKD |
10142 | const struct btf_type *t; |
10143 | u32 res_id; | |
3f50f132 | 10144 | |
00b85860 KKD |
10145 | t = btf_type_skip_modifiers(btf, arg->type, NULL); |
10146 | if (!t) | |
10147 | return false; | |
10148 | if (!btf_type_is_ptr(t)) | |
10149 | return false; | |
10150 | t = btf_type_skip_modifiers(btf, t->type, &res_id); | |
10151 | if (!t) | |
10152 | return false; | |
8cab76ec | 10153 | return btf_types_are_same(btf, res_id, btf_vmlinux, kf_arg_btf_ids[type]); |
3f50f132 JF |
10154 | } |
10155 | ||
8cab76ec | 10156 | static bool is_kfunc_arg_dynptr(const struct btf *btf, const struct btf_param *arg) |
b03c9f9f | 10157 | { |
8cab76ec | 10158 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_DYNPTR_ID); |
969bf05e AS |
10159 | } |
10160 | ||
8cab76ec | 10161 | static bool is_kfunc_arg_list_head(const struct btf *btf, const struct btf_param *arg) |
3f50f132 | 10162 | { |
8cab76ec | 10163 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_HEAD_ID); |
3f50f132 JF |
10164 | } |
10165 | ||
8cab76ec | 10166 | static bool is_kfunc_arg_list_node(const struct btf *btf, const struct btf_param *arg) |
bb7f0f98 | 10167 | { |
8cab76ec | 10168 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_NODE_ID); |
00b85860 KKD |
10169 | } |
10170 | ||
cd6791b4 DM |
10171 | static bool is_kfunc_arg_rbtree_root(const struct btf *btf, const struct btf_param *arg) |
10172 | { | |
10173 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_RB_ROOT_ID); | |
10174 | } | |
10175 | ||
10176 | static bool is_kfunc_arg_rbtree_node(const struct btf *btf, const struct btf_param *arg) | |
10177 | { | |
10178 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_RB_NODE_ID); | |
10179 | } | |
10180 | ||
5d92ddc3 DM |
10181 | static bool is_kfunc_arg_callback(struct bpf_verifier_env *env, const struct btf *btf, |
10182 | const struct btf_param *arg) | |
10183 | { | |
10184 | const struct btf_type *t; | |
10185 | ||
10186 | t = btf_type_resolve_func_ptr(btf, arg->type, NULL); | |
10187 | if (!t) | |
10188 | return false; | |
10189 | ||
10190 | return true; | |
10191 | } | |
10192 | ||
00b85860 KKD |
10193 | /* Returns true if struct is composed of scalars, 4 levels of nesting allowed */ |
10194 | static bool __btf_type_is_scalar_struct(struct bpf_verifier_env *env, | |
10195 | const struct btf *btf, | |
10196 | const struct btf_type *t, int rec) | |
10197 | { | |
10198 | const struct btf_type *member_type; | |
10199 | const struct btf_member *member; | |
10200 | u32 i; | |
10201 | ||
10202 | if (!btf_type_is_struct(t)) | |
10203 | return false; | |
10204 | ||
10205 | for_each_member(i, t, member) { | |
10206 | const struct btf_array *array; | |
10207 | ||
10208 | member_type = btf_type_skip_modifiers(btf, member->type, NULL); | |
10209 | if (btf_type_is_struct(member_type)) { | |
10210 | if (rec >= 3) { | |
10211 | verbose(env, "max struct nesting depth exceeded\n"); | |
10212 | return false; | |
10213 | } | |
10214 | if (!__btf_type_is_scalar_struct(env, btf, member_type, rec + 1)) | |
10215 | return false; | |
10216 | continue; | |
10217 | } | |
10218 | if (btf_type_is_array(member_type)) { | |
10219 | array = btf_array(member_type); | |
10220 | if (!array->nelems) | |
10221 | return false; | |
10222 | member_type = btf_type_skip_modifiers(btf, array->type, NULL); | |
10223 | if (!btf_type_is_scalar(member_type)) | |
10224 | return false; | |
10225 | continue; | |
10226 | } | |
10227 | if (!btf_type_is_scalar(member_type)) | |
10228 | return false; | |
10229 | } | |
10230 | return true; | |
10231 | } | |
10232 | ||
00b85860 KKD |
10233 | enum kfunc_ptr_arg_type { |
10234 | KF_ARG_PTR_TO_CTX, | |
7c50b1cb DM |
10235 | KF_ARG_PTR_TO_ALLOC_BTF_ID, /* Allocated object */ |
10236 | KF_ARG_PTR_TO_REFCOUNTED_KPTR, /* Refcounted local kptr */ | |
00b85860 | 10237 | KF_ARG_PTR_TO_DYNPTR, |
06accc87 | 10238 | KF_ARG_PTR_TO_ITER, |
8cab76ec KKD |
10239 | KF_ARG_PTR_TO_LIST_HEAD, |
10240 | KF_ARG_PTR_TO_LIST_NODE, | |
7c50b1cb | 10241 | KF_ARG_PTR_TO_BTF_ID, /* Also covers reg2btf_ids conversions */ |
00b85860 | 10242 | KF_ARG_PTR_TO_MEM, |
7c50b1cb | 10243 | KF_ARG_PTR_TO_MEM_SIZE, /* Size derived from next argument, skip it */ |
5d92ddc3 | 10244 | KF_ARG_PTR_TO_CALLBACK, |
cd6791b4 DM |
10245 | KF_ARG_PTR_TO_RB_ROOT, |
10246 | KF_ARG_PTR_TO_RB_NODE, | |
00b85860 KKD |
10247 | }; |
10248 | ||
ac9f0605 KKD |
10249 | enum special_kfunc_type { |
10250 | KF_bpf_obj_new_impl, | |
10251 | KF_bpf_obj_drop_impl, | |
7c50b1cb | 10252 | KF_bpf_refcount_acquire_impl, |
d2dcc67d DM |
10253 | KF_bpf_list_push_front_impl, |
10254 | KF_bpf_list_push_back_impl, | |
8cab76ec KKD |
10255 | KF_bpf_list_pop_front, |
10256 | KF_bpf_list_pop_back, | |
fd264ca0 | 10257 | KF_bpf_cast_to_kern_ctx, |
a35b9af4 | 10258 | KF_bpf_rdonly_cast, |
9bb00b28 YS |
10259 | KF_bpf_rcu_read_lock, |
10260 | KF_bpf_rcu_read_unlock, | |
bd1279ae | 10261 | KF_bpf_rbtree_remove, |
d2dcc67d | 10262 | KF_bpf_rbtree_add_impl, |
bd1279ae | 10263 | KF_bpf_rbtree_first, |
b5964b96 | 10264 | KF_bpf_dynptr_from_skb, |
05421aec | 10265 | KF_bpf_dynptr_from_xdp, |
66e3a13e JK |
10266 | KF_bpf_dynptr_slice, |
10267 | KF_bpf_dynptr_slice_rdwr, | |
361f129f | 10268 | KF_bpf_dynptr_clone, |
ac9f0605 KKD |
10269 | }; |
10270 | ||
10271 | BTF_SET_START(special_kfunc_set) | |
10272 | BTF_ID(func, bpf_obj_new_impl) | |
10273 | BTF_ID(func, bpf_obj_drop_impl) | |
7c50b1cb | 10274 | BTF_ID(func, bpf_refcount_acquire_impl) |
d2dcc67d DM |
10275 | BTF_ID(func, bpf_list_push_front_impl) |
10276 | BTF_ID(func, bpf_list_push_back_impl) | |
8cab76ec KKD |
10277 | BTF_ID(func, bpf_list_pop_front) |
10278 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 10279 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 10280 | BTF_ID(func, bpf_rdonly_cast) |
bd1279ae | 10281 | BTF_ID(func, bpf_rbtree_remove) |
d2dcc67d | 10282 | BTF_ID(func, bpf_rbtree_add_impl) |
bd1279ae | 10283 | BTF_ID(func, bpf_rbtree_first) |
b5964b96 | 10284 | BTF_ID(func, bpf_dynptr_from_skb) |
05421aec | 10285 | BTF_ID(func, bpf_dynptr_from_xdp) |
66e3a13e JK |
10286 | BTF_ID(func, bpf_dynptr_slice) |
10287 | BTF_ID(func, bpf_dynptr_slice_rdwr) | |
361f129f | 10288 | BTF_ID(func, bpf_dynptr_clone) |
ac9f0605 KKD |
10289 | BTF_SET_END(special_kfunc_set) |
10290 | ||
10291 | BTF_ID_LIST(special_kfunc_list) | |
10292 | BTF_ID(func, bpf_obj_new_impl) | |
10293 | BTF_ID(func, bpf_obj_drop_impl) | |
7c50b1cb | 10294 | BTF_ID(func, bpf_refcount_acquire_impl) |
d2dcc67d DM |
10295 | BTF_ID(func, bpf_list_push_front_impl) |
10296 | BTF_ID(func, bpf_list_push_back_impl) | |
8cab76ec KKD |
10297 | BTF_ID(func, bpf_list_pop_front) |
10298 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 10299 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 10300 | BTF_ID(func, bpf_rdonly_cast) |
9bb00b28 YS |
10301 | BTF_ID(func, bpf_rcu_read_lock) |
10302 | BTF_ID(func, bpf_rcu_read_unlock) | |
bd1279ae | 10303 | BTF_ID(func, bpf_rbtree_remove) |
d2dcc67d | 10304 | BTF_ID(func, bpf_rbtree_add_impl) |
bd1279ae | 10305 | BTF_ID(func, bpf_rbtree_first) |
b5964b96 | 10306 | BTF_ID(func, bpf_dynptr_from_skb) |
05421aec | 10307 | BTF_ID(func, bpf_dynptr_from_xdp) |
66e3a13e JK |
10308 | BTF_ID(func, bpf_dynptr_slice) |
10309 | BTF_ID(func, bpf_dynptr_slice_rdwr) | |
361f129f | 10310 | BTF_ID(func, bpf_dynptr_clone) |
9bb00b28 | 10311 | |
7793fc3b DM |
10312 | static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta) |
10313 | { | |
10314 | if (meta->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl] && | |
10315 | meta->arg_owning_ref) { | |
10316 | return false; | |
10317 | } | |
10318 | ||
10319 | return meta->kfunc_flags & KF_RET_NULL; | |
10320 | } | |
10321 | ||
9bb00b28 YS |
10322 | static bool is_kfunc_bpf_rcu_read_lock(struct bpf_kfunc_call_arg_meta *meta) |
10323 | { | |
10324 | return meta->func_id == special_kfunc_list[KF_bpf_rcu_read_lock]; | |
10325 | } | |
10326 | ||
10327 | static bool is_kfunc_bpf_rcu_read_unlock(struct bpf_kfunc_call_arg_meta *meta) | |
10328 | { | |
10329 | return meta->func_id == special_kfunc_list[KF_bpf_rcu_read_unlock]; | |
10330 | } | |
ac9f0605 | 10331 | |
00b85860 KKD |
10332 | static enum kfunc_ptr_arg_type |
10333 | get_kfunc_ptr_arg_type(struct bpf_verifier_env *env, | |
10334 | struct bpf_kfunc_call_arg_meta *meta, | |
10335 | const struct btf_type *t, const struct btf_type *ref_t, | |
10336 | const char *ref_tname, const struct btf_param *args, | |
10337 | int argno, int nargs) | |
10338 | { | |
10339 | u32 regno = argno + 1; | |
10340 | struct bpf_reg_state *regs = cur_regs(env); | |
10341 | struct bpf_reg_state *reg = ®s[regno]; | |
10342 | bool arg_mem_size = false; | |
10343 | ||
fd264ca0 YS |
10344 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) |
10345 | return KF_ARG_PTR_TO_CTX; | |
10346 | ||
00b85860 KKD |
10347 | /* In this function, we verify the kfunc's BTF as per the argument type, |
10348 | * leaving the rest of the verification with respect to the register | |
10349 | * type to our caller. When a set of conditions hold in the BTF type of | |
10350 | * arguments, we resolve it to a known kfunc_ptr_arg_type. | |
10351 | */ | |
10352 | if (btf_get_prog_ctx_type(&env->log, meta->btf, t, resolve_prog_type(env->prog), argno)) | |
10353 | return KF_ARG_PTR_TO_CTX; | |
10354 | ||
ac9f0605 KKD |
10355 | if (is_kfunc_arg_alloc_obj(meta->btf, &args[argno])) |
10356 | return KF_ARG_PTR_TO_ALLOC_BTF_ID; | |
10357 | ||
7c50b1cb DM |
10358 | if (is_kfunc_arg_refcounted_kptr(meta->btf, &args[argno])) |
10359 | return KF_ARG_PTR_TO_REFCOUNTED_KPTR; | |
00b85860 KKD |
10360 | |
10361 | if (is_kfunc_arg_dynptr(meta->btf, &args[argno])) | |
10362 | return KF_ARG_PTR_TO_DYNPTR; | |
10363 | ||
06accc87 AN |
10364 | if (is_kfunc_arg_iter(meta, argno)) |
10365 | return KF_ARG_PTR_TO_ITER; | |
10366 | ||
8cab76ec KKD |
10367 | if (is_kfunc_arg_list_head(meta->btf, &args[argno])) |
10368 | return KF_ARG_PTR_TO_LIST_HEAD; | |
10369 | ||
10370 | if (is_kfunc_arg_list_node(meta->btf, &args[argno])) | |
10371 | return KF_ARG_PTR_TO_LIST_NODE; | |
10372 | ||
cd6791b4 DM |
10373 | if (is_kfunc_arg_rbtree_root(meta->btf, &args[argno])) |
10374 | return KF_ARG_PTR_TO_RB_ROOT; | |
10375 | ||
10376 | if (is_kfunc_arg_rbtree_node(meta->btf, &args[argno])) | |
10377 | return KF_ARG_PTR_TO_RB_NODE; | |
10378 | ||
00b85860 KKD |
10379 | if ((base_type(reg->type) == PTR_TO_BTF_ID || reg2btf_ids[base_type(reg->type)])) { |
10380 | if (!btf_type_is_struct(ref_t)) { | |
10381 | verbose(env, "kernel function %s args#%d pointer type %s %s is not supported\n", | |
10382 | meta->func_name, argno, btf_type_str(ref_t), ref_tname); | |
10383 | return -EINVAL; | |
10384 | } | |
10385 | return KF_ARG_PTR_TO_BTF_ID; | |
10386 | } | |
10387 | ||
5d92ddc3 DM |
10388 | if (is_kfunc_arg_callback(env, meta->btf, &args[argno])) |
10389 | return KF_ARG_PTR_TO_CALLBACK; | |
10390 | ||
66e3a13e JK |
10391 | |
10392 | if (argno + 1 < nargs && | |
10393 | (is_kfunc_arg_mem_size(meta->btf, &args[argno + 1], ®s[regno + 1]) || | |
10394 | is_kfunc_arg_const_mem_size(meta->btf, &args[argno + 1], ®s[regno + 1]))) | |
00b85860 KKD |
10395 | arg_mem_size = true; |
10396 | ||
10397 | /* This is the catch all argument type of register types supported by | |
10398 | * check_helper_mem_access. However, we only allow when argument type is | |
10399 | * pointer to scalar, or struct composed (recursively) of scalars. When | |
10400 | * arg_mem_size is true, the pointer can be void *. | |
10401 | */ | |
10402 | if (!btf_type_is_scalar(ref_t) && !__btf_type_is_scalar_struct(env, meta->btf, ref_t, 0) && | |
10403 | (arg_mem_size ? !btf_type_is_void(ref_t) : 1)) { | |
10404 | verbose(env, "arg#%d pointer type %s %s must point to %sscalar, or struct with scalar\n", | |
10405 | argno, btf_type_str(ref_t), ref_tname, arg_mem_size ? "void, " : ""); | |
10406 | return -EINVAL; | |
10407 | } | |
10408 | return arg_mem_size ? KF_ARG_PTR_TO_MEM_SIZE : KF_ARG_PTR_TO_MEM; | |
10409 | } | |
10410 | ||
10411 | static int process_kf_arg_ptr_to_btf_id(struct bpf_verifier_env *env, | |
10412 | struct bpf_reg_state *reg, | |
10413 | const struct btf_type *ref_t, | |
10414 | const char *ref_tname, u32 ref_id, | |
10415 | struct bpf_kfunc_call_arg_meta *meta, | |
10416 | int argno) | |
10417 | { | |
10418 | const struct btf_type *reg_ref_t; | |
10419 | bool strict_type_match = false; | |
10420 | const struct btf *reg_btf; | |
10421 | const char *reg_ref_tname; | |
10422 | u32 reg_ref_id; | |
10423 | ||
3f00c523 | 10424 | if (base_type(reg->type) == PTR_TO_BTF_ID) { |
00b85860 KKD |
10425 | reg_btf = reg->btf; |
10426 | reg_ref_id = reg->btf_id; | |
10427 | } else { | |
10428 | reg_btf = btf_vmlinux; | |
10429 | reg_ref_id = *reg2btf_ids[base_type(reg->type)]; | |
10430 | } | |
10431 | ||
b613d335 DV |
10432 | /* Enforce strict type matching for calls to kfuncs that are acquiring |
10433 | * or releasing a reference, or are no-cast aliases. We do _not_ | |
10434 | * enforce strict matching for plain KF_TRUSTED_ARGS kfuncs by default, | |
10435 | * as we want to enable BPF programs to pass types that are bitwise | |
10436 | * equivalent without forcing them to explicitly cast with something | |
10437 | * like bpf_cast_to_kern_ctx(). | |
10438 | * | |
10439 | * For example, say we had a type like the following: | |
10440 | * | |
10441 | * struct bpf_cpumask { | |
10442 | * cpumask_t cpumask; | |
10443 | * refcount_t usage; | |
10444 | * }; | |
10445 | * | |
10446 | * Note that as specified in <linux/cpumask.h>, cpumask_t is typedef'ed | |
10447 | * to a struct cpumask, so it would be safe to pass a struct | |
10448 | * bpf_cpumask * to a kfunc expecting a struct cpumask *. | |
10449 | * | |
10450 | * The philosophy here is similar to how we allow scalars of different | |
10451 | * types to be passed to kfuncs as long as the size is the same. The | |
10452 | * only difference here is that we're simply allowing | |
10453 | * btf_struct_ids_match() to walk the struct at the 0th offset, and | |
10454 | * resolve types. | |
10455 | */ | |
10456 | if (is_kfunc_acquire(meta) || | |
10457 | (is_kfunc_release(meta) && reg->ref_obj_id) || | |
10458 | btf_type_ids_nocast_alias(&env->log, reg_btf, reg_ref_id, meta->btf, ref_id)) | |
00b85860 KKD |
10459 | strict_type_match = true; |
10460 | ||
b613d335 DV |
10461 | WARN_ON_ONCE(is_kfunc_trusted_args(meta) && reg->off); |
10462 | ||
00b85860 KKD |
10463 | reg_ref_t = btf_type_skip_modifiers(reg_btf, reg_ref_id, ®_ref_id); |
10464 | reg_ref_tname = btf_name_by_offset(reg_btf, reg_ref_t->name_off); | |
10465 | if (!btf_struct_ids_match(&env->log, reg_btf, reg_ref_id, reg->off, meta->btf, ref_id, strict_type_match)) { | |
10466 | verbose(env, "kernel function %s args#%d expected pointer to %s %s but R%d has a pointer to %s %s\n", | |
10467 | meta->func_name, argno, btf_type_str(ref_t), ref_tname, argno + 1, | |
10468 | btf_type_str(reg_ref_t), reg_ref_tname); | |
10469 | return -EINVAL; | |
10470 | } | |
10471 | return 0; | |
10472 | } | |
10473 | ||
6a3cd331 | 10474 | static int ref_set_non_owning(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
534e86bc | 10475 | { |
6a3cd331 | 10476 | struct bpf_verifier_state *state = env->cur_state; |
0816b8c6 | 10477 | struct btf_record *rec = reg_btf_record(reg); |
6a3cd331 DM |
10478 | |
10479 | if (!state->active_lock.ptr) { | |
10480 | verbose(env, "verifier internal error: ref_set_non_owning w/o active lock\n"); | |
10481 | return -EFAULT; | |
10482 | } | |
10483 | ||
10484 | if (type_flag(reg->type) & NON_OWN_REF) { | |
10485 | verbose(env, "verifier internal error: NON_OWN_REF already set\n"); | |
10486 | return -EFAULT; | |
10487 | } | |
10488 | ||
10489 | reg->type |= NON_OWN_REF; | |
0816b8c6 DM |
10490 | if (rec->refcount_off >= 0) |
10491 | reg->type |= MEM_RCU; | |
10492 | ||
6a3cd331 DM |
10493 | return 0; |
10494 | } | |
10495 | ||
10496 | static int ref_convert_owning_non_owning(struct bpf_verifier_env *env, u32 ref_obj_id) | |
10497 | { | |
10498 | struct bpf_func_state *state, *unused; | |
534e86bc KKD |
10499 | struct bpf_reg_state *reg; |
10500 | int i; | |
10501 | ||
6a3cd331 DM |
10502 | state = cur_func(env); |
10503 | ||
534e86bc | 10504 | if (!ref_obj_id) { |
6a3cd331 DM |
10505 | verbose(env, "verifier internal error: ref_obj_id is zero for " |
10506 | "owning -> non-owning conversion\n"); | |
534e86bc KKD |
10507 | return -EFAULT; |
10508 | } | |
6a3cd331 | 10509 | |
534e86bc | 10510 | for (i = 0; i < state->acquired_refs; i++) { |
6a3cd331 DM |
10511 | if (state->refs[i].id != ref_obj_id) |
10512 | continue; | |
10513 | ||
10514 | /* Clear ref_obj_id here so release_reference doesn't clobber | |
10515 | * the whole reg | |
10516 | */ | |
10517 | bpf_for_each_reg_in_vstate(env->cur_state, unused, reg, ({ | |
10518 | if (reg->ref_obj_id == ref_obj_id) { | |
10519 | reg->ref_obj_id = 0; | |
10520 | ref_set_non_owning(env, reg); | |
534e86bc | 10521 | } |
6a3cd331 DM |
10522 | })); |
10523 | return 0; | |
534e86bc | 10524 | } |
6a3cd331 | 10525 | |
534e86bc KKD |
10526 | verbose(env, "verifier internal error: ref state missing for ref_obj_id\n"); |
10527 | return -EFAULT; | |
10528 | } | |
10529 | ||
8cab76ec KKD |
10530 | /* Implementation details: |
10531 | * | |
10532 | * Each register points to some region of memory, which we define as an | |
10533 | * allocation. Each allocation may embed a bpf_spin_lock which protects any | |
10534 | * special BPF objects (bpf_list_head, bpf_rb_root, etc.) part of the same | |
10535 | * allocation. The lock and the data it protects are colocated in the same | |
10536 | * memory region. | |
10537 | * | |
10538 | * Hence, everytime a register holds a pointer value pointing to such | |
10539 | * allocation, the verifier preserves a unique reg->id for it. | |
10540 | * | |
10541 | * The verifier remembers the lock 'ptr' and the lock 'id' whenever | |
10542 | * bpf_spin_lock is called. | |
10543 | * | |
10544 | * To enable this, lock state in the verifier captures two values: | |
10545 | * active_lock.ptr = Register's type specific pointer | |
10546 | * active_lock.id = A unique ID for each register pointer value | |
10547 | * | |
10548 | * Currently, PTR_TO_MAP_VALUE and PTR_TO_BTF_ID | MEM_ALLOC are the two | |
10549 | * supported register types. | |
10550 | * | |
10551 | * The active_lock.ptr in case of map values is the reg->map_ptr, and in case of | |
10552 | * allocated objects is the reg->btf pointer. | |
10553 | * | |
10554 | * The active_lock.id is non-unique for maps supporting direct_value_addr, as we | |
10555 | * can establish the provenance of the map value statically for each distinct | |
10556 | * lookup into such maps. They always contain a single map value hence unique | |
10557 | * IDs for each pseudo load pessimizes the algorithm and rejects valid programs. | |
10558 | * | |
10559 | * So, in case of global variables, they use array maps with max_entries = 1, | |
10560 | * hence their active_lock.ptr becomes map_ptr and id = 0 (since they all point | |
10561 | * into the same map value as max_entries is 1, as described above). | |
10562 | * | |
10563 | * In case of inner map lookups, the inner map pointer has same map_ptr as the | |
10564 | * outer map pointer (in verifier context), but each lookup into an inner map | |
10565 | * assigns a fresh reg->id to the lookup, so while lookups into distinct inner | |
10566 | * maps from the same outer map share the same map_ptr as active_lock.ptr, they | |
10567 | * will get different reg->id assigned to each lookup, hence different | |
10568 | * active_lock.id. | |
10569 | * | |
10570 | * In case of allocated objects, active_lock.ptr is the reg->btf, and the | |
10571 | * reg->id is a unique ID preserved after the NULL pointer check on the pointer | |
10572 | * returned from bpf_obj_new. Each allocation receives a new reg->id. | |
10573 | */ | |
10574 | static int check_reg_allocation_locked(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
10575 | { | |
10576 | void *ptr; | |
10577 | u32 id; | |
10578 | ||
10579 | switch ((int)reg->type) { | |
10580 | case PTR_TO_MAP_VALUE: | |
10581 | ptr = reg->map_ptr; | |
10582 | break; | |
10583 | case PTR_TO_BTF_ID | MEM_ALLOC: | |
10584 | ptr = reg->btf; | |
10585 | break; | |
10586 | default: | |
10587 | verbose(env, "verifier internal error: unknown reg type for lock check\n"); | |
10588 | return -EFAULT; | |
10589 | } | |
10590 | id = reg->id; | |
10591 | ||
10592 | if (!env->cur_state->active_lock.ptr) | |
10593 | return -EINVAL; | |
10594 | if (env->cur_state->active_lock.ptr != ptr || | |
10595 | env->cur_state->active_lock.id != id) { | |
10596 | verbose(env, "held lock and object are not in the same allocation\n"); | |
10597 | return -EINVAL; | |
10598 | } | |
10599 | return 0; | |
10600 | } | |
10601 | ||
10602 | static bool is_bpf_list_api_kfunc(u32 btf_id) | |
10603 | { | |
d2dcc67d DM |
10604 | return btf_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
10605 | btf_id == special_kfunc_list[KF_bpf_list_push_back_impl] || | |
8cab76ec KKD |
10606 | btf_id == special_kfunc_list[KF_bpf_list_pop_front] || |
10607 | btf_id == special_kfunc_list[KF_bpf_list_pop_back]; | |
10608 | } | |
10609 | ||
cd6791b4 DM |
10610 | static bool is_bpf_rbtree_api_kfunc(u32 btf_id) |
10611 | { | |
d2dcc67d | 10612 | return btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl] || |
cd6791b4 DM |
10613 | btf_id == special_kfunc_list[KF_bpf_rbtree_remove] || |
10614 | btf_id == special_kfunc_list[KF_bpf_rbtree_first]; | |
10615 | } | |
10616 | ||
10617 | static bool is_bpf_graph_api_kfunc(u32 btf_id) | |
10618 | { | |
7c50b1cb DM |
10619 | return is_bpf_list_api_kfunc(btf_id) || is_bpf_rbtree_api_kfunc(btf_id) || |
10620 | btf_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]; | |
cd6791b4 DM |
10621 | } |
10622 | ||
5d92ddc3 DM |
10623 | static bool is_callback_calling_kfunc(u32 btf_id) |
10624 | { | |
d2dcc67d | 10625 | return btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl]; |
5d92ddc3 DM |
10626 | } |
10627 | ||
10628 | static bool is_rbtree_lock_required_kfunc(u32 btf_id) | |
10629 | { | |
10630 | return is_bpf_rbtree_api_kfunc(btf_id); | |
10631 | } | |
10632 | ||
cd6791b4 DM |
10633 | static bool check_kfunc_is_graph_root_api(struct bpf_verifier_env *env, |
10634 | enum btf_field_type head_field_type, | |
10635 | u32 kfunc_btf_id) | |
10636 | { | |
10637 | bool ret; | |
10638 | ||
10639 | switch (head_field_type) { | |
10640 | case BPF_LIST_HEAD: | |
10641 | ret = is_bpf_list_api_kfunc(kfunc_btf_id); | |
10642 | break; | |
10643 | case BPF_RB_ROOT: | |
10644 | ret = is_bpf_rbtree_api_kfunc(kfunc_btf_id); | |
10645 | break; | |
10646 | default: | |
10647 | verbose(env, "verifier internal error: unexpected graph root argument type %s\n", | |
10648 | btf_field_type_name(head_field_type)); | |
10649 | return false; | |
10650 | } | |
10651 | ||
10652 | if (!ret) | |
10653 | verbose(env, "verifier internal error: %s head arg for unknown kfunc\n", | |
10654 | btf_field_type_name(head_field_type)); | |
10655 | return ret; | |
10656 | } | |
10657 | ||
10658 | static bool check_kfunc_is_graph_node_api(struct bpf_verifier_env *env, | |
10659 | enum btf_field_type node_field_type, | |
10660 | u32 kfunc_btf_id) | |
8cab76ec | 10661 | { |
cd6791b4 DM |
10662 | bool ret; |
10663 | ||
10664 | switch (node_field_type) { | |
10665 | case BPF_LIST_NODE: | |
d2dcc67d DM |
10666 | ret = (kfunc_btf_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
10667 | kfunc_btf_id == special_kfunc_list[KF_bpf_list_push_back_impl]); | |
cd6791b4 DM |
10668 | break; |
10669 | case BPF_RB_NODE: | |
10670 | ret = (kfunc_btf_id == special_kfunc_list[KF_bpf_rbtree_remove] || | |
d2dcc67d | 10671 | kfunc_btf_id == special_kfunc_list[KF_bpf_rbtree_add_impl]); |
cd6791b4 DM |
10672 | break; |
10673 | default: | |
10674 | verbose(env, "verifier internal error: unexpected graph node argument type %s\n", | |
10675 | btf_field_type_name(node_field_type)); | |
10676 | return false; | |
10677 | } | |
10678 | ||
10679 | if (!ret) | |
10680 | verbose(env, "verifier internal error: %s node arg for unknown kfunc\n", | |
10681 | btf_field_type_name(node_field_type)); | |
10682 | return ret; | |
10683 | } | |
10684 | ||
10685 | static int | |
10686 | __process_kf_arg_ptr_to_graph_root(struct bpf_verifier_env *env, | |
10687 | struct bpf_reg_state *reg, u32 regno, | |
10688 | struct bpf_kfunc_call_arg_meta *meta, | |
10689 | enum btf_field_type head_field_type, | |
10690 | struct btf_field **head_field) | |
10691 | { | |
10692 | const char *head_type_name; | |
8cab76ec KKD |
10693 | struct btf_field *field; |
10694 | struct btf_record *rec; | |
cd6791b4 | 10695 | u32 head_off; |
8cab76ec | 10696 | |
cd6791b4 DM |
10697 | if (meta->btf != btf_vmlinux) { |
10698 | verbose(env, "verifier internal error: unexpected btf mismatch in kfunc call\n"); | |
8cab76ec KKD |
10699 | return -EFAULT; |
10700 | } | |
10701 | ||
cd6791b4 DM |
10702 | if (!check_kfunc_is_graph_root_api(env, head_field_type, meta->func_id)) |
10703 | return -EFAULT; | |
10704 | ||
10705 | head_type_name = btf_field_type_name(head_field_type); | |
8cab76ec KKD |
10706 | if (!tnum_is_const(reg->var_off)) { |
10707 | verbose(env, | |
cd6791b4 DM |
10708 | "R%d doesn't have constant offset. %s has to be at the constant offset\n", |
10709 | regno, head_type_name); | |
8cab76ec KKD |
10710 | return -EINVAL; |
10711 | } | |
10712 | ||
10713 | rec = reg_btf_record(reg); | |
cd6791b4 DM |
10714 | head_off = reg->off + reg->var_off.value; |
10715 | field = btf_record_find(rec, head_off, head_field_type); | |
8cab76ec | 10716 | if (!field) { |
cd6791b4 | 10717 | verbose(env, "%s not found at offset=%u\n", head_type_name, head_off); |
8cab76ec KKD |
10718 | return -EINVAL; |
10719 | } | |
10720 | ||
10721 | /* All functions require bpf_list_head to be protected using a bpf_spin_lock */ | |
10722 | if (check_reg_allocation_locked(env, reg)) { | |
cd6791b4 DM |
10723 | verbose(env, "bpf_spin_lock at off=%d must be held for %s\n", |
10724 | rec->spin_lock_off, head_type_name); | |
8cab76ec KKD |
10725 | return -EINVAL; |
10726 | } | |
10727 | ||
cd6791b4 DM |
10728 | if (*head_field) { |
10729 | verbose(env, "verifier internal error: repeating %s arg\n", head_type_name); | |
8cab76ec KKD |
10730 | return -EFAULT; |
10731 | } | |
cd6791b4 | 10732 | *head_field = field; |
8cab76ec KKD |
10733 | return 0; |
10734 | } | |
10735 | ||
cd6791b4 | 10736 | static int process_kf_arg_ptr_to_list_head(struct bpf_verifier_env *env, |
8cab76ec KKD |
10737 | struct bpf_reg_state *reg, u32 regno, |
10738 | struct bpf_kfunc_call_arg_meta *meta) | |
10739 | { | |
cd6791b4 DM |
10740 | return __process_kf_arg_ptr_to_graph_root(env, reg, regno, meta, BPF_LIST_HEAD, |
10741 | &meta->arg_list_head.field); | |
10742 | } | |
10743 | ||
10744 | static int process_kf_arg_ptr_to_rbtree_root(struct bpf_verifier_env *env, | |
10745 | struct bpf_reg_state *reg, u32 regno, | |
10746 | struct bpf_kfunc_call_arg_meta *meta) | |
10747 | { | |
10748 | return __process_kf_arg_ptr_to_graph_root(env, reg, regno, meta, BPF_RB_ROOT, | |
10749 | &meta->arg_rbtree_root.field); | |
10750 | } | |
10751 | ||
10752 | static int | |
10753 | __process_kf_arg_ptr_to_graph_node(struct bpf_verifier_env *env, | |
10754 | struct bpf_reg_state *reg, u32 regno, | |
10755 | struct bpf_kfunc_call_arg_meta *meta, | |
10756 | enum btf_field_type head_field_type, | |
10757 | enum btf_field_type node_field_type, | |
10758 | struct btf_field **node_field) | |
10759 | { | |
10760 | const char *node_type_name; | |
8cab76ec KKD |
10761 | const struct btf_type *et, *t; |
10762 | struct btf_field *field; | |
cd6791b4 | 10763 | u32 node_off; |
8cab76ec | 10764 | |
cd6791b4 DM |
10765 | if (meta->btf != btf_vmlinux) { |
10766 | verbose(env, "verifier internal error: unexpected btf mismatch in kfunc call\n"); | |
8cab76ec KKD |
10767 | return -EFAULT; |
10768 | } | |
10769 | ||
cd6791b4 DM |
10770 | if (!check_kfunc_is_graph_node_api(env, node_field_type, meta->func_id)) |
10771 | return -EFAULT; | |
10772 | ||
10773 | node_type_name = btf_field_type_name(node_field_type); | |
8cab76ec KKD |
10774 | if (!tnum_is_const(reg->var_off)) { |
10775 | verbose(env, | |
cd6791b4 DM |
10776 | "R%d doesn't have constant offset. %s has to be at the constant offset\n", |
10777 | regno, node_type_name); | |
8cab76ec KKD |
10778 | return -EINVAL; |
10779 | } | |
10780 | ||
cd6791b4 DM |
10781 | node_off = reg->off + reg->var_off.value; |
10782 | field = reg_find_field_offset(reg, node_off, node_field_type); | |
10783 | if (!field || field->offset != node_off) { | |
10784 | verbose(env, "%s not found at offset=%u\n", node_type_name, node_off); | |
8cab76ec KKD |
10785 | return -EINVAL; |
10786 | } | |
10787 | ||
cd6791b4 | 10788 | field = *node_field; |
8cab76ec | 10789 | |
30465003 | 10790 | et = btf_type_by_id(field->graph_root.btf, field->graph_root.value_btf_id); |
8cab76ec | 10791 | t = btf_type_by_id(reg->btf, reg->btf_id); |
30465003 DM |
10792 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, 0, field->graph_root.btf, |
10793 | field->graph_root.value_btf_id, true)) { | |
cd6791b4 | 10794 | verbose(env, "operation on %s expects arg#1 %s at offset=%d " |
8cab76ec | 10795 | "in struct %s, but arg is at offset=%d in struct %s\n", |
cd6791b4 DM |
10796 | btf_field_type_name(head_field_type), |
10797 | btf_field_type_name(node_field_type), | |
30465003 DM |
10798 | field->graph_root.node_offset, |
10799 | btf_name_by_offset(field->graph_root.btf, et->name_off), | |
cd6791b4 | 10800 | node_off, btf_name_by_offset(reg->btf, t->name_off)); |
8cab76ec KKD |
10801 | return -EINVAL; |
10802 | } | |
2140a6e3 DM |
10803 | meta->arg_btf = reg->btf; |
10804 | meta->arg_btf_id = reg->btf_id; | |
8cab76ec | 10805 | |
cd6791b4 DM |
10806 | if (node_off != field->graph_root.node_offset) { |
10807 | verbose(env, "arg#1 offset=%d, but expected %s at offset=%d in struct %s\n", | |
10808 | node_off, btf_field_type_name(node_field_type), | |
10809 | field->graph_root.node_offset, | |
30465003 | 10810 | btf_name_by_offset(field->graph_root.btf, et->name_off)); |
8cab76ec KKD |
10811 | return -EINVAL; |
10812 | } | |
6a3cd331 DM |
10813 | |
10814 | return 0; | |
8cab76ec KKD |
10815 | } |
10816 | ||
cd6791b4 DM |
10817 | static int process_kf_arg_ptr_to_list_node(struct bpf_verifier_env *env, |
10818 | struct bpf_reg_state *reg, u32 regno, | |
10819 | struct bpf_kfunc_call_arg_meta *meta) | |
10820 | { | |
10821 | return __process_kf_arg_ptr_to_graph_node(env, reg, regno, meta, | |
10822 | BPF_LIST_HEAD, BPF_LIST_NODE, | |
10823 | &meta->arg_list_head.field); | |
10824 | } | |
10825 | ||
10826 | static int process_kf_arg_ptr_to_rbtree_node(struct bpf_verifier_env *env, | |
10827 | struct bpf_reg_state *reg, u32 regno, | |
10828 | struct bpf_kfunc_call_arg_meta *meta) | |
10829 | { | |
10830 | return __process_kf_arg_ptr_to_graph_node(env, reg, regno, meta, | |
10831 | BPF_RB_ROOT, BPF_RB_NODE, | |
10832 | &meta->arg_rbtree_root.field); | |
10833 | } | |
10834 | ||
1d18feb2 JK |
10835 | static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_arg_meta *meta, |
10836 | int insn_idx) | |
00b85860 KKD |
10837 | { |
10838 | const char *func_name = meta->func_name, *ref_tname; | |
10839 | const struct btf *btf = meta->btf; | |
10840 | const struct btf_param *args; | |
7c50b1cb | 10841 | struct btf_record *rec; |
00b85860 KKD |
10842 | u32 i, nargs; |
10843 | int ret; | |
10844 | ||
10845 | args = (const struct btf_param *)(meta->func_proto + 1); | |
10846 | nargs = btf_type_vlen(meta->func_proto); | |
10847 | if (nargs > MAX_BPF_FUNC_REG_ARGS) { | |
10848 | verbose(env, "Function %s has %d > %d args\n", func_name, nargs, | |
10849 | MAX_BPF_FUNC_REG_ARGS); | |
10850 | return -EINVAL; | |
10851 | } | |
10852 | ||
10853 | /* Check that BTF function arguments match actual types that the | |
10854 | * verifier sees. | |
10855 | */ | |
10856 | for (i = 0; i < nargs; i++) { | |
10857 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[i + 1]; | |
10858 | const struct btf_type *t, *ref_t, *resolve_ret; | |
10859 | enum bpf_arg_type arg_type = ARG_DONTCARE; | |
10860 | u32 regno = i + 1, ref_id, type_size; | |
10861 | bool is_ret_buf_sz = false; | |
10862 | int kf_arg_type; | |
10863 | ||
10864 | t = btf_type_skip_modifiers(btf, args[i].type, NULL); | |
958cf2e2 KKD |
10865 | |
10866 | if (is_kfunc_arg_ignore(btf, &args[i])) | |
10867 | continue; | |
10868 | ||
00b85860 KKD |
10869 | if (btf_type_is_scalar(t)) { |
10870 | if (reg->type != SCALAR_VALUE) { | |
10871 | verbose(env, "R%d is not a scalar\n", regno); | |
10872 | return -EINVAL; | |
10873 | } | |
a50388db KKD |
10874 | |
10875 | if (is_kfunc_arg_constant(meta->btf, &args[i])) { | |
10876 | if (meta->arg_constant.found) { | |
10877 | verbose(env, "verifier internal error: only one constant argument permitted\n"); | |
10878 | return -EFAULT; | |
10879 | } | |
10880 | if (!tnum_is_const(reg->var_off)) { | |
10881 | verbose(env, "R%d must be a known constant\n", regno); | |
10882 | return -EINVAL; | |
10883 | } | |
10884 | ret = mark_chain_precision(env, regno); | |
10885 | if (ret < 0) | |
10886 | return ret; | |
10887 | meta->arg_constant.found = true; | |
10888 | meta->arg_constant.value = reg->var_off.value; | |
10889 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdonly_buf_size")) { | |
00b85860 KKD |
10890 | meta->r0_rdonly = true; |
10891 | is_ret_buf_sz = true; | |
10892 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdwr_buf_size")) { | |
10893 | is_ret_buf_sz = true; | |
10894 | } | |
10895 | ||
10896 | if (is_ret_buf_sz) { | |
10897 | if (meta->r0_size) { | |
10898 | verbose(env, "2 or more rdonly/rdwr_buf_size parameters for kfunc"); | |
10899 | return -EINVAL; | |
10900 | } | |
10901 | ||
10902 | if (!tnum_is_const(reg->var_off)) { | |
10903 | verbose(env, "R%d is not a const\n", regno); | |
10904 | return -EINVAL; | |
10905 | } | |
10906 | ||
10907 | meta->r0_size = reg->var_off.value; | |
10908 | ret = mark_chain_precision(env, regno); | |
10909 | if (ret) | |
10910 | return ret; | |
10911 | } | |
10912 | continue; | |
10913 | } | |
10914 | ||
10915 | if (!btf_type_is_ptr(t)) { | |
10916 | verbose(env, "Unrecognized arg#%d type %s\n", i, btf_type_str(t)); | |
10917 | return -EINVAL; | |
10918 | } | |
10919 | ||
20c09d92 | 10920 | if ((is_kfunc_trusted_args(meta) || is_kfunc_rcu(meta)) && |
caf713c3 DV |
10921 | (register_is_null(reg) || type_may_be_null(reg->type))) { |
10922 | verbose(env, "Possibly NULL pointer passed to trusted arg%d\n", i); | |
10923 | return -EACCES; | |
10924 | } | |
10925 | ||
00b85860 KKD |
10926 | if (reg->ref_obj_id) { |
10927 | if (is_kfunc_release(meta) && meta->ref_obj_id) { | |
10928 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
10929 | regno, reg->ref_obj_id, | |
10930 | meta->ref_obj_id); | |
10931 | return -EFAULT; | |
10932 | } | |
10933 | meta->ref_obj_id = reg->ref_obj_id; | |
10934 | if (is_kfunc_release(meta)) | |
10935 | meta->release_regno = regno; | |
10936 | } | |
10937 | ||
10938 | ref_t = btf_type_skip_modifiers(btf, t->type, &ref_id); | |
10939 | ref_tname = btf_name_by_offset(btf, ref_t->name_off); | |
10940 | ||
10941 | kf_arg_type = get_kfunc_ptr_arg_type(env, meta, t, ref_t, ref_tname, args, i, nargs); | |
10942 | if (kf_arg_type < 0) | |
10943 | return kf_arg_type; | |
10944 | ||
10945 | switch (kf_arg_type) { | |
ac9f0605 | 10946 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
00b85860 | 10947 | case KF_ARG_PTR_TO_BTF_ID: |
fca1aa75 | 10948 | if (!is_kfunc_trusted_args(meta) && !is_kfunc_rcu(meta)) |
00b85860 | 10949 | break; |
3f00c523 DV |
10950 | |
10951 | if (!is_trusted_reg(reg)) { | |
fca1aa75 YS |
10952 | if (!is_kfunc_rcu(meta)) { |
10953 | verbose(env, "R%d must be referenced or trusted\n", regno); | |
10954 | return -EINVAL; | |
10955 | } | |
10956 | if (!is_rcu_reg(reg)) { | |
10957 | verbose(env, "R%d must be a rcu pointer\n", regno); | |
10958 | return -EINVAL; | |
10959 | } | |
00b85860 | 10960 | } |
fca1aa75 | 10961 | |
00b85860 KKD |
10962 | fallthrough; |
10963 | case KF_ARG_PTR_TO_CTX: | |
10964 | /* Trusted arguments have the same offset checks as release arguments */ | |
10965 | arg_type |= OBJ_RELEASE; | |
10966 | break; | |
00b85860 | 10967 | case KF_ARG_PTR_TO_DYNPTR: |
06accc87 | 10968 | case KF_ARG_PTR_TO_ITER: |
8cab76ec KKD |
10969 | case KF_ARG_PTR_TO_LIST_HEAD: |
10970 | case KF_ARG_PTR_TO_LIST_NODE: | |
cd6791b4 DM |
10971 | case KF_ARG_PTR_TO_RB_ROOT: |
10972 | case KF_ARG_PTR_TO_RB_NODE: | |
00b85860 KKD |
10973 | case KF_ARG_PTR_TO_MEM: |
10974 | case KF_ARG_PTR_TO_MEM_SIZE: | |
5d92ddc3 | 10975 | case KF_ARG_PTR_TO_CALLBACK: |
7c50b1cb | 10976 | case KF_ARG_PTR_TO_REFCOUNTED_KPTR: |
00b85860 KKD |
10977 | /* Trusted by default */ |
10978 | break; | |
10979 | default: | |
10980 | WARN_ON_ONCE(1); | |
10981 | return -EFAULT; | |
10982 | } | |
10983 | ||
10984 | if (is_kfunc_release(meta) && reg->ref_obj_id) | |
10985 | arg_type |= OBJ_RELEASE; | |
10986 | ret = check_func_arg_reg_off(env, reg, regno, arg_type); | |
10987 | if (ret < 0) | |
10988 | return ret; | |
10989 | ||
10990 | switch (kf_arg_type) { | |
10991 | case KF_ARG_PTR_TO_CTX: | |
10992 | if (reg->type != PTR_TO_CTX) { | |
10993 | verbose(env, "arg#%d expected pointer to ctx, but got %s\n", i, btf_type_str(t)); | |
10994 | return -EINVAL; | |
10995 | } | |
fd264ca0 YS |
10996 | |
10997 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { | |
10998 | ret = get_kern_ctx_btf_id(&env->log, resolve_prog_type(env->prog)); | |
10999 | if (ret < 0) | |
11000 | return -EINVAL; | |
11001 | meta->ret_btf_id = ret; | |
11002 | } | |
00b85860 | 11003 | break; |
ac9f0605 KKD |
11004 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
11005 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11006 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
11007 | return -EINVAL; | |
11008 | } | |
11009 | if (!reg->ref_obj_id) { | |
11010 | verbose(env, "allocated object must be referenced\n"); | |
11011 | return -EINVAL; | |
11012 | } | |
11013 | if (meta->btf == btf_vmlinux && | |
11014 | meta->func_id == special_kfunc_list[KF_bpf_obj_drop_impl]) { | |
4d585f48 DM |
11015 | meta->arg_btf = reg->btf; |
11016 | meta->arg_btf_id = reg->btf_id; | |
ac9f0605 KKD |
11017 | } |
11018 | break; | |
00b85860 | 11019 | case KF_ARG_PTR_TO_DYNPTR: |
d96d937d JK |
11020 | { |
11021 | enum bpf_arg_type dynptr_arg_type = ARG_PTR_TO_DYNPTR; | |
361f129f | 11022 | int clone_ref_obj_id = 0; |
d96d937d | 11023 | |
6b75bd3d | 11024 | if (reg->type != PTR_TO_STACK && |
27060531 | 11025 | reg->type != CONST_PTR_TO_DYNPTR) { |
6b75bd3d | 11026 | verbose(env, "arg#%d expected pointer to stack or dynptr_ptr\n", i); |
00b85860 KKD |
11027 | return -EINVAL; |
11028 | } | |
11029 | ||
d96d937d JK |
11030 | if (reg->type == CONST_PTR_TO_DYNPTR) |
11031 | dynptr_arg_type |= MEM_RDONLY; | |
11032 | ||
11033 | if (is_kfunc_arg_uninit(btf, &args[i])) | |
11034 | dynptr_arg_type |= MEM_UNINIT; | |
11035 | ||
361f129f | 11036 | if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_from_skb]) { |
b5964b96 | 11037 | dynptr_arg_type |= DYNPTR_TYPE_SKB; |
361f129f | 11038 | } else if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_from_xdp]) { |
05421aec | 11039 | dynptr_arg_type |= DYNPTR_TYPE_XDP; |
361f129f JK |
11040 | } else if (meta->func_id == special_kfunc_list[KF_bpf_dynptr_clone] && |
11041 | (dynptr_arg_type & MEM_UNINIT)) { | |
11042 | enum bpf_dynptr_type parent_type = meta->initialized_dynptr.type; | |
11043 | ||
11044 | if (parent_type == BPF_DYNPTR_TYPE_INVALID) { | |
11045 | verbose(env, "verifier internal error: no dynptr type for parent of clone\n"); | |
11046 | return -EFAULT; | |
11047 | } | |
11048 | ||
11049 | dynptr_arg_type |= (unsigned int)get_dynptr_type_flag(parent_type); | |
11050 | clone_ref_obj_id = meta->initialized_dynptr.ref_obj_id; | |
11051 | if (dynptr_type_refcounted(parent_type) && !clone_ref_obj_id) { | |
11052 | verbose(env, "verifier internal error: missing ref obj id for parent of clone\n"); | |
11053 | return -EFAULT; | |
11054 | } | |
11055 | } | |
b5964b96 | 11056 | |
361f129f | 11057 | ret = process_dynptr_func(env, regno, insn_idx, dynptr_arg_type, clone_ref_obj_id); |
6b75bd3d KKD |
11058 | if (ret < 0) |
11059 | return ret; | |
66e3a13e JK |
11060 | |
11061 | if (!(dynptr_arg_type & MEM_UNINIT)) { | |
11062 | int id = dynptr_id(env, reg); | |
11063 | ||
11064 | if (id < 0) { | |
11065 | verbose(env, "verifier internal error: failed to obtain dynptr id\n"); | |
11066 | return id; | |
11067 | } | |
11068 | meta->initialized_dynptr.id = id; | |
11069 | meta->initialized_dynptr.type = dynptr_get_type(env, reg); | |
361f129f | 11070 | meta->initialized_dynptr.ref_obj_id = dynptr_ref_obj_id(env, reg); |
66e3a13e JK |
11071 | } |
11072 | ||
00b85860 | 11073 | break; |
d96d937d | 11074 | } |
06accc87 AN |
11075 | case KF_ARG_PTR_TO_ITER: |
11076 | ret = process_iter_arg(env, regno, insn_idx, meta); | |
11077 | if (ret < 0) | |
11078 | return ret; | |
11079 | break; | |
8cab76ec KKD |
11080 | case KF_ARG_PTR_TO_LIST_HEAD: |
11081 | if (reg->type != PTR_TO_MAP_VALUE && | |
11082 | reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11083 | verbose(env, "arg#%d expected pointer to map value or allocated object\n", i); | |
11084 | return -EINVAL; | |
11085 | } | |
11086 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC) && !reg->ref_obj_id) { | |
11087 | verbose(env, "allocated object must be referenced\n"); | |
11088 | return -EINVAL; | |
11089 | } | |
11090 | ret = process_kf_arg_ptr_to_list_head(env, reg, regno, meta); | |
11091 | if (ret < 0) | |
11092 | return ret; | |
11093 | break; | |
cd6791b4 DM |
11094 | case KF_ARG_PTR_TO_RB_ROOT: |
11095 | if (reg->type != PTR_TO_MAP_VALUE && | |
11096 | reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11097 | verbose(env, "arg#%d expected pointer to map value or allocated object\n", i); | |
11098 | return -EINVAL; | |
11099 | } | |
11100 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC) && !reg->ref_obj_id) { | |
11101 | verbose(env, "allocated object must be referenced\n"); | |
11102 | return -EINVAL; | |
11103 | } | |
11104 | ret = process_kf_arg_ptr_to_rbtree_root(env, reg, regno, meta); | |
11105 | if (ret < 0) | |
11106 | return ret; | |
11107 | break; | |
8cab76ec KKD |
11108 | case KF_ARG_PTR_TO_LIST_NODE: |
11109 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11110 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
11111 | return -EINVAL; | |
11112 | } | |
11113 | if (!reg->ref_obj_id) { | |
11114 | verbose(env, "allocated object must be referenced\n"); | |
11115 | return -EINVAL; | |
11116 | } | |
11117 | ret = process_kf_arg_ptr_to_list_node(env, reg, regno, meta); | |
11118 | if (ret < 0) | |
11119 | return ret; | |
11120 | break; | |
cd6791b4 | 11121 | case KF_ARG_PTR_TO_RB_NODE: |
a40d3632 DM |
11122 | if (meta->func_id == special_kfunc_list[KF_bpf_rbtree_remove]) { |
11123 | if (!type_is_non_owning_ref(reg->type) || reg->ref_obj_id) { | |
11124 | verbose(env, "rbtree_remove node input must be non-owning ref\n"); | |
11125 | return -EINVAL; | |
11126 | } | |
11127 | if (in_rbtree_lock_required_cb(env)) { | |
11128 | verbose(env, "rbtree_remove not allowed in rbtree cb\n"); | |
11129 | return -EINVAL; | |
11130 | } | |
11131 | } else { | |
11132 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
11133 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
11134 | return -EINVAL; | |
11135 | } | |
11136 | if (!reg->ref_obj_id) { | |
11137 | verbose(env, "allocated object must be referenced\n"); | |
11138 | return -EINVAL; | |
11139 | } | |
cd6791b4 | 11140 | } |
a40d3632 | 11141 | |
cd6791b4 DM |
11142 | ret = process_kf_arg_ptr_to_rbtree_node(env, reg, regno, meta); |
11143 | if (ret < 0) | |
11144 | return ret; | |
11145 | break; | |
00b85860 KKD |
11146 | case KF_ARG_PTR_TO_BTF_ID: |
11147 | /* Only base_type is checked, further checks are done here */ | |
3f00c523 | 11148 | if ((base_type(reg->type) != PTR_TO_BTF_ID || |
fca1aa75 | 11149 | (bpf_type_has_unsafe_modifiers(reg->type) && !is_rcu_reg(reg))) && |
3f00c523 DV |
11150 | !reg2btf_ids[base_type(reg->type)]) { |
11151 | verbose(env, "arg#%d is %s ", i, reg_type_str(env, reg->type)); | |
11152 | verbose(env, "expected %s or socket\n", | |
11153 | reg_type_str(env, base_type(reg->type) | | |
11154 | (type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS))); | |
00b85860 KKD |
11155 | return -EINVAL; |
11156 | } | |
11157 | ret = process_kf_arg_ptr_to_btf_id(env, reg, ref_t, ref_tname, ref_id, meta, i); | |
11158 | if (ret < 0) | |
11159 | return ret; | |
11160 | break; | |
11161 | case KF_ARG_PTR_TO_MEM: | |
11162 | resolve_ret = btf_resolve_size(btf, ref_t, &type_size); | |
11163 | if (IS_ERR(resolve_ret)) { | |
11164 | verbose(env, "arg#%d reference type('%s %s') size cannot be determined: %ld\n", | |
11165 | i, btf_type_str(ref_t), ref_tname, PTR_ERR(resolve_ret)); | |
11166 | return -EINVAL; | |
11167 | } | |
11168 | ret = check_mem_reg(env, reg, regno, type_size); | |
11169 | if (ret < 0) | |
11170 | return ret; | |
11171 | break; | |
11172 | case KF_ARG_PTR_TO_MEM_SIZE: | |
66e3a13e | 11173 | { |
3bda08b6 DR |
11174 | struct bpf_reg_state *buff_reg = ®s[regno]; |
11175 | const struct btf_param *buff_arg = &args[i]; | |
66e3a13e JK |
11176 | struct bpf_reg_state *size_reg = ®s[regno + 1]; |
11177 | const struct btf_param *size_arg = &args[i + 1]; | |
11178 | ||
3bda08b6 DR |
11179 | if (!register_is_null(buff_reg) || !is_kfunc_arg_optional(meta->btf, buff_arg)) { |
11180 | ret = check_kfunc_mem_size_reg(env, size_reg, regno + 1); | |
11181 | if (ret < 0) { | |
11182 | verbose(env, "arg#%d arg#%d memory, len pair leads to invalid memory access\n", i, i + 1); | |
11183 | return ret; | |
11184 | } | |
00b85860 | 11185 | } |
66e3a13e JK |
11186 | |
11187 | if (is_kfunc_arg_const_mem_size(meta->btf, size_arg, size_reg)) { | |
11188 | if (meta->arg_constant.found) { | |
11189 | verbose(env, "verifier internal error: only one constant argument permitted\n"); | |
11190 | return -EFAULT; | |
11191 | } | |
11192 | if (!tnum_is_const(size_reg->var_off)) { | |
11193 | verbose(env, "R%d must be a known constant\n", regno + 1); | |
11194 | return -EINVAL; | |
11195 | } | |
11196 | meta->arg_constant.found = true; | |
11197 | meta->arg_constant.value = size_reg->var_off.value; | |
11198 | } | |
11199 | ||
11200 | /* Skip next '__sz' or '__szk' argument */ | |
00b85860 KKD |
11201 | i++; |
11202 | break; | |
66e3a13e | 11203 | } |
5d92ddc3 DM |
11204 | case KF_ARG_PTR_TO_CALLBACK: |
11205 | meta->subprogno = reg->subprogno; | |
11206 | break; | |
7c50b1cb | 11207 | case KF_ARG_PTR_TO_REFCOUNTED_KPTR: |
7793fc3b | 11208 | if (!type_is_ptr_alloc_obj(reg->type)) { |
7c50b1cb DM |
11209 | verbose(env, "arg#%d is neither owning or non-owning ref\n", i); |
11210 | return -EINVAL; | |
11211 | } | |
7793fc3b DM |
11212 | if (!type_is_non_owning_ref(reg->type)) |
11213 | meta->arg_owning_ref = true; | |
7c50b1cb DM |
11214 | |
11215 | rec = reg_btf_record(reg); | |
11216 | if (!rec) { | |
11217 | verbose(env, "verifier internal error: Couldn't find btf_record\n"); | |
11218 | return -EFAULT; | |
11219 | } | |
11220 | ||
11221 | if (rec->refcount_off < 0) { | |
11222 | verbose(env, "arg#%d doesn't point to a type with bpf_refcount field\n", i); | |
11223 | return -EINVAL; | |
11224 | } | |
ba2464c8 | 11225 | |
4d585f48 DM |
11226 | meta->arg_btf = reg->btf; |
11227 | meta->arg_btf_id = reg->btf_id; | |
7c50b1cb | 11228 | break; |
00b85860 KKD |
11229 | } |
11230 | } | |
11231 | ||
11232 | if (is_kfunc_release(meta) && !meta->release_regno) { | |
11233 | verbose(env, "release kernel function %s expects refcounted PTR_TO_BTF_ID\n", | |
11234 | func_name); | |
11235 | return -EINVAL; | |
11236 | } | |
11237 | ||
11238 | return 0; | |
11239 | } | |
11240 | ||
07236eab AN |
11241 | static int fetch_kfunc_meta(struct bpf_verifier_env *env, |
11242 | struct bpf_insn *insn, | |
11243 | struct bpf_kfunc_call_arg_meta *meta, | |
11244 | const char **kfunc_name) | |
e6ac2450 | 11245 | { |
07236eab AN |
11246 | const struct btf_type *func, *func_proto; |
11247 | u32 func_id, *kfunc_flags; | |
11248 | const char *func_name; | |
2357672c | 11249 | struct btf *desc_btf; |
e6ac2450 | 11250 | |
07236eab AN |
11251 | if (kfunc_name) |
11252 | *kfunc_name = NULL; | |
11253 | ||
a5d82727 | 11254 | if (!insn->imm) |
07236eab | 11255 | return -EINVAL; |
a5d82727 | 11256 | |
43bf0878 | 11257 | desc_btf = find_kfunc_desc_btf(env, insn->off); |
2357672c KKD |
11258 | if (IS_ERR(desc_btf)) |
11259 | return PTR_ERR(desc_btf); | |
11260 | ||
e6ac2450 | 11261 | func_id = insn->imm; |
2357672c KKD |
11262 | func = btf_type_by_id(desc_btf, func_id); |
11263 | func_name = btf_name_by_offset(desc_btf, func->name_off); | |
07236eab AN |
11264 | if (kfunc_name) |
11265 | *kfunc_name = func_name; | |
2357672c | 11266 | func_proto = btf_type_by_id(desc_btf, func->type); |
e6ac2450 | 11267 | |
e924e80e | 11268 | kfunc_flags = btf_kfunc_id_set_contains(desc_btf, func_id, env->prog); |
a4703e31 | 11269 | if (!kfunc_flags) { |
e6ac2450 MKL |
11270 | return -EACCES; |
11271 | } | |
00b85860 | 11272 | |
07236eab AN |
11273 | memset(meta, 0, sizeof(*meta)); |
11274 | meta->btf = desc_btf; | |
11275 | meta->func_id = func_id; | |
11276 | meta->kfunc_flags = *kfunc_flags; | |
11277 | meta->func_proto = func_proto; | |
11278 | meta->func_name = func_name; | |
11279 | ||
11280 | return 0; | |
11281 | } | |
11282 | ||
11283 | static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
11284 | int *insn_idx_p) | |
11285 | { | |
11286 | const struct btf_type *t, *ptr_type; | |
11287 | u32 i, nargs, ptr_type_id, release_ref_obj_id; | |
11288 | struct bpf_reg_state *regs = cur_regs(env); | |
11289 | const char *func_name, *ptr_type_name; | |
11290 | bool sleepable, rcu_lock, rcu_unlock; | |
11291 | struct bpf_kfunc_call_arg_meta meta; | |
11292 | struct bpf_insn_aux_data *insn_aux; | |
11293 | int err, insn_idx = *insn_idx_p; | |
11294 | const struct btf_param *args; | |
11295 | const struct btf_type *ret_t; | |
11296 | struct btf *desc_btf; | |
11297 | ||
11298 | /* skip for now, but return error when we find this in fixup_kfunc_call */ | |
11299 | if (!insn->imm) | |
11300 | return 0; | |
11301 | ||
11302 | err = fetch_kfunc_meta(env, insn, &meta, &func_name); | |
11303 | if (err == -EACCES && func_name) | |
11304 | verbose(env, "calling kernel function %s is not allowed\n", func_name); | |
11305 | if (err) | |
11306 | return err; | |
11307 | desc_btf = meta.btf; | |
11308 | insn_aux = &env->insn_aux_data[insn_idx]; | |
00b85860 | 11309 | |
06accc87 AN |
11310 | insn_aux->is_iter_next = is_iter_next_kfunc(&meta); |
11311 | ||
00b85860 KKD |
11312 | if (is_kfunc_destructive(&meta) && !capable(CAP_SYS_BOOT)) { |
11313 | verbose(env, "destructive kfunc calls require CAP_SYS_BOOT capability\n"); | |
4dd48c6f AS |
11314 | return -EACCES; |
11315 | } | |
11316 | ||
9bb00b28 YS |
11317 | sleepable = is_kfunc_sleepable(&meta); |
11318 | if (sleepable && !env->prog->aux->sleepable) { | |
00b85860 KKD |
11319 | verbose(env, "program must be sleepable to call sleepable kfunc %s\n", func_name); |
11320 | return -EACCES; | |
11321 | } | |
eb1f7f71 | 11322 | |
9bb00b28 YS |
11323 | rcu_lock = is_kfunc_bpf_rcu_read_lock(&meta); |
11324 | rcu_unlock = is_kfunc_bpf_rcu_read_unlock(&meta); | |
9bb00b28 YS |
11325 | |
11326 | if (env->cur_state->active_rcu_lock) { | |
11327 | struct bpf_func_state *state; | |
11328 | struct bpf_reg_state *reg; | |
11329 | ||
0816b8c6 DM |
11330 | if (in_rbtree_lock_required_cb(env) && (rcu_lock || rcu_unlock)) { |
11331 | verbose(env, "Calling bpf_rcu_read_{lock,unlock} in unnecessary rbtree callback\n"); | |
11332 | return -EACCES; | |
11333 | } | |
11334 | ||
9bb00b28 YS |
11335 | if (rcu_lock) { |
11336 | verbose(env, "nested rcu read lock (kernel function %s)\n", func_name); | |
11337 | return -EINVAL; | |
11338 | } else if (rcu_unlock) { | |
11339 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ | |
11340 | if (reg->type & MEM_RCU) { | |
fca1aa75 | 11341 | reg->type &= ~(MEM_RCU | PTR_MAYBE_NULL); |
9bb00b28 YS |
11342 | reg->type |= PTR_UNTRUSTED; |
11343 | } | |
11344 | })); | |
11345 | env->cur_state->active_rcu_lock = false; | |
11346 | } else if (sleepable) { | |
11347 | verbose(env, "kernel func %s is sleepable within rcu_read_lock region\n", func_name); | |
11348 | return -EACCES; | |
11349 | } | |
11350 | } else if (rcu_lock) { | |
11351 | env->cur_state->active_rcu_lock = true; | |
11352 | } else if (rcu_unlock) { | |
11353 | verbose(env, "unmatched rcu read unlock (kernel function %s)\n", func_name); | |
11354 | return -EINVAL; | |
11355 | } | |
11356 | ||
e6ac2450 | 11357 | /* Check the arguments */ |
1d18feb2 | 11358 | err = check_kfunc_args(env, &meta, insn_idx); |
5c073f26 | 11359 | if (err < 0) |
e6ac2450 | 11360 | return err; |
5c073f26 | 11361 | /* In case of release function, we get register number of refcounted |
00b85860 | 11362 | * PTR_TO_BTF_ID in bpf_kfunc_arg_meta, do the release now. |
5c073f26 | 11363 | */ |
00b85860 KKD |
11364 | if (meta.release_regno) { |
11365 | err = release_reference(env, regs[meta.release_regno].ref_obj_id); | |
5c073f26 KKD |
11366 | if (err) { |
11367 | verbose(env, "kfunc %s#%d reference has not been acquired before\n", | |
07236eab | 11368 | func_name, meta.func_id); |
5c073f26 KKD |
11369 | return err; |
11370 | } | |
11371 | } | |
e6ac2450 | 11372 | |
d2dcc67d DM |
11373 | if (meta.func_id == special_kfunc_list[KF_bpf_list_push_front_impl] || |
11374 | meta.func_id == special_kfunc_list[KF_bpf_list_push_back_impl] || | |
11375 | meta.func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
6a3cd331 | 11376 | release_ref_obj_id = regs[BPF_REG_2].ref_obj_id; |
d2dcc67d | 11377 | insn_aux->insert_off = regs[BPF_REG_2].off; |
2140a6e3 | 11378 | insn_aux->kptr_struct_meta = btf_find_struct_meta(meta.arg_btf, meta.arg_btf_id); |
6a3cd331 DM |
11379 | err = ref_convert_owning_non_owning(env, release_ref_obj_id); |
11380 | if (err) { | |
11381 | verbose(env, "kfunc %s#%d conversion of owning ref to non-owning failed\n", | |
07236eab | 11382 | func_name, meta.func_id); |
6a3cd331 DM |
11383 | return err; |
11384 | } | |
11385 | ||
11386 | err = release_reference(env, release_ref_obj_id); | |
11387 | if (err) { | |
11388 | verbose(env, "kfunc %s#%d reference has not been acquired before\n", | |
07236eab | 11389 | func_name, meta.func_id); |
6a3cd331 DM |
11390 | return err; |
11391 | } | |
11392 | } | |
11393 | ||
d2dcc67d | 11394 | if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { |
5d92ddc3 DM |
11395 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
11396 | set_rbtree_add_callback_state); | |
11397 | if (err) { | |
11398 | verbose(env, "kfunc %s#%d failed callback verification\n", | |
07236eab | 11399 | func_name, meta.func_id); |
5d92ddc3 DM |
11400 | return err; |
11401 | } | |
11402 | } | |
11403 | ||
e6ac2450 MKL |
11404 | for (i = 0; i < CALLER_SAVED_REGS; i++) |
11405 | mark_reg_not_init(env, regs, caller_saved[i]); | |
11406 | ||
11407 | /* Check return type */ | |
07236eab | 11408 | t = btf_type_skip_modifiers(desc_btf, meta.func_proto->type, NULL); |
5c073f26 | 11409 | |
00b85860 | 11410 | if (is_kfunc_acquire(&meta) && !btf_type_is_struct_ptr(meta.btf, t)) { |
958cf2e2 | 11411 | /* Only exception is bpf_obj_new_impl */ |
7c50b1cb DM |
11412 | if (meta.btf != btf_vmlinux || |
11413 | (meta.func_id != special_kfunc_list[KF_bpf_obj_new_impl] && | |
11414 | meta.func_id != special_kfunc_list[KF_bpf_refcount_acquire_impl])) { | |
958cf2e2 KKD |
11415 | verbose(env, "acquire kernel function does not return PTR_TO_BTF_ID\n"); |
11416 | return -EINVAL; | |
11417 | } | |
5c073f26 KKD |
11418 | } |
11419 | ||
e6ac2450 MKL |
11420 | if (btf_type_is_scalar(t)) { |
11421 | mark_reg_unknown(env, regs, BPF_REG_0); | |
11422 | mark_btf_func_reg_size(env, BPF_REG_0, t->size); | |
11423 | } else if (btf_type_is_ptr(t)) { | |
958cf2e2 KKD |
11424 | ptr_type = btf_type_skip_modifiers(desc_btf, t->type, &ptr_type_id); |
11425 | ||
11426 | if (meta.btf == btf_vmlinux && btf_id_set_contains(&special_kfunc_set, meta.func_id)) { | |
11427 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_new_impl]) { | |
958cf2e2 KKD |
11428 | struct btf *ret_btf; |
11429 | u32 ret_btf_id; | |
11430 | ||
e181d3f1 KKD |
11431 | if (unlikely(!bpf_global_ma_set)) |
11432 | return -ENOMEM; | |
11433 | ||
958cf2e2 KKD |
11434 | if (((u64)(u32)meta.arg_constant.value) != meta.arg_constant.value) { |
11435 | verbose(env, "local type ID argument must be in range [0, U32_MAX]\n"); | |
11436 | return -EINVAL; | |
11437 | } | |
11438 | ||
11439 | ret_btf = env->prog->aux->btf; | |
11440 | ret_btf_id = meta.arg_constant.value; | |
11441 | ||
11442 | /* This may be NULL due to user not supplying a BTF */ | |
11443 | if (!ret_btf) { | |
11444 | verbose(env, "bpf_obj_new requires prog BTF\n"); | |
11445 | return -EINVAL; | |
11446 | } | |
11447 | ||
11448 | ret_t = btf_type_by_id(ret_btf, ret_btf_id); | |
11449 | if (!ret_t || !__btf_type_is_struct(ret_t)) { | |
11450 | verbose(env, "bpf_obj_new type ID argument must be of a struct\n"); | |
11451 | return -EINVAL; | |
11452 | } | |
11453 | ||
11454 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11455 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
11456 | regs[BPF_REG_0].btf = ret_btf; | |
11457 | regs[BPF_REG_0].btf_id = ret_btf_id; | |
11458 | ||
07236eab AN |
11459 | insn_aux->obj_new_size = ret_t->size; |
11460 | insn_aux->kptr_struct_meta = | |
958cf2e2 | 11461 | btf_find_struct_meta(ret_btf, ret_btf_id); |
7c50b1cb DM |
11462 | } else if (meta.func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]) { |
11463 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11464 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
4d585f48 DM |
11465 | regs[BPF_REG_0].btf = meta.arg_btf; |
11466 | regs[BPF_REG_0].btf_id = meta.arg_btf_id; | |
7c50b1cb DM |
11467 | |
11468 | insn_aux->kptr_struct_meta = | |
4d585f48 DM |
11469 | btf_find_struct_meta(meta.arg_btf, |
11470 | meta.arg_btf_id); | |
8cab76ec KKD |
11471 | } else if (meta.func_id == special_kfunc_list[KF_bpf_list_pop_front] || |
11472 | meta.func_id == special_kfunc_list[KF_bpf_list_pop_back]) { | |
11473 | struct btf_field *field = meta.arg_list_head.field; | |
11474 | ||
a40d3632 DM |
11475 | mark_reg_graph_node(regs, BPF_REG_0, &field->graph_root); |
11476 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_remove] || | |
11477 | meta.func_id == special_kfunc_list[KF_bpf_rbtree_first]) { | |
11478 | struct btf_field *field = meta.arg_rbtree_root.field; | |
11479 | ||
11480 | mark_reg_graph_node(regs, BPF_REG_0, &field->graph_root); | |
fd264ca0 YS |
11481 | } else if (meta.func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { |
11482 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11483 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_TRUSTED; | |
11484 | regs[BPF_REG_0].btf = desc_btf; | |
11485 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; | |
a35b9af4 YS |
11486 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { |
11487 | ret_t = btf_type_by_id(desc_btf, meta.arg_constant.value); | |
11488 | if (!ret_t || !btf_type_is_struct(ret_t)) { | |
11489 | verbose(env, | |
11490 | "kfunc bpf_rdonly_cast type ID argument must be of a struct\n"); | |
11491 | return -EINVAL; | |
11492 | } | |
11493 | ||
11494 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11495 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_UNTRUSTED; | |
11496 | regs[BPF_REG_0].btf = desc_btf; | |
11497 | regs[BPF_REG_0].btf_id = meta.arg_constant.value; | |
66e3a13e JK |
11498 | } else if (meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice] || |
11499 | meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice_rdwr]) { | |
11500 | enum bpf_type_flag type_flag = get_dynptr_type_flag(meta.initialized_dynptr.type); | |
11501 | ||
11502 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11503 | ||
11504 | if (!meta.arg_constant.found) { | |
11505 | verbose(env, "verifier internal error: bpf_dynptr_slice(_rdwr) no constant size\n"); | |
11506 | return -EFAULT; | |
11507 | } | |
11508 | ||
11509 | regs[BPF_REG_0].mem_size = meta.arg_constant.value; | |
11510 | ||
11511 | /* PTR_MAYBE_NULL will be added when is_kfunc_ret_null is checked */ | |
11512 | regs[BPF_REG_0].type = PTR_TO_MEM | type_flag; | |
11513 | ||
11514 | if (meta.func_id == special_kfunc_list[KF_bpf_dynptr_slice]) { | |
11515 | regs[BPF_REG_0].type |= MEM_RDONLY; | |
11516 | } else { | |
11517 | /* this will set env->seen_direct_write to true */ | |
11518 | if (!may_access_direct_pkt_data(env, NULL, BPF_WRITE)) { | |
11519 | verbose(env, "the prog does not allow writes to packet data\n"); | |
11520 | return -EINVAL; | |
11521 | } | |
11522 | } | |
11523 | ||
11524 | if (!meta.initialized_dynptr.id) { | |
11525 | verbose(env, "verifier internal error: no dynptr id\n"); | |
11526 | return -EFAULT; | |
11527 | } | |
11528 | regs[BPF_REG_0].dynptr_id = meta.initialized_dynptr.id; | |
11529 | ||
11530 | /* we don't need to set BPF_REG_0's ref obj id | |
11531 | * because packet slices are not refcounted (see | |
11532 | * dynptr_type_refcounted) | |
11533 | */ | |
958cf2e2 KKD |
11534 | } else { |
11535 | verbose(env, "kernel function %s unhandled dynamic return type\n", | |
11536 | meta.func_name); | |
11537 | return -EFAULT; | |
11538 | } | |
11539 | } else if (!__btf_type_is_struct(ptr_type)) { | |
f4b4eee6 AN |
11540 | if (!meta.r0_size) { |
11541 | __u32 sz; | |
11542 | ||
11543 | if (!IS_ERR(btf_resolve_size(desc_btf, ptr_type, &sz))) { | |
11544 | meta.r0_size = sz; | |
11545 | meta.r0_rdonly = true; | |
11546 | } | |
11547 | } | |
eb1f7f71 BT |
11548 | if (!meta.r0_size) { |
11549 | ptr_type_name = btf_name_by_offset(desc_btf, | |
11550 | ptr_type->name_off); | |
11551 | verbose(env, | |
11552 | "kernel function %s returns pointer type %s %s is not supported\n", | |
11553 | func_name, | |
11554 | btf_type_str(ptr_type), | |
11555 | ptr_type_name); | |
11556 | return -EINVAL; | |
11557 | } | |
11558 | ||
11559 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11560 | regs[BPF_REG_0].type = PTR_TO_MEM; | |
11561 | regs[BPF_REG_0].mem_size = meta.r0_size; | |
11562 | ||
11563 | if (meta.r0_rdonly) | |
11564 | regs[BPF_REG_0].type |= MEM_RDONLY; | |
11565 | ||
11566 | /* Ensures we don't access the memory after a release_reference() */ | |
11567 | if (meta.ref_obj_id) | |
11568 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
11569 | } else { | |
11570 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
11571 | regs[BPF_REG_0].btf = desc_btf; | |
11572 | regs[BPF_REG_0].type = PTR_TO_BTF_ID; | |
11573 | regs[BPF_REG_0].btf_id = ptr_type_id; | |
e6ac2450 | 11574 | } |
958cf2e2 | 11575 | |
00b85860 | 11576 | if (is_kfunc_ret_null(&meta)) { |
5c073f26 KKD |
11577 | regs[BPF_REG_0].type |= PTR_MAYBE_NULL; |
11578 | /* For mark_ptr_or_null_reg, see 93c230e3f5bd6 */ | |
11579 | regs[BPF_REG_0].id = ++env->id_gen; | |
11580 | } | |
e6ac2450 | 11581 | mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *)); |
00b85860 | 11582 | if (is_kfunc_acquire(&meta)) { |
5c073f26 KKD |
11583 | int id = acquire_reference_state(env, insn_idx); |
11584 | ||
11585 | if (id < 0) | |
11586 | return id; | |
00b85860 KKD |
11587 | if (is_kfunc_ret_null(&meta)) |
11588 | regs[BPF_REG_0].id = id; | |
5c073f26 | 11589 | regs[BPF_REG_0].ref_obj_id = id; |
a40d3632 DM |
11590 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rbtree_first]) { |
11591 | ref_set_non_owning(env, ®s[BPF_REG_0]); | |
5c073f26 | 11592 | } |
a40d3632 | 11593 | |
00b85860 KKD |
11594 | if (reg_may_point_to_spin_lock(®s[BPF_REG_0]) && !regs[BPF_REG_0].id) |
11595 | regs[BPF_REG_0].id = ++env->id_gen; | |
f6a6a5a9 DM |
11596 | } else if (btf_type_is_void(t)) { |
11597 | if (meta.btf == btf_vmlinux && btf_id_set_contains(&special_kfunc_set, meta.func_id)) { | |
11598 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_drop_impl]) { | |
11599 | insn_aux->kptr_struct_meta = | |
4d585f48 DM |
11600 | btf_find_struct_meta(meta.arg_btf, |
11601 | meta.arg_btf_id); | |
f6a6a5a9 DM |
11602 | } |
11603 | } | |
11604 | } | |
e6ac2450 | 11605 | |
07236eab AN |
11606 | nargs = btf_type_vlen(meta.func_proto); |
11607 | args = (const struct btf_param *)(meta.func_proto + 1); | |
e6ac2450 MKL |
11608 | for (i = 0; i < nargs; i++) { |
11609 | u32 regno = i + 1; | |
11610 | ||
2357672c | 11611 | t = btf_type_skip_modifiers(desc_btf, args[i].type, NULL); |
e6ac2450 MKL |
11612 | if (btf_type_is_ptr(t)) |
11613 | mark_btf_func_reg_size(env, regno, sizeof(void *)); | |
11614 | else | |
11615 | /* scalar. ensured by btf_check_kfunc_arg_match() */ | |
11616 | mark_btf_func_reg_size(env, regno, t->size); | |
11617 | } | |
11618 | ||
06accc87 AN |
11619 | if (is_iter_next_kfunc(&meta)) { |
11620 | err = process_iter_next_call(env, insn_idx, &meta); | |
11621 | if (err) | |
11622 | return err; | |
11623 | } | |
11624 | ||
e6ac2450 MKL |
11625 | return 0; |
11626 | } | |
11627 | ||
b03c9f9f EC |
11628 | static bool signed_add_overflows(s64 a, s64 b) |
11629 | { | |
11630 | /* Do the add in u64, where overflow is well-defined */ | |
11631 | s64 res = (s64)((u64)a + (u64)b); | |
11632 | ||
11633 | if (b < 0) | |
11634 | return res > a; | |
11635 | return res < a; | |
11636 | } | |
11637 | ||
bc895e8b | 11638 | static bool signed_add32_overflows(s32 a, s32 b) |
3f50f132 JF |
11639 | { |
11640 | /* Do the add in u32, where overflow is well-defined */ | |
11641 | s32 res = (s32)((u32)a + (u32)b); | |
11642 | ||
11643 | if (b < 0) | |
11644 | return res > a; | |
11645 | return res < a; | |
11646 | } | |
11647 | ||
bc895e8b | 11648 | static bool signed_sub_overflows(s64 a, s64 b) |
b03c9f9f EC |
11649 | { |
11650 | /* Do the sub in u64, where overflow is well-defined */ | |
11651 | s64 res = (s64)((u64)a - (u64)b); | |
11652 | ||
11653 | if (b < 0) | |
11654 | return res < a; | |
11655 | return res > a; | |
969bf05e AS |
11656 | } |
11657 | ||
3f50f132 JF |
11658 | static bool signed_sub32_overflows(s32 a, s32 b) |
11659 | { | |
bc895e8b | 11660 | /* Do the sub in u32, where overflow is well-defined */ |
3f50f132 JF |
11661 | s32 res = (s32)((u32)a - (u32)b); |
11662 | ||
11663 | if (b < 0) | |
11664 | return res < a; | |
11665 | return res > a; | |
11666 | } | |
11667 | ||
bb7f0f98 AS |
11668 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
11669 | const struct bpf_reg_state *reg, | |
11670 | enum bpf_reg_type type) | |
11671 | { | |
11672 | bool known = tnum_is_const(reg->var_off); | |
11673 | s64 val = reg->var_off.value; | |
11674 | s64 smin = reg->smin_value; | |
11675 | ||
11676 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
11677 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
c25b2ae1 | 11678 | reg_type_str(env, type), val); |
bb7f0f98 AS |
11679 | return false; |
11680 | } | |
11681 | ||
11682 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
11683 | verbose(env, "%s pointer offset %d is not allowed\n", | |
c25b2ae1 | 11684 | reg_type_str(env, type), reg->off); |
bb7f0f98 AS |
11685 | return false; |
11686 | } | |
11687 | ||
11688 | if (smin == S64_MIN) { | |
11689 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
c25b2ae1 | 11690 | reg_type_str(env, type)); |
bb7f0f98 AS |
11691 | return false; |
11692 | } | |
11693 | ||
11694 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
11695 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
c25b2ae1 | 11696 | smin, reg_type_str(env, type)); |
bb7f0f98 AS |
11697 | return false; |
11698 | } | |
11699 | ||
11700 | return true; | |
11701 | } | |
11702 | ||
a6aaece0 DB |
11703 | enum { |
11704 | REASON_BOUNDS = -1, | |
11705 | REASON_TYPE = -2, | |
11706 | REASON_PATHS = -3, | |
11707 | REASON_LIMIT = -4, | |
11708 | REASON_STACK = -5, | |
11709 | }; | |
11710 | ||
979d63d5 | 11711 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, |
bb01a1bb | 11712 | u32 *alu_limit, bool mask_to_left) |
979d63d5 | 11713 | { |
7fedb63a | 11714 | u32 max = 0, ptr_limit = 0; |
979d63d5 DB |
11715 | |
11716 | switch (ptr_reg->type) { | |
11717 | case PTR_TO_STACK: | |
1b1597e6 | 11718 | /* Offset 0 is out-of-bounds, but acceptable start for the |
7fedb63a DB |
11719 | * left direction, see BPF_REG_FP. Also, unknown scalar |
11720 | * offset where we would need to deal with min/max bounds is | |
11721 | * currently prohibited for unprivileged. | |
1b1597e6 PK |
11722 | */ |
11723 | max = MAX_BPF_STACK + mask_to_left; | |
7fedb63a | 11724 | ptr_limit = -(ptr_reg->var_off.value + ptr_reg->off); |
b658bbb8 | 11725 | break; |
979d63d5 | 11726 | case PTR_TO_MAP_VALUE: |
1b1597e6 | 11727 | max = ptr_reg->map_ptr->value_size; |
7fedb63a DB |
11728 | ptr_limit = (mask_to_left ? |
11729 | ptr_reg->smin_value : | |
11730 | ptr_reg->umax_value) + ptr_reg->off; | |
b658bbb8 | 11731 | break; |
979d63d5 | 11732 | default: |
a6aaece0 | 11733 | return REASON_TYPE; |
979d63d5 | 11734 | } |
b658bbb8 DB |
11735 | |
11736 | if (ptr_limit >= max) | |
a6aaece0 | 11737 | return REASON_LIMIT; |
b658bbb8 DB |
11738 | *alu_limit = ptr_limit; |
11739 | return 0; | |
979d63d5 DB |
11740 | } |
11741 | ||
d3bd7413 DB |
11742 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
11743 | const struct bpf_insn *insn) | |
11744 | { | |
2c78ee89 | 11745 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
11746 | } |
11747 | ||
11748 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
11749 | u32 alu_state, u32 alu_limit) | |
11750 | { | |
11751 | /* If we arrived here from different branches with different | |
11752 | * state or limits to sanitize, then this won't work. | |
11753 | */ | |
11754 | if (aux->alu_state && | |
11755 | (aux->alu_state != alu_state || | |
11756 | aux->alu_limit != alu_limit)) | |
a6aaece0 | 11757 | return REASON_PATHS; |
d3bd7413 | 11758 | |
e6ac5933 | 11759 | /* Corresponding fixup done in do_misc_fixups(). */ |
d3bd7413 DB |
11760 | aux->alu_state = alu_state; |
11761 | aux->alu_limit = alu_limit; | |
11762 | return 0; | |
11763 | } | |
11764 | ||
11765 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
11766 | struct bpf_insn *insn) | |
11767 | { | |
11768 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
11769 | ||
11770 | if (can_skip_alu_sanitation(env, insn)) | |
11771 | return 0; | |
11772 | ||
11773 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
11774 | } | |
11775 | ||
f5288193 DB |
11776 | static bool sanitize_needed(u8 opcode) |
11777 | { | |
11778 | return opcode == BPF_ADD || opcode == BPF_SUB; | |
11779 | } | |
11780 | ||
3d0220f6 DB |
11781 | struct bpf_sanitize_info { |
11782 | struct bpf_insn_aux_data aux; | |
bb01a1bb | 11783 | bool mask_to_left; |
3d0220f6 DB |
11784 | }; |
11785 | ||
9183671a DB |
11786 | static struct bpf_verifier_state * |
11787 | sanitize_speculative_path(struct bpf_verifier_env *env, | |
11788 | const struct bpf_insn *insn, | |
11789 | u32 next_idx, u32 curr_idx) | |
11790 | { | |
11791 | struct bpf_verifier_state *branch; | |
11792 | struct bpf_reg_state *regs; | |
11793 | ||
11794 | branch = push_stack(env, next_idx, curr_idx, true); | |
11795 | if (branch && insn) { | |
11796 | regs = branch->frame[branch->curframe]->regs; | |
11797 | if (BPF_SRC(insn->code) == BPF_K) { | |
11798 | mark_reg_unknown(env, regs, insn->dst_reg); | |
11799 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
11800 | mark_reg_unknown(env, regs, insn->dst_reg); | |
11801 | mark_reg_unknown(env, regs, insn->src_reg); | |
11802 | } | |
11803 | } | |
11804 | return branch; | |
11805 | } | |
11806 | ||
979d63d5 DB |
11807 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
11808 | struct bpf_insn *insn, | |
11809 | const struct bpf_reg_state *ptr_reg, | |
6f55b2f2 | 11810 | const struct bpf_reg_state *off_reg, |
979d63d5 | 11811 | struct bpf_reg_state *dst_reg, |
3d0220f6 | 11812 | struct bpf_sanitize_info *info, |
7fedb63a | 11813 | const bool commit_window) |
979d63d5 | 11814 | { |
3d0220f6 | 11815 | struct bpf_insn_aux_data *aux = commit_window ? cur_aux(env) : &info->aux; |
979d63d5 | 11816 | struct bpf_verifier_state *vstate = env->cur_state; |
801c6058 | 11817 | bool off_is_imm = tnum_is_const(off_reg->var_off); |
6f55b2f2 | 11818 | bool off_is_neg = off_reg->smin_value < 0; |
979d63d5 DB |
11819 | bool ptr_is_dst_reg = ptr_reg == dst_reg; |
11820 | u8 opcode = BPF_OP(insn->code); | |
11821 | u32 alu_state, alu_limit; | |
11822 | struct bpf_reg_state tmp; | |
11823 | bool ret; | |
f232326f | 11824 | int err; |
979d63d5 | 11825 | |
d3bd7413 | 11826 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
11827 | return 0; |
11828 | ||
11829 | /* We already marked aux for masking from non-speculative | |
11830 | * paths, thus we got here in the first place. We only care | |
11831 | * to explore bad access from here. | |
11832 | */ | |
11833 | if (vstate->speculative) | |
11834 | goto do_sim; | |
11835 | ||
bb01a1bb DB |
11836 | if (!commit_window) { |
11837 | if (!tnum_is_const(off_reg->var_off) && | |
11838 | (off_reg->smin_value < 0) != (off_reg->smax_value < 0)) | |
11839 | return REASON_BOUNDS; | |
11840 | ||
11841 | info->mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
11842 | (opcode == BPF_SUB && !off_is_neg); | |
11843 | } | |
11844 | ||
11845 | err = retrieve_ptr_limit(ptr_reg, &alu_limit, info->mask_to_left); | |
f232326f PK |
11846 | if (err < 0) |
11847 | return err; | |
11848 | ||
7fedb63a DB |
11849 | if (commit_window) { |
11850 | /* In commit phase we narrow the masking window based on | |
11851 | * the observed pointer move after the simulated operation. | |
11852 | */ | |
3d0220f6 DB |
11853 | alu_state = info->aux.alu_state; |
11854 | alu_limit = abs(info->aux.alu_limit - alu_limit); | |
7fedb63a DB |
11855 | } else { |
11856 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
801c6058 | 11857 | alu_state |= off_is_imm ? BPF_ALU_IMMEDIATE : 0; |
7fedb63a DB |
11858 | alu_state |= ptr_is_dst_reg ? |
11859 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
e042aa53 DB |
11860 | |
11861 | /* Limit pruning on unknown scalars to enable deep search for | |
11862 | * potential masking differences from other program paths. | |
11863 | */ | |
11864 | if (!off_is_imm) | |
11865 | env->explore_alu_limits = true; | |
7fedb63a DB |
11866 | } |
11867 | ||
f232326f PK |
11868 | err = update_alu_sanitation_state(aux, alu_state, alu_limit); |
11869 | if (err < 0) | |
11870 | return err; | |
979d63d5 | 11871 | do_sim: |
7fedb63a DB |
11872 | /* If we're in commit phase, we're done here given we already |
11873 | * pushed the truncated dst_reg into the speculative verification | |
11874 | * stack. | |
a7036191 DB |
11875 | * |
11876 | * Also, when register is a known constant, we rewrite register-based | |
11877 | * operation to immediate-based, and thus do not need masking (and as | |
11878 | * a consequence, do not need to simulate the zero-truncation either). | |
7fedb63a | 11879 | */ |
a7036191 | 11880 | if (commit_window || off_is_imm) |
7fedb63a DB |
11881 | return 0; |
11882 | ||
979d63d5 DB |
11883 | /* Simulate and find potential out-of-bounds access under |
11884 | * speculative execution from truncation as a result of | |
11885 | * masking when off was not within expected range. If off | |
11886 | * sits in dst, then we temporarily need to move ptr there | |
11887 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
11888 | * for cases where we use K-based arithmetic in one direction | |
11889 | * and truncated reg-based in the other in order to explore | |
11890 | * bad access. | |
11891 | */ | |
11892 | if (!ptr_is_dst_reg) { | |
11893 | tmp = *dst_reg; | |
71f656a5 | 11894 | copy_register_state(dst_reg, ptr_reg); |
979d63d5 | 11895 | } |
9183671a DB |
11896 | ret = sanitize_speculative_path(env, NULL, env->insn_idx + 1, |
11897 | env->insn_idx); | |
0803278b | 11898 | if (!ptr_is_dst_reg && ret) |
979d63d5 | 11899 | *dst_reg = tmp; |
a6aaece0 DB |
11900 | return !ret ? REASON_STACK : 0; |
11901 | } | |
11902 | ||
fe9a5ca7 DB |
11903 | static void sanitize_mark_insn_seen(struct bpf_verifier_env *env) |
11904 | { | |
11905 | struct bpf_verifier_state *vstate = env->cur_state; | |
11906 | ||
11907 | /* If we simulate paths under speculation, we don't update the | |
11908 | * insn as 'seen' such that when we verify unreachable paths in | |
11909 | * the non-speculative domain, sanitize_dead_code() can still | |
11910 | * rewrite/sanitize them. | |
11911 | */ | |
11912 | if (!vstate->speculative) | |
11913 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; | |
11914 | } | |
11915 | ||
a6aaece0 DB |
11916 | static int sanitize_err(struct bpf_verifier_env *env, |
11917 | const struct bpf_insn *insn, int reason, | |
11918 | const struct bpf_reg_state *off_reg, | |
11919 | const struct bpf_reg_state *dst_reg) | |
11920 | { | |
11921 | static const char *err = "pointer arithmetic with it prohibited for !root"; | |
11922 | const char *op = BPF_OP(insn->code) == BPF_ADD ? "add" : "sub"; | |
11923 | u32 dst = insn->dst_reg, src = insn->src_reg; | |
11924 | ||
11925 | switch (reason) { | |
11926 | case REASON_BOUNDS: | |
11927 | verbose(env, "R%d has unknown scalar with mixed signed bounds, %s\n", | |
11928 | off_reg == dst_reg ? dst : src, err); | |
11929 | break; | |
11930 | case REASON_TYPE: | |
11931 | verbose(env, "R%d has pointer with unsupported alu operation, %s\n", | |
11932 | off_reg == dst_reg ? src : dst, err); | |
11933 | break; | |
11934 | case REASON_PATHS: | |
11935 | verbose(env, "R%d tried to %s from different maps, paths or scalars, %s\n", | |
11936 | dst, op, err); | |
11937 | break; | |
11938 | case REASON_LIMIT: | |
11939 | verbose(env, "R%d tried to %s beyond pointer bounds, %s\n", | |
11940 | dst, op, err); | |
11941 | break; | |
11942 | case REASON_STACK: | |
11943 | verbose(env, "R%d could not be pushed for speculative verification, %s\n", | |
11944 | dst, err); | |
11945 | break; | |
11946 | default: | |
11947 | verbose(env, "verifier internal error: unknown reason (%d)\n", | |
11948 | reason); | |
11949 | break; | |
11950 | } | |
11951 | ||
11952 | return -EACCES; | |
979d63d5 DB |
11953 | } |
11954 | ||
01f810ac AM |
11955 | /* check that stack access falls within stack limits and that 'reg' doesn't |
11956 | * have a variable offset. | |
11957 | * | |
11958 | * Variable offset is prohibited for unprivileged mode for simplicity since it | |
11959 | * requires corresponding support in Spectre masking for stack ALU. See also | |
11960 | * retrieve_ptr_limit(). | |
11961 | * | |
11962 | * | |
11963 | * 'off' includes 'reg->off'. | |
11964 | */ | |
11965 | static int check_stack_access_for_ptr_arithmetic( | |
11966 | struct bpf_verifier_env *env, | |
11967 | int regno, | |
11968 | const struct bpf_reg_state *reg, | |
11969 | int off) | |
11970 | { | |
11971 | if (!tnum_is_const(reg->var_off)) { | |
11972 | char tn_buf[48]; | |
11973 | ||
11974 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
11975 | verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n", | |
11976 | regno, tn_buf, off); | |
11977 | return -EACCES; | |
11978 | } | |
11979 | ||
11980 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
11981 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
11982 | "prohibited for !root; off=%d\n", regno, off); | |
11983 | return -EACCES; | |
11984 | } | |
11985 | ||
11986 | return 0; | |
11987 | } | |
11988 | ||
073815b7 DB |
11989 | static int sanitize_check_bounds(struct bpf_verifier_env *env, |
11990 | const struct bpf_insn *insn, | |
11991 | const struct bpf_reg_state *dst_reg) | |
11992 | { | |
11993 | u32 dst = insn->dst_reg; | |
11994 | ||
11995 | /* For unprivileged we require that resulting offset must be in bounds | |
11996 | * in order to be able to sanitize access later on. | |
11997 | */ | |
11998 | if (env->bypass_spec_v1) | |
11999 | return 0; | |
12000 | ||
12001 | switch (dst_reg->type) { | |
12002 | case PTR_TO_STACK: | |
12003 | if (check_stack_access_for_ptr_arithmetic(env, dst, dst_reg, | |
12004 | dst_reg->off + dst_reg->var_off.value)) | |
12005 | return -EACCES; | |
12006 | break; | |
12007 | case PTR_TO_MAP_VALUE: | |
61df10c7 | 12008 | if (check_map_access(env, dst, dst_reg->off, 1, false, ACCESS_HELPER)) { |
073815b7 DB |
12009 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " |
12010 | "prohibited for !root\n", dst); | |
12011 | return -EACCES; | |
12012 | } | |
12013 | break; | |
12014 | default: | |
12015 | break; | |
12016 | } | |
12017 | ||
12018 | return 0; | |
12019 | } | |
01f810ac | 12020 | |
f1174f77 | 12021 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
12022 | * Caller should also handle BPF_MOV case separately. |
12023 | * If we return -EACCES, caller may want to try again treating pointer as a | |
12024 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
12025 | */ | |
12026 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
12027 | struct bpf_insn *insn, | |
12028 | const struct bpf_reg_state *ptr_reg, | |
12029 | const struct bpf_reg_state *off_reg) | |
969bf05e | 12030 | { |
f4d7e40a AS |
12031 | struct bpf_verifier_state *vstate = env->cur_state; |
12032 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
12033 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 12034 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
12035 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
12036 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
12037 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
12038 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
3d0220f6 | 12039 | struct bpf_sanitize_info info = {}; |
969bf05e | 12040 | u8 opcode = BPF_OP(insn->code); |
24c109bb | 12041 | u32 dst = insn->dst_reg; |
979d63d5 | 12042 | int ret; |
969bf05e | 12043 | |
f1174f77 | 12044 | dst_reg = ®s[dst]; |
969bf05e | 12045 | |
6f16101e DB |
12046 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
12047 | smin_val > smax_val || umin_val > umax_val) { | |
12048 | /* Taint dst register if offset had invalid bounds derived from | |
12049 | * e.g. dead branches. | |
12050 | */ | |
f54c7898 | 12051 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 12052 | return 0; |
f1174f77 EC |
12053 | } |
12054 | ||
12055 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
12056 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
6c693541 YS |
12057 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
12058 | __mark_reg_unknown(env, dst_reg); | |
12059 | return 0; | |
12060 | } | |
12061 | ||
82abbf8d AS |
12062 | verbose(env, |
12063 | "R%d 32-bit pointer arithmetic prohibited\n", | |
12064 | dst); | |
f1174f77 | 12065 | return -EACCES; |
969bf05e AS |
12066 | } |
12067 | ||
c25b2ae1 | 12068 | if (ptr_reg->type & PTR_MAYBE_NULL) { |
aad2eeaf | 12069 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", |
c25b2ae1 | 12070 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 12071 | return -EACCES; |
c25b2ae1 HL |
12072 | } |
12073 | ||
12074 | switch (base_type(ptr_reg->type)) { | |
aad2eeaf | 12075 | case CONST_PTR_TO_MAP: |
7c696732 YS |
12076 | /* smin_val represents the known value */ |
12077 | if (known && smin_val == 0 && opcode == BPF_ADD) | |
12078 | break; | |
8731745e | 12079 | fallthrough; |
aad2eeaf | 12080 | case PTR_TO_PACKET_END: |
c64b7983 | 12081 | case PTR_TO_SOCKET: |
46f8bc92 | 12082 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 12083 | case PTR_TO_TCP_SOCK: |
fada7fdc | 12084 | case PTR_TO_XDP_SOCK: |
aad2eeaf | 12085 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
c25b2ae1 | 12086 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 12087 | return -EACCES; |
aad2eeaf JS |
12088 | default: |
12089 | break; | |
f1174f77 EC |
12090 | } |
12091 | ||
12092 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
12093 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 12094 | */ |
f1174f77 EC |
12095 | dst_reg->type = ptr_reg->type; |
12096 | dst_reg->id = ptr_reg->id; | |
969bf05e | 12097 | |
bb7f0f98 AS |
12098 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
12099 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
12100 | return -EINVAL; | |
12101 | ||
3f50f132 JF |
12102 | /* pointer types do not carry 32-bit bounds at the moment. */ |
12103 | __mark_reg32_unbounded(dst_reg); | |
12104 | ||
7fedb63a DB |
12105 | if (sanitize_needed(opcode)) { |
12106 | ret = sanitize_ptr_alu(env, insn, ptr_reg, off_reg, dst_reg, | |
3d0220f6 | 12107 | &info, false); |
a6aaece0 DB |
12108 | if (ret < 0) |
12109 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
7fedb63a | 12110 | } |
a6aaece0 | 12111 | |
f1174f77 EC |
12112 | switch (opcode) { |
12113 | case BPF_ADD: | |
12114 | /* We can take a fixed offset as long as it doesn't overflow | |
12115 | * the s32 'off' field | |
969bf05e | 12116 | */ |
b03c9f9f EC |
12117 | if (known && (ptr_reg->off + smin_val == |
12118 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 12119 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
12120 | dst_reg->smin_value = smin_ptr; |
12121 | dst_reg->smax_value = smax_ptr; | |
12122 | dst_reg->umin_value = umin_ptr; | |
12123 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 12124 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 12125 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 12126 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
12127 | break; |
12128 | } | |
f1174f77 EC |
12129 | /* A new variable offset is created. Note that off_reg->off |
12130 | * == 0, since it's a scalar. | |
12131 | * dst_reg gets the pointer type and since some positive | |
12132 | * integer value was added to the pointer, give it a new 'id' | |
12133 | * if it's a PTR_TO_PACKET. | |
12134 | * this creates a new 'base' pointer, off_reg (variable) gets | |
12135 | * added into the variable offset, and we copy the fixed offset | |
12136 | * from ptr_reg. | |
969bf05e | 12137 | */ |
b03c9f9f EC |
12138 | if (signed_add_overflows(smin_ptr, smin_val) || |
12139 | signed_add_overflows(smax_ptr, smax_val)) { | |
12140 | dst_reg->smin_value = S64_MIN; | |
12141 | dst_reg->smax_value = S64_MAX; | |
12142 | } else { | |
12143 | dst_reg->smin_value = smin_ptr + smin_val; | |
12144 | dst_reg->smax_value = smax_ptr + smax_val; | |
12145 | } | |
12146 | if (umin_ptr + umin_val < umin_ptr || | |
12147 | umax_ptr + umax_val < umax_ptr) { | |
12148 | dst_reg->umin_value = 0; | |
12149 | dst_reg->umax_value = U64_MAX; | |
12150 | } else { | |
12151 | dst_reg->umin_value = umin_ptr + umin_val; | |
12152 | dst_reg->umax_value = umax_ptr + umax_val; | |
12153 | } | |
f1174f77 EC |
12154 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
12155 | dst_reg->off = ptr_reg->off; | |
0962590e | 12156 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 12157 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
12158 | dst_reg->id = ++env->id_gen; |
12159 | /* something was added to pkt_ptr, set range to zero */ | |
22dc4a0f | 12160 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
f1174f77 EC |
12161 | } |
12162 | break; | |
12163 | case BPF_SUB: | |
12164 | if (dst_reg == off_reg) { | |
12165 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
12166 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
12167 | dst); | |
f1174f77 EC |
12168 | return -EACCES; |
12169 | } | |
12170 | /* We don't allow subtraction from FP, because (according to | |
12171 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
12172 | * be able to deal with it. | |
969bf05e | 12173 | */ |
f1174f77 | 12174 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
12175 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
12176 | dst); | |
f1174f77 EC |
12177 | return -EACCES; |
12178 | } | |
b03c9f9f EC |
12179 | if (known && (ptr_reg->off - smin_val == |
12180 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 12181 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
12182 | dst_reg->smin_value = smin_ptr; |
12183 | dst_reg->smax_value = smax_ptr; | |
12184 | dst_reg->umin_value = umin_ptr; | |
12185 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
12186 | dst_reg->var_off = ptr_reg->var_off; |
12187 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 12188 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 12189 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
12190 | break; |
12191 | } | |
f1174f77 EC |
12192 | /* A new variable offset is created. If the subtrahend is known |
12193 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 12194 | */ |
b03c9f9f EC |
12195 | if (signed_sub_overflows(smin_ptr, smax_val) || |
12196 | signed_sub_overflows(smax_ptr, smin_val)) { | |
12197 | /* Overflow possible, we know nothing */ | |
12198 | dst_reg->smin_value = S64_MIN; | |
12199 | dst_reg->smax_value = S64_MAX; | |
12200 | } else { | |
12201 | dst_reg->smin_value = smin_ptr - smax_val; | |
12202 | dst_reg->smax_value = smax_ptr - smin_val; | |
12203 | } | |
12204 | if (umin_ptr < umax_val) { | |
12205 | /* Overflow possible, we know nothing */ | |
12206 | dst_reg->umin_value = 0; | |
12207 | dst_reg->umax_value = U64_MAX; | |
12208 | } else { | |
12209 | /* Cannot overflow (as long as bounds are consistent) */ | |
12210 | dst_reg->umin_value = umin_ptr - umax_val; | |
12211 | dst_reg->umax_value = umax_ptr - umin_val; | |
12212 | } | |
f1174f77 EC |
12213 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
12214 | dst_reg->off = ptr_reg->off; | |
0962590e | 12215 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 12216 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
12217 | dst_reg->id = ++env->id_gen; |
12218 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 12219 | if (smin_val < 0) |
22dc4a0f | 12220 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
43188702 | 12221 | } |
f1174f77 EC |
12222 | break; |
12223 | case BPF_AND: | |
12224 | case BPF_OR: | |
12225 | case BPF_XOR: | |
82abbf8d AS |
12226 | /* bitwise ops on pointers are troublesome, prohibit. */ |
12227 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
12228 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
12229 | return -EACCES; |
12230 | default: | |
12231 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
12232 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
12233 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 12234 | return -EACCES; |
43188702 JF |
12235 | } |
12236 | ||
bb7f0f98 AS |
12237 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
12238 | return -EINVAL; | |
3844d153 | 12239 | reg_bounds_sync(dst_reg); |
073815b7 DB |
12240 | if (sanitize_check_bounds(env, insn, dst_reg) < 0) |
12241 | return -EACCES; | |
7fedb63a DB |
12242 | if (sanitize_needed(opcode)) { |
12243 | ret = sanitize_ptr_alu(env, insn, dst_reg, off_reg, dst_reg, | |
3d0220f6 | 12244 | &info, true); |
7fedb63a DB |
12245 | if (ret < 0) |
12246 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
0d6303db DB |
12247 | } |
12248 | ||
43188702 JF |
12249 | return 0; |
12250 | } | |
12251 | ||
3f50f132 JF |
12252 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
12253 | struct bpf_reg_state *src_reg) | |
12254 | { | |
12255 | s32 smin_val = src_reg->s32_min_value; | |
12256 | s32 smax_val = src_reg->s32_max_value; | |
12257 | u32 umin_val = src_reg->u32_min_value; | |
12258 | u32 umax_val = src_reg->u32_max_value; | |
12259 | ||
12260 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
12261 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
12262 | dst_reg->s32_min_value = S32_MIN; | |
12263 | dst_reg->s32_max_value = S32_MAX; | |
12264 | } else { | |
12265 | dst_reg->s32_min_value += smin_val; | |
12266 | dst_reg->s32_max_value += smax_val; | |
12267 | } | |
12268 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
12269 | dst_reg->u32_max_value + umax_val < umax_val) { | |
12270 | dst_reg->u32_min_value = 0; | |
12271 | dst_reg->u32_max_value = U32_MAX; | |
12272 | } else { | |
12273 | dst_reg->u32_min_value += umin_val; | |
12274 | dst_reg->u32_max_value += umax_val; | |
12275 | } | |
12276 | } | |
12277 | ||
07cd2631 JF |
12278 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
12279 | struct bpf_reg_state *src_reg) | |
12280 | { | |
12281 | s64 smin_val = src_reg->smin_value; | |
12282 | s64 smax_val = src_reg->smax_value; | |
12283 | u64 umin_val = src_reg->umin_value; | |
12284 | u64 umax_val = src_reg->umax_value; | |
12285 | ||
12286 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
12287 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
12288 | dst_reg->smin_value = S64_MIN; | |
12289 | dst_reg->smax_value = S64_MAX; | |
12290 | } else { | |
12291 | dst_reg->smin_value += smin_val; | |
12292 | dst_reg->smax_value += smax_val; | |
12293 | } | |
12294 | if (dst_reg->umin_value + umin_val < umin_val || | |
12295 | dst_reg->umax_value + umax_val < umax_val) { | |
12296 | dst_reg->umin_value = 0; | |
12297 | dst_reg->umax_value = U64_MAX; | |
12298 | } else { | |
12299 | dst_reg->umin_value += umin_val; | |
12300 | dst_reg->umax_value += umax_val; | |
12301 | } | |
3f50f132 JF |
12302 | } |
12303 | ||
12304 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
12305 | struct bpf_reg_state *src_reg) | |
12306 | { | |
12307 | s32 smin_val = src_reg->s32_min_value; | |
12308 | s32 smax_val = src_reg->s32_max_value; | |
12309 | u32 umin_val = src_reg->u32_min_value; | |
12310 | u32 umax_val = src_reg->u32_max_value; | |
12311 | ||
12312 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
12313 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
12314 | /* Overflow possible, we know nothing */ | |
12315 | dst_reg->s32_min_value = S32_MIN; | |
12316 | dst_reg->s32_max_value = S32_MAX; | |
12317 | } else { | |
12318 | dst_reg->s32_min_value -= smax_val; | |
12319 | dst_reg->s32_max_value -= smin_val; | |
12320 | } | |
12321 | if (dst_reg->u32_min_value < umax_val) { | |
12322 | /* Overflow possible, we know nothing */ | |
12323 | dst_reg->u32_min_value = 0; | |
12324 | dst_reg->u32_max_value = U32_MAX; | |
12325 | } else { | |
12326 | /* Cannot overflow (as long as bounds are consistent) */ | |
12327 | dst_reg->u32_min_value -= umax_val; | |
12328 | dst_reg->u32_max_value -= umin_val; | |
12329 | } | |
07cd2631 JF |
12330 | } |
12331 | ||
12332 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
12333 | struct bpf_reg_state *src_reg) | |
12334 | { | |
12335 | s64 smin_val = src_reg->smin_value; | |
12336 | s64 smax_val = src_reg->smax_value; | |
12337 | u64 umin_val = src_reg->umin_value; | |
12338 | u64 umax_val = src_reg->umax_value; | |
12339 | ||
12340 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
12341 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
12342 | /* Overflow possible, we know nothing */ | |
12343 | dst_reg->smin_value = S64_MIN; | |
12344 | dst_reg->smax_value = S64_MAX; | |
12345 | } else { | |
12346 | dst_reg->smin_value -= smax_val; | |
12347 | dst_reg->smax_value -= smin_val; | |
12348 | } | |
12349 | if (dst_reg->umin_value < umax_val) { | |
12350 | /* Overflow possible, we know nothing */ | |
12351 | dst_reg->umin_value = 0; | |
12352 | dst_reg->umax_value = U64_MAX; | |
12353 | } else { | |
12354 | /* Cannot overflow (as long as bounds are consistent) */ | |
12355 | dst_reg->umin_value -= umax_val; | |
12356 | dst_reg->umax_value -= umin_val; | |
12357 | } | |
3f50f132 JF |
12358 | } |
12359 | ||
12360 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
12361 | struct bpf_reg_state *src_reg) | |
12362 | { | |
12363 | s32 smin_val = src_reg->s32_min_value; | |
12364 | u32 umin_val = src_reg->u32_min_value; | |
12365 | u32 umax_val = src_reg->u32_max_value; | |
12366 | ||
12367 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
12368 | /* Ain't nobody got time to multiply that sign */ | |
12369 | __mark_reg32_unbounded(dst_reg); | |
12370 | return; | |
12371 | } | |
12372 | /* Both values are positive, so we can work with unsigned and | |
12373 | * copy the result to signed (unless it exceeds S32_MAX). | |
12374 | */ | |
12375 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
12376 | /* Potential overflow, we know nothing */ | |
12377 | __mark_reg32_unbounded(dst_reg); | |
12378 | return; | |
12379 | } | |
12380 | dst_reg->u32_min_value *= umin_val; | |
12381 | dst_reg->u32_max_value *= umax_val; | |
12382 | if (dst_reg->u32_max_value > S32_MAX) { | |
12383 | /* Overflow possible, we know nothing */ | |
12384 | dst_reg->s32_min_value = S32_MIN; | |
12385 | dst_reg->s32_max_value = S32_MAX; | |
12386 | } else { | |
12387 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
12388 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
12389 | } | |
07cd2631 JF |
12390 | } |
12391 | ||
12392 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
12393 | struct bpf_reg_state *src_reg) | |
12394 | { | |
12395 | s64 smin_val = src_reg->smin_value; | |
12396 | u64 umin_val = src_reg->umin_value; | |
12397 | u64 umax_val = src_reg->umax_value; | |
12398 | ||
07cd2631 JF |
12399 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
12400 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 12401 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
12402 | return; |
12403 | } | |
12404 | /* Both values are positive, so we can work with unsigned and | |
12405 | * copy the result to signed (unless it exceeds S64_MAX). | |
12406 | */ | |
12407 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
12408 | /* Potential overflow, we know nothing */ | |
3f50f132 | 12409 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
12410 | return; |
12411 | } | |
12412 | dst_reg->umin_value *= umin_val; | |
12413 | dst_reg->umax_value *= umax_val; | |
12414 | if (dst_reg->umax_value > S64_MAX) { | |
12415 | /* Overflow possible, we know nothing */ | |
12416 | dst_reg->smin_value = S64_MIN; | |
12417 | dst_reg->smax_value = S64_MAX; | |
12418 | } else { | |
12419 | dst_reg->smin_value = dst_reg->umin_value; | |
12420 | dst_reg->smax_value = dst_reg->umax_value; | |
12421 | } | |
12422 | } | |
12423 | ||
3f50f132 JF |
12424 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
12425 | struct bpf_reg_state *src_reg) | |
12426 | { | |
12427 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
12428 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
12429 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
12430 | s32 smin_val = src_reg->s32_min_value; | |
12431 | u32 umax_val = src_reg->u32_max_value; | |
12432 | ||
049c4e13 DB |
12433 | if (src_known && dst_known) { |
12434 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 12435 | return; |
049c4e13 | 12436 | } |
3f50f132 JF |
12437 | |
12438 | /* We get our minimum from the var_off, since that's inherently | |
12439 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
12440 | */ | |
12441 | dst_reg->u32_min_value = var32_off.value; | |
12442 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
12443 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
12444 | /* Lose signed bounds when ANDing negative numbers, | |
12445 | * ain't nobody got time for that. | |
12446 | */ | |
12447 | dst_reg->s32_min_value = S32_MIN; | |
12448 | dst_reg->s32_max_value = S32_MAX; | |
12449 | } else { | |
12450 | /* ANDing two positives gives a positive, so safe to | |
12451 | * cast result into s64. | |
12452 | */ | |
12453 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
12454 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
12455 | } | |
3f50f132 JF |
12456 | } |
12457 | ||
07cd2631 JF |
12458 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
12459 | struct bpf_reg_state *src_reg) | |
12460 | { | |
3f50f132 JF |
12461 | bool src_known = tnum_is_const(src_reg->var_off); |
12462 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
12463 | s64 smin_val = src_reg->smin_value; |
12464 | u64 umax_val = src_reg->umax_value; | |
12465 | ||
3f50f132 | 12466 | if (src_known && dst_known) { |
4fbb38a3 | 12467 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
12468 | return; |
12469 | } | |
12470 | ||
07cd2631 JF |
12471 | /* We get our minimum from the var_off, since that's inherently |
12472 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
12473 | */ | |
07cd2631 JF |
12474 | dst_reg->umin_value = dst_reg->var_off.value; |
12475 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
12476 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
12477 | /* Lose signed bounds when ANDing negative numbers, | |
12478 | * ain't nobody got time for that. | |
12479 | */ | |
12480 | dst_reg->smin_value = S64_MIN; | |
12481 | dst_reg->smax_value = S64_MAX; | |
12482 | } else { | |
12483 | /* ANDing two positives gives a positive, so safe to | |
12484 | * cast result into s64. | |
12485 | */ | |
12486 | dst_reg->smin_value = dst_reg->umin_value; | |
12487 | dst_reg->smax_value = dst_reg->umax_value; | |
12488 | } | |
12489 | /* We may learn something more from the var_off */ | |
12490 | __update_reg_bounds(dst_reg); | |
12491 | } | |
12492 | ||
3f50f132 JF |
12493 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
12494 | struct bpf_reg_state *src_reg) | |
12495 | { | |
12496 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
12497 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
12498 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5b9fbeb7 DB |
12499 | s32 smin_val = src_reg->s32_min_value; |
12500 | u32 umin_val = src_reg->u32_min_value; | |
3f50f132 | 12501 | |
049c4e13 DB |
12502 | if (src_known && dst_known) { |
12503 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 12504 | return; |
049c4e13 | 12505 | } |
3f50f132 JF |
12506 | |
12507 | /* We get our maximum from the var_off, and our minimum is the | |
12508 | * maximum of the operands' minima | |
12509 | */ | |
12510 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
12511 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
12512 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
12513 | /* Lose signed bounds when ORing negative numbers, | |
12514 | * ain't nobody got time for that. | |
12515 | */ | |
12516 | dst_reg->s32_min_value = S32_MIN; | |
12517 | dst_reg->s32_max_value = S32_MAX; | |
12518 | } else { | |
12519 | /* ORing two positives gives a positive, so safe to | |
12520 | * cast result into s64. | |
12521 | */ | |
5b9fbeb7 DB |
12522 | dst_reg->s32_min_value = dst_reg->u32_min_value; |
12523 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
3f50f132 JF |
12524 | } |
12525 | } | |
12526 | ||
07cd2631 JF |
12527 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
12528 | struct bpf_reg_state *src_reg) | |
12529 | { | |
3f50f132 JF |
12530 | bool src_known = tnum_is_const(src_reg->var_off); |
12531 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
12532 | s64 smin_val = src_reg->smin_value; |
12533 | u64 umin_val = src_reg->umin_value; | |
12534 | ||
3f50f132 | 12535 | if (src_known && dst_known) { |
4fbb38a3 | 12536 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
12537 | return; |
12538 | } | |
12539 | ||
07cd2631 JF |
12540 | /* We get our maximum from the var_off, and our minimum is the |
12541 | * maximum of the operands' minima | |
12542 | */ | |
07cd2631 JF |
12543 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
12544 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
12545 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
12546 | /* Lose signed bounds when ORing negative numbers, | |
12547 | * ain't nobody got time for that. | |
12548 | */ | |
12549 | dst_reg->smin_value = S64_MIN; | |
12550 | dst_reg->smax_value = S64_MAX; | |
12551 | } else { | |
12552 | /* ORing two positives gives a positive, so safe to | |
12553 | * cast result into s64. | |
12554 | */ | |
12555 | dst_reg->smin_value = dst_reg->umin_value; | |
12556 | dst_reg->smax_value = dst_reg->umax_value; | |
12557 | } | |
12558 | /* We may learn something more from the var_off */ | |
12559 | __update_reg_bounds(dst_reg); | |
12560 | } | |
12561 | ||
2921c90d YS |
12562 | static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg, |
12563 | struct bpf_reg_state *src_reg) | |
12564 | { | |
12565 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
12566 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
12567 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
12568 | s32 smin_val = src_reg->s32_min_value; | |
12569 | ||
049c4e13 DB |
12570 | if (src_known && dst_known) { |
12571 | __mark_reg32_known(dst_reg, var32_off.value); | |
2921c90d | 12572 | return; |
049c4e13 | 12573 | } |
2921c90d YS |
12574 | |
12575 | /* We get both minimum and maximum from the var32_off. */ | |
12576 | dst_reg->u32_min_value = var32_off.value; | |
12577 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
12578 | ||
12579 | if (dst_reg->s32_min_value >= 0 && smin_val >= 0) { | |
12580 | /* XORing two positive sign numbers gives a positive, | |
12581 | * so safe to cast u32 result into s32. | |
12582 | */ | |
12583 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
12584 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
12585 | } else { | |
12586 | dst_reg->s32_min_value = S32_MIN; | |
12587 | dst_reg->s32_max_value = S32_MAX; | |
12588 | } | |
12589 | } | |
12590 | ||
12591 | static void scalar_min_max_xor(struct bpf_reg_state *dst_reg, | |
12592 | struct bpf_reg_state *src_reg) | |
12593 | { | |
12594 | bool src_known = tnum_is_const(src_reg->var_off); | |
12595 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
12596 | s64 smin_val = src_reg->smin_value; | |
12597 | ||
12598 | if (src_known && dst_known) { | |
12599 | /* dst_reg->var_off.value has been updated earlier */ | |
12600 | __mark_reg_known(dst_reg, dst_reg->var_off.value); | |
12601 | return; | |
12602 | } | |
12603 | ||
12604 | /* We get both minimum and maximum from the var_off. */ | |
12605 | dst_reg->umin_value = dst_reg->var_off.value; | |
12606 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
12607 | ||
12608 | if (dst_reg->smin_value >= 0 && smin_val >= 0) { | |
12609 | /* XORing two positive sign numbers gives a positive, | |
12610 | * so safe to cast u64 result into s64. | |
12611 | */ | |
12612 | dst_reg->smin_value = dst_reg->umin_value; | |
12613 | dst_reg->smax_value = dst_reg->umax_value; | |
12614 | } else { | |
12615 | dst_reg->smin_value = S64_MIN; | |
12616 | dst_reg->smax_value = S64_MAX; | |
12617 | } | |
12618 | ||
12619 | __update_reg_bounds(dst_reg); | |
12620 | } | |
12621 | ||
3f50f132 JF |
12622 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
12623 | u64 umin_val, u64 umax_val) | |
07cd2631 | 12624 | { |
07cd2631 JF |
12625 | /* We lose all sign bit information (except what we can pick |
12626 | * up from var_off) | |
12627 | */ | |
3f50f132 JF |
12628 | dst_reg->s32_min_value = S32_MIN; |
12629 | dst_reg->s32_max_value = S32_MAX; | |
12630 | /* If we might shift our top bit out, then we know nothing */ | |
12631 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
12632 | dst_reg->u32_min_value = 0; | |
12633 | dst_reg->u32_max_value = U32_MAX; | |
12634 | } else { | |
12635 | dst_reg->u32_min_value <<= umin_val; | |
12636 | dst_reg->u32_max_value <<= umax_val; | |
12637 | } | |
12638 | } | |
12639 | ||
12640 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
12641 | struct bpf_reg_state *src_reg) | |
12642 | { | |
12643 | u32 umax_val = src_reg->u32_max_value; | |
12644 | u32 umin_val = src_reg->u32_min_value; | |
12645 | /* u32 alu operation will zext upper bits */ | |
12646 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
12647 | ||
12648 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
12649 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
12650 | /* Not required but being careful mark reg64 bounds as unknown so | |
12651 | * that we are forced to pick them up from tnum and zext later and | |
12652 | * if some path skips this step we are still safe. | |
12653 | */ | |
12654 | __mark_reg64_unbounded(dst_reg); | |
12655 | __update_reg32_bounds(dst_reg); | |
12656 | } | |
12657 | ||
12658 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
12659 | u64 umin_val, u64 umax_val) | |
12660 | { | |
12661 | /* Special case <<32 because it is a common compiler pattern to sign | |
12662 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
12663 | * positive we know this shift will also be positive so we can track | |
12664 | * bounds correctly. Otherwise we lose all sign bit information except | |
12665 | * what we can pick up from var_off. Perhaps we can generalize this | |
12666 | * later to shifts of any length. | |
12667 | */ | |
12668 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
12669 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
12670 | else | |
12671 | dst_reg->smax_value = S64_MAX; | |
12672 | ||
12673 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
12674 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
12675 | else | |
12676 | dst_reg->smin_value = S64_MIN; | |
12677 | ||
07cd2631 JF |
12678 | /* If we might shift our top bit out, then we know nothing */ |
12679 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
12680 | dst_reg->umin_value = 0; | |
12681 | dst_reg->umax_value = U64_MAX; | |
12682 | } else { | |
12683 | dst_reg->umin_value <<= umin_val; | |
12684 | dst_reg->umax_value <<= umax_val; | |
12685 | } | |
3f50f132 JF |
12686 | } |
12687 | ||
12688 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
12689 | struct bpf_reg_state *src_reg) | |
12690 | { | |
12691 | u64 umax_val = src_reg->umax_value; | |
12692 | u64 umin_val = src_reg->umin_value; | |
12693 | ||
12694 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
12695 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
12696 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
12697 | ||
07cd2631 JF |
12698 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
12699 | /* We may learn something more from the var_off */ | |
12700 | __update_reg_bounds(dst_reg); | |
12701 | } | |
12702 | ||
3f50f132 JF |
12703 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
12704 | struct bpf_reg_state *src_reg) | |
12705 | { | |
12706 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
12707 | u32 umax_val = src_reg->u32_max_value; | |
12708 | u32 umin_val = src_reg->u32_min_value; | |
12709 | ||
12710 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
12711 | * be negative, then either: | |
12712 | * 1) src_reg might be zero, so the sign bit of the result is | |
12713 | * unknown, so we lose our signed bounds | |
12714 | * 2) it's known negative, thus the unsigned bounds capture the | |
12715 | * signed bounds | |
12716 | * 3) the signed bounds cross zero, so they tell us nothing | |
12717 | * about the result | |
12718 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 12719 | * unsigned bounds capture the signed bounds. |
3f50f132 JF |
12720 | * Thus, in all cases it suffices to blow away our signed bounds |
12721 | * and rely on inferring new ones from the unsigned bounds and | |
12722 | * var_off of the result. | |
12723 | */ | |
12724 | dst_reg->s32_min_value = S32_MIN; | |
12725 | dst_reg->s32_max_value = S32_MAX; | |
12726 | ||
12727 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
12728 | dst_reg->u32_min_value >>= umax_val; | |
12729 | dst_reg->u32_max_value >>= umin_val; | |
12730 | ||
12731 | __mark_reg64_unbounded(dst_reg); | |
12732 | __update_reg32_bounds(dst_reg); | |
12733 | } | |
12734 | ||
07cd2631 JF |
12735 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
12736 | struct bpf_reg_state *src_reg) | |
12737 | { | |
12738 | u64 umax_val = src_reg->umax_value; | |
12739 | u64 umin_val = src_reg->umin_value; | |
12740 | ||
12741 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
12742 | * be negative, then either: | |
12743 | * 1) src_reg might be zero, so the sign bit of the result is | |
12744 | * unknown, so we lose our signed bounds | |
12745 | * 2) it's known negative, thus the unsigned bounds capture the | |
12746 | * signed bounds | |
12747 | * 3) the signed bounds cross zero, so they tell us nothing | |
12748 | * about the result | |
12749 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 12750 | * unsigned bounds capture the signed bounds. |
07cd2631 JF |
12751 | * Thus, in all cases it suffices to blow away our signed bounds |
12752 | * and rely on inferring new ones from the unsigned bounds and | |
12753 | * var_off of the result. | |
12754 | */ | |
12755 | dst_reg->smin_value = S64_MIN; | |
12756 | dst_reg->smax_value = S64_MAX; | |
12757 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
12758 | dst_reg->umin_value >>= umax_val; | |
12759 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
12760 | |
12761 | /* Its not easy to operate on alu32 bounds here because it depends | |
12762 | * on bits being shifted in. Take easy way out and mark unbounded | |
12763 | * so we can recalculate later from tnum. | |
12764 | */ | |
12765 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
12766 | __update_reg_bounds(dst_reg); |
12767 | } | |
12768 | ||
3f50f132 JF |
12769 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
12770 | struct bpf_reg_state *src_reg) | |
07cd2631 | 12771 | { |
3f50f132 | 12772 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
12773 | |
12774 | /* Upon reaching here, src_known is true and | |
12775 | * umax_val is equal to umin_val. | |
12776 | */ | |
3f50f132 JF |
12777 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
12778 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 12779 | |
3f50f132 JF |
12780 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
12781 | ||
12782 | /* blow away the dst_reg umin_value/umax_value and rely on | |
12783 | * dst_reg var_off to refine the result. | |
12784 | */ | |
12785 | dst_reg->u32_min_value = 0; | |
12786 | dst_reg->u32_max_value = U32_MAX; | |
12787 | ||
12788 | __mark_reg64_unbounded(dst_reg); | |
12789 | __update_reg32_bounds(dst_reg); | |
12790 | } | |
12791 | ||
12792 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
12793 | struct bpf_reg_state *src_reg) | |
12794 | { | |
12795 | u64 umin_val = src_reg->umin_value; | |
12796 | ||
12797 | /* Upon reaching here, src_known is true and umax_val is equal | |
12798 | * to umin_val. | |
12799 | */ | |
12800 | dst_reg->smin_value >>= umin_val; | |
12801 | dst_reg->smax_value >>= umin_val; | |
12802 | ||
12803 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
12804 | |
12805 | /* blow away the dst_reg umin_value/umax_value and rely on | |
12806 | * dst_reg var_off to refine the result. | |
12807 | */ | |
12808 | dst_reg->umin_value = 0; | |
12809 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
12810 | |
12811 | /* Its not easy to operate on alu32 bounds here because it depends | |
12812 | * on bits being shifted in from upper 32-bits. Take easy way out | |
12813 | * and mark unbounded so we can recalculate later from tnum. | |
12814 | */ | |
12815 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
12816 | __update_reg_bounds(dst_reg); |
12817 | } | |
12818 | ||
468f6eaf JH |
12819 | /* WARNING: This function does calculations on 64-bit values, but the actual |
12820 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
12821 | * need extra checks in the 32-bit case. | |
12822 | */ | |
f1174f77 EC |
12823 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
12824 | struct bpf_insn *insn, | |
12825 | struct bpf_reg_state *dst_reg, | |
12826 | struct bpf_reg_state src_reg) | |
969bf05e | 12827 | { |
638f5b90 | 12828 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 12829 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 12830 | bool src_known; |
b03c9f9f EC |
12831 | s64 smin_val, smax_val; |
12832 | u64 umin_val, umax_val; | |
3f50f132 JF |
12833 | s32 s32_min_val, s32_max_val; |
12834 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 12835 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
3f50f132 | 12836 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
a6aaece0 | 12837 | int ret; |
b799207e | 12838 | |
b03c9f9f EC |
12839 | smin_val = src_reg.smin_value; |
12840 | smax_val = src_reg.smax_value; | |
12841 | umin_val = src_reg.umin_value; | |
12842 | umax_val = src_reg.umax_value; | |
f23cc643 | 12843 | |
3f50f132 JF |
12844 | s32_min_val = src_reg.s32_min_value; |
12845 | s32_max_val = src_reg.s32_max_value; | |
12846 | u32_min_val = src_reg.u32_min_value; | |
12847 | u32_max_val = src_reg.u32_max_value; | |
12848 | ||
12849 | if (alu32) { | |
12850 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
12851 | if ((src_known && |
12852 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
12853 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
12854 | /* Taint dst register if offset had invalid bounds | |
12855 | * derived from e.g. dead branches. | |
12856 | */ | |
12857 | __mark_reg_unknown(env, dst_reg); | |
12858 | return 0; | |
12859 | } | |
12860 | } else { | |
12861 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
12862 | if ((src_known && |
12863 | (smin_val != smax_val || umin_val != umax_val)) || | |
12864 | smin_val > smax_val || umin_val > umax_val) { | |
12865 | /* Taint dst register if offset had invalid bounds | |
12866 | * derived from e.g. dead branches. | |
12867 | */ | |
12868 | __mark_reg_unknown(env, dst_reg); | |
12869 | return 0; | |
12870 | } | |
6f16101e DB |
12871 | } |
12872 | ||
bb7f0f98 AS |
12873 | if (!src_known && |
12874 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 12875 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
12876 | return 0; |
12877 | } | |
12878 | ||
f5288193 DB |
12879 | if (sanitize_needed(opcode)) { |
12880 | ret = sanitize_val_alu(env, insn); | |
12881 | if (ret < 0) | |
12882 | return sanitize_err(env, insn, ret, NULL, NULL); | |
12883 | } | |
12884 | ||
3f50f132 JF |
12885 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
12886 | * There are two classes of instructions: The first class we track both | |
12887 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
12888 | * greatest amount of precision when alu operations are mixed with jmp32 | |
12889 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
12890 | * and BPF_OR. This is possible because these ops have fairly easy to | |
12891 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
12892 | * See alu32 verifier tests for examples. The second class of | |
12893 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
12894 | * with regards to tracking sign/unsigned bounds because the bits may | |
12895 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
12896 | * the reg unbounded in the subreg bound space and use the resulting | |
12897 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
12898 | */ | |
48461135 JB |
12899 | switch (opcode) { |
12900 | case BPF_ADD: | |
3f50f132 | 12901 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 12902 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 12903 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
12904 | break; |
12905 | case BPF_SUB: | |
3f50f132 | 12906 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 12907 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 12908 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
12909 | break; |
12910 | case BPF_MUL: | |
3f50f132 JF |
12911 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
12912 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 12913 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
12914 | break; |
12915 | case BPF_AND: | |
3f50f132 JF |
12916 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
12917 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 12918 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
12919 | break; |
12920 | case BPF_OR: | |
3f50f132 JF |
12921 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
12922 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 12923 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 | 12924 | break; |
2921c90d YS |
12925 | case BPF_XOR: |
12926 | dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off); | |
12927 | scalar32_min_max_xor(dst_reg, &src_reg); | |
12928 | scalar_min_max_xor(dst_reg, &src_reg); | |
12929 | break; | |
48461135 | 12930 | case BPF_LSH: |
468f6eaf JH |
12931 | if (umax_val >= insn_bitness) { |
12932 | /* Shifts greater than 31 or 63 are undefined. | |
12933 | * This includes shifts by a negative number. | |
b03c9f9f | 12934 | */ |
61bd5218 | 12935 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
12936 | break; |
12937 | } | |
3f50f132 JF |
12938 | if (alu32) |
12939 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
12940 | else | |
12941 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
12942 | break; |
12943 | case BPF_RSH: | |
468f6eaf JH |
12944 | if (umax_val >= insn_bitness) { |
12945 | /* Shifts greater than 31 or 63 are undefined. | |
12946 | * This includes shifts by a negative number. | |
b03c9f9f | 12947 | */ |
61bd5218 | 12948 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
12949 | break; |
12950 | } | |
3f50f132 JF |
12951 | if (alu32) |
12952 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
12953 | else | |
12954 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 12955 | break; |
9cbe1f5a YS |
12956 | case BPF_ARSH: |
12957 | if (umax_val >= insn_bitness) { | |
12958 | /* Shifts greater than 31 or 63 are undefined. | |
12959 | * This includes shifts by a negative number. | |
12960 | */ | |
12961 | mark_reg_unknown(env, regs, insn->dst_reg); | |
12962 | break; | |
12963 | } | |
3f50f132 JF |
12964 | if (alu32) |
12965 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
12966 | else | |
12967 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 12968 | break; |
48461135 | 12969 | default: |
61bd5218 | 12970 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
12971 | break; |
12972 | } | |
12973 | ||
3f50f132 JF |
12974 | /* ALU32 ops are zero extended into 64bit register */ |
12975 | if (alu32) | |
12976 | zext_32_to_64(dst_reg); | |
3844d153 | 12977 | reg_bounds_sync(dst_reg); |
f1174f77 EC |
12978 | return 0; |
12979 | } | |
12980 | ||
12981 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
12982 | * and var_off. | |
12983 | */ | |
12984 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
12985 | struct bpf_insn *insn) | |
12986 | { | |
f4d7e40a AS |
12987 | struct bpf_verifier_state *vstate = env->cur_state; |
12988 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
12989 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
12990 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
12991 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 12992 | int err; |
f1174f77 EC |
12993 | |
12994 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
12995 | src_reg = NULL; |
12996 | if (dst_reg->type != SCALAR_VALUE) | |
12997 | ptr_reg = dst_reg; | |
75748837 AS |
12998 | else |
12999 | /* Make sure ID is cleared otherwise dst_reg min/max could be | |
13000 | * incorrectly propagated into other registers by find_equal_scalars() | |
13001 | */ | |
13002 | dst_reg->id = 0; | |
f1174f77 EC |
13003 | if (BPF_SRC(insn->code) == BPF_X) { |
13004 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
13005 | if (src_reg->type != SCALAR_VALUE) { |
13006 | if (dst_reg->type != SCALAR_VALUE) { | |
13007 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
13008 | * an arbitrary scalar. Disallow all math except |
13009 | * pointer subtraction | |
f1174f77 | 13010 | */ |
dd066823 | 13011 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
13012 | mark_reg_unknown(env, regs, insn->dst_reg); |
13013 | return 0; | |
f1174f77 | 13014 | } |
82abbf8d AS |
13015 | verbose(env, "R%d pointer %s pointer prohibited\n", |
13016 | insn->dst_reg, | |
13017 | bpf_alu_string[opcode >> 4]); | |
13018 | return -EACCES; | |
f1174f77 EC |
13019 | } else { |
13020 | /* scalar += pointer | |
13021 | * This is legal, but we have to reverse our | |
13022 | * src/dest handling in computing the range | |
13023 | */ | |
b5dc0163 AS |
13024 | err = mark_chain_precision(env, insn->dst_reg); |
13025 | if (err) | |
13026 | return err; | |
82abbf8d AS |
13027 | return adjust_ptr_min_max_vals(env, insn, |
13028 | src_reg, dst_reg); | |
f1174f77 EC |
13029 | } |
13030 | } else if (ptr_reg) { | |
13031 | /* pointer += scalar */ | |
b5dc0163 AS |
13032 | err = mark_chain_precision(env, insn->src_reg); |
13033 | if (err) | |
13034 | return err; | |
82abbf8d AS |
13035 | return adjust_ptr_min_max_vals(env, insn, |
13036 | dst_reg, src_reg); | |
a3b666bf AN |
13037 | } else if (dst_reg->precise) { |
13038 | /* if dst_reg is precise, src_reg should be precise as well */ | |
13039 | err = mark_chain_precision(env, insn->src_reg); | |
13040 | if (err) | |
13041 | return err; | |
f1174f77 EC |
13042 | } |
13043 | } else { | |
13044 | /* Pretend the src is a reg with a known value, since we only | |
13045 | * need to be able to read from this state. | |
13046 | */ | |
13047 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 13048 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 13049 | src_reg = &off_reg; |
82abbf8d AS |
13050 | if (ptr_reg) /* pointer += K */ |
13051 | return adjust_ptr_min_max_vals(env, insn, | |
13052 | ptr_reg, src_reg); | |
f1174f77 EC |
13053 | } |
13054 | ||
13055 | /* Got here implies adding two SCALAR_VALUEs */ | |
13056 | if (WARN_ON_ONCE(ptr_reg)) { | |
0f55f9ed | 13057 | print_verifier_state(env, state, true); |
61bd5218 | 13058 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
13059 | return -EINVAL; |
13060 | } | |
13061 | if (WARN_ON(!src_reg)) { | |
0f55f9ed | 13062 | print_verifier_state(env, state, true); |
61bd5218 | 13063 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
13064 | return -EINVAL; |
13065 | } | |
13066 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
13067 | } |
13068 | ||
17a52670 | 13069 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 13070 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 13071 | { |
638f5b90 | 13072 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
13073 | u8 opcode = BPF_OP(insn->code); |
13074 | int err; | |
13075 | ||
13076 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
13077 | if (opcode == BPF_NEG) { | |
395e942d | 13078 | if (BPF_SRC(insn->code) != BPF_K || |
17a52670 AS |
13079 | insn->src_reg != BPF_REG_0 || |
13080 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 13081 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
13082 | return -EINVAL; |
13083 | } | |
13084 | } else { | |
13085 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 | 13086 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
0845c3db YS |
13087 | (BPF_CLASS(insn->code) == BPF_ALU64 && |
13088 | BPF_SRC(insn->code) != BPF_TO_LE)) { | |
61bd5218 | 13089 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
13090 | return -EINVAL; |
13091 | } | |
13092 | } | |
13093 | ||
13094 | /* check src operand */ | |
dc503a8a | 13095 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
13096 | if (err) |
13097 | return err; | |
13098 | ||
1be7f75d | 13099 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 13100 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
13101 | insn->dst_reg); |
13102 | return -EACCES; | |
13103 | } | |
13104 | ||
17a52670 | 13105 | /* check dest operand */ |
dc503a8a | 13106 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
13107 | if (err) |
13108 | return err; | |
13109 | ||
13110 | } else if (opcode == BPF_MOV) { | |
13111 | ||
13112 | if (BPF_SRC(insn->code) == BPF_X) { | |
8100928c | 13113 | if (insn->imm != 0) { |
61bd5218 | 13114 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
13115 | return -EINVAL; |
13116 | } | |
13117 | ||
8100928c YS |
13118 | if (BPF_CLASS(insn->code) == BPF_ALU) { |
13119 | if (insn->off != 0 && insn->off != 8 && insn->off != 16) { | |
13120 | verbose(env, "BPF_MOV uses reserved fields\n"); | |
13121 | return -EINVAL; | |
13122 | } | |
13123 | } else { | |
13124 | if (insn->off != 0 && insn->off != 8 && insn->off != 16 && | |
13125 | insn->off != 32) { | |
13126 | verbose(env, "BPF_MOV uses reserved fields\n"); | |
13127 | return -EINVAL; | |
13128 | } | |
13129 | } | |
13130 | ||
17a52670 | 13131 | /* check src operand */ |
dc503a8a | 13132 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
13133 | if (err) |
13134 | return err; | |
13135 | } else { | |
13136 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 13137 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
13138 | return -EINVAL; |
13139 | } | |
13140 | } | |
13141 | ||
fbeb1603 AF |
13142 | /* check dest operand, mark as required later */ |
13143 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
13144 | if (err) |
13145 | return err; | |
13146 | ||
13147 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
13148 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
13149 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
1ffc85d9 EZ |
13150 | bool need_id = src_reg->type == SCALAR_VALUE && !src_reg->id && |
13151 | !tnum_is_const(src_reg->var_off); | |
e434b8cd | 13152 | |
17a52670 | 13153 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
8100928c YS |
13154 | if (insn->off == 0) { |
13155 | /* case: R1 = R2 | |
13156 | * copy register state to dest reg | |
75748837 | 13157 | */ |
8100928c YS |
13158 | if (need_id) |
13159 | /* Assign src and dst registers the same ID | |
13160 | * that will be used by find_equal_scalars() | |
13161 | * to propagate min/max range. | |
13162 | */ | |
13163 | src_reg->id = ++env->id_gen; | |
13164 | copy_register_state(dst_reg, src_reg); | |
13165 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13166 | dst_reg->subreg_def = DEF_NOT_SUBREG; | |
13167 | } else { | |
13168 | /* case: R1 = (s8, s16 s32)R2 */ | |
db2baf82 YS |
13169 | if (is_pointer_value(env, insn->src_reg)) { |
13170 | verbose(env, | |
13171 | "R%d sign-extension part of pointer\n", | |
13172 | insn->src_reg); | |
13173 | return -EACCES; | |
13174 | } else if (src_reg->type == SCALAR_VALUE) { | |
13175 | bool no_sext; | |
13176 | ||
13177 | no_sext = src_reg->umax_value < (1ULL << (insn->off - 1)); | |
13178 | if (no_sext && need_id) | |
13179 | src_reg->id = ++env->id_gen; | |
13180 | copy_register_state(dst_reg, src_reg); | |
13181 | if (!no_sext) | |
13182 | dst_reg->id = 0; | |
13183 | coerce_reg_to_size_sx(dst_reg, insn->off >> 3); | |
13184 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13185 | dst_reg->subreg_def = DEF_NOT_SUBREG; | |
13186 | } else { | |
13187 | mark_reg_unknown(env, regs, insn->dst_reg); | |
13188 | } | |
8100928c | 13189 | } |
17a52670 | 13190 | } else { |
f1174f77 | 13191 | /* R1 = (u32) R2 */ |
1be7f75d | 13192 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
13193 | verbose(env, |
13194 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
13195 | insn->src_reg); |
13196 | return -EACCES; | |
e434b8cd | 13197 | } else if (src_reg->type == SCALAR_VALUE) { |
8100928c YS |
13198 | if (insn->off == 0) { |
13199 | bool is_src_reg_u32 = src_reg->umax_value <= U32_MAX; | |
13200 | ||
13201 | if (is_src_reg_u32 && need_id) | |
13202 | src_reg->id = ++env->id_gen; | |
13203 | copy_register_state(dst_reg, src_reg); | |
13204 | /* Make sure ID is cleared if src_reg is not in u32 | |
13205 | * range otherwise dst_reg min/max could be incorrectly | |
13206 | * propagated into src_reg by find_equal_scalars() | |
13207 | */ | |
13208 | if (!is_src_reg_u32) | |
13209 | dst_reg->id = 0; | |
13210 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13211 | dst_reg->subreg_def = env->insn_idx + 1; | |
13212 | } else { | |
13213 | /* case: W1 = (s8, s16)W2 */ | |
13214 | bool no_sext = src_reg->umax_value < (1ULL << (insn->off - 1)); | |
13215 | ||
13216 | if (no_sext && need_id) | |
13217 | src_reg->id = ++env->id_gen; | |
13218 | copy_register_state(dst_reg, src_reg); | |
13219 | if (!no_sext) | |
13220 | dst_reg->id = 0; | |
13221 | dst_reg->live |= REG_LIVE_WRITTEN; | |
13222 | dst_reg->subreg_def = env->insn_idx + 1; | |
13223 | coerce_subreg_to_size_sx(dst_reg, insn->off >> 3); | |
13224 | } | |
e434b8cd JW |
13225 | } else { |
13226 | mark_reg_unknown(env, regs, | |
13227 | insn->dst_reg); | |
1be7f75d | 13228 | } |
3f50f132 | 13229 | zext_32_to_64(dst_reg); |
3844d153 | 13230 | reg_bounds_sync(dst_reg); |
17a52670 AS |
13231 | } |
13232 | } else { | |
13233 | /* case: R = imm | |
13234 | * remember the value we stored into this reg | |
13235 | */ | |
fbeb1603 AF |
13236 | /* clear any state __mark_reg_known doesn't set */ |
13237 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 13238 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
13239 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
13240 | __mark_reg_known(regs + insn->dst_reg, | |
13241 | insn->imm); | |
13242 | } else { | |
13243 | __mark_reg_known(regs + insn->dst_reg, | |
13244 | (u32)insn->imm); | |
13245 | } | |
17a52670 AS |
13246 | } |
13247 | ||
13248 | } else if (opcode > BPF_END) { | |
61bd5218 | 13249 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
13250 | return -EINVAL; |
13251 | ||
13252 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
13253 | ||
17a52670 | 13254 | if (BPF_SRC(insn->code) == BPF_X) { |
ec0e2da9 YS |
13255 | if (insn->imm != 0 || insn->off > 1 || |
13256 | (insn->off == 1 && opcode != BPF_MOD && opcode != BPF_DIV)) { | |
61bd5218 | 13257 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
13258 | return -EINVAL; |
13259 | } | |
13260 | /* check src1 operand */ | |
dc503a8a | 13261 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
13262 | if (err) |
13263 | return err; | |
13264 | } else { | |
ec0e2da9 YS |
13265 | if (insn->src_reg != BPF_REG_0 || insn->off > 1 || |
13266 | (insn->off == 1 && opcode != BPF_MOD && opcode != BPF_DIV)) { | |
61bd5218 | 13267 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
13268 | return -EINVAL; |
13269 | } | |
13270 | } | |
13271 | ||
13272 | /* check src2 operand */ | |
dc503a8a | 13273 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
13274 | if (err) |
13275 | return err; | |
13276 | ||
13277 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
13278 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 13279 | verbose(env, "div by zero\n"); |
17a52670 AS |
13280 | return -EINVAL; |
13281 | } | |
13282 | ||
229394e8 RV |
13283 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
13284 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
13285 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
13286 | ||
13287 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 13288 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
13289 | return -EINVAL; |
13290 | } | |
13291 | } | |
13292 | ||
1a0dc1ac | 13293 | /* check dest operand */ |
dc503a8a | 13294 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
13295 | if (err) |
13296 | return err; | |
13297 | ||
f1174f77 | 13298 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
13299 | } |
13300 | ||
13301 | return 0; | |
13302 | } | |
13303 | ||
f4d7e40a | 13304 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 13305 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 13306 | enum bpf_reg_type type, |
fb2a311a | 13307 | bool range_right_open) |
969bf05e | 13308 | { |
b239da34 KKD |
13309 | struct bpf_func_state *state; |
13310 | struct bpf_reg_state *reg; | |
13311 | int new_range; | |
2d2be8ca | 13312 | |
fb2a311a DB |
13313 | if (dst_reg->off < 0 || |
13314 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
13315 | /* This doesn't give us any range */ |
13316 | return; | |
13317 | ||
b03c9f9f EC |
13318 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
13319 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
13320 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
13321 | * than pkt_end, but that's because it's also less than pkt. | |
13322 | */ | |
13323 | return; | |
13324 | ||
fb2a311a DB |
13325 | new_range = dst_reg->off; |
13326 | if (range_right_open) | |
2fa7d94a | 13327 | new_range++; |
fb2a311a DB |
13328 | |
13329 | /* Examples for register markings: | |
2d2be8ca | 13330 | * |
fb2a311a | 13331 | * pkt_data in dst register: |
2d2be8ca DB |
13332 | * |
13333 | * r2 = r3; | |
13334 | * r2 += 8; | |
13335 | * if (r2 > pkt_end) goto <handle exception> | |
13336 | * <access okay> | |
13337 | * | |
b4e432f1 DB |
13338 | * r2 = r3; |
13339 | * r2 += 8; | |
13340 | * if (r2 < pkt_end) goto <access okay> | |
13341 | * <handle exception> | |
13342 | * | |
2d2be8ca DB |
13343 | * Where: |
13344 | * r2 == dst_reg, pkt_end == src_reg | |
13345 | * r2=pkt(id=n,off=8,r=0) | |
13346 | * r3=pkt(id=n,off=0,r=0) | |
13347 | * | |
fb2a311a | 13348 | * pkt_data in src register: |
2d2be8ca DB |
13349 | * |
13350 | * r2 = r3; | |
13351 | * r2 += 8; | |
13352 | * if (pkt_end >= r2) goto <access okay> | |
13353 | * <handle exception> | |
13354 | * | |
b4e432f1 DB |
13355 | * r2 = r3; |
13356 | * r2 += 8; | |
13357 | * if (pkt_end <= r2) goto <handle exception> | |
13358 | * <access okay> | |
13359 | * | |
2d2be8ca DB |
13360 | * Where: |
13361 | * pkt_end == dst_reg, r2 == src_reg | |
13362 | * r2=pkt(id=n,off=8,r=0) | |
13363 | * r3=pkt(id=n,off=0,r=0) | |
13364 | * | |
13365 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
13366 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
13367 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
13368 | * the check. | |
969bf05e | 13369 | */ |
2d2be8ca | 13370 | |
f1174f77 EC |
13371 | /* If our ids match, then we must have the same max_value. And we |
13372 | * don't care about the other reg's fixed offset, since if it's too big | |
13373 | * the range won't allow anything. | |
13374 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
13375 | */ | |
b239da34 KKD |
13376 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
13377 | if (reg->type == type && reg->id == dst_reg->id) | |
13378 | /* keep the maximum range already checked */ | |
13379 | reg->range = max(reg->range, new_range); | |
13380 | })); | |
969bf05e AS |
13381 | } |
13382 | ||
3f50f132 | 13383 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 13384 | { |
3f50f132 JF |
13385 | struct tnum subreg = tnum_subreg(reg->var_off); |
13386 | s32 sval = (s32)val; | |
a72dafaf | 13387 | |
3f50f132 JF |
13388 | switch (opcode) { |
13389 | case BPF_JEQ: | |
13390 | if (tnum_is_const(subreg)) | |
13391 | return !!tnum_equals_const(subreg, val); | |
13fbcee5 YS |
13392 | else if (val < reg->u32_min_value || val > reg->u32_max_value) |
13393 | return 0; | |
3f50f132 JF |
13394 | break; |
13395 | case BPF_JNE: | |
13396 | if (tnum_is_const(subreg)) | |
13397 | return !tnum_equals_const(subreg, val); | |
13fbcee5 YS |
13398 | else if (val < reg->u32_min_value || val > reg->u32_max_value) |
13399 | return 1; | |
3f50f132 JF |
13400 | break; |
13401 | case BPF_JSET: | |
13402 | if ((~subreg.mask & subreg.value) & val) | |
13403 | return 1; | |
13404 | if (!((subreg.mask | subreg.value) & val)) | |
13405 | return 0; | |
13406 | break; | |
13407 | case BPF_JGT: | |
13408 | if (reg->u32_min_value > val) | |
13409 | return 1; | |
13410 | else if (reg->u32_max_value <= val) | |
13411 | return 0; | |
13412 | break; | |
13413 | case BPF_JSGT: | |
13414 | if (reg->s32_min_value > sval) | |
13415 | return 1; | |
ee114dd6 | 13416 | else if (reg->s32_max_value <= sval) |
3f50f132 JF |
13417 | return 0; |
13418 | break; | |
13419 | case BPF_JLT: | |
13420 | if (reg->u32_max_value < val) | |
13421 | return 1; | |
13422 | else if (reg->u32_min_value >= val) | |
13423 | return 0; | |
13424 | break; | |
13425 | case BPF_JSLT: | |
13426 | if (reg->s32_max_value < sval) | |
13427 | return 1; | |
13428 | else if (reg->s32_min_value >= sval) | |
13429 | return 0; | |
13430 | break; | |
13431 | case BPF_JGE: | |
13432 | if (reg->u32_min_value >= val) | |
13433 | return 1; | |
13434 | else if (reg->u32_max_value < val) | |
13435 | return 0; | |
13436 | break; | |
13437 | case BPF_JSGE: | |
13438 | if (reg->s32_min_value >= sval) | |
13439 | return 1; | |
13440 | else if (reg->s32_max_value < sval) | |
13441 | return 0; | |
13442 | break; | |
13443 | case BPF_JLE: | |
13444 | if (reg->u32_max_value <= val) | |
13445 | return 1; | |
13446 | else if (reg->u32_min_value > val) | |
13447 | return 0; | |
13448 | break; | |
13449 | case BPF_JSLE: | |
13450 | if (reg->s32_max_value <= sval) | |
13451 | return 1; | |
13452 | else if (reg->s32_min_value > sval) | |
13453 | return 0; | |
13454 | break; | |
13455 | } | |
4f7b3e82 | 13456 | |
3f50f132 JF |
13457 | return -1; |
13458 | } | |
092ed096 | 13459 | |
3f50f132 JF |
13460 | |
13461 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
13462 | { | |
13463 | s64 sval = (s64)val; | |
a72dafaf | 13464 | |
4f7b3e82 AS |
13465 | switch (opcode) { |
13466 | case BPF_JEQ: | |
13467 | if (tnum_is_const(reg->var_off)) | |
13468 | return !!tnum_equals_const(reg->var_off, val); | |
13fbcee5 YS |
13469 | else if (val < reg->umin_value || val > reg->umax_value) |
13470 | return 0; | |
4f7b3e82 AS |
13471 | break; |
13472 | case BPF_JNE: | |
13473 | if (tnum_is_const(reg->var_off)) | |
13474 | return !tnum_equals_const(reg->var_off, val); | |
13fbcee5 YS |
13475 | else if (val < reg->umin_value || val > reg->umax_value) |
13476 | return 1; | |
4f7b3e82 | 13477 | break; |
960ea056 JK |
13478 | case BPF_JSET: |
13479 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
13480 | return 1; | |
13481 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
13482 | return 0; | |
13483 | break; | |
4f7b3e82 AS |
13484 | case BPF_JGT: |
13485 | if (reg->umin_value > val) | |
13486 | return 1; | |
13487 | else if (reg->umax_value <= val) | |
13488 | return 0; | |
13489 | break; | |
13490 | case BPF_JSGT: | |
a72dafaf | 13491 | if (reg->smin_value > sval) |
4f7b3e82 | 13492 | return 1; |
ee114dd6 | 13493 | else if (reg->smax_value <= sval) |
4f7b3e82 AS |
13494 | return 0; |
13495 | break; | |
13496 | case BPF_JLT: | |
13497 | if (reg->umax_value < val) | |
13498 | return 1; | |
13499 | else if (reg->umin_value >= val) | |
13500 | return 0; | |
13501 | break; | |
13502 | case BPF_JSLT: | |
a72dafaf | 13503 | if (reg->smax_value < sval) |
4f7b3e82 | 13504 | return 1; |
a72dafaf | 13505 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
13506 | return 0; |
13507 | break; | |
13508 | case BPF_JGE: | |
13509 | if (reg->umin_value >= val) | |
13510 | return 1; | |
13511 | else if (reg->umax_value < val) | |
13512 | return 0; | |
13513 | break; | |
13514 | case BPF_JSGE: | |
a72dafaf | 13515 | if (reg->smin_value >= sval) |
4f7b3e82 | 13516 | return 1; |
a72dafaf | 13517 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
13518 | return 0; |
13519 | break; | |
13520 | case BPF_JLE: | |
13521 | if (reg->umax_value <= val) | |
13522 | return 1; | |
13523 | else if (reg->umin_value > val) | |
13524 | return 0; | |
13525 | break; | |
13526 | case BPF_JSLE: | |
a72dafaf | 13527 | if (reg->smax_value <= sval) |
4f7b3e82 | 13528 | return 1; |
a72dafaf | 13529 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
13530 | return 0; |
13531 | break; | |
13532 | } | |
13533 | ||
13534 | return -1; | |
13535 | } | |
13536 | ||
3f50f132 JF |
13537 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
13538 | * and return: | |
13539 | * 1 - branch will be taken and "goto target" will be executed | |
13540 | * 0 - branch will not be taken and fall-through to next insn | |
13541 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
13542 | * range [0,10] | |
604dca5e | 13543 | */ |
3f50f132 JF |
13544 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
13545 | bool is_jmp32) | |
604dca5e | 13546 | { |
cac616db | 13547 | if (__is_pointer_value(false, reg)) { |
51302c95 | 13548 | if (!reg_not_null(reg)) |
cac616db JF |
13549 | return -1; |
13550 | ||
13551 | /* If pointer is valid tests against zero will fail so we can | |
13552 | * use this to direct branch taken. | |
13553 | */ | |
13554 | if (val != 0) | |
13555 | return -1; | |
13556 | ||
13557 | switch (opcode) { | |
13558 | case BPF_JEQ: | |
13559 | return 0; | |
13560 | case BPF_JNE: | |
13561 | return 1; | |
13562 | default: | |
13563 | return -1; | |
13564 | } | |
13565 | } | |
604dca5e | 13566 | |
3f50f132 JF |
13567 | if (is_jmp32) |
13568 | return is_branch32_taken(reg, val, opcode); | |
13569 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
13570 | } |
13571 | ||
6d94e741 AS |
13572 | static int flip_opcode(u32 opcode) |
13573 | { | |
13574 | /* How can we transform "a <op> b" into "b <op> a"? */ | |
13575 | static const u8 opcode_flip[16] = { | |
13576 | /* these stay the same */ | |
13577 | [BPF_JEQ >> 4] = BPF_JEQ, | |
13578 | [BPF_JNE >> 4] = BPF_JNE, | |
13579 | [BPF_JSET >> 4] = BPF_JSET, | |
13580 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
13581 | [BPF_JGE >> 4] = BPF_JLE, | |
13582 | [BPF_JGT >> 4] = BPF_JLT, | |
13583 | [BPF_JLE >> 4] = BPF_JGE, | |
13584 | [BPF_JLT >> 4] = BPF_JGT, | |
13585 | [BPF_JSGE >> 4] = BPF_JSLE, | |
13586 | [BPF_JSGT >> 4] = BPF_JSLT, | |
13587 | [BPF_JSLE >> 4] = BPF_JSGE, | |
13588 | [BPF_JSLT >> 4] = BPF_JSGT | |
13589 | }; | |
13590 | return opcode_flip[opcode >> 4]; | |
13591 | } | |
13592 | ||
13593 | static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg, | |
13594 | struct bpf_reg_state *src_reg, | |
13595 | u8 opcode) | |
13596 | { | |
13597 | struct bpf_reg_state *pkt; | |
13598 | ||
13599 | if (src_reg->type == PTR_TO_PACKET_END) { | |
13600 | pkt = dst_reg; | |
13601 | } else if (dst_reg->type == PTR_TO_PACKET_END) { | |
13602 | pkt = src_reg; | |
13603 | opcode = flip_opcode(opcode); | |
13604 | } else { | |
13605 | return -1; | |
13606 | } | |
13607 | ||
13608 | if (pkt->range >= 0) | |
13609 | return -1; | |
13610 | ||
13611 | switch (opcode) { | |
13612 | case BPF_JLE: | |
13613 | /* pkt <= pkt_end */ | |
13614 | fallthrough; | |
13615 | case BPF_JGT: | |
13616 | /* pkt > pkt_end */ | |
13617 | if (pkt->range == BEYOND_PKT_END) | |
13618 | /* pkt has at last one extra byte beyond pkt_end */ | |
13619 | return opcode == BPF_JGT; | |
13620 | break; | |
13621 | case BPF_JLT: | |
13622 | /* pkt < pkt_end */ | |
13623 | fallthrough; | |
13624 | case BPF_JGE: | |
13625 | /* pkt >= pkt_end */ | |
13626 | if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END) | |
13627 | return opcode == BPF_JGE; | |
13628 | break; | |
13629 | } | |
13630 | return -1; | |
13631 | } | |
13632 | ||
48461135 JB |
13633 | /* Adjusts the register min/max values in the case that the dst_reg is the |
13634 | * variable register that we are working on, and src_reg is a constant or we're | |
13635 | * simply doing a BPF_K check. | |
f1174f77 | 13636 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
13637 | */ |
13638 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
13639 | struct bpf_reg_state *false_reg, |
13640 | u64 val, u32 val32, | |
092ed096 | 13641 | u8 opcode, bool is_jmp32) |
48461135 | 13642 | { |
3f50f132 JF |
13643 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
13644 | struct tnum false_64off = false_reg->var_off; | |
13645 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
13646 | struct tnum true_64off = true_reg->var_off; | |
13647 | s64 sval = (s64)val; | |
13648 | s32 sval32 = (s32)val32; | |
a72dafaf | 13649 | |
f1174f77 EC |
13650 | /* If the dst_reg is a pointer, we can't learn anything about its |
13651 | * variable offset from the compare (unless src_reg were a pointer into | |
13652 | * the same object, but we don't bother with that. | |
13653 | * Since false_reg and true_reg have the same type by construction, we | |
13654 | * only need to check one of them for pointerness. | |
13655 | */ | |
13656 | if (__is_pointer_value(false, false_reg)) | |
13657 | return; | |
4cabc5b1 | 13658 | |
48461135 | 13659 | switch (opcode) { |
a12ca627 DB |
13660 | /* JEQ/JNE comparison doesn't change the register equivalence. |
13661 | * | |
13662 | * r1 = r2; | |
13663 | * if (r1 == 42) goto label; | |
13664 | * ... | |
13665 | * label: // here both r1 and r2 are known to be 42. | |
13666 | * | |
13667 | * Hence when marking register as known preserve it's ID. | |
13668 | */ | |
48461135 | 13669 | case BPF_JEQ: |
a12ca627 DB |
13670 | if (is_jmp32) { |
13671 | __mark_reg32_known(true_reg, val32); | |
13672 | true_32off = tnum_subreg(true_reg->var_off); | |
13673 | } else { | |
13674 | ___mark_reg_known(true_reg, val); | |
13675 | true_64off = true_reg->var_off; | |
13676 | } | |
13677 | break; | |
48461135 | 13678 | case BPF_JNE: |
a12ca627 DB |
13679 | if (is_jmp32) { |
13680 | __mark_reg32_known(false_reg, val32); | |
13681 | false_32off = tnum_subreg(false_reg->var_off); | |
13682 | } else { | |
13683 | ___mark_reg_known(false_reg, val); | |
13684 | false_64off = false_reg->var_off; | |
13685 | } | |
48461135 | 13686 | break; |
960ea056 | 13687 | case BPF_JSET: |
3f50f132 JF |
13688 | if (is_jmp32) { |
13689 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
13690 | if (is_power_of_2(val32)) | |
13691 | true_32off = tnum_or(true_32off, | |
13692 | tnum_const(val32)); | |
13693 | } else { | |
13694 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
13695 | if (is_power_of_2(val)) | |
13696 | true_64off = tnum_or(true_64off, | |
13697 | tnum_const(val)); | |
13698 | } | |
960ea056 | 13699 | break; |
48461135 | 13700 | case BPF_JGE: |
a72dafaf JW |
13701 | case BPF_JGT: |
13702 | { | |
3f50f132 JF |
13703 | if (is_jmp32) { |
13704 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
13705 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
13706 | ||
13707 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
13708 | false_umax); | |
13709 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
13710 | true_umin); | |
13711 | } else { | |
13712 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
13713 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
13714 | ||
13715 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
13716 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
13717 | } | |
b03c9f9f | 13718 | break; |
a72dafaf | 13719 | } |
48461135 | 13720 | case BPF_JSGE: |
a72dafaf JW |
13721 | case BPF_JSGT: |
13722 | { | |
3f50f132 JF |
13723 | if (is_jmp32) { |
13724 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
13725 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 13726 | |
3f50f132 JF |
13727 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
13728 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
13729 | } else { | |
13730 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
13731 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
13732 | ||
13733 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
13734 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
13735 | } | |
48461135 | 13736 | break; |
a72dafaf | 13737 | } |
b4e432f1 | 13738 | case BPF_JLE: |
a72dafaf JW |
13739 | case BPF_JLT: |
13740 | { | |
3f50f132 JF |
13741 | if (is_jmp32) { |
13742 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
13743 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
13744 | ||
13745 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
13746 | false_umin); | |
13747 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
13748 | true_umax); | |
13749 | } else { | |
13750 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
13751 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
13752 | ||
13753 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
13754 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
13755 | } | |
b4e432f1 | 13756 | break; |
a72dafaf | 13757 | } |
b4e432f1 | 13758 | case BPF_JSLE: |
a72dafaf JW |
13759 | case BPF_JSLT: |
13760 | { | |
3f50f132 JF |
13761 | if (is_jmp32) { |
13762 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
13763 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 13764 | |
3f50f132 JF |
13765 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
13766 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
13767 | } else { | |
13768 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
13769 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
13770 | ||
13771 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
13772 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
13773 | } | |
b4e432f1 | 13774 | break; |
a72dafaf | 13775 | } |
48461135 | 13776 | default: |
0fc31b10 | 13777 | return; |
48461135 JB |
13778 | } |
13779 | ||
3f50f132 JF |
13780 | if (is_jmp32) { |
13781 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
13782 | tnum_subreg(false_32off)); | |
13783 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
13784 | tnum_subreg(true_32off)); | |
13785 | __reg_combine_32_into_64(false_reg); | |
13786 | __reg_combine_32_into_64(true_reg); | |
13787 | } else { | |
13788 | false_reg->var_off = false_64off; | |
13789 | true_reg->var_off = true_64off; | |
13790 | __reg_combine_64_into_32(false_reg); | |
13791 | __reg_combine_64_into_32(true_reg); | |
13792 | } | |
48461135 JB |
13793 | } |
13794 | ||
f1174f77 EC |
13795 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
13796 | * the variable reg. | |
48461135 JB |
13797 | */ |
13798 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
13799 | struct bpf_reg_state *false_reg, |
13800 | u64 val, u32 val32, | |
092ed096 | 13801 | u8 opcode, bool is_jmp32) |
48461135 | 13802 | { |
6d94e741 | 13803 | opcode = flip_opcode(opcode); |
0fc31b10 JH |
13804 | /* This uses zero as "not present in table"; luckily the zero opcode, |
13805 | * BPF_JA, can't get here. | |
b03c9f9f | 13806 | */ |
0fc31b10 | 13807 | if (opcode) |
3f50f132 | 13808 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
13809 | } |
13810 | ||
13811 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
13812 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
13813 | struct bpf_reg_state *dst_reg) | |
13814 | { | |
b03c9f9f EC |
13815 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
13816 | dst_reg->umin_value); | |
13817 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
13818 | dst_reg->umax_value); | |
13819 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
13820 | dst_reg->smin_value); | |
13821 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
13822 | dst_reg->smax_value); | |
f1174f77 EC |
13823 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
13824 | dst_reg->var_off); | |
3844d153 DB |
13825 | reg_bounds_sync(src_reg); |
13826 | reg_bounds_sync(dst_reg); | |
f1174f77 EC |
13827 | } |
13828 | ||
13829 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
13830 | struct bpf_reg_state *true_dst, | |
13831 | struct bpf_reg_state *false_src, | |
13832 | struct bpf_reg_state *false_dst, | |
13833 | u8 opcode) | |
13834 | { | |
13835 | switch (opcode) { | |
13836 | case BPF_JEQ: | |
13837 | __reg_combine_min_max(true_src, true_dst); | |
13838 | break; | |
13839 | case BPF_JNE: | |
13840 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 13841 | break; |
4cabc5b1 | 13842 | } |
48461135 JB |
13843 | } |
13844 | ||
fd978bf7 JS |
13845 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
13846 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 13847 | bool is_null) |
57a09bf0 | 13848 | { |
c25b2ae1 | 13849 | if (type_may_be_null(reg->type) && reg->id == id && |
fca1aa75 | 13850 | (is_rcu_reg(reg) || !WARN_ON_ONCE(!reg->id))) { |
df57f38a KKD |
13851 | /* Old offset (both fixed and variable parts) should have been |
13852 | * known-zero, because we don't allow pointer arithmetic on | |
13853 | * pointers that might be NULL. If we see this happening, don't | |
13854 | * convert the register. | |
13855 | * | |
13856 | * But in some cases, some helpers that return local kptrs | |
13857 | * advance offset for the returned pointer. In those cases, it | |
13858 | * is fine to expect to see reg->off. | |
13859 | */ | |
13860 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || !tnum_equals_const(reg->var_off, 0))) | |
13861 | return; | |
6a3cd331 DM |
13862 | if (!(type_is_ptr_alloc_obj(reg->type) || type_is_non_owning_ref(reg->type)) && |
13863 | WARN_ON_ONCE(reg->off)) | |
e60b0d12 | 13864 | return; |
6a3cd331 | 13865 | |
f1174f77 EC |
13866 | if (is_null) { |
13867 | reg->type = SCALAR_VALUE; | |
1b986589 MKL |
13868 | /* We don't need id and ref_obj_id from this point |
13869 | * onwards anymore, thus we should better reset it, | |
13870 | * so that state pruning has chances to take effect. | |
13871 | */ | |
13872 | reg->id = 0; | |
13873 | reg->ref_obj_id = 0; | |
4ddb7416 DB |
13874 | |
13875 | return; | |
13876 | } | |
13877 | ||
13878 | mark_ptr_not_null_reg(reg); | |
13879 | ||
13880 | if (!reg_may_point_to_spin_lock(reg)) { | |
1b986589 | 13881 | /* For not-NULL ptr, reg->ref_obj_id will be reset |
b239da34 | 13882 | * in release_reference(). |
1b986589 MKL |
13883 | * |
13884 | * reg->id is still used by spin_lock ptr. Other | |
13885 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
13886 | */ |
13887 | reg->id = 0; | |
56f668df | 13888 | } |
57a09bf0 TG |
13889 | } |
13890 | } | |
13891 | ||
13892 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
13893 | * be folded together at some point. | |
13894 | */ | |
840b9615 JS |
13895 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
13896 | bool is_null) | |
57a09bf0 | 13897 | { |
f4d7e40a | 13898 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
b239da34 | 13899 | struct bpf_reg_state *regs = state->regs, *reg; |
1b986589 | 13900 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 13901 | u32 id = regs[regno].id; |
57a09bf0 | 13902 | |
1b986589 MKL |
13903 | if (ref_obj_id && ref_obj_id == id && is_null) |
13904 | /* regs[regno] is in the " == NULL" branch. | |
13905 | * No one could have freed the reference state before | |
13906 | * doing the NULL check. | |
13907 | */ | |
13908 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 13909 | |
b239da34 KKD |
13910 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
13911 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
13912 | })); | |
57a09bf0 TG |
13913 | } |
13914 | ||
5beca081 DB |
13915 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
13916 | struct bpf_reg_state *dst_reg, | |
13917 | struct bpf_reg_state *src_reg, | |
13918 | struct bpf_verifier_state *this_branch, | |
13919 | struct bpf_verifier_state *other_branch) | |
13920 | { | |
13921 | if (BPF_SRC(insn->code) != BPF_X) | |
13922 | return false; | |
13923 | ||
092ed096 JW |
13924 | /* Pointers are always 64-bit. */ |
13925 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
13926 | return false; | |
13927 | ||
5beca081 DB |
13928 | switch (BPF_OP(insn->code)) { |
13929 | case BPF_JGT: | |
13930 | if ((dst_reg->type == PTR_TO_PACKET && | |
13931 | src_reg->type == PTR_TO_PACKET_END) || | |
13932 | (dst_reg->type == PTR_TO_PACKET_META && | |
13933 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
13934 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
13935 | find_good_pkt_pointers(this_branch, dst_reg, | |
13936 | dst_reg->type, false); | |
6d94e741 | 13937 | mark_pkt_end(other_branch, insn->dst_reg, true); |
5beca081 DB |
13938 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
13939 | src_reg->type == PTR_TO_PACKET) || | |
13940 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
13941 | src_reg->type == PTR_TO_PACKET_META)) { | |
13942 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
13943 | find_good_pkt_pointers(other_branch, src_reg, | |
13944 | src_reg->type, true); | |
6d94e741 | 13945 | mark_pkt_end(this_branch, insn->src_reg, false); |
5beca081 DB |
13946 | } else { |
13947 | return false; | |
13948 | } | |
13949 | break; | |
13950 | case BPF_JLT: | |
13951 | if ((dst_reg->type == PTR_TO_PACKET && | |
13952 | src_reg->type == PTR_TO_PACKET_END) || | |
13953 | (dst_reg->type == PTR_TO_PACKET_META && | |
13954 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
13955 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
13956 | find_good_pkt_pointers(other_branch, dst_reg, | |
13957 | dst_reg->type, true); | |
6d94e741 | 13958 | mark_pkt_end(this_branch, insn->dst_reg, false); |
5beca081 DB |
13959 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
13960 | src_reg->type == PTR_TO_PACKET) || | |
13961 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
13962 | src_reg->type == PTR_TO_PACKET_META)) { | |
13963 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
13964 | find_good_pkt_pointers(this_branch, src_reg, | |
13965 | src_reg->type, false); | |
6d94e741 | 13966 | mark_pkt_end(other_branch, insn->src_reg, true); |
5beca081 DB |
13967 | } else { |
13968 | return false; | |
13969 | } | |
13970 | break; | |
13971 | case BPF_JGE: | |
13972 | if ((dst_reg->type == PTR_TO_PACKET && | |
13973 | src_reg->type == PTR_TO_PACKET_END) || | |
13974 | (dst_reg->type == PTR_TO_PACKET_META && | |
13975 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
13976 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
13977 | find_good_pkt_pointers(this_branch, dst_reg, | |
13978 | dst_reg->type, true); | |
6d94e741 | 13979 | mark_pkt_end(other_branch, insn->dst_reg, false); |
5beca081 DB |
13980 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
13981 | src_reg->type == PTR_TO_PACKET) || | |
13982 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
13983 | src_reg->type == PTR_TO_PACKET_META)) { | |
13984 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
13985 | find_good_pkt_pointers(other_branch, src_reg, | |
13986 | src_reg->type, false); | |
6d94e741 | 13987 | mark_pkt_end(this_branch, insn->src_reg, true); |
5beca081 DB |
13988 | } else { |
13989 | return false; | |
13990 | } | |
13991 | break; | |
13992 | case BPF_JLE: | |
13993 | if ((dst_reg->type == PTR_TO_PACKET && | |
13994 | src_reg->type == PTR_TO_PACKET_END) || | |
13995 | (dst_reg->type == PTR_TO_PACKET_META && | |
13996 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
13997 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
13998 | find_good_pkt_pointers(other_branch, dst_reg, | |
13999 | dst_reg->type, false); | |
6d94e741 | 14000 | mark_pkt_end(this_branch, insn->dst_reg, true); |
5beca081 DB |
14001 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
14002 | src_reg->type == PTR_TO_PACKET) || | |
14003 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
14004 | src_reg->type == PTR_TO_PACKET_META)) { | |
14005 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
14006 | find_good_pkt_pointers(this_branch, src_reg, | |
14007 | src_reg->type, true); | |
6d94e741 | 14008 | mark_pkt_end(other_branch, insn->src_reg, false); |
5beca081 DB |
14009 | } else { |
14010 | return false; | |
14011 | } | |
14012 | break; | |
14013 | default: | |
14014 | return false; | |
14015 | } | |
14016 | ||
14017 | return true; | |
14018 | } | |
14019 | ||
75748837 AS |
14020 | static void find_equal_scalars(struct bpf_verifier_state *vstate, |
14021 | struct bpf_reg_state *known_reg) | |
14022 | { | |
14023 | struct bpf_func_state *state; | |
14024 | struct bpf_reg_state *reg; | |
75748837 | 14025 | |
b239da34 KKD |
14026 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
14027 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
71f656a5 | 14028 | copy_register_state(reg, known_reg); |
b239da34 | 14029 | })); |
75748837 AS |
14030 | } |
14031 | ||
58e2af8b | 14032 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
14033 | struct bpf_insn *insn, int *insn_idx) |
14034 | { | |
f4d7e40a AS |
14035 | struct bpf_verifier_state *this_branch = env->cur_state; |
14036 | struct bpf_verifier_state *other_branch; | |
14037 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 14038 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
befae758 | 14039 | struct bpf_reg_state *eq_branch_regs; |
17a52670 | 14040 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 14041 | bool is_jmp32; |
fb8d251e | 14042 | int pred = -1; |
17a52670 AS |
14043 | int err; |
14044 | ||
092ed096 JW |
14045 | /* Only conditional jumps are expected to reach here. */ |
14046 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
14047 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
14048 | return -EINVAL; |
14049 | } | |
14050 | ||
d75e30dd YS |
14051 | /* check src2 operand */ |
14052 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); | |
14053 | if (err) | |
14054 | return err; | |
14055 | ||
14056 | dst_reg = ®s[insn->dst_reg]; | |
17a52670 AS |
14057 | if (BPF_SRC(insn->code) == BPF_X) { |
14058 | if (insn->imm != 0) { | |
092ed096 | 14059 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
14060 | return -EINVAL; |
14061 | } | |
14062 | ||
14063 | /* check src1 operand */ | |
dc503a8a | 14064 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
14065 | if (err) |
14066 | return err; | |
1be7f75d | 14067 | |
d75e30dd YS |
14068 | src_reg = ®s[insn->src_reg]; |
14069 | if (!(reg_is_pkt_pointer_any(dst_reg) && reg_is_pkt_pointer_any(src_reg)) && | |
14070 | is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 14071 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
14072 | insn->src_reg); |
14073 | return -EACCES; | |
14074 | } | |
17a52670 AS |
14075 | } else { |
14076 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 14077 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
14078 | return -EINVAL; |
14079 | } | |
14080 | } | |
14081 | ||
092ed096 | 14082 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 14083 | |
3f50f132 JF |
14084 | if (BPF_SRC(insn->code) == BPF_K) { |
14085 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
14086 | } else if (src_reg->type == SCALAR_VALUE && | |
14087 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
14088 | pred = is_branch_taken(dst_reg, | |
14089 | tnum_subreg(src_reg->var_off).value, | |
14090 | opcode, | |
14091 | is_jmp32); | |
14092 | } else if (src_reg->type == SCALAR_VALUE && | |
14093 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
14094 | pred = is_branch_taken(dst_reg, | |
14095 | src_reg->var_off.value, | |
14096 | opcode, | |
14097 | is_jmp32); | |
953d9f5b YS |
14098 | } else if (dst_reg->type == SCALAR_VALUE && |
14099 | is_jmp32 && tnum_is_const(tnum_subreg(dst_reg->var_off))) { | |
14100 | pred = is_branch_taken(src_reg, | |
14101 | tnum_subreg(dst_reg->var_off).value, | |
14102 | flip_opcode(opcode), | |
14103 | is_jmp32); | |
14104 | } else if (dst_reg->type == SCALAR_VALUE && | |
14105 | !is_jmp32 && tnum_is_const(dst_reg->var_off)) { | |
14106 | pred = is_branch_taken(src_reg, | |
14107 | dst_reg->var_off.value, | |
14108 | flip_opcode(opcode), | |
14109 | is_jmp32); | |
6d94e741 AS |
14110 | } else if (reg_is_pkt_pointer_any(dst_reg) && |
14111 | reg_is_pkt_pointer_any(src_reg) && | |
14112 | !is_jmp32) { | |
14113 | pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode); | |
3f50f132 JF |
14114 | } |
14115 | ||
b5dc0163 | 14116 | if (pred >= 0) { |
cac616db JF |
14117 | /* If we get here with a dst_reg pointer type it is because |
14118 | * above is_branch_taken() special cased the 0 comparison. | |
14119 | */ | |
14120 | if (!__is_pointer_value(false, dst_reg)) | |
14121 | err = mark_chain_precision(env, insn->dst_reg); | |
6d94e741 AS |
14122 | if (BPF_SRC(insn->code) == BPF_X && !err && |
14123 | !__is_pointer_value(false, src_reg)) | |
b5dc0163 AS |
14124 | err = mark_chain_precision(env, insn->src_reg); |
14125 | if (err) | |
14126 | return err; | |
14127 | } | |
9183671a | 14128 | |
fb8d251e | 14129 | if (pred == 1) { |
9183671a DB |
14130 | /* Only follow the goto, ignore fall-through. If needed, push |
14131 | * the fall-through branch for simulation under speculative | |
14132 | * execution. | |
14133 | */ | |
14134 | if (!env->bypass_spec_v1 && | |
14135 | !sanitize_speculative_path(env, insn, *insn_idx + 1, | |
14136 | *insn_idx)) | |
14137 | return -EFAULT; | |
fb8d251e AS |
14138 | *insn_idx += insn->off; |
14139 | return 0; | |
14140 | } else if (pred == 0) { | |
9183671a DB |
14141 | /* Only follow the fall-through branch, since that's where the |
14142 | * program will go. If needed, push the goto branch for | |
14143 | * simulation under speculative execution. | |
fb8d251e | 14144 | */ |
9183671a DB |
14145 | if (!env->bypass_spec_v1 && |
14146 | !sanitize_speculative_path(env, insn, | |
14147 | *insn_idx + insn->off + 1, | |
14148 | *insn_idx)) | |
14149 | return -EFAULT; | |
fb8d251e | 14150 | return 0; |
17a52670 AS |
14151 | } |
14152 | ||
979d63d5 DB |
14153 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
14154 | false); | |
17a52670 AS |
14155 | if (!other_branch) |
14156 | return -EFAULT; | |
f4d7e40a | 14157 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 14158 | |
48461135 JB |
14159 | /* detect if we are comparing against a constant value so we can adjust |
14160 | * our min/max values for our dst register. | |
f1174f77 | 14161 | * this is only legit if both are scalars (or pointers to the same |
befae758 EZ |
14162 | * object, I suppose, see the PTR_MAYBE_NULL related if block below), |
14163 | * because otherwise the different base pointers mean the offsets aren't | |
f1174f77 | 14164 | * comparable. |
48461135 JB |
14165 | */ |
14166 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 14167 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 14168 | |
f1174f77 | 14169 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
14170 | src_reg->type == SCALAR_VALUE) { |
14171 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
14172 | (is_jmp32 && |
14173 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 14174 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 14175 | dst_reg, |
3f50f132 JF |
14176 | src_reg->var_off.value, |
14177 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
14178 | opcode, is_jmp32); |
14179 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
14180 | (is_jmp32 && |
14181 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 14182 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 14183 | src_reg, |
3f50f132 JF |
14184 | dst_reg->var_off.value, |
14185 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
14186 | opcode, is_jmp32); |
14187 | else if (!is_jmp32 && | |
14188 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 14189 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
14190 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
14191 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 14192 | src_reg, dst_reg, opcode); |
e688c3db AS |
14193 | if (src_reg->id && |
14194 | !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) { | |
75748837 AS |
14195 | find_equal_scalars(this_branch, src_reg); |
14196 | find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]); | |
14197 | } | |
14198 | ||
f1174f77 EC |
14199 | } |
14200 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 14201 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
14202 | dst_reg, insn->imm, (u32)insn->imm, |
14203 | opcode, is_jmp32); | |
48461135 JB |
14204 | } |
14205 | ||
e688c3db AS |
14206 | if (dst_reg->type == SCALAR_VALUE && dst_reg->id && |
14207 | !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) { | |
75748837 AS |
14208 | find_equal_scalars(this_branch, dst_reg); |
14209 | find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]); | |
14210 | } | |
14211 | ||
befae758 EZ |
14212 | /* if one pointer register is compared to another pointer |
14213 | * register check if PTR_MAYBE_NULL could be lifted. | |
14214 | * E.g. register A - maybe null | |
14215 | * register B - not null | |
14216 | * for JNE A, B, ... - A is not null in the false branch; | |
14217 | * for JEQ A, B, ... - A is not null in the true branch. | |
8374bfd5 HS |
14218 | * |
14219 | * Since PTR_TO_BTF_ID points to a kernel struct that does | |
14220 | * not need to be null checked by the BPF program, i.e., | |
14221 | * could be null even without PTR_MAYBE_NULL marking, so | |
14222 | * only propagate nullness when neither reg is that type. | |
befae758 EZ |
14223 | */ |
14224 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_X && | |
14225 | __is_pointer_value(false, src_reg) && __is_pointer_value(false, dst_reg) && | |
8374bfd5 HS |
14226 | type_may_be_null(src_reg->type) != type_may_be_null(dst_reg->type) && |
14227 | base_type(src_reg->type) != PTR_TO_BTF_ID && | |
14228 | base_type(dst_reg->type) != PTR_TO_BTF_ID) { | |
befae758 EZ |
14229 | eq_branch_regs = NULL; |
14230 | switch (opcode) { | |
14231 | case BPF_JEQ: | |
14232 | eq_branch_regs = other_branch_regs; | |
14233 | break; | |
14234 | case BPF_JNE: | |
14235 | eq_branch_regs = regs; | |
14236 | break; | |
14237 | default: | |
14238 | /* do nothing */ | |
14239 | break; | |
14240 | } | |
14241 | if (eq_branch_regs) { | |
14242 | if (type_may_be_null(src_reg->type)) | |
14243 | mark_ptr_not_null_reg(&eq_branch_regs[insn->src_reg]); | |
14244 | else | |
14245 | mark_ptr_not_null_reg(&eq_branch_regs[insn->dst_reg]); | |
14246 | } | |
14247 | } | |
14248 | ||
092ed096 JW |
14249 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
14250 | * NOTE: these optimizations below are related with pointer comparison | |
14251 | * which will never be JMP32. | |
14252 | */ | |
14253 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 14254 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
c25b2ae1 | 14255 | type_may_be_null(dst_reg->type)) { |
840b9615 | 14256 | /* Mark all identical registers in each branch as either |
57a09bf0 TG |
14257 | * safe or unknown depending R == 0 or R != 0 conditional. |
14258 | */ | |
840b9615 JS |
14259 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
14260 | opcode == BPF_JNE); | |
14261 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
14262 | opcode == BPF_JEQ); | |
5beca081 DB |
14263 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
14264 | this_branch, other_branch) && | |
14265 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
14266 | verbose(env, "R%d pointer comparison prohibited\n", |
14267 | insn->dst_reg); | |
1be7f75d | 14268 | return -EACCES; |
17a52670 | 14269 | } |
06ee7115 | 14270 | if (env->log.level & BPF_LOG_LEVEL) |
2e576648 | 14271 | print_insn_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
14272 | return 0; |
14273 | } | |
14274 | ||
17a52670 | 14275 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 14276 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 14277 | { |
d8eca5bb | 14278 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 14279 | struct bpf_reg_state *regs = cur_regs(env); |
4976b718 | 14280 | struct bpf_reg_state *dst_reg; |
d8eca5bb | 14281 | struct bpf_map *map; |
17a52670 AS |
14282 | int err; |
14283 | ||
14284 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 14285 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
14286 | return -EINVAL; |
14287 | } | |
14288 | if (insn->off != 0) { | |
61bd5218 | 14289 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
14290 | return -EINVAL; |
14291 | } | |
14292 | ||
dc503a8a | 14293 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
14294 | if (err) |
14295 | return err; | |
14296 | ||
4976b718 | 14297 | dst_reg = ®s[insn->dst_reg]; |
6b173873 | 14298 | if (insn->src_reg == 0) { |
6b173873 JK |
14299 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
14300 | ||
4976b718 | 14301 | dst_reg->type = SCALAR_VALUE; |
b03c9f9f | 14302 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 14303 | return 0; |
6b173873 | 14304 | } |
17a52670 | 14305 | |
d400a6cf DB |
14306 | /* All special src_reg cases are listed below. From this point onwards |
14307 | * we either succeed and assign a corresponding dst_reg->type after | |
14308 | * zeroing the offset, or fail and reject the program. | |
14309 | */ | |
14310 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
4976b718 | 14311 | |
d400a6cf | 14312 | if (insn->src_reg == BPF_PSEUDO_BTF_ID) { |
4976b718 | 14313 | dst_reg->type = aux->btf_var.reg_type; |
34d3a78c | 14314 | switch (base_type(dst_reg->type)) { |
4976b718 HL |
14315 | case PTR_TO_MEM: |
14316 | dst_reg->mem_size = aux->btf_var.mem_size; | |
14317 | break; | |
14318 | case PTR_TO_BTF_ID: | |
22dc4a0f | 14319 | dst_reg->btf = aux->btf_var.btf; |
4976b718 HL |
14320 | dst_reg->btf_id = aux->btf_var.btf_id; |
14321 | break; | |
14322 | default: | |
14323 | verbose(env, "bpf verifier is misconfigured\n"); | |
14324 | return -EFAULT; | |
14325 | } | |
14326 | return 0; | |
14327 | } | |
14328 | ||
69c087ba YS |
14329 | if (insn->src_reg == BPF_PSEUDO_FUNC) { |
14330 | struct bpf_prog_aux *aux = env->prog->aux; | |
3990ed4c MKL |
14331 | u32 subprogno = find_subprog(env, |
14332 | env->insn_idx + insn->imm + 1); | |
69c087ba YS |
14333 | |
14334 | if (!aux->func_info) { | |
14335 | verbose(env, "missing btf func_info\n"); | |
14336 | return -EINVAL; | |
14337 | } | |
14338 | if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) { | |
14339 | verbose(env, "callback function not static\n"); | |
14340 | return -EINVAL; | |
14341 | } | |
14342 | ||
14343 | dst_reg->type = PTR_TO_FUNC; | |
14344 | dst_reg->subprogno = subprogno; | |
14345 | return 0; | |
14346 | } | |
14347 | ||
d8eca5bb | 14348 | map = env->used_maps[aux->map_index]; |
4976b718 | 14349 | dst_reg->map_ptr = map; |
d8eca5bb | 14350 | |
387544bf AS |
14351 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE || |
14352 | insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE) { | |
4976b718 HL |
14353 | dst_reg->type = PTR_TO_MAP_VALUE; |
14354 | dst_reg->off = aux->map_off; | |
d0d78c1d KKD |
14355 | WARN_ON_ONCE(map->max_entries != 1); |
14356 | /* We want reg->id to be same (0) as map_value is not distinct */ | |
387544bf AS |
14357 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD || |
14358 | insn->src_reg == BPF_PSEUDO_MAP_IDX) { | |
4976b718 | 14359 | dst_reg->type = CONST_PTR_TO_MAP; |
d8eca5bb DB |
14360 | } else { |
14361 | verbose(env, "bpf verifier is misconfigured\n"); | |
14362 | return -EINVAL; | |
14363 | } | |
17a52670 | 14364 | |
17a52670 AS |
14365 | return 0; |
14366 | } | |
14367 | ||
96be4325 DB |
14368 | static bool may_access_skb(enum bpf_prog_type type) |
14369 | { | |
14370 | switch (type) { | |
14371 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
14372 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 14373 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
14374 | return true; |
14375 | default: | |
14376 | return false; | |
14377 | } | |
14378 | } | |
14379 | ||
ddd872bc AS |
14380 | /* verify safety of LD_ABS|LD_IND instructions: |
14381 | * - they can only appear in the programs where ctx == skb | |
14382 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
14383 | * preserve R6-R9, and store return value into R0 | |
14384 | * | |
14385 | * Implicit input: | |
14386 | * ctx == skb == R6 == CTX | |
14387 | * | |
14388 | * Explicit input: | |
14389 | * SRC == any register | |
14390 | * IMM == 32-bit immediate | |
14391 | * | |
14392 | * Output: | |
14393 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
14394 | */ | |
58e2af8b | 14395 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 14396 | { |
638f5b90 | 14397 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 14398 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 14399 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
14400 | int i, err; |
14401 | ||
7e40781c | 14402 | if (!may_access_skb(resolve_prog_type(env->prog))) { |
61bd5218 | 14403 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
14404 | return -EINVAL; |
14405 | } | |
14406 | ||
e0cea7ce DB |
14407 | if (!env->ops->gen_ld_abs) { |
14408 | verbose(env, "bpf verifier is misconfigured\n"); | |
14409 | return -EINVAL; | |
14410 | } | |
14411 | ||
ddd872bc | 14412 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 14413 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 14414 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 14415 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
14416 | return -EINVAL; |
14417 | } | |
14418 | ||
14419 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 14420 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
14421 | if (err) |
14422 | return err; | |
14423 | ||
fd978bf7 JS |
14424 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
14425 | * gen_ld_abs() may terminate the program at runtime, leading to | |
14426 | * reference leak. | |
14427 | */ | |
14428 | err = check_reference_leak(env); | |
14429 | if (err) { | |
14430 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
14431 | return err; | |
14432 | } | |
14433 | ||
d0d78c1d | 14434 | if (env->cur_state->active_lock.ptr) { |
d83525ca AS |
14435 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); |
14436 | return -EINVAL; | |
14437 | } | |
14438 | ||
9bb00b28 YS |
14439 | if (env->cur_state->active_rcu_lock) { |
14440 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_rcu_read_lock-ed region\n"); | |
14441 | return -EINVAL; | |
14442 | } | |
14443 | ||
6d4f151a | 14444 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
14445 | verbose(env, |
14446 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
14447 | return -EINVAL; |
14448 | } | |
14449 | ||
14450 | if (mode == BPF_IND) { | |
14451 | /* check explicit source operand */ | |
dc503a8a | 14452 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
14453 | if (err) |
14454 | return err; | |
14455 | } | |
14456 | ||
be80a1d3 | 14457 | err = check_ptr_off_reg(env, ®s[ctx_reg], ctx_reg); |
6d4f151a DB |
14458 | if (err < 0) |
14459 | return err; | |
14460 | ||
ddd872bc | 14461 | /* reset caller saved regs to unreadable */ |
dc503a8a | 14462 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 14463 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
14464 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
14465 | } | |
ddd872bc AS |
14466 | |
14467 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
14468 | * the value fetched from the packet. |
14469 | * Already marked as written above. | |
ddd872bc | 14470 | */ |
61bd5218 | 14471 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
14472 | /* ld_abs load up to 32-bit skb data. */ |
14473 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
14474 | return 0; |
14475 | } | |
14476 | ||
390ee7e2 AS |
14477 | static int check_return_code(struct bpf_verifier_env *env) |
14478 | { | |
5cf1e914 | 14479 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 14480 | const struct bpf_prog *prog = env->prog; |
390ee7e2 AS |
14481 | struct bpf_reg_state *reg; |
14482 | struct tnum range = tnum_range(0, 1); | |
7e40781c | 14483 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
27ae7997 | 14484 | int err; |
bfc6bb74 AS |
14485 | struct bpf_func_state *frame = env->cur_state->frame[0]; |
14486 | const bool is_subprog = frame->subprogno; | |
27ae7997 | 14487 | |
9e4e01df | 14488 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
d1a6edec SF |
14489 | if (!is_subprog) { |
14490 | switch (prog_type) { | |
14491 | case BPF_PROG_TYPE_LSM: | |
14492 | if (prog->expected_attach_type == BPF_LSM_CGROUP) | |
14493 | /* See below, can be 0 or 0-1 depending on hook. */ | |
14494 | break; | |
14495 | fallthrough; | |
14496 | case BPF_PROG_TYPE_STRUCT_OPS: | |
14497 | if (!prog->aux->attach_func_proto->type) | |
14498 | return 0; | |
14499 | break; | |
14500 | default: | |
14501 | break; | |
14502 | } | |
14503 | } | |
27ae7997 | 14504 | |
8fb33b60 | 14505 | /* eBPF calling convention is such that R0 is used |
27ae7997 MKL |
14506 | * to return the value from eBPF program. |
14507 | * Make sure that it's readable at this time | |
14508 | * of bpf_exit, which means that program wrote | |
14509 | * something into it earlier | |
14510 | */ | |
14511 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
14512 | if (err) | |
14513 | return err; | |
14514 | ||
14515 | if (is_pointer_value(env, BPF_REG_0)) { | |
14516 | verbose(env, "R0 leaks addr as return value\n"); | |
14517 | return -EACCES; | |
14518 | } | |
390ee7e2 | 14519 | |
f782e2c3 | 14520 | reg = cur_regs(env) + BPF_REG_0; |
bfc6bb74 AS |
14521 | |
14522 | if (frame->in_async_callback_fn) { | |
14523 | /* enforce return zero from async callbacks like timer */ | |
14524 | if (reg->type != SCALAR_VALUE) { | |
14525 | verbose(env, "In async callback the register R0 is not a known value (%s)\n", | |
c25b2ae1 | 14526 | reg_type_str(env, reg->type)); |
bfc6bb74 AS |
14527 | return -EINVAL; |
14528 | } | |
14529 | ||
14530 | if (!tnum_in(tnum_const(0), reg->var_off)) { | |
14531 | verbose_invalid_scalar(env, reg, &range, "async callback", "R0"); | |
14532 | return -EINVAL; | |
14533 | } | |
14534 | return 0; | |
14535 | } | |
14536 | ||
f782e2c3 DB |
14537 | if (is_subprog) { |
14538 | if (reg->type != SCALAR_VALUE) { | |
14539 | verbose(env, "At subprogram exit the register R0 is not a scalar value (%s)\n", | |
c25b2ae1 | 14540 | reg_type_str(env, reg->type)); |
f782e2c3 DB |
14541 | return -EINVAL; |
14542 | } | |
14543 | return 0; | |
14544 | } | |
14545 | ||
7e40781c | 14546 | switch (prog_type) { |
983695fa DB |
14547 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
14548 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 DB |
14549 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
14550 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || | |
14551 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
14552 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || | |
14553 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME) | |
983695fa | 14554 | range = tnum_range(1, 1); |
77241217 SF |
14555 | if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND || |
14556 | env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND) | |
14557 | range = tnum_range(0, 3); | |
ed4ed404 | 14558 | break; |
390ee7e2 | 14559 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 14560 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
14561 | range = tnum_range(0, 3); | |
14562 | enforce_attach_type_range = tnum_range(2, 3); | |
14563 | } | |
ed4ed404 | 14564 | break; |
390ee7e2 AS |
14565 | case BPF_PROG_TYPE_CGROUP_SOCK: |
14566 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 14567 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 14568 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 14569 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 14570 | break; |
15ab09bd AS |
14571 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
14572 | if (!env->prog->aux->attach_btf_id) | |
14573 | return 0; | |
14574 | range = tnum_const(0); | |
14575 | break; | |
15d83c4d | 14576 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
14577 | switch (env->prog->expected_attach_type) { |
14578 | case BPF_TRACE_FENTRY: | |
14579 | case BPF_TRACE_FEXIT: | |
14580 | range = tnum_const(0); | |
14581 | break; | |
14582 | case BPF_TRACE_RAW_TP: | |
14583 | case BPF_MODIFY_RETURN: | |
15d83c4d | 14584 | return 0; |
2ec0616e DB |
14585 | case BPF_TRACE_ITER: |
14586 | break; | |
e92888c7 YS |
14587 | default: |
14588 | return -ENOTSUPP; | |
14589 | } | |
15d83c4d | 14590 | break; |
e9ddbb77 JS |
14591 | case BPF_PROG_TYPE_SK_LOOKUP: |
14592 | range = tnum_range(SK_DROP, SK_PASS); | |
14593 | break; | |
69fd337a SF |
14594 | |
14595 | case BPF_PROG_TYPE_LSM: | |
14596 | if (env->prog->expected_attach_type != BPF_LSM_CGROUP) { | |
14597 | /* Regular BPF_PROG_TYPE_LSM programs can return | |
14598 | * any value. | |
14599 | */ | |
14600 | return 0; | |
14601 | } | |
14602 | if (!env->prog->aux->attach_func_proto->type) { | |
14603 | /* Make sure programs that attach to void | |
14604 | * hooks don't try to modify return value. | |
14605 | */ | |
14606 | range = tnum_range(1, 1); | |
14607 | } | |
14608 | break; | |
14609 | ||
fd9c663b FW |
14610 | case BPF_PROG_TYPE_NETFILTER: |
14611 | range = tnum_range(NF_DROP, NF_ACCEPT); | |
14612 | break; | |
e92888c7 YS |
14613 | case BPF_PROG_TYPE_EXT: |
14614 | /* freplace program can return anything as its return value | |
14615 | * depends on the to-be-replaced kernel func or bpf program. | |
14616 | */ | |
390ee7e2 AS |
14617 | default: |
14618 | return 0; | |
14619 | } | |
14620 | ||
390ee7e2 | 14621 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 14622 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
c25b2ae1 | 14623 | reg_type_str(env, reg->type)); |
390ee7e2 AS |
14624 | return -EINVAL; |
14625 | } | |
14626 | ||
14627 | if (!tnum_in(range, reg->var_off)) { | |
bc2591d6 | 14628 | verbose_invalid_scalar(env, reg, &range, "program exit", "R0"); |
69fd337a | 14629 | if (prog->expected_attach_type == BPF_LSM_CGROUP && |
d1a6edec | 14630 | prog_type == BPF_PROG_TYPE_LSM && |
69fd337a SF |
14631 | !prog->aux->attach_func_proto->type) |
14632 | verbose(env, "Note, BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
390ee7e2 AS |
14633 | return -EINVAL; |
14634 | } | |
5cf1e914 | 14635 | |
14636 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
14637 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
14638 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
14639 | return 0; |
14640 | } | |
14641 | ||
475fb78f AS |
14642 | /* non-recursive DFS pseudo code |
14643 | * 1 procedure DFS-iterative(G,v): | |
14644 | * 2 label v as discovered | |
14645 | * 3 let S be a stack | |
14646 | * 4 S.push(v) | |
14647 | * 5 while S is not empty | |
b6d20799 | 14648 | * 6 t <- S.peek() |
475fb78f AS |
14649 | * 7 if t is what we're looking for: |
14650 | * 8 return t | |
14651 | * 9 for all edges e in G.adjacentEdges(t) do | |
14652 | * 10 if edge e is already labelled | |
14653 | * 11 continue with the next edge | |
14654 | * 12 w <- G.adjacentVertex(t,e) | |
14655 | * 13 if vertex w is not discovered and not explored | |
14656 | * 14 label e as tree-edge | |
14657 | * 15 label w as discovered | |
14658 | * 16 S.push(w) | |
14659 | * 17 continue at 5 | |
14660 | * 18 else if vertex w is discovered | |
14661 | * 19 label e as back-edge | |
14662 | * 20 else | |
14663 | * 21 // vertex w is explored | |
14664 | * 22 label e as forward- or cross-edge | |
14665 | * 23 label t as explored | |
14666 | * 24 S.pop() | |
14667 | * | |
14668 | * convention: | |
14669 | * 0x10 - discovered | |
14670 | * 0x11 - discovered and fall-through edge labelled | |
14671 | * 0x12 - discovered and fall-through and branch edges labelled | |
14672 | * 0x20 - explored | |
14673 | */ | |
14674 | ||
14675 | enum { | |
14676 | DISCOVERED = 0x10, | |
14677 | EXPLORED = 0x20, | |
14678 | FALLTHROUGH = 1, | |
14679 | BRANCH = 2, | |
14680 | }; | |
14681 | ||
dc2a4ebc AS |
14682 | static u32 state_htab_size(struct bpf_verifier_env *env) |
14683 | { | |
14684 | return env->prog->len; | |
14685 | } | |
14686 | ||
5d839021 AS |
14687 | static struct bpf_verifier_state_list **explored_state( |
14688 | struct bpf_verifier_env *env, | |
14689 | int idx) | |
14690 | { | |
dc2a4ebc AS |
14691 | struct bpf_verifier_state *cur = env->cur_state; |
14692 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
14693 | ||
14694 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
14695 | } |
14696 | ||
bffdeaa8 | 14697 | static void mark_prune_point(struct bpf_verifier_env *env, int idx) |
5d839021 | 14698 | { |
a8f500af | 14699 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 14700 | } |
f1bca824 | 14701 | |
bffdeaa8 AN |
14702 | static bool is_prune_point(struct bpf_verifier_env *env, int insn_idx) |
14703 | { | |
14704 | return env->insn_aux_data[insn_idx].prune_point; | |
14705 | } | |
14706 | ||
4b5ce570 AN |
14707 | static void mark_force_checkpoint(struct bpf_verifier_env *env, int idx) |
14708 | { | |
14709 | env->insn_aux_data[idx].force_checkpoint = true; | |
14710 | } | |
14711 | ||
14712 | static bool is_force_checkpoint(struct bpf_verifier_env *env, int insn_idx) | |
14713 | { | |
14714 | return env->insn_aux_data[insn_idx].force_checkpoint; | |
14715 | } | |
14716 | ||
14717 | ||
59e2e27d WAF |
14718 | enum { |
14719 | DONE_EXPLORING = 0, | |
14720 | KEEP_EXPLORING = 1, | |
14721 | }; | |
14722 | ||
475fb78f AS |
14723 | /* t, w, e - match pseudo-code above: |
14724 | * t - index of current instruction | |
14725 | * w - next instruction | |
14726 | * e - edge | |
14727 | */ | |
2589726d AS |
14728 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env, |
14729 | bool loop_ok) | |
475fb78f | 14730 | { |
7df737e9 AS |
14731 | int *insn_stack = env->cfg.insn_stack; |
14732 | int *insn_state = env->cfg.insn_state; | |
14733 | ||
475fb78f | 14734 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
59e2e27d | 14735 | return DONE_EXPLORING; |
475fb78f AS |
14736 | |
14737 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
59e2e27d | 14738 | return DONE_EXPLORING; |
475fb78f AS |
14739 | |
14740 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 14741 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 14742 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
14743 | return -EINVAL; |
14744 | } | |
14745 | ||
bffdeaa8 | 14746 | if (e == BRANCH) { |
f1bca824 | 14747 | /* mark branch target for state pruning */ |
bffdeaa8 AN |
14748 | mark_prune_point(env, w); |
14749 | mark_jmp_point(env, w); | |
14750 | } | |
f1bca824 | 14751 | |
475fb78f AS |
14752 | if (insn_state[w] == 0) { |
14753 | /* tree-edge */ | |
14754 | insn_state[t] = DISCOVERED | e; | |
14755 | insn_state[w] = DISCOVERED; | |
7df737e9 | 14756 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 14757 | return -E2BIG; |
7df737e9 | 14758 | insn_stack[env->cfg.cur_stack++] = w; |
59e2e27d | 14759 | return KEEP_EXPLORING; |
475fb78f | 14760 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { |
2c78ee89 | 14761 | if (loop_ok && env->bpf_capable) |
59e2e27d | 14762 | return DONE_EXPLORING; |
d9762e84 MKL |
14763 | verbose_linfo(env, t, "%d: ", t); |
14764 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 14765 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
14766 | return -EINVAL; |
14767 | } else if (insn_state[w] == EXPLORED) { | |
14768 | /* forward- or cross-edge */ | |
14769 | insn_state[t] = DISCOVERED | e; | |
14770 | } else { | |
61bd5218 | 14771 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
14772 | return -EFAULT; |
14773 | } | |
59e2e27d WAF |
14774 | return DONE_EXPLORING; |
14775 | } | |
14776 | ||
dcb2288b | 14777 | static int visit_func_call_insn(int t, struct bpf_insn *insns, |
efdb22de YS |
14778 | struct bpf_verifier_env *env, |
14779 | bool visit_callee) | |
14780 | { | |
14781 | int ret; | |
14782 | ||
14783 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); | |
14784 | if (ret) | |
14785 | return ret; | |
14786 | ||
618945fb AN |
14787 | mark_prune_point(env, t + 1); |
14788 | /* when we exit from subprog, we need to record non-linear history */ | |
14789 | mark_jmp_point(env, t + 1); | |
14790 | ||
efdb22de | 14791 | if (visit_callee) { |
bffdeaa8 | 14792 | mark_prune_point(env, t); |
86fc6ee6 AS |
14793 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env, |
14794 | /* It's ok to allow recursion from CFG point of | |
14795 | * view. __check_func_call() will do the actual | |
14796 | * check. | |
14797 | */ | |
14798 | bpf_pseudo_func(insns + t)); | |
efdb22de YS |
14799 | } |
14800 | return ret; | |
14801 | } | |
14802 | ||
59e2e27d WAF |
14803 | /* Visits the instruction at index t and returns one of the following: |
14804 | * < 0 - an error occurred | |
14805 | * DONE_EXPLORING - the instruction was fully explored | |
14806 | * KEEP_EXPLORING - there is still work to be done before it is fully explored | |
14807 | */ | |
dcb2288b | 14808 | static int visit_insn(int t, struct bpf_verifier_env *env) |
59e2e27d | 14809 | { |
653ae3a8 | 14810 | struct bpf_insn *insns = env->prog->insnsi, *insn = &insns[t]; |
4cd58e9a | 14811 | int ret, off; |
59e2e27d | 14812 | |
653ae3a8 | 14813 | if (bpf_pseudo_func(insn)) |
dcb2288b | 14814 | return visit_func_call_insn(t, insns, env, true); |
69c087ba | 14815 | |
59e2e27d | 14816 | /* All non-branch instructions have a single fall-through edge. */ |
653ae3a8 AN |
14817 | if (BPF_CLASS(insn->code) != BPF_JMP && |
14818 | BPF_CLASS(insn->code) != BPF_JMP32) | |
59e2e27d WAF |
14819 | return push_insn(t, t + 1, FALLTHROUGH, env, false); |
14820 | ||
653ae3a8 | 14821 | switch (BPF_OP(insn->code)) { |
59e2e27d WAF |
14822 | case BPF_EXIT: |
14823 | return DONE_EXPLORING; | |
14824 | ||
14825 | case BPF_CALL: | |
c1ee85a9 | 14826 | if (insn->src_reg == 0 && insn->imm == BPF_FUNC_timer_set_callback) |
618945fb AN |
14827 | /* Mark this call insn as a prune point to trigger |
14828 | * is_state_visited() check before call itself is | |
14829 | * processed by __check_func_call(). Otherwise new | |
14830 | * async state will be pushed for further exploration. | |
bfc6bb74 | 14831 | */ |
bffdeaa8 | 14832 | mark_prune_point(env, t); |
06accc87 AN |
14833 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
14834 | struct bpf_kfunc_call_arg_meta meta; | |
14835 | ||
14836 | ret = fetch_kfunc_meta(env, insn, &meta, NULL); | |
4b5ce570 | 14837 | if (ret == 0 && is_iter_next_kfunc(&meta)) { |
06accc87 | 14838 | mark_prune_point(env, t); |
4b5ce570 AN |
14839 | /* Checking and saving state checkpoints at iter_next() call |
14840 | * is crucial for fast convergence of open-coded iterator loop | |
14841 | * logic, so we need to force it. If we don't do that, | |
14842 | * is_state_visited() might skip saving a checkpoint, causing | |
14843 | * unnecessarily long sequence of not checkpointed | |
14844 | * instructions and jumps, leading to exhaustion of jump | |
14845 | * history buffer, and potentially other undesired outcomes. | |
14846 | * It is expected that with correct open-coded iterators | |
14847 | * convergence will happen quickly, so we don't run a risk of | |
14848 | * exhausting memory. | |
14849 | */ | |
14850 | mark_force_checkpoint(env, t); | |
14851 | } | |
06accc87 | 14852 | } |
653ae3a8 | 14853 | return visit_func_call_insn(t, insns, env, insn->src_reg == BPF_PSEUDO_CALL); |
59e2e27d WAF |
14854 | |
14855 | case BPF_JA: | |
653ae3a8 | 14856 | if (BPF_SRC(insn->code) != BPF_K) |
59e2e27d WAF |
14857 | return -EINVAL; |
14858 | ||
4cd58e9a YS |
14859 | if (BPF_CLASS(insn->code) == BPF_JMP) |
14860 | off = insn->off; | |
14861 | else | |
14862 | off = insn->imm; | |
14863 | ||
59e2e27d | 14864 | /* unconditional jump with single edge */ |
4cd58e9a | 14865 | ret = push_insn(t, t + off + 1, FALLTHROUGH, env, |
59e2e27d WAF |
14866 | true); |
14867 | if (ret) | |
14868 | return ret; | |
14869 | ||
4cd58e9a YS |
14870 | mark_prune_point(env, t + off + 1); |
14871 | mark_jmp_point(env, t + off + 1); | |
59e2e27d WAF |
14872 | |
14873 | return ret; | |
14874 | ||
14875 | default: | |
14876 | /* conditional jump with two edges */ | |
bffdeaa8 | 14877 | mark_prune_point(env, t); |
618945fb | 14878 | |
59e2e27d WAF |
14879 | ret = push_insn(t, t + 1, FALLTHROUGH, env, true); |
14880 | if (ret) | |
14881 | return ret; | |
14882 | ||
653ae3a8 | 14883 | return push_insn(t, t + insn->off + 1, BRANCH, env, true); |
59e2e27d | 14884 | } |
475fb78f AS |
14885 | } |
14886 | ||
14887 | /* non-recursive depth-first-search to detect loops in BPF program | |
14888 | * loop == back-edge in directed graph | |
14889 | */ | |
58e2af8b | 14890 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f | 14891 | { |
475fb78f | 14892 | int insn_cnt = env->prog->len; |
7df737e9 | 14893 | int *insn_stack, *insn_state; |
475fb78f | 14894 | int ret = 0; |
59e2e27d | 14895 | int i; |
475fb78f | 14896 | |
7df737e9 | 14897 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
14898 | if (!insn_state) |
14899 | return -ENOMEM; | |
14900 | ||
7df737e9 | 14901 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 14902 | if (!insn_stack) { |
71dde681 | 14903 | kvfree(insn_state); |
475fb78f AS |
14904 | return -ENOMEM; |
14905 | } | |
14906 | ||
14907 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
14908 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 14909 | env->cfg.cur_stack = 1; |
475fb78f | 14910 | |
59e2e27d WAF |
14911 | while (env->cfg.cur_stack > 0) { |
14912 | int t = insn_stack[env->cfg.cur_stack - 1]; | |
475fb78f | 14913 | |
dcb2288b | 14914 | ret = visit_insn(t, env); |
59e2e27d WAF |
14915 | switch (ret) { |
14916 | case DONE_EXPLORING: | |
14917 | insn_state[t] = EXPLORED; | |
14918 | env->cfg.cur_stack--; | |
14919 | break; | |
14920 | case KEEP_EXPLORING: | |
14921 | break; | |
14922 | default: | |
14923 | if (ret > 0) { | |
14924 | verbose(env, "visit_insn internal bug\n"); | |
14925 | ret = -EFAULT; | |
475fb78f | 14926 | } |
475fb78f | 14927 | goto err_free; |
59e2e27d | 14928 | } |
475fb78f AS |
14929 | } |
14930 | ||
59e2e27d | 14931 | if (env->cfg.cur_stack < 0) { |
61bd5218 | 14932 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
14933 | ret = -EFAULT; |
14934 | goto err_free; | |
14935 | } | |
475fb78f | 14936 | |
475fb78f AS |
14937 | for (i = 0; i < insn_cnt; i++) { |
14938 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 14939 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
14940 | ret = -EINVAL; |
14941 | goto err_free; | |
14942 | } | |
14943 | } | |
14944 | ret = 0; /* cfg looks good */ | |
14945 | ||
14946 | err_free: | |
71dde681 AS |
14947 | kvfree(insn_state); |
14948 | kvfree(insn_stack); | |
7df737e9 | 14949 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
14950 | return ret; |
14951 | } | |
14952 | ||
09b28d76 AS |
14953 | static int check_abnormal_return(struct bpf_verifier_env *env) |
14954 | { | |
14955 | int i; | |
14956 | ||
14957 | for (i = 1; i < env->subprog_cnt; i++) { | |
14958 | if (env->subprog_info[i].has_ld_abs) { | |
14959 | verbose(env, "LD_ABS is not allowed in subprogs without BTF\n"); | |
14960 | return -EINVAL; | |
14961 | } | |
14962 | if (env->subprog_info[i].has_tail_call) { | |
14963 | verbose(env, "tail_call is not allowed in subprogs without BTF\n"); | |
14964 | return -EINVAL; | |
14965 | } | |
14966 | } | |
14967 | return 0; | |
14968 | } | |
14969 | ||
838e9690 YS |
14970 | /* The minimum supported BTF func info size */ |
14971 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
14972 | #define MAX_FUNCINFO_REC_SIZE 252 | |
14973 | ||
c454a46b MKL |
14974 | static int check_btf_func(struct bpf_verifier_env *env, |
14975 | const union bpf_attr *attr, | |
af2ac3e1 | 14976 | bpfptr_t uattr) |
838e9690 | 14977 | { |
09b28d76 | 14978 | const struct btf_type *type, *func_proto, *ret_type; |
d0b2818e | 14979 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 14980 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 14981 | struct bpf_func_info *krecord; |
8c1b6e69 | 14982 | struct bpf_func_info_aux *info_aux = NULL; |
c454a46b MKL |
14983 | struct bpf_prog *prog; |
14984 | const struct btf *btf; | |
af2ac3e1 | 14985 | bpfptr_t urecord; |
d0b2818e | 14986 | u32 prev_offset = 0; |
09b28d76 | 14987 | bool scalar_return; |
e7ed83d6 | 14988 | int ret = -ENOMEM; |
838e9690 YS |
14989 | |
14990 | nfuncs = attr->func_info_cnt; | |
09b28d76 AS |
14991 | if (!nfuncs) { |
14992 | if (check_abnormal_return(env)) | |
14993 | return -EINVAL; | |
838e9690 | 14994 | return 0; |
09b28d76 | 14995 | } |
838e9690 YS |
14996 | |
14997 | if (nfuncs != env->subprog_cnt) { | |
14998 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
14999 | return -EINVAL; | |
15000 | } | |
15001 | ||
15002 | urec_size = attr->func_info_rec_size; | |
15003 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
15004 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
15005 | urec_size % sizeof(u32)) { | |
15006 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
15007 | return -EINVAL; | |
15008 | } | |
15009 | ||
c454a46b MKL |
15010 | prog = env->prog; |
15011 | btf = prog->aux->btf; | |
838e9690 | 15012 | |
af2ac3e1 | 15013 | urecord = make_bpfptr(attr->func_info, uattr.is_kernel); |
838e9690 YS |
15014 | min_size = min_t(u32, krec_size, urec_size); |
15015 | ||
ba64e7d8 | 15016 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
15017 | if (!krecord) |
15018 | return -ENOMEM; | |
8c1b6e69 AS |
15019 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); |
15020 | if (!info_aux) | |
15021 | goto err_free; | |
ba64e7d8 | 15022 | |
838e9690 YS |
15023 | for (i = 0; i < nfuncs; i++) { |
15024 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
15025 | if (ret) { | |
15026 | if (ret == -E2BIG) { | |
15027 | verbose(env, "nonzero tailing record in func info"); | |
15028 | /* set the size kernel expects so loader can zero | |
15029 | * out the rest of the record. | |
15030 | */ | |
af2ac3e1 AS |
15031 | if (copy_to_bpfptr_offset(uattr, |
15032 | offsetof(union bpf_attr, func_info_rec_size), | |
15033 | &min_size, sizeof(min_size))) | |
838e9690 YS |
15034 | ret = -EFAULT; |
15035 | } | |
c454a46b | 15036 | goto err_free; |
838e9690 YS |
15037 | } |
15038 | ||
af2ac3e1 | 15039 | if (copy_from_bpfptr(&krecord[i], urecord, min_size)) { |
838e9690 | 15040 | ret = -EFAULT; |
c454a46b | 15041 | goto err_free; |
838e9690 YS |
15042 | } |
15043 | ||
d30d42e0 | 15044 | /* check insn_off */ |
09b28d76 | 15045 | ret = -EINVAL; |
838e9690 | 15046 | if (i == 0) { |
d30d42e0 | 15047 | if (krecord[i].insn_off) { |
838e9690 | 15048 | verbose(env, |
d30d42e0 MKL |
15049 | "nonzero insn_off %u for the first func info record", |
15050 | krecord[i].insn_off); | |
c454a46b | 15051 | goto err_free; |
838e9690 | 15052 | } |
d30d42e0 | 15053 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
15054 | verbose(env, |
15055 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 15056 | krecord[i].insn_off, prev_offset); |
c454a46b | 15057 | goto err_free; |
838e9690 YS |
15058 | } |
15059 | ||
d30d42e0 | 15060 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 | 15061 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
c454a46b | 15062 | goto err_free; |
838e9690 YS |
15063 | } |
15064 | ||
15065 | /* check type_id */ | |
ba64e7d8 | 15066 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 15067 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 15068 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 15069 | krecord[i].type_id); |
c454a46b | 15070 | goto err_free; |
838e9690 | 15071 | } |
51c39bb1 | 15072 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); |
09b28d76 AS |
15073 | |
15074 | func_proto = btf_type_by_id(btf, type->type); | |
15075 | if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto))) | |
15076 | /* btf_func_check() already verified it during BTF load */ | |
15077 | goto err_free; | |
15078 | ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL); | |
15079 | scalar_return = | |
6089fb32 | 15080 | btf_type_is_small_int(ret_type) || btf_is_any_enum(ret_type); |
09b28d76 AS |
15081 | if (i && !scalar_return && env->subprog_info[i].has_ld_abs) { |
15082 | verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n"); | |
15083 | goto err_free; | |
15084 | } | |
15085 | if (i && !scalar_return && env->subprog_info[i].has_tail_call) { | |
15086 | verbose(env, "tail_call is only allowed in functions that return 'int'.\n"); | |
15087 | goto err_free; | |
15088 | } | |
15089 | ||
d30d42e0 | 15090 | prev_offset = krecord[i].insn_off; |
af2ac3e1 | 15091 | bpfptr_add(&urecord, urec_size); |
838e9690 YS |
15092 | } |
15093 | ||
ba64e7d8 YS |
15094 | prog->aux->func_info = krecord; |
15095 | prog->aux->func_info_cnt = nfuncs; | |
8c1b6e69 | 15096 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
15097 | return 0; |
15098 | ||
c454a46b | 15099 | err_free: |
ba64e7d8 | 15100 | kvfree(krecord); |
8c1b6e69 | 15101 | kfree(info_aux); |
838e9690 YS |
15102 | return ret; |
15103 | } | |
15104 | ||
ba64e7d8 YS |
15105 | static void adjust_btf_func(struct bpf_verifier_env *env) |
15106 | { | |
8c1b6e69 | 15107 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
15108 | int i; |
15109 | ||
8c1b6e69 | 15110 | if (!aux->func_info) |
ba64e7d8 YS |
15111 | return; |
15112 | ||
15113 | for (i = 0; i < env->subprog_cnt; i++) | |
8c1b6e69 | 15114 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
15115 | } |
15116 | ||
1b773d00 | 15117 | #define MIN_BPF_LINEINFO_SIZE offsetofend(struct bpf_line_info, line_col) |
c454a46b MKL |
15118 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE |
15119 | ||
15120 | static int check_btf_line(struct bpf_verifier_env *env, | |
15121 | const union bpf_attr *attr, | |
af2ac3e1 | 15122 | bpfptr_t uattr) |
c454a46b MKL |
15123 | { |
15124 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
15125 | struct bpf_subprog_info *sub; | |
15126 | struct bpf_line_info *linfo; | |
15127 | struct bpf_prog *prog; | |
15128 | const struct btf *btf; | |
af2ac3e1 | 15129 | bpfptr_t ulinfo; |
c454a46b MKL |
15130 | int err; |
15131 | ||
15132 | nr_linfo = attr->line_info_cnt; | |
15133 | if (!nr_linfo) | |
15134 | return 0; | |
0e6491b5 BC |
15135 | if (nr_linfo > INT_MAX / sizeof(struct bpf_line_info)) |
15136 | return -EINVAL; | |
c454a46b MKL |
15137 | |
15138 | rec_size = attr->line_info_rec_size; | |
15139 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
15140 | rec_size > MAX_LINEINFO_REC_SIZE || | |
15141 | rec_size & (sizeof(u32) - 1)) | |
15142 | return -EINVAL; | |
15143 | ||
15144 | /* Need to zero it in case the userspace may | |
15145 | * pass in a smaller bpf_line_info object. | |
15146 | */ | |
15147 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
15148 | GFP_KERNEL | __GFP_NOWARN); | |
15149 | if (!linfo) | |
15150 | return -ENOMEM; | |
15151 | ||
15152 | prog = env->prog; | |
15153 | btf = prog->aux->btf; | |
15154 | ||
15155 | s = 0; | |
15156 | sub = env->subprog_info; | |
af2ac3e1 | 15157 | ulinfo = make_bpfptr(attr->line_info, uattr.is_kernel); |
c454a46b MKL |
15158 | expected_size = sizeof(struct bpf_line_info); |
15159 | ncopy = min_t(u32, expected_size, rec_size); | |
15160 | for (i = 0; i < nr_linfo; i++) { | |
15161 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
15162 | if (err) { | |
15163 | if (err == -E2BIG) { | |
15164 | verbose(env, "nonzero tailing record in line_info"); | |
af2ac3e1 AS |
15165 | if (copy_to_bpfptr_offset(uattr, |
15166 | offsetof(union bpf_attr, line_info_rec_size), | |
15167 | &expected_size, sizeof(expected_size))) | |
c454a46b MKL |
15168 | err = -EFAULT; |
15169 | } | |
15170 | goto err_free; | |
15171 | } | |
15172 | ||
af2ac3e1 | 15173 | if (copy_from_bpfptr(&linfo[i], ulinfo, ncopy)) { |
c454a46b MKL |
15174 | err = -EFAULT; |
15175 | goto err_free; | |
15176 | } | |
15177 | ||
15178 | /* | |
15179 | * Check insn_off to ensure | |
15180 | * 1) strictly increasing AND | |
15181 | * 2) bounded by prog->len | |
15182 | * | |
15183 | * The linfo[0].insn_off == 0 check logically falls into | |
15184 | * the later "missing bpf_line_info for func..." case | |
15185 | * because the first linfo[0].insn_off must be the | |
15186 | * first sub also and the first sub must have | |
15187 | * subprog_info[0].start == 0. | |
15188 | */ | |
15189 | if ((i && linfo[i].insn_off <= prev_offset) || | |
15190 | linfo[i].insn_off >= prog->len) { | |
15191 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
15192 | i, linfo[i].insn_off, prev_offset, | |
15193 | prog->len); | |
15194 | err = -EINVAL; | |
15195 | goto err_free; | |
15196 | } | |
15197 | ||
fdbaa0be MKL |
15198 | if (!prog->insnsi[linfo[i].insn_off].code) { |
15199 | verbose(env, | |
15200 | "Invalid insn code at line_info[%u].insn_off\n", | |
15201 | i); | |
15202 | err = -EINVAL; | |
15203 | goto err_free; | |
15204 | } | |
15205 | ||
23127b33 MKL |
15206 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
15207 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
15208 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
15209 | err = -EINVAL; | |
15210 | goto err_free; | |
15211 | } | |
15212 | ||
15213 | if (s != env->subprog_cnt) { | |
15214 | if (linfo[i].insn_off == sub[s].start) { | |
15215 | sub[s].linfo_idx = i; | |
15216 | s++; | |
15217 | } else if (sub[s].start < linfo[i].insn_off) { | |
15218 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
15219 | err = -EINVAL; | |
15220 | goto err_free; | |
15221 | } | |
15222 | } | |
15223 | ||
15224 | prev_offset = linfo[i].insn_off; | |
af2ac3e1 | 15225 | bpfptr_add(&ulinfo, rec_size); |
c454a46b MKL |
15226 | } |
15227 | ||
15228 | if (s != env->subprog_cnt) { | |
15229 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
15230 | env->subprog_cnt - s, s); | |
15231 | err = -EINVAL; | |
15232 | goto err_free; | |
15233 | } | |
15234 | ||
15235 | prog->aux->linfo = linfo; | |
15236 | prog->aux->nr_linfo = nr_linfo; | |
15237 | ||
15238 | return 0; | |
15239 | ||
15240 | err_free: | |
15241 | kvfree(linfo); | |
15242 | return err; | |
15243 | } | |
15244 | ||
fbd94c7a AS |
15245 | #define MIN_CORE_RELO_SIZE sizeof(struct bpf_core_relo) |
15246 | #define MAX_CORE_RELO_SIZE MAX_FUNCINFO_REC_SIZE | |
15247 | ||
15248 | static int check_core_relo(struct bpf_verifier_env *env, | |
15249 | const union bpf_attr *attr, | |
15250 | bpfptr_t uattr) | |
15251 | { | |
15252 | u32 i, nr_core_relo, ncopy, expected_size, rec_size; | |
15253 | struct bpf_core_relo core_relo = {}; | |
15254 | struct bpf_prog *prog = env->prog; | |
15255 | const struct btf *btf = prog->aux->btf; | |
15256 | struct bpf_core_ctx ctx = { | |
15257 | .log = &env->log, | |
15258 | .btf = btf, | |
15259 | }; | |
15260 | bpfptr_t u_core_relo; | |
15261 | int err; | |
15262 | ||
15263 | nr_core_relo = attr->core_relo_cnt; | |
15264 | if (!nr_core_relo) | |
15265 | return 0; | |
15266 | if (nr_core_relo > INT_MAX / sizeof(struct bpf_core_relo)) | |
15267 | return -EINVAL; | |
15268 | ||
15269 | rec_size = attr->core_relo_rec_size; | |
15270 | if (rec_size < MIN_CORE_RELO_SIZE || | |
15271 | rec_size > MAX_CORE_RELO_SIZE || | |
15272 | rec_size % sizeof(u32)) | |
15273 | return -EINVAL; | |
15274 | ||
15275 | u_core_relo = make_bpfptr(attr->core_relos, uattr.is_kernel); | |
15276 | expected_size = sizeof(struct bpf_core_relo); | |
15277 | ncopy = min_t(u32, expected_size, rec_size); | |
15278 | ||
15279 | /* Unlike func_info and line_info, copy and apply each CO-RE | |
15280 | * relocation record one at a time. | |
15281 | */ | |
15282 | for (i = 0; i < nr_core_relo; i++) { | |
15283 | /* future proofing when sizeof(bpf_core_relo) changes */ | |
15284 | err = bpf_check_uarg_tail_zero(u_core_relo, expected_size, rec_size); | |
15285 | if (err) { | |
15286 | if (err == -E2BIG) { | |
15287 | verbose(env, "nonzero tailing record in core_relo"); | |
15288 | if (copy_to_bpfptr_offset(uattr, | |
15289 | offsetof(union bpf_attr, core_relo_rec_size), | |
15290 | &expected_size, sizeof(expected_size))) | |
15291 | err = -EFAULT; | |
15292 | } | |
15293 | break; | |
15294 | } | |
15295 | ||
15296 | if (copy_from_bpfptr(&core_relo, u_core_relo, ncopy)) { | |
15297 | err = -EFAULT; | |
15298 | break; | |
15299 | } | |
15300 | ||
15301 | if (core_relo.insn_off % 8 || core_relo.insn_off / 8 >= prog->len) { | |
15302 | verbose(env, "Invalid core_relo[%u].insn_off:%u prog->len:%u\n", | |
15303 | i, core_relo.insn_off, prog->len); | |
15304 | err = -EINVAL; | |
15305 | break; | |
15306 | } | |
15307 | ||
15308 | err = bpf_core_apply(&ctx, &core_relo, i, | |
15309 | &prog->insnsi[core_relo.insn_off / 8]); | |
15310 | if (err) | |
15311 | break; | |
15312 | bpfptr_add(&u_core_relo, rec_size); | |
15313 | } | |
15314 | return err; | |
15315 | } | |
15316 | ||
c454a46b MKL |
15317 | static int check_btf_info(struct bpf_verifier_env *env, |
15318 | const union bpf_attr *attr, | |
af2ac3e1 | 15319 | bpfptr_t uattr) |
c454a46b MKL |
15320 | { |
15321 | struct btf *btf; | |
15322 | int err; | |
15323 | ||
09b28d76 AS |
15324 | if (!attr->func_info_cnt && !attr->line_info_cnt) { |
15325 | if (check_abnormal_return(env)) | |
15326 | return -EINVAL; | |
c454a46b | 15327 | return 0; |
09b28d76 | 15328 | } |
c454a46b MKL |
15329 | |
15330 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
15331 | if (IS_ERR(btf)) | |
15332 | return PTR_ERR(btf); | |
350a5c4d AS |
15333 | if (btf_is_kernel(btf)) { |
15334 | btf_put(btf); | |
15335 | return -EACCES; | |
15336 | } | |
c454a46b MKL |
15337 | env->prog->aux->btf = btf; |
15338 | ||
15339 | err = check_btf_func(env, attr, uattr); | |
15340 | if (err) | |
15341 | return err; | |
15342 | ||
15343 | err = check_btf_line(env, attr, uattr); | |
15344 | if (err) | |
15345 | return err; | |
15346 | ||
fbd94c7a AS |
15347 | err = check_core_relo(env, attr, uattr); |
15348 | if (err) | |
15349 | return err; | |
15350 | ||
c454a46b | 15351 | return 0; |
ba64e7d8 YS |
15352 | } |
15353 | ||
f1174f77 EC |
15354 | /* check %cur's range satisfies %old's */ |
15355 | static bool range_within(struct bpf_reg_state *old, | |
15356 | struct bpf_reg_state *cur) | |
15357 | { | |
b03c9f9f EC |
15358 | return old->umin_value <= cur->umin_value && |
15359 | old->umax_value >= cur->umax_value && | |
15360 | old->smin_value <= cur->smin_value && | |
fd675184 DB |
15361 | old->smax_value >= cur->smax_value && |
15362 | old->u32_min_value <= cur->u32_min_value && | |
15363 | old->u32_max_value >= cur->u32_max_value && | |
15364 | old->s32_min_value <= cur->s32_min_value && | |
15365 | old->s32_max_value >= cur->s32_max_value; | |
f1174f77 EC |
15366 | } |
15367 | ||
f1174f77 EC |
15368 | /* If in the old state two registers had the same id, then they need to have |
15369 | * the same id in the new state as well. But that id could be different from | |
15370 | * the old state, so we need to track the mapping from old to new ids. | |
15371 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
15372 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
15373 | * regs with a different old id could still have new id 9, we don't care about | |
15374 | * that. | |
15375 | * So we look through our idmap to see if this old id has been seen before. If | |
15376 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 15377 | */ |
1ffc85d9 | 15378 | static bool check_ids(u32 old_id, u32 cur_id, struct bpf_idmap *idmap) |
969bf05e | 15379 | { |
1ffc85d9 | 15380 | struct bpf_id_pair *map = idmap->map; |
f1174f77 | 15381 | unsigned int i; |
969bf05e | 15382 | |
4633a006 AN |
15383 | /* either both IDs should be set or both should be zero */ |
15384 | if (!!old_id != !!cur_id) | |
15385 | return false; | |
15386 | ||
15387 | if (old_id == 0) /* cur_id == 0 as well */ | |
15388 | return true; | |
15389 | ||
c9e73e3d | 15390 | for (i = 0; i < BPF_ID_MAP_SIZE; i++) { |
1ffc85d9 | 15391 | if (!map[i].old) { |
f1174f77 | 15392 | /* Reached an empty slot; haven't seen this id before */ |
1ffc85d9 EZ |
15393 | map[i].old = old_id; |
15394 | map[i].cur = cur_id; | |
f1174f77 EC |
15395 | return true; |
15396 | } | |
1ffc85d9 EZ |
15397 | if (map[i].old == old_id) |
15398 | return map[i].cur == cur_id; | |
15399 | if (map[i].cur == cur_id) | |
15400 | return false; | |
f1174f77 EC |
15401 | } |
15402 | /* We ran out of idmap slots, which should be impossible */ | |
15403 | WARN_ON_ONCE(1); | |
15404 | return false; | |
15405 | } | |
15406 | ||
1ffc85d9 EZ |
15407 | /* Similar to check_ids(), but allocate a unique temporary ID |
15408 | * for 'old_id' or 'cur_id' of zero. | |
15409 | * This makes pairs like '0 vs unique ID', 'unique ID vs 0' valid. | |
15410 | */ | |
15411 | static bool check_scalar_ids(u32 old_id, u32 cur_id, struct bpf_idmap *idmap) | |
15412 | { | |
15413 | old_id = old_id ? old_id : ++idmap->tmp_id_gen; | |
15414 | cur_id = cur_id ? cur_id : ++idmap->tmp_id_gen; | |
15415 | ||
15416 | return check_ids(old_id, cur_id, idmap); | |
15417 | } | |
15418 | ||
9242b5f5 AS |
15419 | static void clean_func_state(struct bpf_verifier_env *env, |
15420 | struct bpf_func_state *st) | |
15421 | { | |
15422 | enum bpf_reg_liveness live; | |
15423 | int i, j; | |
15424 | ||
15425 | for (i = 0; i < BPF_REG_FP; i++) { | |
15426 | live = st->regs[i].live; | |
15427 | /* liveness must not touch this register anymore */ | |
15428 | st->regs[i].live |= REG_LIVE_DONE; | |
15429 | if (!(live & REG_LIVE_READ)) | |
15430 | /* since the register is unused, clear its state | |
15431 | * to make further comparison simpler | |
15432 | */ | |
f54c7898 | 15433 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
15434 | } |
15435 | ||
15436 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
15437 | live = st->stack[i].spilled_ptr.live; | |
15438 | /* liveness must not touch this stack slot anymore */ | |
15439 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
15440 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 15441 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
15442 | for (j = 0; j < BPF_REG_SIZE; j++) |
15443 | st->stack[i].slot_type[j] = STACK_INVALID; | |
15444 | } | |
15445 | } | |
15446 | } | |
15447 | ||
15448 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
15449 | struct bpf_verifier_state *st) | |
15450 | { | |
15451 | int i; | |
15452 | ||
15453 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
15454 | /* all regs in this state in all frames were already marked */ | |
15455 | return; | |
15456 | ||
15457 | for (i = 0; i <= st->curframe; i++) | |
15458 | clean_func_state(env, st->frame[i]); | |
15459 | } | |
15460 | ||
15461 | /* the parentage chains form a tree. | |
15462 | * the verifier states are added to state lists at given insn and | |
15463 | * pushed into state stack for future exploration. | |
15464 | * when the verifier reaches bpf_exit insn some of the verifer states | |
15465 | * stored in the state lists have their final liveness state already, | |
15466 | * but a lot of states will get revised from liveness point of view when | |
15467 | * the verifier explores other branches. | |
15468 | * Example: | |
15469 | * 1: r0 = 1 | |
15470 | * 2: if r1 == 100 goto pc+1 | |
15471 | * 3: r0 = 2 | |
15472 | * 4: exit | |
15473 | * when the verifier reaches exit insn the register r0 in the state list of | |
15474 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
15475 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
15476 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
15477 | * | |
15478 | * Since the verifier pushes the branch states as it sees them while exploring | |
15479 | * the program the condition of walking the branch instruction for the second | |
15480 | * time means that all states below this branch were already explored and | |
8fb33b60 | 15481 | * their final liveness marks are already propagated. |
9242b5f5 AS |
15482 | * Hence when the verifier completes the search of state list in is_state_visited() |
15483 | * we can call this clean_live_states() function to mark all liveness states | |
15484 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
15485 | * will not be used. | |
15486 | * This function also clears the registers and stack for states that !READ | |
15487 | * to simplify state merging. | |
15488 | * | |
15489 | * Important note here that walking the same branch instruction in the callee | |
15490 | * doesn't meant that the states are DONE. The verifier has to compare | |
15491 | * the callsites | |
15492 | */ | |
15493 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
15494 | struct bpf_verifier_state *cur) | |
15495 | { | |
15496 | struct bpf_verifier_state_list *sl; | |
15497 | int i; | |
15498 | ||
5d839021 | 15499 | sl = *explored_state(env, insn); |
a8f500af | 15500 | while (sl) { |
2589726d AS |
15501 | if (sl->state.branches) |
15502 | goto next; | |
dc2a4ebc AS |
15503 | if (sl->state.insn_idx != insn || |
15504 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
15505 | goto next; |
15506 | for (i = 0; i <= cur->curframe; i++) | |
15507 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
15508 | goto next; | |
15509 | clean_verifier_state(env, &sl->state); | |
15510 | next: | |
15511 | sl = sl->next; | |
15512 | } | |
15513 | } | |
15514 | ||
4a95c85c | 15515 | static bool regs_exact(const struct bpf_reg_state *rold, |
4633a006 | 15516 | const struct bpf_reg_state *rcur, |
1ffc85d9 | 15517 | struct bpf_idmap *idmap) |
4a95c85c | 15518 | { |
d2dcc67d | 15519 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && |
4633a006 AN |
15520 | check_ids(rold->id, rcur->id, idmap) && |
15521 | check_ids(rold->ref_obj_id, rcur->ref_obj_id, idmap); | |
4a95c85c AN |
15522 | } |
15523 | ||
f1174f77 | 15524 | /* Returns true if (rold safe implies rcur safe) */ |
e042aa53 | 15525 | static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold, |
1ffc85d9 | 15526 | struct bpf_reg_state *rcur, struct bpf_idmap *idmap) |
f1174f77 | 15527 | { |
dc503a8a EC |
15528 | if (!(rold->live & REG_LIVE_READ)) |
15529 | /* explored state didn't use this */ | |
15530 | return true; | |
f1174f77 EC |
15531 | if (rold->type == NOT_INIT) |
15532 | /* explored state can't have used this */ | |
969bf05e | 15533 | return true; |
f1174f77 EC |
15534 | if (rcur->type == NOT_INIT) |
15535 | return false; | |
7f4ce97c | 15536 | |
910f6999 AN |
15537 | /* Enforce that register types have to match exactly, including their |
15538 | * modifiers (like PTR_MAYBE_NULL, MEM_RDONLY, etc), as a general | |
15539 | * rule. | |
15540 | * | |
15541 | * One can make a point that using a pointer register as unbounded | |
15542 | * SCALAR would be technically acceptable, but this could lead to | |
15543 | * pointer leaks because scalars are allowed to leak while pointers | |
15544 | * are not. We could make this safe in special cases if root is | |
15545 | * calling us, but it's probably not worth the hassle. | |
15546 | * | |
15547 | * Also, register types that are *not* MAYBE_NULL could technically be | |
15548 | * safe to use as their MAYBE_NULL variants (e.g., PTR_TO_MAP_VALUE | |
15549 | * is safe to be used as PTR_TO_MAP_VALUE_OR_NULL, provided both point | |
15550 | * to the same map). | |
7f4ce97c AN |
15551 | * However, if the old MAYBE_NULL register then got NULL checked, |
15552 | * doing so could have affected others with the same id, and we can't | |
15553 | * check for that because we lost the id when we converted to | |
15554 | * a non-MAYBE_NULL variant. | |
15555 | * So, as a general rule we don't allow mixing MAYBE_NULL and | |
910f6999 | 15556 | * non-MAYBE_NULL registers as well. |
7f4ce97c | 15557 | */ |
910f6999 | 15558 | if (rold->type != rcur->type) |
7f4ce97c AN |
15559 | return false; |
15560 | ||
c25b2ae1 | 15561 | switch (base_type(rold->type)) { |
f1174f77 | 15562 | case SCALAR_VALUE: |
1ffc85d9 EZ |
15563 | if (env->explore_alu_limits) { |
15564 | /* explore_alu_limits disables tnum_in() and range_within() | |
15565 | * logic and requires everything to be strict | |
15566 | */ | |
15567 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
15568 | check_scalar_ids(rold->id, rcur->id, idmap); | |
15569 | } | |
910f6999 AN |
15570 | if (!rold->precise) |
15571 | return true; | |
1ffc85d9 EZ |
15572 | /* Why check_ids() for scalar registers? |
15573 | * | |
15574 | * Consider the following BPF code: | |
15575 | * 1: r6 = ... unbound scalar, ID=a ... | |
15576 | * 2: r7 = ... unbound scalar, ID=b ... | |
15577 | * 3: if (r6 > r7) goto +1 | |
15578 | * 4: r6 = r7 | |
15579 | * 5: if (r6 > X) goto ... | |
15580 | * 6: ... memory operation using r7 ... | |
15581 | * | |
15582 | * First verification path is [1-6]: | |
15583 | * - at (4) same bpf_reg_state::id (b) would be assigned to r6 and r7; | |
15584 | * - at (5) r6 would be marked <= X, find_equal_scalars() would also mark | |
15585 | * r7 <= X, because r6 and r7 share same id. | |
15586 | * Next verification path is [1-4, 6]. | |
15587 | * | |
15588 | * Instruction (6) would be reached in two states: | |
15589 | * I. r6{.id=b}, r7{.id=b} via path 1-6; | |
15590 | * II. r6{.id=a}, r7{.id=b} via path 1-4, 6. | |
15591 | * | |
15592 | * Use check_ids() to distinguish these states. | |
15593 | * --- | |
15594 | * Also verify that new value satisfies old value range knowledge. | |
15595 | */ | |
910f6999 | 15596 | return range_within(rold, rcur) && |
1ffc85d9 EZ |
15597 | tnum_in(rold->var_off, rcur->var_off) && |
15598 | check_scalar_ids(rold->id, rcur->id, idmap); | |
69c087ba | 15599 | case PTR_TO_MAP_KEY: |
f1174f77 | 15600 | case PTR_TO_MAP_VALUE: |
567da5d2 AN |
15601 | case PTR_TO_MEM: |
15602 | case PTR_TO_BUF: | |
15603 | case PTR_TO_TP_BUFFER: | |
1b688a19 EC |
15604 | /* If the new min/max/var_off satisfy the old ones and |
15605 | * everything else matches, we are OK. | |
1b688a19 | 15606 | */ |
a73bf9f2 | 15607 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, var_off)) == 0 && |
1b688a19 | 15608 | range_within(rold, rcur) && |
4ea2bb15 | 15609 | tnum_in(rold->var_off, rcur->var_off) && |
567da5d2 AN |
15610 | check_ids(rold->id, rcur->id, idmap) && |
15611 | check_ids(rold->ref_obj_id, rcur->ref_obj_id, idmap); | |
de8f3a83 | 15612 | case PTR_TO_PACKET_META: |
f1174f77 | 15613 | case PTR_TO_PACKET: |
f1174f77 EC |
15614 | /* We must have at least as much range as the old ptr |
15615 | * did, so that any accesses which were safe before are | |
15616 | * still safe. This is true even if old range < old off, | |
15617 | * since someone could have accessed through (ptr - k), or | |
15618 | * even done ptr -= k in a register, to get a safe access. | |
15619 | */ | |
15620 | if (rold->range > rcur->range) | |
15621 | return false; | |
15622 | /* If the offsets don't match, we can't trust our alignment; | |
15623 | * nor can we be sure that we won't fall out of range. | |
15624 | */ | |
15625 | if (rold->off != rcur->off) | |
15626 | return false; | |
15627 | /* id relations must be preserved */ | |
4633a006 | 15628 | if (!check_ids(rold->id, rcur->id, idmap)) |
f1174f77 EC |
15629 | return false; |
15630 | /* new val must satisfy old val knowledge */ | |
15631 | return range_within(rold, rcur) && | |
15632 | tnum_in(rold->var_off, rcur->var_off); | |
7c884339 EZ |
15633 | case PTR_TO_STACK: |
15634 | /* two stack pointers are equal only if they're pointing to | |
15635 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
f1174f77 | 15636 | */ |
4633a006 | 15637 | return regs_exact(rold, rcur, idmap) && rold->frameno == rcur->frameno; |
f1174f77 | 15638 | default: |
4633a006 | 15639 | return regs_exact(rold, rcur, idmap); |
f1174f77 | 15640 | } |
969bf05e AS |
15641 | } |
15642 | ||
e042aa53 | 15643 | static bool stacksafe(struct bpf_verifier_env *env, struct bpf_func_state *old, |
1ffc85d9 | 15644 | struct bpf_func_state *cur, struct bpf_idmap *idmap) |
638f5b90 AS |
15645 | { |
15646 | int i, spi; | |
15647 | ||
638f5b90 AS |
15648 | /* walk slots of the explored stack and ignore any additional |
15649 | * slots in the current stack, since explored(safe) state | |
15650 | * didn't use them | |
15651 | */ | |
15652 | for (i = 0; i < old->allocated_stack; i++) { | |
06accc87 AN |
15653 | struct bpf_reg_state *old_reg, *cur_reg; |
15654 | ||
638f5b90 AS |
15655 | spi = i / BPF_REG_SIZE; |
15656 | ||
b233920c AS |
15657 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
15658 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 15659 | /* explored state didn't use this */ |
fd05e57b | 15660 | continue; |
b233920c | 15661 | } |
cc2b14d5 | 15662 | |
638f5b90 AS |
15663 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
15664 | continue; | |
19e2dbb7 | 15665 | |
6715df8d EZ |
15666 | if (env->allow_uninit_stack && |
15667 | old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC) | |
15668 | continue; | |
15669 | ||
19e2dbb7 AS |
15670 | /* explored stack has more populated slots than current stack |
15671 | * and these slots were used | |
15672 | */ | |
15673 | if (i >= cur->allocated_stack) | |
15674 | return false; | |
15675 | ||
cc2b14d5 AS |
15676 | /* if old state was safe with misc data in the stack |
15677 | * it will be safe with zero-initialized stack. | |
15678 | * The opposite is not true | |
15679 | */ | |
15680 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
15681 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
15682 | continue; | |
638f5b90 AS |
15683 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
15684 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
15685 | /* Ex: old explored (safe) state has STACK_SPILL in | |
b8c1a309 | 15686 | * this stack slot, but current has STACK_MISC -> |
638f5b90 AS |
15687 | * this verifier states are not equivalent, |
15688 | * return false to continue verification of this path | |
15689 | */ | |
15690 | return false; | |
27113c59 | 15691 | if (i % BPF_REG_SIZE != BPF_REG_SIZE - 1) |
638f5b90 | 15692 | continue; |
d6fefa11 KKD |
15693 | /* Both old and cur are having same slot_type */ |
15694 | switch (old->stack[spi].slot_type[BPF_REG_SIZE - 1]) { | |
15695 | case STACK_SPILL: | |
638f5b90 AS |
15696 | /* when explored and current stack slot are both storing |
15697 | * spilled registers, check that stored pointers types | |
15698 | * are the same as well. | |
15699 | * Ex: explored safe path could have stored | |
15700 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
15701 | * but current path has stored: | |
15702 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
15703 | * such verifier states are not equivalent. | |
15704 | * return false to continue verification of this path | |
15705 | */ | |
d6fefa11 KKD |
15706 | if (!regsafe(env, &old->stack[spi].spilled_ptr, |
15707 | &cur->stack[spi].spilled_ptr, idmap)) | |
15708 | return false; | |
15709 | break; | |
15710 | case STACK_DYNPTR: | |
d6fefa11 KKD |
15711 | old_reg = &old->stack[spi].spilled_ptr; |
15712 | cur_reg = &cur->stack[spi].spilled_ptr; | |
15713 | if (old_reg->dynptr.type != cur_reg->dynptr.type || | |
15714 | old_reg->dynptr.first_slot != cur_reg->dynptr.first_slot || | |
15715 | !check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap)) | |
15716 | return false; | |
15717 | break; | |
06accc87 AN |
15718 | case STACK_ITER: |
15719 | old_reg = &old->stack[spi].spilled_ptr; | |
15720 | cur_reg = &cur->stack[spi].spilled_ptr; | |
15721 | /* iter.depth is not compared between states as it | |
15722 | * doesn't matter for correctness and would otherwise | |
15723 | * prevent convergence; we maintain it only to prevent | |
15724 | * infinite loop check triggering, see | |
15725 | * iter_active_depths_differ() | |
15726 | */ | |
15727 | if (old_reg->iter.btf != cur_reg->iter.btf || | |
15728 | old_reg->iter.btf_id != cur_reg->iter.btf_id || | |
15729 | old_reg->iter.state != cur_reg->iter.state || | |
15730 | /* ignore {old_reg,cur_reg}->iter.depth, see above */ | |
15731 | !check_ids(old_reg->ref_obj_id, cur_reg->ref_obj_id, idmap)) | |
15732 | return false; | |
15733 | break; | |
d6fefa11 KKD |
15734 | case STACK_MISC: |
15735 | case STACK_ZERO: | |
15736 | case STACK_INVALID: | |
15737 | continue; | |
15738 | /* Ensure that new unhandled slot types return false by default */ | |
15739 | default: | |
638f5b90 | 15740 | return false; |
d6fefa11 | 15741 | } |
638f5b90 AS |
15742 | } |
15743 | return true; | |
15744 | } | |
15745 | ||
e8f55fcf | 15746 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur, |
1ffc85d9 | 15747 | struct bpf_idmap *idmap) |
fd978bf7 | 15748 | { |
e8f55fcf AN |
15749 | int i; |
15750 | ||
fd978bf7 JS |
15751 | if (old->acquired_refs != cur->acquired_refs) |
15752 | return false; | |
e8f55fcf AN |
15753 | |
15754 | for (i = 0; i < old->acquired_refs; i++) { | |
15755 | if (!check_ids(old->refs[i].id, cur->refs[i].id, idmap)) | |
15756 | return false; | |
15757 | } | |
15758 | ||
15759 | return true; | |
fd978bf7 JS |
15760 | } |
15761 | ||
f1bca824 AS |
15762 | /* compare two verifier states |
15763 | * | |
15764 | * all states stored in state_list are known to be valid, since | |
15765 | * verifier reached 'bpf_exit' instruction through them | |
15766 | * | |
15767 | * this function is called when verifier exploring different branches of | |
15768 | * execution popped from the state stack. If it sees an old state that has | |
15769 | * more strict register state and more strict stack state then this execution | |
15770 | * branch doesn't need to be explored further, since verifier already | |
15771 | * concluded that more strict state leads to valid finish. | |
15772 | * | |
15773 | * Therefore two states are equivalent if register state is more conservative | |
15774 | * and explored stack state is more conservative than the current one. | |
15775 | * Example: | |
15776 | * explored current | |
15777 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
15778 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
15779 | * | |
15780 | * In other words if current stack state (one being explored) has more | |
15781 | * valid slots than old one that already passed validation, it means | |
15782 | * the verifier can stop exploring and conclude that current state is valid too | |
15783 | * | |
15784 | * Similarly with registers. If explored state has register type as invalid | |
15785 | * whereas register type in current state is meaningful, it means that | |
15786 | * the current state will reach 'bpf_exit' instruction safely | |
15787 | */ | |
c9e73e3d | 15788 | static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_state *old, |
f4d7e40a | 15789 | struct bpf_func_state *cur) |
f1bca824 AS |
15790 | { |
15791 | int i; | |
15792 | ||
c9e73e3d | 15793 | for (i = 0; i < MAX_BPF_REG; i++) |
e042aa53 | 15794 | if (!regsafe(env, &old->regs[i], &cur->regs[i], |
1ffc85d9 | 15795 | &env->idmap_scratch)) |
c9e73e3d | 15796 | return false; |
f1bca824 | 15797 | |
1ffc85d9 | 15798 | if (!stacksafe(env, old, cur, &env->idmap_scratch)) |
c9e73e3d | 15799 | return false; |
fd978bf7 | 15800 | |
1ffc85d9 | 15801 | if (!refsafe(old, cur, &env->idmap_scratch)) |
c9e73e3d LB |
15802 | return false; |
15803 | ||
15804 | return true; | |
f1bca824 AS |
15805 | } |
15806 | ||
f4d7e40a AS |
15807 | static bool states_equal(struct bpf_verifier_env *env, |
15808 | struct bpf_verifier_state *old, | |
15809 | struct bpf_verifier_state *cur) | |
15810 | { | |
15811 | int i; | |
15812 | ||
15813 | if (old->curframe != cur->curframe) | |
15814 | return false; | |
15815 | ||
1ffc85d9 EZ |
15816 | env->idmap_scratch.tmp_id_gen = env->id_gen; |
15817 | memset(&env->idmap_scratch.map, 0, sizeof(env->idmap_scratch.map)); | |
5dd9cdbc | 15818 | |
979d63d5 DB |
15819 | /* Verification state from speculative execution simulation |
15820 | * must never prune a non-speculative execution one. | |
15821 | */ | |
15822 | if (old->speculative && !cur->speculative) | |
15823 | return false; | |
15824 | ||
4ea2bb15 EZ |
15825 | if (old->active_lock.ptr != cur->active_lock.ptr) |
15826 | return false; | |
15827 | ||
15828 | /* Old and cur active_lock's have to be either both present | |
15829 | * or both absent. | |
15830 | */ | |
15831 | if (!!old->active_lock.id != !!cur->active_lock.id) | |
15832 | return false; | |
15833 | ||
15834 | if (old->active_lock.id && | |
1ffc85d9 | 15835 | !check_ids(old->active_lock.id, cur->active_lock.id, &env->idmap_scratch)) |
d83525ca AS |
15836 | return false; |
15837 | ||
9bb00b28 | 15838 | if (old->active_rcu_lock != cur->active_rcu_lock) |
d83525ca AS |
15839 | return false; |
15840 | ||
f4d7e40a AS |
15841 | /* for states to be equal callsites have to be the same |
15842 | * and all frame states need to be equivalent | |
15843 | */ | |
15844 | for (i = 0; i <= old->curframe; i++) { | |
15845 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
15846 | return false; | |
c9e73e3d | 15847 | if (!func_states_equal(env, old->frame[i], cur->frame[i])) |
f4d7e40a AS |
15848 | return false; |
15849 | } | |
15850 | return true; | |
15851 | } | |
15852 | ||
5327ed3d JW |
15853 | /* Return 0 if no propagation happened. Return negative error code if error |
15854 | * happened. Otherwise, return the propagated bit. | |
15855 | */ | |
55e7f3b5 JW |
15856 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
15857 | struct bpf_reg_state *reg, | |
15858 | struct bpf_reg_state *parent_reg) | |
15859 | { | |
5327ed3d JW |
15860 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
15861 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
15862 | int err; |
15863 | ||
5327ed3d JW |
15864 | /* When comes here, read flags of PARENT_REG or REG could be any of |
15865 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
15866 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
15867 | */ | |
15868 | if (parent_flag == REG_LIVE_READ64 || | |
15869 | /* Or if there is no read flag from REG. */ | |
15870 | !flag || | |
15871 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
15872 | parent_flag == flag) | |
55e7f3b5 JW |
15873 | return 0; |
15874 | ||
5327ed3d | 15875 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
15876 | if (err) |
15877 | return err; | |
15878 | ||
5327ed3d | 15879 | return flag; |
55e7f3b5 JW |
15880 | } |
15881 | ||
8e9cd9ce | 15882 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
15883 | * straight-line code between a state and its parent. When we arrive at an |
15884 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
15885 | * code, so read marks in the state must propagate to the parent regardless | |
15886 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 15887 | * in mark_reg_read() is for. |
8e9cd9ce | 15888 | */ |
f4d7e40a AS |
15889 | static int propagate_liveness(struct bpf_verifier_env *env, |
15890 | const struct bpf_verifier_state *vstate, | |
15891 | struct bpf_verifier_state *vparent) | |
dc503a8a | 15892 | { |
3f8cafa4 | 15893 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 15894 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 15895 | int i, frame, err = 0; |
dc503a8a | 15896 | |
f4d7e40a AS |
15897 | if (vparent->curframe != vstate->curframe) { |
15898 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
15899 | vparent->curframe, vstate->curframe); | |
15900 | return -EFAULT; | |
15901 | } | |
dc503a8a EC |
15902 | /* Propagate read liveness of registers... */ |
15903 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 15904 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
15905 | parent = vparent->frame[frame]; |
15906 | state = vstate->frame[frame]; | |
15907 | parent_reg = parent->regs; | |
15908 | state_reg = state->regs; | |
83d16312 JK |
15909 | /* We don't need to worry about FP liveness, it's read-only */ |
15910 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
15911 | err = propagate_liveness_reg(env, &state_reg[i], |
15912 | &parent_reg[i]); | |
5327ed3d | 15913 | if (err < 0) |
3f8cafa4 | 15914 | return err; |
5327ed3d JW |
15915 | if (err == REG_LIVE_READ64) |
15916 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 15917 | } |
f4d7e40a | 15918 | |
1b04aee7 | 15919 | /* Propagate stack slots. */ |
f4d7e40a AS |
15920 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
15921 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
15922 | parent_reg = &parent->stack[i].spilled_ptr; |
15923 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
15924 | err = propagate_liveness_reg(env, state_reg, |
15925 | parent_reg); | |
5327ed3d | 15926 | if (err < 0) |
3f8cafa4 | 15927 | return err; |
dc503a8a EC |
15928 | } |
15929 | } | |
5327ed3d | 15930 | return 0; |
dc503a8a EC |
15931 | } |
15932 | ||
a3ce685d AS |
15933 | /* find precise scalars in the previous equivalent state and |
15934 | * propagate them into the current state | |
15935 | */ | |
15936 | static int propagate_precision(struct bpf_verifier_env *env, | |
15937 | const struct bpf_verifier_state *old) | |
15938 | { | |
15939 | struct bpf_reg_state *state_reg; | |
15940 | struct bpf_func_state *state; | |
529409ea | 15941 | int i, err = 0, fr; |
f655badf | 15942 | bool first; |
a3ce685d | 15943 | |
529409ea AN |
15944 | for (fr = old->curframe; fr >= 0; fr--) { |
15945 | state = old->frame[fr]; | |
15946 | state_reg = state->regs; | |
f655badf | 15947 | first = true; |
529409ea AN |
15948 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { |
15949 | if (state_reg->type != SCALAR_VALUE || | |
52c2b005 AN |
15950 | !state_reg->precise || |
15951 | !(state_reg->live & REG_LIVE_READ)) | |
529409ea | 15952 | continue; |
f655badf AN |
15953 | if (env->log.level & BPF_LOG_LEVEL2) { |
15954 | if (first) | |
15955 | verbose(env, "frame %d: propagating r%d", fr, i); | |
15956 | else | |
15957 | verbose(env, ",r%d", i); | |
15958 | } | |
15959 | bt_set_frame_reg(&env->bt, fr, i); | |
15960 | first = false; | |
529409ea | 15961 | } |
a3ce685d | 15962 | |
529409ea AN |
15963 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
15964 | if (!is_spilled_reg(&state->stack[i])) | |
15965 | continue; | |
15966 | state_reg = &state->stack[i].spilled_ptr; | |
15967 | if (state_reg->type != SCALAR_VALUE || | |
52c2b005 AN |
15968 | !state_reg->precise || |
15969 | !(state_reg->live & REG_LIVE_READ)) | |
529409ea | 15970 | continue; |
f655badf AN |
15971 | if (env->log.level & BPF_LOG_LEVEL2) { |
15972 | if (first) | |
15973 | verbose(env, "frame %d: propagating fp%d", | |
15974 | fr, (-i - 1) * BPF_REG_SIZE); | |
15975 | else | |
15976 | verbose(env, ",fp%d", (-i - 1) * BPF_REG_SIZE); | |
15977 | } | |
15978 | bt_set_frame_slot(&env->bt, fr, i); | |
15979 | first = false; | |
529409ea | 15980 | } |
f655badf AN |
15981 | if (!first) |
15982 | verbose(env, "\n"); | |
a3ce685d | 15983 | } |
f655badf AN |
15984 | |
15985 | err = mark_chain_precision_batch(env); | |
15986 | if (err < 0) | |
15987 | return err; | |
15988 | ||
a3ce685d AS |
15989 | return 0; |
15990 | } | |
15991 | ||
2589726d AS |
15992 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
15993 | struct bpf_verifier_state *cur) | |
15994 | { | |
15995 | struct bpf_func_state *fold, *fcur; | |
15996 | int i, fr = cur->curframe; | |
15997 | ||
15998 | if (old->curframe != fr) | |
15999 | return false; | |
16000 | ||
16001 | fold = old->frame[fr]; | |
16002 | fcur = cur->frame[fr]; | |
16003 | for (i = 0; i < MAX_BPF_REG; i++) | |
16004 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
16005 | offsetof(struct bpf_reg_state, parent))) | |
16006 | return false; | |
16007 | return true; | |
16008 | } | |
16009 | ||
06accc87 AN |
16010 | static bool is_iter_next_insn(struct bpf_verifier_env *env, int insn_idx) |
16011 | { | |
16012 | return env->insn_aux_data[insn_idx].is_iter_next; | |
16013 | } | |
16014 | ||
16015 | /* is_state_visited() handles iter_next() (see process_iter_next_call() for | |
16016 | * terminology) calls specially: as opposed to bounded BPF loops, it *expects* | |
16017 | * states to match, which otherwise would look like an infinite loop. So while | |
16018 | * iter_next() calls are taken care of, we still need to be careful and | |
16019 | * prevent erroneous and too eager declaration of "ininite loop", when | |
16020 | * iterators are involved. | |
16021 | * | |
16022 | * Here's a situation in pseudo-BPF assembly form: | |
16023 | * | |
16024 | * 0: again: ; set up iter_next() call args | |
16025 | * 1: r1 = &it ; <CHECKPOINT HERE> | |
16026 | * 2: call bpf_iter_num_next ; this is iter_next() call | |
16027 | * 3: if r0 == 0 goto done | |
16028 | * 4: ... something useful here ... | |
16029 | * 5: goto again ; another iteration | |
16030 | * 6: done: | |
16031 | * 7: r1 = &it | |
16032 | * 8: call bpf_iter_num_destroy ; clean up iter state | |
16033 | * 9: exit | |
16034 | * | |
16035 | * This is a typical loop. Let's assume that we have a prune point at 1:, | |
16036 | * before we get to `call bpf_iter_num_next` (e.g., because of that `goto | |
16037 | * again`, assuming other heuristics don't get in a way). | |
16038 | * | |
16039 | * When we first time come to 1:, let's say we have some state X. We proceed | |
16040 | * to 2:, fork states, enqueue ACTIVE, validate NULL case successfully, exit. | |
16041 | * Now we come back to validate that forked ACTIVE state. We proceed through | |
16042 | * 3-5, come to goto, jump to 1:. Let's assume our state didn't change, so we | |
16043 | * are converging. But the problem is that we don't know that yet, as this | |
16044 | * convergence has to happen at iter_next() call site only. So if nothing is | |
16045 | * done, at 1: verifier will use bounded loop logic and declare infinite | |
16046 | * looping (and would be *technically* correct, if not for iterator's | |
16047 | * "eventual sticky NULL" contract, see process_iter_next_call()). But we | |
16048 | * don't want that. So what we do in process_iter_next_call() when we go on | |
16049 | * another ACTIVE iteration, we bump slot->iter.depth, to mark that it's | |
16050 | * a different iteration. So when we suspect an infinite loop, we additionally | |
16051 | * check if any of the *ACTIVE* iterator states depths differ. If yes, we | |
16052 | * pretend we are not looping and wait for next iter_next() call. | |
16053 | * | |
16054 | * This only applies to ACTIVE state. In DRAINED state we don't expect to | |
16055 | * loop, because that would actually mean infinite loop, as DRAINED state is | |
16056 | * "sticky", and so we'll keep returning into the same instruction with the | |
16057 | * same state (at least in one of possible code paths). | |
16058 | * | |
16059 | * This approach allows to keep infinite loop heuristic even in the face of | |
16060 | * active iterator. E.g., C snippet below is and will be detected as | |
16061 | * inifintely looping: | |
16062 | * | |
16063 | * struct bpf_iter_num it; | |
16064 | * int *p, x; | |
16065 | * | |
16066 | * bpf_iter_num_new(&it, 0, 10); | |
16067 | * while ((p = bpf_iter_num_next(&t))) { | |
16068 | * x = p; | |
16069 | * while (x--) {} // <<-- infinite loop here | |
16070 | * } | |
16071 | * | |
16072 | */ | |
16073 | static bool iter_active_depths_differ(struct bpf_verifier_state *old, struct bpf_verifier_state *cur) | |
16074 | { | |
16075 | struct bpf_reg_state *slot, *cur_slot; | |
16076 | struct bpf_func_state *state; | |
16077 | int i, fr; | |
16078 | ||
16079 | for (fr = old->curframe; fr >= 0; fr--) { | |
16080 | state = old->frame[fr]; | |
16081 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
16082 | if (state->stack[i].slot_type[0] != STACK_ITER) | |
16083 | continue; | |
16084 | ||
16085 | slot = &state->stack[i].spilled_ptr; | |
16086 | if (slot->iter.state != BPF_ITER_STATE_ACTIVE) | |
16087 | continue; | |
16088 | ||
16089 | cur_slot = &cur->frame[fr]->stack[i].spilled_ptr; | |
16090 | if (cur_slot->iter.depth != slot->iter.depth) | |
16091 | return true; | |
16092 | } | |
16093 | } | |
16094 | return false; | |
16095 | } | |
2589726d | 16096 | |
58e2af8b | 16097 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 16098 | { |
58e2af8b | 16099 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 16100 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 16101 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 16102 | int i, j, err, states_cnt = 0; |
4b5ce570 AN |
16103 | bool force_new_state = env->test_state_freq || is_force_checkpoint(env, insn_idx); |
16104 | bool add_new_state = force_new_state; | |
f1bca824 | 16105 | |
2589726d AS |
16106 | /* bpf progs typically have pruning point every 4 instructions |
16107 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
16108 | * Do not add new state for future pruning if the verifier hasn't seen | |
16109 | * at least 2 jumps and at least 8 instructions. | |
16110 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
16111 | * In tests that amounts to up to 50% reduction into total verifier | |
16112 | * memory consumption and 20% verifier time speedup. | |
16113 | */ | |
16114 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
16115 | env->insn_processed - env->prev_insn_processed >= 8) | |
16116 | add_new_state = true; | |
16117 | ||
a8f500af AS |
16118 | pprev = explored_state(env, insn_idx); |
16119 | sl = *pprev; | |
16120 | ||
9242b5f5 AS |
16121 | clean_live_states(env, insn_idx, cur); |
16122 | ||
a8f500af | 16123 | while (sl) { |
dc2a4ebc AS |
16124 | states_cnt++; |
16125 | if (sl->state.insn_idx != insn_idx) | |
16126 | goto next; | |
bfc6bb74 | 16127 | |
2589726d | 16128 | if (sl->state.branches) { |
bfc6bb74 AS |
16129 | struct bpf_func_state *frame = sl->state.frame[sl->state.curframe]; |
16130 | ||
16131 | if (frame->in_async_callback_fn && | |
16132 | frame->async_entry_cnt != cur->frame[cur->curframe]->async_entry_cnt) { | |
16133 | /* Different async_entry_cnt means that the verifier is | |
16134 | * processing another entry into async callback. | |
16135 | * Seeing the same state is not an indication of infinite | |
16136 | * loop or infinite recursion. | |
16137 | * But finding the same state doesn't mean that it's safe | |
16138 | * to stop processing the current state. The previous state | |
16139 | * hasn't yet reached bpf_exit, since state.branches > 0. | |
16140 | * Checking in_async_callback_fn alone is not enough either. | |
16141 | * Since the verifier still needs to catch infinite loops | |
16142 | * inside async callbacks. | |
16143 | */ | |
06accc87 AN |
16144 | goto skip_inf_loop_check; |
16145 | } | |
16146 | /* BPF open-coded iterators loop detection is special. | |
16147 | * states_maybe_looping() logic is too simplistic in detecting | |
16148 | * states that *might* be equivalent, because it doesn't know | |
16149 | * about ID remapping, so don't even perform it. | |
16150 | * See process_iter_next_call() and iter_active_depths_differ() | |
16151 | * for overview of the logic. When current and one of parent | |
16152 | * states are detected as equivalent, it's a good thing: we prove | |
16153 | * convergence and can stop simulating further iterations. | |
16154 | * It's safe to assume that iterator loop will finish, taking into | |
16155 | * account iter_next() contract of eventually returning | |
16156 | * sticky NULL result. | |
16157 | */ | |
16158 | if (is_iter_next_insn(env, insn_idx)) { | |
16159 | if (states_equal(env, &sl->state, cur)) { | |
16160 | struct bpf_func_state *cur_frame; | |
16161 | struct bpf_reg_state *iter_state, *iter_reg; | |
16162 | int spi; | |
16163 | ||
16164 | cur_frame = cur->frame[cur->curframe]; | |
16165 | /* btf_check_iter_kfuncs() enforces that | |
16166 | * iter state pointer is always the first arg | |
16167 | */ | |
16168 | iter_reg = &cur_frame->regs[BPF_REG_1]; | |
16169 | /* current state is valid due to states_equal(), | |
16170 | * so we can assume valid iter and reg state, | |
16171 | * no need for extra (re-)validations | |
16172 | */ | |
16173 | spi = __get_spi(iter_reg->off + iter_reg->var_off.value); | |
16174 | iter_state = &func(env, iter_reg)->stack[spi].spilled_ptr; | |
16175 | if (iter_state->iter.state == BPF_ITER_STATE_ACTIVE) | |
16176 | goto hit; | |
16177 | } | |
16178 | goto skip_inf_loop_check; | |
16179 | } | |
16180 | /* attempt to detect infinite loop to avoid unnecessary doomed work */ | |
16181 | if (states_maybe_looping(&sl->state, cur) && | |
16182 | states_equal(env, &sl->state, cur) && | |
16183 | !iter_active_depths_differ(&sl->state, cur)) { | |
2589726d AS |
16184 | verbose_linfo(env, insn_idx, "; "); |
16185 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
16186 | return -EINVAL; | |
16187 | } | |
16188 | /* if the verifier is processing a loop, avoid adding new state | |
16189 | * too often, since different loop iterations have distinct | |
16190 | * states and may not help future pruning. | |
16191 | * This threshold shouldn't be too low to make sure that | |
16192 | * a loop with large bound will be rejected quickly. | |
16193 | * The most abusive loop will be: | |
16194 | * r1 += 1 | |
16195 | * if r1 < 1000000 goto pc-2 | |
16196 | * 1M insn_procssed limit / 100 == 10k peak states. | |
16197 | * This threshold shouldn't be too high either, since states | |
16198 | * at the end of the loop are likely to be useful in pruning. | |
16199 | */ | |
06accc87 | 16200 | skip_inf_loop_check: |
4b5ce570 | 16201 | if (!force_new_state && |
98ddcf38 | 16202 | env->jmps_processed - env->prev_jmps_processed < 20 && |
2589726d AS |
16203 | env->insn_processed - env->prev_insn_processed < 100) |
16204 | add_new_state = false; | |
16205 | goto miss; | |
16206 | } | |
638f5b90 | 16207 | if (states_equal(env, &sl->state, cur)) { |
06accc87 | 16208 | hit: |
9f4686c4 | 16209 | sl->hit_cnt++; |
f1bca824 | 16210 | /* reached equivalent register/stack state, |
dc503a8a EC |
16211 | * prune the search. |
16212 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
16213 | * If we have any write marks in env->cur_state, they |
16214 | * will prevent corresponding reads in the continuation | |
16215 | * from reaching our parent (an explored_state). Our | |
16216 | * own state will get the read marks recorded, but | |
16217 | * they'll be immediately forgotten as we're pruning | |
16218 | * this state and will pop a new one. | |
f1bca824 | 16219 | */ |
f4d7e40a | 16220 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
16221 | |
16222 | /* if previous state reached the exit with precision and | |
16223 | * current state is equivalent to it (except precsion marks) | |
16224 | * the precision needs to be propagated back in | |
16225 | * the current state. | |
16226 | */ | |
16227 | err = err ? : push_jmp_history(env, cur); | |
16228 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
16229 | if (err) |
16230 | return err; | |
f1bca824 | 16231 | return 1; |
dc503a8a | 16232 | } |
2589726d AS |
16233 | miss: |
16234 | /* when new state is not going to be added do not increase miss count. | |
16235 | * Otherwise several loop iterations will remove the state | |
16236 | * recorded earlier. The goal of these heuristics is to have | |
16237 | * states from some iterations of the loop (some in the beginning | |
16238 | * and some at the end) to help pruning. | |
16239 | */ | |
16240 | if (add_new_state) | |
16241 | sl->miss_cnt++; | |
9f4686c4 AS |
16242 | /* heuristic to determine whether this state is beneficial |
16243 | * to keep checking from state equivalence point of view. | |
16244 | * Higher numbers increase max_states_per_insn and verification time, | |
16245 | * but do not meaningfully decrease insn_processed. | |
16246 | */ | |
16247 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
16248 | /* the state is unlikely to be useful. Remove it to | |
16249 | * speed up verification | |
16250 | */ | |
16251 | *pprev = sl->next; | |
16252 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
2589726d AS |
16253 | u32 br = sl->state.branches; |
16254 | ||
16255 | WARN_ONCE(br, | |
16256 | "BUG live_done but branches_to_explore %d\n", | |
16257 | br); | |
9f4686c4 AS |
16258 | free_verifier_state(&sl->state, false); |
16259 | kfree(sl); | |
16260 | env->peak_states--; | |
16261 | } else { | |
16262 | /* cannot free this state, since parentage chain may | |
16263 | * walk it later. Add it for free_list instead to | |
16264 | * be freed at the end of verification | |
16265 | */ | |
16266 | sl->next = env->free_list; | |
16267 | env->free_list = sl; | |
16268 | } | |
16269 | sl = *pprev; | |
16270 | continue; | |
16271 | } | |
dc2a4ebc | 16272 | next: |
9f4686c4 AS |
16273 | pprev = &sl->next; |
16274 | sl = *pprev; | |
f1bca824 AS |
16275 | } |
16276 | ||
06ee7115 AS |
16277 | if (env->max_states_per_insn < states_cnt) |
16278 | env->max_states_per_insn = states_cnt; | |
16279 | ||
2c78ee89 | 16280 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
a095f421 | 16281 | return 0; |
ceefbc96 | 16282 | |
2589726d | 16283 | if (!add_new_state) |
a095f421 | 16284 | return 0; |
ceefbc96 | 16285 | |
2589726d AS |
16286 | /* There were no equivalent states, remember the current one. |
16287 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 16288 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 16289 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 16290 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
16291 | * again on the way to bpf_exit. |
16292 | * When looping the sl->state.branches will be > 0 and this state | |
16293 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 16294 | */ |
638f5b90 | 16295 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
16296 | if (!new_sl) |
16297 | return -ENOMEM; | |
06ee7115 AS |
16298 | env->total_states++; |
16299 | env->peak_states++; | |
2589726d AS |
16300 | env->prev_jmps_processed = env->jmps_processed; |
16301 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 | 16302 | |
7a830b53 AN |
16303 | /* forget precise markings we inherited, see __mark_chain_precision */ |
16304 | if (env->bpf_capable) | |
16305 | mark_all_scalars_imprecise(env, cur); | |
16306 | ||
f1bca824 | 16307 | /* add new state to the head of linked list */ |
679c782d EC |
16308 | new = &new_sl->state; |
16309 | err = copy_verifier_state(new, cur); | |
1969db47 | 16310 | if (err) { |
679c782d | 16311 | free_verifier_state(new, false); |
1969db47 AS |
16312 | kfree(new_sl); |
16313 | return err; | |
16314 | } | |
dc2a4ebc | 16315 | new->insn_idx = insn_idx; |
2589726d AS |
16316 | WARN_ONCE(new->branches != 1, |
16317 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 16318 | |
2589726d | 16319 | cur->parent = new; |
b5dc0163 AS |
16320 | cur->first_insn_idx = insn_idx; |
16321 | clear_jmp_history(cur); | |
5d839021 AS |
16322 | new_sl->next = *explored_state(env, insn_idx); |
16323 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
16324 | /* connect new state to parentage chain. Current frame needs all |
16325 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
16326 | * to the stack implicitly by JITs) so in callers' frames connect just | |
16327 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
16328 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
16329 | * from callee with its full parentage chain, anyway. | |
16330 | */ | |
8e9cd9ce EC |
16331 | /* clear write marks in current state: the writes we did are not writes |
16332 | * our child did, so they don't screen off its reads from us. | |
16333 | * (There are no read marks in current state, because reads always mark | |
16334 | * their parent and current state never has children yet. Only | |
16335 | * explored_states can get read marks.) | |
16336 | */ | |
eea1c227 AS |
16337 | for (j = 0; j <= cur->curframe; j++) { |
16338 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
16339 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
16340 | for (i = 0; i < BPF_REG_FP; i++) | |
16341 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
16342 | } | |
f4d7e40a AS |
16343 | |
16344 | /* all stack frames are accessible from callee, clear them all */ | |
16345 | for (j = 0; j <= cur->curframe; j++) { | |
16346 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 16347 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 16348 | |
679c782d | 16349 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 16350 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
16351 | frame->stack[i].spilled_ptr.parent = |
16352 | &newframe->stack[i].spilled_ptr; | |
16353 | } | |
f4d7e40a | 16354 | } |
f1bca824 AS |
16355 | return 0; |
16356 | } | |
16357 | ||
c64b7983 JS |
16358 | /* Return true if it's OK to have the same insn return a different type. */ |
16359 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
16360 | { | |
c25b2ae1 | 16361 | switch (base_type(type)) { |
c64b7983 JS |
16362 | case PTR_TO_CTX: |
16363 | case PTR_TO_SOCKET: | |
46f8bc92 | 16364 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 16365 | case PTR_TO_TCP_SOCK: |
fada7fdc | 16366 | case PTR_TO_XDP_SOCK: |
2a02759e | 16367 | case PTR_TO_BTF_ID: |
c64b7983 JS |
16368 | return false; |
16369 | default: | |
16370 | return true; | |
16371 | } | |
16372 | } | |
16373 | ||
16374 | /* If an instruction was previously used with particular pointer types, then we | |
16375 | * need to be careful to avoid cases such as the below, where it may be ok | |
16376 | * for one branch accessing the pointer, but not ok for the other branch: | |
16377 | * | |
16378 | * R1 = sock_ptr | |
16379 | * goto X; | |
16380 | * ... | |
16381 | * R1 = some_other_valid_ptr; | |
16382 | * goto X; | |
16383 | * ... | |
16384 | * R2 = *(u32 *)(R1 + 0); | |
16385 | */ | |
16386 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
16387 | { | |
16388 | return src != prev && (!reg_type_mismatch_ok(src) || | |
16389 | !reg_type_mismatch_ok(prev)); | |
16390 | } | |
16391 | ||
0d80a619 EZ |
16392 | static int save_aux_ptr_type(struct bpf_verifier_env *env, enum bpf_reg_type type, |
16393 | bool allow_trust_missmatch) | |
16394 | { | |
16395 | enum bpf_reg_type *prev_type = &env->insn_aux_data[env->insn_idx].ptr_type; | |
16396 | ||
16397 | if (*prev_type == NOT_INIT) { | |
16398 | /* Saw a valid insn | |
16399 | * dst_reg = *(u32 *)(src_reg + off) | |
16400 | * save type to validate intersecting paths | |
16401 | */ | |
16402 | *prev_type = type; | |
16403 | } else if (reg_type_mismatch(type, *prev_type)) { | |
16404 | /* Abuser program is trying to use the same insn | |
16405 | * dst_reg = *(u32*) (src_reg + off) | |
16406 | * with different pointer types: | |
16407 | * src_reg == ctx in one branch and | |
16408 | * src_reg == stack|map in some other branch. | |
16409 | * Reject it. | |
16410 | */ | |
16411 | if (allow_trust_missmatch && | |
16412 | base_type(type) == PTR_TO_BTF_ID && | |
16413 | base_type(*prev_type) == PTR_TO_BTF_ID) { | |
16414 | /* | |
16415 | * Have to support a use case when one path through | |
16416 | * the program yields TRUSTED pointer while another | |
16417 | * is UNTRUSTED. Fallback to UNTRUSTED to generate | |
1f9a1ea8 | 16418 | * BPF_PROBE_MEM/BPF_PROBE_MEMSX. |
0d80a619 EZ |
16419 | */ |
16420 | *prev_type = PTR_TO_BTF_ID | PTR_UNTRUSTED; | |
16421 | } else { | |
16422 | verbose(env, "same insn cannot be used with different pointers\n"); | |
16423 | return -EINVAL; | |
16424 | } | |
16425 | } | |
16426 | ||
16427 | return 0; | |
16428 | } | |
16429 | ||
58e2af8b | 16430 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 16431 | { |
6f8a57cc | 16432 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 16433 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 16434 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 16435 | struct bpf_reg_state *regs; |
06ee7115 | 16436 | int insn_cnt = env->prog->len; |
17a52670 | 16437 | bool do_print_state = false; |
b5dc0163 | 16438 | int prev_insn_idx = -1; |
17a52670 | 16439 | |
17a52670 AS |
16440 | for (;;) { |
16441 | struct bpf_insn *insn; | |
16442 | u8 class; | |
16443 | int err; | |
16444 | ||
b5dc0163 | 16445 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 16446 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 16447 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 16448 | env->insn_idx, insn_cnt); |
17a52670 AS |
16449 | return -EFAULT; |
16450 | } | |
16451 | ||
c08435ec | 16452 | insn = &insns[env->insn_idx]; |
17a52670 AS |
16453 | class = BPF_CLASS(insn->code); |
16454 | ||
06ee7115 | 16455 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
16456 | verbose(env, |
16457 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 16458 | env->insn_processed); |
17a52670 AS |
16459 | return -E2BIG; |
16460 | } | |
16461 | ||
a095f421 AN |
16462 | state->last_insn_idx = env->prev_insn_idx; |
16463 | ||
16464 | if (is_prune_point(env, env->insn_idx)) { | |
16465 | err = is_state_visited(env, env->insn_idx); | |
16466 | if (err < 0) | |
16467 | return err; | |
16468 | if (err == 1) { | |
16469 | /* found equivalent state, can prune the search */ | |
16470 | if (env->log.level & BPF_LOG_LEVEL) { | |
16471 | if (do_print_state) | |
16472 | verbose(env, "\nfrom %d to %d%s: safe\n", | |
16473 | env->prev_insn_idx, env->insn_idx, | |
16474 | env->cur_state->speculative ? | |
16475 | " (speculative execution)" : ""); | |
16476 | else | |
16477 | verbose(env, "%d: safe\n", env->insn_idx); | |
16478 | } | |
16479 | goto process_bpf_exit; | |
f1bca824 | 16480 | } |
a095f421 AN |
16481 | } |
16482 | ||
16483 | if (is_jmp_point(env, env->insn_idx)) { | |
16484 | err = push_jmp_history(env, state); | |
16485 | if (err) | |
16486 | return err; | |
f1bca824 AS |
16487 | } |
16488 | ||
c3494801 AS |
16489 | if (signal_pending(current)) |
16490 | return -EAGAIN; | |
16491 | ||
3c2ce60b DB |
16492 | if (need_resched()) |
16493 | cond_resched(); | |
16494 | ||
2e576648 CL |
16495 | if (env->log.level & BPF_LOG_LEVEL2 && do_print_state) { |
16496 | verbose(env, "\nfrom %d to %d%s:", | |
16497 | env->prev_insn_idx, env->insn_idx, | |
16498 | env->cur_state->speculative ? | |
16499 | " (speculative execution)" : ""); | |
16500 | print_verifier_state(env, state->frame[state->curframe], true); | |
17a52670 AS |
16501 | do_print_state = false; |
16502 | } | |
16503 | ||
06ee7115 | 16504 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 | 16505 | const struct bpf_insn_cbs cbs = { |
e6ac2450 | 16506 | .cb_call = disasm_kfunc_name, |
7105e828 | 16507 | .cb_print = verbose, |
abe08840 | 16508 | .private_data = env, |
7105e828 DB |
16509 | }; |
16510 | ||
2e576648 CL |
16511 | if (verifier_state_scratched(env)) |
16512 | print_insn_state(env, state->frame[state->curframe]); | |
16513 | ||
c08435ec | 16514 | verbose_linfo(env, env->insn_idx, "; "); |
12166409 | 16515 | env->prev_log_pos = env->log.end_pos; |
c08435ec | 16516 | verbose(env, "%d: ", env->insn_idx); |
abe08840 | 16517 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
12166409 AN |
16518 | env->prev_insn_print_pos = env->log.end_pos - env->prev_log_pos; |
16519 | env->prev_log_pos = env->log.end_pos; | |
17a52670 AS |
16520 | } |
16521 | ||
9d03ebc7 | 16522 | if (bpf_prog_is_offloaded(env->prog->aux)) { |
c08435ec DB |
16523 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
16524 | env->prev_insn_idx); | |
cae1927c JK |
16525 | if (err) |
16526 | return err; | |
16527 | } | |
13a27dfc | 16528 | |
638f5b90 | 16529 | regs = cur_regs(env); |
fe9a5ca7 | 16530 | sanitize_mark_insn_seen(env); |
b5dc0163 | 16531 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 16532 | |
17a52670 | 16533 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 16534 | err = check_alu_op(env, insn); |
17a52670 AS |
16535 | if (err) |
16536 | return err; | |
16537 | ||
16538 | } else if (class == BPF_LDX) { | |
0d80a619 | 16539 | enum bpf_reg_type src_reg_type; |
9bac3d6d AS |
16540 | |
16541 | /* check for reserved fields is already done */ | |
16542 | ||
17a52670 | 16543 | /* check src operand */ |
dc503a8a | 16544 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
16545 | if (err) |
16546 | return err; | |
16547 | ||
dc503a8a | 16548 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
16549 | if (err) |
16550 | return err; | |
16551 | ||
725f9dcd AS |
16552 | src_reg_type = regs[insn->src_reg].type; |
16553 | ||
17a52670 AS |
16554 | /* check that memory (src_reg + off) is readable, |
16555 | * the state of dst_reg will be updated by this func | |
16556 | */ | |
c08435ec DB |
16557 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
16558 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 YS |
16559 | BPF_READ, insn->dst_reg, false, |
16560 | BPF_MODE(insn->code) == BPF_MEMSX); | |
17a52670 AS |
16561 | if (err) |
16562 | return err; | |
16563 | ||
0d80a619 EZ |
16564 | err = save_aux_ptr_type(env, src_reg_type, true); |
16565 | if (err) | |
16566 | return err; | |
17a52670 | 16567 | } else if (class == BPF_STX) { |
0d80a619 | 16568 | enum bpf_reg_type dst_reg_type; |
d691f9e8 | 16569 | |
91c960b0 BJ |
16570 | if (BPF_MODE(insn->code) == BPF_ATOMIC) { |
16571 | err = check_atomic(env, env->insn_idx, insn); | |
17a52670 AS |
16572 | if (err) |
16573 | return err; | |
c08435ec | 16574 | env->insn_idx++; |
17a52670 AS |
16575 | continue; |
16576 | } | |
16577 | ||
5ca419f2 BJ |
16578 | if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) { |
16579 | verbose(env, "BPF_STX uses reserved fields\n"); | |
16580 | return -EINVAL; | |
16581 | } | |
16582 | ||
17a52670 | 16583 | /* check src1 operand */ |
dc503a8a | 16584 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
16585 | if (err) |
16586 | return err; | |
16587 | /* check src2 operand */ | |
dc503a8a | 16588 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
16589 | if (err) |
16590 | return err; | |
16591 | ||
d691f9e8 AS |
16592 | dst_reg_type = regs[insn->dst_reg].type; |
16593 | ||
17a52670 | 16594 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
16595 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
16596 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 | 16597 | BPF_WRITE, insn->src_reg, false, false); |
17a52670 AS |
16598 | if (err) |
16599 | return err; | |
16600 | ||
0d80a619 EZ |
16601 | err = save_aux_ptr_type(env, dst_reg_type, false); |
16602 | if (err) | |
16603 | return err; | |
17a52670 | 16604 | } else if (class == BPF_ST) { |
0d80a619 EZ |
16605 | enum bpf_reg_type dst_reg_type; |
16606 | ||
17a52670 AS |
16607 | if (BPF_MODE(insn->code) != BPF_MEM || |
16608 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 16609 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
16610 | return -EINVAL; |
16611 | } | |
16612 | /* check src operand */ | |
dc503a8a | 16613 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
16614 | if (err) |
16615 | return err; | |
16616 | ||
0d80a619 | 16617 | dst_reg_type = regs[insn->dst_reg].type; |
f37a8cb8 | 16618 | |
17a52670 | 16619 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
16620 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
16621 | insn->off, BPF_SIZE(insn->code), | |
1f9a1ea8 | 16622 | BPF_WRITE, -1, false, false); |
17a52670 AS |
16623 | if (err) |
16624 | return err; | |
16625 | ||
0d80a619 EZ |
16626 | err = save_aux_ptr_type(env, dst_reg_type, false); |
16627 | if (err) | |
16628 | return err; | |
092ed096 | 16629 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
16630 | u8 opcode = BPF_OP(insn->code); |
16631 | ||
2589726d | 16632 | env->jmps_processed++; |
17a52670 AS |
16633 | if (opcode == BPF_CALL) { |
16634 | if (BPF_SRC(insn->code) != BPF_K || | |
2357672c KKD |
16635 | (insn->src_reg != BPF_PSEUDO_KFUNC_CALL |
16636 | && insn->off != 0) || | |
f4d7e40a | 16637 | (insn->src_reg != BPF_REG_0 && |
e6ac2450 MKL |
16638 | insn->src_reg != BPF_PSEUDO_CALL && |
16639 | insn->src_reg != BPF_PSEUDO_KFUNC_CALL) || | |
092ed096 JW |
16640 | insn->dst_reg != BPF_REG_0 || |
16641 | class == BPF_JMP32) { | |
61bd5218 | 16642 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
16643 | return -EINVAL; |
16644 | } | |
16645 | ||
8cab76ec KKD |
16646 | if (env->cur_state->active_lock.ptr) { |
16647 | if ((insn->src_reg == BPF_REG_0 && insn->imm != BPF_FUNC_spin_unlock) || | |
16648 | (insn->src_reg == BPF_PSEUDO_CALL) || | |
16649 | (insn->src_reg == BPF_PSEUDO_KFUNC_CALL && | |
cd6791b4 | 16650 | (insn->off != 0 || !is_bpf_graph_api_kfunc(insn->imm)))) { |
8cab76ec KKD |
16651 | verbose(env, "function calls are not allowed while holding a lock\n"); |
16652 | return -EINVAL; | |
16653 | } | |
d83525ca | 16654 | } |
f4d7e40a | 16655 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 16656 | err = check_func_call(env, insn, &env->insn_idx); |
e6ac2450 | 16657 | else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) |
5c073f26 | 16658 | err = check_kfunc_call(env, insn, &env->insn_idx); |
f4d7e40a | 16659 | else |
69c087ba | 16660 | err = check_helper_call(env, insn, &env->insn_idx); |
17a52670 AS |
16661 | if (err) |
16662 | return err; | |
553a64a8 AN |
16663 | |
16664 | mark_reg_scratched(env, BPF_REG_0); | |
17a52670 AS |
16665 | } else if (opcode == BPF_JA) { |
16666 | if (BPF_SRC(insn->code) != BPF_K || | |
17a52670 | 16667 | insn->src_reg != BPF_REG_0 || |
092ed096 | 16668 | insn->dst_reg != BPF_REG_0 || |
4cd58e9a YS |
16669 | (class == BPF_JMP && insn->imm != 0) || |
16670 | (class == BPF_JMP32 && insn->off != 0)) { | |
61bd5218 | 16671 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
16672 | return -EINVAL; |
16673 | } | |
16674 | ||
4cd58e9a YS |
16675 | if (class == BPF_JMP) |
16676 | env->insn_idx += insn->off + 1; | |
16677 | else | |
16678 | env->insn_idx += insn->imm + 1; | |
17a52670 AS |
16679 | continue; |
16680 | ||
16681 | } else if (opcode == BPF_EXIT) { | |
16682 | if (BPF_SRC(insn->code) != BPF_K || | |
16683 | insn->imm != 0 || | |
16684 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
16685 | insn->dst_reg != BPF_REG_0 || |
16686 | class == BPF_JMP32) { | |
61bd5218 | 16687 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
16688 | return -EINVAL; |
16689 | } | |
16690 | ||
5d92ddc3 DM |
16691 | if (env->cur_state->active_lock.ptr && |
16692 | !in_rbtree_lock_required_cb(env)) { | |
d83525ca AS |
16693 | verbose(env, "bpf_spin_unlock is missing\n"); |
16694 | return -EINVAL; | |
16695 | } | |
16696 | ||
0816b8c6 DM |
16697 | if (env->cur_state->active_rcu_lock && |
16698 | !in_rbtree_lock_required_cb(env)) { | |
9bb00b28 YS |
16699 | verbose(env, "bpf_rcu_read_unlock is missing\n"); |
16700 | return -EINVAL; | |
16701 | } | |
16702 | ||
9d9d00ac KKD |
16703 | /* We must do check_reference_leak here before |
16704 | * prepare_func_exit to handle the case when | |
16705 | * state->curframe > 0, it may be a callback | |
16706 | * function, for which reference_state must | |
16707 | * match caller reference state when it exits. | |
16708 | */ | |
16709 | err = check_reference_leak(env); | |
16710 | if (err) | |
16711 | return err; | |
16712 | ||
f4d7e40a AS |
16713 | if (state->curframe) { |
16714 | /* exit from nested function */ | |
c08435ec | 16715 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
16716 | if (err) |
16717 | return err; | |
16718 | do_print_state = true; | |
16719 | continue; | |
16720 | } | |
16721 | ||
390ee7e2 AS |
16722 | err = check_return_code(env); |
16723 | if (err) | |
16724 | return err; | |
f1bca824 | 16725 | process_bpf_exit: |
0f55f9ed | 16726 | mark_verifier_state_scratched(env); |
2589726d | 16727 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 16728 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 16729 | &env->insn_idx, pop_log); |
638f5b90 AS |
16730 | if (err < 0) { |
16731 | if (err != -ENOENT) | |
16732 | return err; | |
17a52670 AS |
16733 | break; |
16734 | } else { | |
16735 | do_print_state = true; | |
16736 | continue; | |
16737 | } | |
16738 | } else { | |
c08435ec | 16739 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
16740 | if (err) |
16741 | return err; | |
16742 | } | |
16743 | } else if (class == BPF_LD) { | |
16744 | u8 mode = BPF_MODE(insn->code); | |
16745 | ||
16746 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
16747 | err = check_ld_abs(env, insn); |
16748 | if (err) | |
16749 | return err; | |
16750 | ||
17a52670 AS |
16751 | } else if (mode == BPF_IMM) { |
16752 | err = check_ld_imm(env, insn); | |
16753 | if (err) | |
16754 | return err; | |
16755 | ||
c08435ec | 16756 | env->insn_idx++; |
fe9a5ca7 | 16757 | sanitize_mark_insn_seen(env); |
17a52670 | 16758 | } else { |
61bd5218 | 16759 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
16760 | return -EINVAL; |
16761 | } | |
16762 | } else { | |
61bd5218 | 16763 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
16764 | return -EINVAL; |
16765 | } | |
16766 | ||
c08435ec | 16767 | env->insn_idx++; |
17a52670 AS |
16768 | } |
16769 | ||
16770 | return 0; | |
16771 | } | |
16772 | ||
541c3bad AN |
16773 | static int find_btf_percpu_datasec(struct btf *btf) |
16774 | { | |
16775 | const struct btf_type *t; | |
16776 | const char *tname; | |
16777 | int i, n; | |
16778 | ||
16779 | /* | |
16780 | * Both vmlinux and module each have their own ".data..percpu" | |
16781 | * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF | |
16782 | * types to look at only module's own BTF types. | |
16783 | */ | |
16784 | n = btf_nr_types(btf); | |
16785 | if (btf_is_module(btf)) | |
16786 | i = btf_nr_types(btf_vmlinux); | |
16787 | else | |
16788 | i = 1; | |
16789 | ||
16790 | for(; i < n; i++) { | |
16791 | t = btf_type_by_id(btf, i); | |
16792 | if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC) | |
16793 | continue; | |
16794 | ||
16795 | tname = btf_name_by_offset(btf, t->name_off); | |
16796 | if (!strcmp(tname, ".data..percpu")) | |
16797 | return i; | |
16798 | } | |
16799 | ||
16800 | return -ENOENT; | |
16801 | } | |
16802 | ||
4976b718 HL |
16803 | /* replace pseudo btf_id with kernel symbol address */ |
16804 | static int check_pseudo_btf_id(struct bpf_verifier_env *env, | |
16805 | struct bpf_insn *insn, | |
16806 | struct bpf_insn_aux_data *aux) | |
16807 | { | |
eaa6bcb7 HL |
16808 | const struct btf_var_secinfo *vsi; |
16809 | const struct btf_type *datasec; | |
541c3bad | 16810 | struct btf_mod_pair *btf_mod; |
4976b718 HL |
16811 | const struct btf_type *t; |
16812 | const char *sym_name; | |
eaa6bcb7 | 16813 | bool percpu = false; |
f16e6313 | 16814 | u32 type, id = insn->imm; |
541c3bad | 16815 | struct btf *btf; |
f16e6313 | 16816 | s32 datasec_id; |
4976b718 | 16817 | u64 addr; |
541c3bad | 16818 | int i, btf_fd, err; |
4976b718 | 16819 | |
541c3bad AN |
16820 | btf_fd = insn[1].imm; |
16821 | if (btf_fd) { | |
16822 | btf = btf_get_by_fd(btf_fd); | |
16823 | if (IS_ERR(btf)) { | |
16824 | verbose(env, "invalid module BTF object FD specified.\n"); | |
16825 | return -EINVAL; | |
16826 | } | |
16827 | } else { | |
16828 | if (!btf_vmlinux) { | |
16829 | verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n"); | |
16830 | return -EINVAL; | |
16831 | } | |
16832 | btf = btf_vmlinux; | |
16833 | btf_get(btf); | |
4976b718 HL |
16834 | } |
16835 | ||
541c3bad | 16836 | t = btf_type_by_id(btf, id); |
4976b718 HL |
16837 | if (!t) { |
16838 | verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id); | |
541c3bad AN |
16839 | err = -ENOENT; |
16840 | goto err_put; | |
4976b718 HL |
16841 | } |
16842 | ||
58aa2afb AS |
16843 | if (!btf_type_is_var(t) && !btf_type_is_func(t)) { |
16844 | verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR or KIND_FUNC\n", id); | |
541c3bad AN |
16845 | err = -EINVAL; |
16846 | goto err_put; | |
4976b718 HL |
16847 | } |
16848 | ||
541c3bad | 16849 | sym_name = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
16850 | addr = kallsyms_lookup_name(sym_name); |
16851 | if (!addr) { | |
16852 | verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n", | |
16853 | sym_name); | |
541c3bad AN |
16854 | err = -ENOENT; |
16855 | goto err_put; | |
4976b718 | 16856 | } |
58aa2afb AS |
16857 | insn[0].imm = (u32)addr; |
16858 | insn[1].imm = addr >> 32; | |
16859 | ||
16860 | if (btf_type_is_func(t)) { | |
16861 | aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY; | |
16862 | aux->btf_var.mem_size = 0; | |
16863 | goto check_btf; | |
16864 | } | |
4976b718 | 16865 | |
541c3bad | 16866 | datasec_id = find_btf_percpu_datasec(btf); |
eaa6bcb7 | 16867 | if (datasec_id > 0) { |
541c3bad | 16868 | datasec = btf_type_by_id(btf, datasec_id); |
eaa6bcb7 HL |
16869 | for_each_vsi(i, datasec, vsi) { |
16870 | if (vsi->type == id) { | |
16871 | percpu = true; | |
16872 | break; | |
16873 | } | |
16874 | } | |
16875 | } | |
16876 | ||
4976b718 | 16877 | type = t->type; |
541c3bad | 16878 | t = btf_type_skip_modifiers(btf, type, NULL); |
eaa6bcb7 | 16879 | if (percpu) { |
5844101a | 16880 | aux->btf_var.reg_type = PTR_TO_BTF_ID | MEM_PERCPU; |
541c3bad | 16881 | aux->btf_var.btf = btf; |
eaa6bcb7 HL |
16882 | aux->btf_var.btf_id = type; |
16883 | } else if (!btf_type_is_struct(t)) { | |
4976b718 HL |
16884 | const struct btf_type *ret; |
16885 | const char *tname; | |
16886 | u32 tsize; | |
16887 | ||
16888 | /* resolve the type size of ksym. */ | |
541c3bad | 16889 | ret = btf_resolve_size(btf, t, &tsize); |
4976b718 | 16890 | if (IS_ERR(ret)) { |
541c3bad | 16891 | tname = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
16892 | verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n", |
16893 | tname, PTR_ERR(ret)); | |
541c3bad AN |
16894 | err = -EINVAL; |
16895 | goto err_put; | |
4976b718 | 16896 | } |
34d3a78c | 16897 | aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY; |
4976b718 HL |
16898 | aux->btf_var.mem_size = tsize; |
16899 | } else { | |
16900 | aux->btf_var.reg_type = PTR_TO_BTF_ID; | |
541c3bad | 16901 | aux->btf_var.btf = btf; |
4976b718 HL |
16902 | aux->btf_var.btf_id = type; |
16903 | } | |
58aa2afb | 16904 | check_btf: |
541c3bad AN |
16905 | /* check whether we recorded this BTF (and maybe module) already */ |
16906 | for (i = 0; i < env->used_btf_cnt; i++) { | |
16907 | if (env->used_btfs[i].btf == btf) { | |
16908 | btf_put(btf); | |
16909 | return 0; | |
16910 | } | |
16911 | } | |
16912 | ||
16913 | if (env->used_btf_cnt >= MAX_USED_BTFS) { | |
16914 | err = -E2BIG; | |
16915 | goto err_put; | |
16916 | } | |
16917 | ||
16918 | btf_mod = &env->used_btfs[env->used_btf_cnt]; | |
16919 | btf_mod->btf = btf; | |
16920 | btf_mod->module = NULL; | |
16921 | ||
16922 | /* if we reference variables from kernel module, bump its refcount */ | |
16923 | if (btf_is_module(btf)) { | |
16924 | btf_mod->module = btf_try_get_module(btf); | |
16925 | if (!btf_mod->module) { | |
16926 | err = -ENXIO; | |
16927 | goto err_put; | |
16928 | } | |
16929 | } | |
16930 | ||
16931 | env->used_btf_cnt++; | |
16932 | ||
4976b718 | 16933 | return 0; |
541c3bad AN |
16934 | err_put: |
16935 | btf_put(btf); | |
16936 | return err; | |
4976b718 HL |
16937 | } |
16938 | ||
d83525ca AS |
16939 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
16940 | { | |
16941 | switch (type) { | |
16942 | case BPF_PROG_TYPE_KPROBE: | |
16943 | case BPF_PROG_TYPE_TRACEPOINT: | |
16944 | case BPF_PROG_TYPE_PERF_EVENT: | |
16945 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
5002615a | 16946 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: |
d83525ca AS |
16947 | return true; |
16948 | default: | |
16949 | return false; | |
16950 | } | |
16951 | } | |
16952 | ||
61bd5218 JK |
16953 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
16954 | struct bpf_map *map, | |
fdc15d38 AS |
16955 | struct bpf_prog *prog) |
16956 | ||
16957 | { | |
7e40781c | 16958 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
a3884572 | 16959 | |
9c395c1b DM |
16960 | if (btf_record_has_field(map->record, BPF_LIST_HEAD) || |
16961 | btf_record_has_field(map->record, BPF_RB_ROOT)) { | |
f0c5941f | 16962 | if (is_tracing_prog_type(prog_type)) { |
9c395c1b | 16963 | verbose(env, "tracing progs cannot use bpf_{list_head,rb_root} yet\n"); |
f0c5941f KKD |
16964 | return -EINVAL; |
16965 | } | |
16966 | } | |
16967 | ||
db559117 | 16968 | if (btf_record_has_field(map->record, BPF_SPIN_LOCK)) { |
9e7a4d98 KS |
16969 | if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) { |
16970 | verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n"); | |
16971 | return -EINVAL; | |
16972 | } | |
16973 | ||
16974 | if (is_tracing_prog_type(prog_type)) { | |
16975 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
16976 | return -EINVAL; | |
16977 | } | |
16978 | ||
16979 | if (prog->aux->sleepable) { | |
16980 | verbose(env, "sleepable progs cannot use bpf_spin_lock yet\n"); | |
16981 | return -EINVAL; | |
16982 | } | |
d83525ca AS |
16983 | } |
16984 | ||
db559117 | 16985 | if (btf_record_has_field(map->record, BPF_TIMER)) { |
5e0bc308 DB |
16986 | if (is_tracing_prog_type(prog_type)) { |
16987 | verbose(env, "tracing progs cannot use bpf_timer yet\n"); | |
16988 | return -EINVAL; | |
16989 | } | |
16990 | } | |
16991 | ||
9d03ebc7 | 16992 | if ((bpf_prog_is_offloaded(prog->aux) || bpf_map_is_offloaded(map)) && |
09728266 | 16993 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
16994 | verbose(env, "offload device mismatch between prog and map\n"); |
16995 | return -EINVAL; | |
16996 | } | |
16997 | ||
85d33df3 MKL |
16998 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
16999 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
17000 | return -EINVAL; | |
17001 | } | |
17002 | ||
1e6c62a8 AS |
17003 | if (prog->aux->sleepable) |
17004 | switch (map->map_type) { | |
17005 | case BPF_MAP_TYPE_HASH: | |
17006 | case BPF_MAP_TYPE_LRU_HASH: | |
17007 | case BPF_MAP_TYPE_ARRAY: | |
638e4b82 AS |
17008 | case BPF_MAP_TYPE_PERCPU_HASH: |
17009 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
17010 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
17011 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
17012 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
ba90c2cc | 17013 | case BPF_MAP_TYPE_RINGBUF: |
583c1f42 | 17014 | case BPF_MAP_TYPE_USER_RINGBUF: |
0fe4b381 KS |
17015 | case BPF_MAP_TYPE_INODE_STORAGE: |
17016 | case BPF_MAP_TYPE_SK_STORAGE: | |
17017 | case BPF_MAP_TYPE_TASK_STORAGE: | |
2c40d97d | 17018 | case BPF_MAP_TYPE_CGRP_STORAGE: |
ba90c2cc | 17019 | break; |
1e6c62a8 AS |
17020 | default: |
17021 | verbose(env, | |
2c40d97d | 17022 | "Sleepable programs can only use array, hash, ringbuf and local storage maps\n"); |
1e6c62a8 AS |
17023 | return -EINVAL; |
17024 | } | |
17025 | ||
fdc15d38 AS |
17026 | return 0; |
17027 | } | |
17028 | ||
b741f163 RG |
17029 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
17030 | { | |
17031 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
17032 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
17033 | } | |
17034 | ||
4976b718 HL |
17035 | /* find and rewrite pseudo imm in ld_imm64 instructions: |
17036 | * | |
17037 | * 1. if it accesses map FD, replace it with actual map pointer. | |
17038 | * 2. if it accesses btf_id of a VAR, replace it with pointer to the var. | |
17039 | * | |
17040 | * NOTE: btf_vmlinux is required for converting pseudo btf_id. | |
0246e64d | 17041 | */ |
4976b718 | 17042 | static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env) |
0246e64d AS |
17043 | { |
17044 | struct bpf_insn *insn = env->prog->insnsi; | |
17045 | int insn_cnt = env->prog->len; | |
fdc15d38 | 17046 | int i, j, err; |
0246e64d | 17047 | |
f1f7714e | 17048 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
17049 | if (err) |
17050 | return err; | |
17051 | ||
0246e64d | 17052 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 17053 | if (BPF_CLASS(insn->code) == BPF_LDX && |
1f9a1ea8 YS |
17054 | ((BPF_MODE(insn->code) != BPF_MEM && BPF_MODE(insn->code) != BPF_MEMSX) || |
17055 | insn->imm != 0)) { | |
61bd5218 | 17056 | verbose(env, "BPF_LDX uses reserved fields\n"); |
d691f9e8 AS |
17057 | return -EINVAL; |
17058 | } | |
17059 | ||
0246e64d | 17060 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 17061 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
17062 | struct bpf_map *map; |
17063 | struct fd f; | |
d8eca5bb | 17064 | u64 addr; |
387544bf | 17065 | u32 fd; |
0246e64d AS |
17066 | |
17067 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
17068 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
17069 | insn[1].off != 0) { | |
61bd5218 | 17070 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
17071 | return -EINVAL; |
17072 | } | |
17073 | ||
d8eca5bb | 17074 | if (insn[0].src_reg == 0) |
0246e64d AS |
17075 | /* valid generic load 64-bit imm */ |
17076 | goto next_insn; | |
17077 | ||
4976b718 HL |
17078 | if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) { |
17079 | aux = &env->insn_aux_data[i]; | |
17080 | err = check_pseudo_btf_id(env, insn, aux); | |
17081 | if (err) | |
17082 | return err; | |
17083 | goto next_insn; | |
17084 | } | |
17085 | ||
69c087ba YS |
17086 | if (insn[0].src_reg == BPF_PSEUDO_FUNC) { |
17087 | aux = &env->insn_aux_data[i]; | |
17088 | aux->ptr_type = PTR_TO_FUNC; | |
17089 | goto next_insn; | |
17090 | } | |
17091 | ||
d8eca5bb DB |
17092 | /* In final convert_pseudo_ld_imm64() step, this is |
17093 | * converted into regular 64-bit imm load insn. | |
17094 | */ | |
387544bf AS |
17095 | switch (insn[0].src_reg) { |
17096 | case BPF_PSEUDO_MAP_VALUE: | |
17097 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
17098 | break; | |
17099 | case BPF_PSEUDO_MAP_FD: | |
17100 | case BPF_PSEUDO_MAP_IDX: | |
17101 | if (insn[1].imm == 0) | |
17102 | break; | |
17103 | fallthrough; | |
17104 | default: | |
17105 | verbose(env, "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
17106 | return -EINVAL; |
17107 | } | |
17108 | ||
387544bf AS |
17109 | switch (insn[0].src_reg) { |
17110 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
17111 | case BPF_PSEUDO_MAP_IDX: | |
17112 | if (bpfptr_is_null(env->fd_array)) { | |
17113 | verbose(env, "fd_idx without fd_array is invalid\n"); | |
17114 | return -EPROTO; | |
17115 | } | |
17116 | if (copy_from_bpfptr_offset(&fd, env->fd_array, | |
17117 | insn[0].imm * sizeof(fd), | |
17118 | sizeof(fd))) | |
17119 | return -EFAULT; | |
17120 | break; | |
17121 | default: | |
17122 | fd = insn[0].imm; | |
17123 | break; | |
17124 | } | |
17125 | ||
17126 | f = fdget(fd); | |
c2101297 | 17127 | map = __bpf_map_get(f); |
0246e64d | 17128 | if (IS_ERR(map)) { |
61bd5218 | 17129 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 17130 | insn[0].imm); |
0246e64d AS |
17131 | return PTR_ERR(map); |
17132 | } | |
17133 | ||
61bd5218 | 17134 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
17135 | if (err) { |
17136 | fdput(f); | |
17137 | return err; | |
17138 | } | |
17139 | ||
d8eca5bb | 17140 | aux = &env->insn_aux_data[i]; |
387544bf AS |
17141 | if (insn[0].src_reg == BPF_PSEUDO_MAP_FD || |
17142 | insn[0].src_reg == BPF_PSEUDO_MAP_IDX) { | |
d8eca5bb DB |
17143 | addr = (unsigned long)map; |
17144 | } else { | |
17145 | u32 off = insn[1].imm; | |
17146 | ||
17147 | if (off >= BPF_MAX_VAR_OFF) { | |
17148 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
17149 | fdput(f); | |
17150 | return -EINVAL; | |
17151 | } | |
17152 | ||
17153 | if (!map->ops->map_direct_value_addr) { | |
17154 | verbose(env, "no direct value access support for this map type\n"); | |
17155 | fdput(f); | |
17156 | return -EINVAL; | |
17157 | } | |
17158 | ||
17159 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
17160 | if (err) { | |
17161 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
17162 | map->value_size, off); | |
17163 | fdput(f); | |
17164 | return err; | |
17165 | } | |
17166 | ||
17167 | aux->map_off = off; | |
17168 | addr += off; | |
17169 | } | |
17170 | ||
17171 | insn[0].imm = (u32)addr; | |
17172 | insn[1].imm = addr >> 32; | |
0246e64d AS |
17173 | |
17174 | /* check whether we recorded this map already */ | |
d8eca5bb | 17175 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 17176 | if (env->used_maps[j] == map) { |
d8eca5bb | 17177 | aux->map_index = j; |
0246e64d AS |
17178 | fdput(f); |
17179 | goto next_insn; | |
17180 | } | |
d8eca5bb | 17181 | } |
0246e64d AS |
17182 | |
17183 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
17184 | fdput(f); | |
17185 | return -E2BIG; | |
17186 | } | |
17187 | ||
0246e64d AS |
17188 | /* hold the map. If the program is rejected by verifier, |
17189 | * the map will be released by release_maps() or it | |
17190 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 17191 | * and all maps are released in free_used_maps() |
0246e64d | 17192 | */ |
1e0bd5a0 | 17193 | bpf_map_inc(map); |
d8eca5bb DB |
17194 | |
17195 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
17196 | env->used_maps[env->used_map_cnt++] = map; |
17197 | ||
b741f163 | 17198 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 17199 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 17200 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
17201 | fdput(f); |
17202 | return -EBUSY; | |
17203 | } | |
17204 | ||
0246e64d AS |
17205 | fdput(f); |
17206 | next_insn: | |
17207 | insn++; | |
17208 | i++; | |
5e581dad DB |
17209 | continue; |
17210 | } | |
17211 | ||
17212 | /* Basic sanity check before we invest more work here. */ | |
17213 | if (!bpf_opcode_in_insntable(insn->code)) { | |
17214 | verbose(env, "unknown opcode %02x\n", insn->code); | |
17215 | return -EINVAL; | |
0246e64d AS |
17216 | } |
17217 | } | |
17218 | ||
17219 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
17220 | * 'struct bpf_map *' into a register instead of user map_fd. | |
17221 | * These pointers will be used later by verifier to validate map access. | |
17222 | */ | |
17223 | return 0; | |
17224 | } | |
17225 | ||
17226 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 17227 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 17228 | { |
a2ea0746 DB |
17229 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
17230 | env->used_map_cnt); | |
0246e64d AS |
17231 | } |
17232 | ||
541c3bad AN |
17233 | /* drop refcnt of maps used by the rejected program */ |
17234 | static void release_btfs(struct bpf_verifier_env *env) | |
17235 | { | |
17236 | __bpf_free_used_btfs(env->prog->aux, env->used_btfs, | |
17237 | env->used_btf_cnt); | |
17238 | } | |
17239 | ||
0246e64d | 17240 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ |
58e2af8b | 17241 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
17242 | { |
17243 | struct bpf_insn *insn = env->prog->insnsi; | |
17244 | int insn_cnt = env->prog->len; | |
17245 | int i; | |
17246 | ||
69c087ba YS |
17247 | for (i = 0; i < insn_cnt; i++, insn++) { |
17248 | if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) | |
17249 | continue; | |
17250 | if (insn->src_reg == BPF_PSEUDO_FUNC) | |
17251 | continue; | |
17252 | insn->src_reg = 0; | |
17253 | } | |
0246e64d AS |
17254 | } |
17255 | ||
8041902d AS |
17256 | /* single env->prog->insni[off] instruction was replaced with the range |
17257 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
17258 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
17259 | */ | |
75f0fc7b HF |
17260 | static void adjust_insn_aux_data(struct bpf_verifier_env *env, |
17261 | struct bpf_insn_aux_data *new_data, | |
17262 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d | 17263 | { |
75f0fc7b | 17264 | struct bpf_insn_aux_data *old_data = env->insn_aux_data; |
b325fbca | 17265 | struct bpf_insn *insn = new_prog->insnsi; |
d203b0fd | 17266 | u32 old_seen = old_data[off].seen; |
b325fbca | 17267 | u32 prog_len; |
c131187d | 17268 | int i; |
8041902d | 17269 | |
b325fbca JW |
17270 | /* aux info at OFF always needs adjustment, no matter fast path |
17271 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
17272 | * original insn at old prog. | |
17273 | */ | |
17274 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
17275 | ||
8041902d | 17276 | if (cnt == 1) |
75f0fc7b | 17277 | return; |
b325fbca | 17278 | prog_len = new_prog->len; |
75f0fc7b | 17279 | |
8041902d AS |
17280 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); |
17281 | memcpy(new_data + off + cnt - 1, old_data + off, | |
17282 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 17283 | for (i = off; i < off + cnt - 1; i++) { |
d203b0fd DB |
17284 | /* Expand insni[off]'s seen count to the patched range. */ |
17285 | new_data[i].seen = old_seen; | |
b325fbca JW |
17286 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
17287 | } | |
8041902d AS |
17288 | env->insn_aux_data = new_data; |
17289 | vfree(old_data); | |
8041902d AS |
17290 | } |
17291 | ||
cc8b0b92 AS |
17292 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
17293 | { | |
17294 | int i; | |
17295 | ||
17296 | if (len == 1) | |
17297 | return; | |
4cb3d99c JW |
17298 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
17299 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 17300 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 17301 | continue; |
9c8105bd | 17302 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
17303 | } |
17304 | } | |
17305 | ||
7506d211 | 17306 | static void adjust_poke_descs(struct bpf_prog *prog, u32 off, u32 len) |
a748c697 MF |
17307 | { |
17308 | struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab; | |
17309 | int i, sz = prog->aux->size_poke_tab; | |
17310 | struct bpf_jit_poke_descriptor *desc; | |
17311 | ||
17312 | for (i = 0; i < sz; i++) { | |
17313 | desc = &tab[i]; | |
7506d211 JF |
17314 | if (desc->insn_idx <= off) |
17315 | continue; | |
a748c697 MF |
17316 | desc->insn_idx += len - 1; |
17317 | } | |
17318 | } | |
17319 | ||
8041902d AS |
17320 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
17321 | const struct bpf_insn *patch, u32 len) | |
17322 | { | |
17323 | struct bpf_prog *new_prog; | |
75f0fc7b HF |
17324 | struct bpf_insn_aux_data *new_data = NULL; |
17325 | ||
17326 | if (len > 1) { | |
17327 | new_data = vzalloc(array_size(env->prog->len + len - 1, | |
17328 | sizeof(struct bpf_insn_aux_data))); | |
17329 | if (!new_data) | |
17330 | return NULL; | |
17331 | } | |
8041902d AS |
17332 | |
17333 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
17334 | if (IS_ERR(new_prog)) { |
17335 | if (PTR_ERR(new_prog) == -ERANGE) | |
17336 | verbose(env, | |
17337 | "insn %d cannot be patched due to 16-bit range\n", | |
17338 | env->insn_aux_data[off].orig_idx); | |
75f0fc7b | 17339 | vfree(new_data); |
8041902d | 17340 | return NULL; |
4f73379e | 17341 | } |
75f0fc7b | 17342 | adjust_insn_aux_data(env, new_data, new_prog, off, len); |
cc8b0b92 | 17343 | adjust_subprog_starts(env, off, len); |
7506d211 | 17344 | adjust_poke_descs(new_prog, off, len); |
8041902d AS |
17345 | return new_prog; |
17346 | } | |
17347 | ||
52875a04 JK |
17348 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
17349 | u32 off, u32 cnt) | |
17350 | { | |
17351 | int i, j; | |
17352 | ||
17353 | /* find first prog starting at or after off (first to remove) */ | |
17354 | for (i = 0; i < env->subprog_cnt; i++) | |
17355 | if (env->subprog_info[i].start >= off) | |
17356 | break; | |
17357 | /* find first prog starting at or after off + cnt (first to stay) */ | |
17358 | for (j = i; j < env->subprog_cnt; j++) | |
17359 | if (env->subprog_info[j].start >= off + cnt) | |
17360 | break; | |
17361 | /* if j doesn't start exactly at off + cnt, we are just removing | |
17362 | * the front of previous prog | |
17363 | */ | |
17364 | if (env->subprog_info[j].start != off + cnt) | |
17365 | j--; | |
17366 | ||
17367 | if (j > i) { | |
17368 | struct bpf_prog_aux *aux = env->prog->aux; | |
17369 | int move; | |
17370 | ||
17371 | /* move fake 'exit' subprog as well */ | |
17372 | move = env->subprog_cnt + 1 - j; | |
17373 | ||
17374 | memmove(env->subprog_info + i, | |
17375 | env->subprog_info + j, | |
17376 | sizeof(*env->subprog_info) * move); | |
17377 | env->subprog_cnt -= j - i; | |
17378 | ||
17379 | /* remove func_info */ | |
17380 | if (aux->func_info) { | |
17381 | move = aux->func_info_cnt - j; | |
17382 | ||
17383 | memmove(aux->func_info + i, | |
17384 | aux->func_info + j, | |
17385 | sizeof(*aux->func_info) * move); | |
17386 | aux->func_info_cnt -= j - i; | |
17387 | /* func_info->insn_off is set after all code rewrites, | |
17388 | * in adjust_btf_func() - no need to adjust | |
17389 | */ | |
17390 | } | |
17391 | } else { | |
17392 | /* convert i from "first prog to remove" to "first to adjust" */ | |
17393 | if (env->subprog_info[i].start == off) | |
17394 | i++; | |
17395 | } | |
17396 | ||
17397 | /* update fake 'exit' subprog as well */ | |
17398 | for (; i <= env->subprog_cnt; i++) | |
17399 | env->subprog_info[i].start -= cnt; | |
17400 | ||
17401 | return 0; | |
17402 | } | |
17403 | ||
17404 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
17405 | u32 cnt) | |
17406 | { | |
17407 | struct bpf_prog *prog = env->prog; | |
17408 | u32 i, l_off, l_cnt, nr_linfo; | |
17409 | struct bpf_line_info *linfo; | |
17410 | ||
17411 | nr_linfo = prog->aux->nr_linfo; | |
17412 | if (!nr_linfo) | |
17413 | return 0; | |
17414 | ||
17415 | linfo = prog->aux->linfo; | |
17416 | ||
17417 | /* find first line info to remove, count lines to be removed */ | |
17418 | for (i = 0; i < nr_linfo; i++) | |
17419 | if (linfo[i].insn_off >= off) | |
17420 | break; | |
17421 | ||
17422 | l_off = i; | |
17423 | l_cnt = 0; | |
17424 | for (; i < nr_linfo; i++) | |
17425 | if (linfo[i].insn_off < off + cnt) | |
17426 | l_cnt++; | |
17427 | else | |
17428 | break; | |
17429 | ||
17430 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
17431 | * last removed linfo. prog is already modified, so prog->len == off | |
17432 | * means no live instructions after (tail of the program was removed). | |
17433 | */ | |
17434 | if (prog->len != off && l_cnt && | |
17435 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
17436 | l_cnt--; | |
17437 | linfo[--i].insn_off = off + cnt; | |
17438 | } | |
17439 | ||
17440 | /* remove the line info which refer to the removed instructions */ | |
17441 | if (l_cnt) { | |
17442 | memmove(linfo + l_off, linfo + i, | |
17443 | sizeof(*linfo) * (nr_linfo - i)); | |
17444 | ||
17445 | prog->aux->nr_linfo -= l_cnt; | |
17446 | nr_linfo = prog->aux->nr_linfo; | |
17447 | } | |
17448 | ||
17449 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
17450 | for (i = l_off; i < nr_linfo; i++) | |
17451 | linfo[i].insn_off -= cnt; | |
17452 | ||
17453 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
17454 | for (i = 0; i <= env->subprog_cnt; i++) | |
17455 | if (env->subprog_info[i].linfo_idx > l_off) { | |
17456 | /* program may have started in the removed region but | |
17457 | * may not be fully removed | |
17458 | */ | |
17459 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
17460 | env->subprog_info[i].linfo_idx -= l_cnt; | |
17461 | else | |
17462 | env->subprog_info[i].linfo_idx = l_off; | |
17463 | } | |
17464 | ||
17465 | return 0; | |
17466 | } | |
17467 | ||
17468 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
17469 | { | |
17470 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
17471 | unsigned int orig_prog_len = env->prog->len; | |
17472 | int err; | |
17473 | ||
9d03ebc7 | 17474 | if (bpf_prog_is_offloaded(env->prog->aux)) |
08ca90af JK |
17475 | bpf_prog_offload_remove_insns(env, off, cnt); |
17476 | ||
52875a04 JK |
17477 | err = bpf_remove_insns(env->prog, off, cnt); |
17478 | if (err) | |
17479 | return err; | |
17480 | ||
17481 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
17482 | if (err) | |
17483 | return err; | |
17484 | ||
17485 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
17486 | if (err) | |
17487 | return err; | |
17488 | ||
17489 | memmove(aux_data + off, aux_data + off + cnt, | |
17490 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
17491 | ||
17492 | return 0; | |
17493 | } | |
17494 | ||
2a5418a1 DB |
17495 | /* The verifier does more data flow analysis than llvm and will not |
17496 | * explore branches that are dead at run time. Malicious programs can | |
17497 | * have dead code too. Therefore replace all dead at-run-time code | |
17498 | * with 'ja -1'. | |
17499 | * | |
17500 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
17501 | * program and through another bug we would manage to jump there, then | |
17502 | * we'd execute beyond program memory otherwise. Returning exception | |
17503 | * code also wouldn't work since we can have subprogs where the dead | |
17504 | * code could be located. | |
c131187d AS |
17505 | */ |
17506 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
17507 | { | |
17508 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 17509 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
17510 | struct bpf_insn *insn = env->prog->insnsi; |
17511 | const int insn_cnt = env->prog->len; | |
17512 | int i; | |
17513 | ||
17514 | for (i = 0; i < insn_cnt; i++) { | |
17515 | if (aux_data[i].seen) | |
17516 | continue; | |
2a5418a1 | 17517 | memcpy(insn + i, &trap, sizeof(trap)); |
45c709f8 | 17518 | aux_data[i].zext_dst = false; |
c131187d AS |
17519 | } |
17520 | } | |
17521 | ||
e2ae4ca2 JK |
17522 | static bool insn_is_cond_jump(u8 code) |
17523 | { | |
17524 | u8 op; | |
17525 | ||
4cd58e9a | 17526 | op = BPF_OP(code); |
092ed096 | 17527 | if (BPF_CLASS(code) == BPF_JMP32) |
4cd58e9a | 17528 | return op != BPF_JA; |
092ed096 | 17529 | |
e2ae4ca2 JK |
17530 | if (BPF_CLASS(code) != BPF_JMP) |
17531 | return false; | |
17532 | ||
e2ae4ca2 JK |
17533 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; |
17534 | } | |
17535 | ||
17536 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
17537 | { | |
17538 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
17539 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
17540 | struct bpf_insn *insn = env->prog->insnsi; | |
17541 | const int insn_cnt = env->prog->len; | |
17542 | int i; | |
17543 | ||
17544 | for (i = 0; i < insn_cnt; i++, insn++) { | |
17545 | if (!insn_is_cond_jump(insn->code)) | |
17546 | continue; | |
17547 | ||
17548 | if (!aux_data[i + 1].seen) | |
17549 | ja.off = insn->off; | |
17550 | else if (!aux_data[i + 1 + insn->off].seen) | |
17551 | ja.off = 0; | |
17552 | else | |
17553 | continue; | |
17554 | ||
9d03ebc7 | 17555 | if (bpf_prog_is_offloaded(env->prog->aux)) |
08ca90af JK |
17556 | bpf_prog_offload_replace_insn(env, i, &ja); |
17557 | ||
e2ae4ca2 JK |
17558 | memcpy(insn, &ja, sizeof(ja)); |
17559 | } | |
17560 | } | |
17561 | ||
52875a04 JK |
17562 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
17563 | { | |
17564 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
17565 | int insn_cnt = env->prog->len; | |
17566 | int i, err; | |
17567 | ||
17568 | for (i = 0; i < insn_cnt; i++) { | |
17569 | int j; | |
17570 | ||
17571 | j = 0; | |
17572 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
17573 | j++; | |
17574 | if (!j) | |
17575 | continue; | |
17576 | ||
17577 | err = verifier_remove_insns(env, i, j); | |
17578 | if (err) | |
17579 | return err; | |
17580 | insn_cnt = env->prog->len; | |
17581 | } | |
17582 | ||
17583 | return 0; | |
17584 | } | |
17585 | ||
a1b14abc JK |
17586 | static int opt_remove_nops(struct bpf_verifier_env *env) |
17587 | { | |
17588 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
17589 | struct bpf_insn *insn = env->prog->insnsi; | |
17590 | int insn_cnt = env->prog->len; | |
17591 | int i, err; | |
17592 | ||
17593 | for (i = 0; i < insn_cnt; i++) { | |
17594 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
17595 | continue; | |
17596 | ||
17597 | err = verifier_remove_insns(env, i, 1); | |
17598 | if (err) | |
17599 | return err; | |
17600 | insn_cnt--; | |
17601 | i--; | |
17602 | } | |
17603 | ||
17604 | return 0; | |
17605 | } | |
17606 | ||
d6c2308c JW |
17607 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
17608 | const union bpf_attr *attr) | |
a4b1d3c1 | 17609 | { |
d6c2308c | 17610 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 17611 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 17612 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 17613 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 17614 | struct bpf_prog *new_prog; |
d6c2308c | 17615 | bool rnd_hi32; |
a4b1d3c1 | 17616 | |
d6c2308c | 17617 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 17618 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
17619 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
17620 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
17621 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
17622 | for (i = 0; i < len; i++) { |
17623 | int adj_idx = i + delta; | |
17624 | struct bpf_insn insn; | |
83a28819 | 17625 | int load_reg; |
a4b1d3c1 | 17626 | |
d6c2308c | 17627 | insn = insns[adj_idx]; |
83a28819 | 17628 | load_reg = insn_def_regno(&insn); |
d6c2308c JW |
17629 | if (!aux[adj_idx].zext_dst) { |
17630 | u8 code, class; | |
17631 | u32 imm_rnd; | |
17632 | ||
17633 | if (!rnd_hi32) | |
17634 | continue; | |
17635 | ||
17636 | code = insn.code; | |
17637 | class = BPF_CLASS(code); | |
83a28819 | 17638 | if (load_reg == -1) |
d6c2308c JW |
17639 | continue; |
17640 | ||
17641 | /* NOTE: arg "reg" (the fourth one) is only used for | |
83a28819 IL |
17642 | * BPF_STX + SRC_OP, so it is safe to pass NULL |
17643 | * here. | |
d6c2308c | 17644 | */ |
83a28819 | 17645 | if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) { |
d6c2308c JW |
17646 | if (class == BPF_LD && |
17647 | BPF_MODE(code) == BPF_IMM) | |
17648 | i++; | |
17649 | continue; | |
17650 | } | |
17651 | ||
17652 | /* ctx load could be transformed into wider load. */ | |
17653 | if (class == BPF_LDX && | |
17654 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
17655 | continue; | |
17656 | ||
a251c17a | 17657 | imm_rnd = get_random_u32(); |
d6c2308c JW |
17658 | rnd_hi32_patch[0] = insn; |
17659 | rnd_hi32_patch[1].imm = imm_rnd; | |
83a28819 | 17660 | rnd_hi32_patch[3].dst_reg = load_reg; |
d6c2308c JW |
17661 | patch = rnd_hi32_patch; |
17662 | patch_len = 4; | |
17663 | goto apply_patch_buffer; | |
17664 | } | |
17665 | ||
39491867 BJ |
17666 | /* Add in an zero-extend instruction if a) the JIT has requested |
17667 | * it or b) it's a CMPXCHG. | |
17668 | * | |
17669 | * The latter is because: BPF_CMPXCHG always loads a value into | |
17670 | * R0, therefore always zero-extends. However some archs' | |
17671 | * equivalent instruction only does this load when the | |
17672 | * comparison is successful. This detail of CMPXCHG is | |
17673 | * orthogonal to the general zero-extension behaviour of the | |
17674 | * CPU, so it's treated independently of bpf_jit_needs_zext. | |
17675 | */ | |
17676 | if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn)) | |
a4b1d3c1 JW |
17677 | continue; |
17678 | ||
d35af0a7 BT |
17679 | /* Zero-extension is done by the caller. */ |
17680 | if (bpf_pseudo_kfunc_call(&insn)) | |
17681 | continue; | |
17682 | ||
83a28819 IL |
17683 | if (WARN_ON(load_reg == -1)) { |
17684 | verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n"); | |
17685 | return -EFAULT; | |
b2e37a71 IL |
17686 | } |
17687 | ||
a4b1d3c1 | 17688 | zext_patch[0] = insn; |
b2e37a71 IL |
17689 | zext_patch[1].dst_reg = load_reg; |
17690 | zext_patch[1].src_reg = load_reg; | |
d6c2308c JW |
17691 | patch = zext_patch; |
17692 | patch_len = 2; | |
17693 | apply_patch_buffer: | |
17694 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
17695 | if (!new_prog) |
17696 | return -ENOMEM; | |
17697 | env->prog = new_prog; | |
17698 | insns = new_prog->insnsi; | |
17699 | aux = env->insn_aux_data; | |
d6c2308c | 17700 | delta += patch_len - 1; |
a4b1d3c1 JW |
17701 | } |
17702 | ||
17703 | return 0; | |
17704 | } | |
17705 | ||
c64b7983 JS |
17706 | /* convert load instructions that access fields of a context type into a |
17707 | * sequence of instructions that access fields of the underlying structure: | |
17708 | * struct __sk_buff -> struct sk_buff | |
17709 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 17710 | */ |
58e2af8b | 17711 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 17712 | { |
00176a34 | 17713 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 17714 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 17715 | const int insn_cnt = env->prog->len; |
36bbef52 | 17716 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 17717 | u32 target_size, size_default, off; |
9bac3d6d | 17718 | struct bpf_prog *new_prog; |
d691f9e8 | 17719 | enum bpf_access_type type; |
f96da094 | 17720 | bool is_narrower_load; |
9bac3d6d | 17721 | |
b09928b9 DB |
17722 | if (ops->gen_prologue || env->seen_direct_write) { |
17723 | if (!ops->gen_prologue) { | |
17724 | verbose(env, "bpf verifier is misconfigured\n"); | |
17725 | return -EINVAL; | |
17726 | } | |
36bbef52 DB |
17727 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
17728 | env->prog); | |
17729 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 17730 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
17731 | return -EINVAL; |
17732 | } else if (cnt) { | |
8041902d | 17733 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
17734 | if (!new_prog) |
17735 | return -ENOMEM; | |
8041902d | 17736 | |
36bbef52 | 17737 | env->prog = new_prog; |
3df126f3 | 17738 | delta += cnt - 1; |
36bbef52 DB |
17739 | } |
17740 | } | |
17741 | ||
9d03ebc7 | 17742 | if (bpf_prog_is_offloaded(env->prog->aux)) |
9bac3d6d AS |
17743 | return 0; |
17744 | ||
3df126f3 | 17745 | insn = env->prog->insnsi + delta; |
36bbef52 | 17746 | |
9bac3d6d | 17747 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 | 17748 | bpf_convert_ctx_access_t convert_ctx_access; |
1f1e864b | 17749 | u8 mode; |
c64b7983 | 17750 | |
62c7989b DB |
17751 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
17752 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
17753 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
1f9a1ea8 YS |
17754 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW) || |
17755 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_B) || | |
17756 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_H) || | |
17757 | insn->code == (BPF_LDX | BPF_MEMSX | BPF_W)) { | |
d691f9e8 | 17758 | type = BPF_READ; |
2039f26f DB |
17759 | } else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
17760 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
17761 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
17762 | insn->code == (BPF_STX | BPF_MEM | BPF_DW) || | |
17763 | insn->code == (BPF_ST | BPF_MEM | BPF_B) || | |
17764 | insn->code == (BPF_ST | BPF_MEM | BPF_H) || | |
17765 | insn->code == (BPF_ST | BPF_MEM | BPF_W) || | |
17766 | insn->code == (BPF_ST | BPF_MEM | BPF_DW)) { | |
d691f9e8 | 17767 | type = BPF_WRITE; |
2039f26f | 17768 | } else { |
9bac3d6d | 17769 | continue; |
2039f26f | 17770 | } |
9bac3d6d | 17771 | |
af86ca4e | 17772 | if (type == BPF_WRITE && |
2039f26f | 17773 | env->insn_aux_data[i + delta].sanitize_stack_spill) { |
af86ca4e | 17774 | struct bpf_insn patch[] = { |
af86ca4e | 17775 | *insn, |
2039f26f | 17776 | BPF_ST_NOSPEC(), |
af86ca4e AS |
17777 | }; |
17778 | ||
17779 | cnt = ARRAY_SIZE(patch); | |
17780 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
17781 | if (!new_prog) | |
17782 | return -ENOMEM; | |
17783 | ||
17784 | delta += cnt - 1; | |
17785 | env->prog = new_prog; | |
17786 | insn = new_prog->insnsi + i + delta; | |
17787 | continue; | |
17788 | } | |
17789 | ||
6efe152d | 17790 | switch ((int)env->insn_aux_data[i + delta].ptr_type) { |
c64b7983 JS |
17791 | case PTR_TO_CTX: |
17792 | if (!ops->convert_ctx_access) | |
17793 | continue; | |
17794 | convert_ctx_access = ops->convert_ctx_access; | |
17795 | break; | |
17796 | case PTR_TO_SOCKET: | |
46f8bc92 | 17797 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
17798 | convert_ctx_access = bpf_sock_convert_ctx_access; |
17799 | break; | |
655a51e5 MKL |
17800 | case PTR_TO_TCP_SOCK: |
17801 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
17802 | break; | |
fada7fdc JL |
17803 | case PTR_TO_XDP_SOCK: |
17804 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
17805 | break; | |
2a02759e | 17806 | case PTR_TO_BTF_ID: |
6efe152d | 17807 | case PTR_TO_BTF_ID | PTR_UNTRUSTED: |
282de143 KKD |
17808 | /* PTR_TO_BTF_ID | MEM_ALLOC always has a valid lifetime, unlike |
17809 | * PTR_TO_BTF_ID, and an active ref_obj_id, but the same cannot | |
17810 | * be said once it is marked PTR_UNTRUSTED, hence we must handle | |
17811 | * any faults for loads into such types. BPF_WRITE is disallowed | |
17812 | * for this case. | |
17813 | */ | |
17814 | case PTR_TO_BTF_ID | MEM_ALLOC | PTR_UNTRUSTED: | |
27ae7997 | 17815 | if (type == BPF_READ) { |
1f9a1ea8 YS |
17816 | if (BPF_MODE(insn->code) == BPF_MEM) |
17817 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
17818 | BPF_SIZE((insn)->code); | |
17819 | else | |
17820 | insn->code = BPF_LDX | BPF_PROBE_MEMSX | | |
17821 | BPF_SIZE((insn)->code); | |
27ae7997 | 17822 | env->prog->aux->num_exentries++; |
2a02759e | 17823 | } |
2a02759e | 17824 | continue; |
c64b7983 | 17825 | default: |
9bac3d6d | 17826 | continue; |
c64b7983 | 17827 | } |
9bac3d6d | 17828 | |
31fd8581 | 17829 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 17830 | size = BPF_LDST_BYTES(insn); |
1f1e864b | 17831 | mode = BPF_MODE(insn->code); |
31fd8581 YS |
17832 | |
17833 | /* If the read access is a narrower load of the field, | |
17834 | * convert to a 4/8-byte load, to minimum program type specific | |
17835 | * convert_ctx_access changes. If conversion is successful, | |
17836 | * we will apply proper mask to the result. | |
17837 | */ | |
f96da094 | 17838 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
17839 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
17840 | off = insn->off; | |
31fd8581 | 17841 | if (is_narrower_load) { |
f96da094 DB |
17842 | u8 size_code; |
17843 | ||
17844 | if (type == BPF_WRITE) { | |
61bd5218 | 17845 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
17846 | return -EINVAL; |
17847 | } | |
31fd8581 | 17848 | |
f96da094 | 17849 | size_code = BPF_H; |
31fd8581 YS |
17850 | if (ctx_field_size == 4) |
17851 | size_code = BPF_W; | |
17852 | else if (ctx_field_size == 8) | |
17853 | size_code = BPF_DW; | |
f96da094 | 17854 | |
bc23105c | 17855 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
17856 | insn->code = BPF_LDX | BPF_MEM | size_code; |
17857 | } | |
f96da094 DB |
17858 | |
17859 | target_size = 0; | |
c64b7983 JS |
17860 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
17861 | &target_size); | |
f96da094 DB |
17862 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
17863 | (ctx_field_size && !target_size)) { | |
61bd5218 | 17864 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
17865 | return -EINVAL; |
17866 | } | |
f96da094 DB |
17867 | |
17868 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
17869 | u8 shift = bpf_ctx_narrow_access_offset( |
17870 | off, size, size_default) * 8; | |
d7af7e49 AI |
17871 | if (shift && cnt + 1 >= ARRAY_SIZE(insn_buf)) { |
17872 | verbose(env, "bpf verifier narrow ctx load misconfigured\n"); | |
17873 | return -EINVAL; | |
17874 | } | |
46f53a65 AI |
17875 | if (ctx_field_size <= 4) { |
17876 | if (shift) | |
17877 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
17878 | insn->dst_reg, | |
17879 | shift); | |
31fd8581 | 17880 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 17881 | (1 << size * 8) - 1); |
46f53a65 AI |
17882 | } else { |
17883 | if (shift) | |
17884 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
17885 | insn->dst_reg, | |
17886 | shift); | |
0613d8ca | 17887 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 17888 | (1ULL << size * 8) - 1); |
46f53a65 | 17889 | } |
31fd8581 | 17890 | } |
1f1e864b YS |
17891 | if (mode == BPF_MEMSX) |
17892 | insn_buf[cnt++] = BPF_RAW_INSN(BPF_ALU64 | BPF_MOV | BPF_X, | |
17893 | insn->dst_reg, insn->dst_reg, | |
17894 | size * 8, 0); | |
9bac3d6d | 17895 | |
8041902d | 17896 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
17897 | if (!new_prog) |
17898 | return -ENOMEM; | |
17899 | ||
3df126f3 | 17900 | delta += cnt - 1; |
9bac3d6d AS |
17901 | |
17902 | /* keep walking new program and skip insns we just inserted */ | |
17903 | env->prog = new_prog; | |
3df126f3 | 17904 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
17905 | } |
17906 | ||
17907 | return 0; | |
17908 | } | |
17909 | ||
1c2a088a AS |
17910 | static int jit_subprogs(struct bpf_verifier_env *env) |
17911 | { | |
17912 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
17913 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
a748c697 | 17914 | struct bpf_map *map_ptr; |
7105e828 | 17915 | struct bpf_insn *insn; |
1c2a088a | 17916 | void *old_bpf_func; |
c4c0bdc0 | 17917 | int err, num_exentries; |
1c2a088a | 17918 | |
f910cefa | 17919 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
17920 | return 0; |
17921 | ||
7105e828 | 17922 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
3990ed4c | 17923 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn)) |
69c087ba | 17924 | continue; |
69c087ba | 17925 | |
c7a89784 DB |
17926 | /* Upon error here we cannot fall back to interpreter but |
17927 | * need a hard reject of the program. Thus -EFAULT is | |
17928 | * propagated in any case. | |
17929 | */ | |
1c2a088a AS |
17930 | subprog = find_subprog(env, i + insn->imm + 1); |
17931 | if (subprog < 0) { | |
17932 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
17933 | i + insn->imm + 1); | |
17934 | return -EFAULT; | |
17935 | } | |
17936 | /* temporarily remember subprog id inside insn instead of | |
17937 | * aux_data, since next loop will split up all insns into funcs | |
17938 | */ | |
f910cefa | 17939 | insn->off = subprog; |
1c2a088a AS |
17940 | /* remember original imm in case JIT fails and fallback |
17941 | * to interpreter will be needed | |
17942 | */ | |
17943 | env->insn_aux_data[i].call_imm = insn->imm; | |
17944 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
17945 | insn->imm = 1; | |
3990ed4c MKL |
17946 | if (bpf_pseudo_func(insn)) |
17947 | /* jit (e.g. x86_64) may emit fewer instructions | |
17948 | * if it learns a u32 imm is the same as a u64 imm. | |
17949 | * Force a non zero here. | |
17950 | */ | |
17951 | insn[1].imm = 1; | |
1c2a088a AS |
17952 | } |
17953 | ||
c454a46b MKL |
17954 | err = bpf_prog_alloc_jited_linfo(prog); |
17955 | if (err) | |
17956 | goto out_undo_insn; | |
17957 | ||
17958 | err = -ENOMEM; | |
6396bb22 | 17959 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 17960 | if (!func) |
c7a89784 | 17961 | goto out_undo_insn; |
1c2a088a | 17962 | |
f910cefa | 17963 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 17964 | subprog_start = subprog_end; |
4cb3d99c | 17965 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
17966 | |
17967 | len = subprog_end - subprog_start; | |
fb7dd8bc | 17968 | /* bpf_prog_run() doesn't call subprogs directly, |
492ecee8 AS |
17969 | * hence main prog stats include the runtime of subprogs. |
17970 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
700d4796 | 17971 | * func[i]->stats will never be accessed and stays NULL |
492ecee8 AS |
17972 | */ |
17973 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
17974 | if (!func[i]) |
17975 | goto out_free; | |
17976 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
17977 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 17978 | func[i]->type = prog->type; |
1c2a088a | 17979 | func[i]->len = len; |
4f74d809 DB |
17980 | if (bpf_prog_calc_tag(func[i])) |
17981 | goto out_free; | |
1c2a088a | 17982 | func[i]->is_func = 1; |
ba64e7d8 | 17983 | func[i]->aux->func_idx = i; |
f263a814 | 17984 | /* Below members will be freed only at prog->aux */ |
ba64e7d8 YS |
17985 | func[i]->aux->btf = prog->aux->btf; |
17986 | func[i]->aux->func_info = prog->aux->func_info; | |
9c7c48d6 | 17987 | func[i]->aux->func_info_cnt = prog->aux->func_info_cnt; |
f263a814 JF |
17988 | func[i]->aux->poke_tab = prog->aux->poke_tab; |
17989 | func[i]->aux->size_poke_tab = prog->aux->size_poke_tab; | |
ba64e7d8 | 17990 | |
a748c697 | 17991 | for (j = 0; j < prog->aux->size_poke_tab; j++) { |
f263a814 | 17992 | struct bpf_jit_poke_descriptor *poke; |
a748c697 | 17993 | |
f263a814 JF |
17994 | poke = &prog->aux->poke_tab[j]; |
17995 | if (poke->insn_idx < subprog_end && | |
17996 | poke->insn_idx >= subprog_start) | |
17997 | poke->aux = func[i]->aux; | |
a748c697 MF |
17998 | } |
17999 | ||
1c2a088a | 18000 | func[i]->aux->name[0] = 'F'; |
9c8105bd | 18001 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 18002 | func[i]->jit_requested = 1; |
d2a3b7c5 | 18003 | func[i]->blinding_requested = prog->blinding_requested; |
e6ac2450 | 18004 | func[i]->aux->kfunc_tab = prog->aux->kfunc_tab; |
2357672c | 18005 | func[i]->aux->kfunc_btf_tab = prog->aux->kfunc_btf_tab; |
c454a46b MKL |
18006 | func[i]->aux->linfo = prog->aux->linfo; |
18007 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
18008 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
18009 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
c4c0bdc0 YS |
18010 | num_exentries = 0; |
18011 | insn = func[i]->insnsi; | |
18012 | for (j = 0; j < func[i]->len; j++, insn++) { | |
18013 | if (BPF_CLASS(insn->code) == BPF_LDX && | |
1f9a1ea8 YS |
18014 | (BPF_MODE(insn->code) == BPF_PROBE_MEM || |
18015 | BPF_MODE(insn->code) == BPF_PROBE_MEMSX)) | |
c4c0bdc0 YS |
18016 | num_exentries++; |
18017 | } | |
18018 | func[i]->aux->num_exentries = num_exentries; | |
ebf7d1f5 | 18019 | func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable; |
1c2a088a AS |
18020 | func[i] = bpf_int_jit_compile(func[i]); |
18021 | if (!func[i]->jited) { | |
18022 | err = -ENOTSUPP; | |
18023 | goto out_free; | |
18024 | } | |
18025 | cond_resched(); | |
18026 | } | |
a748c697 | 18027 | |
1c2a088a AS |
18028 | /* at this point all bpf functions were successfully JITed |
18029 | * now populate all bpf_calls with correct addresses and | |
18030 | * run last pass of JIT | |
18031 | */ | |
f910cefa | 18032 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
18033 | insn = func[i]->insnsi; |
18034 | for (j = 0; j < func[i]->len; j++, insn++) { | |
69c087ba | 18035 | if (bpf_pseudo_func(insn)) { |
3990ed4c | 18036 | subprog = insn->off; |
69c087ba YS |
18037 | insn[0].imm = (u32)(long)func[subprog]->bpf_func; |
18038 | insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32; | |
18039 | continue; | |
18040 | } | |
23a2d70c | 18041 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
18042 | continue; |
18043 | subprog = insn->off; | |
3d717fad | 18044 | insn->imm = BPF_CALL_IMM(func[subprog]->bpf_func); |
1c2a088a | 18045 | } |
2162fed4 SD |
18046 | |
18047 | /* we use the aux data to keep a list of the start addresses | |
18048 | * of the JITed images for each function in the program | |
18049 | * | |
18050 | * for some architectures, such as powerpc64, the imm field | |
18051 | * might not be large enough to hold the offset of the start | |
18052 | * address of the callee's JITed image from __bpf_call_base | |
18053 | * | |
18054 | * in such cases, we can lookup the start address of a callee | |
18055 | * by using its subprog id, available from the off field of | |
18056 | * the call instruction, as an index for this list | |
18057 | */ | |
18058 | func[i]->aux->func = func; | |
18059 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 18060 | } |
f910cefa | 18061 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
18062 | old_bpf_func = func[i]->bpf_func; |
18063 | tmp = bpf_int_jit_compile(func[i]); | |
18064 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
18065 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 18066 | err = -ENOTSUPP; |
1c2a088a AS |
18067 | goto out_free; |
18068 | } | |
18069 | cond_resched(); | |
18070 | } | |
18071 | ||
18072 | /* finally lock prog and jit images for all functions and | |
0108a4e9 KJ |
18073 | * populate kallsysm. Begin at the first subprogram, since |
18074 | * bpf_prog_load will add the kallsyms for the main program. | |
1c2a088a | 18075 | */ |
0108a4e9 | 18076 | for (i = 1; i < env->subprog_cnt; i++) { |
1c2a088a AS |
18077 | bpf_prog_lock_ro(func[i]); |
18078 | bpf_prog_kallsyms_add(func[i]); | |
18079 | } | |
7105e828 DB |
18080 | |
18081 | /* Last step: make now unused interpreter insns from main | |
18082 | * prog consistent for later dump requests, so they can | |
18083 | * later look the same as if they were interpreted only. | |
18084 | */ | |
18085 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
69c087ba YS |
18086 | if (bpf_pseudo_func(insn)) { |
18087 | insn[0].imm = env->insn_aux_data[i].call_imm; | |
3990ed4c MKL |
18088 | insn[1].imm = insn->off; |
18089 | insn->off = 0; | |
69c087ba YS |
18090 | continue; |
18091 | } | |
23a2d70c | 18092 | if (!bpf_pseudo_call(insn)) |
7105e828 DB |
18093 | continue; |
18094 | insn->off = env->insn_aux_data[i].call_imm; | |
18095 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 18096 | insn->imm = subprog; |
7105e828 DB |
18097 | } |
18098 | ||
1c2a088a AS |
18099 | prog->jited = 1; |
18100 | prog->bpf_func = func[0]->bpf_func; | |
d00c6473 | 18101 | prog->jited_len = func[0]->jited_len; |
0108a4e9 KJ |
18102 | prog->aux->extable = func[0]->aux->extable; |
18103 | prog->aux->num_exentries = func[0]->aux->num_exentries; | |
1c2a088a | 18104 | prog->aux->func = func; |
f910cefa | 18105 | prog->aux->func_cnt = env->subprog_cnt; |
e16301fb | 18106 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
18107 | return 0; |
18108 | out_free: | |
f263a814 JF |
18109 | /* We failed JIT'ing, so at this point we need to unregister poke |
18110 | * descriptors from subprogs, so that kernel is not attempting to | |
18111 | * patch it anymore as we're freeing the subprog JIT memory. | |
18112 | */ | |
18113 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
18114 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
18115 | map_ptr->ops->map_poke_untrack(map_ptr, prog->aux); | |
18116 | } | |
18117 | /* At this point we're guaranteed that poke descriptors are not | |
18118 | * live anymore. We can just unlink its descriptor table as it's | |
18119 | * released with the main prog. | |
18120 | */ | |
a748c697 MF |
18121 | for (i = 0; i < env->subprog_cnt; i++) { |
18122 | if (!func[i]) | |
18123 | continue; | |
f263a814 | 18124 | func[i]->aux->poke_tab = NULL; |
a748c697 MF |
18125 | bpf_jit_free(func[i]); |
18126 | } | |
1c2a088a | 18127 | kfree(func); |
c7a89784 | 18128 | out_undo_insn: |
1c2a088a AS |
18129 | /* cleanup main prog to be interpreted */ |
18130 | prog->jit_requested = 0; | |
d2a3b7c5 | 18131 | prog->blinding_requested = 0; |
1c2a088a | 18132 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
23a2d70c | 18133 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
18134 | continue; |
18135 | insn->off = 0; | |
18136 | insn->imm = env->insn_aux_data[i].call_imm; | |
18137 | } | |
e16301fb | 18138 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
18139 | return err; |
18140 | } | |
18141 | ||
1ea47e01 AS |
18142 | static int fixup_call_args(struct bpf_verifier_env *env) |
18143 | { | |
19d28fbd | 18144 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
18145 | struct bpf_prog *prog = env->prog; |
18146 | struct bpf_insn *insn = prog->insnsi; | |
e6ac2450 | 18147 | bool has_kfunc_call = bpf_prog_has_kfunc_call(prog); |
1ea47e01 | 18148 | int i, depth; |
19d28fbd | 18149 | #endif |
e4052d06 | 18150 | int err = 0; |
1ea47e01 | 18151 | |
e4052d06 | 18152 | if (env->prog->jit_requested && |
9d03ebc7 | 18153 | !bpf_prog_is_offloaded(env->prog->aux)) { |
19d28fbd DM |
18154 | err = jit_subprogs(env); |
18155 | if (err == 0) | |
1c2a088a | 18156 | return 0; |
c7a89784 DB |
18157 | if (err == -EFAULT) |
18158 | return err; | |
19d28fbd DM |
18159 | } |
18160 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
e6ac2450 MKL |
18161 | if (has_kfunc_call) { |
18162 | verbose(env, "calling kernel functions are not allowed in non-JITed programs\n"); | |
18163 | return -EINVAL; | |
18164 | } | |
e411901c MF |
18165 | if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) { |
18166 | /* When JIT fails the progs with bpf2bpf calls and tail_calls | |
18167 | * have to be rejected, since interpreter doesn't support them yet. | |
18168 | */ | |
18169 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
18170 | return -EINVAL; | |
18171 | } | |
1ea47e01 | 18172 | for (i = 0; i < prog->len; i++, insn++) { |
69c087ba YS |
18173 | if (bpf_pseudo_func(insn)) { |
18174 | /* When JIT fails the progs with callback calls | |
18175 | * have to be rejected, since interpreter doesn't support them yet. | |
18176 | */ | |
18177 | verbose(env, "callbacks are not allowed in non-JITed programs\n"); | |
18178 | return -EINVAL; | |
18179 | } | |
18180 | ||
23a2d70c | 18181 | if (!bpf_pseudo_call(insn)) |
1ea47e01 AS |
18182 | continue; |
18183 | depth = get_callee_stack_depth(env, insn, i); | |
18184 | if (depth < 0) | |
18185 | return depth; | |
18186 | bpf_patch_call_args(insn, depth); | |
18187 | } | |
19d28fbd DM |
18188 | err = 0; |
18189 | #endif | |
18190 | return err; | |
1ea47e01 AS |
18191 | } |
18192 | ||
1cf3bfc6 IL |
18193 | /* replace a generic kfunc with a specialized version if necessary */ |
18194 | static void specialize_kfunc(struct bpf_verifier_env *env, | |
18195 | u32 func_id, u16 offset, unsigned long *addr) | |
18196 | { | |
18197 | struct bpf_prog *prog = env->prog; | |
18198 | bool seen_direct_write; | |
18199 | void *xdp_kfunc; | |
18200 | bool is_rdonly; | |
18201 | ||
18202 | if (bpf_dev_bound_kfunc_id(func_id)) { | |
18203 | xdp_kfunc = bpf_dev_bound_resolve_kfunc(prog, func_id); | |
18204 | if (xdp_kfunc) { | |
18205 | *addr = (unsigned long)xdp_kfunc; | |
18206 | return; | |
18207 | } | |
18208 | /* fallback to default kfunc when not supported by netdev */ | |
18209 | } | |
18210 | ||
18211 | if (offset) | |
18212 | return; | |
18213 | ||
18214 | if (func_id == special_kfunc_list[KF_bpf_dynptr_from_skb]) { | |
18215 | seen_direct_write = env->seen_direct_write; | |
18216 | is_rdonly = !may_access_direct_pkt_data(env, NULL, BPF_WRITE); | |
18217 | ||
18218 | if (is_rdonly) | |
18219 | *addr = (unsigned long)bpf_dynptr_from_skb_rdonly; | |
18220 | ||
18221 | /* restore env->seen_direct_write to its original value, since | |
18222 | * may_access_direct_pkt_data mutates it | |
18223 | */ | |
18224 | env->seen_direct_write = seen_direct_write; | |
18225 | } | |
18226 | } | |
18227 | ||
d2dcc67d DM |
18228 | static void __fixup_collection_insert_kfunc(struct bpf_insn_aux_data *insn_aux, |
18229 | u16 struct_meta_reg, | |
18230 | u16 node_offset_reg, | |
18231 | struct bpf_insn *insn, | |
18232 | struct bpf_insn *insn_buf, | |
18233 | int *cnt) | |
18234 | { | |
18235 | struct btf_struct_meta *kptr_struct_meta = insn_aux->kptr_struct_meta; | |
18236 | struct bpf_insn addr[2] = { BPF_LD_IMM64(struct_meta_reg, (long)kptr_struct_meta) }; | |
18237 | ||
18238 | insn_buf[0] = addr[0]; | |
18239 | insn_buf[1] = addr[1]; | |
18240 | insn_buf[2] = BPF_MOV64_IMM(node_offset_reg, insn_aux->insert_off); | |
18241 | insn_buf[3] = *insn; | |
18242 | *cnt = 4; | |
18243 | } | |
18244 | ||
958cf2e2 KKD |
18245 | static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
18246 | struct bpf_insn *insn_buf, int insn_idx, int *cnt) | |
e6ac2450 MKL |
18247 | { |
18248 | const struct bpf_kfunc_desc *desc; | |
18249 | ||
a5d82727 KKD |
18250 | if (!insn->imm) { |
18251 | verbose(env, "invalid kernel function call not eliminated in verifier pass\n"); | |
18252 | return -EINVAL; | |
18253 | } | |
18254 | ||
3d76a4d3 SF |
18255 | *cnt = 0; |
18256 | ||
1cf3bfc6 IL |
18257 | /* insn->imm has the btf func_id. Replace it with an offset relative to |
18258 | * __bpf_call_base, unless the JIT needs to call functions that are | |
18259 | * further than 32 bits away (bpf_jit_supports_far_kfunc_call()). | |
e6ac2450 | 18260 | */ |
2357672c | 18261 | desc = find_kfunc_desc(env->prog, insn->imm, insn->off); |
e6ac2450 MKL |
18262 | if (!desc) { |
18263 | verbose(env, "verifier internal error: kernel function descriptor not found for func_id %u\n", | |
18264 | insn->imm); | |
18265 | return -EFAULT; | |
18266 | } | |
18267 | ||
1cf3bfc6 IL |
18268 | if (!bpf_jit_supports_far_kfunc_call()) |
18269 | insn->imm = BPF_CALL_IMM(desc->addr); | |
958cf2e2 KKD |
18270 | if (insn->off) |
18271 | return 0; | |
18272 | if (desc->func_id == special_kfunc_list[KF_bpf_obj_new_impl]) { | |
18273 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; | |
18274 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
18275 | u64 obj_new_size = env->insn_aux_data[insn_idx].obj_new_size; | |
e6ac2450 | 18276 | |
958cf2e2 KKD |
18277 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_1, obj_new_size); |
18278 | insn_buf[1] = addr[0]; | |
18279 | insn_buf[2] = addr[1]; | |
18280 | insn_buf[3] = *insn; | |
18281 | *cnt = 4; | |
7c50b1cb DM |
18282 | } else if (desc->func_id == special_kfunc_list[KF_bpf_obj_drop_impl] || |
18283 | desc->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl]) { | |
ac9f0605 KKD |
18284 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; |
18285 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
18286 | ||
f0d991a0 DM |
18287 | if (desc->func_id == special_kfunc_list[KF_bpf_refcount_acquire_impl] && |
18288 | !kptr_struct_meta) { | |
18289 | verbose(env, "verifier internal error: kptr_struct_meta expected at insn_idx %d\n", | |
18290 | insn_idx); | |
18291 | return -EFAULT; | |
18292 | } | |
18293 | ||
ac9f0605 KKD |
18294 | insn_buf[0] = addr[0]; |
18295 | insn_buf[1] = addr[1]; | |
18296 | insn_buf[2] = *insn; | |
18297 | *cnt = 3; | |
d2dcc67d DM |
18298 | } else if (desc->func_id == special_kfunc_list[KF_bpf_list_push_back_impl] || |
18299 | desc->func_id == special_kfunc_list[KF_bpf_list_push_front_impl] || | |
18300 | desc->func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
f0d991a0 | 18301 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; |
d2dcc67d DM |
18302 | int struct_meta_reg = BPF_REG_3; |
18303 | int node_offset_reg = BPF_REG_4; | |
18304 | ||
18305 | /* rbtree_add has extra 'less' arg, so args-to-fixup are in diff regs */ | |
18306 | if (desc->func_id == special_kfunc_list[KF_bpf_rbtree_add_impl]) { | |
18307 | struct_meta_reg = BPF_REG_4; | |
18308 | node_offset_reg = BPF_REG_5; | |
18309 | } | |
18310 | ||
f0d991a0 DM |
18311 | if (!kptr_struct_meta) { |
18312 | verbose(env, "verifier internal error: kptr_struct_meta expected at insn_idx %d\n", | |
18313 | insn_idx); | |
18314 | return -EFAULT; | |
18315 | } | |
18316 | ||
d2dcc67d DM |
18317 | __fixup_collection_insert_kfunc(&env->insn_aux_data[insn_idx], struct_meta_reg, |
18318 | node_offset_reg, insn, insn_buf, cnt); | |
a35b9af4 YS |
18319 | } else if (desc->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx] || |
18320 | desc->func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { | |
fd264ca0 YS |
18321 | insn_buf[0] = BPF_MOV64_REG(BPF_REG_0, BPF_REG_1); |
18322 | *cnt = 1; | |
958cf2e2 | 18323 | } |
e6ac2450 MKL |
18324 | return 0; |
18325 | } | |
18326 | ||
e6ac5933 BJ |
18327 | /* Do various post-verification rewrites in a single program pass. |
18328 | * These rewrites simplify JIT and interpreter implementations. | |
e245c5c6 | 18329 | */ |
e6ac5933 | 18330 | static int do_misc_fixups(struct bpf_verifier_env *env) |
e245c5c6 | 18331 | { |
79741b3b | 18332 | struct bpf_prog *prog = env->prog; |
f92c1e18 | 18333 | enum bpf_attach_type eatype = prog->expected_attach_type; |
9b99edca | 18334 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
79741b3b | 18335 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 18336 | const struct bpf_func_proto *fn; |
79741b3b | 18337 | const int insn_cnt = prog->len; |
09772d92 | 18338 | const struct bpf_map_ops *ops; |
c93552c4 | 18339 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
18340 | struct bpf_insn insn_buf[16]; |
18341 | struct bpf_prog *new_prog; | |
18342 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 18343 | int i, ret, cnt, delta = 0; |
e245c5c6 | 18344 | |
79741b3b | 18345 | for (i = 0; i < insn_cnt; i++, insn++) { |
e6ac5933 | 18346 | /* Make divide-by-zero exceptions impossible. */ |
f6b1b3bf DB |
18347 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
18348 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
18349 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 18350 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf | 18351 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
e88b2c6e DB |
18352 | bool isdiv = BPF_OP(insn->code) == BPF_DIV; |
18353 | struct bpf_insn *patchlet; | |
18354 | struct bpf_insn chk_and_div[] = { | |
9b00f1b7 | 18355 | /* [R,W]x div 0 -> 0 */ |
e88b2c6e DB |
18356 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
18357 | BPF_JNE | BPF_K, insn->src_reg, | |
18358 | 0, 2, 0), | |
f6b1b3bf DB |
18359 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), |
18360 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
18361 | *insn, | |
18362 | }; | |
e88b2c6e | 18363 | struct bpf_insn chk_and_mod[] = { |
9b00f1b7 | 18364 | /* [R,W]x mod 0 -> [R,W]x */ |
e88b2c6e DB |
18365 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
18366 | BPF_JEQ | BPF_K, insn->src_reg, | |
9b00f1b7 | 18367 | 0, 1 + (is64 ? 0 : 1), 0), |
f6b1b3bf | 18368 | *insn, |
9b00f1b7 DB |
18369 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), |
18370 | BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), | |
f6b1b3bf | 18371 | }; |
f6b1b3bf | 18372 | |
e88b2c6e DB |
18373 | patchlet = isdiv ? chk_and_div : chk_and_mod; |
18374 | cnt = isdiv ? ARRAY_SIZE(chk_and_div) : | |
9b00f1b7 | 18375 | ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0); |
f6b1b3bf DB |
18376 | |
18377 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
18378 | if (!new_prog) |
18379 | return -ENOMEM; | |
18380 | ||
18381 | delta += cnt - 1; | |
18382 | env->prog = prog = new_prog; | |
18383 | insn = new_prog->insnsi + i + delta; | |
18384 | continue; | |
18385 | } | |
18386 | ||
e6ac5933 | 18387 | /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */ |
e0cea7ce DB |
18388 | if (BPF_CLASS(insn->code) == BPF_LD && |
18389 | (BPF_MODE(insn->code) == BPF_ABS || | |
18390 | BPF_MODE(insn->code) == BPF_IND)) { | |
18391 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
18392 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
18393 | verbose(env, "bpf verifier is misconfigured\n"); | |
18394 | return -EINVAL; | |
18395 | } | |
18396 | ||
18397 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18398 | if (!new_prog) | |
18399 | return -ENOMEM; | |
18400 | ||
18401 | delta += cnt - 1; | |
18402 | env->prog = prog = new_prog; | |
18403 | insn = new_prog->insnsi + i + delta; | |
18404 | continue; | |
18405 | } | |
18406 | ||
e6ac5933 | 18407 | /* Rewrite pointer arithmetic to mitigate speculation attacks. */ |
979d63d5 DB |
18408 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
18409 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
18410 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
18411 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
979d63d5 | 18412 | struct bpf_insn *patch = &insn_buf[0]; |
801c6058 | 18413 | bool issrc, isneg, isimm; |
979d63d5 DB |
18414 | u32 off_reg; |
18415 | ||
18416 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
18417 | if (!aux->alu_state || |
18418 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
18419 | continue; |
18420 | ||
18421 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
18422 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
18423 | BPF_ALU_SANITIZE_SRC; | |
801c6058 | 18424 | isimm = aux->alu_state & BPF_ALU_IMMEDIATE; |
979d63d5 DB |
18425 | |
18426 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
801c6058 DB |
18427 | if (isimm) { |
18428 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
18429 | } else { | |
18430 | if (isneg) | |
18431 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
18432 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
18433 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
18434 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
18435 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
18436 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
18437 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, off_reg); | |
18438 | } | |
b9b34ddb DB |
18439 | if (!issrc) |
18440 | *patch++ = BPF_MOV64_REG(insn->dst_reg, insn->src_reg); | |
18441 | insn->src_reg = BPF_REG_AX; | |
979d63d5 DB |
18442 | if (isneg) |
18443 | insn->code = insn->code == code_add ? | |
18444 | code_sub : code_add; | |
18445 | *patch++ = *insn; | |
801c6058 | 18446 | if (issrc && isneg && !isimm) |
979d63d5 DB |
18447 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); |
18448 | cnt = patch - insn_buf; | |
18449 | ||
18450 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18451 | if (!new_prog) | |
18452 | return -ENOMEM; | |
18453 | ||
18454 | delta += cnt - 1; | |
18455 | env->prog = prog = new_prog; | |
18456 | insn = new_prog->insnsi + i + delta; | |
18457 | continue; | |
18458 | } | |
18459 | ||
79741b3b AS |
18460 | if (insn->code != (BPF_JMP | BPF_CALL)) |
18461 | continue; | |
cc8b0b92 AS |
18462 | if (insn->src_reg == BPF_PSEUDO_CALL) |
18463 | continue; | |
e6ac2450 | 18464 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
958cf2e2 | 18465 | ret = fixup_kfunc_call(env, insn, insn_buf, i + delta, &cnt); |
e6ac2450 MKL |
18466 | if (ret) |
18467 | return ret; | |
958cf2e2 KKD |
18468 | if (cnt == 0) |
18469 | continue; | |
18470 | ||
18471 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18472 | if (!new_prog) | |
18473 | return -ENOMEM; | |
18474 | ||
18475 | delta += cnt - 1; | |
18476 | env->prog = prog = new_prog; | |
18477 | insn = new_prog->insnsi + i + delta; | |
e6ac2450 MKL |
18478 | continue; |
18479 | } | |
e245c5c6 | 18480 | |
79741b3b AS |
18481 | if (insn->imm == BPF_FUNC_get_route_realm) |
18482 | prog->dst_needed = 1; | |
18483 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
18484 | bpf_user_rnd_init_once(); | |
9802d865 JB |
18485 | if (insn->imm == BPF_FUNC_override_return) |
18486 | prog->kprobe_override = 1; | |
79741b3b | 18487 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
18488 | /* If we tail call into other programs, we |
18489 | * cannot make any assumptions since they can | |
18490 | * be replaced dynamically during runtime in | |
18491 | * the program array. | |
18492 | */ | |
18493 | prog->cb_access = 1; | |
e411901c MF |
18494 | if (!allow_tail_call_in_subprogs(env)) |
18495 | prog->aux->stack_depth = MAX_BPF_STACK; | |
18496 | prog->aux->max_pkt_offset = MAX_PACKET_OFF; | |
7b9f6da1 | 18497 | |
79741b3b | 18498 | /* mark bpf_tail_call as different opcode to avoid |
8fb33b60 | 18499 | * conditional branch in the interpreter for every normal |
79741b3b AS |
18500 | * call and to prevent accidental JITing by JIT compiler |
18501 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 18502 | */ |
79741b3b | 18503 | insn->imm = 0; |
71189fa9 | 18504 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 18505 | |
c93552c4 | 18506 | aux = &env->insn_aux_data[i + delta]; |
d2a3b7c5 | 18507 | if (env->bpf_capable && !prog->blinding_requested && |
cc52d914 | 18508 | prog->jit_requested && |
d2e4c1e6 DB |
18509 | !bpf_map_key_poisoned(aux) && |
18510 | !bpf_map_ptr_poisoned(aux) && | |
18511 | !bpf_map_ptr_unpriv(aux)) { | |
18512 | struct bpf_jit_poke_descriptor desc = { | |
18513 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
18514 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
18515 | .tail_call.key = bpf_map_key_immediate(aux), | |
a748c697 | 18516 | .insn_idx = i + delta, |
d2e4c1e6 DB |
18517 | }; |
18518 | ||
18519 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
18520 | if (ret < 0) { | |
18521 | verbose(env, "adding tail call poke descriptor failed\n"); | |
18522 | return ret; | |
18523 | } | |
18524 | ||
18525 | insn->imm = ret + 1; | |
18526 | continue; | |
18527 | } | |
18528 | ||
c93552c4 DB |
18529 | if (!bpf_map_ptr_unpriv(aux)) |
18530 | continue; | |
18531 | ||
b2157399 AS |
18532 | /* instead of changing every JIT dealing with tail_call |
18533 | * emit two extra insns: | |
18534 | * if (index >= max_entries) goto out; | |
18535 | * index &= array->index_mask; | |
18536 | * to avoid out-of-bounds cpu speculation | |
18537 | */ | |
c93552c4 | 18538 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 18539 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
18540 | return -EINVAL; |
18541 | } | |
c93552c4 | 18542 | |
d2e4c1e6 | 18543 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
18544 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
18545 | map_ptr->max_entries, 2); | |
18546 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
18547 | container_of(map_ptr, | |
18548 | struct bpf_array, | |
18549 | map)->index_mask); | |
18550 | insn_buf[2] = *insn; | |
18551 | cnt = 3; | |
18552 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18553 | if (!new_prog) | |
18554 | return -ENOMEM; | |
18555 | ||
18556 | delta += cnt - 1; | |
18557 | env->prog = prog = new_prog; | |
18558 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
18559 | continue; |
18560 | } | |
e245c5c6 | 18561 | |
b00628b1 AS |
18562 | if (insn->imm == BPF_FUNC_timer_set_callback) { |
18563 | /* The verifier will process callback_fn as many times as necessary | |
18564 | * with different maps and the register states prepared by | |
18565 | * set_timer_callback_state will be accurate. | |
18566 | * | |
18567 | * The following use case is valid: | |
18568 | * map1 is shared by prog1, prog2, prog3. | |
18569 | * prog1 calls bpf_timer_init for some map1 elements | |
18570 | * prog2 calls bpf_timer_set_callback for some map1 elements. | |
18571 | * Those that were not bpf_timer_init-ed will return -EINVAL. | |
18572 | * prog3 calls bpf_timer_start for some map1 elements. | |
18573 | * Those that were not both bpf_timer_init-ed and | |
18574 | * bpf_timer_set_callback-ed will return -EINVAL. | |
18575 | */ | |
18576 | struct bpf_insn ld_addrs[2] = { | |
18577 | BPF_LD_IMM64(BPF_REG_3, (long)prog->aux), | |
18578 | }; | |
18579 | ||
18580 | insn_buf[0] = ld_addrs[0]; | |
18581 | insn_buf[1] = ld_addrs[1]; | |
18582 | insn_buf[2] = *insn; | |
18583 | cnt = 3; | |
18584 | ||
18585 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18586 | if (!new_prog) | |
18587 | return -ENOMEM; | |
18588 | ||
18589 | delta += cnt - 1; | |
18590 | env->prog = prog = new_prog; | |
18591 | insn = new_prog->insnsi + i + delta; | |
18592 | goto patch_call_imm; | |
18593 | } | |
18594 | ||
9bb00b28 YS |
18595 | if (is_storage_get_function(insn->imm)) { |
18596 | if (!env->prog->aux->sleepable || | |
18597 | env->insn_aux_data[i + delta].storage_get_func_atomic) | |
d56c9fe6 | 18598 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_ATOMIC); |
9bb00b28 YS |
18599 | else |
18600 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_KERNEL); | |
b00fa38a JK |
18601 | insn_buf[1] = *insn; |
18602 | cnt = 2; | |
18603 | ||
18604 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18605 | if (!new_prog) | |
18606 | return -ENOMEM; | |
18607 | ||
18608 | delta += cnt - 1; | |
18609 | env->prog = prog = new_prog; | |
18610 | insn = new_prog->insnsi + i + delta; | |
18611 | goto patch_call_imm; | |
18612 | } | |
18613 | ||
89c63074 | 18614 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
18615 | * and other inlining handlers are currently limited to 64 bit |
18616 | * only. | |
89c63074 | 18617 | */ |
60b58afc | 18618 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
18619 | (insn->imm == BPF_FUNC_map_lookup_elem || |
18620 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
18621 | insn->imm == BPF_FUNC_map_delete_elem || |
18622 | insn->imm == BPF_FUNC_map_push_elem || | |
18623 | insn->imm == BPF_FUNC_map_pop_elem || | |
e6a4750f | 18624 | insn->imm == BPF_FUNC_map_peek_elem || |
0640c77c | 18625 | insn->imm == BPF_FUNC_redirect_map || |
07343110 FZ |
18626 | insn->imm == BPF_FUNC_for_each_map_elem || |
18627 | insn->imm == BPF_FUNC_map_lookup_percpu_elem)) { | |
c93552c4 DB |
18628 | aux = &env->insn_aux_data[i + delta]; |
18629 | if (bpf_map_ptr_poisoned(aux)) | |
18630 | goto patch_call_imm; | |
18631 | ||
d2e4c1e6 | 18632 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
18633 | ops = map_ptr->ops; |
18634 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
18635 | ops->map_gen_lookup) { | |
18636 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
4a8f87e6 DB |
18637 | if (cnt == -EOPNOTSUPP) |
18638 | goto patch_map_ops_generic; | |
18639 | if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
09772d92 DB |
18640 | verbose(env, "bpf verifier is misconfigured\n"); |
18641 | return -EINVAL; | |
18642 | } | |
81ed18ab | 18643 | |
09772d92 DB |
18644 | new_prog = bpf_patch_insn_data(env, i + delta, |
18645 | insn_buf, cnt); | |
18646 | if (!new_prog) | |
18647 | return -ENOMEM; | |
81ed18ab | 18648 | |
09772d92 DB |
18649 | delta += cnt - 1; |
18650 | env->prog = prog = new_prog; | |
18651 | insn = new_prog->insnsi + i + delta; | |
18652 | continue; | |
18653 | } | |
81ed18ab | 18654 | |
09772d92 DB |
18655 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
18656 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
18657 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
d7ba4cc9 | 18658 | (long (*)(struct bpf_map *map, void *key))NULL)); |
09772d92 | 18659 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, |
d7ba4cc9 | 18660 | (long (*)(struct bpf_map *map, void *key, void *value, |
09772d92 | 18661 | u64 flags))NULL)); |
84430d42 | 18662 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
d7ba4cc9 | 18663 | (long (*)(struct bpf_map *map, void *value, |
84430d42 DB |
18664 | u64 flags))NULL)); |
18665 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
d7ba4cc9 | 18666 | (long (*)(struct bpf_map *map, void *value))NULL)); |
84430d42 | 18667 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, |
d7ba4cc9 | 18668 | (long (*)(struct bpf_map *map, void *value))NULL)); |
e6a4750f | 18669 | BUILD_BUG_ON(!__same_type(ops->map_redirect, |
d7ba4cc9 | 18670 | (long (*)(struct bpf_map *map, u64 index, u64 flags))NULL)); |
0640c77c | 18671 | BUILD_BUG_ON(!__same_type(ops->map_for_each_callback, |
d7ba4cc9 | 18672 | (long (*)(struct bpf_map *map, |
0640c77c AI |
18673 | bpf_callback_t callback_fn, |
18674 | void *callback_ctx, | |
18675 | u64 flags))NULL)); | |
07343110 FZ |
18676 | BUILD_BUG_ON(!__same_type(ops->map_lookup_percpu_elem, |
18677 | (void *(*)(struct bpf_map *map, void *key, u32 cpu))NULL)); | |
e6a4750f | 18678 | |
4a8f87e6 | 18679 | patch_map_ops_generic: |
09772d92 DB |
18680 | switch (insn->imm) { |
18681 | case BPF_FUNC_map_lookup_elem: | |
3d717fad | 18682 | insn->imm = BPF_CALL_IMM(ops->map_lookup_elem); |
09772d92 DB |
18683 | continue; |
18684 | case BPF_FUNC_map_update_elem: | |
3d717fad | 18685 | insn->imm = BPF_CALL_IMM(ops->map_update_elem); |
09772d92 DB |
18686 | continue; |
18687 | case BPF_FUNC_map_delete_elem: | |
3d717fad | 18688 | insn->imm = BPF_CALL_IMM(ops->map_delete_elem); |
09772d92 | 18689 | continue; |
84430d42 | 18690 | case BPF_FUNC_map_push_elem: |
3d717fad | 18691 | insn->imm = BPF_CALL_IMM(ops->map_push_elem); |
84430d42 DB |
18692 | continue; |
18693 | case BPF_FUNC_map_pop_elem: | |
3d717fad | 18694 | insn->imm = BPF_CALL_IMM(ops->map_pop_elem); |
84430d42 DB |
18695 | continue; |
18696 | case BPF_FUNC_map_peek_elem: | |
3d717fad | 18697 | insn->imm = BPF_CALL_IMM(ops->map_peek_elem); |
84430d42 | 18698 | continue; |
e6a4750f | 18699 | case BPF_FUNC_redirect_map: |
3d717fad | 18700 | insn->imm = BPF_CALL_IMM(ops->map_redirect); |
e6a4750f | 18701 | continue; |
0640c77c AI |
18702 | case BPF_FUNC_for_each_map_elem: |
18703 | insn->imm = BPF_CALL_IMM(ops->map_for_each_callback); | |
e6a4750f | 18704 | continue; |
07343110 FZ |
18705 | case BPF_FUNC_map_lookup_percpu_elem: |
18706 | insn->imm = BPF_CALL_IMM(ops->map_lookup_percpu_elem); | |
18707 | continue; | |
09772d92 | 18708 | } |
81ed18ab | 18709 | |
09772d92 | 18710 | goto patch_call_imm; |
81ed18ab AS |
18711 | } |
18712 | ||
e6ac5933 | 18713 | /* Implement bpf_jiffies64 inline. */ |
5576b991 MKL |
18714 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
18715 | insn->imm == BPF_FUNC_jiffies64) { | |
18716 | struct bpf_insn ld_jiffies_addr[2] = { | |
18717 | BPF_LD_IMM64(BPF_REG_0, | |
18718 | (unsigned long)&jiffies), | |
18719 | }; | |
18720 | ||
18721 | insn_buf[0] = ld_jiffies_addr[0]; | |
18722 | insn_buf[1] = ld_jiffies_addr[1]; | |
18723 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
18724 | BPF_REG_0, 0); | |
18725 | cnt = 3; | |
18726 | ||
18727 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
18728 | cnt); | |
18729 | if (!new_prog) | |
18730 | return -ENOMEM; | |
18731 | ||
18732 | delta += cnt - 1; | |
18733 | env->prog = prog = new_prog; | |
18734 | insn = new_prog->insnsi + i + delta; | |
18735 | continue; | |
18736 | } | |
18737 | ||
f92c1e18 JO |
18738 | /* Implement bpf_get_func_arg inline. */ |
18739 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
18740 | insn->imm == BPF_FUNC_get_func_arg) { | |
18741 | /* Load nr_args from ctx - 8 */ | |
18742 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
18743 | insn_buf[1] = BPF_JMP32_REG(BPF_JGE, BPF_REG_2, BPF_REG_0, 6); | |
18744 | insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 3); | |
18745 | insn_buf[3] = BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1); | |
18746 | insn_buf[4] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_2, 0); | |
18747 | insn_buf[5] = BPF_STX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
18748 | insn_buf[6] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
18749 | insn_buf[7] = BPF_JMP_A(1); | |
18750 | insn_buf[8] = BPF_MOV64_IMM(BPF_REG_0, -EINVAL); | |
18751 | cnt = 9; | |
18752 | ||
18753 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18754 | if (!new_prog) | |
18755 | return -ENOMEM; | |
18756 | ||
18757 | delta += cnt - 1; | |
18758 | env->prog = prog = new_prog; | |
18759 | insn = new_prog->insnsi + i + delta; | |
18760 | continue; | |
18761 | } | |
18762 | ||
18763 | /* Implement bpf_get_func_ret inline. */ | |
18764 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
18765 | insn->imm == BPF_FUNC_get_func_ret) { | |
18766 | if (eatype == BPF_TRACE_FEXIT || | |
18767 | eatype == BPF_MODIFY_RETURN) { | |
18768 | /* Load nr_args from ctx - 8 */ | |
18769 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
18770 | insn_buf[1] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3); | |
18771 | insn_buf[2] = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1); | |
18772 | insn_buf[3] = BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
18773 | insn_buf[4] = BPF_STX_MEM(BPF_DW, BPF_REG_2, BPF_REG_3, 0); | |
18774 | insn_buf[5] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
18775 | cnt = 6; | |
18776 | } else { | |
18777 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, -EOPNOTSUPP); | |
18778 | cnt = 1; | |
18779 | } | |
18780 | ||
18781 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
18782 | if (!new_prog) | |
18783 | return -ENOMEM; | |
18784 | ||
18785 | delta += cnt - 1; | |
18786 | env->prog = prog = new_prog; | |
18787 | insn = new_prog->insnsi + i + delta; | |
18788 | continue; | |
18789 | } | |
18790 | ||
18791 | /* Implement get_func_arg_cnt inline. */ | |
18792 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
18793 | insn->imm == BPF_FUNC_get_func_arg_cnt) { | |
18794 | /* Load nr_args from ctx - 8 */ | |
18795 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
18796 | ||
18797 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
18798 | if (!new_prog) | |
18799 | return -ENOMEM; | |
18800 | ||
18801 | env->prog = prog = new_prog; | |
18802 | insn = new_prog->insnsi + i + delta; | |
18803 | continue; | |
18804 | } | |
18805 | ||
f705ec76 | 18806 | /* Implement bpf_get_func_ip inline. */ |
9b99edca JO |
18807 | if (prog_type == BPF_PROG_TYPE_TRACING && |
18808 | insn->imm == BPF_FUNC_get_func_ip) { | |
f92c1e18 JO |
18809 | /* Load IP address from ctx - 16 */ |
18810 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -16); | |
9b99edca JO |
18811 | |
18812 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
18813 | if (!new_prog) | |
18814 | return -ENOMEM; | |
18815 | ||
18816 | env->prog = prog = new_prog; | |
18817 | insn = new_prog->insnsi + i + delta; | |
18818 | continue; | |
18819 | } | |
18820 | ||
81ed18ab | 18821 | patch_call_imm: |
5e43f899 | 18822 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
18823 | /* all functions that have prototype and verifier allowed |
18824 | * programs to call them, must be real in-kernel functions | |
18825 | */ | |
18826 | if (!fn->func) { | |
61bd5218 JK |
18827 | verbose(env, |
18828 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
18829 | func_id_name(insn->imm), insn->imm); |
18830 | return -EFAULT; | |
e245c5c6 | 18831 | } |
79741b3b | 18832 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 18833 | } |
e245c5c6 | 18834 | |
d2e4c1e6 DB |
18835 | /* Since poke tab is now finalized, publish aux to tracker. */ |
18836 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
18837 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
18838 | if (!map_ptr->ops->map_poke_track || | |
18839 | !map_ptr->ops->map_poke_untrack || | |
18840 | !map_ptr->ops->map_poke_run) { | |
18841 | verbose(env, "bpf verifier is misconfigured\n"); | |
18842 | return -EINVAL; | |
18843 | } | |
18844 | ||
18845 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
18846 | if (ret < 0) { | |
18847 | verbose(env, "tracking tail call prog failed\n"); | |
18848 | return ret; | |
18849 | } | |
18850 | } | |
18851 | ||
1cf3bfc6 | 18852 | sort_kfunc_descs_by_imm_off(env->prog); |
e6ac2450 | 18853 | |
79741b3b AS |
18854 | return 0; |
18855 | } | |
e245c5c6 | 18856 | |
1ade2371 EZ |
18857 | static struct bpf_prog *inline_bpf_loop(struct bpf_verifier_env *env, |
18858 | int position, | |
18859 | s32 stack_base, | |
18860 | u32 callback_subprogno, | |
18861 | u32 *cnt) | |
18862 | { | |
18863 | s32 r6_offset = stack_base + 0 * BPF_REG_SIZE; | |
18864 | s32 r7_offset = stack_base + 1 * BPF_REG_SIZE; | |
18865 | s32 r8_offset = stack_base + 2 * BPF_REG_SIZE; | |
18866 | int reg_loop_max = BPF_REG_6; | |
18867 | int reg_loop_cnt = BPF_REG_7; | |
18868 | int reg_loop_ctx = BPF_REG_8; | |
18869 | ||
18870 | struct bpf_prog *new_prog; | |
18871 | u32 callback_start; | |
18872 | u32 call_insn_offset; | |
18873 | s32 callback_offset; | |
18874 | ||
18875 | /* This represents an inlined version of bpf_iter.c:bpf_loop, | |
18876 | * be careful to modify this code in sync. | |
18877 | */ | |
18878 | struct bpf_insn insn_buf[] = { | |
18879 | /* Return error and jump to the end of the patch if | |
18880 | * expected number of iterations is too big. | |
18881 | */ | |
18882 | BPF_JMP_IMM(BPF_JLE, BPF_REG_1, BPF_MAX_LOOPS, 2), | |
18883 | BPF_MOV32_IMM(BPF_REG_0, -E2BIG), | |
18884 | BPF_JMP_IMM(BPF_JA, 0, 0, 16), | |
18885 | /* spill R6, R7, R8 to use these as loop vars */ | |
18886 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_6, r6_offset), | |
18887 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_7, r7_offset), | |
18888 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_8, r8_offset), | |
18889 | /* initialize loop vars */ | |
18890 | BPF_MOV64_REG(reg_loop_max, BPF_REG_1), | |
18891 | BPF_MOV32_IMM(reg_loop_cnt, 0), | |
18892 | BPF_MOV64_REG(reg_loop_ctx, BPF_REG_3), | |
18893 | /* loop header, | |
18894 | * if reg_loop_cnt >= reg_loop_max skip the loop body | |
18895 | */ | |
18896 | BPF_JMP_REG(BPF_JGE, reg_loop_cnt, reg_loop_max, 5), | |
18897 | /* callback call, | |
18898 | * correct callback offset would be set after patching | |
18899 | */ | |
18900 | BPF_MOV64_REG(BPF_REG_1, reg_loop_cnt), | |
18901 | BPF_MOV64_REG(BPF_REG_2, reg_loop_ctx), | |
18902 | BPF_CALL_REL(0), | |
18903 | /* increment loop counter */ | |
18904 | BPF_ALU64_IMM(BPF_ADD, reg_loop_cnt, 1), | |
18905 | /* jump to loop header if callback returned 0 */ | |
18906 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, -6), | |
18907 | /* return value of bpf_loop, | |
18908 | * set R0 to the number of iterations | |
18909 | */ | |
18910 | BPF_MOV64_REG(BPF_REG_0, reg_loop_cnt), | |
18911 | /* restore original values of R6, R7, R8 */ | |
18912 | BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_10, r6_offset), | |
18913 | BPF_LDX_MEM(BPF_DW, BPF_REG_7, BPF_REG_10, r7_offset), | |
18914 | BPF_LDX_MEM(BPF_DW, BPF_REG_8, BPF_REG_10, r8_offset), | |
18915 | }; | |
18916 | ||
18917 | *cnt = ARRAY_SIZE(insn_buf); | |
18918 | new_prog = bpf_patch_insn_data(env, position, insn_buf, *cnt); | |
18919 | if (!new_prog) | |
18920 | return new_prog; | |
18921 | ||
18922 | /* callback start is known only after patching */ | |
18923 | callback_start = env->subprog_info[callback_subprogno].start; | |
18924 | /* Note: insn_buf[12] is an offset of BPF_CALL_REL instruction */ | |
18925 | call_insn_offset = position + 12; | |
18926 | callback_offset = callback_start - call_insn_offset - 1; | |
fb4e3b33 | 18927 | new_prog->insnsi[call_insn_offset].imm = callback_offset; |
1ade2371 EZ |
18928 | |
18929 | return new_prog; | |
18930 | } | |
18931 | ||
18932 | static bool is_bpf_loop_call(struct bpf_insn *insn) | |
18933 | { | |
18934 | return insn->code == (BPF_JMP | BPF_CALL) && | |
18935 | insn->src_reg == 0 && | |
18936 | insn->imm == BPF_FUNC_loop; | |
18937 | } | |
18938 | ||
18939 | /* For all sub-programs in the program (including main) check | |
18940 | * insn_aux_data to see if there are bpf_loop calls that require | |
18941 | * inlining. If such calls are found the calls are replaced with a | |
18942 | * sequence of instructions produced by `inline_bpf_loop` function and | |
18943 | * subprog stack_depth is increased by the size of 3 registers. | |
18944 | * This stack space is used to spill values of the R6, R7, R8. These | |
18945 | * registers are used to store the loop bound, counter and context | |
18946 | * variables. | |
18947 | */ | |
18948 | static int optimize_bpf_loop(struct bpf_verifier_env *env) | |
18949 | { | |
18950 | struct bpf_subprog_info *subprogs = env->subprog_info; | |
18951 | int i, cur_subprog = 0, cnt, delta = 0; | |
18952 | struct bpf_insn *insn = env->prog->insnsi; | |
18953 | int insn_cnt = env->prog->len; | |
18954 | u16 stack_depth = subprogs[cur_subprog].stack_depth; | |
18955 | u16 stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
18956 | u16 stack_depth_extra = 0; | |
18957 | ||
18958 | for (i = 0; i < insn_cnt; i++, insn++) { | |
18959 | struct bpf_loop_inline_state *inline_state = | |
18960 | &env->insn_aux_data[i + delta].loop_inline_state; | |
18961 | ||
18962 | if (is_bpf_loop_call(insn) && inline_state->fit_for_inline) { | |
18963 | struct bpf_prog *new_prog; | |
18964 | ||
18965 | stack_depth_extra = BPF_REG_SIZE * 3 + stack_depth_roundup; | |
18966 | new_prog = inline_bpf_loop(env, | |
18967 | i + delta, | |
18968 | -(stack_depth + stack_depth_extra), | |
18969 | inline_state->callback_subprogno, | |
18970 | &cnt); | |
18971 | if (!new_prog) | |
18972 | return -ENOMEM; | |
18973 | ||
18974 | delta += cnt - 1; | |
18975 | env->prog = new_prog; | |
18976 | insn = new_prog->insnsi + i + delta; | |
18977 | } | |
18978 | ||
18979 | if (subprogs[cur_subprog + 1].start == i + delta + 1) { | |
18980 | subprogs[cur_subprog].stack_depth += stack_depth_extra; | |
18981 | cur_subprog++; | |
18982 | stack_depth = subprogs[cur_subprog].stack_depth; | |
18983 | stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
18984 | stack_depth_extra = 0; | |
18985 | } | |
18986 | } | |
18987 | ||
18988 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
18989 | ||
18990 | return 0; | |
18991 | } | |
18992 | ||
58e2af8b | 18993 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 18994 | { |
58e2af8b | 18995 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
18996 | int i; |
18997 | ||
9f4686c4 AS |
18998 | sl = env->free_list; |
18999 | while (sl) { | |
19000 | sln = sl->next; | |
19001 | free_verifier_state(&sl->state, false); | |
19002 | kfree(sl); | |
19003 | sl = sln; | |
19004 | } | |
51c39bb1 | 19005 | env->free_list = NULL; |
9f4686c4 | 19006 | |
f1bca824 AS |
19007 | if (!env->explored_states) |
19008 | return; | |
19009 | ||
dc2a4ebc | 19010 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
19011 | sl = env->explored_states[i]; |
19012 | ||
a8f500af AS |
19013 | while (sl) { |
19014 | sln = sl->next; | |
19015 | free_verifier_state(&sl->state, false); | |
19016 | kfree(sl); | |
19017 | sl = sln; | |
19018 | } | |
51c39bb1 | 19019 | env->explored_states[i] = NULL; |
f1bca824 | 19020 | } |
51c39bb1 | 19021 | } |
f1bca824 | 19022 | |
51c39bb1 AS |
19023 | static int do_check_common(struct bpf_verifier_env *env, int subprog) |
19024 | { | |
6f8a57cc | 19025 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
19026 | struct bpf_verifier_state *state; |
19027 | struct bpf_reg_state *regs; | |
19028 | int ret, i; | |
19029 | ||
19030 | env->prev_linfo = NULL; | |
19031 | env->pass_cnt++; | |
19032 | ||
19033 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
19034 | if (!state) | |
19035 | return -ENOMEM; | |
19036 | state->curframe = 0; | |
19037 | state->speculative = false; | |
19038 | state->branches = 1; | |
19039 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
19040 | if (!state->frame[0]) { | |
19041 | kfree(state); | |
19042 | return -ENOMEM; | |
19043 | } | |
19044 | env->cur_state = state; | |
19045 | init_func_state(env, state->frame[0], | |
19046 | BPF_MAIN_FUNC /* callsite */, | |
19047 | 0 /* frameno */, | |
19048 | subprog); | |
be2ef816 AN |
19049 | state->first_insn_idx = env->subprog_info[subprog].start; |
19050 | state->last_insn_idx = -1; | |
51c39bb1 AS |
19051 | |
19052 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 19053 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
51c39bb1 AS |
19054 | ret = btf_prepare_func_args(env, subprog, regs); |
19055 | if (ret) | |
19056 | goto out; | |
19057 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
19058 | if (regs[i].type == PTR_TO_CTX) | |
19059 | mark_reg_known_zero(env, regs, i); | |
19060 | else if (regs[i].type == SCALAR_VALUE) | |
19061 | mark_reg_unknown(env, regs, i); | |
cf9f2f8d | 19062 | else if (base_type(regs[i].type) == PTR_TO_MEM) { |
e5069b9c DB |
19063 | const u32 mem_size = regs[i].mem_size; |
19064 | ||
19065 | mark_reg_known_zero(env, regs, i); | |
19066 | regs[i].mem_size = mem_size; | |
19067 | regs[i].id = ++env->id_gen; | |
19068 | } | |
51c39bb1 AS |
19069 | } |
19070 | } else { | |
19071 | /* 1st arg to a function */ | |
19072 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
19073 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
34747c41 | 19074 | ret = btf_check_subprog_arg_match(env, subprog, regs); |
51c39bb1 AS |
19075 | if (ret == -EFAULT) |
19076 | /* unlikely verifier bug. abort. | |
19077 | * ret == 0 and ret < 0 are sadly acceptable for | |
19078 | * main() function due to backward compatibility. | |
19079 | * Like socket filter program may be written as: | |
19080 | * int bpf_prog(struct pt_regs *ctx) | |
19081 | * and never dereference that ctx in the program. | |
19082 | * 'struct pt_regs' is a type mismatch for socket | |
19083 | * filter that should be using 'struct __sk_buff'. | |
19084 | */ | |
19085 | goto out; | |
19086 | } | |
19087 | ||
19088 | ret = do_check(env); | |
19089 | out: | |
f59bbfc2 AS |
19090 | /* check for NULL is necessary, since cur_state can be freed inside |
19091 | * do_check() under memory pressure. | |
19092 | */ | |
19093 | if (env->cur_state) { | |
19094 | free_verifier_state(env->cur_state, true); | |
19095 | env->cur_state = NULL; | |
19096 | } | |
6f8a57cc AN |
19097 | while (!pop_stack(env, NULL, NULL, false)); |
19098 | if (!ret && pop_log) | |
19099 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 | 19100 | free_states(env); |
51c39bb1 AS |
19101 | return ret; |
19102 | } | |
19103 | ||
19104 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
19105 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
19106 | * Consider: | |
19107 | * int bar(int); | |
19108 | * int foo(int f) | |
19109 | * { | |
19110 | * return bar(f); | |
19111 | * } | |
19112 | * int bar(int b) | |
19113 | * { | |
19114 | * ... | |
19115 | * } | |
19116 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
19117 | * will be assumed that bar() already verified successfully and call to bar() | |
19118 | * from foo() will be checked for type match only. Later bar() will be verified | |
19119 | * independently to check that it's safe for R1=any_scalar_value. | |
19120 | */ | |
19121 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
19122 | { | |
19123 | struct bpf_prog_aux *aux = env->prog->aux; | |
19124 | int i, ret; | |
19125 | ||
19126 | if (!aux->func_info) | |
19127 | return 0; | |
19128 | ||
19129 | for (i = 1; i < env->subprog_cnt; i++) { | |
19130 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
19131 | continue; | |
19132 | env->insn_idx = env->subprog_info[i].start; | |
19133 | WARN_ON_ONCE(env->insn_idx == 0); | |
19134 | ret = do_check_common(env, i); | |
19135 | if (ret) { | |
19136 | return ret; | |
19137 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
19138 | verbose(env, | |
19139 | "Func#%d is safe for any args that match its prototype\n", | |
19140 | i); | |
19141 | } | |
19142 | } | |
19143 | return 0; | |
19144 | } | |
19145 | ||
19146 | static int do_check_main(struct bpf_verifier_env *env) | |
19147 | { | |
19148 | int ret; | |
19149 | ||
19150 | env->insn_idx = 0; | |
19151 | ret = do_check_common(env, 0); | |
19152 | if (!ret) | |
19153 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
19154 | return ret; | |
19155 | } | |
19156 | ||
19157 | ||
06ee7115 AS |
19158 | static void print_verification_stats(struct bpf_verifier_env *env) |
19159 | { | |
19160 | int i; | |
19161 | ||
19162 | if (env->log.level & BPF_LOG_STATS) { | |
19163 | verbose(env, "verification time %lld usec\n", | |
19164 | div_u64(env->verification_time, 1000)); | |
19165 | verbose(env, "stack depth "); | |
19166 | for (i = 0; i < env->subprog_cnt; i++) { | |
19167 | u32 depth = env->subprog_info[i].stack_depth; | |
19168 | ||
19169 | verbose(env, "%d", depth); | |
19170 | if (i + 1 < env->subprog_cnt) | |
19171 | verbose(env, "+"); | |
19172 | } | |
19173 | verbose(env, "\n"); | |
19174 | } | |
19175 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
19176 | "total_states %d peak_states %d mark_read %d\n", | |
19177 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
19178 | env->max_states_per_insn, env->total_states, | |
19179 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
19180 | } |
19181 | ||
27ae7997 MKL |
19182 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
19183 | { | |
19184 | const struct btf_type *t, *func_proto; | |
19185 | const struct bpf_struct_ops *st_ops; | |
19186 | const struct btf_member *member; | |
19187 | struct bpf_prog *prog = env->prog; | |
19188 | u32 btf_id, member_idx; | |
19189 | const char *mname; | |
19190 | ||
12aa8a94 THJ |
19191 | if (!prog->gpl_compatible) { |
19192 | verbose(env, "struct ops programs must have a GPL compatible license\n"); | |
19193 | return -EINVAL; | |
19194 | } | |
19195 | ||
27ae7997 MKL |
19196 | btf_id = prog->aux->attach_btf_id; |
19197 | st_ops = bpf_struct_ops_find(btf_id); | |
19198 | if (!st_ops) { | |
19199 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
19200 | btf_id); | |
19201 | return -ENOTSUPP; | |
19202 | } | |
19203 | ||
19204 | t = st_ops->type; | |
19205 | member_idx = prog->expected_attach_type; | |
19206 | if (member_idx >= btf_type_vlen(t)) { | |
19207 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
19208 | member_idx, st_ops->name); | |
19209 | return -EINVAL; | |
19210 | } | |
19211 | ||
19212 | member = &btf_type_member(t)[member_idx]; | |
19213 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
19214 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
19215 | NULL); | |
19216 | if (!func_proto) { | |
19217 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
19218 | mname, member_idx, st_ops->name); | |
19219 | return -EINVAL; | |
19220 | } | |
19221 | ||
19222 | if (st_ops->check_member) { | |
51a52a29 | 19223 | int err = st_ops->check_member(t, member, prog); |
27ae7997 MKL |
19224 | |
19225 | if (err) { | |
19226 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
19227 | mname, st_ops->name); | |
19228 | return err; | |
19229 | } | |
19230 | } | |
19231 | ||
19232 | prog->aux->attach_func_proto = func_proto; | |
19233 | prog->aux->attach_func_name = mname; | |
19234 | env->ops = st_ops->verifier_ops; | |
19235 | ||
19236 | return 0; | |
19237 | } | |
6ba43b76 KS |
19238 | #define SECURITY_PREFIX "security_" |
19239 | ||
f7b12b6f | 19240 | static int check_attach_modify_return(unsigned long addr, const char *func_name) |
6ba43b76 | 19241 | { |
69191754 | 19242 | if (within_error_injection_list(addr) || |
f7b12b6f | 19243 | !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1)) |
6ba43b76 | 19244 | return 0; |
6ba43b76 | 19245 | |
6ba43b76 KS |
19246 | return -EINVAL; |
19247 | } | |
27ae7997 | 19248 | |
1e6c62a8 AS |
19249 | /* list of non-sleepable functions that are otherwise on |
19250 | * ALLOW_ERROR_INJECTION list | |
19251 | */ | |
19252 | BTF_SET_START(btf_non_sleepable_error_inject) | |
19253 | /* Three functions below can be called from sleepable and non-sleepable context. | |
19254 | * Assume non-sleepable from bpf safety point of view. | |
19255 | */ | |
9dd3d069 | 19256 | BTF_ID(func, __filemap_add_folio) |
1e6c62a8 AS |
19257 | BTF_ID(func, should_fail_alloc_page) |
19258 | BTF_ID(func, should_failslab) | |
19259 | BTF_SET_END(btf_non_sleepable_error_inject) | |
19260 | ||
19261 | static int check_non_sleepable_error_inject(u32 btf_id) | |
19262 | { | |
19263 | return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id); | |
19264 | } | |
19265 | ||
f7b12b6f THJ |
19266 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
19267 | const struct bpf_prog *prog, | |
19268 | const struct bpf_prog *tgt_prog, | |
19269 | u32 btf_id, | |
19270 | struct bpf_attach_target_info *tgt_info) | |
38207291 | 19271 | { |
be8704ff | 19272 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
f1b9509c | 19273 | const char prefix[] = "btf_trace_"; |
5b92a28a | 19274 | int ret = 0, subprog = -1, i; |
38207291 | 19275 | const struct btf_type *t; |
5b92a28a | 19276 | bool conservative = true; |
38207291 | 19277 | const char *tname; |
5b92a28a | 19278 | struct btf *btf; |
f7b12b6f | 19279 | long addr = 0; |
31bf1dbc | 19280 | struct module *mod = NULL; |
38207291 | 19281 | |
f1b9509c | 19282 | if (!btf_id) { |
efc68158 | 19283 | bpf_log(log, "Tracing programs must provide btf_id\n"); |
f1b9509c AS |
19284 | return -EINVAL; |
19285 | } | |
22dc4a0f | 19286 | btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf; |
5b92a28a | 19287 | if (!btf) { |
efc68158 | 19288 | bpf_log(log, |
5b92a28a AS |
19289 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); |
19290 | return -EINVAL; | |
19291 | } | |
19292 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c | 19293 | if (!t) { |
efc68158 | 19294 | bpf_log(log, "attach_btf_id %u is invalid\n", btf_id); |
f1b9509c AS |
19295 | return -EINVAL; |
19296 | } | |
5b92a28a | 19297 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c | 19298 | if (!tname) { |
efc68158 | 19299 | bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id); |
f1b9509c AS |
19300 | return -EINVAL; |
19301 | } | |
5b92a28a AS |
19302 | if (tgt_prog) { |
19303 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
19304 | ||
fd7c211d THJ |
19305 | if (bpf_prog_is_dev_bound(prog->aux) && |
19306 | !bpf_prog_dev_bound_match(prog, tgt_prog)) { | |
19307 | bpf_log(log, "Target program bound device mismatch"); | |
3d76a4d3 SF |
19308 | return -EINVAL; |
19309 | } | |
19310 | ||
5b92a28a AS |
19311 | for (i = 0; i < aux->func_info_cnt; i++) |
19312 | if (aux->func_info[i].type_id == btf_id) { | |
19313 | subprog = i; | |
19314 | break; | |
19315 | } | |
19316 | if (subprog == -1) { | |
efc68158 | 19317 | bpf_log(log, "Subprog %s doesn't exist\n", tname); |
5b92a28a AS |
19318 | return -EINVAL; |
19319 | } | |
19320 | conservative = aux->func_info_aux[subprog].unreliable; | |
be8704ff AS |
19321 | if (prog_extension) { |
19322 | if (conservative) { | |
efc68158 | 19323 | bpf_log(log, |
be8704ff AS |
19324 | "Cannot replace static functions\n"); |
19325 | return -EINVAL; | |
19326 | } | |
19327 | if (!prog->jit_requested) { | |
efc68158 | 19328 | bpf_log(log, |
be8704ff AS |
19329 | "Extension programs should be JITed\n"); |
19330 | return -EINVAL; | |
19331 | } | |
be8704ff AS |
19332 | } |
19333 | if (!tgt_prog->jited) { | |
efc68158 | 19334 | bpf_log(log, "Can attach to only JITed progs\n"); |
be8704ff AS |
19335 | return -EINVAL; |
19336 | } | |
19337 | if (tgt_prog->type == prog->type) { | |
19338 | /* Cannot fentry/fexit another fentry/fexit program. | |
19339 | * Cannot attach program extension to another extension. | |
19340 | * It's ok to attach fentry/fexit to extension program. | |
19341 | */ | |
efc68158 | 19342 | bpf_log(log, "Cannot recursively attach\n"); |
be8704ff AS |
19343 | return -EINVAL; |
19344 | } | |
19345 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
19346 | prog_extension && | |
19347 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
19348 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
19349 | /* Program extensions can extend all program types | |
19350 | * except fentry/fexit. The reason is the following. | |
19351 | * The fentry/fexit programs are used for performance | |
19352 | * analysis, stats and can be attached to any program | |
19353 | * type except themselves. When extension program is | |
19354 | * replacing XDP function it is necessary to allow | |
19355 | * performance analysis of all functions. Both original | |
19356 | * XDP program and its program extension. Hence | |
19357 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
19358 | * allowed. If extending of fentry/fexit was allowed it | |
19359 | * would be possible to create long call chain | |
19360 | * fentry->extension->fentry->extension beyond | |
19361 | * reasonable stack size. Hence extending fentry is not | |
19362 | * allowed. | |
19363 | */ | |
efc68158 | 19364 | bpf_log(log, "Cannot extend fentry/fexit\n"); |
be8704ff AS |
19365 | return -EINVAL; |
19366 | } | |
5b92a28a | 19367 | } else { |
be8704ff | 19368 | if (prog_extension) { |
efc68158 | 19369 | bpf_log(log, "Cannot replace kernel functions\n"); |
be8704ff AS |
19370 | return -EINVAL; |
19371 | } | |
5b92a28a | 19372 | } |
f1b9509c AS |
19373 | |
19374 | switch (prog->expected_attach_type) { | |
19375 | case BPF_TRACE_RAW_TP: | |
5b92a28a | 19376 | if (tgt_prog) { |
efc68158 | 19377 | bpf_log(log, |
5b92a28a AS |
19378 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); |
19379 | return -EINVAL; | |
19380 | } | |
38207291 | 19381 | if (!btf_type_is_typedef(t)) { |
efc68158 | 19382 | bpf_log(log, "attach_btf_id %u is not a typedef\n", |
38207291 MKL |
19383 | btf_id); |
19384 | return -EINVAL; | |
19385 | } | |
f1b9509c | 19386 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
efc68158 | 19387 | bpf_log(log, "attach_btf_id %u points to wrong type name %s\n", |
38207291 MKL |
19388 | btf_id, tname); |
19389 | return -EINVAL; | |
19390 | } | |
19391 | tname += sizeof(prefix) - 1; | |
5b92a28a | 19392 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
19393 | if (!btf_type_is_ptr(t)) |
19394 | /* should never happen in valid vmlinux build */ | |
19395 | return -EINVAL; | |
5b92a28a | 19396 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
19397 | if (!btf_type_is_func_proto(t)) |
19398 | /* should never happen in valid vmlinux build */ | |
19399 | return -EINVAL; | |
19400 | ||
f7b12b6f | 19401 | break; |
15d83c4d YS |
19402 | case BPF_TRACE_ITER: |
19403 | if (!btf_type_is_func(t)) { | |
efc68158 | 19404 | bpf_log(log, "attach_btf_id %u is not a function\n", |
15d83c4d YS |
19405 | btf_id); |
19406 | return -EINVAL; | |
19407 | } | |
19408 | t = btf_type_by_id(btf, t->type); | |
19409 | if (!btf_type_is_func_proto(t)) | |
19410 | return -EINVAL; | |
f7b12b6f THJ |
19411 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); |
19412 | if (ret) | |
19413 | return ret; | |
19414 | break; | |
be8704ff AS |
19415 | default: |
19416 | if (!prog_extension) | |
19417 | return -EINVAL; | |
df561f66 | 19418 | fallthrough; |
ae240823 | 19419 | case BPF_MODIFY_RETURN: |
9e4e01df | 19420 | case BPF_LSM_MAC: |
69fd337a | 19421 | case BPF_LSM_CGROUP: |
fec56f58 AS |
19422 | case BPF_TRACE_FENTRY: |
19423 | case BPF_TRACE_FEXIT: | |
19424 | if (!btf_type_is_func(t)) { | |
efc68158 | 19425 | bpf_log(log, "attach_btf_id %u is not a function\n", |
fec56f58 AS |
19426 | btf_id); |
19427 | return -EINVAL; | |
19428 | } | |
be8704ff | 19429 | if (prog_extension && |
efc68158 | 19430 | btf_check_type_match(log, prog, btf, t)) |
be8704ff | 19431 | return -EINVAL; |
5b92a28a | 19432 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
19433 | if (!btf_type_is_func_proto(t)) |
19434 | return -EINVAL; | |
f7b12b6f | 19435 | |
4a1e7c0c THJ |
19436 | if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) && |
19437 | (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type || | |
19438 | prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type)) | |
19439 | return -EINVAL; | |
19440 | ||
f7b12b6f | 19441 | if (tgt_prog && conservative) |
5b92a28a | 19442 | t = NULL; |
f7b12b6f THJ |
19443 | |
19444 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); | |
fec56f58 | 19445 | if (ret < 0) |
f7b12b6f THJ |
19446 | return ret; |
19447 | ||
5b92a28a | 19448 | if (tgt_prog) { |
e9eeec58 YS |
19449 | if (subprog == 0) |
19450 | addr = (long) tgt_prog->bpf_func; | |
19451 | else | |
19452 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a | 19453 | } else { |
31bf1dbc VM |
19454 | if (btf_is_module(btf)) { |
19455 | mod = btf_try_get_module(btf); | |
19456 | if (mod) | |
19457 | addr = find_kallsyms_symbol_value(mod, tname); | |
19458 | else | |
19459 | addr = 0; | |
19460 | } else { | |
19461 | addr = kallsyms_lookup_name(tname); | |
19462 | } | |
5b92a28a | 19463 | if (!addr) { |
31bf1dbc | 19464 | module_put(mod); |
efc68158 | 19465 | bpf_log(log, |
5b92a28a AS |
19466 | "The address of function %s cannot be found\n", |
19467 | tname); | |
f7b12b6f | 19468 | return -ENOENT; |
5b92a28a | 19469 | } |
fec56f58 | 19470 | } |
18644cec | 19471 | |
1e6c62a8 AS |
19472 | if (prog->aux->sleepable) { |
19473 | ret = -EINVAL; | |
19474 | switch (prog->type) { | |
19475 | case BPF_PROG_TYPE_TRACING: | |
5b481aca BT |
19476 | |
19477 | /* fentry/fexit/fmod_ret progs can be sleepable if they are | |
1e6c62a8 AS |
19478 | * attached to ALLOW_ERROR_INJECTION and are not in denylist. |
19479 | */ | |
19480 | if (!check_non_sleepable_error_inject(btf_id) && | |
19481 | within_error_injection_list(addr)) | |
19482 | ret = 0; | |
5b481aca BT |
19483 | /* fentry/fexit/fmod_ret progs can also be sleepable if they are |
19484 | * in the fmodret id set with the KF_SLEEPABLE flag. | |
19485 | */ | |
19486 | else { | |
e924e80e AG |
19487 | u32 *flags = btf_kfunc_is_modify_return(btf, btf_id, |
19488 | prog); | |
5b481aca BT |
19489 | |
19490 | if (flags && (*flags & KF_SLEEPABLE)) | |
19491 | ret = 0; | |
19492 | } | |
1e6c62a8 AS |
19493 | break; |
19494 | case BPF_PROG_TYPE_LSM: | |
19495 | /* LSM progs check that they are attached to bpf_lsm_*() funcs. | |
19496 | * Only some of them are sleepable. | |
19497 | */ | |
423f1610 | 19498 | if (bpf_lsm_is_sleepable_hook(btf_id)) |
1e6c62a8 AS |
19499 | ret = 0; |
19500 | break; | |
19501 | default: | |
19502 | break; | |
19503 | } | |
f7b12b6f | 19504 | if (ret) { |
31bf1dbc | 19505 | module_put(mod); |
f7b12b6f THJ |
19506 | bpf_log(log, "%s is not sleepable\n", tname); |
19507 | return ret; | |
19508 | } | |
1e6c62a8 | 19509 | } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) { |
1af9270e | 19510 | if (tgt_prog) { |
31bf1dbc | 19511 | module_put(mod); |
efc68158 | 19512 | bpf_log(log, "can't modify return codes of BPF programs\n"); |
f7b12b6f THJ |
19513 | return -EINVAL; |
19514 | } | |
5b481aca | 19515 | ret = -EINVAL; |
e924e80e | 19516 | if (btf_kfunc_is_modify_return(btf, btf_id, prog) || |
5b481aca BT |
19517 | !check_attach_modify_return(addr, tname)) |
19518 | ret = 0; | |
f7b12b6f | 19519 | if (ret) { |
31bf1dbc | 19520 | module_put(mod); |
f7b12b6f THJ |
19521 | bpf_log(log, "%s() is not modifiable\n", tname); |
19522 | return ret; | |
1af9270e | 19523 | } |
18644cec | 19524 | } |
f7b12b6f THJ |
19525 | |
19526 | break; | |
19527 | } | |
19528 | tgt_info->tgt_addr = addr; | |
19529 | tgt_info->tgt_name = tname; | |
19530 | tgt_info->tgt_type = t; | |
31bf1dbc | 19531 | tgt_info->tgt_mod = mod; |
f7b12b6f THJ |
19532 | return 0; |
19533 | } | |
19534 | ||
35e3815f JO |
19535 | BTF_SET_START(btf_id_deny) |
19536 | BTF_ID_UNUSED | |
19537 | #ifdef CONFIG_SMP | |
19538 | BTF_ID(func, migrate_disable) | |
19539 | BTF_ID(func, migrate_enable) | |
19540 | #endif | |
19541 | #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU | |
19542 | BTF_ID(func, rcu_read_unlock_strict) | |
19543 | #endif | |
c11bd046 Y |
19544 | #if defined(CONFIG_DEBUG_PREEMPT) || defined(CONFIG_TRACE_PREEMPT_TOGGLE) |
19545 | BTF_ID(func, preempt_count_add) | |
19546 | BTF_ID(func, preempt_count_sub) | |
19547 | #endif | |
a0c109dc YS |
19548 | #ifdef CONFIG_PREEMPT_RCU |
19549 | BTF_ID(func, __rcu_read_lock) | |
19550 | BTF_ID(func, __rcu_read_unlock) | |
19551 | #endif | |
35e3815f JO |
19552 | BTF_SET_END(btf_id_deny) |
19553 | ||
700e6f85 JO |
19554 | static bool can_be_sleepable(struct bpf_prog *prog) |
19555 | { | |
19556 | if (prog->type == BPF_PROG_TYPE_TRACING) { | |
19557 | switch (prog->expected_attach_type) { | |
19558 | case BPF_TRACE_FENTRY: | |
19559 | case BPF_TRACE_FEXIT: | |
19560 | case BPF_MODIFY_RETURN: | |
19561 | case BPF_TRACE_ITER: | |
19562 | return true; | |
19563 | default: | |
19564 | return false; | |
19565 | } | |
19566 | } | |
19567 | return prog->type == BPF_PROG_TYPE_LSM || | |
1e12d3ef DV |
19568 | prog->type == BPF_PROG_TYPE_KPROBE /* only for uprobes */ || |
19569 | prog->type == BPF_PROG_TYPE_STRUCT_OPS; | |
700e6f85 JO |
19570 | } |
19571 | ||
f7b12b6f THJ |
19572 | static int check_attach_btf_id(struct bpf_verifier_env *env) |
19573 | { | |
19574 | struct bpf_prog *prog = env->prog; | |
3aac1ead | 19575 | struct bpf_prog *tgt_prog = prog->aux->dst_prog; |
f7b12b6f THJ |
19576 | struct bpf_attach_target_info tgt_info = {}; |
19577 | u32 btf_id = prog->aux->attach_btf_id; | |
19578 | struct bpf_trampoline *tr; | |
19579 | int ret; | |
19580 | u64 key; | |
19581 | ||
79a7f8bd AS |
19582 | if (prog->type == BPF_PROG_TYPE_SYSCALL) { |
19583 | if (prog->aux->sleepable) | |
19584 | /* attach_btf_id checked to be zero already */ | |
19585 | return 0; | |
19586 | verbose(env, "Syscall programs can only be sleepable\n"); | |
19587 | return -EINVAL; | |
19588 | } | |
19589 | ||
700e6f85 | 19590 | if (prog->aux->sleepable && !can_be_sleepable(prog)) { |
1e12d3ef | 19591 | verbose(env, "Only fentry/fexit/fmod_ret, lsm, iter, uprobe, and struct_ops programs can be sleepable\n"); |
f7b12b6f THJ |
19592 | return -EINVAL; |
19593 | } | |
19594 | ||
19595 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) | |
19596 | return check_struct_ops_btf_id(env); | |
19597 | ||
19598 | if (prog->type != BPF_PROG_TYPE_TRACING && | |
19599 | prog->type != BPF_PROG_TYPE_LSM && | |
19600 | prog->type != BPF_PROG_TYPE_EXT) | |
19601 | return 0; | |
19602 | ||
19603 | ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info); | |
19604 | if (ret) | |
fec56f58 | 19605 | return ret; |
f7b12b6f THJ |
19606 | |
19607 | if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) { | |
3aac1ead THJ |
19608 | /* to make freplace equivalent to their targets, they need to |
19609 | * inherit env->ops and expected_attach_type for the rest of the | |
19610 | * verification | |
19611 | */ | |
f7b12b6f THJ |
19612 | env->ops = bpf_verifier_ops[tgt_prog->type]; |
19613 | prog->expected_attach_type = tgt_prog->expected_attach_type; | |
19614 | } | |
19615 | ||
19616 | /* store info about the attachment target that will be used later */ | |
19617 | prog->aux->attach_func_proto = tgt_info.tgt_type; | |
19618 | prog->aux->attach_func_name = tgt_info.tgt_name; | |
31bf1dbc | 19619 | prog->aux->mod = tgt_info.tgt_mod; |
f7b12b6f | 19620 | |
4a1e7c0c THJ |
19621 | if (tgt_prog) { |
19622 | prog->aux->saved_dst_prog_type = tgt_prog->type; | |
19623 | prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type; | |
19624 | } | |
19625 | ||
f7b12b6f THJ |
19626 | if (prog->expected_attach_type == BPF_TRACE_RAW_TP) { |
19627 | prog->aux->attach_btf_trace = true; | |
19628 | return 0; | |
19629 | } else if (prog->expected_attach_type == BPF_TRACE_ITER) { | |
19630 | if (!bpf_iter_prog_supported(prog)) | |
19631 | return -EINVAL; | |
19632 | return 0; | |
19633 | } | |
19634 | ||
19635 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
19636 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
19637 | if (ret < 0) | |
19638 | return ret; | |
35e3815f JO |
19639 | } else if (prog->type == BPF_PROG_TYPE_TRACING && |
19640 | btf_id_set_contains(&btf_id_deny, btf_id)) { | |
19641 | return -EINVAL; | |
38207291 | 19642 | } |
f7b12b6f | 19643 | |
22dc4a0f | 19644 | key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id); |
f7b12b6f THJ |
19645 | tr = bpf_trampoline_get(key, &tgt_info); |
19646 | if (!tr) | |
19647 | return -ENOMEM; | |
19648 | ||
3aac1ead | 19649 | prog->aux->dst_trampoline = tr; |
f7b12b6f | 19650 | return 0; |
38207291 MKL |
19651 | } |
19652 | ||
76654e67 AM |
19653 | struct btf *bpf_get_btf_vmlinux(void) |
19654 | { | |
19655 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { | |
19656 | mutex_lock(&bpf_verifier_lock); | |
19657 | if (!btf_vmlinux) | |
19658 | btf_vmlinux = btf_parse_vmlinux(); | |
19659 | mutex_unlock(&bpf_verifier_lock); | |
19660 | } | |
19661 | return btf_vmlinux; | |
19662 | } | |
19663 | ||
47a71c1f | 19664 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr, __u32 uattr_size) |
51580e79 | 19665 | { |
06ee7115 | 19666 | u64 start_time = ktime_get_ns(); |
58e2af8b | 19667 | struct bpf_verifier_env *env; |
bdcab414 AN |
19668 | int i, len, ret = -EINVAL, err; |
19669 | u32 log_true_size; | |
e2ae4ca2 | 19670 | bool is_priv; |
51580e79 | 19671 | |
eba0c929 AB |
19672 | /* no program is valid */ |
19673 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
19674 | return -EINVAL; | |
19675 | ||
58e2af8b | 19676 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
19677 | * allocate/free it every time bpf_check() is called |
19678 | */ | |
58e2af8b | 19679 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
19680 | if (!env) |
19681 | return -ENOMEM; | |
19682 | ||
407958a0 AN |
19683 | env->bt.env = env; |
19684 | ||
9e4c24e7 | 19685 | len = (*prog)->len; |
fad953ce | 19686 | env->insn_aux_data = |
9e4c24e7 | 19687 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
19688 | ret = -ENOMEM; |
19689 | if (!env->insn_aux_data) | |
19690 | goto err_free_env; | |
9e4c24e7 JK |
19691 | for (i = 0; i < len; i++) |
19692 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 19693 | env->prog = *prog; |
00176a34 | 19694 | env->ops = bpf_verifier_ops[env->prog->type]; |
387544bf | 19695 | env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel); |
2c78ee89 | 19696 | is_priv = bpf_capable(); |
0246e64d | 19697 | |
76654e67 | 19698 | bpf_get_btf_vmlinux(); |
8580ac94 | 19699 | |
cbd35700 | 19700 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
19701 | if (!is_priv) |
19702 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 | 19703 | |
bdcab414 AN |
19704 | /* user could have requested verbose verifier output |
19705 | * and supplied buffer to store the verification trace | |
19706 | */ | |
19707 | ret = bpf_vlog_init(&env->log, attr->log_level, | |
19708 | (char __user *) (unsigned long) attr->log_buf, | |
19709 | attr->log_size); | |
19710 | if (ret) | |
19711 | goto err_unlock; | |
1ad2f583 | 19712 | |
0f55f9ed CL |
19713 | mark_verifier_state_clean(env); |
19714 | ||
8580ac94 AS |
19715 | if (IS_ERR(btf_vmlinux)) { |
19716 | /* Either gcc or pahole or kernel are broken. */ | |
19717 | verbose(env, "in-kernel BTF is malformed\n"); | |
19718 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 19719 | goto skip_full_check; |
8580ac94 AS |
19720 | } |
19721 | ||
1ad2f583 DB |
19722 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
19723 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 19724 | env->strict_alignment = true; |
e9ee9efc DM |
19725 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
19726 | env->strict_alignment = false; | |
cbd35700 | 19727 | |
2c78ee89 | 19728 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
01f810ac | 19729 | env->allow_uninit_stack = bpf_allow_uninit_stack(); |
2c78ee89 AS |
19730 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); |
19731 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
19732 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 19733 | |
10d274e8 AS |
19734 | if (is_priv) |
19735 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
19736 | ||
dc2a4ebc | 19737 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 19738 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
19739 | GFP_USER); |
19740 | ret = -ENOMEM; | |
19741 | if (!env->explored_states) | |
19742 | goto skip_full_check; | |
19743 | ||
e6ac2450 MKL |
19744 | ret = add_subprog_and_kfunc(env); |
19745 | if (ret < 0) | |
19746 | goto skip_full_check; | |
19747 | ||
d9762e84 | 19748 | ret = check_subprogs(env); |
475fb78f AS |
19749 | if (ret < 0) |
19750 | goto skip_full_check; | |
19751 | ||
c454a46b | 19752 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
19753 | if (ret < 0) |
19754 | goto skip_full_check; | |
19755 | ||
be8704ff AS |
19756 | ret = check_attach_btf_id(env); |
19757 | if (ret) | |
19758 | goto skip_full_check; | |
19759 | ||
4976b718 HL |
19760 | ret = resolve_pseudo_ldimm64(env); |
19761 | if (ret < 0) | |
19762 | goto skip_full_check; | |
19763 | ||
9d03ebc7 | 19764 | if (bpf_prog_is_offloaded(env->prog->aux)) { |
ceb11679 YZ |
19765 | ret = bpf_prog_offload_verifier_prep(env->prog); |
19766 | if (ret) | |
19767 | goto skip_full_check; | |
19768 | } | |
19769 | ||
d9762e84 MKL |
19770 | ret = check_cfg(env); |
19771 | if (ret < 0) | |
19772 | goto skip_full_check; | |
19773 | ||
51c39bb1 AS |
19774 | ret = do_check_subprogs(env); |
19775 | ret = ret ?: do_check_main(env); | |
cbd35700 | 19776 | |
9d03ebc7 | 19777 | if (ret == 0 && bpf_prog_is_offloaded(env->prog->aux)) |
c941ce9c QM |
19778 | ret = bpf_prog_offload_finalize(env); |
19779 | ||
0246e64d | 19780 | skip_full_check: |
51c39bb1 | 19781 | kvfree(env->explored_states); |
0246e64d | 19782 | |
c131187d | 19783 | if (ret == 0) |
9b38c405 | 19784 | ret = check_max_stack_depth(env); |
c131187d | 19785 | |
9b38c405 | 19786 | /* instruction rewrites happen after this point */ |
1ade2371 EZ |
19787 | if (ret == 0) |
19788 | ret = optimize_bpf_loop(env); | |
19789 | ||
e2ae4ca2 JK |
19790 | if (is_priv) { |
19791 | if (ret == 0) | |
19792 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
19793 | if (ret == 0) |
19794 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
19795 | if (ret == 0) |
19796 | ret = opt_remove_nops(env); | |
52875a04 JK |
19797 | } else { |
19798 | if (ret == 0) | |
19799 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
19800 | } |
19801 | ||
9bac3d6d AS |
19802 | if (ret == 0) |
19803 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
19804 | ret = convert_ctx_accesses(env); | |
19805 | ||
e245c5c6 | 19806 | if (ret == 0) |
e6ac5933 | 19807 | ret = do_misc_fixups(env); |
e245c5c6 | 19808 | |
a4b1d3c1 JW |
19809 | /* do 32-bit optimization after insn patching has done so those patched |
19810 | * insns could be handled correctly. | |
19811 | */ | |
9d03ebc7 | 19812 | if (ret == 0 && !bpf_prog_is_offloaded(env->prog->aux)) { |
d6c2308c JW |
19813 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); |
19814 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
19815 | : false; | |
a4b1d3c1 JW |
19816 | } |
19817 | ||
1ea47e01 AS |
19818 | if (ret == 0) |
19819 | ret = fixup_call_args(env); | |
19820 | ||
06ee7115 AS |
19821 | env->verification_time = ktime_get_ns() - start_time; |
19822 | print_verification_stats(env); | |
aba64c7d | 19823 | env->prog->aux->verified_insns = env->insn_processed; |
06ee7115 | 19824 | |
bdcab414 AN |
19825 | /* preserve original error even if log finalization is successful */ |
19826 | err = bpf_vlog_finalize(&env->log, &log_true_size); | |
19827 | if (err) | |
19828 | ret = err; | |
19829 | ||
47a71c1f AN |
19830 | if (uattr_size >= offsetofend(union bpf_attr, log_true_size) && |
19831 | copy_to_bpfptr_offset(uattr, offsetof(union bpf_attr, log_true_size), | |
bdcab414 | 19832 | &log_true_size, sizeof(log_true_size))) { |
47a71c1f AN |
19833 | ret = -EFAULT; |
19834 | goto err_release_maps; | |
19835 | } | |
cbd35700 | 19836 | |
541c3bad AN |
19837 | if (ret) |
19838 | goto err_release_maps; | |
19839 | ||
19840 | if (env->used_map_cnt) { | |
0246e64d | 19841 | /* if program passed verifier, update used_maps in bpf_prog_info */ |
9bac3d6d AS |
19842 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
19843 | sizeof(env->used_maps[0]), | |
19844 | GFP_KERNEL); | |
0246e64d | 19845 | |
9bac3d6d | 19846 | if (!env->prog->aux->used_maps) { |
0246e64d | 19847 | ret = -ENOMEM; |
a2a7d570 | 19848 | goto err_release_maps; |
0246e64d AS |
19849 | } |
19850 | ||
9bac3d6d | 19851 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 19852 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 19853 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
541c3bad AN |
19854 | } |
19855 | if (env->used_btf_cnt) { | |
19856 | /* if program passed verifier, update used_btfs in bpf_prog_aux */ | |
19857 | env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt, | |
19858 | sizeof(env->used_btfs[0]), | |
19859 | GFP_KERNEL); | |
19860 | if (!env->prog->aux->used_btfs) { | |
19861 | ret = -ENOMEM; | |
19862 | goto err_release_maps; | |
19863 | } | |
0246e64d | 19864 | |
541c3bad AN |
19865 | memcpy(env->prog->aux->used_btfs, env->used_btfs, |
19866 | sizeof(env->used_btfs[0]) * env->used_btf_cnt); | |
19867 | env->prog->aux->used_btf_cnt = env->used_btf_cnt; | |
19868 | } | |
19869 | if (env->used_map_cnt || env->used_btf_cnt) { | |
0246e64d AS |
19870 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic |
19871 | * bpf_ld_imm64 instructions | |
19872 | */ | |
19873 | convert_pseudo_ld_imm64(env); | |
19874 | } | |
cbd35700 | 19875 | |
541c3bad | 19876 | adjust_btf_func(env); |
ba64e7d8 | 19877 | |
a2a7d570 | 19878 | err_release_maps: |
9bac3d6d | 19879 | if (!env->prog->aux->used_maps) |
0246e64d | 19880 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 19881 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
19882 | */ |
19883 | release_maps(env); | |
541c3bad AN |
19884 | if (!env->prog->aux->used_btfs) |
19885 | release_btfs(env); | |
03f87c0b THJ |
19886 | |
19887 | /* extension progs temporarily inherit the attach_type of their targets | |
19888 | for verification purposes, so set it back to zero before returning | |
19889 | */ | |
19890 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
19891 | env->prog->expected_attach_type = 0; | |
19892 | ||
9bac3d6d | 19893 | *prog = env->prog; |
3df126f3 | 19894 | err_unlock: |
45a73c17 AS |
19895 | if (!is_priv) |
19896 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
19897 | vfree(env->insn_aux_data); |
19898 | err_free_env: | |
19899 | kfree(env); | |
51580e79 AS |
19900 | return ret; |
19901 | } |