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> |
51580e79 | 27 | |
f4ac7e0b JK |
28 | #include "disasm.h" |
29 | ||
00176a34 | 30 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
91cc1a99 | 31 | #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ |
00176a34 JK |
32 | [_id] = & _name ## _verifier_ops, |
33 | #define BPF_MAP_TYPE(_id, _ops) | |
f2e10bff | 34 | #define BPF_LINK_TYPE(_id, _name) |
00176a34 JK |
35 | #include <linux/bpf_types.h> |
36 | #undef BPF_PROG_TYPE | |
37 | #undef BPF_MAP_TYPE | |
f2e10bff | 38 | #undef BPF_LINK_TYPE |
00176a34 JK |
39 | }; |
40 | ||
51580e79 AS |
41 | /* bpf_check() is a static code analyzer that walks eBPF program |
42 | * instruction by instruction and updates register/stack state. | |
43 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
44 | * | |
45 | * The first pass is depth-first-search to check that the program is a DAG. | |
46 | * It rejects the following programs: | |
47 | * - larger than BPF_MAXINSNS insns | |
48 | * - if loop is present (detected via back-edge) | |
49 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
50 | * - out of bounds or malformed jumps | |
51 | * The second pass is all possible path descent from the 1st insn. | |
8fb33b60 | 52 | * Since it's analyzing all paths through the program, the length of the |
eba38a96 | 53 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
54 | * insn is less then 4K, but there are too many branches that change stack/regs. |
55 | * Number of 'branches to be analyzed' is limited to 1k | |
56 | * | |
57 | * On entry to each instruction, each register has a type, and the instruction | |
58 | * changes the types of the registers depending on instruction semantics. | |
59 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
60 | * copied to R1. | |
61 | * | |
62 | * All registers are 64-bit. | |
63 | * R0 - return register | |
64 | * R1-R5 argument passing registers | |
65 | * R6-R9 callee saved registers | |
66 | * R10 - frame pointer read-only | |
67 | * | |
68 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
69 | * and has type PTR_TO_CTX. | |
70 | * | |
71 | * Verifier tracks arithmetic operations on pointers in case: | |
72 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
73 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
74 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
75 | * and 2nd arithmetic instruction is pattern matched to recognize | |
76 | * that it wants to construct a pointer to some element within stack. | |
77 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
78 | * (and -20 constant is saved for further stack bounds checking). | |
79 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
80 | * | |
f1174f77 | 81 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 82 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 83 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
84 | * |
85 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
86 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
87 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
88 | * |
89 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
90 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
91 | * | |
92 | * registers used to pass values to function calls are checked against | |
93 | * function argument constraints. | |
94 | * | |
95 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
96 | * It means that the register type passed to this function must be | |
97 | * PTR_TO_STACK and it will be used inside the function as | |
98 | * 'pointer to map element key' | |
99 | * | |
100 | * For example the argument constraints for bpf_map_lookup_elem(): | |
101 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
102 | * .arg1_type = ARG_CONST_MAP_PTR, | |
103 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
104 | * | |
105 | * ret_type says that this function returns 'pointer to map elem value or null' | |
106 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
107 | * 2nd argument should be a pointer to stack, which will be used inside | |
108 | * the helper function as a pointer to map element key. | |
109 | * | |
110 | * On the kernel side the helper function looks like: | |
111 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
112 | * { | |
113 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
114 | * void *key = (void *) (unsigned long) r2; | |
115 | * void *value; | |
116 | * | |
117 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
118 | * [key, key + map->key_size) bytes are valid and were initialized on | |
119 | * the stack of eBPF program. | |
120 | * } | |
121 | * | |
122 | * Corresponding eBPF program may look like: | |
123 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
124 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
125 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
126 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
127 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
128 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
129 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
130 | * | |
131 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
132 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
133 | * and were initialized prior to this call. | |
134 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
135 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
136 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
8fb33b60 | 137 | * returns either pointer to map value or NULL. |
51580e79 AS |
138 | * |
139 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
140 | * insn, the register holding that pointer in the true branch changes state to | |
141 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
142 | * branch. See check_cond_jmp_op(). | |
143 | * | |
144 | * After the call R0 is set to return type of the function and registers R1-R5 | |
145 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
146 | * |
147 | * The following reference types represent a potential reference to a kernel | |
148 | * resource which, after first being allocated, must be checked and freed by | |
149 | * the BPF program: | |
150 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
151 | * | |
152 | * When the verifier sees a helper call return a reference type, it allocates a | |
153 | * pointer id for the reference and stores it in the current function state. | |
154 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
155 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
156 | * passes through a NULL-check conditional. For the branch wherein the state is | |
157 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
158 | * |
159 | * For each helper function that allocates a reference, such as | |
160 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
161 | * bpf_sk_release(). When a reference type passes into the release function, | |
162 | * the verifier also releases the reference. If any unchecked or unreleased | |
163 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
164 | */ |
165 | ||
17a52670 | 166 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 167 | struct bpf_verifier_stack_elem { |
17a52670 AS |
168 | /* verifer state is 'st' |
169 | * before processing instruction 'insn_idx' | |
170 | * and after processing instruction 'prev_insn_idx' | |
171 | */ | |
58e2af8b | 172 | struct bpf_verifier_state st; |
17a52670 AS |
173 | int insn_idx; |
174 | int prev_insn_idx; | |
58e2af8b | 175 | struct bpf_verifier_stack_elem *next; |
6f8a57cc AN |
176 | /* length of verifier log at the time this state was pushed on stack */ |
177 | u32 log_pos; | |
cbd35700 AS |
178 | }; |
179 | ||
b285fcb7 | 180 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 181 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 182 | |
d2e4c1e6 DB |
183 | #define BPF_MAP_KEY_POISON (1ULL << 63) |
184 | #define BPF_MAP_KEY_SEEN (1ULL << 62) | |
185 | ||
c93552c4 DB |
186 | #define BPF_MAP_PTR_UNPRIV 1UL |
187 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
188 | POISON_POINTER_DELTA)) | |
189 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
190 | ||
bc34dee6 JK |
191 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx); |
192 | static int release_reference(struct bpf_verifier_env *env, int ref_obj_id); | |
193 | ||
c93552c4 DB |
194 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) |
195 | { | |
d2e4c1e6 | 196 | return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON; |
c93552c4 DB |
197 | } |
198 | ||
199 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
200 | { | |
d2e4c1e6 | 201 | return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV; |
c93552c4 DB |
202 | } |
203 | ||
204 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
205 | const struct bpf_map *map, bool unpriv) | |
206 | { | |
207 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
208 | unpriv |= bpf_map_ptr_unpriv(aux); | |
d2e4c1e6 DB |
209 | aux->map_ptr_state = (unsigned long)map | |
210 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
211 | } | |
212 | ||
213 | static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux) | |
214 | { | |
215 | return aux->map_key_state & BPF_MAP_KEY_POISON; | |
216 | } | |
217 | ||
218 | static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux) | |
219 | { | |
220 | return !(aux->map_key_state & BPF_MAP_KEY_SEEN); | |
221 | } | |
222 | ||
223 | static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux) | |
224 | { | |
225 | return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON); | |
226 | } | |
227 | ||
228 | static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state) | |
229 | { | |
230 | bool poisoned = bpf_map_key_poisoned(aux); | |
231 | ||
232 | aux->map_key_state = state | BPF_MAP_KEY_SEEN | | |
233 | (poisoned ? BPF_MAP_KEY_POISON : 0ULL); | |
c93552c4 | 234 | } |
fad73a1a | 235 | |
23a2d70c YS |
236 | static bool bpf_pseudo_call(const struct bpf_insn *insn) |
237 | { | |
238 | return insn->code == (BPF_JMP | BPF_CALL) && | |
239 | insn->src_reg == BPF_PSEUDO_CALL; | |
240 | } | |
241 | ||
e6ac2450 MKL |
242 | static bool bpf_pseudo_kfunc_call(const struct bpf_insn *insn) |
243 | { | |
244 | return insn->code == (BPF_JMP | BPF_CALL) && | |
245 | insn->src_reg == BPF_PSEUDO_KFUNC_CALL; | |
246 | } | |
247 | ||
33ff9823 DB |
248 | struct bpf_call_arg_meta { |
249 | struct bpf_map *map_ptr; | |
435faee1 | 250 | bool raw_mode; |
36bbef52 | 251 | bool pkt_access; |
8f14852e | 252 | u8 release_regno; |
435faee1 DB |
253 | int regno; |
254 | int access_size; | |
457f4436 | 255 | int mem_size; |
10060503 | 256 | u64 msize_max_value; |
1b986589 | 257 | int ref_obj_id; |
3e8ce298 | 258 | int map_uid; |
d83525ca | 259 | int func_id; |
22dc4a0f | 260 | struct btf *btf; |
eaa6bcb7 | 261 | u32 btf_id; |
22dc4a0f | 262 | struct btf *ret_btf; |
eaa6bcb7 | 263 | u32 ret_btf_id; |
69c087ba | 264 | u32 subprogno; |
aa3496ac | 265 | struct btf_field *kptr_field; |
97e03f52 | 266 | u8 uninit_dynptr_regno; |
33ff9823 DB |
267 | }; |
268 | ||
8580ac94 AS |
269 | struct btf *btf_vmlinux; |
270 | ||
cbd35700 AS |
271 | static DEFINE_MUTEX(bpf_verifier_lock); |
272 | ||
d9762e84 MKL |
273 | static const struct bpf_line_info * |
274 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
275 | { | |
276 | const struct bpf_line_info *linfo; | |
277 | const struct bpf_prog *prog; | |
278 | u32 i, nr_linfo; | |
279 | ||
280 | prog = env->prog; | |
281 | nr_linfo = prog->aux->nr_linfo; | |
282 | ||
283 | if (!nr_linfo || insn_off >= prog->len) | |
284 | return NULL; | |
285 | ||
286 | linfo = prog->aux->linfo; | |
287 | for (i = 1; i < nr_linfo; i++) | |
288 | if (insn_off < linfo[i].insn_off) | |
289 | break; | |
290 | ||
291 | return &linfo[i - 1]; | |
292 | } | |
293 | ||
77d2e05a MKL |
294 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
295 | va_list args) | |
cbd35700 | 296 | { |
a2a7d570 | 297 | unsigned int n; |
cbd35700 | 298 | |
a2a7d570 | 299 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
300 | |
301 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
302 | "verifier log line truncated - local buffer too short\n"); | |
303 | ||
8580ac94 | 304 | if (log->level == BPF_LOG_KERNEL) { |
436d404c HT |
305 | bool newline = n > 0 && log->kbuf[n - 1] == '\n'; |
306 | ||
307 | pr_err("BPF: %s%s", log->kbuf, newline ? "" : "\n"); | |
8580ac94 AS |
308 | return; |
309 | } | |
436d404c HT |
310 | |
311 | n = min(log->len_total - log->len_used - 1, n); | |
312 | log->kbuf[n] = '\0'; | |
a2a7d570 JK |
313 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) |
314 | log->len_used += n; | |
315 | else | |
316 | log->ubuf = NULL; | |
cbd35700 | 317 | } |
abe08840 | 318 | |
6f8a57cc AN |
319 | static void bpf_vlog_reset(struct bpf_verifier_log *log, u32 new_pos) |
320 | { | |
321 | char zero = 0; | |
322 | ||
323 | if (!bpf_verifier_log_needed(log)) | |
324 | return; | |
325 | ||
326 | log->len_used = new_pos; | |
327 | if (put_user(zero, log->ubuf + new_pos)) | |
328 | log->ubuf = NULL; | |
329 | } | |
330 | ||
abe08840 JO |
331 | /* log_level controls verbosity level of eBPF verifier. |
332 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
333 | * so the user can figure out what's wrong with the program | |
430e68d1 | 334 | */ |
abe08840 JO |
335 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
336 | const char *fmt, ...) | |
337 | { | |
338 | va_list args; | |
339 | ||
77d2e05a MKL |
340 | if (!bpf_verifier_log_needed(&env->log)) |
341 | return; | |
342 | ||
abe08840 | 343 | va_start(args, fmt); |
77d2e05a | 344 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
345 | va_end(args); |
346 | } | |
347 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
348 | ||
349 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
350 | { | |
77d2e05a | 351 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
352 | va_list args; |
353 | ||
77d2e05a MKL |
354 | if (!bpf_verifier_log_needed(&env->log)) |
355 | return; | |
356 | ||
abe08840 | 357 | va_start(args, fmt); |
77d2e05a | 358 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
359 | va_end(args); |
360 | } | |
cbd35700 | 361 | |
9e15db66 AS |
362 | __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, |
363 | const char *fmt, ...) | |
364 | { | |
365 | va_list args; | |
366 | ||
367 | if (!bpf_verifier_log_needed(log)) | |
368 | return; | |
369 | ||
370 | va_start(args, fmt); | |
371 | bpf_verifier_vlog(log, fmt, args); | |
372 | va_end(args); | |
373 | } | |
84c6ac41 | 374 | EXPORT_SYMBOL_GPL(bpf_log); |
9e15db66 | 375 | |
d9762e84 MKL |
376 | static const char *ltrim(const char *s) |
377 | { | |
378 | while (isspace(*s)) | |
379 | s++; | |
380 | ||
381 | return s; | |
382 | } | |
383 | ||
384 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
385 | u32 insn_off, | |
386 | const char *prefix_fmt, ...) | |
387 | { | |
388 | const struct bpf_line_info *linfo; | |
389 | ||
390 | if (!bpf_verifier_log_needed(&env->log)) | |
391 | return; | |
392 | ||
393 | linfo = find_linfo(env, insn_off); | |
394 | if (!linfo || linfo == env->prev_linfo) | |
395 | return; | |
396 | ||
397 | if (prefix_fmt) { | |
398 | va_list args; | |
399 | ||
400 | va_start(args, prefix_fmt); | |
401 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
402 | va_end(args); | |
403 | } | |
404 | ||
405 | verbose(env, "%s\n", | |
406 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
407 | linfo->line_off))); | |
408 | ||
409 | env->prev_linfo = linfo; | |
410 | } | |
411 | ||
bc2591d6 YS |
412 | static void verbose_invalid_scalar(struct bpf_verifier_env *env, |
413 | struct bpf_reg_state *reg, | |
414 | struct tnum *range, const char *ctx, | |
415 | const char *reg_name) | |
416 | { | |
417 | char tn_buf[48]; | |
418 | ||
419 | verbose(env, "At %s the register %s ", ctx, reg_name); | |
420 | if (!tnum_is_unknown(reg->var_off)) { | |
421 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
422 | verbose(env, "has value %s", tn_buf); | |
423 | } else { | |
424 | verbose(env, "has unknown scalar value"); | |
425 | } | |
426 | tnum_strn(tn_buf, sizeof(tn_buf), *range); | |
427 | verbose(env, " should have been in %s\n", tn_buf); | |
428 | } | |
429 | ||
de8f3a83 DB |
430 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
431 | { | |
0c9a7a7e | 432 | type = base_type(type); |
de8f3a83 DB |
433 | return type == PTR_TO_PACKET || |
434 | type == PTR_TO_PACKET_META; | |
435 | } | |
436 | ||
46f8bc92 MKL |
437 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
438 | { | |
439 | return type == PTR_TO_SOCKET || | |
655a51e5 | 440 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
441 | type == PTR_TO_TCP_SOCK || |
442 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
443 | } |
444 | ||
cac616db JF |
445 | static bool reg_type_not_null(enum bpf_reg_type type) |
446 | { | |
447 | return type == PTR_TO_SOCKET || | |
448 | type == PTR_TO_TCP_SOCK || | |
449 | type == PTR_TO_MAP_VALUE || | |
69c087ba | 450 | type == PTR_TO_MAP_KEY || |
01c66c48 | 451 | type == PTR_TO_SOCK_COMMON; |
cac616db JF |
452 | } |
453 | ||
4e814da0 KKD |
454 | static struct btf_record *reg_btf_record(const struct bpf_reg_state *reg) |
455 | { | |
456 | struct btf_record *rec = NULL; | |
457 | struct btf_struct_meta *meta; | |
458 | ||
459 | if (reg->type == PTR_TO_MAP_VALUE) { | |
460 | rec = reg->map_ptr->record; | |
461 | } else if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
462 | meta = btf_find_struct_meta(reg->btf, reg->btf_id); | |
463 | if (meta) | |
464 | rec = meta->record; | |
465 | } | |
466 | return rec; | |
467 | } | |
468 | ||
d83525ca AS |
469 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
470 | { | |
4e814da0 | 471 | return btf_record_has_field(reg_btf_record(reg), BPF_SPIN_LOCK); |
cba368c1 MKL |
472 | } |
473 | ||
20b2aff4 HL |
474 | static bool type_is_rdonly_mem(u32 type) |
475 | { | |
476 | return type & MEM_RDONLY; | |
cba368c1 MKL |
477 | } |
478 | ||
48946bd6 | 479 | static bool type_may_be_null(u32 type) |
fd1b0d60 | 480 | { |
48946bd6 | 481 | return type & PTR_MAYBE_NULL; |
fd1b0d60 LB |
482 | } |
483 | ||
64d85290 JS |
484 | static bool is_acquire_function(enum bpf_func_id func_id, |
485 | const struct bpf_map *map) | |
486 | { | |
487 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
488 | ||
489 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
490 | func_id == BPF_FUNC_sk_lookup_udp || | |
457f4436 | 491 | func_id == BPF_FUNC_skc_lookup_tcp || |
c0a5a21c KKD |
492 | func_id == BPF_FUNC_ringbuf_reserve || |
493 | func_id == BPF_FUNC_kptr_xchg) | |
64d85290 JS |
494 | return true; |
495 | ||
496 | if (func_id == BPF_FUNC_map_lookup_elem && | |
497 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
498 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
499 | return true; | |
500 | ||
501 | return false; | |
46f8bc92 MKL |
502 | } |
503 | ||
1b986589 MKL |
504 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
505 | { | |
506 | return func_id == BPF_FUNC_tcp_sock || | |
1df8f55a MKL |
507 | func_id == BPF_FUNC_sk_fullsock || |
508 | func_id == BPF_FUNC_skc_to_tcp_sock || | |
509 | func_id == BPF_FUNC_skc_to_tcp6_sock || | |
510 | func_id == BPF_FUNC_skc_to_udp6_sock || | |
3bc253c2 | 511 | func_id == BPF_FUNC_skc_to_mptcp_sock || |
1df8f55a MKL |
512 | func_id == BPF_FUNC_skc_to_tcp_timewait_sock || |
513 | func_id == BPF_FUNC_skc_to_tcp_request_sock; | |
1b986589 MKL |
514 | } |
515 | ||
88374342 | 516 | static bool is_dynptr_ref_function(enum bpf_func_id func_id) |
b2d8ef19 DM |
517 | { |
518 | return func_id == BPF_FUNC_dynptr_data; | |
519 | } | |
520 | ||
be2ef816 AN |
521 | static bool is_callback_calling_function(enum bpf_func_id func_id) |
522 | { | |
523 | return func_id == BPF_FUNC_for_each_map_elem || | |
524 | func_id == BPF_FUNC_timer_set_callback || | |
525 | func_id == BPF_FUNC_find_vma || | |
526 | func_id == BPF_FUNC_loop || | |
527 | func_id == BPF_FUNC_user_ringbuf_drain; | |
528 | } | |
529 | ||
b2d8ef19 DM |
530 | static bool helper_multiple_ref_obj_use(enum bpf_func_id func_id, |
531 | const struct bpf_map *map) | |
532 | { | |
533 | int ref_obj_uses = 0; | |
534 | ||
535 | if (is_ptr_cast_function(func_id)) | |
536 | ref_obj_uses++; | |
537 | if (is_acquire_function(func_id, map)) | |
538 | ref_obj_uses++; | |
88374342 | 539 | if (is_dynptr_ref_function(func_id)) |
b2d8ef19 DM |
540 | ref_obj_uses++; |
541 | ||
542 | return ref_obj_uses > 1; | |
543 | } | |
544 | ||
39491867 BJ |
545 | static bool is_cmpxchg_insn(const struct bpf_insn *insn) |
546 | { | |
547 | return BPF_CLASS(insn->code) == BPF_STX && | |
548 | BPF_MODE(insn->code) == BPF_ATOMIC && | |
549 | insn->imm == BPF_CMPXCHG; | |
550 | } | |
551 | ||
c25b2ae1 HL |
552 | /* string representation of 'enum bpf_reg_type' |
553 | * | |
554 | * Note that reg_type_str() can not appear more than once in a single verbose() | |
555 | * statement. | |
556 | */ | |
557 | static const char *reg_type_str(struct bpf_verifier_env *env, | |
558 | enum bpf_reg_type type) | |
559 | { | |
ef66c547 | 560 | char postfix[16] = {0}, prefix[64] = {0}; |
c25b2ae1 HL |
561 | static const char * const str[] = { |
562 | [NOT_INIT] = "?", | |
7df5072c | 563 | [SCALAR_VALUE] = "scalar", |
c25b2ae1 HL |
564 | [PTR_TO_CTX] = "ctx", |
565 | [CONST_PTR_TO_MAP] = "map_ptr", | |
566 | [PTR_TO_MAP_VALUE] = "map_value", | |
567 | [PTR_TO_STACK] = "fp", | |
568 | [PTR_TO_PACKET] = "pkt", | |
569 | [PTR_TO_PACKET_META] = "pkt_meta", | |
570 | [PTR_TO_PACKET_END] = "pkt_end", | |
571 | [PTR_TO_FLOW_KEYS] = "flow_keys", | |
572 | [PTR_TO_SOCKET] = "sock", | |
573 | [PTR_TO_SOCK_COMMON] = "sock_common", | |
574 | [PTR_TO_TCP_SOCK] = "tcp_sock", | |
575 | [PTR_TO_TP_BUFFER] = "tp_buffer", | |
576 | [PTR_TO_XDP_SOCK] = "xdp_sock", | |
577 | [PTR_TO_BTF_ID] = "ptr_", | |
c25b2ae1 | 578 | [PTR_TO_MEM] = "mem", |
20b2aff4 | 579 | [PTR_TO_BUF] = "buf", |
c25b2ae1 HL |
580 | [PTR_TO_FUNC] = "func", |
581 | [PTR_TO_MAP_KEY] = "map_key", | |
20571567 | 582 | [PTR_TO_DYNPTR] = "dynptr_ptr", |
c25b2ae1 HL |
583 | }; |
584 | ||
585 | if (type & PTR_MAYBE_NULL) { | |
5844101a | 586 | if (base_type(type) == PTR_TO_BTF_ID) |
c25b2ae1 HL |
587 | strncpy(postfix, "or_null_", 16); |
588 | else | |
589 | strncpy(postfix, "_or_null", 16); | |
590 | } | |
591 | ||
3f00c523 | 592 | snprintf(prefix, sizeof(prefix), "%s%s%s%s%s%s", |
ef66c547 DV |
593 | type & MEM_RDONLY ? "rdonly_" : "", |
594 | type & MEM_RINGBUF ? "ringbuf_" : "", | |
595 | type & MEM_USER ? "user_" : "", | |
596 | type & MEM_PERCPU ? "percpu_" : "", | |
3f00c523 DV |
597 | type & PTR_UNTRUSTED ? "untrusted_" : "", |
598 | type & PTR_TRUSTED ? "trusted_" : "" | |
ef66c547 | 599 | ); |
20b2aff4 HL |
600 | |
601 | snprintf(env->type_str_buf, TYPE_STR_BUF_LEN, "%s%s%s", | |
602 | prefix, str[base_type(type)], postfix); | |
c25b2ae1 HL |
603 | return env->type_str_buf; |
604 | } | |
17a52670 | 605 | |
8efea21d EC |
606 | static char slot_type_char[] = { |
607 | [STACK_INVALID] = '?', | |
608 | [STACK_SPILL] = 'r', | |
609 | [STACK_MISC] = 'm', | |
610 | [STACK_ZERO] = '0', | |
97e03f52 | 611 | [STACK_DYNPTR] = 'd', |
8efea21d EC |
612 | }; |
613 | ||
4e92024a AS |
614 | static void print_liveness(struct bpf_verifier_env *env, |
615 | enum bpf_reg_liveness live) | |
616 | { | |
9242b5f5 | 617 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
618 | verbose(env, "_"); |
619 | if (live & REG_LIVE_READ) | |
620 | verbose(env, "r"); | |
621 | if (live & REG_LIVE_WRITTEN) | |
622 | verbose(env, "w"); | |
9242b5f5 AS |
623 | if (live & REG_LIVE_DONE) |
624 | verbose(env, "D"); | |
4e92024a AS |
625 | } |
626 | ||
97e03f52 JK |
627 | static int get_spi(s32 off) |
628 | { | |
629 | return (-off - 1) / BPF_REG_SIZE; | |
630 | } | |
631 | ||
632 | static bool is_spi_bounds_valid(struct bpf_func_state *state, int spi, int nr_slots) | |
633 | { | |
634 | int allocated_slots = state->allocated_stack / BPF_REG_SIZE; | |
635 | ||
636 | /* We need to check that slots between [spi - nr_slots + 1, spi] are | |
637 | * within [0, allocated_stack). | |
638 | * | |
639 | * Please note that the spi grows downwards. For example, a dynptr | |
640 | * takes the size of two stack slots; the first slot will be at | |
641 | * spi and the second slot will be at spi - 1. | |
642 | */ | |
643 | return spi - nr_slots + 1 >= 0 && spi < allocated_slots; | |
644 | } | |
645 | ||
f4d7e40a AS |
646 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
647 | const struct bpf_reg_state *reg) | |
648 | { | |
649 | struct bpf_verifier_state *cur = env->cur_state; | |
650 | ||
651 | return cur->frame[reg->frameno]; | |
652 | } | |
653 | ||
22dc4a0f | 654 | static const char *kernel_type_name(const struct btf* btf, u32 id) |
9e15db66 | 655 | { |
22dc4a0f | 656 | return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off); |
9e15db66 AS |
657 | } |
658 | ||
0f55f9ed CL |
659 | static void mark_reg_scratched(struct bpf_verifier_env *env, u32 regno) |
660 | { | |
661 | env->scratched_regs |= 1U << regno; | |
662 | } | |
663 | ||
664 | static void mark_stack_slot_scratched(struct bpf_verifier_env *env, u32 spi) | |
665 | { | |
343e5375 | 666 | env->scratched_stack_slots |= 1ULL << spi; |
0f55f9ed CL |
667 | } |
668 | ||
669 | static bool reg_scratched(const struct bpf_verifier_env *env, u32 regno) | |
670 | { | |
671 | return (env->scratched_regs >> regno) & 1; | |
672 | } | |
673 | ||
674 | static bool stack_slot_scratched(const struct bpf_verifier_env *env, u64 regno) | |
675 | { | |
676 | return (env->scratched_stack_slots >> regno) & 1; | |
677 | } | |
678 | ||
679 | static bool verifier_state_scratched(const struct bpf_verifier_env *env) | |
680 | { | |
681 | return env->scratched_regs || env->scratched_stack_slots; | |
682 | } | |
683 | ||
684 | static void mark_verifier_state_clean(struct bpf_verifier_env *env) | |
685 | { | |
686 | env->scratched_regs = 0U; | |
343e5375 | 687 | env->scratched_stack_slots = 0ULL; |
0f55f9ed CL |
688 | } |
689 | ||
690 | /* Used for printing the entire verifier state. */ | |
691 | static void mark_verifier_state_scratched(struct bpf_verifier_env *env) | |
692 | { | |
693 | env->scratched_regs = ~0U; | |
343e5375 | 694 | env->scratched_stack_slots = ~0ULL; |
0f55f9ed CL |
695 | } |
696 | ||
97e03f52 JK |
697 | static enum bpf_dynptr_type arg_to_dynptr_type(enum bpf_arg_type arg_type) |
698 | { | |
699 | switch (arg_type & DYNPTR_TYPE_FLAG_MASK) { | |
700 | case DYNPTR_TYPE_LOCAL: | |
701 | return BPF_DYNPTR_TYPE_LOCAL; | |
bc34dee6 JK |
702 | case DYNPTR_TYPE_RINGBUF: |
703 | return BPF_DYNPTR_TYPE_RINGBUF; | |
97e03f52 JK |
704 | default: |
705 | return BPF_DYNPTR_TYPE_INVALID; | |
706 | } | |
707 | } | |
708 | ||
bc34dee6 JK |
709 | static bool dynptr_type_refcounted(enum bpf_dynptr_type type) |
710 | { | |
711 | return type == BPF_DYNPTR_TYPE_RINGBUF; | |
712 | } | |
713 | ||
97e03f52 JK |
714 | static int mark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
715 | enum bpf_arg_type arg_type, int insn_idx) | |
716 | { | |
717 | struct bpf_func_state *state = func(env, reg); | |
718 | enum bpf_dynptr_type type; | |
bc34dee6 | 719 | int spi, i, id; |
97e03f52 JK |
720 | |
721 | spi = get_spi(reg->off); | |
722 | ||
723 | if (!is_spi_bounds_valid(state, spi, BPF_DYNPTR_NR_SLOTS)) | |
724 | return -EINVAL; | |
725 | ||
726 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
727 | state->stack[spi].slot_type[i] = STACK_DYNPTR; | |
728 | state->stack[spi - 1].slot_type[i] = STACK_DYNPTR; | |
729 | } | |
730 | ||
731 | type = arg_to_dynptr_type(arg_type); | |
732 | if (type == BPF_DYNPTR_TYPE_INVALID) | |
733 | return -EINVAL; | |
734 | ||
735 | state->stack[spi].spilled_ptr.dynptr.first_slot = true; | |
736 | state->stack[spi].spilled_ptr.dynptr.type = type; | |
737 | state->stack[spi - 1].spilled_ptr.dynptr.type = type; | |
738 | ||
bc34dee6 JK |
739 | if (dynptr_type_refcounted(type)) { |
740 | /* The id is used to track proper releasing */ | |
741 | id = acquire_reference_state(env, insn_idx); | |
742 | if (id < 0) | |
743 | return id; | |
744 | ||
745 | state->stack[spi].spilled_ptr.id = id; | |
746 | state->stack[spi - 1].spilled_ptr.id = id; | |
747 | } | |
748 | ||
97e03f52 JK |
749 | return 0; |
750 | } | |
751 | ||
752 | static int unmark_stack_slots_dynptr(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
753 | { | |
754 | struct bpf_func_state *state = func(env, reg); | |
755 | int spi, i; | |
756 | ||
757 | spi = get_spi(reg->off); | |
758 | ||
759 | if (!is_spi_bounds_valid(state, spi, BPF_DYNPTR_NR_SLOTS)) | |
760 | return -EINVAL; | |
761 | ||
762 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
763 | state->stack[spi].slot_type[i] = STACK_INVALID; | |
764 | state->stack[spi - 1].slot_type[i] = STACK_INVALID; | |
765 | } | |
766 | ||
bc34dee6 JK |
767 | /* Invalidate any slices associated with this dynptr */ |
768 | if (dynptr_type_refcounted(state->stack[spi].spilled_ptr.dynptr.type)) { | |
769 | release_reference(env, state->stack[spi].spilled_ptr.id); | |
770 | state->stack[spi].spilled_ptr.id = 0; | |
771 | state->stack[spi - 1].spilled_ptr.id = 0; | |
772 | } | |
773 | ||
97e03f52 JK |
774 | state->stack[spi].spilled_ptr.dynptr.first_slot = false; |
775 | state->stack[spi].spilled_ptr.dynptr.type = 0; | |
776 | state->stack[spi - 1].spilled_ptr.dynptr.type = 0; | |
777 | ||
778 | return 0; | |
779 | } | |
780 | ||
781 | static bool is_dynptr_reg_valid_uninit(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
782 | { | |
783 | struct bpf_func_state *state = func(env, reg); | |
784 | int spi = get_spi(reg->off); | |
785 | int i; | |
786 | ||
787 | if (!is_spi_bounds_valid(state, spi, BPF_DYNPTR_NR_SLOTS)) | |
788 | return true; | |
789 | ||
790 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
791 | if (state->stack[spi].slot_type[i] == STACK_DYNPTR || | |
792 | state->stack[spi - 1].slot_type[i] == STACK_DYNPTR) | |
793 | return false; | |
794 | } | |
795 | ||
796 | return true; | |
797 | } | |
798 | ||
b8d31762 RS |
799 | bool is_dynptr_reg_valid_init(struct bpf_verifier_env *env, |
800 | struct bpf_reg_state *reg) | |
97e03f52 JK |
801 | { |
802 | struct bpf_func_state *state = func(env, reg); | |
803 | int spi = get_spi(reg->off); | |
804 | int i; | |
805 | ||
806 | if (!is_spi_bounds_valid(state, spi, BPF_DYNPTR_NR_SLOTS) || | |
807 | !state->stack[spi].spilled_ptr.dynptr.first_slot) | |
808 | return false; | |
809 | ||
810 | for (i = 0; i < BPF_REG_SIZE; i++) { | |
811 | if (state->stack[spi].slot_type[i] != STACK_DYNPTR || | |
812 | state->stack[spi - 1].slot_type[i] != STACK_DYNPTR) | |
813 | return false; | |
814 | } | |
815 | ||
e9e315b4 RS |
816 | return true; |
817 | } | |
818 | ||
b8d31762 RS |
819 | bool is_dynptr_type_expected(struct bpf_verifier_env *env, |
820 | struct bpf_reg_state *reg, | |
821 | enum bpf_arg_type arg_type) | |
e9e315b4 RS |
822 | { |
823 | struct bpf_func_state *state = func(env, reg); | |
824 | enum bpf_dynptr_type dynptr_type; | |
825 | int spi = get_spi(reg->off); | |
826 | ||
97e03f52 JK |
827 | /* ARG_PTR_TO_DYNPTR takes any type of dynptr */ |
828 | if (arg_type == ARG_PTR_TO_DYNPTR) | |
829 | return true; | |
830 | ||
e9e315b4 RS |
831 | dynptr_type = arg_to_dynptr_type(arg_type); |
832 | ||
833 | return state->stack[spi].spilled_ptr.dynptr.type == dynptr_type; | |
97e03f52 JK |
834 | } |
835 | ||
27113c59 MKL |
836 | /* The reg state of a pointer or a bounded scalar was saved when |
837 | * it was spilled to the stack. | |
838 | */ | |
839 | static bool is_spilled_reg(const struct bpf_stack_state *stack) | |
840 | { | |
841 | return stack->slot_type[BPF_REG_SIZE - 1] == STACK_SPILL; | |
842 | } | |
843 | ||
354e8f19 MKL |
844 | static void scrub_spilled_slot(u8 *stype) |
845 | { | |
846 | if (*stype != STACK_INVALID) | |
847 | *stype = STACK_MISC; | |
848 | } | |
849 | ||
61bd5218 | 850 | static void print_verifier_state(struct bpf_verifier_env *env, |
0f55f9ed CL |
851 | const struct bpf_func_state *state, |
852 | bool print_all) | |
17a52670 | 853 | { |
f4d7e40a | 854 | const struct bpf_reg_state *reg; |
17a52670 AS |
855 | enum bpf_reg_type t; |
856 | int i; | |
857 | ||
f4d7e40a AS |
858 | if (state->frameno) |
859 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 860 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
861 | reg = &state->regs[i]; |
862 | t = reg->type; | |
17a52670 AS |
863 | if (t == NOT_INIT) |
864 | continue; | |
0f55f9ed CL |
865 | if (!print_all && !reg_scratched(env, i)) |
866 | continue; | |
4e92024a AS |
867 | verbose(env, " R%d", i); |
868 | print_liveness(env, reg->live); | |
7df5072c | 869 | verbose(env, "="); |
b5dc0163 AS |
870 | if (t == SCALAR_VALUE && reg->precise) |
871 | verbose(env, "P"); | |
f1174f77 EC |
872 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
873 | tnum_is_const(reg->var_off)) { | |
874 | /* reg->off should be 0 for SCALAR_VALUE */ | |
7df5072c | 875 | verbose(env, "%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
61bd5218 | 876 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 877 | } else { |
7df5072c ML |
878 | const char *sep = ""; |
879 | ||
880 | verbose(env, "%s", reg_type_str(env, t)); | |
5844101a | 881 | if (base_type(t) == PTR_TO_BTF_ID) |
22dc4a0f | 882 | verbose(env, "%s", kernel_type_name(reg->btf, reg->btf_id)); |
7df5072c ML |
883 | verbose(env, "("); |
884 | /* | |
885 | * _a stands for append, was shortened to avoid multiline statements below. | |
886 | * This macro is used to output a comma separated list of attributes. | |
887 | */ | |
888 | #define verbose_a(fmt, ...) ({ verbose(env, "%s" fmt, sep, __VA_ARGS__); sep = ","; }) | |
889 | ||
890 | if (reg->id) | |
891 | verbose_a("id=%d", reg->id); | |
a28ace78 | 892 | if (reg->ref_obj_id) |
7df5072c | 893 | verbose_a("ref_obj_id=%d", reg->ref_obj_id); |
f1174f77 | 894 | if (t != SCALAR_VALUE) |
7df5072c | 895 | verbose_a("off=%d", reg->off); |
de8f3a83 | 896 | if (type_is_pkt_pointer(t)) |
7df5072c | 897 | verbose_a("r=%d", reg->range); |
c25b2ae1 HL |
898 | else if (base_type(t) == CONST_PTR_TO_MAP || |
899 | base_type(t) == PTR_TO_MAP_KEY || | |
900 | base_type(t) == PTR_TO_MAP_VALUE) | |
7df5072c ML |
901 | verbose_a("ks=%d,vs=%d", |
902 | reg->map_ptr->key_size, | |
903 | reg->map_ptr->value_size); | |
7d1238f2 EC |
904 | if (tnum_is_const(reg->var_off)) { |
905 | /* Typically an immediate SCALAR_VALUE, but | |
906 | * could be a pointer whose offset is too big | |
907 | * for reg->off | |
908 | */ | |
7df5072c | 909 | verbose_a("imm=%llx", reg->var_off.value); |
7d1238f2 EC |
910 | } else { |
911 | if (reg->smin_value != reg->umin_value && | |
912 | reg->smin_value != S64_MIN) | |
7df5072c | 913 | verbose_a("smin=%lld", (long long)reg->smin_value); |
7d1238f2 EC |
914 | if (reg->smax_value != reg->umax_value && |
915 | reg->smax_value != S64_MAX) | |
7df5072c | 916 | verbose_a("smax=%lld", (long long)reg->smax_value); |
7d1238f2 | 917 | if (reg->umin_value != 0) |
7df5072c | 918 | verbose_a("umin=%llu", (unsigned long long)reg->umin_value); |
7d1238f2 | 919 | if (reg->umax_value != U64_MAX) |
7df5072c | 920 | verbose_a("umax=%llu", (unsigned long long)reg->umax_value); |
7d1238f2 EC |
921 | if (!tnum_is_unknown(reg->var_off)) { |
922 | char tn_buf[48]; | |
f1174f77 | 923 | |
7d1238f2 | 924 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
7df5072c | 925 | verbose_a("var_off=%s", tn_buf); |
7d1238f2 | 926 | } |
3f50f132 JF |
927 | if (reg->s32_min_value != reg->smin_value && |
928 | reg->s32_min_value != S32_MIN) | |
7df5072c | 929 | verbose_a("s32_min=%d", (int)(reg->s32_min_value)); |
3f50f132 JF |
930 | if (reg->s32_max_value != reg->smax_value && |
931 | reg->s32_max_value != S32_MAX) | |
7df5072c | 932 | verbose_a("s32_max=%d", (int)(reg->s32_max_value)); |
3f50f132 JF |
933 | if (reg->u32_min_value != reg->umin_value && |
934 | reg->u32_min_value != U32_MIN) | |
7df5072c | 935 | verbose_a("u32_min=%d", (int)(reg->u32_min_value)); |
3f50f132 JF |
936 | if (reg->u32_max_value != reg->umax_value && |
937 | reg->u32_max_value != U32_MAX) | |
7df5072c | 938 | verbose_a("u32_max=%d", (int)(reg->u32_max_value)); |
f1174f77 | 939 | } |
7df5072c ML |
940 | #undef verbose_a |
941 | ||
61bd5218 | 942 | verbose(env, ")"); |
f1174f77 | 943 | } |
17a52670 | 944 | } |
638f5b90 | 945 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
946 | char types_buf[BPF_REG_SIZE + 1]; |
947 | bool valid = false; | |
948 | int j; | |
949 | ||
950 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
951 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
952 | valid = true; | |
953 | types_buf[j] = slot_type_char[ | |
954 | state->stack[i].slot_type[j]]; | |
955 | } | |
956 | types_buf[BPF_REG_SIZE] = 0; | |
957 | if (!valid) | |
958 | continue; | |
0f55f9ed CL |
959 | if (!print_all && !stack_slot_scratched(env, i)) |
960 | continue; | |
8efea21d EC |
961 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); |
962 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
27113c59 | 963 | if (is_spilled_reg(&state->stack[i])) { |
b5dc0163 AS |
964 | reg = &state->stack[i].spilled_ptr; |
965 | t = reg->type; | |
7df5072c | 966 | verbose(env, "=%s", t == SCALAR_VALUE ? "" : reg_type_str(env, t)); |
b5dc0163 AS |
967 | if (t == SCALAR_VALUE && reg->precise) |
968 | verbose(env, "P"); | |
969 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
970 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
971 | } else { | |
8efea21d | 972 | verbose(env, "=%s", types_buf); |
b5dc0163 | 973 | } |
17a52670 | 974 | } |
fd978bf7 JS |
975 | if (state->acquired_refs && state->refs[0].id) { |
976 | verbose(env, " refs=%d", state->refs[0].id); | |
977 | for (i = 1; i < state->acquired_refs; i++) | |
978 | if (state->refs[i].id) | |
979 | verbose(env, ",%d", state->refs[i].id); | |
980 | } | |
bfc6bb74 AS |
981 | if (state->in_callback_fn) |
982 | verbose(env, " cb"); | |
983 | if (state->in_async_callback_fn) | |
984 | verbose(env, " async_cb"); | |
61bd5218 | 985 | verbose(env, "\n"); |
0f55f9ed | 986 | mark_verifier_state_clean(env); |
17a52670 AS |
987 | } |
988 | ||
2e576648 CL |
989 | static inline u32 vlog_alignment(u32 pos) |
990 | { | |
991 | return round_up(max(pos + BPF_LOG_MIN_ALIGNMENT / 2, BPF_LOG_ALIGNMENT), | |
992 | BPF_LOG_MIN_ALIGNMENT) - pos - 1; | |
993 | } | |
994 | ||
995 | static void print_insn_state(struct bpf_verifier_env *env, | |
996 | const struct bpf_func_state *state) | |
997 | { | |
998 | if (env->prev_log_len && env->prev_log_len == env->log.len_used) { | |
999 | /* remove new line character */ | |
1000 | bpf_vlog_reset(&env->log, env->prev_log_len - 1); | |
1001 | verbose(env, "%*c;", vlog_alignment(env->prev_insn_print_len), ' '); | |
1002 | } else { | |
1003 | verbose(env, "%d:", env->insn_idx); | |
1004 | } | |
1005 | print_verifier_state(env, state, false); | |
17a52670 AS |
1006 | } |
1007 | ||
c69431aa LB |
1008 | /* copy array src of length n * size bytes to dst. dst is reallocated if it's too |
1009 | * small to hold src. This is different from krealloc since we don't want to preserve | |
1010 | * the contents of dst. | |
1011 | * | |
1012 | * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could | |
1013 | * not be allocated. | |
638f5b90 | 1014 | */ |
c69431aa | 1015 | static void *copy_array(void *dst, const void *src, size_t n, size_t size, gfp_t flags) |
638f5b90 | 1016 | { |
c69431aa LB |
1017 | size_t bytes; |
1018 | ||
1019 | if (ZERO_OR_NULL_PTR(src)) | |
1020 | goto out; | |
1021 | ||
1022 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
1023 | return NULL; | |
1024 | ||
1025 | if (ksize(dst) < bytes) { | |
1026 | kfree(dst); | |
1027 | dst = kmalloc_track_caller(bytes, flags); | |
1028 | if (!dst) | |
1029 | return NULL; | |
1030 | } | |
1031 | ||
1032 | memcpy(dst, src, bytes); | |
1033 | out: | |
1034 | return dst ? dst : ZERO_SIZE_PTR; | |
1035 | } | |
1036 | ||
1037 | /* resize an array from old_n items to new_n items. the array is reallocated if it's too | |
1038 | * small to hold new_n items. new items are zeroed out if the array grows. | |
1039 | * | |
1040 | * Contrary to krealloc_array, does not free arr if new_n is zero. | |
1041 | */ | |
1042 | static void *realloc_array(void *arr, size_t old_n, size_t new_n, size_t size) | |
1043 | { | |
42378a9c KC |
1044 | void *new_arr; |
1045 | ||
c69431aa LB |
1046 | if (!new_n || old_n == new_n) |
1047 | goto out; | |
1048 | ||
42378a9c KC |
1049 | new_arr = krealloc_array(arr, new_n, size, GFP_KERNEL); |
1050 | if (!new_arr) { | |
1051 | kfree(arr); | |
c69431aa | 1052 | return NULL; |
42378a9c KC |
1053 | } |
1054 | arr = new_arr; | |
c69431aa LB |
1055 | |
1056 | if (new_n > old_n) | |
1057 | memset(arr + old_n * size, 0, (new_n - old_n) * size); | |
1058 | ||
1059 | out: | |
1060 | return arr ? arr : ZERO_SIZE_PTR; | |
1061 | } | |
1062 | ||
1063 | static int copy_reference_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1064 | { | |
1065 | dst->refs = copy_array(dst->refs, src->refs, src->acquired_refs, | |
1066 | sizeof(struct bpf_reference_state), GFP_KERNEL); | |
1067 | if (!dst->refs) | |
1068 | return -ENOMEM; | |
1069 | ||
1070 | dst->acquired_refs = src->acquired_refs; | |
1071 | return 0; | |
1072 | } | |
1073 | ||
1074 | static int copy_stack_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
1075 | { | |
1076 | size_t n = src->allocated_stack / BPF_REG_SIZE; | |
1077 | ||
1078 | dst->stack = copy_array(dst->stack, src->stack, n, sizeof(struct bpf_stack_state), | |
1079 | GFP_KERNEL); | |
1080 | if (!dst->stack) | |
1081 | return -ENOMEM; | |
1082 | ||
1083 | dst->allocated_stack = src->allocated_stack; | |
1084 | return 0; | |
1085 | } | |
1086 | ||
1087 | static int resize_reference_state(struct bpf_func_state *state, size_t n) | |
1088 | { | |
1089 | state->refs = realloc_array(state->refs, state->acquired_refs, n, | |
1090 | sizeof(struct bpf_reference_state)); | |
1091 | if (!state->refs) | |
1092 | return -ENOMEM; | |
1093 | ||
1094 | state->acquired_refs = n; | |
1095 | return 0; | |
1096 | } | |
1097 | ||
1098 | static int grow_stack_state(struct bpf_func_state *state, int size) | |
1099 | { | |
1100 | size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE; | |
1101 | ||
1102 | if (old_n >= n) | |
1103 | return 0; | |
1104 | ||
1105 | state->stack = realloc_array(state->stack, old_n, n, sizeof(struct bpf_stack_state)); | |
1106 | if (!state->stack) | |
1107 | return -ENOMEM; | |
1108 | ||
1109 | state->allocated_stack = size; | |
1110 | return 0; | |
fd978bf7 JS |
1111 | } |
1112 | ||
1113 | /* Acquire a pointer id from the env and update the state->refs to include | |
1114 | * this new pointer reference. | |
1115 | * On success, returns a valid pointer id to associate with the register | |
1116 | * On failure, returns a negative errno. | |
638f5b90 | 1117 | */ |
fd978bf7 | 1118 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 1119 | { |
fd978bf7 JS |
1120 | struct bpf_func_state *state = cur_func(env); |
1121 | int new_ofs = state->acquired_refs; | |
1122 | int id, err; | |
1123 | ||
c69431aa | 1124 | err = resize_reference_state(state, state->acquired_refs + 1); |
fd978bf7 JS |
1125 | if (err) |
1126 | return err; | |
1127 | id = ++env->id_gen; | |
1128 | state->refs[new_ofs].id = id; | |
1129 | state->refs[new_ofs].insn_idx = insn_idx; | |
9d9d00ac | 1130 | state->refs[new_ofs].callback_ref = state->in_callback_fn ? state->frameno : 0; |
638f5b90 | 1131 | |
fd978bf7 JS |
1132 | return id; |
1133 | } | |
1134 | ||
1135 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 1136 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
1137 | { |
1138 | int i, last_idx; | |
1139 | ||
fd978bf7 JS |
1140 | last_idx = state->acquired_refs - 1; |
1141 | for (i = 0; i < state->acquired_refs; i++) { | |
1142 | if (state->refs[i].id == ptr_id) { | |
9d9d00ac KKD |
1143 | /* Cannot release caller references in callbacks */ |
1144 | if (state->in_callback_fn && state->refs[i].callback_ref != state->frameno) | |
1145 | return -EINVAL; | |
fd978bf7 JS |
1146 | if (last_idx && i != last_idx) |
1147 | memcpy(&state->refs[i], &state->refs[last_idx], | |
1148 | sizeof(*state->refs)); | |
1149 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
1150 | state->acquired_refs--; | |
638f5b90 | 1151 | return 0; |
638f5b90 | 1152 | } |
638f5b90 | 1153 | } |
46f8bc92 | 1154 | return -EINVAL; |
fd978bf7 JS |
1155 | } |
1156 | ||
f4d7e40a AS |
1157 | static void free_func_state(struct bpf_func_state *state) |
1158 | { | |
5896351e AS |
1159 | if (!state) |
1160 | return; | |
fd978bf7 | 1161 | kfree(state->refs); |
f4d7e40a AS |
1162 | kfree(state->stack); |
1163 | kfree(state); | |
1164 | } | |
1165 | ||
b5dc0163 AS |
1166 | static void clear_jmp_history(struct bpf_verifier_state *state) |
1167 | { | |
1168 | kfree(state->jmp_history); | |
1169 | state->jmp_history = NULL; | |
1170 | state->jmp_history_cnt = 0; | |
1171 | } | |
1172 | ||
1969db47 AS |
1173 | static void free_verifier_state(struct bpf_verifier_state *state, |
1174 | bool free_self) | |
638f5b90 | 1175 | { |
f4d7e40a AS |
1176 | int i; |
1177 | ||
1178 | for (i = 0; i <= state->curframe; i++) { | |
1179 | free_func_state(state->frame[i]); | |
1180 | state->frame[i] = NULL; | |
1181 | } | |
b5dc0163 | 1182 | clear_jmp_history(state); |
1969db47 AS |
1183 | if (free_self) |
1184 | kfree(state); | |
638f5b90 AS |
1185 | } |
1186 | ||
1187 | /* copy verifier state from src to dst growing dst stack space | |
1188 | * when necessary to accommodate larger src stack | |
1189 | */ | |
f4d7e40a AS |
1190 | static int copy_func_state(struct bpf_func_state *dst, |
1191 | const struct bpf_func_state *src) | |
638f5b90 AS |
1192 | { |
1193 | int err; | |
1194 | ||
fd978bf7 JS |
1195 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); |
1196 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
1197 | if (err) |
1198 | return err; | |
638f5b90 AS |
1199 | return copy_stack_state(dst, src); |
1200 | } | |
1201 | ||
f4d7e40a AS |
1202 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
1203 | const struct bpf_verifier_state *src) | |
1204 | { | |
1205 | struct bpf_func_state *dst; | |
1206 | int i, err; | |
1207 | ||
06ab6a50 LB |
1208 | dst_state->jmp_history = copy_array(dst_state->jmp_history, src->jmp_history, |
1209 | src->jmp_history_cnt, sizeof(struct bpf_idx_pair), | |
1210 | GFP_USER); | |
1211 | if (!dst_state->jmp_history) | |
1212 | return -ENOMEM; | |
b5dc0163 AS |
1213 | dst_state->jmp_history_cnt = src->jmp_history_cnt; |
1214 | ||
f4d7e40a AS |
1215 | /* if dst has more stack frames then src frame, free them */ |
1216 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
1217 | free_func_state(dst_state->frame[i]); | |
1218 | dst_state->frame[i] = NULL; | |
1219 | } | |
979d63d5 | 1220 | dst_state->speculative = src->speculative; |
f4d7e40a | 1221 | dst_state->curframe = src->curframe; |
d0d78c1d KKD |
1222 | dst_state->active_lock.ptr = src->active_lock.ptr; |
1223 | dst_state->active_lock.id = src->active_lock.id; | |
2589726d AS |
1224 | dst_state->branches = src->branches; |
1225 | dst_state->parent = src->parent; | |
b5dc0163 AS |
1226 | dst_state->first_insn_idx = src->first_insn_idx; |
1227 | dst_state->last_insn_idx = src->last_insn_idx; | |
f4d7e40a AS |
1228 | for (i = 0; i <= src->curframe; i++) { |
1229 | dst = dst_state->frame[i]; | |
1230 | if (!dst) { | |
1231 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
1232 | if (!dst) | |
1233 | return -ENOMEM; | |
1234 | dst_state->frame[i] = dst; | |
1235 | } | |
1236 | err = copy_func_state(dst, src->frame[i]); | |
1237 | if (err) | |
1238 | return err; | |
1239 | } | |
1240 | return 0; | |
1241 | } | |
1242 | ||
2589726d AS |
1243 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
1244 | { | |
1245 | while (st) { | |
1246 | u32 br = --st->branches; | |
1247 | ||
1248 | /* WARN_ON(br > 1) technically makes sense here, | |
1249 | * but see comment in push_stack(), hence: | |
1250 | */ | |
1251 | WARN_ONCE((int)br < 0, | |
1252 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
1253 | br); | |
1254 | if (br) | |
1255 | break; | |
1256 | st = st->parent; | |
1257 | } | |
1258 | } | |
1259 | ||
638f5b90 | 1260 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 1261 | int *insn_idx, bool pop_log) |
638f5b90 AS |
1262 | { |
1263 | struct bpf_verifier_state *cur = env->cur_state; | |
1264 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
1265 | int err; | |
17a52670 AS |
1266 | |
1267 | if (env->head == NULL) | |
638f5b90 | 1268 | return -ENOENT; |
17a52670 | 1269 | |
638f5b90 AS |
1270 | if (cur) { |
1271 | err = copy_verifier_state(cur, &head->st); | |
1272 | if (err) | |
1273 | return err; | |
1274 | } | |
6f8a57cc AN |
1275 | if (pop_log) |
1276 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
1277 | if (insn_idx) |
1278 | *insn_idx = head->insn_idx; | |
17a52670 | 1279 | if (prev_insn_idx) |
638f5b90 AS |
1280 | *prev_insn_idx = head->prev_insn_idx; |
1281 | elem = head->next; | |
1969db47 | 1282 | free_verifier_state(&head->st, false); |
638f5b90 | 1283 | kfree(head); |
17a52670 AS |
1284 | env->head = elem; |
1285 | env->stack_size--; | |
638f5b90 | 1286 | return 0; |
17a52670 AS |
1287 | } |
1288 | ||
58e2af8b | 1289 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
1290 | int insn_idx, int prev_insn_idx, |
1291 | bool speculative) | |
17a52670 | 1292 | { |
638f5b90 | 1293 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 1294 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 1295 | int err; |
17a52670 | 1296 | |
638f5b90 | 1297 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
1298 | if (!elem) |
1299 | goto err; | |
1300 | ||
17a52670 AS |
1301 | elem->insn_idx = insn_idx; |
1302 | elem->prev_insn_idx = prev_insn_idx; | |
1303 | elem->next = env->head; | |
6f8a57cc | 1304 | elem->log_pos = env->log.len_used; |
17a52670 AS |
1305 | env->head = elem; |
1306 | env->stack_size++; | |
1969db47 AS |
1307 | err = copy_verifier_state(&elem->st, cur); |
1308 | if (err) | |
1309 | goto err; | |
979d63d5 | 1310 | elem->st.speculative |= speculative; |
b285fcb7 AS |
1311 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
1312 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
1313 | env->stack_size); | |
17a52670 AS |
1314 | goto err; |
1315 | } | |
2589726d AS |
1316 | if (elem->st.parent) { |
1317 | ++elem->st.parent->branches; | |
1318 | /* WARN_ON(branches > 2) technically makes sense here, | |
1319 | * but | |
1320 | * 1. speculative states will bump 'branches' for non-branch | |
1321 | * instructions | |
1322 | * 2. is_state_visited() heuristics may decide not to create | |
1323 | * a new state for a sequence of branches and all such current | |
1324 | * and cloned states will be pointing to a single parent state | |
1325 | * which might have large 'branches' count. | |
1326 | */ | |
1327 | } | |
17a52670 AS |
1328 | return &elem->st; |
1329 | err: | |
5896351e AS |
1330 | free_verifier_state(env->cur_state, true); |
1331 | env->cur_state = NULL; | |
17a52670 | 1332 | /* pop all elements and return */ |
6f8a57cc | 1333 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
1334 | return NULL; |
1335 | } | |
1336 | ||
1337 | #define CALLER_SAVED_REGS 6 | |
1338 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
1339 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
1340 | }; | |
1341 | ||
f54c7898 DB |
1342 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1343 | struct bpf_reg_state *reg); | |
f1174f77 | 1344 | |
e688c3db AS |
1345 | /* This helper doesn't clear reg->id */ |
1346 | static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
b03c9f9f | 1347 | { |
b03c9f9f EC |
1348 | reg->var_off = tnum_const(imm); |
1349 | reg->smin_value = (s64)imm; | |
1350 | reg->smax_value = (s64)imm; | |
1351 | reg->umin_value = imm; | |
1352 | reg->umax_value = imm; | |
3f50f132 JF |
1353 | |
1354 | reg->s32_min_value = (s32)imm; | |
1355 | reg->s32_max_value = (s32)imm; | |
1356 | reg->u32_min_value = (u32)imm; | |
1357 | reg->u32_max_value = (u32)imm; | |
1358 | } | |
1359 | ||
e688c3db AS |
1360 | /* Mark the unknown part of a register (variable offset or scalar value) as |
1361 | * known to have the value @imm. | |
1362 | */ | |
1363 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
1364 | { | |
1365 | /* Clear id, off, and union(map_ptr, range) */ | |
1366 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
1367 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
1368 | ___mark_reg_known(reg, imm); | |
1369 | } | |
1370 | ||
3f50f132 JF |
1371 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) |
1372 | { | |
1373 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
1374 | reg->s32_min_value = (s32)imm; | |
1375 | reg->s32_max_value = (s32)imm; | |
1376 | reg->u32_min_value = (u32)imm; | |
1377 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
1378 | } |
1379 | ||
f1174f77 EC |
1380 | /* Mark the 'variable offset' part of a register as zero. This should be |
1381 | * used only on registers holding a pointer type. | |
1382 | */ | |
1383 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 1384 | { |
b03c9f9f | 1385 | __mark_reg_known(reg, 0); |
f1174f77 | 1386 | } |
a9789ef9 | 1387 | |
cc2b14d5 AS |
1388 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
1389 | { | |
1390 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
1391 | reg->type = SCALAR_VALUE; |
1392 | } | |
1393 | ||
61bd5218 JK |
1394 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
1395 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1396 | { |
1397 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1398 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
1399 | /* Something bad happened, let's kill all regs */ |
1400 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 1401 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1402 | return; |
1403 | } | |
1404 | __mark_reg_known_zero(regs + regno); | |
1405 | } | |
1406 | ||
4ddb7416 DB |
1407 | static void mark_ptr_not_null_reg(struct bpf_reg_state *reg) |
1408 | { | |
c25b2ae1 | 1409 | if (base_type(reg->type) == PTR_TO_MAP_VALUE) { |
4ddb7416 DB |
1410 | const struct bpf_map *map = reg->map_ptr; |
1411 | ||
1412 | if (map->inner_map_meta) { | |
1413 | reg->type = CONST_PTR_TO_MAP; | |
1414 | reg->map_ptr = map->inner_map_meta; | |
3e8ce298 AS |
1415 | /* transfer reg's id which is unique for every map_lookup_elem |
1416 | * as UID of the inner map. | |
1417 | */ | |
db559117 | 1418 | if (btf_record_has_field(map->inner_map_meta->record, BPF_TIMER)) |
34d11a44 | 1419 | reg->map_uid = reg->id; |
4ddb7416 DB |
1420 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { |
1421 | reg->type = PTR_TO_XDP_SOCK; | |
1422 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || | |
1423 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
1424 | reg->type = PTR_TO_SOCKET; | |
1425 | } else { | |
1426 | reg->type = PTR_TO_MAP_VALUE; | |
1427 | } | |
c25b2ae1 | 1428 | return; |
4ddb7416 | 1429 | } |
c25b2ae1 HL |
1430 | |
1431 | reg->type &= ~PTR_MAYBE_NULL; | |
4ddb7416 DB |
1432 | } |
1433 | ||
de8f3a83 DB |
1434 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
1435 | { | |
1436 | return type_is_pkt_pointer(reg->type); | |
1437 | } | |
1438 | ||
1439 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
1440 | { | |
1441 | return reg_is_pkt_pointer(reg) || | |
1442 | reg->type == PTR_TO_PACKET_END; | |
1443 | } | |
1444 | ||
1445 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
1446 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
1447 | enum bpf_reg_type which) | |
1448 | { | |
1449 | /* The register can already have a range from prior markings. | |
1450 | * This is fine as long as it hasn't been advanced from its | |
1451 | * origin. | |
1452 | */ | |
1453 | return reg->type == which && | |
1454 | reg->id == 0 && | |
1455 | reg->off == 0 && | |
1456 | tnum_equals_const(reg->var_off, 0); | |
1457 | } | |
1458 | ||
3f50f132 JF |
1459 | /* Reset the min/max bounds of a register */ |
1460 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
1461 | { | |
1462 | reg->smin_value = S64_MIN; | |
1463 | reg->smax_value = S64_MAX; | |
1464 | reg->umin_value = 0; | |
1465 | reg->umax_value = U64_MAX; | |
1466 | ||
1467 | reg->s32_min_value = S32_MIN; | |
1468 | reg->s32_max_value = S32_MAX; | |
1469 | reg->u32_min_value = 0; | |
1470 | reg->u32_max_value = U32_MAX; | |
1471 | } | |
1472 | ||
1473 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
1474 | { | |
1475 | reg->smin_value = S64_MIN; | |
1476 | reg->smax_value = S64_MAX; | |
1477 | reg->umin_value = 0; | |
1478 | reg->umax_value = U64_MAX; | |
1479 | } | |
1480 | ||
1481 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
1482 | { | |
1483 | reg->s32_min_value = S32_MIN; | |
1484 | reg->s32_max_value = S32_MAX; | |
1485 | reg->u32_min_value = 0; | |
1486 | reg->u32_max_value = U32_MAX; | |
1487 | } | |
1488 | ||
1489 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
1490 | { | |
1491 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
1492 | ||
1493 | /* min signed is max(sign bit) | min(other bits) */ | |
1494 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
1495 | var32_off.value | (var32_off.mask & S32_MIN)); | |
1496 | /* max signed is min(sign bit) | max(other bits) */ | |
1497 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
1498 | var32_off.value | (var32_off.mask & S32_MAX)); | |
1499 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
1500 | reg->u32_max_value = min(reg->u32_max_value, | |
1501 | (u32)(var32_off.value | var32_off.mask)); | |
1502 | } | |
1503 | ||
1504 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1505 | { |
1506 | /* min signed is max(sign bit) | min(other bits) */ | |
1507 | reg->smin_value = max_t(s64, reg->smin_value, | |
1508 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
1509 | /* max signed is min(sign bit) | max(other bits) */ | |
1510 | reg->smax_value = min_t(s64, reg->smax_value, | |
1511 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
1512 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
1513 | reg->umax_value = min(reg->umax_value, | |
1514 | reg->var_off.value | reg->var_off.mask); | |
1515 | } | |
1516 | ||
3f50f132 JF |
1517 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
1518 | { | |
1519 | __update_reg32_bounds(reg); | |
1520 | __update_reg64_bounds(reg); | |
1521 | } | |
1522 | ||
b03c9f9f | 1523 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
1524 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
1525 | { | |
1526 | /* Learn sign from signed bounds. | |
1527 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1528 | * are the same, so combine. This works even in the negative case, e.g. | |
1529 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1530 | */ | |
1531 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
1532 | reg->s32_min_value = reg->u32_min_value = | |
1533 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1534 | reg->s32_max_value = reg->u32_max_value = | |
1535 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1536 | return; | |
1537 | } | |
1538 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1539 | * boundary, so we must be careful. | |
1540 | */ | |
1541 | if ((s32)reg->u32_max_value >= 0) { | |
1542 | /* Positive. We can't learn anything from the smin, but smax | |
1543 | * is positive, hence safe. | |
1544 | */ | |
1545 | reg->s32_min_value = reg->u32_min_value; | |
1546 | reg->s32_max_value = reg->u32_max_value = | |
1547 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1548 | } else if ((s32)reg->u32_min_value < 0) { | |
1549 | /* Negative. We can't learn anything from the smax, but smin | |
1550 | * is negative, hence safe. | |
1551 | */ | |
1552 | reg->s32_min_value = reg->u32_min_value = | |
1553 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1554 | reg->s32_max_value = reg->u32_max_value; | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1559 | { |
1560 | /* Learn sign from signed bounds. | |
1561 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1562 | * are the same, so combine. This works even in the negative case, e.g. | |
1563 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1564 | */ | |
1565 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
1566 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1567 | reg->umin_value); | |
1568 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1569 | reg->umax_value); | |
1570 | return; | |
1571 | } | |
1572 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1573 | * boundary, so we must be careful. | |
1574 | */ | |
1575 | if ((s64)reg->umax_value >= 0) { | |
1576 | /* Positive. We can't learn anything from the smin, but smax | |
1577 | * is positive, hence safe. | |
1578 | */ | |
1579 | reg->smin_value = reg->umin_value; | |
1580 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1581 | reg->umax_value); | |
1582 | } else if ((s64)reg->umin_value < 0) { | |
1583 | /* Negative. We can't learn anything from the smax, but smin | |
1584 | * is negative, hence safe. | |
1585 | */ | |
1586 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1587 | reg->umin_value); | |
1588 | reg->smax_value = reg->umax_value; | |
1589 | } | |
1590 | } | |
1591 | ||
3f50f132 JF |
1592 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
1593 | { | |
1594 | __reg32_deduce_bounds(reg); | |
1595 | __reg64_deduce_bounds(reg); | |
1596 | } | |
1597 | ||
b03c9f9f EC |
1598 | /* Attempts to improve var_off based on unsigned min/max information */ |
1599 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
1600 | { | |
3f50f132 JF |
1601 | struct tnum var64_off = tnum_intersect(reg->var_off, |
1602 | tnum_range(reg->umin_value, | |
1603 | reg->umax_value)); | |
1604 | struct tnum var32_off = tnum_intersect(tnum_subreg(reg->var_off), | |
1605 | tnum_range(reg->u32_min_value, | |
1606 | reg->u32_max_value)); | |
1607 | ||
1608 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
1609 | } |
1610 | ||
3844d153 DB |
1611 | static void reg_bounds_sync(struct bpf_reg_state *reg) |
1612 | { | |
1613 | /* We might have learned new bounds from the var_off. */ | |
1614 | __update_reg_bounds(reg); | |
1615 | /* We might have learned something about the sign bit. */ | |
1616 | __reg_deduce_bounds(reg); | |
1617 | /* We might have learned some bits from the bounds. */ | |
1618 | __reg_bound_offset(reg); | |
1619 | /* Intersecting with the old var_off might have improved our bounds | |
1620 | * slightly, e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1621 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1622 | */ | |
1623 | __update_reg_bounds(reg); | |
1624 | } | |
1625 | ||
e572ff80 DB |
1626 | static bool __reg32_bound_s64(s32 a) |
1627 | { | |
1628 | return a >= 0 && a <= S32_MAX; | |
1629 | } | |
1630 | ||
3f50f132 | 1631 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 1632 | { |
3f50f132 JF |
1633 | reg->umin_value = reg->u32_min_value; |
1634 | reg->umax_value = reg->u32_max_value; | |
e572ff80 DB |
1635 | |
1636 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds but must | |
1637 | * be positive otherwise set to worse case bounds and refine later | |
1638 | * from tnum. | |
3f50f132 | 1639 | */ |
e572ff80 DB |
1640 | if (__reg32_bound_s64(reg->s32_min_value) && |
1641 | __reg32_bound_s64(reg->s32_max_value)) { | |
3a71dc36 | 1642 | reg->smin_value = reg->s32_min_value; |
e572ff80 DB |
1643 | reg->smax_value = reg->s32_max_value; |
1644 | } else { | |
3a71dc36 | 1645 | reg->smin_value = 0; |
e572ff80 DB |
1646 | reg->smax_value = U32_MAX; |
1647 | } | |
3f50f132 JF |
1648 | } |
1649 | ||
1650 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
1651 | { | |
1652 | /* special case when 64-bit register has upper 32-bit register | |
1653 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
1654 | * allowing us to use 32-bit bounds directly, | |
1655 | */ | |
1656 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
1657 | __reg_assign_32_into_64(reg); | |
1658 | } else { | |
1659 | /* Otherwise the best we can do is push lower 32bit known and | |
1660 | * unknown bits into register (var_off set from jmp logic) | |
1661 | * then learn as much as possible from the 64-bit tnum | |
1662 | * known and unknown bits. The previous smin/smax bounds are | |
1663 | * invalid here because of jmp32 compare so mark them unknown | |
1664 | * so they do not impact tnum bounds calculation. | |
1665 | */ | |
1666 | __mark_reg64_unbounded(reg); | |
3f50f132 | 1667 | } |
3844d153 | 1668 | reg_bounds_sync(reg); |
3f50f132 JF |
1669 | } |
1670 | ||
1671 | static bool __reg64_bound_s32(s64 a) | |
1672 | { | |
388e2c0b | 1673 | return a >= S32_MIN && a <= S32_MAX; |
3f50f132 JF |
1674 | } |
1675 | ||
1676 | static bool __reg64_bound_u32(u64 a) | |
1677 | { | |
b9979db8 | 1678 | return a >= U32_MIN && a <= U32_MAX; |
3f50f132 JF |
1679 | } |
1680 | ||
1681 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
1682 | { | |
1683 | __mark_reg32_unbounded(reg); | |
b0270958 | 1684 | if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) { |
3f50f132 | 1685 | reg->s32_min_value = (s32)reg->smin_value; |
3f50f132 | 1686 | reg->s32_max_value = (s32)reg->smax_value; |
b0270958 | 1687 | } |
10bf4e83 | 1688 | if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value)) { |
3f50f132 | 1689 | reg->u32_min_value = (u32)reg->umin_value; |
3f50f132 | 1690 | reg->u32_max_value = (u32)reg->umax_value; |
10bf4e83 | 1691 | } |
3844d153 | 1692 | reg_bounds_sync(reg); |
b03c9f9f EC |
1693 | } |
1694 | ||
f1174f77 | 1695 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
1696 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1697 | struct bpf_reg_state *reg) | |
f1174f77 | 1698 | { |
a9c676bc AS |
1699 | /* |
1700 | * Clear type, id, off, and union(map_ptr, range) and | |
1701 | * padding between 'type' and union | |
1702 | */ | |
1703 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 1704 | reg->type = SCALAR_VALUE; |
f1174f77 | 1705 | reg->var_off = tnum_unknown; |
f4d7e40a | 1706 | reg->frameno = 0; |
be2ef816 | 1707 | reg->precise = !env->bpf_capable; |
b03c9f9f | 1708 | __mark_reg_unbounded(reg); |
f1174f77 EC |
1709 | } |
1710 | ||
61bd5218 JK |
1711 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
1712 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1713 | { |
1714 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1715 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
1716 | /* Something bad happened, let's kill all regs except FP */ |
1717 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1718 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1719 | return; |
1720 | } | |
f54c7898 | 1721 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
1722 | } |
1723 | ||
f54c7898 DB |
1724 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1725 | struct bpf_reg_state *reg) | |
f1174f77 | 1726 | { |
f54c7898 | 1727 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
1728 | reg->type = NOT_INIT; |
1729 | } | |
1730 | ||
61bd5218 JK |
1731 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
1732 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1733 | { |
1734 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1735 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
1736 | /* Something bad happened, let's kill all regs except FP */ |
1737 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1738 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1739 | return; |
1740 | } | |
f54c7898 | 1741 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
1742 | } |
1743 | ||
41c48f3a AI |
1744 | static void mark_btf_ld_reg(struct bpf_verifier_env *env, |
1745 | struct bpf_reg_state *regs, u32 regno, | |
22dc4a0f | 1746 | enum bpf_reg_type reg_type, |
c6f1bfe8 YS |
1747 | struct btf *btf, u32 btf_id, |
1748 | enum bpf_type_flag flag) | |
41c48f3a AI |
1749 | { |
1750 | if (reg_type == SCALAR_VALUE) { | |
1751 | mark_reg_unknown(env, regs, regno); | |
1752 | return; | |
1753 | } | |
1754 | mark_reg_known_zero(env, regs, regno); | |
c6f1bfe8 | 1755 | regs[regno].type = PTR_TO_BTF_ID | flag; |
22dc4a0f | 1756 | regs[regno].btf = btf; |
41c48f3a AI |
1757 | regs[regno].btf_id = btf_id; |
1758 | } | |
1759 | ||
5327ed3d | 1760 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 1761 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 1762 | struct bpf_func_state *state) |
17a52670 | 1763 | { |
f4d7e40a | 1764 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
1765 | int i; |
1766 | ||
dc503a8a | 1767 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 1768 | mark_reg_not_init(env, regs, i); |
dc503a8a | 1769 | regs[i].live = REG_LIVE_NONE; |
679c782d | 1770 | regs[i].parent = NULL; |
5327ed3d | 1771 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 1772 | } |
17a52670 AS |
1773 | |
1774 | /* frame pointer */ | |
f1174f77 | 1775 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 1776 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 1777 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
1778 | } |
1779 | ||
f4d7e40a AS |
1780 | #define BPF_MAIN_FUNC (-1) |
1781 | static void init_func_state(struct bpf_verifier_env *env, | |
1782 | struct bpf_func_state *state, | |
1783 | int callsite, int frameno, int subprogno) | |
1784 | { | |
1785 | state->callsite = callsite; | |
1786 | state->frameno = frameno; | |
1787 | state->subprogno = subprogno; | |
1bfe26fb | 1788 | state->callback_ret_range = tnum_range(0, 0); |
f4d7e40a | 1789 | init_reg_state(env, state); |
0f55f9ed | 1790 | mark_verifier_state_scratched(env); |
f4d7e40a AS |
1791 | } |
1792 | ||
bfc6bb74 AS |
1793 | /* Similar to push_stack(), but for async callbacks */ |
1794 | static struct bpf_verifier_state *push_async_cb(struct bpf_verifier_env *env, | |
1795 | int insn_idx, int prev_insn_idx, | |
1796 | int subprog) | |
1797 | { | |
1798 | struct bpf_verifier_stack_elem *elem; | |
1799 | struct bpf_func_state *frame; | |
1800 | ||
1801 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); | |
1802 | if (!elem) | |
1803 | goto err; | |
1804 | ||
1805 | elem->insn_idx = insn_idx; | |
1806 | elem->prev_insn_idx = prev_insn_idx; | |
1807 | elem->next = env->head; | |
1808 | elem->log_pos = env->log.len_used; | |
1809 | env->head = elem; | |
1810 | env->stack_size++; | |
1811 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { | |
1812 | verbose(env, | |
1813 | "The sequence of %d jumps is too complex for async cb.\n", | |
1814 | env->stack_size); | |
1815 | goto err; | |
1816 | } | |
1817 | /* Unlike push_stack() do not copy_verifier_state(). | |
1818 | * The caller state doesn't matter. | |
1819 | * This is async callback. It starts in a fresh stack. | |
1820 | * Initialize it similar to do_check_common(). | |
1821 | */ | |
1822 | elem->st.branches = 1; | |
1823 | frame = kzalloc(sizeof(*frame), GFP_KERNEL); | |
1824 | if (!frame) | |
1825 | goto err; | |
1826 | init_func_state(env, frame, | |
1827 | BPF_MAIN_FUNC /* callsite */, | |
1828 | 0 /* frameno within this callchain */, | |
1829 | subprog /* subprog number within this prog */); | |
1830 | elem->st.frame[0] = frame; | |
1831 | return &elem->st; | |
1832 | err: | |
1833 | free_verifier_state(env->cur_state, true); | |
1834 | env->cur_state = NULL; | |
1835 | /* pop all elements and return */ | |
1836 | while (!pop_stack(env, NULL, NULL, false)); | |
1837 | return NULL; | |
1838 | } | |
1839 | ||
1840 | ||
17a52670 AS |
1841 | enum reg_arg_type { |
1842 | SRC_OP, /* register is used as source operand */ | |
1843 | DST_OP, /* register is used as destination operand */ | |
1844 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1845 | }; | |
1846 | ||
cc8b0b92 AS |
1847 | static int cmp_subprogs(const void *a, const void *b) |
1848 | { | |
9c8105bd JW |
1849 | return ((struct bpf_subprog_info *)a)->start - |
1850 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1851 | } |
1852 | ||
1853 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1854 | { | |
9c8105bd | 1855 | struct bpf_subprog_info *p; |
cc8b0b92 | 1856 | |
9c8105bd JW |
1857 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1858 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1859 | if (!p) |
1860 | return -ENOENT; | |
9c8105bd | 1861 | return p - env->subprog_info; |
cc8b0b92 AS |
1862 | |
1863 | } | |
1864 | ||
1865 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1866 | { | |
1867 | int insn_cnt = env->prog->len; | |
1868 | int ret; | |
1869 | ||
1870 | if (off >= insn_cnt || off < 0) { | |
1871 | verbose(env, "call to invalid destination\n"); | |
1872 | return -EINVAL; | |
1873 | } | |
1874 | ret = find_subprog(env, off); | |
1875 | if (ret >= 0) | |
282a0f46 | 1876 | return ret; |
4cb3d99c | 1877 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1878 | verbose(env, "too many subprograms\n"); |
1879 | return -E2BIG; | |
1880 | } | |
e6ac2450 | 1881 | /* determine subprog starts. The end is one before the next starts */ |
9c8105bd JW |
1882 | env->subprog_info[env->subprog_cnt++].start = off; |
1883 | sort(env->subprog_info, env->subprog_cnt, | |
1884 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
282a0f46 | 1885 | return env->subprog_cnt - 1; |
cc8b0b92 AS |
1886 | } |
1887 | ||
2357672c KKD |
1888 | #define MAX_KFUNC_DESCS 256 |
1889 | #define MAX_KFUNC_BTFS 256 | |
1890 | ||
e6ac2450 MKL |
1891 | struct bpf_kfunc_desc { |
1892 | struct btf_func_model func_model; | |
1893 | u32 func_id; | |
1894 | s32 imm; | |
2357672c KKD |
1895 | u16 offset; |
1896 | }; | |
1897 | ||
1898 | struct bpf_kfunc_btf { | |
1899 | struct btf *btf; | |
1900 | struct module *module; | |
1901 | u16 offset; | |
e6ac2450 MKL |
1902 | }; |
1903 | ||
e6ac2450 MKL |
1904 | struct bpf_kfunc_desc_tab { |
1905 | struct bpf_kfunc_desc descs[MAX_KFUNC_DESCS]; | |
1906 | u32 nr_descs; | |
1907 | }; | |
1908 | ||
2357672c KKD |
1909 | struct bpf_kfunc_btf_tab { |
1910 | struct bpf_kfunc_btf descs[MAX_KFUNC_BTFS]; | |
1911 | u32 nr_descs; | |
1912 | }; | |
1913 | ||
1914 | static int kfunc_desc_cmp_by_id_off(const void *a, const void *b) | |
e6ac2450 MKL |
1915 | { |
1916 | const struct bpf_kfunc_desc *d0 = a; | |
1917 | const struct bpf_kfunc_desc *d1 = b; | |
1918 | ||
1919 | /* func_id is not greater than BTF_MAX_TYPE */ | |
2357672c KKD |
1920 | return d0->func_id - d1->func_id ?: d0->offset - d1->offset; |
1921 | } | |
1922 | ||
1923 | static int kfunc_btf_cmp_by_off(const void *a, const void *b) | |
1924 | { | |
1925 | const struct bpf_kfunc_btf *d0 = a; | |
1926 | const struct bpf_kfunc_btf *d1 = b; | |
1927 | ||
1928 | return d0->offset - d1->offset; | |
e6ac2450 MKL |
1929 | } |
1930 | ||
1931 | static const struct bpf_kfunc_desc * | |
2357672c | 1932 | find_kfunc_desc(const struct bpf_prog *prog, u32 func_id, u16 offset) |
e6ac2450 MKL |
1933 | { |
1934 | struct bpf_kfunc_desc desc = { | |
1935 | .func_id = func_id, | |
2357672c | 1936 | .offset = offset, |
e6ac2450 MKL |
1937 | }; |
1938 | struct bpf_kfunc_desc_tab *tab; | |
1939 | ||
1940 | tab = prog->aux->kfunc_tab; | |
1941 | return bsearch(&desc, tab->descs, tab->nr_descs, | |
2357672c KKD |
1942 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_id_off); |
1943 | } | |
1944 | ||
1945 | static struct btf *__find_kfunc_desc_btf(struct bpf_verifier_env *env, | |
b202d844 | 1946 | s16 offset) |
2357672c KKD |
1947 | { |
1948 | struct bpf_kfunc_btf kf_btf = { .offset = offset }; | |
1949 | struct bpf_kfunc_btf_tab *tab; | |
1950 | struct bpf_kfunc_btf *b; | |
1951 | struct module *mod; | |
1952 | struct btf *btf; | |
1953 | int btf_fd; | |
1954 | ||
1955 | tab = env->prog->aux->kfunc_btf_tab; | |
1956 | b = bsearch(&kf_btf, tab->descs, tab->nr_descs, | |
1957 | sizeof(tab->descs[0]), kfunc_btf_cmp_by_off); | |
1958 | if (!b) { | |
1959 | if (tab->nr_descs == MAX_KFUNC_BTFS) { | |
1960 | verbose(env, "too many different module BTFs\n"); | |
1961 | return ERR_PTR(-E2BIG); | |
1962 | } | |
1963 | ||
1964 | if (bpfptr_is_null(env->fd_array)) { | |
1965 | verbose(env, "kfunc offset > 0 without fd_array is invalid\n"); | |
1966 | return ERR_PTR(-EPROTO); | |
1967 | } | |
1968 | ||
1969 | if (copy_from_bpfptr_offset(&btf_fd, env->fd_array, | |
1970 | offset * sizeof(btf_fd), | |
1971 | sizeof(btf_fd))) | |
1972 | return ERR_PTR(-EFAULT); | |
1973 | ||
1974 | btf = btf_get_by_fd(btf_fd); | |
588cd7ef KKD |
1975 | if (IS_ERR(btf)) { |
1976 | verbose(env, "invalid module BTF fd specified\n"); | |
2357672c | 1977 | return btf; |
588cd7ef | 1978 | } |
2357672c KKD |
1979 | |
1980 | if (!btf_is_module(btf)) { | |
1981 | verbose(env, "BTF fd for kfunc is not a module BTF\n"); | |
1982 | btf_put(btf); | |
1983 | return ERR_PTR(-EINVAL); | |
1984 | } | |
1985 | ||
1986 | mod = btf_try_get_module(btf); | |
1987 | if (!mod) { | |
1988 | btf_put(btf); | |
1989 | return ERR_PTR(-ENXIO); | |
1990 | } | |
1991 | ||
1992 | b = &tab->descs[tab->nr_descs++]; | |
1993 | b->btf = btf; | |
1994 | b->module = mod; | |
1995 | b->offset = offset; | |
1996 | ||
1997 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
1998 | kfunc_btf_cmp_by_off, NULL); | |
1999 | } | |
2357672c | 2000 | return b->btf; |
e6ac2450 MKL |
2001 | } |
2002 | ||
2357672c KKD |
2003 | void bpf_free_kfunc_btf_tab(struct bpf_kfunc_btf_tab *tab) |
2004 | { | |
2005 | if (!tab) | |
2006 | return; | |
2007 | ||
2008 | while (tab->nr_descs--) { | |
2009 | module_put(tab->descs[tab->nr_descs].module); | |
2010 | btf_put(tab->descs[tab->nr_descs].btf); | |
2011 | } | |
2012 | kfree(tab); | |
2013 | } | |
2014 | ||
43bf0878 | 2015 | static struct btf *find_kfunc_desc_btf(struct bpf_verifier_env *env, s16 offset) |
2357672c | 2016 | { |
2357672c KKD |
2017 | if (offset) { |
2018 | if (offset < 0) { | |
2019 | /* In the future, this can be allowed to increase limit | |
2020 | * of fd index into fd_array, interpreted as u16. | |
2021 | */ | |
2022 | verbose(env, "negative offset disallowed for kernel module function call\n"); | |
2023 | return ERR_PTR(-EINVAL); | |
2024 | } | |
2025 | ||
b202d844 | 2026 | return __find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2027 | } |
2028 | return btf_vmlinux ?: ERR_PTR(-ENOENT); | |
e6ac2450 MKL |
2029 | } |
2030 | ||
2357672c | 2031 | static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id, s16 offset) |
e6ac2450 MKL |
2032 | { |
2033 | const struct btf_type *func, *func_proto; | |
2357672c | 2034 | struct bpf_kfunc_btf_tab *btf_tab; |
e6ac2450 MKL |
2035 | struct bpf_kfunc_desc_tab *tab; |
2036 | struct bpf_prog_aux *prog_aux; | |
2037 | struct bpf_kfunc_desc *desc; | |
2038 | const char *func_name; | |
2357672c | 2039 | struct btf *desc_btf; |
8cbf062a | 2040 | unsigned long call_imm; |
e6ac2450 MKL |
2041 | unsigned long addr; |
2042 | int err; | |
2043 | ||
2044 | prog_aux = env->prog->aux; | |
2045 | tab = prog_aux->kfunc_tab; | |
2357672c | 2046 | btf_tab = prog_aux->kfunc_btf_tab; |
e6ac2450 MKL |
2047 | if (!tab) { |
2048 | if (!btf_vmlinux) { | |
2049 | verbose(env, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
2050 | return -ENOTSUPP; | |
2051 | } | |
2052 | ||
2053 | if (!env->prog->jit_requested) { | |
2054 | verbose(env, "JIT is required for calling kernel function\n"); | |
2055 | return -ENOTSUPP; | |
2056 | } | |
2057 | ||
2058 | if (!bpf_jit_supports_kfunc_call()) { | |
2059 | verbose(env, "JIT does not support calling kernel function\n"); | |
2060 | return -ENOTSUPP; | |
2061 | } | |
2062 | ||
2063 | if (!env->prog->gpl_compatible) { | |
2064 | verbose(env, "cannot call kernel function from non-GPL compatible program\n"); | |
2065 | return -EINVAL; | |
2066 | } | |
2067 | ||
2068 | tab = kzalloc(sizeof(*tab), GFP_KERNEL); | |
2069 | if (!tab) | |
2070 | return -ENOMEM; | |
2071 | prog_aux->kfunc_tab = tab; | |
2072 | } | |
2073 | ||
a5d82727 KKD |
2074 | /* func_id == 0 is always invalid, but instead of returning an error, be |
2075 | * conservative and wait until the code elimination pass before returning | |
2076 | * error, so that invalid calls that get pruned out can be in BPF programs | |
2077 | * loaded from userspace. It is also required that offset be untouched | |
2078 | * for such calls. | |
2079 | */ | |
2080 | if (!func_id && !offset) | |
2081 | return 0; | |
2082 | ||
2357672c KKD |
2083 | if (!btf_tab && offset) { |
2084 | btf_tab = kzalloc(sizeof(*btf_tab), GFP_KERNEL); | |
2085 | if (!btf_tab) | |
2086 | return -ENOMEM; | |
2087 | prog_aux->kfunc_btf_tab = btf_tab; | |
2088 | } | |
2089 | ||
43bf0878 | 2090 | desc_btf = find_kfunc_desc_btf(env, offset); |
2357672c KKD |
2091 | if (IS_ERR(desc_btf)) { |
2092 | verbose(env, "failed to find BTF for kernel function\n"); | |
2093 | return PTR_ERR(desc_btf); | |
2094 | } | |
2095 | ||
2096 | if (find_kfunc_desc(env->prog, func_id, offset)) | |
e6ac2450 MKL |
2097 | return 0; |
2098 | ||
2099 | if (tab->nr_descs == MAX_KFUNC_DESCS) { | |
2100 | verbose(env, "too many different kernel function calls\n"); | |
2101 | return -E2BIG; | |
2102 | } | |
2103 | ||
2357672c | 2104 | func = btf_type_by_id(desc_btf, func_id); |
e6ac2450 MKL |
2105 | if (!func || !btf_type_is_func(func)) { |
2106 | verbose(env, "kernel btf_id %u is not a function\n", | |
2107 | func_id); | |
2108 | return -EINVAL; | |
2109 | } | |
2357672c | 2110 | func_proto = btf_type_by_id(desc_btf, func->type); |
e6ac2450 MKL |
2111 | if (!func_proto || !btf_type_is_func_proto(func_proto)) { |
2112 | verbose(env, "kernel function btf_id %u does not have a valid func_proto\n", | |
2113 | func_id); | |
2114 | return -EINVAL; | |
2115 | } | |
2116 | ||
2357672c | 2117 | func_name = btf_name_by_offset(desc_btf, func->name_off); |
e6ac2450 MKL |
2118 | addr = kallsyms_lookup_name(func_name); |
2119 | if (!addr) { | |
2120 | verbose(env, "cannot find address for kernel function %s\n", | |
2121 | func_name); | |
2122 | return -EINVAL; | |
2123 | } | |
2124 | ||
8cbf062a HT |
2125 | call_imm = BPF_CALL_IMM(addr); |
2126 | /* Check whether or not the relative offset overflows desc->imm */ | |
2127 | if ((unsigned long)(s32)call_imm != call_imm) { | |
2128 | verbose(env, "address of kernel function %s is out of range\n", | |
2129 | func_name); | |
2130 | return -EINVAL; | |
2131 | } | |
2132 | ||
e6ac2450 MKL |
2133 | desc = &tab->descs[tab->nr_descs++]; |
2134 | desc->func_id = func_id; | |
8cbf062a | 2135 | desc->imm = call_imm; |
2357672c KKD |
2136 | desc->offset = offset; |
2137 | err = btf_distill_func_proto(&env->log, desc_btf, | |
e6ac2450 MKL |
2138 | func_proto, func_name, |
2139 | &desc->func_model); | |
2140 | if (!err) | |
2141 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2357672c | 2142 | kfunc_desc_cmp_by_id_off, NULL); |
e6ac2450 MKL |
2143 | return err; |
2144 | } | |
2145 | ||
2146 | static int kfunc_desc_cmp_by_imm(const void *a, const void *b) | |
2147 | { | |
2148 | const struct bpf_kfunc_desc *d0 = a; | |
2149 | const struct bpf_kfunc_desc *d1 = b; | |
2150 | ||
2151 | if (d0->imm > d1->imm) | |
2152 | return 1; | |
2153 | else if (d0->imm < d1->imm) | |
2154 | return -1; | |
2155 | return 0; | |
2156 | } | |
2157 | ||
2158 | static void sort_kfunc_descs_by_imm(struct bpf_prog *prog) | |
2159 | { | |
2160 | struct bpf_kfunc_desc_tab *tab; | |
2161 | ||
2162 | tab = prog->aux->kfunc_tab; | |
2163 | if (!tab) | |
2164 | return; | |
2165 | ||
2166 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
2167 | kfunc_desc_cmp_by_imm, NULL); | |
2168 | } | |
2169 | ||
2170 | bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) | |
2171 | { | |
2172 | return !!prog->aux->kfunc_tab; | |
2173 | } | |
2174 | ||
2175 | const struct btf_func_model * | |
2176 | bpf_jit_find_kfunc_model(const struct bpf_prog *prog, | |
2177 | const struct bpf_insn *insn) | |
2178 | { | |
2179 | const struct bpf_kfunc_desc desc = { | |
2180 | .imm = insn->imm, | |
2181 | }; | |
2182 | const struct bpf_kfunc_desc *res; | |
2183 | struct bpf_kfunc_desc_tab *tab; | |
2184 | ||
2185 | tab = prog->aux->kfunc_tab; | |
2186 | res = bsearch(&desc, tab->descs, tab->nr_descs, | |
2187 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm); | |
2188 | ||
2189 | return res ? &res->func_model : NULL; | |
2190 | } | |
2191 | ||
2192 | static int add_subprog_and_kfunc(struct bpf_verifier_env *env) | |
cc8b0b92 | 2193 | { |
9c8105bd | 2194 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 | 2195 | struct bpf_insn *insn = env->prog->insnsi; |
e6ac2450 | 2196 | int i, ret, insn_cnt = env->prog->len; |
cc8b0b92 | 2197 | |
f910cefa JW |
2198 | /* Add entry function. */ |
2199 | ret = add_subprog(env, 0); | |
e6ac2450 | 2200 | if (ret) |
f910cefa JW |
2201 | return ret; |
2202 | ||
e6ac2450 MKL |
2203 | for (i = 0; i < insn_cnt; i++, insn++) { |
2204 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn) && | |
2205 | !bpf_pseudo_kfunc_call(insn)) | |
cc8b0b92 | 2206 | continue; |
e6ac2450 | 2207 | |
2c78ee89 | 2208 | if (!env->bpf_capable) { |
e6ac2450 | 2209 | verbose(env, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); |
cc8b0b92 AS |
2210 | return -EPERM; |
2211 | } | |
e6ac2450 | 2212 | |
3990ed4c | 2213 | if (bpf_pseudo_func(insn) || bpf_pseudo_call(insn)) |
e6ac2450 | 2214 | ret = add_subprog(env, i + insn->imm + 1); |
3990ed4c | 2215 | else |
2357672c | 2216 | ret = add_kfunc_call(env, insn->imm, insn->off); |
e6ac2450 | 2217 | |
cc8b0b92 AS |
2218 | if (ret < 0) |
2219 | return ret; | |
2220 | } | |
2221 | ||
4cb3d99c JW |
2222 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
2223 | * logic. 'subprog_cnt' should not be increased. | |
2224 | */ | |
2225 | subprog[env->subprog_cnt].start = insn_cnt; | |
2226 | ||
06ee7115 | 2227 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 2228 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 2229 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 | 2230 | |
e6ac2450 MKL |
2231 | return 0; |
2232 | } | |
2233 | ||
2234 | static int check_subprogs(struct bpf_verifier_env *env) | |
2235 | { | |
2236 | int i, subprog_start, subprog_end, off, cur_subprog = 0; | |
2237 | struct bpf_subprog_info *subprog = env->subprog_info; | |
2238 | struct bpf_insn *insn = env->prog->insnsi; | |
2239 | int insn_cnt = env->prog->len; | |
2240 | ||
cc8b0b92 | 2241 | /* now check that all jumps are within the same subprog */ |
4cb3d99c JW |
2242 | subprog_start = subprog[cur_subprog].start; |
2243 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
2244 | for (i = 0; i < insn_cnt; i++) { |
2245 | u8 code = insn[i].code; | |
2246 | ||
7f6e4312 MF |
2247 | if (code == (BPF_JMP | BPF_CALL) && |
2248 | insn[i].imm == BPF_FUNC_tail_call && | |
2249 | insn[i].src_reg != BPF_PSEUDO_CALL) | |
2250 | subprog[cur_subprog].has_tail_call = true; | |
09b28d76 AS |
2251 | if (BPF_CLASS(code) == BPF_LD && |
2252 | (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND)) | |
2253 | subprog[cur_subprog].has_ld_abs = true; | |
092ed096 | 2254 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
2255 | goto next; |
2256 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
2257 | goto next; | |
2258 | off = i + insn[i].off + 1; | |
2259 | if (off < subprog_start || off >= subprog_end) { | |
2260 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
2261 | return -EINVAL; | |
2262 | } | |
2263 | next: | |
2264 | if (i == subprog_end - 1) { | |
2265 | /* to avoid fall-through from one subprog into another | |
2266 | * the last insn of the subprog should be either exit | |
2267 | * or unconditional jump back | |
2268 | */ | |
2269 | if (code != (BPF_JMP | BPF_EXIT) && | |
2270 | code != (BPF_JMP | BPF_JA)) { | |
2271 | verbose(env, "last insn is not an exit or jmp\n"); | |
2272 | return -EINVAL; | |
2273 | } | |
2274 | subprog_start = subprog_end; | |
4cb3d99c JW |
2275 | cur_subprog++; |
2276 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 2277 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
2278 | } |
2279 | } | |
2280 | return 0; | |
2281 | } | |
2282 | ||
679c782d EC |
2283 | /* Parentage chain of this register (or stack slot) should take care of all |
2284 | * issues like callee-saved registers, stack slot allocation time, etc. | |
2285 | */ | |
f4d7e40a | 2286 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 2287 | const struct bpf_reg_state *state, |
5327ed3d | 2288 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
2289 | { |
2290 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 2291 | int cnt = 0; |
dc503a8a EC |
2292 | |
2293 | while (parent) { | |
2294 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 2295 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 2296 | break; |
9242b5f5 AS |
2297 | if (parent->live & REG_LIVE_DONE) { |
2298 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
c25b2ae1 | 2299 | reg_type_str(env, parent->type), |
9242b5f5 AS |
2300 | parent->var_off.value, parent->off); |
2301 | return -EFAULT; | |
2302 | } | |
5327ed3d JW |
2303 | /* The first condition is more likely to be true than the |
2304 | * second, checked it first. | |
2305 | */ | |
2306 | if ((parent->live & REG_LIVE_READ) == flag || | |
2307 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
2308 | /* The parentage chain never changes and |
2309 | * this parent was already marked as LIVE_READ. | |
2310 | * There is no need to keep walking the chain again and | |
2311 | * keep re-marking all parents as LIVE_READ. | |
2312 | * This case happens when the same register is read | |
2313 | * multiple times without writes into it in-between. | |
5327ed3d JW |
2314 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
2315 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
2316 | */ |
2317 | break; | |
dc503a8a | 2318 | /* ... then we depend on parent's value */ |
5327ed3d JW |
2319 | parent->live |= flag; |
2320 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
2321 | if (flag == REG_LIVE_READ64) | |
2322 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
2323 | state = parent; |
2324 | parent = state->parent; | |
f4d7e40a | 2325 | writes = true; |
06ee7115 | 2326 | cnt++; |
dc503a8a | 2327 | } |
06ee7115 AS |
2328 | |
2329 | if (env->longest_mark_read_walk < cnt) | |
2330 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 2331 | return 0; |
dc503a8a EC |
2332 | } |
2333 | ||
5327ed3d JW |
2334 | /* This function is supposed to be used by the following 32-bit optimization |
2335 | * code only. It returns TRUE if the source or destination register operates | |
2336 | * on 64-bit, otherwise return FALSE. | |
2337 | */ | |
2338 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
2339 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
2340 | { | |
2341 | u8 code, class, op; | |
2342 | ||
2343 | code = insn->code; | |
2344 | class = BPF_CLASS(code); | |
2345 | op = BPF_OP(code); | |
2346 | if (class == BPF_JMP) { | |
2347 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
2348 | * conservatively. | |
2349 | */ | |
2350 | if (op == BPF_EXIT) | |
2351 | return true; | |
2352 | if (op == BPF_CALL) { | |
2353 | /* BPF to BPF call will reach here because of marking | |
2354 | * caller saved clobber with DST_OP_NO_MARK for which we | |
2355 | * don't care the register def because they are anyway | |
2356 | * marked as NOT_INIT already. | |
2357 | */ | |
2358 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
2359 | return false; | |
2360 | /* Helper call will reach here because of arg type | |
2361 | * check, conservatively return TRUE. | |
2362 | */ | |
2363 | if (t == SRC_OP) | |
2364 | return true; | |
2365 | ||
2366 | return false; | |
2367 | } | |
2368 | } | |
2369 | ||
2370 | if (class == BPF_ALU64 || class == BPF_JMP || | |
2371 | /* BPF_END always use BPF_ALU class. */ | |
2372 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) | |
2373 | return true; | |
2374 | ||
2375 | if (class == BPF_ALU || class == BPF_JMP32) | |
2376 | return false; | |
2377 | ||
2378 | if (class == BPF_LDX) { | |
2379 | if (t != SRC_OP) | |
2380 | return BPF_SIZE(code) == BPF_DW; | |
2381 | /* LDX source must be ptr. */ | |
2382 | return true; | |
2383 | } | |
2384 | ||
2385 | if (class == BPF_STX) { | |
83a28819 IL |
2386 | /* BPF_STX (including atomic variants) has multiple source |
2387 | * operands, one of which is a ptr. Check whether the caller is | |
2388 | * asking about it. | |
2389 | */ | |
2390 | if (t == SRC_OP && reg->type != SCALAR_VALUE) | |
5327ed3d JW |
2391 | return true; |
2392 | return BPF_SIZE(code) == BPF_DW; | |
2393 | } | |
2394 | ||
2395 | if (class == BPF_LD) { | |
2396 | u8 mode = BPF_MODE(code); | |
2397 | ||
2398 | /* LD_IMM64 */ | |
2399 | if (mode == BPF_IMM) | |
2400 | return true; | |
2401 | ||
2402 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
2403 | if (t != SRC_OP) | |
2404 | return false; | |
2405 | ||
2406 | /* Implicit ctx ptr. */ | |
2407 | if (regno == BPF_REG_6) | |
2408 | return true; | |
2409 | ||
2410 | /* Explicit source could be any width. */ | |
2411 | return true; | |
2412 | } | |
2413 | ||
2414 | if (class == BPF_ST) | |
2415 | /* The only source register for BPF_ST is a ptr. */ | |
2416 | return true; | |
2417 | ||
2418 | /* Conservatively return true at default. */ | |
2419 | return true; | |
2420 | } | |
2421 | ||
83a28819 IL |
2422 | /* Return the regno defined by the insn, or -1. */ |
2423 | static int insn_def_regno(const struct bpf_insn *insn) | |
b325fbca | 2424 | { |
83a28819 IL |
2425 | switch (BPF_CLASS(insn->code)) { |
2426 | case BPF_JMP: | |
2427 | case BPF_JMP32: | |
2428 | case BPF_ST: | |
2429 | return -1; | |
2430 | case BPF_STX: | |
2431 | if (BPF_MODE(insn->code) == BPF_ATOMIC && | |
2432 | (insn->imm & BPF_FETCH)) { | |
2433 | if (insn->imm == BPF_CMPXCHG) | |
2434 | return BPF_REG_0; | |
2435 | else | |
2436 | return insn->src_reg; | |
2437 | } else { | |
2438 | return -1; | |
2439 | } | |
2440 | default: | |
2441 | return insn->dst_reg; | |
2442 | } | |
b325fbca JW |
2443 | } |
2444 | ||
2445 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
2446 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
2447 | { | |
83a28819 IL |
2448 | int dst_reg = insn_def_regno(insn); |
2449 | ||
2450 | if (dst_reg == -1) | |
b325fbca JW |
2451 | return false; |
2452 | ||
83a28819 | 2453 | return !is_reg64(env, insn, dst_reg, NULL, DST_OP); |
b325fbca JW |
2454 | } |
2455 | ||
5327ed3d JW |
2456 | static void mark_insn_zext(struct bpf_verifier_env *env, |
2457 | struct bpf_reg_state *reg) | |
2458 | { | |
2459 | s32 def_idx = reg->subreg_def; | |
2460 | ||
2461 | if (def_idx == DEF_NOT_SUBREG) | |
2462 | return; | |
2463 | ||
2464 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
2465 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
2466 | reg->subreg_def = DEF_NOT_SUBREG; | |
2467 | } | |
2468 | ||
dc503a8a | 2469 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
2470 | enum reg_arg_type t) |
2471 | { | |
f4d7e40a AS |
2472 | struct bpf_verifier_state *vstate = env->cur_state; |
2473 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 2474 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 2475 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 2476 | bool rw64; |
dc503a8a | 2477 | |
17a52670 | 2478 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 2479 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
2480 | return -EINVAL; |
2481 | } | |
2482 | ||
0f55f9ed CL |
2483 | mark_reg_scratched(env, regno); |
2484 | ||
c342dc10 | 2485 | reg = ®s[regno]; |
5327ed3d | 2486 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
2487 | if (t == SRC_OP) { |
2488 | /* check whether register used as source operand can be read */ | |
c342dc10 | 2489 | if (reg->type == NOT_INIT) { |
61bd5218 | 2490 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
2491 | return -EACCES; |
2492 | } | |
679c782d | 2493 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
2494 | if (regno == BPF_REG_FP) |
2495 | return 0; | |
2496 | ||
5327ed3d JW |
2497 | if (rw64) |
2498 | mark_insn_zext(env, reg); | |
2499 | ||
2500 | return mark_reg_read(env, reg, reg->parent, | |
2501 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
2502 | } else { |
2503 | /* check whether register used as dest operand can be written to */ | |
2504 | if (regno == BPF_REG_FP) { | |
61bd5218 | 2505 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
2506 | return -EACCES; |
2507 | } | |
c342dc10 | 2508 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 2509 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 2510 | if (t == DST_OP) |
61bd5218 | 2511 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
2512 | } |
2513 | return 0; | |
2514 | } | |
2515 | ||
b5dc0163 AS |
2516 | /* for any branch, call, exit record the history of jmps in the given state */ |
2517 | static int push_jmp_history(struct bpf_verifier_env *env, | |
2518 | struct bpf_verifier_state *cur) | |
2519 | { | |
2520 | u32 cnt = cur->jmp_history_cnt; | |
2521 | struct bpf_idx_pair *p; | |
2522 | ||
2523 | cnt++; | |
2524 | p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER); | |
2525 | if (!p) | |
2526 | return -ENOMEM; | |
2527 | p[cnt - 1].idx = env->insn_idx; | |
2528 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
2529 | cur->jmp_history = p; | |
2530 | cur->jmp_history_cnt = cnt; | |
2531 | return 0; | |
2532 | } | |
2533 | ||
2534 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
2535 | * history then previous instruction came from straight line execution. | |
2536 | */ | |
2537 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
2538 | u32 *history) | |
2539 | { | |
2540 | u32 cnt = *history; | |
2541 | ||
2542 | if (cnt && st->jmp_history[cnt - 1].idx == i) { | |
2543 | i = st->jmp_history[cnt - 1].prev_idx; | |
2544 | (*history)--; | |
2545 | } else { | |
2546 | i--; | |
2547 | } | |
2548 | return i; | |
2549 | } | |
2550 | ||
e6ac2450 MKL |
2551 | static const char *disasm_kfunc_name(void *data, const struct bpf_insn *insn) |
2552 | { | |
2553 | const struct btf_type *func; | |
2357672c | 2554 | struct btf *desc_btf; |
e6ac2450 MKL |
2555 | |
2556 | if (insn->src_reg != BPF_PSEUDO_KFUNC_CALL) | |
2557 | return NULL; | |
2558 | ||
43bf0878 | 2559 | desc_btf = find_kfunc_desc_btf(data, insn->off); |
2357672c KKD |
2560 | if (IS_ERR(desc_btf)) |
2561 | return "<error>"; | |
2562 | ||
2563 | func = btf_type_by_id(desc_btf, insn->imm); | |
2564 | return btf_name_by_offset(desc_btf, func->name_off); | |
e6ac2450 MKL |
2565 | } |
2566 | ||
b5dc0163 AS |
2567 | /* For given verifier state backtrack_insn() is called from the last insn to |
2568 | * the first insn. Its purpose is to compute a bitmask of registers and | |
2569 | * stack slots that needs precision in the parent verifier state. | |
2570 | */ | |
2571 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, | |
2572 | u32 *reg_mask, u64 *stack_mask) | |
2573 | { | |
2574 | const struct bpf_insn_cbs cbs = { | |
e6ac2450 | 2575 | .cb_call = disasm_kfunc_name, |
b5dc0163 AS |
2576 | .cb_print = verbose, |
2577 | .private_data = env, | |
2578 | }; | |
2579 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
2580 | u8 class = BPF_CLASS(insn->code); | |
2581 | u8 opcode = BPF_OP(insn->code); | |
2582 | u8 mode = BPF_MODE(insn->code); | |
2583 | u32 dreg = 1u << insn->dst_reg; | |
2584 | u32 sreg = 1u << insn->src_reg; | |
2585 | u32 spi; | |
2586 | ||
2587 | if (insn->code == 0) | |
2588 | return 0; | |
496f3324 | 2589 | if (env->log.level & BPF_LOG_LEVEL2) { |
b5dc0163 AS |
2590 | verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask); |
2591 | verbose(env, "%d: ", idx); | |
2592 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
2593 | } | |
2594 | ||
2595 | if (class == BPF_ALU || class == BPF_ALU64) { | |
2596 | if (!(*reg_mask & dreg)) | |
2597 | return 0; | |
2598 | if (opcode == BPF_MOV) { | |
2599 | if (BPF_SRC(insn->code) == BPF_X) { | |
2600 | /* dreg = sreg | |
2601 | * dreg needs precision after this insn | |
2602 | * sreg needs precision before this insn | |
2603 | */ | |
2604 | *reg_mask &= ~dreg; | |
2605 | *reg_mask |= sreg; | |
2606 | } else { | |
2607 | /* dreg = K | |
2608 | * dreg needs precision after this insn. | |
2609 | * Corresponding register is already marked | |
2610 | * as precise=true in this verifier state. | |
2611 | * No further markings in parent are necessary | |
2612 | */ | |
2613 | *reg_mask &= ~dreg; | |
2614 | } | |
2615 | } else { | |
2616 | if (BPF_SRC(insn->code) == BPF_X) { | |
2617 | /* dreg += sreg | |
2618 | * both dreg and sreg need precision | |
2619 | * before this insn | |
2620 | */ | |
2621 | *reg_mask |= sreg; | |
2622 | } /* else dreg += K | |
2623 | * dreg still needs precision before this insn | |
2624 | */ | |
2625 | } | |
2626 | } else if (class == BPF_LDX) { | |
2627 | if (!(*reg_mask & dreg)) | |
2628 | return 0; | |
2629 | *reg_mask &= ~dreg; | |
2630 | ||
2631 | /* scalars can only be spilled into stack w/o losing precision. | |
2632 | * Load from any other memory can be zero extended. | |
2633 | * The desire to keep that precision is already indicated | |
2634 | * by 'precise' mark in corresponding register of this state. | |
2635 | * No further tracking necessary. | |
2636 | */ | |
2637 | if (insn->src_reg != BPF_REG_FP) | |
2638 | return 0; | |
b5dc0163 AS |
2639 | |
2640 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
2641 | * that [fp - off] slot contains scalar that needs to be | |
2642 | * tracked with precision | |
2643 | */ | |
2644 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
2645 | if (spi >= 64) { | |
2646 | verbose(env, "BUG spi %d\n", spi); | |
2647 | WARN_ONCE(1, "verifier backtracking bug"); | |
2648 | return -EFAULT; | |
2649 | } | |
2650 | *stack_mask |= 1ull << spi; | |
b3b50f05 | 2651 | } else if (class == BPF_STX || class == BPF_ST) { |
b5dc0163 | 2652 | if (*reg_mask & dreg) |
b3b50f05 | 2653 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
2654 | * to access memory. It means backtracking |
2655 | * encountered a case of pointer subtraction. | |
2656 | */ | |
2657 | return -ENOTSUPP; | |
2658 | /* scalars can only be spilled into stack */ | |
2659 | if (insn->dst_reg != BPF_REG_FP) | |
2660 | return 0; | |
b5dc0163 AS |
2661 | spi = (-insn->off - 1) / BPF_REG_SIZE; |
2662 | if (spi >= 64) { | |
2663 | verbose(env, "BUG spi %d\n", spi); | |
2664 | WARN_ONCE(1, "verifier backtracking bug"); | |
2665 | return -EFAULT; | |
2666 | } | |
2667 | if (!(*stack_mask & (1ull << spi))) | |
2668 | return 0; | |
2669 | *stack_mask &= ~(1ull << spi); | |
b3b50f05 AN |
2670 | if (class == BPF_STX) |
2671 | *reg_mask |= sreg; | |
b5dc0163 AS |
2672 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
2673 | if (opcode == BPF_CALL) { | |
2674 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
2675 | return -ENOTSUPP; | |
be2ef816 AN |
2676 | /* BPF helpers that invoke callback subprogs are |
2677 | * equivalent to BPF_PSEUDO_CALL above | |
2678 | */ | |
2679 | if (insn->src_reg == 0 && is_callback_calling_function(insn->imm)) | |
2680 | return -ENOTSUPP; | |
b5dc0163 AS |
2681 | /* regular helper call sets R0 */ |
2682 | *reg_mask &= ~1; | |
2683 | if (*reg_mask & 0x3f) { | |
2684 | /* if backtracing was looking for registers R1-R5 | |
2685 | * they should have been found already. | |
2686 | */ | |
2687 | verbose(env, "BUG regs %x\n", *reg_mask); | |
2688 | WARN_ONCE(1, "verifier backtracking bug"); | |
2689 | return -EFAULT; | |
2690 | } | |
2691 | } else if (opcode == BPF_EXIT) { | |
2692 | return -ENOTSUPP; | |
2693 | } | |
2694 | } else if (class == BPF_LD) { | |
2695 | if (!(*reg_mask & dreg)) | |
2696 | return 0; | |
2697 | *reg_mask &= ~dreg; | |
2698 | /* It's ld_imm64 or ld_abs or ld_ind. | |
2699 | * For ld_imm64 no further tracking of precision | |
2700 | * into parent is necessary | |
2701 | */ | |
2702 | if (mode == BPF_IND || mode == BPF_ABS) | |
2703 | /* to be analyzed */ | |
2704 | return -ENOTSUPP; | |
b5dc0163 AS |
2705 | } |
2706 | return 0; | |
2707 | } | |
2708 | ||
2709 | /* the scalar precision tracking algorithm: | |
2710 | * . at the start all registers have precise=false. | |
2711 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
2712 | * . once precise value of the scalar register is used in: | |
2713 | * . ptr + scalar alu | |
2714 | * . if (scalar cond K|scalar) | |
2715 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
2716 | * backtrack through the verifier states and mark all registers and | |
2717 | * stack slots with spilled constants that these scalar regisers | |
2718 | * should be precise. | |
2719 | * . during state pruning two registers (or spilled stack slots) | |
2720 | * are equivalent if both are not precise. | |
2721 | * | |
2722 | * Note the verifier cannot simply walk register parentage chain, | |
2723 | * since many different registers and stack slots could have been | |
2724 | * used to compute single precise scalar. | |
2725 | * | |
2726 | * The approach of starting with precise=true for all registers and then | |
2727 | * backtrack to mark a register as not precise when the verifier detects | |
2728 | * that program doesn't care about specific value (e.g., when helper | |
2729 | * takes register as ARG_ANYTHING parameter) is not safe. | |
2730 | * | |
2731 | * It's ok to walk single parentage chain of the verifier states. | |
2732 | * It's possible that this backtracking will go all the way till 1st insn. | |
2733 | * All other branches will be explored for needing precision later. | |
2734 | * | |
2735 | * The backtracking needs to deal with cases like: | |
2736 | * 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) | |
2737 | * r9 -= r8 | |
2738 | * r5 = r9 | |
2739 | * if r5 > 0x79f goto pc+7 | |
2740 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
2741 | * r5 += 1 | |
2742 | * ... | |
2743 | * call bpf_perf_event_output#25 | |
2744 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
2745 | * | |
2746 | * and this case: | |
2747 | * r6 = 1 | |
2748 | * call foo // uses callee's r6 inside to compute r0 | |
2749 | * r0 += r6 | |
2750 | * if r0 == 0 goto | |
2751 | * | |
2752 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
2753 | * | |
2754 | * Also if parent's curframe > frame where backtracking started, | |
2755 | * the verifier need to mark registers in both frames, otherwise callees | |
2756 | * may incorrectly prune callers. This is similar to | |
2757 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
2758 | * | |
2759 | * For now backtracking falls back into conservative marking. | |
2760 | */ | |
2761 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
2762 | struct bpf_verifier_state *st) | |
2763 | { | |
2764 | struct bpf_func_state *func; | |
2765 | struct bpf_reg_state *reg; | |
2766 | int i, j; | |
2767 | ||
2768 | /* big hammer: mark all scalars precise in this path. | |
2769 | * pop_stack may still get !precise scalars. | |
f63181b6 AN |
2770 | * We also skip current state and go straight to first parent state, |
2771 | * because precision markings in current non-checkpointed state are | |
2772 | * not needed. See why in the comment in __mark_chain_precision below. | |
b5dc0163 | 2773 | */ |
f63181b6 | 2774 | for (st = st->parent; st; st = st->parent) { |
b5dc0163 AS |
2775 | for (i = 0; i <= st->curframe; i++) { |
2776 | func = st->frame[i]; | |
2777 | for (j = 0; j < BPF_REG_FP; j++) { | |
2778 | reg = &func->regs[j]; | |
2779 | if (reg->type != SCALAR_VALUE) | |
2780 | continue; | |
2781 | reg->precise = true; | |
2782 | } | |
2783 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
27113c59 | 2784 | if (!is_spilled_reg(&func->stack[j])) |
b5dc0163 AS |
2785 | continue; |
2786 | reg = &func->stack[j].spilled_ptr; | |
2787 | if (reg->type != SCALAR_VALUE) | |
2788 | continue; | |
2789 | reg->precise = true; | |
2790 | } | |
2791 | } | |
f63181b6 | 2792 | } |
b5dc0163 AS |
2793 | } |
2794 | ||
7a830b53 AN |
2795 | static void mark_all_scalars_imprecise(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
2796 | { | |
2797 | struct bpf_func_state *func; | |
2798 | struct bpf_reg_state *reg; | |
2799 | int i, j; | |
2800 | ||
2801 | for (i = 0; i <= st->curframe; i++) { | |
2802 | func = st->frame[i]; | |
2803 | for (j = 0; j < BPF_REG_FP; j++) { | |
2804 | reg = &func->regs[j]; | |
2805 | if (reg->type != SCALAR_VALUE) | |
2806 | continue; | |
2807 | reg->precise = false; | |
2808 | } | |
2809 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
2810 | if (!is_spilled_reg(&func->stack[j])) | |
2811 | continue; | |
2812 | reg = &func->stack[j].spilled_ptr; | |
2813 | if (reg->type != SCALAR_VALUE) | |
2814 | continue; | |
2815 | reg->precise = false; | |
2816 | } | |
2817 | } | |
2818 | } | |
2819 | ||
f63181b6 AN |
2820 | /* |
2821 | * __mark_chain_precision() backtracks BPF program instruction sequence and | |
2822 | * chain of verifier states making sure that register *regno* (if regno >= 0) | |
2823 | * and/or stack slot *spi* (if spi >= 0) are marked as precisely tracked | |
2824 | * SCALARS, as well as any other registers and slots that contribute to | |
2825 | * a tracked state of given registers/stack slots, depending on specific BPF | |
2826 | * assembly instructions (see backtrack_insns() for exact instruction handling | |
2827 | * logic). This backtracking relies on recorded jmp_history and is able to | |
2828 | * traverse entire chain of parent states. This process ends only when all the | |
2829 | * necessary registers/slots and their transitive dependencies are marked as | |
2830 | * precise. | |
2831 | * | |
2832 | * One important and subtle aspect is that precise marks *do not matter* in | |
2833 | * the currently verified state (current state). It is important to understand | |
2834 | * why this is the case. | |
2835 | * | |
2836 | * First, note that current state is the state that is not yet "checkpointed", | |
2837 | * i.e., it is not yet put into env->explored_states, and it has no children | |
2838 | * states as well. It's ephemeral, and can end up either a) being discarded if | |
2839 | * compatible explored state is found at some point or BPF_EXIT instruction is | |
2840 | * reached or b) checkpointed and put into env->explored_states, branching out | |
2841 | * into one or more children states. | |
2842 | * | |
2843 | * In the former case, precise markings in current state are completely | |
2844 | * ignored by state comparison code (see regsafe() for details). Only | |
2845 | * checkpointed ("old") state precise markings are important, and if old | |
2846 | * state's register/slot is precise, regsafe() assumes current state's | |
2847 | * register/slot as precise and checks value ranges exactly and precisely. If | |
2848 | * states turn out to be compatible, current state's necessary precise | |
2849 | * markings and any required parent states' precise markings are enforced | |
2850 | * after the fact with propagate_precision() logic, after the fact. But it's | |
2851 | * important to realize that in this case, even after marking current state | |
2852 | * registers/slots as precise, we immediately discard current state. So what | |
2853 | * actually matters is any of the precise markings propagated into current | |
2854 | * state's parent states, which are always checkpointed (due to b) case above). | |
2855 | * As such, for scenario a) it doesn't matter if current state has precise | |
2856 | * markings set or not. | |
2857 | * | |
2858 | * Now, for the scenario b), checkpointing and forking into child(ren) | |
2859 | * state(s). Note that before current state gets to checkpointing step, any | |
2860 | * processed instruction always assumes precise SCALAR register/slot | |
2861 | * knowledge: if precise value or range is useful to prune jump branch, BPF | |
2862 | * verifier takes this opportunity enthusiastically. Similarly, when | |
2863 | * register's value is used to calculate offset or memory address, exact | |
2864 | * knowledge of SCALAR range is assumed, checked, and enforced. So, similar to | |
2865 | * what we mentioned above about state comparison ignoring precise markings | |
2866 | * during state comparison, BPF verifier ignores and also assumes precise | |
2867 | * markings *at will* during instruction verification process. But as verifier | |
2868 | * assumes precision, it also propagates any precision dependencies across | |
2869 | * parent states, which are not yet finalized, so can be further restricted | |
2870 | * based on new knowledge gained from restrictions enforced by their children | |
2871 | * states. This is so that once those parent states are finalized, i.e., when | |
2872 | * they have no more active children state, state comparison logic in | |
2873 | * is_state_visited() would enforce strict and precise SCALAR ranges, if | |
2874 | * required for correctness. | |
2875 | * | |
2876 | * To build a bit more intuition, note also that once a state is checkpointed, | |
2877 | * the path we took to get to that state is not important. This is crucial | |
2878 | * property for state pruning. When state is checkpointed and finalized at | |
2879 | * some instruction index, it can be correctly and safely used to "short | |
2880 | * circuit" any *compatible* state that reaches exactly the same instruction | |
2881 | * index. I.e., if we jumped to that instruction from a completely different | |
2882 | * code path than original finalized state was derived from, it doesn't | |
2883 | * matter, current state can be discarded because from that instruction | |
2884 | * forward having a compatible state will ensure we will safely reach the | |
2885 | * exit. States describe preconditions for further exploration, but completely | |
2886 | * forget the history of how we got here. | |
2887 | * | |
2888 | * This also means that even if we needed precise SCALAR range to get to | |
2889 | * finalized state, but from that point forward *that same* SCALAR register is | |
2890 | * never used in a precise context (i.e., it's precise value is not needed for | |
2891 | * correctness), it's correct and safe to mark such register as "imprecise" | |
2892 | * (i.e., precise marking set to false). This is what we rely on when we do | |
2893 | * not set precise marking in current state. If no child state requires | |
2894 | * precision for any given SCALAR register, it's safe to dictate that it can | |
2895 | * be imprecise. If any child state does require this register to be precise, | |
2896 | * we'll mark it precise later retroactively during precise markings | |
2897 | * propagation from child state to parent states. | |
7a830b53 AN |
2898 | * |
2899 | * Skipping precise marking setting in current state is a mild version of | |
2900 | * relying on the above observation. But we can utilize this property even | |
2901 | * more aggressively by proactively forgetting any precise marking in the | |
2902 | * current state (which we inherited from the parent state), right before we | |
2903 | * checkpoint it and branch off into new child state. This is done by | |
2904 | * mark_all_scalars_imprecise() to hopefully get more permissive and generic | |
2905 | * finalized states which help in short circuiting more future states. | |
f63181b6 | 2906 | */ |
529409ea | 2907 | static int __mark_chain_precision(struct bpf_verifier_env *env, int frame, int regno, |
a3ce685d | 2908 | int spi) |
b5dc0163 AS |
2909 | { |
2910 | struct bpf_verifier_state *st = env->cur_state; | |
2911 | int first_idx = st->first_insn_idx; | |
2912 | int last_idx = env->insn_idx; | |
2913 | struct bpf_func_state *func; | |
2914 | struct bpf_reg_state *reg; | |
a3ce685d AS |
2915 | u32 reg_mask = regno >= 0 ? 1u << regno : 0; |
2916 | u64 stack_mask = spi >= 0 ? 1ull << spi : 0; | |
b5dc0163 | 2917 | bool skip_first = true; |
a3ce685d | 2918 | bool new_marks = false; |
b5dc0163 AS |
2919 | int i, err; |
2920 | ||
2c78ee89 | 2921 | if (!env->bpf_capable) |
b5dc0163 AS |
2922 | return 0; |
2923 | ||
f63181b6 AN |
2924 | /* Do sanity checks against current state of register and/or stack |
2925 | * slot, but don't set precise flag in current state, as precision | |
2926 | * tracking in the current state is unnecessary. | |
2927 | */ | |
529409ea | 2928 | func = st->frame[frame]; |
a3ce685d AS |
2929 | if (regno >= 0) { |
2930 | reg = &func->regs[regno]; | |
2931 | if (reg->type != SCALAR_VALUE) { | |
2932 | WARN_ONCE(1, "backtracing misuse"); | |
2933 | return -EFAULT; | |
2934 | } | |
f63181b6 | 2935 | new_marks = true; |
b5dc0163 | 2936 | } |
b5dc0163 | 2937 | |
a3ce685d | 2938 | while (spi >= 0) { |
27113c59 | 2939 | if (!is_spilled_reg(&func->stack[spi])) { |
a3ce685d AS |
2940 | stack_mask = 0; |
2941 | break; | |
2942 | } | |
2943 | reg = &func->stack[spi].spilled_ptr; | |
2944 | if (reg->type != SCALAR_VALUE) { | |
2945 | stack_mask = 0; | |
2946 | break; | |
2947 | } | |
f63181b6 | 2948 | new_marks = true; |
a3ce685d AS |
2949 | break; |
2950 | } | |
2951 | ||
2952 | if (!new_marks) | |
2953 | return 0; | |
2954 | if (!reg_mask && !stack_mask) | |
2955 | return 0; | |
be2ef816 | 2956 | |
b5dc0163 AS |
2957 | for (;;) { |
2958 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
2959 | u32 history = st->jmp_history_cnt; |
2960 | ||
496f3324 | 2961 | if (env->log.level & BPF_LOG_LEVEL2) |
b5dc0163 | 2962 | verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx); |
be2ef816 AN |
2963 | |
2964 | if (last_idx < 0) { | |
2965 | /* we are at the entry into subprog, which | |
2966 | * is expected for global funcs, but only if | |
2967 | * requested precise registers are R1-R5 | |
2968 | * (which are global func's input arguments) | |
2969 | */ | |
2970 | if (st->curframe == 0 && | |
2971 | st->frame[0]->subprogno > 0 && | |
2972 | st->frame[0]->callsite == BPF_MAIN_FUNC && | |
2973 | stack_mask == 0 && (reg_mask & ~0x3e) == 0) { | |
2974 | bitmap_from_u64(mask, reg_mask); | |
2975 | for_each_set_bit(i, mask, 32) { | |
2976 | reg = &st->frame[0]->regs[i]; | |
2977 | if (reg->type != SCALAR_VALUE) { | |
2978 | reg_mask &= ~(1u << i); | |
2979 | continue; | |
2980 | } | |
2981 | reg->precise = true; | |
2982 | } | |
2983 | return 0; | |
2984 | } | |
2985 | ||
2986 | verbose(env, "BUG backtracing func entry subprog %d reg_mask %x stack_mask %llx\n", | |
2987 | st->frame[0]->subprogno, reg_mask, stack_mask); | |
2988 | WARN_ONCE(1, "verifier backtracking bug"); | |
2989 | return -EFAULT; | |
2990 | } | |
2991 | ||
b5dc0163 AS |
2992 | for (i = last_idx;;) { |
2993 | if (skip_first) { | |
2994 | err = 0; | |
2995 | skip_first = false; | |
2996 | } else { | |
2997 | err = backtrack_insn(env, i, ®_mask, &stack_mask); | |
2998 | } | |
2999 | if (err == -ENOTSUPP) { | |
3000 | mark_all_scalars_precise(env, st); | |
3001 | return 0; | |
3002 | } else if (err) { | |
3003 | return err; | |
3004 | } | |
3005 | if (!reg_mask && !stack_mask) | |
3006 | /* Found assignment(s) into tracked register in this state. | |
3007 | * Since this state is already marked, just return. | |
3008 | * Nothing to be tracked further in the parent state. | |
3009 | */ | |
3010 | return 0; | |
3011 | if (i == first_idx) | |
3012 | break; | |
3013 | i = get_prev_insn_idx(st, i, &history); | |
3014 | if (i >= env->prog->len) { | |
3015 | /* This can happen if backtracking reached insn 0 | |
3016 | * and there are still reg_mask or stack_mask | |
3017 | * to backtrack. | |
3018 | * It means the backtracking missed the spot where | |
3019 | * particular register was initialized with a constant. | |
3020 | */ | |
3021 | verbose(env, "BUG backtracking idx %d\n", i); | |
3022 | WARN_ONCE(1, "verifier backtracking bug"); | |
3023 | return -EFAULT; | |
3024 | } | |
3025 | } | |
3026 | st = st->parent; | |
3027 | if (!st) | |
3028 | break; | |
3029 | ||
a3ce685d | 3030 | new_marks = false; |
529409ea | 3031 | func = st->frame[frame]; |
b5dc0163 AS |
3032 | bitmap_from_u64(mask, reg_mask); |
3033 | for_each_set_bit(i, mask, 32) { | |
3034 | reg = &func->regs[i]; | |
a3ce685d AS |
3035 | if (reg->type != SCALAR_VALUE) { |
3036 | reg_mask &= ~(1u << i); | |
b5dc0163 | 3037 | continue; |
a3ce685d | 3038 | } |
b5dc0163 AS |
3039 | if (!reg->precise) |
3040 | new_marks = true; | |
3041 | reg->precise = true; | |
3042 | } | |
3043 | ||
3044 | bitmap_from_u64(mask, stack_mask); | |
3045 | for_each_set_bit(i, mask, 64) { | |
3046 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
2339cd6c AS |
3047 | /* the sequence of instructions: |
3048 | * 2: (bf) r3 = r10 | |
3049 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
3050 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
3051 | * doesn't contain jmps. It's backtracked | |
3052 | * as a single block. | |
3053 | * During backtracking insn 3 is not recognized as | |
3054 | * stack access, so at the end of backtracking | |
3055 | * stack slot fp-8 is still marked in stack_mask. | |
3056 | * However the parent state may not have accessed | |
3057 | * fp-8 and it's "unallocated" stack space. | |
3058 | * In such case fallback to conservative. | |
b5dc0163 | 3059 | */ |
2339cd6c AS |
3060 | mark_all_scalars_precise(env, st); |
3061 | return 0; | |
b5dc0163 AS |
3062 | } |
3063 | ||
27113c59 | 3064 | if (!is_spilled_reg(&func->stack[i])) { |
a3ce685d | 3065 | stack_mask &= ~(1ull << i); |
b5dc0163 | 3066 | continue; |
a3ce685d | 3067 | } |
b5dc0163 | 3068 | reg = &func->stack[i].spilled_ptr; |
a3ce685d AS |
3069 | if (reg->type != SCALAR_VALUE) { |
3070 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 3071 | continue; |
a3ce685d | 3072 | } |
b5dc0163 AS |
3073 | if (!reg->precise) |
3074 | new_marks = true; | |
3075 | reg->precise = true; | |
3076 | } | |
496f3324 | 3077 | if (env->log.level & BPF_LOG_LEVEL2) { |
2e576648 | 3078 | verbose(env, "parent %s regs=%x stack=%llx marks:", |
b5dc0163 AS |
3079 | new_marks ? "didn't have" : "already had", |
3080 | reg_mask, stack_mask); | |
2e576648 | 3081 | print_verifier_state(env, func, true); |
b5dc0163 AS |
3082 | } |
3083 | ||
a3ce685d AS |
3084 | if (!reg_mask && !stack_mask) |
3085 | break; | |
b5dc0163 AS |
3086 | if (!new_marks) |
3087 | break; | |
3088 | ||
3089 | last_idx = st->last_insn_idx; | |
3090 | first_idx = st->first_insn_idx; | |
3091 | } | |
3092 | return 0; | |
3093 | } | |
3094 | ||
eb1f7f71 | 3095 | int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
a3ce685d | 3096 | { |
529409ea | 3097 | return __mark_chain_precision(env, env->cur_state->curframe, regno, -1); |
a3ce685d AS |
3098 | } |
3099 | ||
529409ea | 3100 | static int mark_chain_precision_frame(struct bpf_verifier_env *env, int frame, int regno) |
a3ce685d | 3101 | { |
529409ea | 3102 | return __mark_chain_precision(env, frame, regno, -1); |
a3ce685d AS |
3103 | } |
3104 | ||
529409ea | 3105 | static int mark_chain_precision_stack_frame(struct bpf_verifier_env *env, int frame, int spi) |
a3ce685d | 3106 | { |
529409ea | 3107 | return __mark_chain_precision(env, frame, -1, spi); |
a3ce685d | 3108 | } |
b5dc0163 | 3109 | |
1be7f75d AS |
3110 | static bool is_spillable_regtype(enum bpf_reg_type type) |
3111 | { | |
c25b2ae1 | 3112 | switch (base_type(type)) { |
1be7f75d | 3113 | case PTR_TO_MAP_VALUE: |
1be7f75d AS |
3114 | case PTR_TO_STACK: |
3115 | case PTR_TO_CTX: | |
969bf05e | 3116 | case PTR_TO_PACKET: |
de8f3a83 | 3117 | case PTR_TO_PACKET_META: |
969bf05e | 3118 | case PTR_TO_PACKET_END: |
d58e468b | 3119 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 3120 | case CONST_PTR_TO_MAP: |
c64b7983 | 3121 | case PTR_TO_SOCKET: |
46f8bc92 | 3122 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 3123 | case PTR_TO_TCP_SOCK: |
fada7fdc | 3124 | case PTR_TO_XDP_SOCK: |
65726b5b | 3125 | case PTR_TO_BTF_ID: |
20b2aff4 | 3126 | case PTR_TO_BUF: |
744ea4e3 | 3127 | case PTR_TO_MEM: |
69c087ba YS |
3128 | case PTR_TO_FUNC: |
3129 | case PTR_TO_MAP_KEY: | |
1be7f75d AS |
3130 | return true; |
3131 | default: | |
3132 | return false; | |
3133 | } | |
3134 | } | |
3135 | ||
cc2b14d5 AS |
3136 | /* Does this register contain a constant zero? */ |
3137 | static bool register_is_null(struct bpf_reg_state *reg) | |
3138 | { | |
3139 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
3140 | } | |
3141 | ||
f7cf25b2 AS |
3142 | static bool register_is_const(struct bpf_reg_state *reg) |
3143 | { | |
3144 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
3145 | } | |
3146 | ||
5689d49b YS |
3147 | static bool __is_scalar_unbounded(struct bpf_reg_state *reg) |
3148 | { | |
3149 | return tnum_is_unknown(reg->var_off) && | |
3150 | reg->smin_value == S64_MIN && reg->smax_value == S64_MAX && | |
3151 | reg->umin_value == 0 && reg->umax_value == U64_MAX && | |
3152 | reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX && | |
3153 | reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX; | |
3154 | } | |
3155 | ||
3156 | static bool register_is_bounded(struct bpf_reg_state *reg) | |
3157 | { | |
3158 | return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg); | |
3159 | } | |
3160 | ||
6e7e63cb JH |
3161 | static bool __is_pointer_value(bool allow_ptr_leaks, |
3162 | const struct bpf_reg_state *reg) | |
3163 | { | |
3164 | if (allow_ptr_leaks) | |
3165 | return false; | |
3166 | ||
3167 | return reg->type != SCALAR_VALUE; | |
3168 | } | |
3169 | ||
f7cf25b2 | 3170 | static void save_register_state(struct bpf_func_state *state, |
354e8f19 MKL |
3171 | int spi, struct bpf_reg_state *reg, |
3172 | int size) | |
f7cf25b2 AS |
3173 | { |
3174 | int i; | |
3175 | ||
3176 | state->stack[spi].spilled_ptr = *reg; | |
354e8f19 MKL |
3177 | if (size == BPF_REG_SIZE) |
3178 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
f7cf25b2 | 3179 | |
354e8f19 MKL |
3180 | for (i = BPF_REG_SIZE; i > BPF_REG_SIZE - size; i--) |
3181 | state->stack[spi].slot_type[i - 1] = STACK_SPILL; | |
f7cf25b2 | 3182 | |
354e8f19 MKL |
3183 | /* size < 8 bytes spill */ |
3184 | for (; i; i--) | |
3185 | scrub_spilled_slot(&state->stack[spi].slot_type[i - 1]); | |
f7cf25b2 AS |
3186 | } |
3187 | ||
01f810ac | 3188 | /* check_stack_{read,write}_fixed_off functions track spill/fill of registers, |
17a52670 AS |
3189 | * stack boundary and alignment are checked in check_mem_access() |
3190 | */ | |
01f810ac AM |
3191 | static int check_stack_write_fixed_off(struct bpf_verifier_env *env, |
3192 | /* stack frame we're writing to */ | |
3193 | struct bpf_func_state *state, | |
3194 | int off, int size, int value_regno, | |
3195 | int insn_idx) | |
17a52670 | 3196 | { |
f4d7e40a | 3197 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 3198 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
b5dc0163 | 3199 | u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg; |
f7cf25b2 | 3200 | struct bpf_reg_state *reg = NULL; |
638f5b90 | 3201 | |
c69431aa | 3202 | err = grow_stack_state(state, round_up(slot + 1, BPF_REG_SIZE)); |
638f5b90 AS |
3203 | if (err) |
3204 | return err; | |
9c399760 AS |
3205 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
3206 | * so it's aligned access and [off, off + size) are within stack limits | |
3207 | */ | |
638f5b90 AS |
3208 | if (!env->allow_ptr_leaks && |
3209 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
3210 | size != BPF_REG_SIZE) { | |
3211 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
3212 | return -EACCES; | |
3213 | } | |
17a52670 | 3214 | |
f4d7e40a | 3215 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
3216 | if (value_regno >= 0) |
3217 | reg = &cur->regs[value_regno]; | |
2039f26f DB |
3218 | if (!env->bypass_spec_v4) { |
3219 | bool sanitize = reg && is_spillable_regtype(reg->type); | |
3220 | ||
3221 | for (i = 0; i < size; i++) { | |
3222 | if (state->stack[spi].slot_type[i] == STACK_INVALID) { | |
3223 | sanitize = true; | |
3224 | break; | |
3225 | } | |
3226 | } | |
3227 | ||
3228 | if (sanitize) | |
3229 | env->insn_aux_data[insn_idx].sanitize_stack_spill = true; | |
3230 | } | |
17a52670 | 3231 | |
0f55f9ed | 3232 | mark_stack_slot_scratched(env, spi); |
354e8f19 | 3233 | if (reg && !(off % BPF_REG_SIZE) && register_is_bounded(reg) && |
2c78ee89 | 3234 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
3235 | if (dst_reg != BPF_REG_FP) { |
3236 | /* The backtracking logic can only recognize explicit | |
3237 | * stack slot address like [fp - 8]. Other spill of | |
8fb33b60 | 3238 | * scalar via different register has to be conservative. |
b5dc0163 AS |
3239 | * Backtrack from here and mark all registers as precise |
3240 | * that contributed into 'reg' being a constant. | |
3241 | */ | |
3242 | err = mark_chain_precision(env, value_regno); | |
3243 | if (err) | |
3244 | return err; | |
3245 | } | |
354e8f19 | 3246 | save_register_state(state, spi, reg, size); |
f7cf25b2 | 3247 | } else if (reg && is_spillable_regtype(reg->type)) { |
17a52670 | 3248 | /* register containing pointer is being spilled into stack */ |
9c399760 | 3249 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 3250 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 3251 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
3252 | return -EACCES; |
3253 | } | |
f7cf25b2 | 3254 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
3255 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
3256 | return -EINVAL; | |
3257 | } | |
354e8f19 | 3258 | save_register_state(state, spi, reg, size); |
9c399760 | 3259 | } else { |
cc2b14d5 AS |
3260 | u8 type = STACK_MISC; |
3261 | ||
679c782d EC |
3262 | /* regular write of data into stack destroys any spilled ptr */ |
3263 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d | 3264 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
27113c59 | 3265 | if (is_spilled_reg(&state->stack[spi])) |
0bae2d4d | 3266 | for (i = 0; i < BPF_REG_SIZE; i++) |
354e8f19 | 3267 | scrub_spilled_slot(&state->stack[spi].slot_type[i]); |
9c399760 | 3268 | |
cc2b14d5 AS |
3269 | /* only mark the slot as written if all 8 bytes were written |
3270 | * otherwise read propagation may incorrectly stop too soon | |
3271 | * when stack slots are partially written. | |
3272 | * This heuristic means that read propagation will be | |
3273 | * conservative, since it will add reg_live_read marks | |
3274 | * to stack slots all the way to first state when programs | |
3275 | * writes+reads less than 8 bytes | |
3276 | */ | |
3277 | if (size == BPF_REG_SIZE) | |
3278 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
3279 | ||
3280 | /* when we zero initialize stack slots mark them as such */ | |
b5dc0163 AS |
3281 | if (reg && register_is_null(reg)) { |
3282 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
3283 | err = mark_chain_precision(env, value_regno); | |
3284 | if (err) | |
3285 | return err; | |
cc2b14d5 | 3286 | type = STACK_ZERO; |
b5dc0163 | 3287 | } |
cc2b14d5 | 3288 | |
0bae2d4d | 3289 | /* Mark slots affected by this stack write. */ |
9c399760 | 3290 | for (i = 0; i < size; i++) |
638f5b90 | 3291 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 3292 | type; |
17a52670 AS |
3293 | } |
3294 | return 0; | |
3295 | } | |
3296 | ||
01f810ac AM |
3297 | /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is |
3298 | * known to contain a variable offset. | |
3299 | * This function checks whether the write is permitted and conservatively | |
3300 | * tracks the effects of the write, considering that each stack slot in the | |
3301 | * dynamic range is potentially written to. | |
3302 | * | |
3303 | * 'off' includes 'regno->off'. | |
3304 | * 'value_regno' can be -1, meaning that an unknown value is being written to | |
3305 | * the stack. | |
3306 | * | |
3307 | * Spilled pointers in range are not marked as written because we don't know | |
3308 | * what's going to be actually written. This means that read propagation for | |
3309 | * future reads cannot be terminated by this write. | |
3310 | * | |
3311 | * For privileged programs, uninitialized stack slots are considered | |
3312 | * initialized by this write (even though we don't know exactly what offsets | |
3313 | * are going to be written to). The idea is that we don't want the verifier to | |
3314 | * reject future reads that access slots written to through variable offsets. | |
3315 | */ | |
3316 | static int check_stack_write_var_off(struct bpf_verifier_env *env, | |
3317 | /* func where register points to */ | |
3318 | struct bpf_func_state *state, | |
3319 | int ptr_regno, int off, int size, | |
3320 | int value_regno, int insn_idx) | |
3321 | { | |
3322 | struct bpf_func_state *cur; /* state of the current function */ | |
3323 | int min_off, max_off; | |
3324 | int i, err; | |
3325 | struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL; | |
3326 | bool writing_zero = false; | |
3327 | /* set if the fact that we're writing a zero is used to let any | |
3328 | * stack slots remain STACK_ZERO | |
3329 | */ | |
3330 | bool zero_used = false; | |
3331 | ||
3332 | cur = env->cur_state->frame[env->cur_state->curframe]; | |
3333 | ptr_reg = &cur->regs[ptr_regno]; | |
3334 | min_off = ptr_reg->smin_value + off; | |
3335 | max_off = ptr_reg->smax_value + off + size; | |
3336 | if (value_regno >= 0) | |
3337 | value_reg = &cur->regs[value_regno]; | |
3338 | if (value_reg && register_is_null(value_reg)) | |
3339 | writing_zero = true; | |
3340 | ||
c69431aa | 3341 | err = grow_stack_state(state, round_up(-min_off, BPF_REG_SIZE)); |
01f810ac AM |
3342 | if (err) |
3343 | return err; | |
3344 | ||
3345 | ||
3346 | /* Variable offset writes destroy any spilled pointers in range. */ | |
3347 | for (i = min_off; i < max_off; i++) { | |
3348 | u8 new_type, *stype; | |
3349 | int slot, spi; | |
3350 | ||
3351 | slot = -i - 1; | |
3352 | spi = slot / BPF_REG_SIZE; | |
3353 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
0f55f9ed | 3354 | mark_stack_slot_scratched(env, spi); |
01f810ac | 3355 | |
f5e477a8 KKD |
3356 | if (!env->allow_ptr_leaks && *stype != STACK_MISC && *stype != STACK_ZERO) { |
3357 | /* Reject the write if range we may write to has not | |
3358 | * been initialized beforehand. If we didn't reject | |
3359 | * here, the ptr status would be erased below (even | |
3360 | * though not all slots are actually overwritten), | |
3361 | * possibly opening the door to leaks. | |
3362 | * | |
3363 | * We do however catch STACK_INVALID case below, and | |
3364 | * only allow reading possibly uninitialized memory | |
3365 | * later for CAP_PERFMON, as the write may not happen to | |
3366 | * that slot. | |
01f810ac AM |
3367 | */ |
3368 | verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d", | |
3369 | insn_idx, i); | |
3370 | return -EINVAL; | |
3371 | } | |
3372 | ||
3373 | /* Erase all spilled pointers. */ | |
3374 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
3375 | ||
3376 | /* Update the slot type. */ | |
3377 | new_type = STACK_MISC; | |
3378 | if (writing_zero && *stype == STACK_ZERO) { | |
3379 | new_type = STACK_ZERO; | |
3380 | zero_used = true; | |
3381 | } | |
3382 | /* If the slot is STACK_INVALID, we check whether it's OK to | |
3383 | * pretend that it will be initialized by this write. The slot | |
3384 | * might not actually be written to, and so if we mark it as | |
3385 | * initialized future reads might leak uninitialized memory. | |
3386 | * For privileged programs, we will accept such reads to slots | |
3387 | * that may or may not be written because, if we're reject | |
3388 | * them, the error would be too confusing. | |
3389 | */ | |
3390 | if (*stype == STACK_INVALID && !env->allow_uninit_stack) { | |
3391 | verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d", | |
3392 | insn_idx, i); | |
3393 | return -EINVAL; | |
3394 | } | |
3395 | *stype = new_type; | |
3396 | } | |
3397 | if (zero_used) { | |
3398 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
3399 | err = mark_chain_precision(env, value_regno); | |
3400 | if (err) | |
3401 | return err; | |
3402 | } | |
3403 | return 0; | |
3404 | } | |
3405 | ||
3406 | /* When register 'dst_regno' is assigned some values from stack[min_off, | |
3407 | * max_off), we set the register's type according to the types of the | |
3408 | * respective stack slots. If all the stack values are known to be zeros, then | |
3409 | * so is the destination reg. Otherwise, the register is considered to be | |
3410 | * SCALAR. This function does not deal with register filling; the caller must | |
3411 | * ensure that all spilled registers in the stack range have been marked as | |
3412 | * read. | |
3413 | */ | |
3414 | static void mark_reg_stack_read(struct bpf_verifier_env *env, | |
3415 | /* func where src register points to */ | |
3416 | struct bpf_func_state *ptr_state, | |
3417 | int min_off, int max_off, int dst_regno) | |
3418 | { | |
3419 | struct bpf_verifier_state *vstate = env->cur_state; | |
3420 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3421 | int i, slot, spi; | |
3422 | u8 *stype; | |
3423 | int zeros = 0; | |
3424 | ||
3425 | for (i = min_off; i < max_off; i++) { | |
3426 | slot = -i - 1; | |
3427 | spi = slot / BPF_REG_SIZE; | |
3428 | stype = ptr_state->stack[spi].slot_type; | |
3429 | if (stype[slot % BPF_REG_SIZE] != STACK_ZERO) | |
3430 | break; | |
3431 | zeros++; | |
3432 | } | |
3433 | if (zeros == max_off - min_off) { | |
3434 | /* any access_size read into register is zero extended, | |
3435 | * so the whole register == const_zero | |
3436 | */ | |
3437 | __mark_reg_const_zero(&state->regs[dst_regno]); | |
3438 | /* backtracking doesn't support STACK_ZERO yet, | |
3439 | * so mark it precise here, so that later | |
3440 | * backtracking can stop here. | |
3441 | * Backtracking may not need this if this register | |
3442 | * doesn't participate in pointer adjustment. | |
3443 | * Forward propagation of precise flag is not | |
3444 | * necessary either. This mark is only to stop | |
3445 | * backtracking. Any register that contributed | |
3446 | * to const 0 was marked precise before spill. | |
3447 | */ | |
3448 | state->regs[dst_regno].precise = true; | |
3449 | } else { | |
3450 | /* have read misc data from the stack */ | |
3451 | mark_reg_unknown(env, state->regs, dst_regno); | |
3452 | } | |
3453 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
3454 | } | |
3455 | ||
3456 | /* Read the stack at 'off' and put the results into the register indicated by | |
3457 | * 'dst_regno'. It handles reg filling if the addressed stack slot is a | |
3458 | * spilled reg. | |
3459 | * | |
3460 | * 'dst_regno' can be -1, meaning that the read value is not going to a | |
3461 | * register. | |
3462 | * | |
3463 | * The access is assumed to be within the current stack bounds. | |
3464 | */ | |
3465 | static int check_stack_read_fixed_off(struct bpf_verifier_env *env, | |
3466 | /* func where src register points to */ | |
3467 | struct bpf_func_state *reg_state, | |
3468 | int off, int size, int dst_regno) | |
17a52670 | 3469 | { |
f4d7e40a AS |
3470 | struct bpf_verifier_state *vstate = env->cur_state; |
3471 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 3472 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 3473 | struct bpf_reg_state *reg; |
354e8f19 | 3474 | u8 *stype, type; |
17a52670 | 3475 | |
f4d7e40a | 3476 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 3477 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 3478 | |
27113c59 | 3479 | if (is_spilled_reg(®_state->stack[spi])) { |
f30d4968 MKL |
3480 | u8 spill_size = 1; |
3481 | ||
3482 | for (i = BPF_REG_SIZE - 1; i > 0 && stype[i - 1] == STACK_SPILL; i--) | |
3483 | spill_size++; | |
354e8f19 | 3484 | |
f30d4968 | 3485 | if (size != BPF_REG_SIZE || spill_size != BPF_REG_SIZE) { |
f7cf25b2 AS |
3486 | if (reg->type != SCALAR_VALUE) { |
3487 | verbose_linfo(env, env->insn_idx, "; "); | |
3488 | verbose(env, "invalid size of register fill\n"); | |
3489 | return -EACCES; | |
3490 | } | |
354e8f19 MKL |
3491 | |
3492 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
3493 | if (dst_regno < 0) | |
3494 | return 0; | |
3495 | ||
f30d4968 | 3496 | if (!(off % BPF_REG_SIZE) && size == spill_size) { |
354e8f19 MKL |
3497 | /* The earlier check_reg_arg() has decided the |
3498 | * subreg_def for this insn. Save it first. | |
3499 | */ | |
3500 | s32 subreg_def = state->regs[dst_regno].subreg_def; | |
3501 | ||
3502 | state->regs[dst_regno] = *reg; | |
3503 | state->regs[dst_regno].subreg_def = subreg_def; | |
3504 | } else { | |
3505 | for (i = 0; i < size; i++) { | |
3506 | type = stype[(slot - i) % BPF_REG_SIZE]; | |
3507 | if (type == STACK_SPILL) | |
3508 | continue; | |
3509 | if (type == STACK_MISC) | |
3510 | continue; | |
3511 | verbose(env, "invalid read from stack off %d+%d size %d\n", | |
3512 | off, i, size); | |
3513 | return -EACCES; | |
3514 | } | |
01f810ac | 3515 | mark_reg_unknown(env, state->regs, dst_regno); |
f7cf25b2 | 3516 | } |
354e8f19 | 3517 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
f7cf25b2 | 3518 | return 0; |
17a52670 | 3519 | } |
17a52670 | 3520 | |
01f810ac | 3521 | if (dst_regno >= 0) { |
17a52670 | 3522 | /* restore register state from stack */ |
01f810ac | 3523 | state->regs[dst_regno] = *reg; |
2f18f62e AS |
3524 | /* mark reg as written since spilled pointer state likely |
3525 | * has its liveness marks cleared by is_state_visited() | |
3526 | * which resets stack/reg liveness for state transitions | |
3527 | */ | |
01f810ac | 3528 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
6e7e63cb | 3529 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
01f810ac | 3530 | /* If dst_regno==-1, the caller is asking us whether |
6e7e63cb JH |
3531 | * it is acceptable to use this value as a SCALAR_VALUE |
3532 | * (e.g. for XADD). | |
3533 | * We must not allow unprivileged callers to do that | |
3534 | * with spilled pointers. | |
3535 | */ | |
3536 | verbose(env, "leaking pointer from stack off %d\n", | |
3537 | off); | |
3538 | return -EACCES; | |
dc503a8a | 3539 | } |
f7cf25b2 | 3540 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 AS |
3541 | } else { |
3542 | for (i = 0; i < size; i++) { | |
01f810ac AM |
3543 | type = stype[(slot - i) % BPF_REG_SIZE]; |
3544 | if (type == STACK_MISC) | |
cc2b14d5 | 3545 | continue; |
01f810ac | 3546 | if (type == STACK_ZERO) |
cc2b14d5 | 3547 | continue; |
cc2b14d5 AS |
3548 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
3549 | off, i, size); | |
3550 | return -EACCES; | |
3551 | } | |
f7cf25b2 | 3552 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
01f810ac AM |
3553 | if (dst_regno >= 0) |
3554 | mark_reg_stack_read(env, reg_state, off, off + size, dst_regno); | |
17a52670 | 3555 | } |
f7cf25b2 | 3556 | return 0; |
17a52670 AS |
3557 | } |
3558 | ||
61df10c7 | 3559 | enum bpf_access_src { |
01f810ac AM |
3560 | ACCESS_DIRECT = 1, /* the access is performed by an instruction */ |
3561 | ACCESS_HELPER = 2, /* the access is performed by a helper */ | |
3562 | }; | |
3563 | ||
3564 | static int check_stack_range_initialized(struct bpf_verifier_env *env, | |
3565 | int regno, int off, int access_size, | |
3566 | bool zero_size_allowed, | |
61df10c7 | 3567 | enum bpf_access_src type, |
01f810ac AM |
3568 | struct bpf_call_arg_meta *meta); |
3569 | ||
3570 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) | |
3571 | { | |
3572 | return cur_regs(env) + regno; | |
3573 | } | |
3574 | ||
3575 | /* Read the stack at 'ptr_regno + off' and put the result into the register | |
3576 | * 'dst_regno'. | |
3577 | * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'), | |
3578 | * but not its variable offset. | |
3579 | * 'size' is assumed to be <= reg size and the access is assumed to be aligned. | |
3580 | * | |
3581 | * As opposed to check_stack_read_fixed_off, this function doesn't deal with | |
3582 | * filling registers (i.e. reads of spilled register cannot be detected when | |
3583 | * the offset is not fixed). We conservatively mark 'dst_regno' as containing | |
3584 | * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable | |
3585 | * offset; for a fixed offset check_stack_read_fixed_off should be used | |
3586 | * instead. | |
3587 | */ | |
3588 | static int check_stack_read_var_off(struct bpf_verifier_env *env, | |
3589 | int ptr_regno, int off, int size, int dst_regno) | |
e4298d25 | 3590 | { |
01f810ac AM |
3591 | /* The state of the source register. */ |
3592 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
3593 | struct bpf_func_state *ptr_state = func(env, reg); | |
3594 | int err; | |
3595 | int min_off, max_off; | |
3596 | ||
3597 | /* Note that we pass a NULL meta, so raw access will not be permitted. | |
e4298d25 | 3598 | */ |
01f810ac AM |
3599 | err = check_stack_range_initialized(env, ptr_regno, off, size, |
3600 | false, ACCESS_DIRECT, NULL); | |
3601 | if (err) | |
3602 | return err; | |
3603 | ||
3604 | min_off = reg->smin_value + off; | |
3605 | max_off = reg->smax_value + off; | |
3606 | mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno); | |
3607 | return 0; | |
3608 | } | |
3609 | ||
3610 | /* check_stack_read dispatches to check_stack_read_fixed_off or | |
3611 | * check_stack_read_var_off. | |
3612 | * | |
3613 | * The caller must ensure that the offset falls within the allocated stack | |
3614 | * bounds. | |
3615 | * | |
3616 | * 'dst_regno' is a register which will receive the value from the stack. It | |
3617 | * can be -1, meaning that the read value is not going to a register. | |
3618 | */ | |
3619 | static int check_stack_read(struct bpf_verifier_env *env, | |
3620 | int ptr_regno, int off, int size, | |
3621 | int dst_regno) | |
3622 | { | |
3623 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
3624 | struct bpf_func_state *state = func(env, reg); | |
3625 | int err; | |
3626 | /* Some accesses are only permitted with a static offset. */ | |
3627 | bool var_off = !tnum_is_const(reg->var_off); | |
3628 | ||
3629 | /* The offset is required to be static when reads don't go to a | |
3630 | * register, in order to not leak pointers (see | |
3631 | * check_stack_read_fixed_off). | |
3632 | */ | |
3633 | if (dst_regno < 0 && var_off) { | |
e4298d25 DB |
3634 | char tn_buf[48]; |
3635 | ||
3636 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac | 3637 | verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n", |
e4298d25 DB |
3638 | tn_buf, off, size); |
3639 | return -EACCES; | |
3640 | } | |
01f810ac AM |
3641 | /* Variable offset is prohibited for unprivileged mode for simplicity |
3642 | * since it requires corresponding support in Spectre masking for stack | |
3643 | * ALU. See also retrieve_ptr_limit(). | |
3644 | */ | |
3645 | if (!env->bypass_spec_v1 && var_off) { | |
3646 | char tn_buf[48]; | |
e4298d25 | 3647 | |
01f810ac AM |
3648 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
3649 | verbose(env, "R%d variable offset stack access prohibited for !root, var_off=%s\n", | |
3650 | ptr_regno, tn_buf); | |
e4298d25 DB |
3651 | return -EACCES; |
3652 | } | |
3653 | ||
01f810ac AM |
3654 | if (!var_off) { |
3655 | off += reg->var_off.value; | |
3656 | err = check_stack_read_fixed_off(env, state, off, size, | |
3657 | dst_regno); | |
3658 | } else { | |
3659 | /* Variable offset stack reads need more conservative handling | |
3660 | * than fixed offset ones. Note that dst_regno >= 0 on this | |
3661 | * branch. | |
3662 | */ | |
3663 | err = check_stack_read_var_off(env, ptr_regno, off, size, | |
3664 | dst_regno); | |
3665 | } | |
3666 | return err; | |
3667 | } | |
3668 | ||
3669 | ||
3670 | /* check_stack_write dispatches to check_stack_write_fixed_off or | |
3671 | * check_stack_write_var_off. | |
3672 | * | |
3673 | * 'ptr_regno' is the register used as a pointer into the stack. | |
3674 | * 'off' includes 'ptr_regno->off', but not its variable offset (if any). | |
3675 | * 'value_regno' is the register whose value we're writing to the stack. It can | |
3676 | * be -1, meaning that we're not writing from a register. | |
3677 | * | |
3678 | * The caller must ensure that the offset falls within the maximum stack size. | |
3679 | */ | |
3680 | static int check_stack_write(struct bpf_verifier_env *env, | |
3681 | int ptr_regno, int off, int size, | |
3682 | int value_regno, int insn_idx) | |
3683 | { | |
3684 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
3685 | struct bpf_func_state *state = func(env, reg); | |
3686 | int err; | |
3687 | ||
3688 | if (tnum_is_const(reg->var_off)) { | |
3689 | off += reg->var_off.value; | |
3690 | err = check_stack_write_fixed_off(env, state, off, size, | |
3691 | value_regno, insn_idx); | |
3692 | } else { | |
3693 | /* Variable offset stack reads need more conservative handling | |
3694 | * than fixed offset ones. | |
3695 | */ | |
3696 | err = check_stack_write_var_off(env, state, | |
3697 | ptr_regno, off, size, | |
3698 | value_regno, insn_idx); | |
3699 | } | |
3700 | return err; | |
e4298d25 DB |
3701 | } |
3702 | ||
591fe988 DB |
3703 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
3704 | int off, int size, enum bpf_access_type type) | |
3705 | { | |
3706 | struct bpf_reg_state *regs = cur_regs(env); | |
3707 | struct bpf_map *map = regs[regno].map_ptr; | |
3708 | u32 cap = bpf_map_flags_to_cap(map); | |
3709 | ||
3710 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
3711 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
3712 | map->value_size, off, size); | |
3713 | return -EACCES; | |
3714 | } | |
3715 | ||
3716 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
3717 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
3718 | map->value_size, off, size); | |
3719 | return -EACCES; | |
3720 | } | |
3721 | ||
3722 | return 0; | |
3723 | } | |
3724 | ||
457f4436 AN |
3725 | /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */ |
3726 | static int __check_mem_access(struct bpf_verifier_env *env, int regno, | |
3727 | int off, int size, u32 mem_size, | |
3728 | bool zero_size_allowed) | |
17a52670 | 3729 | { |
457f4436 AN |
3730 | bool size_ok = size > 0 || (size == 0 && zero_size_allowed); |
3731 | struct bpf_reg_state *reg; | |
3732 | ||
3733 | if (off >= 0 && size_ok && (u64)off + size <= mem_size) | |
3734 | return 0; | |
17a52670 | 3735 | |
457f4436 AN |
3736 | reg = &cur_regs(env)[regno]; |
3737 | switch (reg->type) { | |
69c087ba YS |
3738 | case PTR_TO_MAP_KEY: |
3739 | verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n", | |
3740 | mem_size, off, size); | |
3741 | break; | |
457f4436 | 3742 | case PTR_TO_MAP_VALUE: |
61bd5218 | 3743 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
457f4436 AN |
3744 | mem_size, off, size); |
3745 | break; | |
3746 | case PTR_TO_PACKET: | |
3747 | case PTR_TO_PACKET_META: | |
3748 | case PTR_TO_PACKET_END: | |
3749 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", | |
3750 | off, size, regno, reg->id, off, mem_size); | |
3751 | break; | |
3752 | case PTR_TO_MEM: | |
3753 | default: | |
3754 | verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n", | |
3755 | mem_size, off, size); | |
17a52670 | 3756 | } |
457f4436 AN |
3757 | |
3758 | return -EACCES; | |
17a52670 AS |
3759 | } |
3760 | ||
457f4436 AN |
3761 | /* check read/write into a memory region with possible variable offset */ |
3762 | static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno, | |
3763 | int off, int size, u32 mem_size, | |
3764 | bool zero_size_allowed) | |
dbcfe5f7 | 3765 | { |
f4d7e40a AS |
3766 | struct bpf_verifier_state *vstate = env->cur_state; |
3767 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
3768 | struct bpf_reg_state *reg = &state->regs[regno]; |
3769 | int err; | |
3770 | ||
457f4436 | 3771 | /* We may have adjusted the register pointing to memory region, so we |
f1174f77 EC |
3772 | * need to try adding each of min_value and max_value to off |
3773 | * to make sure our theoretical access will be safe. | |
2e576648 CL |
3774 | * |
3775 | * The minimum value is only important with signed | |
dbcfe5f7 GB |
3776 | * comparisons where we can't assume the floor of a |
3777 | * value is 0. If we are using signed variables for our | |
3778 | * index'es we need to make sure that whatever we use | |
3779 | * will have a set floor within our range. | |
3780 | */ | |
b7137c4e DB |
3781 | if (reg->smin_value < 0 && |
3782 | (reg->smin_value == S64_MIN || | |
3783 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
3784 | reg->smin_value + off < 0)) { | |
61bd5218 | 3785 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
3786 | regno); |
3787 | return -EACCES; | |
3788 | } | |
457f4436 AN |
3789 | err = __check_mem_access(env, regno, reg->smin_value + off, size, |
3790 | mem_size, zero_size_allowed); | |
dbcfe5f7 | 3791 | if (err) { |
457f4436 | 3792 | verbose(env, "R%d min value is outside of the allowed memory range\n", |
61bd5218 | 3793 | regno); |
dbcfe5f7 GB |
3794 | return err; |
3795 | } | |
3796 | ||
b03c9f9f EC |
3797 | /* If we haven't set a max value then we need to bail since we can't be |
3798 | * sure we won't do bad things. | |
3799 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 3800 | */ |
b03c9f9f | 3801 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
457f4436 | 3802 | verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n", |
dbcfe5f7 GB |
3803 | regno); |
3804 | return -EACCES; | |
3805 | } | |
457f4436 AN |
3806 | err = __check_mem_access(env, regno, reg->umax_value + off, size, |
3807 | mem_size, zero_size_allowed); | |
3808 | if (err) { | |
3809 | verbose(env, "R%d max value is outside of the allowed memory range\n", | |
61bd5218 | 3810 | regno); |
457f4436 AN |
3811 | return err; |
3812 | } | |
3813 | ||
3814 | return 0; | |
3815 | } | |
d83525ca | 3816 | |
e9147b44 KKD |
3817 | static int __check_ptr_off_reg(struct bpf_verifier_env *env, |
3818 | const struct bpf_reg_state *reg, int regno, | |
3819 | bool fixed_off_ok) | |
3820 | { | |
3821 | /* Access to this pointer-typed register or passing it to a helper | |
3822 | * is only allowed in its original, unmodified form. | |
3823 | */ | |
3824 | ||
3825 | if (reg->off < 0) { | |
3826 | verbose(env, "negative offset %s ptr R%d off=%d disallowed\n", | |
3827 | reg_type_str(env, reg->type), regno, reg->off); | |
3828 | return -EACCES; | |
3829 | } | |
3830 | ||
3831 | if (!fixed_off_ok && reg->off) { | |
3832 | verbose(env, "dereference of modified %s ptr R%d off=%d disallowed\n", | |
3833 | reg_type_str(env, reg->type), regno, reg->off); | |
3834 | return -EACCES; | |
3835 | } | |
3836 | ||
3837 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3838 | char tn_buf[48]; | |
3839 | ||
3840 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3841 | verbose(env, "variable %s access var_off=%s disallowed\n", | |
3842 | reg_type_str(env, reg->type), tn_buf); | |
3843 | return -EACCES; | |
3844 | } | |
3845 | ||
3846 | return 0; | |
3847 | } | |
3848 | ||
3849 | int check_ptr_off_reg(struct bpf_verifier_env *env, | |
3850 | const struct bpf_reg_state *reg, int regno) | |
3851 | { | |
3852 | return __check_ptr_off_reg(env, reg, regno, false); | |
3853 | } | |
3854 | ||
61df10c7 | 3855 | static int map_kptr_match_type(struct bpf_verifier_env *env, |
aa3496ac | 3856 | struct btf_field *kptr_field, |
61df10c7 KKD |
3857 | struct bpf_reg_state *reg, u32 regno) |
3858 | { | |
aa3496ac | 3859 | const char *targ_name = kernel_type_name(kptr_field->kptr.btf, kptr_field->kptr.btf_id); |
3f00c523 | 3860 | int perm_flags = PTR_MAYBE_NULL | PTR_TRUSTED; |
61df10c7 KKD |
3861 | const char *reg_name = ""; |
3862 | ||
6efe152d | 3863 | /* Only unreferenced case accepts untrusted pointers */ |
aa3496ac | 3864 | if (kptr_field->type == BPF_KPTR_UNREF) |
6efe152d KKD |
3865 | perm_flags |= PTR_UNTRUSTED; |
3866 | ||
3867 | if (base_type(reg->type) != PTR_TO_BTF_ID || (type_flag(reg->type) & ~perm_flags)) | |
61df10c7 KKD |
3868 | goto bad_type; |
3869 | ||
3870 | if (!btf_is_kernel(reg->btf)) { | |
3871 | verbose(env, "R%d must point to kernel BTF\n", regno); | |
3872 | return -EINVAL; | |
3873 | } | |
3874 | /* We need to verify reg->type and reg->btf, before accessing reg->btf */ | |
3875 | reg_name = kernel_type_name(reg->btf, reg->btf_id); | |
3876 | ||
c0a5a21c KKD |
3877 | /* For ref_ptr case, release function check should ensure we get one |
3878 | * referenced PTR_TO_BTF_ID, and that its fixed offset is 0. For the | |
3879 | * normal store of unreferenced kptr, we must ensure var_off is zero. | |
3880 | * Since ref_ptr cannot be accessed directly by BPF insns, checks for | |
3881 | * reg->off and reg->ref_obj_id are not needed here. | |
3882 | */ | |
61df10c7 KKD |
3883 | if (__check_ptr_off_reg(env, reg, regno, true)) |
3884 | return -EACCES; | |
3885 | ||
3886 | /* A full type match is needed, as BTF can be vmlinux or module BTF, and | |
3887 | * we also need to take into account the reg->off. | |
3888 | * | |
3889 | * We want to support cases like: | |
3890 | * | |
3891 | * struct foo { | |
3892 | * struct bar br; | |
3893 | * struct baz bz; | |
3894 | * }; | |
3895 | * | |
3896 | * struct foo *v; | |
3897 | * v = func(); // PTR_TO_BTF_ID | |
3898 | * val->foo = v; // reg->off is zero, btf and btf_id match type | |
3899 | * val->bar = &v->br; // reg->off is still zero, but we need to retry with | |
3900 | * // first member type of struct after comparison fails | |
3901 | * val->baz = &v->bz; // reg->off is non-zero, so struct needs to be walked | |
3902 | * // to match type | |
3903 | * | |
3904 | * In the kptr_ref case, check_func_arg_reg_off already ensures reg->off | |
2ab3b380 KKD |
3905 | * is zero. We must also ensure that btf_struct_ids_match does not walk |
3906 | * the struct to match type against first member of struct, i.e. reject | |
3907 | * second case from above. Hence, when type is BPF_KPTR_REF, we set | |
3908 | * strict mode to true for type match. | |
61df10c7 KKD |
3909 | */ |
3910 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
aa3496ac KKD |
3911 | kptr_field->kptr.btf, kptr_field->kptr.btf_id, |
3912 | kptr_field->type == BPF_KPTR_REF)) | |
61df10c7 KKD |
3913 | goto bad_type; |
3914 | return 0; | |
3915 | bad_type: | |
3916 | verbose(env, "invalid kptr access, R%d type=%s%s ", regno, | |
3917 | reg_type_str(env, reg->type), reg_name); | |
6efe152d | 3918 | verbose(env, "expected=%s%s", reg_type_str(env, PTR_TO_BTF_ID), targ_name); |
aa3496ac | 3919 | if (kptr_field->type == BPF_KPTR_UNREF) |
6efe152d KKD |
3920 | verbose(env, " or %s%s\n", reg_type_str(env, PTR_TO_BTF_ID | PTR_UNTRUSTED), |
3921 | targ_name); | |
3922 | else | |
3923 | verbose(env, "\n"); | |
61df10c7 KKD |
3924 | return -EINVAL; |
3925 | } | |
3926 | ||
3927 | static int check_map_kptr_access(struct bpf_verifier_env *env, u32 regno, | |
3928 | int value_regno, int insn_idx, | |
aa3496ac | 3929 | struct btf_field *kptr_field) |
61df10c7 KKD |
3930 | { |
3931 | struct bpf_insn *insn = &env->prog->insnsi[insn_idx]; | |
3932 | int class = BPF_CLASS(insn->code); | |
3933 | struct bpf_reg_state *val_reg; | |
3934 | ||
3935 | /* Things we already checked for in check_map_access and caller: | |
3936 | * - Reject cases where variable offset may touch kptr | |
3937 | * - size of access (must be BPF_DW) | |
3938 | * - tnum_is_const(reg->var_off) | |
aa3496ac | 3939 | * - kptr_field->offset == off + reg->var_off.value |
61df10c7 KKD |
3940 | */ |
3941 | /* Only BPF_[LDX,STX,ST] | BPF_MEM | BPF_DW is supported */ | |
3942 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
3943 | verbose(env, "kptr in map can only be accessed using BPF_MEM instruction mode\n"); | |
3944 | return -EACCES; | |
3945 | } | |
3946 | ||
6efe152d KKD |
3947 | /* We only allow loading referenced kptr, since it will be marked as |
3948 | * untrusted, similar to unreferenced kptr. | |
3949 | */ | |
aa3496ac | 3950 | if (class != BPF_LDX && kptr_field->type == BPF_KPTR_REF) { |
6efe152d | 3951 | verbose(env, "store to referenced kptr disallowed\n"); |
c0a5a21c KKD |
3952 | return -EACCES; |
3953 | } | |
3954 | ||
61df10c7 KKD |
3955 | if (class == BPF_LDX) { |
3956 | val_reg = reg_state(env, value_regno); | |
3957 | /* We can simply mark the value_regno receiving the pointer | |
3958 | * value from map as PTR_TO_BTF_ID, with the correct type. | |
3959 | */ | |
aa3496ac KKD |
3960 | mark_btf_ld_reg(env, cur_regs(env), value_regno, PTR_TO_BTF_ID, kptr_field->kptr.btf, |
3961 | kptr_field->kptr.btf_id, PTR_MAYBE_NULL | PTR_UNTRUSTED); | |
61df10c7 KKD |
3962 | /* For mark_ptr_or_null_reg */ |
3963 | val_reg->id = ++env->id_gen; | |
3964 | } else if (class == BPF_STX) { | |
3965 | val_reg = reg_state(env, value_regno); | |
3966 | if (!register_is_null(val_reg) && | |
aa3496ac | 3967 | map_kptr_match_type(env, kptr_field, val_reg, value_regno)) |
61df10c7 KKD |
3968 | return -EACCES; |
3969 | } else if (class == BPF_ST) { | |
3970 | if (insn->imm) { | |
3971 | verbose(env, "BPF_ST imm must be 0 when storing to kptr at off=%u\n", | |
aa3496ac | 3972 | kptr_field->offset); |
61df10c7 KKD |
3973 | return -EACCES; |
3974 | } | |
3975 | } else { | |
3976 | verbose(env, "kptr in map can only be accessed using BPF_LDX/BPF_STX/BPF_ST\n"); | |
3977 | return -EACCES; | |
3978 | } | |
3979 | return 0; | |
3980 | } | |
3981 | ||
457f4436 AN |
3982 | /* check read/write into a map element with possible variable offset */ |
3983 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
61df10c7 KKD |
3984 | int off, int size, bool zero_size_allowed, |
3985 | enum bpf_access_src src) | |
457f4436 AN |
3986 | { |
3987 | struct bpf_verifier_state *vstate = env->cur_state; | |
3988 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3989 | struct bpf_reg_state *reg = &state->regs[regno]; | |
3990 | struct bpf_map *map = reg->map_ptr; | |
aa3496ac KKD |
3991 | struct btf_record *rec; |
3992 | int err, i; | |
457f4436 AN |
3993 | |
3994 | err = check_mem_region_access(env, regno, off, size, map->value_size, | |
3995 | zero_size_allowed); | |
3996 | if (err) | |
3997 | return err; | |
3998 | ||
aa3496ac KKD |
3999 | if (IS_ERR_OR_NULL(map->record)) |
4000 | return 0; | |
4001 | rec = map->record; | |
4002 | for (i = 0; i < rec->cnt; i++) { | |
4003 | struct btf_field *field = &rec->fields[i]; | |
4004 | u32 p = field->offset; | |
d83525ca | 4005 | |
db559117 KKD |
4006 | /* If any part of a field can be touched by load/store, reject |
4007 | * this program. To check that [x1, x2) overlaps with [y1, y2), | |
d83525ca AS |
4008 | * it is sufficient to check x1 < y2 && y1 < x2. |
4009 | */ | |
aa3496ac KKD |
4010 | if (reg->smin_value + off < p + btf_field_type_size(field->type) && |
4011 | p < reg->umax_value + off + size) { | |
4012 | switch (field->type) { | |
4013 | case BPF_KPTR_UNREF: | |
4014 | case BPF_KPTR_REF: | |
61df10c7 KKD |
4015 | if (src != ACCESS_DIRECT) { |
4016 | verbose(env, "kptr cannot be accessed indirectly by helper\n"); | |
4017 | return -EACCES; | |
4018 | } | |
4019 | if (!tnum_is_const(reg->var_off)) { | |
4020 | verbose(env, "kptr access cannot have variable offset\n"); | |
4021 | return -EACCES; | |
4022 | } | |
4023 | if (p != off + reg->var_off.value) { | |
4024 | verbose(env, "kptr access misaligned expected=%u off=%llu\n", | |
4025 | p, off + reg->var_off.value); | |
4026 | return -EACCES; | |
4027 | } | |
4028 | if (size != bpf_size_to_bytes(BPF_DW)) { | |
4029 | verbose(env, "kptr access size must be BPF_DW\n"); | |
4030 | return -EACCES; | |
4031 | } | |
4032 | break; | |
aa3496ac | 4033 | default: |
db559117 KKD |
4034 | verbose(env, "%s cannot be accessed directly by load/store\n", |
4035 | btf_field_type_name(field->type)); | |
aa3496ac | 4036 | return -EACCES; |
61df10c7 KKD |
4037 | } |
4038 | } | |
4039 | } | |
aa3496ac | 4040 | return 0; |
dbcfe5f7 GB |
4041 | } |
4042 | ||
969bf05e AS |
4043 | #define MAX_PACKET_OFF 0xffff |
4044 | ||
58e2af8b | 4045 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
4046 | const struct bpf_call_arg_meta *meta, |
4047 | enum bpf_access_type t) | |
4acf6c0b | 4048 | { |
7e40781c UP |
4049 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
4050 | ||
4051 | switch (prog_type) { | |
5d66fa7d | 4052 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
4053 | case BPF_PROG_TYPE_LWT_IN: |
4054 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 4055 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 4056 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 4057 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 4058 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
4059 | if (t == BPF_WRITE) |
4060 | return false; | |
8731745e | 4061 | fallthrough; |
5d66fa7d DB |
4062 | |
4063 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
4064 | case BPF_PROG_TYPE_SCHED_CLS: |
4065 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 4066 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 4067 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 4068 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 4069 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
4070 | if (meta) |
4071 | return meta->pkt_access; | |
4072 | ||
4073 | env->seen_direct_write = true; | |
4acf6c0b | 4074 | return true; |
0d01da6a SF |
4075 | |
4076 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
4077 | if (t == BPF_WRITE) | |
4078 | env->seen_direct_write = true; | |
4079 | ||
4080 | return true; | |
4081 | ||
4acf6c0b BB |
4082 | default: |
4083 | return false; | |
4084 | } | |
4085 | } | |
4086 | ||
f1174f77 | 4087 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 4088 | int size, bool zero_size_allowed) |
f1174f77 | 4089 | { |
638f5b90 | 4090 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
4091 | struct bpf_reg_state *reg = ®s[regno]; |
4092 | int err; | |
4093 | ||
4094 | /* We may have added a variable offset to the packet pointer; but any | |
4095 | * reg->range we have comes after that. We are only checking the fixed | |
4096 | * offset. | |
4097 | */ | |
4098 | ||
4099 | /* We don't allow negative numbers, because we aren't tracking enough | |
4100 | * detail to prove they're safe. | |
4101 | */ | |
b03c9f9f | 4102 | if (reg->smin_value < 0) { |
61bd5218 | 4103 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
4104 | regno); |
4105 | return -EACCES; | |
4106 | } | |
6d94e741 AS |
4107 | |
4108 | err = reg->range < 0 ? -EINVAL : | |
4109 | __check_mem_access(env, regno, off, size, reg->range, | |
457f4436 | 4110 | zero_size_allowed); |
f1174f77 | 4111 | if (err) { |
61bd5218 | 4112 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
4113 | return err; |
4114 | } | |
e647815a | 4115 | |
457f4436 | 4116 | /* __check_mem_access has made sure "off + size - 1" is within u16. |
e647815a JW |
4117 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, |
4118 | * otherwise find_good_pkt_pointers would have refused to set range info | |
457f4436 | 4119 | * that __check_mem_access would have rejected this pkt access. |
e647815a JW |
4120 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. |
4121 | */ | |
4122 | env->prog->aux->max_pkt_offset = | |
4123 | max_t(u32, env->prog->aux->max_pkt_offset, | |
4124 | off + reg->umax_value + size - 1); | |
4125 | ||
f1174f77 EC |
4126 | return err; |
4127 | } | |
4128 | ||
4129 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 4130 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 | 4131 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
22dc4a0f | 4132 | struct btf **btf, u32 *btf_id) |
17a52670 | 4133 | { |
f96da094 DB |
4134 | struct bpf_insn_access_aux info = { |
4135 | .reg_type = *reg_type, | |
9e15db66 | 4136 | .log = &env->log, |
f96da094 | 4137 | }; |
31fd8581 | 4138 | |
4f9218aa | 4139 | if (env->ops->is_valid_access && |
5e43f899 | 4140 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
4141 | /* A non zero info.ctx_field_size indicates that this field is a |
4142 | * candidate for later verifier transformation to load the whole | |
4143 | * field and then apply a mask when accessed with a narrower | |
4144 | * access than actual ctx access size. A zero info.ctx_field_size | |
4145 | * will only allow for whole field access and rejects any other | |
4146 | * type of narrower access. | |
31fd8581 | 4147 | */ |
23994631 | 4148 | *reg_type = info.reg_type; |
31fd8581 | 4149 | |
c25b2ae1 | 4150 | if (base_type(*reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 4151 | *btf = info.btf; |
9e15db66 | 4152 | *btf_id = info.btf_id; |
22dc4a0f | 4153 | } else { |
9e15db66 | 4154 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
22dc4a0f | 4155 | } |
32bbe007 AS |
4156 | /* remember the offset of last byte accessed in ctx */ |
4157 | if (env->prog->aux->max_ctx_offset < off + size) | |
4158 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 4159 | return 0; |
32bbe007 | 4160 | } |
17a52670 | 4161 | |
61bd5218 | 4162 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
4163 | return -EACCES; |
4164 | } | |
4165 | ||
d58e468b PP |
4166 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
4167 | int size) | |
4168 | { | |
4169 | if (size < 0 || off < 0 || | |
4170 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
4171 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
4172 | off, size); | |
4173 | return -EACCES; | |
4174 | } | |
4175 | return 0; | |
4176 | } | |
4177 | ||
5f456649 MKL |
4178 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
4179 | u32 regno, int off, int size, | |
4180 | enum bpf_access_type t) | |
c64b7983 JS |
4181 | { |
4182 | struct bpf_reg_state *regs = cur_regs(env); | |
4183 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 4184 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 4185 | bool valid; |
c64b7983 JS |
4186 | |
4187 | if (reg->smin_value < 0) { | |
4188 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
4189 | regno); | |
4190 | return -EACCES; | |
4191 | } | |
4192 | ||
46f8bc92 MKL |
4193 | switch (reg->type) { |
4194 | case PTR_TO_SOCK_COMMON: | |
4195 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
4196 | break; | |
4197 | case PTR_TO_SOCKET: | |
4198 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
4199 | break; | |
655a51e5 MKL |
4200 | case PTR_TO_TCP_SOCK: |
4201 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
4202 | break; | |
fada7fdc JL |
4203 | case PTR_TO_XDP_SOCK: |
4204 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
4205 | break; | |
46f8bc92 MKL |
4206 | default: |
4207 | valid = false; | |
c64b7983 JS |
4208 | } |
4209 | ||
5f456649 | 4210 | |
46f8bc92 MKL |
4211 | if (valid) { |
4212 | env->insn_aux_data[insn_idx].ctx_field_size = | |
4213 | info.ctx_field_size; | |
4214 | return 0; | |
4215 | } | |
4216 | ||
4217 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
c25b2ae1 | 4218 | regno, reg_type_str(env, reg->type), off, size); |
46f8bc92 MKL |
4219 | |
4220 | return -EACCES; | |
c64b7983 JS |
4221 | } |
4222 | ||
4cabc5b1 DB |
4223 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
4224 | { | |
2a159c6f | 4225 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
4226 | } |
4227 | ||
f37a8cb8 DB |
4228 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
4229 | { | |
2a159c6f | 4230 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 4231 | |
46f8bc92 MKL |
4232 | return reg->type == PTR_TO_CTX; |
4233 | } | |
4234 | ||
4235 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
4236 | { | |
4237 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
4238 | ||
4239 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
4240 | } |
4241 | ||
ca369602 DB |
4242 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
4243 | { | |
2a159c6f | 4244 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
4245 | |
4246 | return type_is_pkt_pointer(reg->type); | |
4247 | } | |
4248 | ||
4b5defde DB |
4249 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
4250 | { | |
4251 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
4252 | ||
4253 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
4254 | return reg->type == PTR_TO_FLOW_KEYS; | |
4255 | } | |
4256 | ||
61bd5218 JK |
4257 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
4258 | const struct bpf_reg_state *reg, | |
d1174416 | 4259 | int off, int size, bool strict) |
969bf05e | 4260 | { |
f1174f77 | 4261 | struct tnum reg_off; |
e07b98d9 | 4262 | int ip_align; |
d1174416 DM |
4263 | |
4264 | /* Byte size accesses are always allowed. */ | |
4265 | if (!strict || size == 1) | |
4266 | return 0; | |
4267 | ||
e4eda884 DM |
4268 | /* For platforms that do not have a Kconfig enabling |
4269 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
4270 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
4271 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
4272 | * to this code only in strict mode where we want to emulate | |
4273 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
4274 | * unconditional IP align value of '2'. | |
e07b98d9 | 4275 | */ |
e4eda884 | 4276 | ip_align = 2; |
f1174f77 EC |
4277 | |
4278 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
4279 | if (!tnum_is_aligned(reg_off, size)) { | |
4280 | char tn_buf[48]; | |
4281 | ||
4282 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
4283 | verbose(env, |
4284 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 4285 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
4286 | return -EACCES; |
4287 | } | |
79adffcd | 4288 | |
969bf05e AS |
4289 | return 0; |
4290 | } | |
4291 | ||
61bd5218 JK |
4292 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
4293 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
4294 | const char *pointer_desc, |
4295 | int off, int size, bool strict) | |
79adffcd | 4296 | { |
f1174f77 EC |
4297 | struct tnum reg_off; |
4298 | ||
4299 | /* Byte size accesses are always allowed. */ | |
4300 | if (!strict || size == 1) | |
4301 | return 0; | |
4302 | ||
4303 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
4304 | if (!tnum_is_aligned(reg_off, size)) { | |
4305 | char tn_buf[48]; | |
4306 | ||
4307 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 4308 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 4309 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
4310 | return -EACCES; |
4311 | } | |
4312 | ||
969bf05e AS |
4313 | return 0; |
4314 | } | |
4315 | ||
e07b98d9 | 4316 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
4317 | const struct bpf_reg_state *reg, int off, |
4318 | int size, bool strict_alignment_once) | |
79adffcd | 4319 | { |
ca369602 | 4320 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 4321 | const char *pointer_desc = ""; |
d1174416 | 4322 | |
79adffcd DB |
4323 | switch (reg->type) { |
4324 | case PTR_TO_PACKET: | |
de8f3a83 DB |
4325 | case PTR_TO_PACKET_META: |
4326 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
4327 | * right in front, treat it the very same way. | |
4328 | */ | |
61bd5218 | 4329 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
4330 | case PTR_TO_FLOW_KEYS: |
4331 | pointer_desc = "flow keys "; | |
4332 | break; | |
69c087ba YS |
4333 | case PTR_TO_MAP_KEY: |
4334 | pointer_desc = "key "; | |
4335 | break; | |
f1174f77 EC |
4336 | case PTR_TO_MAP_VALUE: |
4337 | pointer_desc = "value "; | |
4338 | break; | |
4339 | case PTR_TO_CTX: | |
4340 | pointer_desc = "context "; | |
4341 | break; | |
4342 | case PTR_TO_STACK: | |
4343 | pointer_desc = "stack "; | |
01f810ac AM |
4344 | /* The stack spill tracking logic in check_stack_write_fixed_off() |
4345 | * and check_stack_read_fixed_off() relies on stack accesses being | |
a5ec6ae1 JH |
4346 | * aligned. |
4347 | */ | |
4348 | strict = true; | |
f1174f77 | 4349 | break; |
c64b7983 JS |
4350 | case PTR_TO_SOCKET: |
4351 | pointer_desc = "sock "; | |
4352 | break; | |
46f8bc92 MKL |
4353 | case PTR_TO_SOCK_COMMON: |
4354 | pointer_desc = "sock_common "; | |
4355 | break; | |
655a51e5 MKL |
4356 | case PTR_TO_TCP_SOCK: |
4357 | pointer_desc = "tcp_sock "; | |
4358 | break; | |
fada7fdc JL |
4359 | case PTR_TO_XDP_SOCK: |
4360 | pointer_desc = "xdp_sock "; | |
4361 | break; | |
79adffcd | 4362 | default: |
f1174f77 | 4363 | break; |
79adffcd | 4364 | } |
61bd5218 JK |
4365 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
4366 | strict); | |
79adffcd DB |
4367 | } |
4368 | ||
f4d7e40a AS |
4369 | static int update_stack_depth(struct bpf_verifier_env *env, |
4370 | const struct bpf_func_state *func, | |
4371 | int off) | |
4372 | { | |
9c8105bd | 4373 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
4374 | |
4375 | if (stack >= -off) | |
4376 | return 0; | |
4377 | ||
4378 | /* update known max for given subprogram */ | |
9c8105bd | 4379 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
4380 | return 0; |
4381 | } | |
f4d7e40a | 4382 | |
70a87ffe AS |
4383 | /* starting from main bpf function walk all instructions of the function |
4384 | * and recursively walk all callees that given function can call. | |
4385 | * Ignore jump and exit insns. | |
4386 | * Since recursion is prevented by check_cfg() this algorithm | |
4387 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
4388 | */ | |
4389 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
4390 | { | |
9c8105bd JW |
4391 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
4392 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 4393 | struct bpf_insn *insn = env->prog->insnsi; |
ebf7d1f5 | 4394 | bool tail_call_reachable = false; |
70a87ffe AS |
4395 | int ret_insn[MAX_CALL_FRAMES]; |
4396 | int ret_prog[MAX_CALL_FRAMES]; | |
ebf7d1f5 | 4397 | int j; |
f4d7e40a | 4398 | |
70a87ffe | 4399 | process_func: |
7f6e4312 MF |
4400 | /* protect against potential stack overflow that might happen when |
4401 | * bpf2bpf calls get combined with tailcalls. Limit the caller's stack | |
4402 | * depth for such case down to 256 so that the worst case scenario | |
4403 | * would result in 8k stack size (32 which is tailcall limit * 256 = | |
4404 | * 8k). | |
4405 | * | |
4406 | * To get the idea what might happen, see an example: | |
4407 | * func1 -> sub rsp, 128 | |
4408 | * subfunc1 -> sub rsp, 256 | |
4409 | * tailcall1 -> add rsp, 256 | |
4410 | * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320) | |
4411 | * subfunc2 -> sub rsp, 64 | |
4412 | * subfunc22 -> sub rsp, 128 | |
4413 | * tailcall2 -> add rsp, 128 | |
4414 | * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416) | |
4415 | * | |
4416 | * tailcall will unwind the current stack frame but it will not get rid | |
4417 | * of caller's stack as shown on the example above. | |
4418 | */ | |
4419 | if (idx && subprog[idx].has_tail_call && depth >= 256) { | |
4420 | verbose(env, | |
4421 | "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n", | |
4422 | depth); | |
4423 | return -EACCES; | |
4424 | } | |
70a87ffe AS |
4425 | /* round up to 32-bytes, since this is granularity |
4426 | * of interpreter stack size | |
4427 | */ | |
9c8105bd | 4428 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 4429 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 4430 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 4431 | frame + 1, depth); |
f4d7e40a AS |
4432 | return -EACCES; |
4433 | } | |
70a87ffe | 4434 | continue_func: |
4cb3d99c | 4435 | subprog_end = subprog[idx + 1].start; |
70a87ffe | 4436 | for (; i < subprog_end; i++) { |
7ddc80a4 AS |
4437 | int next_insn; |
4438 | ||
69c087ba | 4439 | if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i)) |
70a87ffe AS |
4440 | continue; |
4441 | /* remember insn and function to return to */ | |
4442 | ret_insn[frame] = i + 1; | |
9c8105bd | 4443 | ret_prog[frame] = idx; |
70a87ffe AS |
4444 | |
4445 | /* find the callee */ | |
7ddc80a4 AS |
4446 | next_insn = i + insn[i].imm + 1; |
4447 | idx = find_subprog(env, next_insn); | |
9c8105bd | 4448 | if (idx < 0) { |
70a87ffe | 4449 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
7ddc80a4 | 4450 | next_insn); |
70a87ffe AS |
4451 | return -EFAULT; |
4452 | } | |
7ddc80a4 AS |
4453 | if (subprog[idx].is_async_cb) { |
4454 | if (subprog[idx].has_tail_call) { | |
4455 | verbose(env, "verifier bug. subprog has tail_call and async cb\n"); | |
4456 | return -EFAULT; | |
4457 | } | |
4458 | /* async callbacks don't increase bpf prog stack size */ | |
4459 | continue; | |
4460 | } | |
4461 | i = next_insn; | |
ebf7d1f5 MF |
4462 | |
4463 | if (subprog[idx].has_tail_call) | |
4464 | tail_call_reachable = true; | |
4465 | ||
70a87ffe AS |
4466 | frame++; |
4467 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
4468 | verbose(env, "the call stack of %d frames is too deep !\n", |
4469 | frame); | |
4470 | return -E2BIG; | |
70a87ffe AS |
4471 | } |
4472 | goto process_func; | |
4473 | } | |
ebf7d1f5 MF |
4474 | /* if tail call got detected across bpf2bpf calls then mark each of the |
4475 | * currently present subprog frames as tail call reachable subprogs; | |
4476 | * this info will be utilized by JIT so that we will be preserving the | |
4477 | * tail call counter throughout bpf2bpf calls combined with tailcalls | |
4478 | */ | |
4479 | if (tail_call_reachable) | |
4480 | for (j = 0; j < frame; j++) | |
4481 | subprog[ret_prog[j]].tail_call_reachable = true; | |
5dd0a6b8 DB |
4482 | if (subprog[0].tail_call_reachable) |
4483 | env->prog->aux->tail_call_reachable = true; | |
ebf7d1f5 | 4484 | |
70a87ffe AS |
4485 | /* end of for() loop means the last insn of the 'subprog' |
4486 | * was reached. Doesn't matter whether it was JA or EXIT | |
4487 | */ | |
4488 | if (frame == 0) | |
4489 | return 0; | |
9c8105bd | 4490 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
4491 | frame--; |
4492 | i = ret_insn[frame]; | |
9c8105bd | 4493 | idx = ret_prog[frame]; |
70a87ffe | 4494 | goto continue_func; |
f4d7e40a AS |
4495 | } |
4496 | ||
19d28fbd | 4497 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
4498 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
4499 | const struct bpf_insn *insn, int idx) | |
4500 | { | |
4501 | int start = idx + insn->imm + 1, subprog; | |
4502 | ||
4503 | subprog = find_subprog(env, start); | |
4504 | if (subprog < 0) { | |
4505 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
4506 | start); | |
4507 | return -EFAULT; | |
4508 | } | |
9c8105bd | 4509 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 4510 | } |
19d28fbd | 4511 | #endif |
1ea47e01 | 4512 | |
afbf21dc YS |
4513 | static int __check_buffer_access(struct bpf_verifier_env *env, |
4514 | const char *buf_info, | |
4515 | const struct bpf_reg_state *reg, | |
4516 | int regno, int off, int size) | |
9df1c28b MM |
4517 | { |
4518 | if (off < 0) { | |
4519 | verbose(env, | |
4fc00b79 | 4520 | "R%d invalid %s buffer access: off=%d, size=%d\n", |
afbf21dc | 4521 | regno, buf_info, off, size); |
9df1c28b MM |
4522 | return -EACCES; |
4523 | } | |
4524 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
4525 | char tn_buf[48]; | |
4526 | ||
4527 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
4528 | verbose(env, | |
4fc00b79 | 4529 | "R%d invalid variable buffer offset: off=%d, var_off=%s\n", |
9df1c28b MM |
4530 | regno, off, tn_buf); |
4531 | return -EACCES; | |
4532 | } | |
afbf21dc YS |
4533 | |
4534 | return 0; | |
4535 | } | |
4536 | ||
4537 | static int check_tp_buffer_access(struct bpf_verifier_env *env, | |
4538 | const struct bpf_reg_state *reg, | |
4539 | int regno, int off, int size) | |
4540 | { | |
4541 | int err; | |
4542 | ||
4543 | err = __check_buffer_access(env, "tracepoint", reg, regno, off, size); | |
4544 | if (err) | |
4545 | return err; | |
4546 | ||
9df1c28b MM |
4547 | if (off + size > env->prog->aux->max_tp_access) |
4548 | env->prog->aux->max_tp_access = off + size; | |
4549 | ||
4550 | return 0; | |
4551 | } | |
4552 | ||
afbf21dc YS |
4553 | static int check_buffer_access(struct bpf_verifier_env *env, |
4554 | const struct bpf_reg_state *reg, | |
4555 | int regno, int off, int size, | |
4556 | bool zero_size_allowed, | |
afbf21dc YS |
4557 | u32 *max_access) |
4558 | { | |
44e9a741 | 4559 | const char *buf_info = type_is_rdonly_mem(reg->type) ? "rdonly" : "rdwr"; |
afbf21dc YS |
4560 | int err; |
4561 | ||
4562 | err = __check_buffer_access(env, buf_info, reg, regno, off, size); | |
4563 | if (err) | |
4564 | return err; | |
4565 | ||
4566 | if (off + size > *max_access) | |
4567 | *max_access = off + size; | |
4568 | ||
4569 | return 0; | |
4570 | } | |
4571 | ||
3f50f132 JF |
4572 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
4573 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
4574 | { | |
4575 | reg->var_off = tnum_subreg(reg->var_off); | |
4576 | __reg_assign_32_into_64(reg); | |
4577 | } | |
9df1c28b | 4578 | |
0c17d1d2 JH |
4579 | /* truncate register to smaller size (in bytes) |
4580 | * must be called with size < BPF_REG_SIZE | |
4581 | */ | |
4582 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
4583 | { | |
4584 | u64 mask; | |
4585 | ||
4586 | /* clear high bits in bit representation */ | |
4587 | reg->var_off = tnum_cast(reg->var_off, size); | |
4588 | ||
4589 | /* fix arithmetic bounds */ | |
4590 | mask = ((u64)1 << (size * 8)) - 1; | |
4591 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
4592 | reg->umin_value &= mask; | |
4593 | reg->umax_value &= mask; | |
4594 | } else { | |
4595 | reg->umin_value = 0; | |
4596 | reg->umax_value = mask; | |
4597 | } | |
4598 | reg->smin_value = reg->umin_value; | |
4599 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
4600 | |
4601 | /* If size is smaller than 32bit register the 32bit register | |
4602 | * values are also truncated so we push 64-bit bounds into | |
4603 | * 32-bit bounds. Above were truncated < 32-bits already. | |
4604 | */ | |
4605 | if (size >= 4) | |
4606 | return; | |
4607 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
4608 | } |
4609 | ||
a23740ec AN |
4610 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
4611 | { | |
353050be DB |
4612 | /* A map is considered read-only if the following condition are true: |
4613 | * | |
4614 | * 1) BPF program side cannot change any of the map content. The | |
4615 | * BPF_F_RDONLY_PROG flag is throughout the lifetime of a map | |
4616 | * and was set at map creation time. | |
4617 | * 2) The map value(s) have been initialized from user space by a | |
4618 | * loader and then "frozen", such that no new map update/delete | |
4619 | * operations from syscall side are possible for the rest of | |
4620 | * the map's lifetime from that point onwards. | |
4621 | * 3) Any parallel/pending map update/delete operations from syscall | |
4622 | * side have been completed. Only after that point, it's safe to | |
4623 | * assume that map value(s) are immutable. | |
4624 | */ | |
4625 | return (map->map_flags & BPF_F_RDONLY_PROG) && | |
4626 | READ_ONCE(map->frozen) && | |
4627 | !bpf_map_write_active(map); | |
a23740ec AN |
4628 | } |
4629 | ||
4630 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val) | |
4631 | { | |
4632 | void *ptr; | |
4633 | u64 addr; | |
4634 | int err; | |
4635 | ||
4636 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
4637 | if (err) | |
4638 | return err; | |
2dedd7d2 | 4639 | ptr = (void *)(long)addr + off; |
a23740ec AN |
4640 | |
4641 | switch (size) { | |
4642 | case sizeof(u8): | |
4643 | *val = (u64)*(u8 *)ptr; | |
4644 | break; | |
4645 | case sizeof(u16): | |
4646 | *val = (u64)*(u16 *)ptr; | |
4647 | break; | |
4648 | case sizeof(u32): | |
4649 | *val = (u64)*(u32 *)ptr; | |
4650 | break; | |
4651 | case sizeof(u64): | |
4652 | *val = *(u64 *)ptr; | |
4653 | break; | |
4654 | default: | |
4655 | return -EINVAL; | |
4656 | } | |
4657 | return 0; | |
4658 | } | |
4659 | ||
9e15db66 AS |
4660 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
4661 | struct bpf_reg_state *regs, | |
4662 | int regno, int off, int size, | |
4663 | enum bpf_access_type atype, | |
4664 | int value_regno) | |
4665 | { | |
4666 | struct bpf_reg_state *reg = regs + regno; | |
22dc4a0f AN |
4667 | const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id); |
4668 | const char *tname = btf_name_by_offset(reg->btf, t->name_off); | |
c6f1bfe8 | 4669 | enum bpf_type_flag flag = 0; |
9e15db66 AS |
4670 | u32 btf_id; |
4671 | int ret; | |
4672 | ||
9e15db66 AS |
4673 | if (off < 0) { |
4674 | verbose(env, | |
4675 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
4676 | regno, tname, off); | |
4677 | return -EACCES; | |
4678 | } | |
4679 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
4680 | char tn_buf[48]; | |
4681 | ||
4682 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
4683 | verbose(env, | |
4684 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
4685 | regno, tname, off, tn_buf); | |
4686 | return -EACCES; | |
4687 | } | |
4688 | ||
c6f1bfe8 YS |
4689 | if (reg->type & MEM_USER) { |
4690 | verbose(env, | |
4691 | "R%d is ptr_%s access user memory: off=%d\n", | |
4692 | regno, tname, off); | |
4693 | return -EACCES; | |
4694 | } | |
4695 | ||
5844101a HL |
4696 | if (reg->type & MEM_PERCPU) { |
4697 | verbose(env, | |
4698 | "R%d is ptr_%s access percpu memory: off=%d\n", | |
4699 | regno, tname, off); | |
4700 | return -EACCES; | |
4701 | } | |
4702 | ||
282de143 KKD |
4703 | if (env->ops->btf_struct_access && !type_is_alloc(reg->type)) { |
4704 | if (!btf_is_kernel(reg->btf)) { | |
4705 | verbose(env, "verifier internal error: reg->btf must be kernel btf\n"); | |
4706 | return -EFAULT; | |
4707 | } | |
6728aea7 | 4708 | ret = env->ops->btf_struct_access(&env->log, reg, off, size, atype, &btf_id, &flag); |
27ae7997 | 4709 | } else { |
282de143 KKD |
4710 | /* Writes are permitted with default btf_struct_access for |
4711 | * program allocated objects (which always have ref_obj_id > 0), | |
4712 | * but not for untrusted PTR_TO_BTF_ID | MEM_ALLOC. | |
4713 | */ | |
4714 | if (atype != BPF_READ && reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
27ae7997 MKL |
4715 | verbose(env, "only read is supported\n"); |
4716 | return -EACCES; | |
4717 | } | |
4718 | ||
282de143 KKD |
4719 | if (type_is_alloc(reg->type) && !reg->ref_obj_id) { |
4720 | verbose(env, "verifier internal error: ref_obj_id for allocated object must be non-zero\n"); | |
4721 | return -EFAULT; | |
4722 | } | |
4723 | ||
6728aea7 | 4724 | ret = btf_struct_access(&env->log, reg, off, size, atype, &btf_id, &flag); |
27ae7997 MKL |
4725 | } |
4726 | ||
9e15db66 AS |
4727 | if (ret < 0) |
4728 | return ret; | |
4729 | ||
6efe152d KKD |
4730 | /* If this is an untrusted pointer, all pointers formed by walking it |
4731 | * also inherit the untrusted flag. | |
4732 | */ | |
4733 | if (type_flag(reg->type) & PTR_UNTRUSTED) | |
4734 | flag |= PTR_UNTRUSTED; | |
4735 | ||
3f00c523 DV |
4736 | /* Any pointer obtained from walking a trusted pointer is no longer trusted. */ |
4737 | flag &= ~PTR_TRUSTED; | |
4738 | ||
41c48f3a | 4739 | if (atype == BPF_READ && value_regno >= 0) |
c6f1bfe8 | 4740 | mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id, flag); |
41c48f3a AI |
4741 | |
4742 | return 0; | |
4743 | } | |
4744 | ||
4745 | static int check_ptr_to_map_access(struct bpf_verifier_env *env, | |
4746 | struct bpf_reg_state *regs, | |
4747 | int regno, int off, int size, | |
4748 | enum bpf_access_type atype, | |
4749 | int value_regno) | |
4750 | { | |
4751 | struct bpf_reg_state *reg = regs + regno; | |
4752 | struct bpf_map *map = reg->map_ptr; | |
6728aea7 | 4753 | struct bpf_reg_state map_reg; |
c6f1bfe8 | 4754 | enum bpf_type_flag flag = 0; |
41c48f3a AI |
4755 | const struct btf_type *t; |
4756 | const char *tname; | |
4757 | u32 btf_id; | |
4758 | int ret; | |
4759 | ||
4760 | if (!btf_vmlinux) { | |
4761 | verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
4762 | return -ENOTSUPP; | |
4763 | } | |
4764 | ||
4765 | if (!map->ops->map_btf_id || !*map->ops->map_btf_id) { | |
4766 | verbose(env, "map_ptr access not supported for map type %d\n", | |
4767 | map->map_type); | |
4768 | return -ENOTSUPP; | |
4769 | } | |
4770 | ||
4771 | t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id); | |
4772 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
4773 | ||
4774 | if (!env->allow_ptr_to_map_access) { | |
4775 | verbose(env, | |
4776 | "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", | |
4777 | tname); | |
4778 | return -EPERM; | |
9e15db66 | 4779 | } |
27ae7997 | 4780 | |
41c48f3a AI |
4781 | if (off < 0) { |
4782 | verbose(env, "R%d is %s invalid negative access: off=%d\n", | |
4783 | regno, tname, off); | |
4784 | return -EACCES; | |
4785 | } | |
4786 | ||
4787 | if (atype != BPF_READ) { | |
4788 | verbose(env, "only read from %s is supported\n", tname); | |
4789 | return -EACCES; | |
4790 | } | |
4791 | ||
6728aea7 KKD |
4792 | /* Simulate access to a PTR_TO_BTF_ID */ |
4793 | memset(&map_reg, 0, sizeof(map_reg)); | |
4794 | mark_btf_ld_reg(env, &map_reg, 0, PTR_TO_BTF_ID, btf_vmlinux, *map->ops->map_btf_id, 0); | |
4795 | ret = btf_struct_access(&env->log, &map_reg, off, size, atype, &btf_id, &flag); | |
41c48f3a AI |
4796 | if (ret < 0) |
4797 | return ret; | |
4798 | ||
4799 | if (value_regno >= 0) | |
c6f1bfe8 | 4800 | mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id, flag); |
41c48f3a | 4801 | |
9e15db66 AS |
4802 | return 0; |
4803 | } | |
4804 | ||
01f810ac AM |
4805 | /* Check that the stack access at the given offset is within bounds. The |
4806 | * maximum valid offset is -1. | |
4807 | * | |
4808 | * The minimum valid offset is -MAX_BPF_STACK for writes, and | |
4809 | * -state->allocated_stack for reads. | |
4810 | */ | |
4811 | static int check_stack_slot_within_bounds(int off, | |
4812 | struct bpf_func_state *state, | |
4813 | enum bpf_access_type t) | |
4814 | { | |
4815 | int min_valid_off; | |
4816 | ||
4817 | if (t == BPF_WRITE) | |
4818 | min_valid_off = -MAX_BPF_STACK; | |
4819 | else | |
4820 | min_valid_off = -state->allocated_stack; | |
4821 | ||
4822 | if (off < min_valid_off || off > -1) | |
4823 | return -EACCES; | |
4824 | return 0; | |
4825 | } | |
4826 | ||
4827 | /* Check that the stack access at 'regno + off' falls within the maximum stack | |
4828 | * bounds. | |
4829 | * | |
4830 | * 'off' includes `regno->offset`, but not its dynamic part (if any). | |
4831 | */ | |
4832 | static int check_stack_access_within_bounds( | |
4833 | struct bpf_verifier_env *env, | |
4834 | int regno, int off, int access_size, | |
61df10c7 | 4835 | enum bpf_access_src src, enum bpf_access_type type) |
01f810ac AM |
4836 | { |
4837 | struct bpf_reg_state *regs = cur_regs(env); | |
4838 | struct bpf_reg_state *reg = regs + regno; | |
4839 | struct bpf_func_state *state = func(env, reg); | |
4840 | int min_off, max_off; | |
4841 | int err; | |
4842 | char *err_extra; | |
4843 | ||
4844 | if (src == ACCESS_HELPER) | |
4845 | /* We don't know if helpers are reading or writing (or both). */ | |
4846 | err_extra = " indirect access to"; | |
4847 | else if (type == BPF_READ) | |
4848 | err_extra = " read from"; | |
4849 | else | |
4850 | err_extra = " write to"; | |
4851 | ||
4852 | if (tnum_is_const(reg->var_off)) { | |
4853 | min_off = reg->var_off.value + off; | |
4854 | if (access_size > 0) | |
4855 | max_off = min_off + access_size - 1; | |
4856 | else | |
4857 | max_off = min_off; | |
4858 | } else { | |
4859 | if (reg->smax_value >= BPF_MAX_VAR_OFF || | |
4860 | reg->smin_value <= -BPF_MAX_VAR_OFF) { | |
4861 | verbose(env, "invalid unbounded variable-offset%s stack R%d\n", | |
4862 | err_extra, regno); | |
4863 | return -EACCES; | |
4864 | } | |
4865 | min_off = reg->smin_value + off; | |
4866 | if (access_size > 0) | |
4867 | max_off = reg->smax_value + off + access_size - 1; | |
4868 | else | |
4869 | max_off = min_off; | |
4870 | } | |
4871 | ||
4872 | err = check_stack_slot_within_bounds(min_off, state, type); | |
4873 | if (!err) | |
4874 | err = check_stack_slot_within_bounds(max_off, state, type); | |
4875 | ||
4876 | if (err) { | |
4877 | if (tnum_is_const(reg->var_off)) { | |
4878 | verbose(env, "invalid%s stack R%d off=%d size=%d\n", | |
4879 | err_extra, regno, off, access_size); | |
4880 | } else { | |
4881 | char tn_buf[48]; | |
4882 | ||
4883 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
4884 | verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n", | |
4885 | err_extra, regno, tn_buf, access_size); | |
4886 | } | |
4887 | } | |
4888 | return err; | |
4889 | } | |
41c48f3a | 4890 | |
17a52670 AS |
4891 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
4892 | * if t==write, value_regno is a register which value is stored into memory | |
4893 | * if t==read, value_regno is a register which will receive the value from memory | |
4894 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
4895 | * if t==read && value_regno==-1, don't care what we read from memory | |
4896 | */ | |
ca369602 DB |
4897 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
4898 | int off, int bpf_size, enum bpf_access_type t, | |
4899 | int value_regno, bool strict_alignment_once) | |
17a52670 | 4900 | { |
638f5b90 AS |
4901 | struct bpf_reg_state *regs = cur_regs(env); |
4902 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 4903 | struct bpf_func_state *state; |
17a52670 AS |
4904 | int size, err = 0; |
4905 | ||
4906 | size = bpf_size_to_bytes(bpf_size); | |
4907 | if (size < 0) | |
4908 | return size; | |
4909 | ||
f1174f77 | 4910 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 4911 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
4912 | if (err) |
4913 | return err; | |
17a52670 | 4914 | |
f1174f77 EC |
4915 | /* for access checks, reg->off is just part of off */ |
4916 | off += reg->off; | |
4917 | ||
69c087ba YS |
4918 | if (reg->type == PTR_TO_MAP_KEY) { |
4919 | if (t == BPF_WRITE) { | |
4920 | verbose(env, "write to change key R%d not allowed\n", regno); | |
4921 | return -EACCES; | |
4922 | } | |
4923 | ||
4924 | err = check_mem_region_access(env, regno, off, size, | |
4925 | reg->map_ptr->key_size, false); | |
4926 | if (err) | |
4927 | return err; | |
4928 | if (value_regno >= 0) | |
4929 | mark_reg_unknown(env, regs, value_regno); | |
4930 | } else if (reg->type == PTR_TO_MAP_VALUE) { | |
aa3496ac | 4931 | struct btf_field *kptr_field = NULL; |
61df10c7 | 4932 | |
1be7f75d AS |
4933 | if (t == BPF_WRITE && value_regno >= 0 && |
4934 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 4935 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
4936 | return -EACCES; |
4937 | } | |
591fe988 DB |
4938 | err = check_map_access_type(env, regno, off, size, t); |
4939 | if (err) | |
4940 | return err; | |
61df10c7 KKD |
4941 | err = check_map_access(env, regno, off, size, false, ACCESS_DIRECT); |
4942 | if (err) | |
4943 | return err; | |
4944 | if (tnum_is_const(reg->var_off)) | |
aa3496ac KKD |
4945 | kptr_field = btf_record_find(reg->map_ptr->record, |
4946 | off + reg->var_off.value, BPF_KPTR); | |
4947 | if (kptr_field) { | |
4948 | err = check_map_kptr_access(env, regno, value_regno, insn_idx, kptr_field); | |
61df10c7 | 4949 | } else if (t == BPF_READ && value_regno >= 0) { |
a23740ec AN |
4950 | struct bpf_map *map = reg->map_ptr; |
4951 | ||
4952 | /* if map is read-only, track its contents as scalars */ | |
4953 | if (tnum_is_const(reg->var_off) && | |
4954 | bpf_map_is_rdonly(map) && | |
4955 | map->ops->map_direct_value_addr) { | |
4956 | int map_off = off + reg->var_off.value; | |
4957 | u64 val = 0; | |
4958 | ||
4959 | err = bpf_map_direct_read(map, map_off, size, | |
4960 | &val); | |
4961 | if (err) | |
4962 | return err; | |
4963 | ||
4964 | regs[value_regno].type = SCALAR_VALUE; | |
4965 | __mark_reg_known(®s[value_regno], val); | |
4966 | } else { | |
4967 | mark_reg_unknown(env, regs, value_regno); | |
4968 | } | |
4969 | } | |
34d3a78c HL |
4970 | } else if (base_type(reg->type) == PTR_TO_MEM) { |
4971 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
4972 | ||
4973 | if (type_may_be_null(reg->type)) { | |
4974 | verbose(env, "R%d invalid mem access '%s'\n", regno, | |
4975 | reg_type_str(env, reg->type)); | |
4976 | return -EACCES; | |
4977 | } | |
4978 | ||
4979 | if (t == BPF_WRITE && rdonly_mem) { | |
4980 | verbose(env, "R%d cannot write into %s\n", | |
4981 | regno, reg_type_str(env, reg->type)); | |
4982 | return -EACCES; | |
4983 | } | |
4984 | ||
457f4436 AN |
4985 | if (t == BPF_WRITE && value_regno >= 0 && |
4986 | is_pointer_value(env, value_regno)) { | |
4987 | verbose(env, "R%d leaks addr into mem\n", value_regno); | |
4988 | return -EACCES; | |
4989 | } | |
34d3a78c | 4990 | |
457f4436 AN |
4991 | err = check_mem_region_access(env, regno, off, size, |
4992 | reg->mem_size, false); | |
34d3a78c | 4993 | if (!err && value_regno >= 0 && (t == BPF_READ || rdonly_mem)) |
457f4436 | 4994 | mark_reg_unknown(env, regs, value_regno); |
1a0dc1ac | 4995 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 4996 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
22dc4a0f | 4997 | struct btf *btf = NULL; |
9e15db66 | 4998 | u32 btf_id = 0; |
19de99f7 | 4999 | |
1be7f75d AS |
5000 | if (t == BPF_WRITE && value_regno >= 0 && |
5001 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 5002 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
5003 | return -EACCES; |
5004 | } | |
f1174f77 | 5005 | |
be80a1d3 | 5006 | err = check_ptr_off_reg(env, reg, regno); |
58990d1f DB |
5007 | if (err < 0) |
5008 | return err; | |
5009 | ||
c6f1bfe8 YS |
5010 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf, |
5011 | &btf_id); | |
9e15db66 AS |
5012 | if (err) |
5013 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 5014 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 5015 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
5016 | * PTR_TO_PACKET[_META,_END]. In the latter |
5017 | * case, we know the offset is zero. | |
f1174f77 | 5018 | */ |
46f8bc92 | 5019 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 5020 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 5021 | } else { |
638f5b90 | 5022 | mark_reg_known_zero(env, regs, |
61bd5218 | 5023 | value_regno); |
c25b2ae1 | 5024 | if (type_may_be_null(reg_type)) |
46f8bc92 | 5025 | regs[value_regno].id = ++env->id_gen; |
5327ed3d JW |
5026 | /* A load of ctx field could have different |
5027 | * actual load size with the one encoded in the | |
5028 | * insn. When the dst is PTR, it is for sure not | |
5029 | * a sub-register. | |
5030 | */ | |
5031 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
c25b2ae1 | 5032 | if (base_type(reg_type) == PTR_TO_BTF_ID) { |
22dc4a0f | 5033 | regs[value_regno].btf = btf; |
9e15db66 | 5034 | regs[value_regno].btf_id = btf_id; |
22dc4a0f | 5035 | } |
46f8bc92 | 5036 | } |
638f5b90 | 5037 | regs[value_regno].type = reg_type; |
969bf05e | 5038 | } |
17a52670 | 5039 | |
f1174f77 | 5040 | } else if (reg->type == PTR_TO_STACK) { |
01f810ac AM |
5041 | /* Basic bounds checks. */ |
5042 | err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t); | |
e4298d25 DB |
5043 | if (err) |
5044 | return err; | |
8726679a | 5045 | |
f4d7e40a AS |
5046 | state = func(env, reg); |
5047 | err = update_stack_depth(env, state, off); | |
5048 | if (err) | |
5049 | return err; | |
8726679a | 5050 | |
01f810ac AM |
5051 | if (t == BPF_READ) |
5052 | err = check_stack_read(env, regno, off, size, | |
61bd5218 | 5053 | value_regno); |
01f810ac AM |
5054 | else |
5055 | err = check_stack_write(env, regno, off, size, | |
5056 | value_regno, insn_idx); | |
de8f3a83 | 5057 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 5058 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 5059 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
5060 | return -EACCES; |
5061 | } | |
4acf6c0b BB |
5062 | if (t == BPF_WRITE && value_regno >= 0 && |
5063 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
5064 | verbose(env, "R%d leaks addr into packet\n", |
5065 | value_regno); | |
4acf6c0b BB |
5066 | return -EACCES; |
5067 | } | |
9fd29c08 | 5068 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 5069 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 5070 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
5071 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
5072 | if (t == BPF_WRITE && value_regno >= 0 && | |
5073 | is_pointer_value(env, value_regno)) { | |
5074 | verbose(env, "R%d leaks addr into flow keys\n", | |
5075 | value_regno); | |
5076 | return -EACCES; | |
5077 | } | |
5078 | ||
5079 | err = check_flow_keys_access(env, off, size); | |
5080 | if (!err && t == BPF_READ && value_regno >= 0) | |
5081 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 5082 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 5083 | if (t == BPF_WRITE) { |
46f8bc92 | 5084 | verbose(env, "R%d cannot write into %s\n", |
c25b2ae1 | 5085 | regno, reg_type_str(env, reg->type)); |
c64b7983 JS |
5086 | return -EACCES; |
5087 | } | |
5f456649 | 5088 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
5089 | if (!err && value_regno >= 0) |
5090 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
5091 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
5092 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
5093 | if (!err && t == BPF_READ && value_regno >= 0) | |
5094 | mark_reg_unknown(env, regs, value_regno); | |
bff61f6f HL |
5095 | } else if (base_type(reg->type) == PTR_TO_BTF_ID && |
5096 | !type_may_be_null(reg->type)) { | |
9e15db66 AS |
5097 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, |
5098 | value_regno); | |
41c48f3a AI |
5099 | } else if (reg->type == CONST_PTR_TO_MAP) { |
5100 | err = check_ptr_to_map_access(env, regs, regno, off, size, t, | |
5101 | value_regno); | |
20b2aff4 HL |
5102 | } else if (base_type(reg->type) == PTR_TO_BUF) { |
5103 | bool rdonly_mem = type_is_rdonly_mem(reg->type); | |
20b2aff4 HL |
5104 | u32 *max_access; |
5105 | ||
5106 | if (rdonly_mem) { | |
5107 | if (t == BPF_WRITE) { | |
5108 | verbose(env, "R%d cannot write into %s\n", | |
5109 | regno, reg_type_str(env, reg->type)); | |
5110 | return -EACCES; | |
5111 | } | |
20b2aff4 HL |
5112 | max_access = &env->prog->aux->max_rdonly_access; |
5113 | } else { | |
20b2aff4 | 5114 | max_access = &env->prog->aux->max_rdwr_access; |
afbf21dc | 5115 | } |
20b2aff4 | 5116 | |
f6dfbe31 | 5117 | err = check_buffer_access(env, reg, regno, off, size, false, |
44e9a741 | 5118 | max_access); |
20b2aff4 HL |
5119 | |
5120 | if (!err && value_regno >= 0 && (rdonly_mem || t == BPF_READ)) | |
afbf21dc | 5121 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 5122 | } else { |
61bd5218 | 5123 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
c25b2ae1 | 5124 | reg_type_str(env, reg->type)); |
17a52670 AS |
5125 | return -EACCES; |
5126 | } | |
969bf05e | 5127 | |
f1174f77 | 5128 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 5129 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 5130 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 5131 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 5132 | } |
17a52670 AS |
5133 | return err; |
5134 | } | |
5135 | ||
91c960b0 | 5136 | static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 5137 | { |
5ffa2550 | 5138 | int load_reg; |
17a52670 AS |
5139 | int err; |
5140 | ||
5ca419f2 BJ |
5141 | switch (insn->imm) { |
5142 | case BPF_ADD: | |
5143 | case BPF_ADD | BPF_FETCH: | |
981f94c3 BJ |
5144 | case BPF_AND: |
5145 | case BPF_AND | BPF_FETCH: | |
5146 | case BPF_OR: | |
5147 | case BPF_OR | BPF_FETCH: | |
5148 | case BPF_XOR: | |
5149 | case BPF_XOR | BPF_FETCH: | |
5ffa2550 BJ |
5150 | case BPF_XCHG: |
5151 | case BPF_CMPXCHG: | |
5ca419f2 BJ |
5152 | break; |
5153 | default: | |
91c960b0 BJ |
5154 | verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm); |
5155 | return -EINVAL; | |
5156 | } | |
5157 | ||
5158 | if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) { | |
5159 | verbose(env, "invalid atomic operand size\n"); | |
17a52670 AS |
5160 | return -EINVAL; |
5161 | } | |
5162 | ||
5163 | /* check src1 operand */ | |
dc503a8a | 5164 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
5165 | if (err) |
5166 | return err; | |
5167 | ||
5168 | /* check src2 operand */ | |
dc503a8a | 5169 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
5170 | if (err) |
5171 | return err; | |
5172 | ||
5ffa2550 BJ |
5173 | if (insn->imm == BPF_CMPXCHG) { |
5174 | /* Check comparison of R0 with memory location */ | |
a82fe085 DB |
5175 | const u32 aux_reg = BPF_REG_0; |
5176 | ||
5177 | err = check_reg_arg(env, aux_reg, SRC_OP); | |
5ffa2550 BJ |
5178 | if (err) |
5179 | return err; | |
a82fe085 DB |
5180 | |
5181 | if (is_pointer_value(env, aux_reg)) { | |
5182 | verbose(env, "R%d leaks addr into mem\n", aux_reg); | |
5183 | return -EACCES; | |
5184 | } | |
5ffa2550 BJ |
5185 | } |
5186 | ||
6bdf6abc | 5187 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 5188 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
5189 | return -EACCES; |
5190 | } | |
5191 | ||
ca369602 | 5192 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 5193 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
5194 | is_flow_key_reg(env, insn->dst_reg) || |
5195 | is_sk_reg(env, insn->dst_reg)) { | |
91c960b0 | 5196 | verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n", |
2a159c6f | 5197 | insn->dst_reg, |
c25b2ae1 | 5198 | reg_type_str(env, reg_state(env, insn->dst_reg)->type)); |
f37a8cb8 DB |
5199 | return -EACCES; |
5200 | } | |
5201 | ||
37086bfd BJ |
5202 | if (insn->imm & BPF_FETCH) { |
5203 | if (insn->imm == BPF_CMPXCHG) | |
5204 | load_reg = BPF_REG_0; | |
5205 | else | |
5206 | load_reg = insn->src_reg; | |
5207 | ||
5208 | /* check and record load of old value */ | |
5209 | err = check_reg_arg(env, load_reg, DST_OP); | |
5210 | if (err) | |
5211 | return err; | |
5212 | } else { | |
5213 | /* This instruction accesses a memory location but doesn't | |
5214 | * actually load it into a register. | |
5215 | */ | |
5216 | load_reg = -1; | |
5217 | } | |
5218 | ||
7d3baf0a DB |
5219 | /* Check whether we can read the memory, with second call for fetch |
5220 | * case to simulate the register fill. | |
5221 | */ | |
31fd8581 | 5222 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
7d3baf0a DB |
5223 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
5224 | if (!err && load_reg >= 0) | |
5225 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, | |
5226 | BPF_SIZE(insn->code), BPF_READ, load_reg, | |
5227 | true); | |
17a52670 AS |
5228 | if (err) |
5229 | return err; | |
5230 | ||
7d3baf0a | 5231 | /* Check whether we can write into the same memory. */ |
5ca419f2 BJ |
5232 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
5233 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); | |
5234 | if (err) | |
5235 | return err; | |
5236 | ||
5ca419f2 | 5237 | return 0; |
17a52670 AS |
5238 | } |
5239 | ||
01f810ac AM |
5240 | /* When register 'regno' is used to read the stack (either directly or through |
5241 | * a helper function) make sure that it's within stack boundary and, depending | |
5242 | * on the access type, that all elements of the stack are initialized. | |
5243 | * | |
5244 | * 'off' includes 'regno->off', but not its dynamic part (if any). | |
5245 | * | |
5246 | * All registers that have been spilled on the stack in the slots within the | |
5247 | * read offsets are marked as read. | |
5248 | */ | |
5249 | static int check_stack_range_initialized( | |
5250 | struct bpf_verifier_env *env, int regno, int off, | |
5251 | int access_size, bool zero_size_allowed, | |
61df10c7 | 5252 | enum bpf_access_src type, struct bpf_call_arg_meta *meta) |
2011fccf AI |
5253 | { |
5254 | struct bpf_reg_state *reg = reg_state(env, regno); | |
01f810ac AM |
5255 | struct bpf_func_state *state = func(env, reg); |
5256 | int err, min_off, max_off, i, j, slot, spi; | |
5257 | char *err_extra = type == ACCESS_HELPER ? " indirect" : ""; | |
5258 | enum bpf_access_type bounds_check_type; | |
5259 | /* Some accesses can write anything into the stack, others are | |
5260 | * read-only. | |
5261 | */ | |
5262 | bool clobber = false; | |
2011fccf | 5263 | |
01f810ac AM |
5264 | if (access_size == 0 && !zero_size_allowed) { |
5265 | verbose(env, "invalid zero-sized read\n"); | |
2011fccf AI |
5266 | return -EACCES; |
5267 | } | |
2011fccf | 5268 | |
01f810ac AM |
5269 | if (type == ACCESS_HELPER) { |
5270 | /* The bounds checks for writes are more permissive than for | |
5271 | * reads. However, if raw_mode is not set, we'll do extra | |
5272 | * checks below. | |
5273 | */ | |
5274 | bounds_check_type = BPF_WRITE; | |
5275 | clobber = true; | |
5276 | } else { | |
5277 | bounds_check_type = BPF_READ; | |
5278 | } | |
5279 | err = check_stack_access_within_bounds(env, regno, off, access_size, | |
5280 | type, bounds_check_type); | |
5281 | if (err) | |
5282 | return err; | |
5283 | ||
17a52670 | 5284 | |
2011fccf | 5285 | if (tnum_is_const(reg->var_off)) { |
01f810ac | 5286 | min_off = max_off = reg->var_off.value + off; |
2011fccf | 5287 | } else { |
088ec26d AI |
5288 | /* Variable offset is prohibited for unprivileged mode for |
5289 | * simplicity since it requires corresponding support in | |
5290 | * Spectre masking for stack ALU. | |
5291 | * See also retrieve_ptr_limit(). | |
5292 | */ | |
2c78ee89 | 5293 | if (!env->bypass_spec_v1) { |
088ec26d | 5294 | char tn_buf[48]; |
f1174f77 | 5295 | |
088ec26d | 5296 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
01f810ac AM |
5297 | verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n", |
5298 | regno, err_extra, tn_buf); | |
088ec26d AI |
5299 | return -EACCES; |
5300 | } | |
f2bcd05e AI |
5301 | /* Only initialized buffer on stack is allowed to be accessed |
5302 | * with variable offset. With uninitialized buffer it's hard to | |
5303 | * guarantee that whole memory is marked as initialized on | |
5304 | * helper return since specific bounds are unknown what may | |
5305 | * cause uninitialized stack leaking. | |
5306 | */ | |
5307 | if (meta && meta->raw_mode) | |
5308 | meta = NULL; | |
5309 | ||
01f810ac AM |
5310 | min_off = reg->smin_value + off; |
5311 | max_off = reg->smax_value + off; | |
17a52670 AS |
5312 | } |
5313 | ||
435faee1 DB |
5314 | if (meta && meta->raw_mode) { |
5315 | meta->access_size = access_size; | |
5316 | meta->regno = regno; | |
5317 | return 0; | |
5318 | } | |
5319 | ||
2011fccf | 5320 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
5321 | u8 *stype; |
5322 | ||
2011fccf | 5323 | slot = -i - 1; |
638f5b90 | 5324 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
5325 | if (state->allocated_stack <= slot) |
5326 | goto err; | |
5327 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
5328 | if (*stype == STACK_MISC) | |
5329 | goto mark; | |
5330 | if (*stype == STACK_ZERO) { | |
01f810ac AM |
5331 | if (clobber) { |
5332 | /* helper can write anything into the stack */ | |
5333 | *stype = STACK_MISC; | |
5334 | } | |
cc2b14d5 | 5335 | goto mark; |
17a52670 | 5336 | } |
1d68f22b | 5337 | |
27113c59 | 5338 | if (is_spilled_reg(&state->stack[spi]) && |
cd17d38f YS |
5339 | (state->stack[spi].spilled_ptr.type == SCALAR_VALUE || |
5340 | env->allow_ptr_leaks)) { | |
01f810ac AM |
5341 | if (clobber) { |
5342 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); | |
5343 | for (j = 0; j < BPF_REG_SIZE; j++) | |
354e8f19 | 5344 | scrub_spilled_slot(&state->stack[spi].slot_type[j]); |
01f810ac | 5345 | } |
f7cf25b2 AS |
5346 | goto mark; |
5347 | } | |
5348 | ||
cc2b14d5 | 5349 | err: |
2011fccf | 5350 | if (tnum_is_const(reg->var_off)) { |
01f810ac AM |
5351 | verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n", |
5352 | err_extra, regno, min_off, i - min_off, access_size); | |
2011fccf AI |
5353 | } else { |
5354 | char tn_buf[48]; | |
5355 | ||
5356 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac AM |
5357 | verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n", |
5358 | err_extra, regno, tn_buf, i - min_off, access_size); | |
2011fccf | 5359 | } |
cc2b14d5 AS |
5360 | return -EACCES; |
5361 | mark: | |
5362 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
5363 | * the whole slot to be marked as 'read' | |
5364 | */ | |
679c782d | 5365 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
5366 | state->stack[spi].spilled_ptr.parent, |
5367 | REG_LIVE_READ64); | |
261f4664 KKD |
5368 | /* We do not set REG_LIVE_WRITTEN for stack slot, as we can not |
5369 | * be sure that whether stack slot is written to or not. Hence, | |
5370 | * we must still conservatively propagate reads upwards even if | |
5371 | * helper may write to the entire memory range. | |
5372 | */ | |
17a52670 | 5373 | } |
2011fccf | 5374 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
5375 | } |
5376 | ||
06c1c049 GB |
5377 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
5378 | int access_size, bool zero_size_allowed, | |
5379 | struct bpf_call_arg_meta *meta) | |
5380 | { | |
638f5b90 | 5381 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
20b2aff4 | 5382 | u32 *max_access; |
06c1c049 | 5383 | |
20b2aff4 | 5384 | switch (base_type(reg->type)) { |
06c1c049 | 5385 | case PTR_TO_PACKET: |
de8f3a83 | 5386 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
5387 | return check_packet_access(env, regno, reg->off, access_size, |
5388 | zero_size_allowed); | |
69c087ba | 5389 | case PTR_TO_MAP_KEY: |
7b3552d3 KKD |
5390 | if (meta && meta->raw_mode) { |
5391 | verbose(env, "R%d cannot write into %s\n", regno, | |
5392 | reg_type_str(env, reg->type)); | |
5393 | return -EACCES; | |
5394 | } | |
69c087ba YS |
5395 | return check_mem_region_access(env, regno, reg->off, access_size, |
5396 | reg->map_ptr->key_size, false); | |
06c1c049 | 5397 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
5398 | if (check_map_access_type(env, regno, reg->off, access_size, |
5399 | meta && meta->raw_mode ? BPF_WRITE : | |
5400 | BPF_READ)) | |
5401 | return -EACCES; | |
9fd29c08 | 5402 | return check_map_access(env, regno, reg->off, access_size, |
61df10c7 | 5403 | zero_size_allowed, ACCESS_HELPER); |
457f4436 | 5404 | case PTR_TO_MEM: |
97e6d7da KKD |
5405 | if (type_is_rdonly_mem(reg->type)) { |
5406 | if (meta && meta->raw_mode) { | |
5407 | verbose(env, "R%d cannot write into %s\n", regno, | |
5408 | reg_type_str(env, reg->type)); | |
5409 | return -EACCES; | |
5410 | } | |
5411 | } | |
457f4436 AN |
5412 | return check_mem_region_access(env, regno, reg->off, |
5413 | access_size, reg->mem_size, | |
5414 | zero_size_allowed); | |
20b2aff4 HL |
5415 | case PTR_TO_BUF: |
5416 | if (type_is_rdonly_mem(reg->type)) { | |
97e6d7da KKD |
5417 | if (meta && meta->raw_mode) { |
5418 | verbose(env, "R%d cannot write into %s\n", regno, | |
5419 | reg_type_str(env, reg->type)); | |
20b2aff4 | 5420 | return -EACCES; |
97e6d7da | 5421 | } |
20b2aff4 | 5422 | |
20b2aff4 HL |
5423 | max_access = &env->prog->aux->max_rdonly_access; |
5424 | } else { | |
20b2aff4 HL |
5425 | max_access = &env->prog->aux->max_rdwr_access; |
5426 | } | |
afbf21dc YS |
5427 | return check_buffer_access(env, reg, regno, reg->off, |
5428 | access_size, zero_size_allowed, | |
44e9a741 | 5429 | max_access); |
0d004c02 | 5430 | case PTR_TO_STACK: |
01f810ac AM |
5431 | return check_stack_range_initialized( |
5432 | env, | |
5433 | regno, reg->off, access_size, | |
5434 | zero_size_allowed, ACCESS_HELPER, meta); | |
15baa55f BT |
5435 | case PTR_TO_CTX: |
5436 | /* in case the function doesn't know how to access the context, | |
5437 | * (because we are in a program of type SYSCALL for example), we | |
5438 | * can not statically check its size. | |
5439 | * Dynamically check it now. | |
5440 | */ | |
5441 | if (!env->ops->convert_ctx_access) { | |
5442 | enum bpf_access_type atype = meta && meta->raw_mode ? BPF_WRITE : BPF_READ; | |
5443 | int offset = access_size - 1; | |
5444 | ||
5445 | /* Allow zero-byte read from PTR_TO_CTX */ | |
5446 | if (access_size == 0) | |
5447 | return zero_size_allowed ? 0 : -EACCES; | |
5448 | ||
5449 | return check_mem_access(env, env->insn_idx, regno, offset, BPF_B, | |
5450 | atype, -1, false); | |
5451 | } | |
5452 | ||
5453 | fallthrough; | |
0d004c02 LB |
5454 | default: /* scalar_value or invalid ptr */ |
5455 | /* Allow zero-byte read from NULL, regardless of pointer type */ | |
5456 | if (zero_size_allowed && access_size == 0 && | |
5457 | register_is_null(reg)) | |
5458 | return 0; | |
5459 | ||
c25b2ae1 HL |
5460 | verbose(env, "R%d type=%s ", regno, |
5461 | reg_type_str(env, reg->type)); | |
5462 | verbose(env, "expected=%s\n", reg_type_str(env, PTR_TO_STACK)); | |
0d004c02 | 5463 | return -EACCES; |
06c1c049 GB |
5464 | } |
5465 | } | |
5466 | ||
d583691c KKD |
5467 | static int check_mem_size_reg(struct bpf_verifier_env *env, |
5468 | struct bpf_reg_state *reg, u32 regno, | |
5469 | bool zero_size_allowed, | |
5470 | struct bpf_call_arg_meta *meta) | |
5471 | { | |
5472 | int err; | |
5473 | ||
5474 | /* This is used to refine r0 return value bounds for helpers | |
5475 | * that enforce this value as an upper bound on return values. | |
5476 | * See do_refine_retval_range() for helpers that can refine | |
5477 | * the return value. C type of helper is u32 so we pull register | |
5478 | * bound from umax_value however, if negative verifier errors | |
5479 | * out. Only upper bounds can be learned because retval is an | |
5480 | * int type and negative retvals are allowed. | |
5481 | */ | |
be77354a | 5482 | meta->msize_max_value = reg->umax_value; |
d583691c KKD |
5483 | |
5484 | /* The register is SCALAR_VALUE; the access check | |
5485 | * happens using its boundaries. | |
5486 | */ | |
5487 | if (!tnum_is_const(reg->var_off)) | |
5488 | /* For unprivileged variable accesses, disable raw | |
5489 | * mode so that the program is required to | |
5490 | * initialize all the memory that the helper could | |
5491 | * just partially fill up. | |
5492 | */ | |
5493 | meta = NULL; | |
5494 | ||
5495 | if (reg->smin_value < 0) { | |
5496 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", | |
5497 | regno); | |
5498 | return -EACCES; | |
5499 | } | |
5500 | ||
5501 | if (reg->umin_value == 0) { | |
5502 | err = check_helper_mem_access(env, regno - 1, 0, | |
5503 | zero_size_allowed, | |
5504 | meta); | |
5505 | if (err) | |
5506 | return err; | |
5507 | } | |
5508 | ||
5509 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { | |
5510 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", | |
5511 | regno); | |
5512 | return -EACCES; | |
5513 | } | |
5514 | err = check_helper_mem_access(env, regno - 1, | |
5515 | reg->umax_value, | |
5516 | zero_size_allowed, meta); | |
5517 | if (!err) | |
5518 | err = mark_chain_precision(env, regno); | |
5519 | return err; | |
5520 | } | |
5521 | ||
e5069b9c DB |
5522 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
5523 | u32 regno, u32 mem_size) | |
5524 | { | |
be77354a KKD |
5525 | bool may_be_null = type_may_be_null(reg->type); |
5526 | struct bpf_reg_state saved_reg; | |
5527 | struct bpf_call_arg_meta meta; | |
5528 | int err; | |
5529 | ||
e5069b9c DB |
5530 | if (register_is_null(reg)) |
5531 | return 0; | |
5532 | ||
be77354a KKD |
5533 | memset(&meta, 0, sizeof(meta)); |
5534 | /* Assuming that the register contains a value check if the memory | |
5535 | * access is safe. Temporarily save and restore the register's state as | |
5536 | * the conversion shouldn't be visible to a caller. | |
5537 | */ | |
5538 | if (may_be_null) { | |
5539 | saved_reg = *reg; | |
e5069b9c | 5540 | mark_ptr_not_null_reg(reg); |
e5069b9c DB |
5541 | } |
5542 | ||
be77354a KKD |
5543 | err = check_helper_mem_access(env, regno, mem_size, true, &meta); |
5544 | /* Check access for BPF_WRITE */ | |
5545 | meta.raw_mode = true; | |
5546 | err = err ?: check_helper_mem_access(env, regno, mem_size, true, &meta); | |
5547 | ||
5548 | if (may_be_null) | |
5549 | *reg = saved_reg; | |
5550 | ||
5551 | return err; | |
e5069b9c DB |
5552 | } |
5553 | ||
00b85860 KKD |
5554 | static int check_kfunc_mem_size_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
5555 | u32 regno) | |
d583691c KKD |
5556 | { |
5557 | struct bpf_reg_state *mem_reg = &cur_regs(env)[regno - 1]; | |
5558 | bool may_be_null = type_may_be_null(mem_reg->type); | |
5559 | struct bpf_reg_state saved_reg; | |
be77354a | 5560 | struct bpf_call_arg_meta meta; |
d583691c KKD |
5561 | int err; |
5562 | ||
5563 | WARN_ON_ONCE(regno < BPF_REG_2 || regno > BPF_REG_5); | |
5564 | ||
be77354a KKD |
5565 | memset(&meta, 0, sizeof(meta)); |
5566 | ||
d583691c KKD |
5567 | if (may_be_null) { |
5568 | saved_reg = *mem_reg; | |
5569 | mark_ptr_not_null_reg(mem_reg); | |
5570 | } | |
5571 | ||
be77354a KKD |
5572 | err = check_mem_size_reg(env, reg, regno, true, &meta); |
5573 | /* Check access for BPF_WRITE */ | |
5574 | meta.raw_mode = true; | |
5575 | err = err ?: check_mem_size_reg(env, reg, regno, true, &meta); | |
d583691c KKD |
5576 | |
5577 | if (may_be_null) | |
5578 | *mem_reg = saved_reg; | |
5579 | return err; | |
5580 | } | |
5581 | ||
d83525ca | 5582 | /* Implementation details: |
4e814da0 KKD |
5583 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL. |
5584 | * bpf_obj_new returns PTR_TO_BTF_ID | MEM_ALLOC | PTR_MAYBE_NULL. | |
d83525ca | 5585 | * Two bpf_map_lookups (even with the same key) will have different reg->id. |
4e814da0 KKD |
5586 | * Two separate bpf_obj_new will also have different reg->id. |
5587 | * For traditional PTR_TO_MAP_VALUE or PTR_TO_BTF_ID | MEM_ALLOC, the verifier | |
5588 | * clears reg->id after value_or_null->value transition, since the verifier only | |
5589 | * cares about the range of access to valid map value pointer and doesn't care | |
5590 | * about actual address of the map element. | |
d83525ca AS |
5591 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps |
5592 | * reg->id > 0 after value_or_null->value transition. By doing so | |
5593 | * two bpf_map_lookups will be considered two different pointers that | |
4e814da0 KKD |
5594 | * point to different bpf_spin_locks. Likewise for pointers to allocated objects |
5595 | * returned from bpf_obj_new. | |
d83525ca AS |
5596 | * The verifier allows taking only one bpf_spin_lock at a time to avoid |
5597 | * dead-locks. | |
5598 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
5599 | * reg_is_refcounted() logic. The verifier needs to remember only | |
5600 | * one spin_lock instead of array of acquired_refs. | |
d0d78c1d | 5601 | * cur_state->active_lock remembers which map value element or allocated |
4e814da0 | 5602 | * object got locked and clears it after bpf_spin_unlock. |
d83525ca AS |
5603 | */ |
5604 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
5605 | bool is_lock) | |
5606 | { | |
5607 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
5608 | struct bpf_verifier_state *cur = env->cur_state; | |
5609 | bool is_const = tnum_is_const(reg->var_off); | |
d83525ca | 5610 | u64 val = reg->var_off.value; |
4e814da0 KKD |
5611 | struct bpf_map *map = NULL; |
5612 | struct btf *btf = NULL; | |
5613 | struct btf_record *rec; | |
d83525ca | 5614 | |
d83525ca AS |
5615 | if (!is_const) { |
5616 | verbose(env, | |
5617 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
5618 | regno); | |
5619 | return -EINVAL; | |
5620 | } | |
4e814da0 KKD |
5621 | if (reg->type == PTR_TO_MAP_VALUE) { |
5622 | map = reg->map_ptr; | |
5623 | if (!map->btf) { | |
5624 | verbose(env, | |
5625 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
5626 | map->name); | |
5627 | return -EINVAL; | |
5628 | } | |
5629 | } else { | |
5630 | btf = reg->btf; | |
d83525ca | 5631 | } |
4e814da0 KKD |
5632 | |
5633 | rec = reg_btf_record(reg); | |
5634 | if (!btf_record_has_field(rec, BPF_SPIN_LOCK)) { | |
5635 | verbose(env, "%s '%s' has no valid bpf_spin_lock\n", map ? "map" : "local", | |
5636 | map ? map->name : "kptr"); | |
d83525ca AS |
5637 | return -EINVAL; |
5638 | } | |
4e814da0 | 5639 | if (rec->spin_lock_off != val + reg->off) { |
db559117 | 5640 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock' that is at %d\n", |
4e814da0 | 5641 | val + reg->off, rec->spin_lock_off); |
d83525ca AS |
5642 | return -EINVAL; |
5643 | } | |
5644 | if (is_lock) { | |
d0d78c1d | 5645 | if (cur->active_lock.ptr) { |
d83525ca AS |
5646 | verbose(env, |
5647 | "Locking two bpf_spin_locks are not allowed\n"); | |
5648 | return -EINVAL; | |
5649 | } | |
d0d78c1d KKD |
5650 | if (map) |
5651 | cur->active_lock.ptr = map; | |
5652 | else | |
5653 | cur->active_lock.ptr = btf; | |
5654 | cur->active_lock.id = reg->id; | |
d83525ca | 5655 | } else { |
534e86bc | 5656 | struct bpf_func_state *fstate = cur_func(env); |
d0d78c1d | 5657 | void *ptr; |
534e86bc | 5658 | int i; |
d0d78c1d KKD |
5659 | |
5660 | if (map) | |
5661 | ptr = map; | |
5662 | else | |
5663 | ptr = btf; | |
5664 | ||
5665 | if (!cur->active_lock.ptr) { | |
d83525ca AS |
5666 | verbose(env, "bpf_spin_unlock without taking a lock\n"); |
5667 | return -EINVAL; | |
5668 | } | |
d0d78c1d KKD |
5669 | if (cur->active_lock.ptr != ptr || |
5670 | cur->active_lock.id != reg->id) { | |
d83525ca AS |
5671 | verbose(env, "bpf_spin_unlock of different lock\n"); |
5672 | return -EINVAL; | |
5673 | } | |
d0d78c1d KKD |
5674 | cur->active_lock.ptr = NULL; |
5675 | cur->active_lock.id = 0; | |
534e86bc KKD |
5676 | |
5677 | for (i = 0; i < fstate->acquired_refs; i++) { | |
5678 | int err; | |
5679 | ||
5680 | /* Complain on error because this reference state cannot | |
5681 | * be freed before this point, as bpf_spin_lock critical | |
5682 | * section does not allow functions that release the | |
5683 | * allocated object immediately. | |
5684 | */ | |
5685 | if (!fstate->refs[i].release_on_unlock) | |
5686 | continue; | |
5687 | err = release_reference(env, fstate->refs[i].id); | |
5688 | if (err) { | |
5689 | verbose(env, "failed to release release_on_unlock reference"); | |
5690 | return err; | |
5691 | } | |
5692 | } | |
d83525ca AS |
5693 | } |
5694 | return 0; | |
5695 | } | |
5696 | ||
b00628b1 AS |
5697 | static int process_timer_func(struct bpf_verifier_env *env, int regno, |
5698 | struct bpf_call_arg_meta *meta) | |
5699 | { | |
5700 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
5701 | bool is_const = tnum_is_const(reg->var_off); | |
5702 | struct bpf_map *map = reg->map_ptr; | |
5703 | u64 val = reg->var_off.value; | |
5704 | ||
5705 | if (!is_const) { | |
5706 | verbose(env, | |
5707 | "R%d doesn't have constant offset. bpf_timer has to be at the constant offset\n", | |
5708 | regno); | |
5709 | return -EINVAL; | |
5710 | } | |
5711 | if (!map->btf) { | |
5712 | verbose(env, "map '%s' has to have BTF in order to use bpf_timer\n", | |
5713 | map->name); | |
5714 | return -EINVAL; | |
5715 | } | |
db559117 KKD |
5716 | if (!btf_record_has_field(map->record, BPF_TIMER)) { |
5717 | verbose(env, "map '%s' has no valid bpf_timer\n", map->name); | |
68134668 AS |
5718 | return -EINVAL; |
5719 | } | |
db559117 | 5720 | if (map->record->timer_off != val + reg->off) { |
68134668 | 5721 | verbose(env, "off %lld doesn't point to 'struct bpf_timer' that is at %d\n", |
db559117 | 5722 | val + reg->off, map->record->timer_off); |
b00628b1 AS |
5723 | return -EINVAL; |
5724 | } | |
5725 | if (meta->map_ptr) { | |
5726 | verbose(env, "verifier bug. Two map pointers in a timer helper\n"); | |
5727 | return -EFAULT; | |
5728 | } | |
3e8ce298 | 5729 | meta->map_uid = reg->map_uid; |
b00628b1 AS |
5730 | meta->map_ptr = map; |
5731 | return 0; | |
5732 | } | |
5733 | ||
c0a5a21c KKD |
5734 | static int process_kptr_func(struct bpf_verifier_env *env, int regno, |
5735 | struct bpf_call_arg_meta *meta) | |
5736 | { | |
5737 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
c0a5a21c | 5738 | struct bpf_map *map_ptr = reg->map_ptr; |
aa3496ac | 5739 | struct btf_field *kptr_field; |
c0a5a21c | 5740 | u32 kptr_off; |
c0a5a21c KKD |
5741 | |
5742 | if (!tnum_is_const(reg->var_off)) { | |
5743 | verbose(env, | |
5744 | "R%d doesn't have constant offset. kptr has to be at the constant offset\n", | |
5745 | regno); | |
5746 | return -EINVAL; | |
5747 | } | |
5748 | if (!map_ptr->btf) { | |
5749 | verbose(env, "map '%s' has to have BTF in order to use bpf_kptr_xchg\n", | |
5750 | map_ptr->name); | |
5751 | return -EINVAL; | |
5752 | } | |
aa3496ac KKD |
5753 | if (!btf_record_has_field(map_ptr->record, BPF_KPTR)) { |
5754 | verbose(env, "map '%s' has no valid kptr\n", map_ptr->name); | |
c0a5a21c KKD |
5755 | return -EINVAL; |
5756 | } | |
5757 | ||
5758 | meta->map_ptr = map_ptr; | |
5759 | kptr_off = reg->off + reg->var_off.value; | |
aa3496ac KKD |
5760 | kptr_field = btf_record_find(map_ptr->record, kptr_off, BPF_KPTR); |
5761 | if (!kptr_field) { | |
c0a5a21c KKD |
5762 | verbose(env, "off=%d doesn't point to kptr\n", kptr_off); |
5763 | return -EACCES; | |
5764 | } | |
aa3496ac | 5765 | if (kptr_field->type != BPF_KPTR_REF) { |
c0a5a21c KKD |
5766 | verbose(env, "off=%d kptr isn't referenced kptr\n", kptr_off); |
5767 | return -EACCES; | |
5768 | } | |
aa3496ac | 5769 | meta->kptr_field = kptr_field; |
c0a5a21c KKD |
5770 | return 0; |
5771 | } | |
5772 | ||
90133415 DB |
5773 | static bool arg_type_is_mem_size(enum bpf_arg_type type) |
5774 | { | |
5775 | return type == ARG_CONST_SIZE || | |
5776 | type == ARG_CONST_SIZE_OR_ZERO; | |
5777 | } | |
5778 | ||
8f14852e KKD |
5779 | static bool arg_type_is_release(enum bpf_arg_type type) |
5780 | { | |
5781 | return type & OBJ_RELEASE; | |
5782 | } | |
5783 | ||
97e03f52 JK |
5784 | static bool arg_type_is_dynptr(enum bpf_arg_type type) |
5785 | { | |
5786 | return base_type(type) == ARG_PTR_TO_DYNPTR; | |
5787 | } | |
5788 | ||
57c3bb72 AI |
5789 | static int int_ptr_type_to_size(enum bpf_arg_type type) |
5790 | { | |
5791 | if (type == ARG_PTR_TO_INT) | |
5792 | return sizeof(u32); | |
5793 | else if (type == ARG_PTR_TO_LONG) | |
5794 | return sizeof(u64); | |
5795 | ||
5796 | return -EINVAL; | |
5797 | } | |
5798 | ||
912f442c LB |
5799 | static int resolve_map_arg_type(struct bpf_verifier_env *env, |
5800 | const struct bpf_call_arg_meta *meta, | |
5801 | enum bpf_arg_type *arg_type) | |
5802 | { | |
5803 | if (!meta->map_ptr) { | |
5804 | /* kernel subsystem misconfigured verifier */ | |
5805 | verbose(env, "invalid map_ptr to access map->type\n"); | |
5806 | return -EACCES; | |
5807 | } | |
5808 | ||
5809 | switch (meta->map_ptr->map_type) { | |
5810 | case BPF_MAP_TYPE_SOCKMAP: | |
5811 | case BPF_MAP_TYPE_SOCKHASH: | |
5812 | if (*arg_type == ARG_PTR_TO_MAP_VALUE) { | |
6550f2dd | 5813 | *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON; |
912f442c LB |
5814 | } else { |
5815 | verbose(env, "invalid arg_type for sockmap/sockhash\n"); | |
5816 | return -EINVAL; | |
5817 | } | |
5818 | break; | |
9330986c JK |
5819 | case BPF_MAP_TYPE_BLOOM_FILTER: |
5820 | if (meta->func_id == BPF_FUNC_map_peek_elem) | |
5821 | *arg_type = ARG_PTR_TO_MAP_VALUE; | |
5822 | break; | |
912f442c LB |
5823 | default: |
5824 | break; | |
5825 | } | |
5826 | return 0; | |
5827 | } | |
5828 | ||
f79e7ea5 LB |
5829 | struct bpf_reg_types { |
5830 | const enum bpf_reg_type types[10]; | |
1df8f55a | 5831 | u32 *btf_id; |
f79e7ea5 LB |
5832 | }; |
5833 | ||
f79e7ea5 LB |
5834 | static const struct bpf_reg_types sock_types = { |
5835 | .types = { | |
5836 | PTR_TO_SOCK_COMMON, | |
5837 | PTR_TO_SOCKET, | |
5838 | PTR_TO_TCP_SOCK, | |
5839 | PTR_TO_XDP_SOCK, | |
5840 | }, | |
5841 | }; | |
5842 | ||
49a2a4d4 | 5843 | #ifdef CONFIG_NET |
1df8f55a MKL |
5844 | static const struct bpf_reg_types btf_id_sock_common_types = { |
5845 | .types = { | |
5846 | PTR_TO_SOCK_COMMON, | |
5847 | PTR_TO_SOCKET, | |
5848 | PTR_TO_TCP_SOCK, | |
5849 | PTR_TO_XDP_SOCK, | |
5850 | PTR_TO_BTF_ID, | |
3f00c523 | 5851 | PTR_TO_BTF_ID | PTR_TRUSTED, |
1df8f55a MKL |
5852 | }, |
5853 | .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
5854 | }; | |
49a2a4d4 | 5855 | #endif |
1df8f55a | 5856 | |
f79e7ea5 LB |
5857 | static const struct bpf_reg_types mem_types = { |
5858 | .types = { | |
5859 | PTR_TO_STACK, | |
5860 | PTR_TO_PACKET, | |
5861 | PTR_TO_PACKET_META, | |
69c087ba | 5862 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
5863 | PTR_TO_MAP_VALUE, |
5864 | PTR_TO_MEM, | |
894f2a8b | 5865 | PTR_TO_MEM | MEM_RINGBUF, |
20b2aff4 | 5866 | PTR_TO_BUF, |
f79e7ea5 LB |
5867 | }, |
5868 | }; | |
5869 | ||
5870 | static const struct bpf_reg_types int_ptr_types = { | |
5871 | .types = { | |
5872 | PTR_TO_STACK, | |
5873 | PTR_TO_PACKET, | |
5874 | PTR_TO_PACKET_META, | |
69c087ba | 5875 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
5876 | PTR_TO_MAP_VALUE, |
5877 | }, | |
5878 | }; | |
5879 | ||
4e814da0 KKD |
5880 | static const struct bpf_reg_types spin_lock_types = { |
5881 | .types = { | |
5882 | PTR_TO_MAP_VALUE, | |
5883 | PTR_TO_BTF_ID | MEM_ALLOC, | |
5884 | } | |
5885 | }; | |
5886 | ||
f79e7ea5 LB |
5887 | static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } }; |
5888 | static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } }; | |
5889 | static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } }; | |
894f2a8b | 5890 | static const struct bpf_reg_types ringbuf_mem_types = { .types = { PTR_TO_MEM | MEM_RINGBUF } }; |
f79e7ea5 | 5891 | static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } }; |
3f00c523 DV |
5892 | static const struct bpf_reg_types btf_ptr_types = { |
5893 | .types = { | |
5894 | PTR_TO_BTF_ID, | |
5895 | PTR_TO_BTF_ID | PTR_TRUSTED, | |
5896 | }, | |
5897 | }; | |
5898 | static const struct bpf_reg_types percpu_btf_ptr_types = { | |
5899 | .types = { | |
5900 | PTR_TO_BTF_ID | MEM_PERCPU, | |
5901 | PTR_TO_BTF_ID | MEM_PERCPU | PTR_TRUSTED, | |
5902 | } | |
5903 | }; | |
69c087ba YS |
5904 | static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } }; |
5905 | static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } }; | |
fff13c4b | 5906 | static const struct bpf_reg_types const_str_ptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
b00628b1 | 5907 | static const struct bpf_reg_types timer_types = { .types = { PTR_TO_MAP_VALUE } }; |
c0a5a21c | 5908 | static const struct bpf_reg_types kptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
20571567 DV |
5909 | static const struct bpf_reg_types dynptr_types = { |
5910 | .types = { | |
5911 | PTR_TO_STACK, | |
5912 | PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL, | |
5913 | } | |
5914 | }; | |
f79e7ea5 | 5915 | |
0789e13b | 5916 | static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = { |
d1673304 DM |
5917 | [ARG_PTR_TO_MAP_KEY] = &mem_types, |
5918 | [ARG_PTR_TO_MAP_VALUE] = &mem_types, | |
f79e7ea5 LB |
5919 | [ARG_CONST_SIZE] = &scalar_types, |
5920 | [ARG_CONST_SIZE_OR_ZERO] = &scalar_types, | |
5921 | [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types, | |
5922 | [ARG_CONST_MAP_PTR] = &const_map_ptr_types, | |
5923 | [ARG_PTR_TO_CTX] = &context_types, | |
f79e7ea5 | 5924 | [ARG_PTR_TO_SOCK_COMMON] = &sock_types, |
49a2a4d4 | 5925 | #ifdef CONFIG_NET |
1df8f55a | 5926 | [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types, |
49a2a4d4 | 5927 | #endif |
f79e7ea5 | 5928 | [ARG_PTR_TO_SOCKET] = &fullsock_types, |
f79e7ea5 LB |
5929 | [ARG_PTR_TO_BTF_ID] = &btf_ptr_types, |
5930 | [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types, | |
5931 | [ARG_PTR_TO_MEM] = &mem_types, | |
894f2a8b | 5932 | [ARG_PTR_TO_RINGBUF_MEM] = &ringbuf_mem_types, |
f79e7ea5 LB |
5933 | [ARG_PTR_TO_INT] = &int_ptr_types, |
5934 | [ARG_PTR_TO_LONG] = &int_ptr_types, | |
eaa6bcb7 | 5935 | [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types, |
69c087ba | 5936 | [ARG_PTR_TO_FUNC] = &func_ptr_types, |
48946bd6 | 5937 | [ARG_PTR_TO_STACK] = &stack_ptr_types, |
fff13c4b | 5938 | [ARG_PTR_TO_CONST_STR] = &const_str_ptr_types, |
b00628b1 | 5939 | [ARG_PTR_TO_TIMER] = &timer_types, |
c0a5a21c | 5940 | [ARG_PTR_TO_KPTR] = &kptr_types, |
20571567 | 5941 | [ARG_PTR_TO_DYNPTR] = &dynptr_types, |
f79e7ea5 LB |
5942 | }; |
5943 | ||
5944 | static int check_reg_type(struct bpf_verifier_env *env, u32 regno, | |
a968d5e2 | 5945 | enum bpf_arg_type arg_type, |
c0a5a21c KKD |
5946 | const u32 *arg_btf_id, |
5947 | struct bpf_call_arg_meta *meta) | |
f79e7ea5 LB |
5948 | { |
5949 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
5950 | enum bpf_reg_type expected, type = reg->type; | |
a968d5e2 | 5951 | const struct bpf_reg_types *compatible; |
f79e7ea5 LB |
5952 | int i, j; |
5953 | ||
48946bd6 | 5954 | compatible = compatible_reg_types[base_type(arg_type)]; |
a968d5e2 MKL |
5955 | if (!compatible) { |
5956 | verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type); | |
5957 | return -EFAULT; | |
5958 | } | |
5959 | ||
216e3cd2 HL |
5960 | /* ARG_PTR_TO_MEM + RDONLY is compatible with PTR_TO_MEM and PTR_TO_MEM + RDONLY, |
5961 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM and NOT with PTR_TO_MEM + RDONLY | |
5962 | * | |
5963 | * Same for MAYBE_NULL: | |
5964 | * | |
5965 | * ARG_PTR_TO_MEM + MAYBE_NULL is compatible with PTR_TO_MEM and PTR_TO_MEM + MAYBE_NULL, | |
5966 | * but ARG_PTR_TO_MEM is compatible only with PTR_TO_MEM but NOT with PTR_TO_MEM + MAYBE_NULL | |
5967 | * | |
5968 | * Therefore we fold these flags depending on the arg_type before comparison. | |
5969 | */ | |
5970 | if (arg_type & MEM_RDONLY) | |
5971 | type &= ~MEM_RDONLY; | |
5972 | if (arg_type & PTR_MAYBE_NULL) | |
5973 | type &= ~PTR_MAYBE_NULL; | |
5974 | ||
f79e7ea5 LB |
5975 | for (i = 0; i < ARRAY_SIZE(compatible->types); i++) { |
5976 | expected = compatible->types[i]; | |
5977 | if (expected == NOT_INIT) | |
5978 | break; | |
5979 | ||
5980 | if (type == expected) | |
a968d5e2 | 5981 | goto found; |
f79e7ea5 LB |
5982 | } |
5983 | ||
216e3cd2 | 5984 | verbose(env, "R%d type=%s expected=", regno, reg_type_str(env, reg->type)); |
f79e7ea5 | 5985 | for (j = 0; j + 1 < i; j++) |
c25b2ae1 HL |
5986 | verbose(env, "%s, ", reg_type_str(env, compatible->types[j])); |
5987 | verbose(env, "%s\n", reg_type_str(env, compatible->types[j])); | |
f79e7ea5 | 5988 | return -EACCES; |
a968d5e2 MKL |
5989 | |
5990 | found: | |
3f00c523 | 5991 | if (reg->type == PTR_TO_BTF_ID || reg->type & PTR_TRUSTED) { |
2ab3b380 KKD |
5992 | /* For bpf_sk_release, it needs to match against first member |
5993 | * 'struct sock_common', hence make an exception for it. This | |
5994 | * allows bpf_sk_release to work for multiple socket types. | |
5995 | */ | |
5996 | bool strict_type_match = arg_type_is_release(arg_type) && | |
5997 | meta->func_id != BPF_FUNC_sk_release; | |
5998 | ||
1df8f55a MKL |
5999 | if (!arg_btf_id) { |
6000 | if (!compatible->btf_id) { | |
6001 | verbose(env, "verifier internal error: missing arg compatible BTF ID\n"); | |
6002 | return -EFAULT; | |
6003 | } | |
6004 | arg_btf_id = compatible->btf_id; | |
6005 | } | |
6006 | ||
c0a5a21c | 6007 | if (meta->func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac | 6008 | if (map_kptr_match_type(env, meta->kptr_field, reg, regno)) |
c0a5a21c | 6009 | return -EACCES; |
47e34cb7 DM |
6010 | } else { |
6011 | if (arg_btf_id == BPF_PTR_POISON) { | |
6012 | verbose(env, "verifier internal error:"); | |
6013 | verbose(env, "R%d has non-overwritten BPF_PTR_POISON type\n", | |
6014 | regno); | |
6015 | return -EACCES; | |
6016 | } | |
6017 | ||
6018 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, | |
6019 | btf_vmlinux, *arg_btf_id, | |
6020 | strict_type_match)) { | |
6021 | verbose(env, "R%d is of type %s but %s is expected\n", | |
6022 | regno, kernel_type_name(reg->btf, reg->btf_id), | |
6023 | kernel_type_name(btf_vmlinux, *arg_btf_id)); | |
6024 | return -EACCES; | |
6025 | } | |
a968d5e2 | 6026 | } |
4e814da0 KKD |
6027 | } else if (type_is_alloc(reg->type)) { |
6028 | if (meta->func_id != BPF_FUNC_spin_lock && meta->func_id != BPF_FUNC_spin_unlock) { | |
6029 | verbose(env, "verifier internal error: unimplemented handling of MEM_ALLOC\n"); | |
6030 | return -EFAULT; | |
6031 | } | |
a968d5e2 MKL |
6032 | } |
6033 | ||
6034 | return 0; | |
f79e7ea5 LB |
6035 | } |
6036 | ||
25b35dd2 KKD |
6037 | int check_func_arg_reg_off(struct bpf_verifier_env *env, |
6038 | const struct bpf_reg_state *reg, int regno, | |
8f14852e | 6039 | enum bpf_arg_type arg_type) |
25b35dd2 KKD |
6040 | { |
6041 | enum bpf_reg_type type = reg->type; | |
8f14852e | 6042 | bool fixed_off_ok = false; |
25b35dd2 KKD |
6043 | |
6044 | switch ((u32)type) { | |
25b35dd2 | 6045 | /* Pointer types where reg offset is explicitly allowed: */ |
97e03f52 JK |
6046 | case PTR_TO_STACK: |
6047 | if (arg_type_is_dynptr(arg_type) && reg->off % BPF_REG_SIZE) { | |
6048 | verbose(env, "cannot pass in dynptr at an offset\n"); | |
6049 | return -EINVAL; | |
6050 | } | |
6051 | fallthrough; | |
25b35dd2 KKD |
6052 | case PTR_TO_PACKET: |
6053 | case PTR_TO_PACKET_META: | |
6054 | case PTR_TO_MAP_KEY: | |
6055 | case PTR_TO_MAP_VALUE: | |
6056 | case PTR_TO_MEM: | |
6057 | case PTR_TO_MEM | MEM_RDONLY: | |
894f2a8b | 6058 | case PTR_TO_MEM | MEM_RINGBUF: |
25b35dd2 KKD |
6059 | case PTR_TO_BUF: |
6060 | case PTR_TO_BUF | MEM_RDONLY: | |
97e03f52 | 6061 | case SCALAR_VALUE: |
25b35dd2 KKD |
6062 | /* Some of the argument types nevertheless require a |
6063 | * zero register offset. | |
6064 | */ | |
894f2a8b | 6065 | if (base_type(arg_type) != ARG_PTR_TO_RINGBUF_MEM) |
25b35dd2 KKD |
6066 | return 0; |
6067 | break; | |
6068 | /* All the rest must be rejected, except PTR_TO_BTF_ID which allows | |
6069 | * fixed offset. | |
6070 | */ | |
6071 | case PTR_TO_BTF_ID: | |
282de143 | 6072 | case PTR_TO_BTF_ID | MEM_ALLOC: |
3f00c523 DV |
6073 | case PTR_TO_BTF_ID | PTR_TRUSTED: |
6074 | case PTR_TO_BTF_ID | MEM_ALLOC | PTR_TRUSTED: | |
24d5bb80 | 6075 | /* When referenced PTR_TO_BTF_ID is passed to release function, |
8f14852e KKD |
6076 | * it's fixed offset must be 0. In the other cases, fixed offset |
6077 | * can be non-zero. | |
24d5bb80 | 6078 | */ |
8f14852e | 6079 | if (arg_type_is_release(arg_type) && reg->off) { |
24d5bb80 KKD |
6080 | verbose(env, "R%d must have zero offset when passed to release func\n", |
6081 | regno); | |
6082 | return -EINVAL; | |
6083 | } | |
8f14852e KKD |
6084 | /* For arg is release pointer, fixed_off_ok must be false, but |
6085 | * we already checked and rejected reg->off != 0 above, so set | |
6086 | * to true to allow fixed offset for all other cases. | |
24d5bb80 | 6087 | */ |
25b35dd2 KKD |
6088 | fixed_off_ok = true; |
6089 | break; | |
6090 | default: | |
6091 | break; | |
6092 | } | |
6093 | return __check_ptr_off_reg(env, reg, regno, fixed_off_ok); | |
6094 | } | |
6095 | ||
34d4ef57 JK |
6096 | static u32 stack_slot_get_id(struct bpf_verifier_env *env, struct bpf_reg_state *reg) |
6097 | { | |
6098 | struct bpf_func_state *state = func(env, reg); | |
6099 | int spi = get_spi(reg->off); | |
6100 | ||
6101 | return state->stack[spi].spilled_ptr.id; | |
6102 | } | |
6103 | ||
af7ec138 YS |
6104 | static int check_func_arg(struct bpf_verifier_env *env, u32 arg, |
6105 | struct bpf_call_arg_meta *meta, | |
6106 | const struct bpf_func_proto *fn) | |
17a52670 | 6107 | { |
af7ec138 | 6108 | u32 regno = BPF_REG_1 + arg; |
638f5b90 | 6109 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
af7ec138 | 6110 | enum bpf_arg_type arg_type = fn->arg_type[arg]; |
f79e7ea5 | 6111 | enum bpf_reg_type type = reg->type; |
508362ac | 6112 | u32 *arg_btf_id = NULL; |
17a52670 AS |
6113 | int err = 0; |
6114 | ||
80f1d68c | 6115 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
6116 | return 0; |
6117 | ||
dc503a8a EC |
6118 | err = check_reg_arg(env, regno, SRC_OP); |
6119 | if (err) | |
6120 | return err; | |
17a52670 | 6121 | |
1be7f75d AS |
6122 | if (arg_type == ARG_ANYTHING) { |
6123 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
6124 | verbose(env, "R%d leaks addr into helper function\n", |
6125 | regno); | |
1be7f75d AS |
6126 | return -EACCES; |
6127 | } | |
80f1d68c | 6128 | return 0; |
1be7f75d | 6129 | } |
80f1d68c | 6130 | |
de8f3a83 | 6131 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 6132 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 6133 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
6134 | return -EACCES; |
6135 | } | |
6136 | ||
16d1e00c | 6137 | if (base_type(arg_type) == ARG_PTR_TO_MAP_VALUE) { |
912f442c LB |
6138 | err = resolve_map_arg_type(env, meta, &arg_type); |
6139 | if (err) | |
6140 | return err; | |
6141 | } | |
6142 | ||
48946bd6 | 6143 | if (register_is_null(reg) && type_may_be_null(arg_type)) |
fd1b0d60 LB |
6144 | /* A NULL register has a SCALAR_VALUE type, so skip |
6145 | * type checking. | |
6146 | */ | |
6147 | goto skip_type_check; | |
6148 | ||
508362ac | 6149 | /* arg_btf_id and arg_size are in a union. */ |
4e814da0 KKD |
6150 | if (base_type(arg_type) == ARG_PTR_TO_BTF_ID || |
6151 | base_type(arg_type) == ARG_PTR_TO_SPIN_LOCK) | |
508362ac MM |
6152 | arg_btf_id = fn->arg_btf_id[arg]; |
6153 | ||
6154 | err = check_reg_type(env, regno, arg_type, arg_btf_id, meta); | |
f79e7ea5 LB |
6155 | if (err) |
6156 | return err; | |
6157 | ||
8f14852e | 6158 | err = check_func_arg_reg_off(env, reg, regno, arg_type); |
25b35dd2 KKD |
6159 | if (err) |
6160 | return err; | |
d7b9454a | 6161 | |
fd1b0d60 | 6162 | skip_type_check: |
8f14852e | 6163 | if (arg_type_is_release(arg_type)) { |
bc34dee6 JK |
6164 | if (arg_type_is_dynptr(arg_type)) { |
6165 | struct bpf_func_state *state = func(env, reg); | |
6166 | int spi = get_spi(reg->off); | |
6167 | ||
6168 | if (!is_spi_bounds_valid(state, spi, BPF_DYNPTR_NR_SLOTS) || | |
6169 | !state->stack[spi].spilled_ptr.id) { | |
6170 | verbose(env, "arg %d is an unacquired reference\n", regno); | |
6171 | return -EINVAL; | |
6172 | } | |
6173 | } else if (!reg->ref_obj_id && !register_is_null(reg)) { | |
8f14852e KKD |
6174 | verbose(env, "R%d must be referenced when passed to release function\n", |
6175 | regno); | |
6176 | return -EINVAL; | |
6177 | } | |
6178 | if (meta->release_regno) { | |
6179 | verbose(env, "verifier internal error: more than one release argument\n"); | |
6180 | return -EFAULT; | |
6181 | } | |
6182 | meta->release_regno = regno; | |
6183 | } | |
6184 | ||
02f7c958 | 6185 | if (reg->ref_obj_id) { |
457f4436 AN |
6186 | if (meta->ref_obj_id) { |
6187 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
6188 | regno, reg->ref_obj_id, | |
6189 | meta->ref_obj_id); | |
6190 | return -EFAULT; | |
6191 | } | |
6192 | meta->ref_obj_id = reg->ref_obj_id; | |
17a52670 AS |
6193 | } |
6194 | ||
8ab4cdcf JK |
6195 | switch (base_type(arg_type)) { |
6196 | case ARG_CONST_MAP_PTR: | |
17a52670 | 6197 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ |
3e8ce298 AS |
6198 | if (meta->map_ptr) { |
6199 | /* Use map_uid (which is unique id of inner map) to reject: | |
6200 | * inner_map1 = bpf_map_lookup_elem(outer_map, key1) | |
6201 | * inner_map2 = bpf_map_lookup_elem(outer_map, key2) | |
6202 | * if (inner_map1 && inner_map2) { | |
6203 | * timer = bpf_map_lookup_elem(inner_map1); | |
6204 | * if (timer) | |
6205 | * // mismatch would have been allowed | |
6206 | * bpf_timer_init(timer, inner_map2); | |
6207 | * } | |
6208 | * | |
6209 | * Comparing map_ptr is enough to distinguish normal and outer maps. | |
6210 | */ | |
6211 | if (meta->map_ptr != reg->map_ptr || | |
6212 | meta->map_uid != reg->map_uid) { | |
6213 | verbose(env, | |
6214 | "timer pointer in R1 map_uid=%d doesn't match map pointer in R2 map_uid=%d\n", | |
6215 | meta->map_uid, reg->map_uid); | |
6216 | return -EINVAL; | |
6217 | } | |
b00628b1 | 6218 | } |
33ff9823 | 6219 | meta->map_ptr = reg->map_ptr; |
3e8ce298 | 6220 | meta->map_uid = reg->map_uid; |
8ab4cdcf JK |
6221 | break; |
6222 | case ARG_PTR_TO_MAP_KEY: | |
17a52670 AS |
6223 | /* bpf_map_xxx(..., map_ptr, ..., key) call: |
6224 | * check that [key, key + map->key_size) are within | |
6225 | * stack limits and initialized | |
6226 | */ | |
33ff9823 | 6227 | if (!meta->map_ptr) { |
17a52670 AS |
6228 | /* in function declaration map_ptr must come before |
6229 | * map_key, so that it's verified and known before | |
6230 | * we have to check map_key here. Otherwise it means | |
6231 | * that kernel subsystem misconfigured verifier | |
6232 | */ | |
61bd5218 | 6233 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
6234 | return -EACCES; |
6235 | } | |
d71962f3 PC |
6236 | err = check_helper_mem_access(env, regno, |
6237 | meta->map_ptr->key_size, false, | |
6238 | NULL); | |
8ab4cdcf JK |
6239 | break; |
6240 | case ARG_PTR_TO_MAP_VALUE: | |
48946bd6 HL |
6241 | if (type_may_be_null(arg_type) && register_is_null(reg)) |
6242 | return 0; | |
6243 | ||
17a52670 AS |
6244 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
6245 | * check [value, value + map->value_size) validity | |
6246 | */ | |
33ff9823 | 6247 | if (!meta->map_ptr) { |
17a52670 | 6248 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 6249 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
6250 | return -EACCES; |
6251 | } | |
16d1e00c | 6252 | meta->raw_mode = arg_type & MEM_UNINIT; |
d71962f3 PC |
6253 | err = check_helper_mem_access(env, regno, |
6254 | meta->map_ptr->value_size, false, | |
2ea864c5 | 6255 | meta); |
8ab4cdcf JK |
6256 | break; |
6257 | case ARG_PTR_TO_PERCPU_BTF_ID: | |
eaa6bcb7 HL |
6258 | if (!reg->btf_id) { |
6259 | verbose(env, "Helper has invalid btf_id in R%d\n", regno); | |
6260 | return -EACCES; | |
6261 | } | |
22dc4a0f | 6262 | meta->ret_btf = reg->btf; |
eaa6bcb7 | 6263 | meta->ret_btf_id = reg->btf_id; |
8ab4cdcf JK |
6264 | break; |
6265 | case ARG_PTR_TO_SPIN_LOCK: | |
c18f0b6a LB |
6266 | if (meta->func_id == BPF_FUNC_spin_lock) { |
6267 | if (process_spin_lock(env, regno, true)) | |
6268 | return -EACCES; | |
6269 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { | |
6270 | if (process_spin_lock(env, regno, false)) | |
6271 | return -EACCES; | |
6272 | } else { | |
6273 | verbose(env, "verifier internal error\n"); | |
6274 | return -EFAULT; | |
6275 | } | |
8ab4cdcf JK |
6276 | break; |
6277 | case ARG_PTR_TO_TIMER: | |
b00628b1 AS |
6278 | if (process_timer_func(env, regno, meta)) |
6279 | return -EACCES; | |
8ab4cdcf JK |
6280 | break; |
6281 | case ARG_PTR_TO_FUNC: | |
69c087ba | 6282 | meta->subprogno = reg->subprogno; |
8ab4cdcf JK |
6283 | break; |
6284 | case ARG_PTR_TO_MEM: | |
a2bbe7cc LB |
6285 | /* The access to this pointer is only checked when we hit the |
6286 | * next is_mem_size argument below. | |
6287 | */ | |
16d1e00c | 6288 | meta->raw_mode = arg_type & MEM_UNINIT; |
508362ac MM |
6289 | if (arg_type & MEM_FIXED_SIZE) { |
6290 | err = check_helper_mem_access(env, regno, | |
6291 | fn->arg_size[arg], false, | |
6292 | meta); | |
6293 | } | |
8ab4cdcf JK |
6294 | break; |
6295 | case ARG_CONST_SIZE: | |
6296 | err = check_mem_size_reg(env, reg, regno, false, meta); | |
6297 | break; | |
6298 | case ARG_CONST_SIZE_OR_ZERO: | |
6299 | err = check_mem_size_reg(env, reg, regno, true, meta); | |
6300 | break; | |
6301 | case ARG_PTR_TO_DYNPTR: | |
20571567 DV |
6302 | /* We only need to check for initialized / uninitialized helper |
6303 | * dynptr args if the dynptr is not PTR_TO_DYNPTR, as the | |
6304 | * assumption is that if it is, that a helper function | |
6305 | * initialized the dynptr on behalf of the BPF program. | |
6306 | */ | |
6307 | if (base_type(reg->type) == PTR_TO_DYNPTR) | |
6308 | break; | |
97e03f52 JK |
6309 | if (arg_type & MEM_UNINIT) { |
6310 | if (!is_dynptr_reg_valid_uninit(env, reg)) { | |
6311 | verbose(env, "Dynptr has to be an uninitialized dynptr\n"); | |
6312 | return -EINVAL; | |
6313 | } | |
6314 | ||
6315 | /* We only support one dynptr being uninitialized at the moment, | |
6316 | * which is sufficient for the helper functions we have right now. | |
6317 | */ | |
6318 | if (meta->uninit_dynptr_regno) { | |
6319 | verbose(env, "verifier internal error: multiple uninitialized dynptr args\n"); | |
6320 | return -EFAULT; | |
6321 | } | |
6322 | ||
6323 | meta->uninit_dynptr_regno = regno; | |
e9e315b4 RS |
6324 | } else if (!is_dynptr_reg_valid_init(env, reg)) { |
6325 | verbose(env, | |
6326 | "Expected an initialized dynptr as arg #%d\n", | |
6327 | arg + 1); | |
6328 | return -EINVAL; | |
6329 | } else if (!is_dynptr_type_expected(env, reg, arg_type)) { | |
97e03f52 JK |
6330 | const char *err_extra = ""; |
6331 | ||
6332 | switch (arg_type & DYNPTR_TYPE_FLAG_MASK) { | |
6333 | case DYNPTR_TYPE_LOCAL: | |
e9e315b4 | 6334 | err_extra = "local"; |
97e03f52 | 6335 | break; |
bc34dee6 | 6336 | case DYNPTR_TYPE_RINGBUF: |
e9e315b4 | 6337 | err_extra = "ringbuf"; |
bc34dee6 | 6338 | break; |
97e03f52 | 6339 | default: |
e9e315b4 | 6340 | err_extra = "<unknown>"; |
97e03f52 JK |
6341 | break; |
6342 | } | |
e9e315b4 RS |
6343 | verbose(env, |
6344 | "Expected a dynptr of type %s as arg #%d\n", | |
97e03f52 JK |
6345 | err_extra, arg + 1); |
6346 | return -EINVAL; | |
6347 | } | |
8ab4cdcf JK |
6348 | break; |
6349 | case ARG_CONST_ALLOC_SIZE_OR_ZERO: | |
457f4436 | 6350 | if (!tnum_is_const(reg->var_off)) { |
28a8add6 | 6351 | verbose(env, "R%d is not a known constant'\n", |
457f4436 AN |
6352 | regno); |
6353 | return -EACCES; | |
6354 | } | |
6355 | meta->mem_size = reg->var_off.value; | |
2fc31465 KKD |
6356 | err = mark_chain_precision(env, regno); |
6357 | if (err) | |
6358 | return err; | |
8ab4cdcf JK |
6359 | break; |
6360 | case ARG_PTR_TO_INT: | |
6361 | case ARG_PTR_TO_LONG: | |
6362 | { | |
57c3bb72 AI |
6363 | int size = int_ptr_type_to_size(arg_type); |
6364 | ||
6365 | err = check_helper_mem_access(env, regno, size, false, meta); | |
6366 | if (err) | |
6367 | return err; | |
6368 | err = check_ptr_alignment(env, reg, 0, size, true); | |
8ab4cdcf JK |
6369 | break; |
6370 | } | |
6371 | case ARG_PTR_TO_CONST_STR: | |
6372 | { | |
fff13c4b FR |
6373 | struct bpf_map *map = reg->map_ptr; |
6374 | int map_off; | |
6375 | u64 map_addr; | |
6376 | char *str_ptr; | |
6377 | ||
a8fad73e | 6378 | if (!bpf_map_is_rdonly(map)) { |
fff13c4b FR |
6379 | verbose(env, "R%d does not point to a readonly map'\n", regno); |
6380 | return -EACCES; | |
6381 | } | |
6382 | ||
6383 | if (!tnum_is_const(reg->var_off)) { | |
6384 | verbose(env, "R%d is not a constant address'\n", regno); | |
6385 | return -EACCES; | |
6386 | } | |
6387 | ||
6388 | if (!map->ops->map_direct_value_addr) { | |
6389 | verbose(env, "no direct value access support for this map type\n"); | |
6390 | return -EACCES; | |
6391 | } | |
6392 | ||
6393 | err = check_map_access(env, regno, reg->off, | |
61df10c7 KKD |
6394 | map->value_size - reg->off, false, |
6395 | ACCESS_HELPER); | |
fff13c4b FR |
6396 | if (err) |
6397 | return err; | |
6398 | ||
6399 | map_off = reg->off + reg->var_off.value; | |
6400 | err = map->ops->map_direct_value_addr(map, &map_addr, map_off); | |
6401 | if (err) { | |
6402 | verbose(env, "direct value access on string failed\n"); | |
6403 | return err; | |
6404 | } | |
6405 | ||
6406 | str_ptr = (char *)(long)(map_addr); | |
6407 | if (!strnchr(str_ptr + map_off, map->value_size - map_off, 0)) { | |
6408 | verbose(env, "string is not zero-terminated\n"); | |
6409 | return -EINVAL; | |
6410 | } | |
8ab4cdcf JK |
6411 | break; |
6412 | } | |
6413 | case ARG_PTR_TO_KPTR: | |
c0a5a21c KKD |
6414 | if (process_kptr_func(env, regno, meta)) |
6415 | return -EACCES; | |
8ab4cdcf | 6416 | break; |
17a52670 AS |
6417 | } |
6418 | ||
6419 | return err; | |
6420 | } | |
6421 | ||
0126240f LB |
6422 | static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) |
6423 | { | |
6424 | enum bpf_attach_type eatype = env->prog->expected_attach_type; | |
7e40781c | 6425 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
0126240f LB |
6426 | |
6427 | if (func_id != BPF_FUNC_map_update_elem) | |
6428 | return false; | |
6429 | ||
6430 | /* It's not possible to get access to a locked struct sock in these | |
6431 | * contexts, so updating is safe. | |
6432 | */ | |
6433 | switch (type) { | |
6434 | case BPF_PROG_TYPE_TRACING: | |
6435 | if (eatype == BPF_TRACE_ITER) | |
6436 | return true; | |
6437 | break; | |
6438 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
6439 | case BPF_PROG_TYPE_SCHED_CLS: | |
6440 | case BPF_PROG_TYPE_SCHED_ACT: | |
6441 | case BPF_PROG_TYPE_XDP: | |
6442 | case BPF_PROG_TYPE_SK_REUSEPORT: | |
6443 | case BPF_PROG_TYPE_FLOW_DISSECTOR: | |
6444 | case BPF_PROG_TYPE_SK_LOOKUP: | |
6445 | return true; | |
6446 | default: | |
6447 | break; | |
6448 | } | |
6449 | ||
6450 | verbose(env, "cannot update sockmap in this context\n"); | |
6451 | return false; | |
6452 | } | |
6453 | ||
e411901c MF |
6454 | static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env) |
6455 | { | |
95acd881 TA |
6456 | return env->prog->jit_requested && |
6457 | bpf_jit_supports_subprog_tailcalls(); | |
e411901c MF |
6458 | } |
6459 | ||
61bd5218 JK |
6460 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
6461 | struct bpf_map *map, int func_id) | |
35578d79 | 6462 | { |
35578d79 KX |
6463 | if (!map) |
6464 | return 0; | |
6465 | ||
6aff67c8 AS |
6466 | /* We need a two way check, first is from map perspective ... */ |
6467 | switch (map->map_type) { | |
6468 | case BPF_MAP_TYPE_PROG_ARRAY: | |
6469 | if (func_id != BPF_FUNC_tail_call) | |
6470 | goto error; | |
6471 | break; | |
6472 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
6473 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 6474 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 6475 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
6476 | func_id != BPF_FUNC_perf_event_read_value && |
6477 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
6478 | goto error; |
6479 | break; | |
457f4436 AN |
6480 | case BPF_MAP_TYPE_RINGBUF: |
6481 | if (func_id != BPF_FUNC_ringbuf_output && | |
6482 | func_id != BPF_FUNC_ringbuf_reserve && | |
bc34dee6 JK |
6483 | func_id != BPF_FUNC_ringbuf_query && |
6484 | func_id != BPF_FUNC_ringbuf_reserve_dynptr && | |
6485 | func_id != BPF_FUNC_ringbuf_submit_dynptr && | |
6486 | func_id != BPF_FUNC_ringbuf_discard_dynptr) | |
457f4436 AN |
6487 | goto error; |
6488 | break; | |
583c1f42 | 6489 | case BPF_MAP_TYPE_USER_RINGBUF: |
20571567 DV |
6490 | if (func_id != BPF_FUNC_user_ringbuf_drain) |
6491 | goto error; | |
6492 | break; | |
6aff67c8 AS |
6493 | case BPF_MAP_TYPE_STACK_TRACE: |
6494 | if (func_id != BPF_FUNC_get_stackid) | |
6495 | goto error; | |
6496 | break; | |
4ed8ec52 | 6497 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 6498 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 6499 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
6500 | goto error; |
6501 | break; | |
cd339431 | 6502 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 6503 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
6504 | if (func_id != BPF_FUNC_get_local_storage) |
6505 | goto error; | |
6506 | break; | |
546ac1ff | 6507 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 6508 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
6509 | if (func_id != BPF_FUNC_redirect_map && |
6510 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
6511 | goto error; |
6512 | break; | |
fbfc504a BT |
6513 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
6514 | * appear. | |
6515 | */ | |
6710e112 JDB |
6516 | case BPF_MAP_TYPE_CPUMAP: |
6517 | if (func_id != BPF_FUNC_redirect_map) | |
6518 | goto error; | |
6519 | break; | |
fada7fdc JL |
6520 | case BPF_MAP_TYPE_XSKMAP: |
6521 | if (func_id != BPF_FUNC_redirect_map && | |
6522 | func_id != BPF_FUNC_map_lookup_elem) | |
6523 | goto error; | |
6524 | break; | |
56f668df | 6525 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 6526 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
6527 | if (func_id != BPF_FUNC_map_lookup_elem) |
6528 | goto error; | |
16a43625 | 6529 | break; |
174a79ff JF |
6530 | case BPF_MAP_TYPE_SOCKMAP: |
6531 | if (func_id != BPF_FUNC_sk_redirect_map && | |
6532 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 6533 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 6534 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 | 6535 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
6536 | func_id != BPF_FUNC_map_lookup_elem && |
6537 | !may_update_sockmap(env, func_id)) | |
174a79ff JF |
6538 | goto error; |
6539 | break; | |
81110384 JF |
6540 | case BPF_MAP_TYPE_SOCKHASH: |
6541 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
6542 | func_id != BPF_FUNC_sock_hash_update && | |
6543 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 6544 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 | 6545 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
6546 | func_id != BPF_FUNC_map_lookup_elem && |
6547 | !may_update_sockmap(env, func_id)) | |
81110384 JF |
6548 | goto error; |
6549 | break; | |
2dbb9b9e MKL |
6550 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
6551 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
6552 | goto error; | |
6553 | break; | |
f1a2e44a MV |
6554 | case BPF_MAP_TYPE_QUEUE: |
6555 | case BPF_MAP_TYPE_STACK: | |
6556 | if (func_id != BPF_FUNC_map_peek_elem && | |
6557 | func_id != BPF_FUNC_map_pop_elem && | |
6558 | func_id != BPF_FUNC_map_push_elem) | |
6559 | goto error; | |
6560 | break; | |
6ac99e8f MKL |
6561 | case BPF_MAP_TYPE_SK_STORAGE: |
6562 | if (func_id != BPF_FUNC_sk_storage_get && | |
6563 | func_id != BPF_FUNC_sk_storage_delete) | |
6564 | goto error; | |
6565 | break; | |
8ea63684 KS |
6566 | case BPF_MAP_TYPE_INODE_STORAGE: |
6567 | if (func_id != BPF_FUNC_inode_storage_get && | |
6568 | func_id != BPF_FUNC_inode_storage_delete) | |
6569 | goto error; | |
6570 | break; | |
4cf1bc1f KS |
6571 | case BPF_MAP_TYPE_TASK_STORAGE: |
6572 | if (func_id != BPF_FUNC_task_storage_get && | |
6573 | func_id != BPF_FUNC_task_storage_delete) | |
6574 | goto error; | |
6575 | break; | |
c4bcfb38 YS |
6576 | case BPF_MAP_TYPE_CGRP_STORAGE: |
6577 | if (func_id != BPF_FUNC_cgrp_storage_get && | |
6578 | func_id != BPF_FUNC_cgrp_storage_delete) | |
6579 | goto error; | |
6580 | break; | |
9330986c JK |
6581 | case BPF_MAP_TYPE_BLOOM_FILTER: |
6582 | if (func_id != BPF_FUNC_map_peek_elem && | |
6583 | func_id != BPF_FUNC_map_push_elem) | |
6584 | goto error; | |
6585 | break; | |
6aff67c8 AS |
6586 | default: |
6587 | break; | |
6588 | } | |
6589 | ||
6590 | /* ... and second from the function itself. */ | |
6591 | switch (func_id) { | |
6592 | case BPF_FUNC_tail_call: | |
6593 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
6594 | goto error; | |
e411901c MF |
6595 | if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) { |
6596 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
f4d7e40a AS |
6597 | return -EINVAL; |
6598 | } | |
6aff67c8 AS |
6599 | break; |
6600 | case BPF_FUNC_perf_event_read: | |
6601 | case BPF_FUNC_perf_event_output: | |
908432ca | 6602 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 6603 | case BPF_FUNC_skb_output: |
d831ee84 | 6604 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
6605 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
6606 | goto error; | |
6607 | break; | |
5b029a32 DB |
6608 | case BPF_FUNC_ringbuf_output: |
6609 | case BPF_FUNC_ringbuf_reserve: | |
6610 | case BPF_FUNC_ringbuf_query: | |
bc34dee6 JK |
6611 | case BPF_FUNC_ringbuf_reserve_dynptr: |
6612 | case BPF_FUNC_ringbuf_submit_dynptr: | |
6613 | case BPF_FUNC_ringbuf_discard_dynptr: | |
5b029a32 DB |
6614 | if (map->map_type != BPF_MAP_TYPE_RINGBUF) |
6615 | goto error; | |
6616 | break; | |
20571567 DV |
6617 | case BPF_FUNC_user_ringbuf_drain: |
6618 | if (map->map_type != BPF_MAP_TYPE_USER_RINGBUF) | |
6619 | goto error; | |
6620 | break; | |
6aff67c8 AS |
6621 | case BPF_FUNC_get_stackid: |
6622 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
6623 | goto error; | |
6624 | break; | |
60d20f91 | 6625 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 6626 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
6627 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
6628 | goto error; | |
6629 | break; | |
97f91a7c | 6630 | case BPF_FUNC_redirect_map: |
9c270af3 | 6631 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 6632 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
6633 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
6634 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
6635 | goto error; |
6636 | break; | |
174a79ff | 6637 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 6638 | case BPF_FUNC_msg_redirect_map: |
81110384 | 6639 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
6640 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
6641 | goto error; | |
6642 | break; | |
81110384 JF |
6643 | case BPF_FUNC_sk_redirect_hash: |
6644 | case BPF_FUNC_msg_redirect_hash: | |
6645 | case BPF_FUNC_sock_hash_update: | |
6646 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
6647 | goto error; |
6648 | break; | |
cd339431 | 6649 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
6650 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
6651 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
6652 | goto error; |
6653 | break; | |
2dbb9b9e | 6654 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
6655 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
6656 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
6657 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
6658 | goto error; |
6659 | break; | |
f1a2e44a | 6660 | case BPF_FUNC_map_pop_elem: |
f1a2e44a MV |
6661 | if (map->map_type != BPF_MAP_TYPE_QUEUE && |
6662 | map->map_type != BPF_MAP_TYPE_STACK) | |
6663 | goto error; | |
6664 | break; | |
9330986c JK |
6665 | case BPF_FUNC_map_peek_elem: |
6666 | case BPF_FUNC_map_push_elem: | |
6667 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
6668 | map->map_type != BPF_MAP_TYPE_STACK && | |
6669 | map->map_type != BPF_MAP_TYPE_BLOOM_FILTER) | |
6670 | goto error; | |
6671 | break; | |
07343110 FZ |
6672 | case BPF_FUNC_map_lookup_percpu_elem: |
6673 | if (map->map_type != BPF_MAP_TYPE_PERCPU_ARRAY && | |
6674 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && | |
6675 | map->map_type != BPF_MAP_TYPE_LRU_PERCPU_HASH) | |
6676 | goto error; | |
6677 | break; | |
6ac99e8f MKL |
6678 | case BPF_FUNC_sk_storage_get: |
6679 | case BPF_FUNC_sk_storage_delete: | |
6680 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
6681 | goto error; | |
6682 | break; | |
8ea63684 KS |
6683 | case BPF_FUNC_inode_storage_get: |
6684 | case BPF_FUNC_inode_storage_delete: | |
6685 | if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE) | |
6686 | goto error; | |
6687 | break; | |
4cf1bc1f KS |
6688 | case BPF_FUNC_task_storage_get: |
6689 | case BPF_FUNC_task_storage_delete: | |
6690 | if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) | |
6691 | goto error; | |
6692 | break; | |
c4bcfb38 YS |
6693 | case BPF_FUNC_cgrp_storage_get: |
6694 | case BPF_FUNC_cgrp_storage_delete: | |
6695 | if (map->map_type != BPF_MAP_TYPE_CGRP_STORAGE) | |
6696 | goto error; | |
6697 | break; | |
6aff67c8 AS |
6698 | default: |
6699 | break; | |
35578d79 KX |
6700 | } |
6701 | ||
6702 | return 0; | |
6aff67c8 | 6703 | error: |
61bd5218 | 6704 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 6705 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 6706 | return -EINVAL; |
35578d79 KX |
6707 | } |
6708 | ||
90133415 | 6709 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
6710 | { |
6711 | int count = 0; | |
6712 | ||
39f19ebb | 6713 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 6714 | count++; |
39f19ebb | 6715 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 6716 | count++; |
39f19ebb | 6717 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 6718 | count++; |
39f19ebb | 6719 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 6720 | count++; |
39f19ebb | 6721 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
6722 | count++; |
6723 | ||
90133415 DB |
6724 | /* We only support one arg being in raw mode at the moment, |
6725 | * which is sufficient for the helper functions we have | |
6726 | * right now. | |
6727 | */ | |
6728 | return count <= 1; | |
6729 | } | |
6730 | ||
508362ac | 6731 | static bool check_args_pair_invalid(const struct bpf_func_proto *fn, int arg) |
90133415 | 6732 | { |
508362ac MM |
6733 | bool is_fixed = fn->arg_type[arg] & MEM_FIXED_SIZE; |
6734 | bool has_size = fn->arg_size[arg] != 0; | |
6735 | bool is_next_size = false; | |
6736 | ||
6737 | if (arg + 1 < ARRAY_SIZE(fn->arg_type)) | |
6738 | is_next_size = arg_type_is_mem_size(fn->arg_type[arg + 1]); | |
6739 | ||
6740 | if (base_type(fn->arg_type[arg]) != ARG_PTR_TO_MEM) | |
6741 | return is_next_size; | |
6742 | ||
6743 | return has_size == is_next_size || is_next_size == is_fixed; | |
90133415 DB |
6744 | } |
6745 | ||
6746 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
6747 | { | |
6748 | /* bpf_xxx(..., buf, len) call will access 'len' | |
6749 | * bytes from memory 'buf'. Both arg types need | |
6750 | * to be paired, so make sure there's no buggy | |
6751 | * helper function specification. | |
6752 | */ | |
6753 | if (arg_type_is_mem_size(fn->arg1_type) || | |
508362ac MM |
6754 | check_args_pair_invalid(fn, 0) || |
6755 | check_args_pair_invalid(fn, 1) || | |
6756 | check_args_pair_invalid(fn, 2) || | |
6757 | check_args_pair_invalid(fn, 3) || | |
6758 | check_args_pair_invalid(fn, 4)) | |
90133415 DB |
6759 | return false; |
6760 | ||
6761 | return true; | |
6762 | } | |
6763 | ||
9436ef6e LB |
6764 | static bool check_btf_id_ok(const struct bpf_func_proto *fn) |
6765 | { | |
6766 | int i; | |
6767 | ||
1df8f55a | 6768 | for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) { |
4e814da0 KKD |
6769 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_BTF_ID) |
6770 | return !!fn->arg_btf_id[i]; | |
6771 | if (base_type(fn->arg_type[i]) == ARG_PTR_TO_SPIN_LOCK) | |
6772 | return fn->arg_btf_id[i] == BPF_PTR_POISON; | |
508362ac MM |
6773 | if (base_type(fn->arg_type[i]) != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i] && |
6774 | /* arg_btf_id and arg_size are in a union. */ | |
6775 | (base_type(fn->arg_type[i]) != ARG_PTR_TO_MEM || | |
6776 | !(fn->arg_type[i] & MEM_FIXED_SIZE))) | |
1df8f55a MKL |
6777 | return false; |
6778 | } | |
6779 | ||
9436ef6e LB |
6780 | return true; |
6781 | } | |
6782 | ||
0c9a7a7e | 6783 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
6784 | { |
6785 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 6786 | check_arg_pair_ok(fn) && |
b2d8ef19 | 6787 | check_btf_id_ok(fn) ? 0 : -EINVAL; |
435faee1 DB |
6788 | } |
6789 | ||
de8f3a83 DB |
6790 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
6791 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 6792 | */ |
b239da34 | 6793 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 6794 | { |
b239da34 KKD |
6795 | struct bpf_func_state *state; |
6796 | struct bpf_reg_state *reg; | |
969bf05e | 6797 | |
b239da34 | 6798 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
de8f3a83 | 6799 | if (reg_is_pkt_pointer_any(reg)) |
f54c7898 | 6800 | __mark_reg_unknown(env, reg); |
b239da34 | 6801 | })); |
f4d7e40a AS |
6802 | } |
6803 | ||
6d94e741 AS |
6804 | enum { |
6805 | AT_PKT_END = -1, | |
6806 | BEYOND_PKT_END = -2, | |
6807 | }; | |
6808 | ||
6809 | static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open) | |
6810 | { | |
6811 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
6812 | struct bpf_reg_state *reg = &state->regs[regn]; | |
6813 | ||
6814 | if (reg->type != PTR_TO_PACKET) | |
6815 | /* PTR_TO_PACKET_META is not supported yet */ | |
6816 | return; | |
6817 | ||
6818 | /* The 'reg' is pkt > pkt_end or pkt >= pkt_end. | |
6819 | * How far beyond pkt_end it goes is unknown. | |
6820 | * if (!range_open) it's the case of pkt >= pkt_end | |
6821 | * if (range_open) it's the case of pkt > pkt_end | |
6822 | * hence this pointer is at least 1 byte bigger than pkt_end | |
6823 | */ | |
6824 | if (range_open) | |
6825 | reg->range = BEYOND_PKT_END; | |
6826 | else | |
6827 | reg->range = AT_PKT_END; | |
6828 | } | |
6829 | ||
fd978bf7 JS |
6830 | /* The pointer with the specified id has released its reference to kernel |
6831 | * resources. Identify all copies of the same pointer and clear the reference. | |
6832 | */ | |
6833 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 6834 | int ref_obj_id) |
fd978bf7 | 6835 | { |
b239da34 KKD |
6836 | struct bpf_func_state *state; |
6837 | struct bpf_reg_state *reg; | |
1b986589 | 6838 | int err; |
fd978bf7 | 6839 | |
1b986589 MKL |
6840 | err = release_reference_state(cur_func(env), ref_obj_id); |
6841 | if (err) | |
6842 | return err; | |
6843 | ||
b239da34 | 6844 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ |
f1db2081 YL |
6845 | if (reg->ref_obj_id == ref_obj_id) { |
6846 | if (!env->allow_ptr_leaks) | |
6847 | __mark_reg_not_init(env, reg); | |
6848 | else | |
6849 | __mark_reg_unknown(env, reg); | |
6850 | } | |
b239da34 | 6851 | })); |
fd978bf7 | 6852 | |
1b986589 | 6853 | return 0; |
fd978bf7 JS |
6854 | } |
6855 | ||
51c39bb1 AS |
6856 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
6857 | struct bpf_reg_state *regs) | |
6858 | { | |
6859 | int i; | |
6860 | ||
6861 | /* after the call registers r0 - r5 were scratched */ | |
6862 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
6863 | mark_reg_not_init(env, regs, caller_saved[i]); | |
6864 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
6865 | } | |
6866 | } | |
6867 | ||
14351375 YS |
6868 | typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env, |
6869 | struct bpf_func_state *caller, | |
6870 | struct bpf_func_state *callee, | |
6871 | int insn_idx); | |
6872 | ||
be2ef816 AN |
6873 | static int set_callee_state(struct bpf_verifier_env *env, |
6874 | struct bpf_func_state *caller, | |
6875 | struct bpf_func_state *callee, int insn_idx); | |
6876 | ||
14351375 YS |
6877 | static int __check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
6878 | int *insn_idx, int subprog, | |
6879 | set_callee_state_fn set_callee_state_cb) | |
f4d7e40a AS |
6880 | { |
6881 | struct bpf_verifier_state *state = env->cur_state; | |
51c39bb1 | 6882 | struct bpf_func_info_aux *func_info_aux; |
f4d7e40a | 6883 | struct bpf_func_state *caller, *callee; |
14351375 | 6884 | int err; |
51c39bb1 | 6885 | bool is_global = false; |
f4d7e40a | 6886 | |
aada9ce6 | 6887 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 6888 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 6889 | state->curframe + 2); |
f4d7e40a AS |
6890 | return -E2BIG; |
6891 | } | |
6892 | ||
f4d7e40a AS |
6893 | caller = state->frame[state->curframe]; |
6894 | if (state->frame[state->curframe + 1]) { | |
6895 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
6896 | state->curframe + 1); | |
6897 | return -EFAULT; | |
6898 | } | |
6899 | ||
51c39bb1 AS |
6900 | func_info_aux = env->prog->aux->func_info_aux; |
6901 | if (func_info_aux) | |
6902 | is_global = func_info_aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
95f2f26f | 6903 | err = btf_check_subprog_call(env, subprog, caller->regs); |
51c39bb1 AS |
6904 | if (err == -EFAULT) |
6905 | return err; | |
6906 | if (is_global) { | |
6907 | if (err) { | |
6908 | verbose(env, "Caller passes invalid args into func#%d\n", | |
6909 | subprog); | |
6910 | return err; | |
6911 | } else { | |
6912 | if (env->log.level & BPF_LOG_LEVEL) | |
6913 | verbose(env, | |
6914 | "Func#%d is global and valid. Skipping.\n", | |
6915 | subprog); | |
6916 | clear_caller_saved_regs(env, caller->regs); | |
6917 | ||
45159b27 | 6918 | /* All global functions return a 64-bit SCALAR_VALUE */ |
51c39bb1 | 6919 | mark_reg_unknown(env, caller->regs, BPF_REG_0); |
45159b27 | 6920 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; |
51c39bb1 AS |
6921 | |
6922 | /* continue with next insn after call */ | |
6923 | return 0; | |
6924 | } | |
6925 | } | |
6926 | ||
be2ef816 AN |
6927 | /* set_callee_state is used for direct subprog calls, but we are |
6928 | * interested in validating only BPF helpers that can call subprogs as | |
6929 | * callbacks | |
6930 | */ | |
6931 | if (set_callee_state_cb != set_callee_state && !is_callback_calling_function(insn->imm)) { | |
6932 | verbose(env, "verifier bug: helper %s#%d is not marked as callback-calling\n", | |
6933 | func_id_name(insn->imm), insn->imm); | |
6934 | return -EFAULT; | |
6935 | } | |
6936 | ||
bfc6bb74 | 6937 | if (insn->code == (BPF_JMP | BPF_CALL) && |
a5bebc4f | 6938 | insn->src_reg == 0 && |
bfc6bb74 AS |
6939 | insn->imm == BPF_FUNC_timer_set_callback) { |
6940 | struct bpf_verifier_state *async_cb; | |
6941 | ||
6942 | /* there is no real recursion here. timer callbacks are async */ | |
7ddc80a4 | 6943 | env->subprog_info[subprog].is_async_cb = true; |
bfc6bb74 AS |
6944 | async_cb = push_async_cb(env, env->subprog_info[subprog].start, |
6945 | *insn_idx, subprog); | |
6946 | if (!async_cb) | |
6947 | return -EFAULT; | |
6948 | callee = async_cb->frame[0]; | |
6949 | callee->async_entry_cnt = caller->async_entry_cnt + 1; | |
6950 | ||
6951 | /* Convert bpf_timer_set_callback() args into timer callback args */ | |
6952 | err = set_callee_state_cb(env, caller, callee, *insn_idx); | |
6953 | if (err) | |
6954 | return err; | |
6955 | ||
6956 | clear_caller_saved_regs(env, caller->regs); | |
6957 | mark_reg_unknown(env, caller->regs, BPF_REG_0); | |
6958 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
6959 | /* continue with next insn after call */ | |
6960 | return 0; | |
6961 | } | |
6962 | ||
f4d7e40a AS |
6963 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); |
6964 | if (!callee) | |
6965 | return -ENOMEM; | |
6966 | state->frame[state->curframe + 1] = callee; | |
6967 | ||
6968 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
6969 | * into its own stack before reading from it. | |
6970 | * callee can read/write into caller's stack | |
6971 | */ | |
6972 | init_func_state(env, callee, | |
6973 | /* remember the callsite, it will be used by bpf_exit */ | |
6974 | *insn_idx /* callsite */, | |
6975 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 6976 | subprog /* subprog number within this prog */); |
f4d7e40a | 6977 | |
fd978bf7 | 6978 | /* Transfer references to the callee */ |
c69431aa | 6979 | err = copy_reference_state(callee, caller); |
fd978bf7 JS |
6980 | if (err) |
6981 | return err; | |
6982 | ||
14351375 YS |
6983 | err = set_callee_state_cb(env, caller, callee, *insn_idx); |
6984 | if (err) | |
6985 | return err; | |
f4d7e40a | 6986 | |
51c39bb1 | 6987 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a AS |
6988 | |
6989 | /* only increment it after check_reg_arg() finished */ | |
6990 | state->curframe++; | |
6991 | ||
6992 | /* and go analyze first insn of the callee */ | |
14351375 | 6993 | *insn_idx = env->subprog_info[subprog].start - 1; |
f4d7e40a | 6994 | |
06ee7115 | 6995 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 6996 | verbose(env, "caller:\n"); |
0f55f9ed | 6997 | print_verifier_state(env, caller, true); |
f4d7e40a | 6998 | verbose(env, "callee:\n"); |
0f55f9ed | 6999 | print_verifier_state(env, callee, true); |
f4d7e40a AS |
7000 | } |
7001 | return 0; | |
7002 | } | |
7003 | ||
314ee05e YS |
7004 | int map_set_for_each_callback_args(struct bpf_verifier_env *env, |
7005 | struct bpf_func_state *caller, | |
7006 | struct bpf_func_state *callee) | |
7007 | { | |
7008 | /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn, | |
7009 | * void *callback_ctx, u64 flags); | |
7010 | * callback_fn(struct bpf_map *map, void *key, void *value, | |
7011 | * void *callback_ctx); | |
7012 | */ | |
7013 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
7014 | ||
7015 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
7016 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
7017 | callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
7018 | ||
7019 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
7020 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
7021 | callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
7022 | ||
7023 | /* pointer to stack or null */ | |
7024 | callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3]; | |
7025 | ||
7026 | /* unused */ | |
7027 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
7028 | return 0; | |
7029 | } | |
7030 | ||
14351375 YS |
7031 | static int set_callee_state(struct bpf_verifier_env *env, |
7032 | struct bpf_func_state *caller, | |
7033 | struct bpf_func_state *callee, int insn_idx) | |
7034 | { | |
7035 | int i; | |
7036 | ||
7037 | /* copy r1 - r5 args that callee can access. The copy includes parent | |
7038 | * pointers, which connects us up to the liveness chain | |
7039 | */ | |
7040 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
7041 | callee->regs[i] = caller->regs[i]; | |
7042 | return 0; | |
7043 | } | |
7044 | ||
7045 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
7046 | int *insn_idx) | |
7047 | { | |
7048 | int subprog, target_insn; | |
7049 | ||
7050 | target_insn = *insn_idx + insn->imm + 1; | |
7051 | subprog = find_subprog(env, target_insn); | |
7052 | if (subprog < 0) { | |
7053 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
7054 | target_insn); | |
7055 | return -EFAULT; | |
7056 | } | |
7057 | ||
7058 | return __check_func_call(env, insn, insn_idx, subprog, set_callee_state); | |
7059 | } | |
7060 | ||
69c087ba YS |
7061 | static int set_map_elem_callback_state(struct bpf_verifier_env *env, |
7062 | struct bpf_func_state *caller, | |
7063 | struct bpf_func_state *callee, | |
7064 | int insn_idx) | |
7065 | { | |
7066 | struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx]; | |
7067 | struct bpf_map *map; | |
7068 | int err; | |
7069 | ||
7070 | if (bpf_map_ptr_poisoned(insn_aux)) { | |
7071 | verbose(env, "tail_call abusing map_ptr\n"); | |
7072 | return -EINVAL; | |
7073 | } | |
7074 | ||
7075 | map = BPF_MAP_PTR(insn_aux->map_ptr_state); | |
7076 | if (!map->ops->map_set_for_each_callback_args || | |
7077 | !map->ops->map_for_each_callback) { | |
7078 | verbose(env, "callback function not allowed for map\n"); | |
7079 | return -ENOTSUPP; | |
7080 | } | |
7081 | ||
7082 | err = map->ops->map_set_for_each_callback_args(env, caller, callee); | |
7083 | if (err) | |
7084 | return err; | |
7085 | ||
7086 | callee->in_callback_fn = true; | |
1bfe26fb | 7087 | callee->callback_ret_range = tnum_range(0, 1); |
69c087ba YS |
7088 | return 0; |
7089 | } | |
7090 | ||
e6f2dd0f JK |
7091 | static int set_loop_callback_state(struct bpf_verifier_env *env, |
7092 | struct bpf_func_state *caller, | |
7093 | struct bpf_func_state *callee, | |
7094 | int insn_idx) | |
7095 | { | |
7096 | /* bpf_loop(u32 nr_loops, void *callback_fn, void *callback_ctx, | |
7097 | * u64 flags); | |
7098 | * callback_fn(u32 index, void *callback_ctx); | |
7099 | */ | |
7100 | callee->regs[BPF_REG_1].type = SCALAR_VALUE; | |
7101 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; | |
7102 | ||
7103 | /* unused */ | |
7104 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
7105 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
7106 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
7107 | ||
7108 | callee->in_callback_fn = true; | |
1bfe26fb | 7109 | callee->callback_ret_range = tnum_range(0, 1); |
e6f2dd0f JK |
7110 | return 0; |
7111 | } | |
7112 | ||
b00628b1 AS |
7113 | static int set_timer_callback_state(struct bpf_verifier_env *env, |
7114 | struct bpf_func_state *caller, | |
7115 | struct bpf_func_state *callee, | |
7116 | int insn_idx) | |
7117 | { | |
7118 | struct bpf_map *map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
7119 | ||
7120 | /* bpf_timer_set_callback(struct bpf_timer *timer, void *callback_fn); | |
7121 | * callback_fn(struct bpf_map *map, void *key, void *value); | |
7122 | */ | |
7123 | callee->regs[BPF_REG_1].type = CONST_PTR_TO_MAP; | |
7124 | __mark_reg_known_zero(&callee->regs[BPF_REG_1]); | |
7125 | callee->regs[BPF_REG_1].map_ptr = map_ptr; | |
7126 | ||
7127 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
7128 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
7129 | callee->regs[BPF_REG_2].map_ptr = map_ptr; | |
7130 | ||
7131 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
7132 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
7133 | callee->regs[BPF_REG_3].map_ptr = map_ptr; | |
7134 | ||
7135 | /* unused */ | |
7136 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
7137 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
bfc6bb74 | 7138 | callee->in_async_callback_fn = true; |
1bfe26fb | 7139 | callee->callback_ret_range = tnum_range(0, 1); |
b00628b1 AS |
7140 | return 0; |
7141 | } | |
7142 | ||
7c7e3d31 SL |
7143 | static int set_find_vma_callback_state(struct bpf_verifier_env *env, |
7144 | struct bpf_func_state *caller, | |
7145 | struct bpf_func_state *callee, | |
7146 | int insn_idx) | |
7147 | { | |
7148 | /* bpf_find_vma(struct task_struct *task, u64 addr, | |
7149 | * void *callback_fn, void *callback_ctx, u64 flags) | |
7150 | * (callback_fn)(struct task_struct *task, | |
7151 | * struct vm_area_struct *vma, void *callback_ctx); | |
7152 | */ | |
7153 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
7154 | ||
7155 | callee->regs[BPF_REG_2].type = PTR_TO_BTF_ID; | |
7156 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
7157 | callee->regs[BPF_REG_2].btf = btf_vmlinux; | |
d19ddb47 | 7158 | callee->regs[BPF_REG_2].btf_id = btf_tracing_ids[BTF_TRACING_TYPE_VMA], |
7c7e3d31 SL |
7159 | |
7160 | /* pointer to stack or null */ | |
7161 | callee->regs[BPF_REG_3] = caller->regs[BPF_REG_4]; | |
7162 | ||
7163 | /* unused */ | |
7164 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
7165 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
7166 | callee->in_callback_fn = true; | |
1bfe26fb | 7167 | callee->callback_ret_range = tnum_range(0, 1); |
7c7e3d31 SL |
7168 | return 0; |
7169 | } | |
7170 | ||
20571567 DV |
7171 | static int set_user_ringbuf_callback_state(struct bpf_verifier_env *env, |
7172 | struct bpf_func_state *caller, | |
7173 | struct bpf_func_state *callee, | |
7174 | int insn_idx) | |
7175 | { | |
7176 | /* bpf_user_ringbuf_drain(struct bpf_map *map, void *callback_fn, void | |
7177 | * callback_ctx, u64 flags); | |
7178 | * callback_fn(struct bpf_dynptr_t* dynptr, void *callback_ctx); | |
7179 | */ | |
7180 | __mark_reg_not_init(env, &callee->regs[BPF_REG_0]); | |
7181 | callee->regs[BPF_REG_1].type = PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL; | |
7182 | __mark_reg_known_zero(&callee->regs[BPF_REG_1]); | |
7183 | callee->regs[BPF_REG_2] = caller->regs[BPF_REG_3]; | |
7184 | ||
7185 | /* unused */ | |
7186 | __mark_reg_not_init(env, &callee->regs[BPF_REG_3]); | |
7187 | __mark_reg_not_init(env, &callee->regs[BPF_REG_4]); | |
7188 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
7189 | ||
7190 | callee->in_callback_fn = true; | |
c92a7a52 | 7191 | callee->callback_ret_range = tnum_range(0, 1); |
20571567 DV |
7192 | return 0; |
7193 | } | |
7194 | ||
f4d7e40a AS |
7195 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) |
7196 | { | |
7197 | struct bpf_verifier_state *state = env->cur_state; | |
7198 | struct bpf_func_state *caller, *callee; | |
7199 | struct bpf_reg_state *r0; | |
fd978bf7 | 7200 | int err; |
f4d7e40a AS |
7201 | |
7202 | callee = state->frame[state->curframe]; | |
7203 | r0 = &callee->regs[BPF_REG_0]; | |
7204 | if (r0->type == PTR_TO_STACK) { | |
7205 | /* technically it's ok to return caller's stack pointer | |
7206 | * (or caller's caller's pointer) back to the caller, | |
7207 | * since these pointers are valid. Only current stack | |
7208 | * pointer will be invalid as soon as function exits, | |
7209 | * but let's be conservative | |
7210 | */ | |
7211 | verbose(env, "cannot return stack pointer to the caller\n"); | |
7212 | return -EINVAL; | |
7213 | } | |
7214 | ||
7215 | state->curframe--; | |
7216 | caller = state->frame[state->curframe]; | |
69c087ba YS |
7217 | if (callee->in_callback_fn) { |
7218 | /* enforce R0 return value range [0, 1]. */ | |
1bfe26fb | 7219 | struct tnum range = callee->callback_ret_range; |
69c087ba YS |
7220 | |
7221 | if (r0->type != SCALAR_VALUE) { | |
7222 | verbose(env, "R0 not a scalar value\n"); | |
7223 | return -EACCES; | |
7224 | } | |
7225 | if (!tnum_in(range, r0->var_off)) { | |
7226 | verbose_invalid_scalar(env, r0, &range, "callback return", "R0"); | |
7227 | return -EINVAL; | |
7228 | } | |
7229 | } else { | |
7230 | /* return to the caller whatever r0 had in the callee */ | |
7231 | caller->regs[BPF_REG_0] = *r0; | |
7232 | } | |
f4d7e40a | 7233 | |
9d9d00ac KKD |
7234 | /* callback_fn frame should have released its own additions to parent's |
7235 | * reference state at this point, or check_reference_leak would | |
7236 | * complain, hence it must be the same as the caller. There is no need | |
7237 | * to copy it back. | |
7238 | */ | |
7239 | if (!callee->in_callback_fn) { | |
7240 | /* Transfer references to the caller */ | |
7241 | err = copy_reference_state(caller, callee); | |
7242 | if (err) | |
7243 | return err; | |
7244 | } | |
fd978bf7 | 7245 | |
f4d7e40a | 7246 | *insn_idx = callee->callsite + 1; |
06ee7115 | 7247 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a | 7248 | verbose(env, "returning from callee:\n"); |
0f55f9ed | 7249 | print_verifier_state(env, callee, true); |
f4d7e40a | 7250 | verbose(env, "to caller at %d:\n", *insn_idx); |
0f55f9ed | 7251 | print_verifier_state(env, caller, true); |
f4d7e40a AS |
7252 | } |
7253 | /* clear everything in the callee */ | |
7254 | free_func_state(callee); | |
7255 | state->frame[state->curframe + 1] = NULL; | |
7256 | return 0; | |
7257 | } | |
7258 | ||
849fa506 YS |
7259 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
7260 | int func_id, | |
7261 | struct bpf_call_arg_meta *meta) | |
7262 | { | |
7263 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
7264 | ||
7265 | if (ret_type != RET_INTEGER || | |
7266 | (func_id != BPF_FUNC_get_stack && | |
fd0b88f7 | 7267 | func_id != BPF_FUNC_get_task_stack && |
47cc0ed5 DB |
7268 | func_id != BPF_FUNC_probe_read_str && |
7269 | func_id != BPF_FUNC_probe_read_kernel_str && | |
7270 | func_id != BPF_FUNC_probe_read_user_str)) | |
849fa506 YS |
7271 | return; |
7272 | ||
10060503 | 7273 | ret_reg->smax_value = meta->msize_max_value; |
fa123ac0 | 7274 | ret_reg->s32_max_value = meta->msize_max_value; |
b0270958 AS |
7275 | ret_reg->smin_value = -MAX_ERRNO; |
7276 | ret_reg->s32_min_value = -MAX_ERRNO; | |
3844d153 | 7277 | reg_bounds_sync(ret_reg); |
849fa506 YS |
7278 | } |
7279 | ||
c93552c4 DB |
7280 | static int |
7281 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
7282 | int func_id, int insn_idx) | |
7283 | { | |
7284 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 7285 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
7286 | |
7287 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
7288 | func_id != BPF_FUNC_map_lookup_elem && |
7289 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
7290 | func_id != BPF_FUNC_map_delete_elem && |
7291 | func_id != BPF_FUNC_map_push_elem && | |
7292 | func_id != BPF_FUNC_map_pop_elem && | |
69c087ba | 7293 | func_id != BPF_FUNC_map_peek_elem && |
e6a4750f | 7294 | func_id != BPF_FUNC_for_each_map_elem && |
07343110 FZ |
7295 | func_id != BPF_FUNC_redirect_map && |
7296 | func_id != BPF_FUNC_map_lookup_percpu_elem) | |
c93552c4 | 7297 | return 0; |
09772d92 | 7298 | |
591fe988 | 7299 | if (map == NULL) { |
c93552c4 DB |
7300 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
7301 | return -EINVAL; | |
7302 | } | |
7303 | ||
591fe988 DB |
7304 | /* In case of read-only, some additional restrictions |
7305 | * need to be applied in order to prevent altering the | |
7306 | * state of the map from program side. | |
7307 | */ | |
7308 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
7309 | (func_id == BPF_FUNC_map_delete_elem || | |
7310 | func_id == BPF_FUNC_map_update_elem || | |
7311 | func_id == BPF_FUNC_map_push_elem || | |
7312 | func_id == BPF_FUNC_map_pop_elem)) { | |
7313 | verbose(env, "write into map forbidden\n"); | |
7314 | return -EACCES; | |
7315 | } | |
7316 | ||
d2e4c1e6 | 7317 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 7318 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 7319 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 7320 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 7321 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 7322 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
7323 | return 0; |
7324 | } | |
7325 | ||
d2e4c1e6 DB |
7326 | static int |
7327 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
7328 | int func_id, int insn_idx) | |
7329 | { | |
7330 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
7331 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
7332 | struct bpf_map *map = meta->map_ptr; | |
a657182a | 7333 | u64 val, max; |
cc52d914 | 7334 | int err; |
d2e4c1e6 DB |
7335 | |
7336 | if (func_id != BPF_FUNC_tail_call) | |
7337 | return 0; | |
7338 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
7339 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
7340 | return -EINVAL; | |
7341 | } | |
7342 | ||
d2e4c1e6 | 7343 | reg = ®s[BPF_REG_3]; |
a657182a DB |
7344 | val = reg->var_off.value; |
7345 | max = map->max_entries; | |
d2e4c1e6 | 7346 | |
a657182a | 7347 | if (!(register_is_const(reg) && val < max)) { |
d2e4c1e6 DB |
7348 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); |
7349 | return 0; | |
7350 | } | |
7351 | ||
cc52d914 DB |
7352 | err = mark_chain_precision(env, BPF_REG_3); |
7353 | if (err) | |
7354 | return err; | |
d2e4c1e6 DB |
7355 | if (bpf_map_key_unseen(aux)) |
7356 | bpf_map_key_store(aux, val); | |
7357 | else if (!bpf_map_key_poisoned(aux) && | |
7358 | bpf_map_key_immediate(aux) != val) | |
7359 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
7360 | return 0; | |
7361 | } | |
7362 | ||
fd978bf7 JS |
7363 | static int check_reference_leak(struct bpf_verifier_env *env) |
7364 | { | |
7365 | struct bpf_func_state *state = cur_func(env); | |
9d9d00ac | 7366 | bool refs_lingering = false; |
fd978bf7 JS |
7367 | int i; |
7368 | ||
9d9d00ac KKD |
7369 | if (state->frameno && !state->in_callback_fn) |
7370 | return 0; | |
7371 | ||
fd978bf7 | 7372 | for (i = 0; i < state->acquired_refs; i++) { |
9d9d00ac KKD |
7373 | if (state->in_callback_fn && state->refs[i].callback_ref != state->frameno) |
7374 | continue; | |
fd978bf7 JS |
7375 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", |
7376 | state->refs[i].id, state->refs[i].insn_idx); | |
9d9d00ac | 7377 | refs_lingering = true; |
fd978bf7 | 7378 | } |
9d9d00ac | 7379 | return refs_lingering ? -EINVAL : 0; |
fd978bf7 JS |
7380 | } |
7381 | ||
7b15523a FR |
7382 | static int check_bpf_snprintf_call(struct bpf_verifier_env *env, |
7383 | struct bpf_reg_state *regs) | |
7384 | { | |
7385 | struct bpf_reg_state *fmt_reg = ®s[BPF_REG_3]; | |
7386 | struct bpf_reg_state *data_len_reg = ®s[BPF_REG_5]; | |
7387 | struct bpf_map *fmt_map = fmt_reg->map_ptr; | |
7388 | int err, fmt_map_off, num_args; | |
7389 | u64 fmt_addr; | |
7390 | char *fmt; | |
7391 | ||
7392 | /* data must be an array of u64 */ | |
7393 | if (data_len_reg->var_off.value % 8) | |
7394 | return -EINVAL; | |
7395 | num_args = data_len_reg->var_off.value / 8; | |
7396 | ||
7397 | /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const | |
7398 | * and map_direct_value_addr is set. | |
7399 | */ | |
7400 | fmt_map_off = fmt_reg->off + fmt_reg->var_off.value; | |
7401 | err = fmt_map->ops->map_direct_value_addr(fmt_map, &fmt_addr, | |
7402 | fmt_map_off); | |
8e8ee109 FR |
7403 | if (err) { |
7404 | verbose(env, "verifier bug\n"); | |
7405 | return -EFAULT; | |
7406 | } | |
7b15523a FR |
7407 | fmt = (char *)(long)fmt_addr + fmt_map_off; |
7408 | ||
7409 | /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we | |
7410 | * can focus on validating the format specifiers. | |
7411 | */ | |
48cac3f4 | 7412 | err = bpf_bprintf_prepare(fmt, UINT_MAX, NULL, NULL, num_args); |
7b15523a FR |
7413 | if (err < 0) |
7414 | verbose(env, "Invalid format string\n"); | |
7415 | ||
7416 | return err; | |
7417 | } | |
7418 | ||
9b99edca JO |
7419 | static int check_get_func_ip(struct bpf_verifier_env *env) |
7420 | { | |
9b99edca JO |
7421 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
7422 | int func_id = BPF_FUNC_get_func_ip; | |
7423 | ||
7424 | if (type == BPF_PROG_TYPE_TRACING) { | |
f92c1e18 | 7425 | if (!bpf_prog_has_trampoline(env->prog)) { |
9b99edca JO |
7426 | verbose(env, "func %s#%d supported only for fentry/fexit/fmod_ret programs\n", |
7427 | func_id_name(func_id), func_id); | |
7428 | return -ENOTSUPP; | |
7429 | } | |
7430 | return 0; | |
9ffd9f3f JO |
7431 | } else if (type == BPF_PROG_TYPE_KPROBE) { |
7432 | return 0; | |
9b99edca JO |
7433 | } |
7434 | ||
7435 | verbose(env, "func %s#%d not supported for program type %d\n", | |
7436 | func_id_name(func_id), func_id, type); | |
7437 | return -ENOTSUPP; | |
7438 | } | |
7439 | ||
1ade2371 EZ |
7440 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
7441 | { | |
7442 | return &env->insn_aux_data[env->insn_idx]; | |
7443 | } | |
7444 | ||
7445 | static bool loop_flag_is_zero(struct bpf_verifier_env *env) | |
7446 | { | |
7447 | struct bpf_reg_state *regs = cur_regs(env); | |
7448 | struct bpf_reg_state *reg = ®s[BPF_REG_4]; | |
7449 | bool reg_is_null = register_is_null(reg); | |
7450 | ||
7451 | if (reg_is_null) | |
7452 | mark_chain_precision(env, BPF_REG_4); | |
7453 | ||
7454 | return reg_is_null; | |
7455 | } | |
7456 | ||
7457 | static void update_loop_inline_state(struct bpf_verifier_env *env, u32 subprogno) | |
7458 | { | |
7459 | struct bpf_loop_inline_state *state = &cur_aux(env)->loop_inline_state; | |
7460 | ||
7461 | if (!state->initialized) { | |
7462 | state->initialized = 1; | |
7463 | state->fit_for_inline = loop_flag_is_zero(env); | |
7464 | state->callback_subprogno = subprogno; | |
7465 | return; | |
7466 | } | |
7467 | ||
7468 | if (!state->fit_for_inline) | |
7469 | return; | |
7470 | ||
7471 | state->fit_for_inline = (loop_flag_is_zero(env) && | |
7472 | state->callback_subprogno == subprogno); | |
7473 | } | |
7474 | ||
69c087ba YS |
7475 | static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
7476 | int *insn_idx_p) | |
17a52670 | 7477 | { |
aef9d4a3 | 7478 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
17a52670 | 7479 | const struct bpf_func_proto *fn = NULL; |
3c480732 | 7480 | enum bpf_return_type ret_type; |
c25b2ae1 | 7481 | enum bpf_type_flag ret_flag; |
638f5b90 | 7482 | struct bpf_reg_state *regs; |
33ff9823 | 7483 | struct bpf_call_arg_meta meta; |
69c087ba | 7484 | int insn_idx = *insn_idx_p; |
969bf05e | 7485 | bool changes_data; |
69c087ba | 7486 | int i, err, func_id; |
17a52670 AS |
7487 | |
7488 | /* find function prototype */ | |
69c087ba | 7489 | func_id = insn->imm; |
17a52670 | 7490 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { |
61bd5218 JK |
7491 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
7492 | func_id); | |
17a52670 AS |
7493 | return -EINVAL; |
7494 | } | |
7495 | ||
00176a34 | 7496 | if (env->ops->get_func_proto) |
5e43f899 | 7497 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 7498 | if (!fn) { |
61bd5218 JK |
7499 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
7500 | func_id); | |
17a52670 AS |
7501 | return -EINVAL; |
7502 | } | |
7503 | ||
7504 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 7505 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 7506 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
7507 | return -EINVAL; |
7508 | } | |
7509 | ||
eae2e83e JO |
7510 | if (fn->allowed && !fn->allowed(env->prog)) { |
7511 | verbose(env, "helper call is not allowed in probe\n"); | |
7512 | return -EINVAL; | |
7513 | } | |
7514 | ||
04514d13 | 7515 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 7516 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
7517 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
7518 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
7519 | func_id_name(func_id), func_id); | |
7520 | return -EINVAL; | |
7521 | } | |
969bf05e | 7522 | |
33ff9823 | 7523 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 7524 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 7525 | |
0c9a7a7e | 7526 | err = check_func_proto(fn, func_id); |
435faee1 | 7527 | if (err) { |
61bd5218 | 7528 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 7529 | func_id_name(func_id), func_id); |
435faee1 DB |
7530 | return err; |
7531 | } | |
7532 | ||
d83525ca | 7533 | meta.func_id = func_id; |
17a52670 | 7534 | /* check args */ |
523a4cf4 | 7535 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { |
af7ec138 | 7536 | err = check_func_arg(env, i, &meta, fn); |
a7658e1a AS |
7537 | if (err) |
7538 | return err; | |
7539 | } | |
17a52670 | 7540 | |
c93552c4 DB |
7541 | err = record_func_map(env, &meta, func_id, insn_idx); |
7542 | if (err) | |
7543 | return err; | |
7544 | ||
d2e4c1e6 DB |
7545 | err = record_func_key(env, &meta, func_id, insn_idx); |
7546 | if (err) | |
7547 | return err; | |
7548 | ||
435faee1 DB |
7549 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
7550 | * is inferred from register state. | |
7551 | */ | |
7552 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
7553 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
7554 | BPF_WRITE, -1, false); | |
435faee1 DB |
7555 | if (err) |
7556 | return err; | |
7557 | } | |
7558 | ||
8f14852e KKD |
7559 | regs = cur_regs(env); |
7560 | ||
97e03f52 JK |
7561 | if (meta.uninit_dynptr_regno) { |
7562 | /* we write BPF_DW bits (8 bytes) at a time */ | |
7563 | for (i = 0; i < BPF_DYNPTR_SIZE; i += 8) { | |
7564 | err = check_mem_access(env, insn_idx, meta.uninit_dynptr_regno, | |
7565 | i, BPF_DW, BPF_WRITE, -1, false); | |
7566 | if (err) | |
7567 | return err; | |
7568 | } | |
7569 | ||
7570 | err = mark_stack_slots_dynptr(env, ®s[meta.uninit_dynptr_regno], | |
7571 | fn->arg_type[meta.uninit_dynptr_regno - BPF_REG_1], | |
7572 | insn_idx); | |
7573 | if (err) | |
7574 | return err; | |
7575 | } | |
7576 | ||
8f14852e KKD |
7577 | if (meta.release_regno) { |
7578 | err = -EINVAL; | |
97e03f52 JK |
7579 | if (arg_type_is_dynptr(fn->arg_type[meta.release_regno - BPF_REG_1])) |
7580 | err = unmark_stack_slots_dynptr(env, ®s[meta.release_regno]); | |
7581 | else if (meta.ref_obj_id) | |
8f14852e KKD |
7582 | err = release_reference(env, meta.ref_obj_id); |
7583 | /* meta.ref_obj_id can only be 0 if register that is meant to be | |
7584 | * released is NULL, which must be > R0. | |
7585 | */ | |
7586 | else if (register_is_null(®s[meta.release_regno])) | |
7587 | err = 0; | |
46f8bc92 MKL |
7588 | if (err) { |
7589 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
7590 | func_id_name(func_id), func_id); | |
fd978bf7 | 7591 | return err; |
46f8bc92 | 7592 | } |
fd978bf7 JS |
7593 | } |
7594 | ||
e6f2dd0f JK |
7595 | switch (func_id) { |
7596 | case BPF_FUNC_tail_call: | |
7597 | err = check_reference_leak(env); | |
7598 | if (err) { | |
7599 | verbose(env, "tail_call would lead to reference leak\n"); | |
7600 | return err; | |
7601 | } | |
7602 | break; | |
7603 | case BPF_FUNC_get_local_storage: | |
7604 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
7605 | * this is required because get_local_storage() can't return an error. | |
7606 | */ | |
7607 | if (!register_is_null(®s[BPF_REG_2])) { | |
7608 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
7609 | return -EINVAL; | |
7610 | } | |
7611 | break; | |
7612 | case BPF_FUNC_for_each_map_elem: | |
69c087ba YS |
7613 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
7614 | set_map_elem_callback_state); | |
e6f2dd0f JK |
7615 | break; |
7616 | case BPF_FUNC_timer_set_callback: | |
b00628b1 AS |
7617 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
7618 | set_timer_callback_state); | |
e6f2dd0f JK |
7619 | break; |
7620 | case BPF_FUNC_find_vma: | |
7c7e3d31 SL |
7621 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
7622 | set_find_vma_callback_state); | |
e6f2dd0f JK |
7623 | break; |
7624 | case BPF_FUNC_snprintf: | |
7b15523a | 7625 | err = check_bpf_snprintf_call(env, regs); |
e6f2dd0f JK |
7626 | break; |
7627 | case BPF_FUNC_loop: | |
1ade2371 | 7628 | update_loop_inline_state(env, meta.subprogno); |
e6f2dd0f JK |
7629 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, |
7630 | set_loop_callback_state); | |
7631 | break; | |
263ae152 JK |
7632 | case BPF_FUNC_dynptr_from_mem: |
7633 | if (regs[BPF_REG_1].type != PTR_TO_MAP_VALUE) { | |
7634 | verbose(env, "Unsupported reg type %s for bpf_dynptr_from_mem data\n", | |
7635 | reg_type_str(env, regs[BPF_REG_1].type)); | |
7636 | return -EACCES; | |
7637 | } | |
69fd337a SF |
7638 | break; |
7639 | case BPF_FUNC_set_retval: | |
aef9d4a3 SF |
7640 | if (prog_type == BPF_PROG_TYPE_LSM && |
7641 | env->prog->expected_attach_type == BPF_LSM_CGROUP) { | |
69fd337a SF |
7642 | if (!env->prog->aux->attach_func_proto->type) { |
7643 | /* Make sure programs that attach to void | |
7644 | * hooks don't try to modify return value. | |
7645 | */ | |
7646 | verbose(env, "BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
7647 | return -EINVAL; | |
7648 | } | |
7649 | } | |
7650 | break; | |
88374342 JK |
7651 | case BPF_FUNC_dynptr_data: |
7652 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { | |
7653 | if (arg_type_is_dynptr(fn->arg_type[i])) { | |
20571567 DV |
7654 | struct bpf_reg_state *reg = ®s[BPF_REG_1 + i]; |
7655 | ||
88374342 JK |
7656 | if (meta.ref_obj_id) { |
7657 | verbose(env, "verifier internal error: meta.ref_obj_id already set\n"); | |
7658 | return -EFAULT; | |
7659 | } | |
20571567 DV |
7660 | |
7661 | if (base_type(reg->type) != PTR_TO_DYNPTR) | |
7662 | /* Find the id of the dynptr we're | |
7663 | * tracking the reference of | |
7664 | */ | |
7665 | meta.ref_obj_id = stack_slot_get_id(env, reg); | |
88374342 JK |
7666 | break; |
7667 | } | |
7668 | } | |
7669 | if (i == MAX_BPF_FUNC_REG_ARGS) { | |
7670 | verbose(env, "verifier internal error: no dynptr in bpf_dynptr_data()\n"); | |
7671 | return -EFAULT; | |
7672 | } | |
7673 | break; | |
20571567 DV |
7674 | case BPF_FUNC_user_ringbuf_drain: |
7675 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, | |
7676 | set_user_ringbuf_callback_state); | |
7677 | break; | |
7b15523a FR |
7678 | } |
7679 | ||
e6f2dd0f JK |
7680 | if (err) |
7681 | return err; | |
7682 | ||
17a52670 | 7683 | /* reset caller saved regs */ |
dc503a8a | 7684 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 7685 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
7686 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
7687 | } | |
17a52670 | 7688 | |
5327ed3d JW |
7689 | /* helper call returns 64-bit value. */ |
7690 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
7691 | ||
dc503a8a | 7692 | /* update return register (already marked as written above) */ |
3c480732 | 7693 | ret_type = fn->ret_type; |
0c9a7a7e JK |
7694 | ret_flag = type_flag(ret_type); |
7695 | ||
7696 | switch (base_type(ret_type)) { | |
7697 | case RET_INTEGER: | |
f1174f77 | 7698 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 7699 | mark_reg_unknown(env, regs, BPF_REG_0); |
0c9a7a7e JK |
7700 | break; |
7701 | case RET_VOID: | |
17a52670 | 7702 | regs[BPF_REG_0].type = NOT_INIT; |
0c9a7a7e JK |
7703 | break; |
7704 | case RET_PTR_TO_MAP_VALUE: | |
f1174f77 | 7705 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 7706 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
7707 | /* remember map_ptr, so that check_map_access() |
7708 | * can check 'value_size' boundary of memory access | |
7709 | * to map element returned from bpf_map_lookup_elem() | |
7710 | */ | |
33ff9823 | 7711 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
7712 | verbose(env, |
7713 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
7714 | return -EINVAL; |
7715 | } | |
33ff9823 | 7716 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
3e8ce298 | 7717 | regs[BPF_REG_0].map_uid = meta.map_uid; |
c25b2ae1 HL |
7718 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE | ret_flag; |
7719 | if (!type_may_be_null(ret_type) && | |
db559117 | 7720 | btf_record_has_field(meta.map_ptr->record, BPF_SPIN_LOCK)) { |
c25b2ae1 | 7721 | regs[BPF_REG_0].id = ++env->id_gen; |
4d31f301 | 7722 | } |
0c9a7a7e JK |
7723 | break; |
7724 | case RET_PTR_TO_SOCKET: | |
c64b7983 | 7725 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 7726 | regs[BPF_REG_0].type = PTR_TO_SOCKET | ret_flag; |
0c9a7a7e JK |
7727 | break; |
7728 | case RET_PTR_TO_SOCK_COMMON: | |
85a51f8c | 7729 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 7730 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON | ret_flag; |
0c9a7a7e JK |
7731 | break; |
7732 | case RET_PTR_TO_TCP_SOCK: | |
655a51e5 | 7733 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 7734 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK | ret_flag; |
0c9a7a7e | 7735 | break; |
2de2669b | 7736 | case RET_PTR_TO_MEM: |
457f4436 | 7737 | mark_reg_known_zero(env, regs, BPF_REG_0); |
c25b2ae1 | 7738 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
457f4436 | 7739 | regs[BPF_REG_0].mem_size = meta.mem_size; |
0c9a7a7e JK |
7740 | break; |
7741 | case RET_PTR_TO_MEM_OR_BTF_ID: | |
7742 | { | |
eaa6bcb7 HL |
7743 | const struct btf_type *t; |
7744 | ||
7745 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
22dc4a0f | 7746 | t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL); |
eaa6bcb7 HL |
7747 | if (!btf_type_is_struct(t)) { |
7748 | u32 tsize; | |
7749 | const struct btf_type *ret; | |
7750 | const char *tname; | |
7751 | ||
7752 | /* resolve the type size of ksym. */ | |
22dc4a0f | 7753 | ret = btf_resolve_size(meta.ret_btf, t, &tsize); |
eaa6bcb7 | 7754 | if (IS_ERR(ret)) { |
22dc4a0f | 7755 | tname = btf_name_by_offset(meta.ret_btf, t->name_off); |
eaa6bcb7 HL |
7756 | verbose(env, "unable to resolve the size of type '%s': %ld\n", |
7757 | tname, PTR_ERR(ret)); | |
7758 | return -EINVAL; | |
7759 | } | |
c25b2ae1 | 7760 | regs[BPF_REG_0].type = PTR_TO_MEM | ret_flag; |
eaa6bcb7 HL |
7761 | regs[BPF_REG_0].mem_size = tsize; |
7762 | } else { | |
34d3a78c HL |
7763 | /* MEM_RDONLY may be carried from ret_flag, but it |
7764 | * doesn't apply on PTR_TO_BTF_ID. Fold it, otherwise | |
7765 | * it will confuse the check of PTR_TO_BTF_ID in | |
7766 | * check_mem_access(). | |
7767 | */ | |
7768 | ret_flag &= ~MEM_RDONLY; | |
7769 | ||
c25b2ae1 | 7770 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; |
22dc4a0f | 7771 | regs[BPF_REG_0].btf = meta.ret_btf; |
eaa6bcb7 HL |
7772 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; |
7773 | } | |
0c9a7a7e JK |
7774 | break; |
7775 | } | |
7776 | case RET_PTR_TO_BTF_ID: | |
7777 | { | |
c0a5a21c | 7778 | struct btf *ret_btf; |
af7ec138 YS |
7779 | int ret_btf_id; |
7780 | ||
7781 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
c25b2ae1 | 7782 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | ret_flag; |
c0a5a21c | 7783 | if (func_id == BPF_FUNC_kptr_xchg) { |
aa3496ac KKD |
7784 | ret_btf = meta.kptr_field->kptr.btf; |
7785 | ret_btf_id = meta.kptr_field->kptr.btf_id; | |
c0a5a21c | 7786 | } else { |
47e34cb7 DM |
7787 | if (fn->ret_btf_id == BPF_PTR_POISON) { |
7788 | verbose(env, "verifier internal error:"); | |
7789 | verbose(env, "func %s has non-overwritten BPF_PTR_POISON return type\n", | |
7790 | func_id_name(func_id)); | |
7791 | return -EINVAL; | |
7792 | } | |
c0a5a21c KKD |
7793 | ret_btf = btf_vmlinux; |
7794 | ret_btf_id = *fn->ret_btf_id; | |
7795 | } | |
af7ec138 | 7796 | if (ret_btf_id == 0) { |
3c480732 HL |
7797 | verbose(env, "invalid return type %u of func %s#%d\n", |
7798 | base_type(ret_type), func_id_name(func_id), | |
7799 | func_id); | |
af7ec138 YS |
7800 | return -EINVAL; |
7801 | } | |
c0a5a21c | 7802 | regs[BPF_REG_0].btf = ret_btf; |
af7ec138 | 7803 | regs[BPF_REG_0].btf_id = ret_btf_id; |
0c9a7a7e JK |
7804 | break; |
7805 | } | |
7806 | default: | |
3c480732 HL |
7807 | verbose(env, "unknown return type %u of func %s#%d\n", |
7808 | base_type(ret_type), func_id_name(func_id), func_id); | |
17a52670 AS |
7809 | return -EINVAL; |
7810 | } | |
04fd61ab | 7811 | |
c25b2ae1 | 7812 | if (type_may_be_null(regs[BPF_REG_0].type)) |
93c230e3 MKL |
7813 | regs[BPF_REG_0].id = ++env->id_gen; |
7814 | ||
b2d8ef19 DM |
7815 | if (helper_multiple_ref_obj_use(func_id, meta.map_ptr)) { |
7816 | verbose(env, "verifier internal error: func %s#%d sets ref_obj_id more than once\n", | |
7817 | func_id_name(func_id), func_id); | |
7818 | return -EFAULT; | |
7819 | } | |
7820 | ||
88374342 | 7821 | if (is_ptr_cast_function(func_id) || is_dynptr_ref_function(func_id)) { |
1b986589 MKL |
7822 | /* For release_reference() */ |
7823 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 7824 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
7825 | int id = acquire_reference_state(env, insn_idx); |
7826 | ||
7827 | if (id < 0) | |
7828 | return id; | |
7829 | /* For mark_ptr_or_null_reg() */ | |
7830 | regs[BPF_REG_0].id = id; | |
7831 | /* For release_reference() */ | |
7832 | regs[BPF_REG_0].ref_obj_id = id; | |
7833 | } | |
1b986589 | 7834 | |
849fa506 YS |
7835 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
7836 | ||
61bd5218 | 7837 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
7838 | if (err) |
7839 | return err; | |
04fd61ab | 7840 | |
fa28dcb8 SL |
7841 | if ((func_id == BPF_FUNC_get_stack || |
7842 | func_id == BPF_FUNC_get_task_stack) && | |
7843 | !env->prog->has_callchain_buf) { | |
c195651e YS |
7844 | const char *err_str; |
7845 | ||
7846 | #ifdef CONFIG_PERF_EVENTS | |
7847 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
7848 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
7849 | #else | |
7850 | err = -ENOTSUPP; | |
7851 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
7852 | #endif | |
7853 | if (err) { | |
7854 | verbose(env, err_str, func_id_name(func_id), func_id); | |
7855 | return err; | |
7856 | } | |
7857 | ||
7858 | env->prog->has_callchain_buf = true; | |
7859 | } | |
7860 | ||
5d99cb2c SL |
7861 | if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack) |
7862 | env->prog->call_get_stack = true; | |
7863 | ||
9b99edca JO |
7864 | if (func_id == BPF_FUNC_get_func_ip) { |
7865 | if (check_get_func_ip(env)) | |
7866 | return -ENOTSUPP; | |
7867 | env->prog->call_get_func_ip = true; | |
7868 | } | |
7869 | ||
969bf05e AS |
7870 | if (changes_data) |
7871 | clear_all_pkt_pointers(env); | |
7872 | return 0; | |
7873 | } | |
7874 | ||
e6ac2450 MKL |
7875 | /* mark_btf_func_reg_size() is used when the reg size is determined by |
7876 | * the BTF func_proto's return value size and argument. | |
7877 | */ | |
7878 | static void mark_btf_func_reg_size(struct bpf_verifier_env *env, u32 regno, | |
7879 | size_t reg_size) | |
7880 | { | |
7881 | struct bpf_reg_state *reg = &cur_regs(env)[regno]; | |
7882 | ||
7883 | if (regno == BPF_REG_0) { | |
7884 | /* Function return value */ | |
7885 | reg->live |= REG_LIVE_WRITTEN; | |
7886 | reg->subreg_def = reg_size == sizeof(u64) ? | |
7887 | DEF_NOT_SUBREG : env->insn_idx + 1; | |
7888 | } else { | |
7889 | /* Function argument */ | |
7890 | if (reg_size == sizeof(u64)) { | |
7891 | mark_insn_zext(env, reg); | |
7892 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
7893 | } else { | |
7894 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ32); | |
7895 | } | |
7896 | } | |
7897 | } | |
7898 | ||
00b85860 KKD |
7899 | struct bpf_kfunc_call_arg_meta { |
7900 | /* In parameters */ | |
7901 | struct btf *btf; | |
7902 | u32 func_id; | |
7903 | u32 kfunc_flags; | |
7904 | const struct btf_type *func_proto; | |
7905 | const char *func_name; | |
7906 | /* Out parameters */ | |
7907 | u32 ref_obj_id; | |
7908 | u8 release_regno; | |
7909 | bool r0_rdonly; | |
fd264ca0 | 7910 | u32 ret_btf_id; |
00b85860 | 7911 | u64 r0_size; |
a50388db KKD |
7912 | struct { |
7913 | u64 value; | |
7914 | bool found; | |
7915 | } arg_constant; | |
ac9f0605 KKD |
7916 | struct { |
7917 | struct btf *btf; | |
7918 | u32 btf_id; | |
7919 | } arg_obj_drop; | |
8cab76ec KKD |
7920 | struct { |
7921 | struct btf_field *field; | |
7922 | } arg_list_head; | |
00b85860 KKD |
7923 | }; |
7924 | ||
7925 | static bool is_kfunc_acquire(struct bpf_kfunc_call_arg_meta *meta) | |
7926 | { | |
7927 | return meta->kfunc_flags & KF_ACQUIRE; | |
7928 | } | |
7929 | ||
7930 | static bool is_kfunc_ret_null(struct bpf_kfunc_call_arg_meta *meta) | |
7931 | { | |
7932 | return meta->kfunc_flags & KF_RET_NULL; | |
7933 | } | |
7934 | ||
7935 | static bool is_kfunc_release(struct bpf_kfunc_call_arg_meta *meta) | |
7936 | { | |
7937 | return meta->kfunc_flags & KF_RELEASE; | |
7938 | } | |
7939 | ||
7940 | static bool is_kfunc_trusted_args(struct bpf_kfunc_call_arg_meta *meta) | |
7941 | { | |
7942 | return meta->kfunc_flags & KF_TRUSTED_ARGS; | |
7943 | } | |
7944 | ||
7945 | static bool is_kfunc_sleepable(struct bpf_kfunc_call_arg_meta *meta) | |
7946 | { | |
7947 | return meta->kfunc_flags & KF_SLEEPABLE; | |
7948 | } | |
7949 | ||
7950 | static bool is_kfunc_destructive(struct bpf_kfunc_call_arg_meta *meta) | |
7951 | { | |
7952 | return meta->kfunc_flags & KF_DESTRUCTIVE; | |
7953 | } | |
7954 | ||
7955 | static bool is_kfunc_arg_kptr_get(struct bpf_kfunc_call_arg_meta *meta, int arg) | |
7956 | { | |
7957 | return arg == 0 && (meta->kfunc_flags & KF_KPTR_GET); | |
7958 | } | |
7959 | ||
3f00c523 DV |
7960 | static bool is_trusted_reg(const struct bpf_reg_state *reg) |
7961 | { | |
7962 | /* A referenced register is always trusted. */ | |
7963 | if (reg->ref_obj_id) | |
7964 | return true; | |
7965 | ||
7966 | /* If a register is not referenced, it is trusted if it has either the | |
7967 | * MEM_ALLOC or PTR_TRUSTED type modifiers, and no others. Some of the | |
7968 | * other type modifiers may be safe, but we elect to take an opt-in | |
7969 | * approach here as some (e.g. PTR_UNTRUSTED and PTR_MAYBE_NULL) are | |
7970 | * not. | |
7971 | * | |
7972 | * Eventually, we should make PTR_TRUSTED the single source of truth | |
7973 | * for whether a register is trusted. | |
7974 | */ | |
7975 | return type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS && | |
7976 | !bpf_type_has_unsafe_modifiers(reg->type); | |
7977 | } | |
7978 | ||
a50388db KKD |
7979 | static bool __kfunc_param_match_suffix(const struct btf *btf, |
7980 | const struct btf_param *arg, | |
7981 | const char *suffix) | |
00b85860 | 7982 | { |
a50388db | 7983 | int suffix_len = strlen(suffix), len; |
00b85860 KKD |
7984 | const char *param_name; |
7985 | ||
00b85860 KKD |
7986 | /* In the future, this can be ported to use BTF tagging */ |
7987 | param_name = btf_name_by_offset(btf, arg->name_off); | |
7988 | if (str_is_empty(param_name)) | |
7989 | return false; | |
7990 | len = strlen(param_name); | |
a50388db | 7991 | if (len < suffix_len) |
00b85860 | 7992 | return false; |
a50388db KKD |
7993 | param_name += len - suffix_len; |
7994 | return !strncmp(param_name, suffix, suffix_len); | |
7995 | } | |
7996 | ||
7997 | static bool is_kfunc_arg_mem_size(const struct btf *btf, | |
7998 | const struct btf_param *arg, | |
7999 | const struct bpf_reg_state *reg) | |
8000 | { | |
8001 | const struct btf_type *t; | |
8002 | ||
8003 | t = btf_type_skip_modifiers(btf, arg->type, NULL); | |
8004 | if (!btf_type_is_scalar(t) || reg->type != SCALAR_VALUE) | |
00b85860 KKD |
8005 | return false; |
8006 | ||
a50388db KKD |
8007 | return __kfunc_param_match_suffix(btf, arg, "__sz"); |
8008 | } | |
8009 | ||
8010 | static bool is_kfunc_arg_constant(const struct btf *btf, const struct btf_param *arg) | |
8011 | { | |
8012 | return __kfunc_param_match_suffix(btf, arg, "__k"); | |
00b85860 KKD |
8013 | } |
8014 | ||
958cf2e2 KKD |
8015 | static bool is_kfunc_arg_ignore(const struct btf *btf, const struct btf_param *arg) |
8016 | { | |
8017 | return __kfunc_param_match_suffix(btf, arg, "__ign"); | |
8018 | } | |
8019 | ||
ac9f0605 KKD |
8020 | static bool is_kfunc_arg_alloc_obj(const struct btf *btf, const struct btf_param *arg) |
8021 | { | |
8022 | return __kfunc_param_match_suffix(btf, arg, "__alloc"); | |
8023 | } | |
8024 | ||
00b85860 KKD |
8025 | static bool is_kfunc_arg_scalar_with_name(const struct btf *btf, |
8026 | const struct btf_param *arg, | |
8027 | const char *name) | |
8028 | { | |
8029 | int len, target_len = strlen(name); | |
8030 | const char *param_name; | |
8031 | ||
8032 | param_name = btf_name_by_offset(btf, arg->name_off); | |
8033 | if (str_is_empty(param_name)) | |
8034 | return false; | |
8035 | len = strlen(param_name); | |
8036 | if (len != target_len) | |
8037 | return false; | |
8038 | if (strcmp(param_name, name)) | |
8039 | return false; | |
8040 | ||
8041 | return true; | |
8042 | } | |
8043 | ||
8044 | enum { | |
8045 | KF_ARG_DYNPTR_ID, | |
8cab76ec KKD |
8046 | KF_ARG_LIST_HEAD_ID, |
8047 | KF_ARG_LIST_NODE_ID, | |
00b85860 KKD |
8048 | }; |
8049 | ||
8050 | BTF_ID_LIST(kf_arg_btf_ids) | |
8051 | BTF_ID(struct, bpf_dynptr_kern) | |
8cab76ec KKD |
8052 | BTF_ID(struct, bpf_list_head) |
8053 | BTF_ID(struct, bpf_list_node) | |
00b85860 | 8054 | |
8cab76ec KKD |
8055 | static bool __is_kfunc_ptr_arg_type(const struct btf *btf, |
8056 | const struct btf_param *arg, int type) | |
00b85860 KKD |
8057 | { |
8058 | const struct btf_type *t; | |
8059 | u32 res_id; | |
8060 | ||
8061 | t = btf_type_skip_modifiers(btf, arg->type, NULL); | |
8062 | if (!t) | |
8063 | return false; | |
8064 | if (!btf_type_is_ptr(t)) | |
8065 | return false; | |
8066 | t = btf_type_skip_modifiers(btf, t->type, &res_id); | |
8067 | if (!t) | |
8068 | return false; | |
8cab76ec KKD |
8069 | return btf_types_are_same(btf, res_id, btf_vmlinux, kf_arg_btf_ids[type]); |
8070 | } | |
8071 | ||
8072 | static bool is_kfunc_arg_dynptr(const struct btf *btf, const struct btf_param *arg) | |
8073 | { | |
8074 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_DYNPTR_ID); | |
8075 | } | |
8076 | ||
8077 | static bool is_kfunc_arg_list_head(const struct btf *btf, const struct btf_param *arg) | |
8078 | { | |
8079 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_HEAD_ID); | |
8080 | } | |
8081 | ||
8082 | static bool is_kfunc_arg_list_node(const struct btf *btf, const struct btf_param *arg) | |
8083 | { | |
8084 | return __is_kfunc_ptr_arg_type(btf, arg, KF_ARG_LIST_NODE_ID); | |
00b85860 KKD |
8085 | } |
8086 | ||
8087 | /* Returns true if struct is composed of scalars, 4 levels of nesting allowed */ | |
8088 | static bool __btf_type_is_scalar_struct(struct bpf_verifier_env *env, | |
8089 | const struct btf *btf, | |
8090 | const struct btf_type *t, int rec) | |
8091 | { | |
8092 | const struct btf_type *member_type; | |
8093 | const struct btf_member *member; | |
8094 | u32 i; | |
8095 | ||
8096 | if (!btf_type_is_struct(t)) | |
8097 | return false; | |
8098 | ||
8099 | for_each_member(i, t, member) { | |
8100 | const struct btf_array *array; | |
8101 | ||
8102 | member_type = btf_type_skip_modifiers(btf, member->type, NULL); | |
8103 | if (btf_type_is_struct(member_type)) { | |
8104 | if (rec >= 3) { | |
8105 | verbose(env, "max struct nesting depth exceeded\n"); | |
8106 | return false; | |
8107 | } | |
8108 | if (!__btf_type_is_scalar_struct(env, btf, member_type, rec + 1)) | |
8109 | return false; | |
8110 | continue; | |
8111 | } | |
8112 | if (btf_type_is_array(member_type)) { | |
8113 | array = btf_array(member_type); | |
8114 | if (!array->nelems) | |
8115 | return false; | |
8116 | member_type = btf_type_skip_modifiers(btf, array->type, NULL); | |
8117 | if (!btf_type_is_scalar(member_type)) | |
8118 | return false; | |
8119 | continue; | |
8120 | } | |
8121 | if (!btf_type_is_scalar(member_type)) | |
8122 | return false; | |
8123 | } | |
8124 | return true; | |
8125 | } | |
8126 | ||
8127 | ||
8128 | static u32 *reg2btf_ids[__BPF_REG_TYPE_MAX] = { | |
8129 | #ifdef CONFIG_NET | |
8130 | [PTR_TO_SOCKET] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK], | |
8131 | [PTR_TO_SOCK_COMMON] = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
8132 | [PTR_TO_TCP_SOCK] = &btf_sock_ids[BTF_SOCK_TYPE_TCP], | |
8133 | #endif | |
8134 | }; | |
8135 | ||
8136 | enum kfunc_ptr_arg_type { | |
8137 | KF_ARG_PTR_TO_CTX, | |
ac9f0605 | 8138 | KF_ARG_PTR_TO_ALLOC_BTF_ID, /* Allocated object */ |
00b85860 KKD |
8139 | KF_ARG_PTR_TO_KPTR, /* PTR_TO_KPTR but type specific */ |
8140 | KF_ARG_PTR_TO_DYNPTR, | |
8cab76ec KKD |
8141 | KF_ARG_PTR_TO_LIST_HEAD, |
8142 | KF_ARG_PTR_TO_LIST_NODE, | |
00b85860 KKD |
8143 | KF_ARG_PTR_TO_BTF_ID, /* Also covers reg2btf_ids conversions */ |
8144 | KF_ARG_PTR_TO_MEM, | |
8145 | KF_ARG_PTR_TO_MEM_SIZE, /* Size derived from next argument, skip it */ | |
8146 | }; | |
8147 | ||
ac9f0605 KKD |
8148 | enum special_kfunc_type { |
8149 | KF_bpf_obj_new_impl, | |
8150 | KF_bpf_obj_drop_impl, | |
8cab76ec KKD |
8151 | KF_bpf_list_push_front, |
8152 | KF_bpf_list_push_back, | |
8153 | KF_bpf_list_pop_front, | |
8154 | KF_bpf_list_pop_back, | |
fd264ca0 | 8155 | KF_bpf_cast_to_kern_ctx, |
a35b9af4 | 8156 | KF_bpf_rdonly_cast, |
ac9f0605 KKD |
8157 | }; |
8158 | ||
8159 | BTF_SET_START(special_kfunc_set) | |
8160 | BTF_ID(func, bpf_obj_new_impl) | |
8161 | BTF_ID(func, bpf_obj_drop_impl) | |
8cab76ec KKD |
8162 | BTF_ID(func, bpf_list_push_front) |
8163 | BTF_ID(func, bpf_list_push_back) | |
8164 | BTF_ID(func, bpf_list_pop_front) | |
8165 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 8166 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 8167 | BTF_ID(func, bpf_rdonly_cast) |
ac9f0605 KKD |
8168 | BTF_SET_END(special_kfunc_set) |
8169 | ||
8170 | BTF_ID_LIST(special_kfunc_list) | |
8171 | BTF_ID(func, bpf_obj_new_impl) | |
8172 | BTF_ID(func, bpf_obj_drop_impl) | |
8cab76ec KKD |
8173 | BTF_ID(func, bpf_list_push_front) |
8174 | BTF_ID(func, bpf_list_push_back) | |
8175 | BTF_ID(func, bpf_list_pop_front) | |
8176 | BTF_ID(func, bpf_list_pop_back) | |
fd264ca0 | 8177 | BTF_ID(func, bpf_cast_to_kern_ctx) |
a35b9af4 | 8178 | BTF_ID(func, bpf_rdonly_cast) |
ac9f0605 | 8179 | |
00b85860 KKD |
8180 | static enum kfunc_ptr_arg_type |
8181 | get_kfunc_ptr_arg_type(struct bpf_verifier_env *env, | |
8182 | struct bpf_kfunc_call_arg_meta *meta, | |
8183 | const struct btf_type *t, const struct btf_type *ref_t, | |
8184 | const char *ref_tname, const struct btf_param *args, | |
8185 | int argno, int nargs) | |
8186 | { | |
8187 | u32 regno = argno + 1; | |
8188 | struct bpf_reg_state *regs = cur_regs(env); | |
8189 | struct bpf_reg_state *reg = ®s[regno]; | |
8190 | bool arg_mem_size = false; | |
8191 | ||
fd264ca0 YS |
8192 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) |
8193 | return KF_ARG_PTR_TO_CTX; | |
8194 | ||
00b85860 KKD |
8195 | /* In this function, we verify the kfunc's BTF as per the argument type, |
8196 | * leaving the rest of the verification with respect to the register | |
8197 | * type to our caller. When a set of conditions hold in the BTF type of | |
8198 | * arguments, we resolve it to a known kfunc_ptr_arg_type. | |
8199 | */ | |
8200 | if (btf_get_prog_ctx_type(&env->log, meta->btf, t, resolve_prog_type(env->prog), argno)) | |
8201 | return KF_ARG_PTR_TO_CTX; | |
8202 | ||
ac9f0605 KKD |
8203 | if (is_kfunc_arg_alloc_obj(meta->btf, &args[argno])) |
8204 | return KF_ARG_PTR_TO_ALLOC_BTF_ID; | |
8205 | ||
00b85860 KKD |
8206 | if (is_kfunc_arg_kptr_get(meta, argno)) { |
8207 | if (!btf_type_is_ptr(ref_t)) { | |
8208 | verbose(env, "arg#0 BTF type must be a double pointer for kptr_get kfunc\n"); | |
8209 | return -EINVAL; | |
8210 | } | |
8211 | ref_t = btf_type_by_id(meta->btf, ref_t->type); | |
8212 | ref_tname = btf_name_by_offset(meta->btf, ref_t->name_off); | |
8213 | if (!btf_type_is_struct(ref_t)) { | |
8214 | verbose(env, "kernel function %s args#0 pointer type %s %s is not supported\n", | |
8215 | meta->func_name, btf_type_str(ref_t), ref_tname); | |
8216 | return -EINVAL; | |
8217 | } | |
8218 | return KF_ARG_PTR_TO_KPTR; | |
8219 | } | |
8220 | ||
8221 | if (is_kfunc_arg_dynptr(meta->btf, &args[argno])) | |
8222 | return KF_ARG_PTR_TO_DYNPTR; | |
8223 | ||
8cab76ec KKD |
8224 | if (is_kfunc_arg_list_head(meta->btf, &args[argno])) |
8225 | return KF_ARG_PTR_TO_LIST_HEAD; | |
8226 | ||
8227 | if (is_kfunc_arg_list_node(meta->btf, &args[argno])) | |
8228 | return KF_ARG_PTR_TO_LIST_NODE; | |
8229 | ||
00b85860 KKD |
8230 | if ((base_type(reg->type) == PTR_TO_BTF_ID || reg2btf_ids[base_type(reg->type)])) { |
8231 | if (!btf_type_is_struct(ref_t)) { | |
8232 | verbose(env, "kernel function %s args#%d pointer type %s %s is not supported\n", | |
8233 | meta->func_name, argno, btf_type_str(ref_t), ref_tname); | |
8234 | return -EINVAL; | |
8235 | } | |
8236 | return KF_ARG_PTR_TO_BTF_ID; | |
8237 | } | |
8238 | ||
8239 | if (argno + 1 < nargs && is_kfunc_arg_mem_size(meta->btf, &args[argno + 1], ®s[regno + 1])) | |
8240 | arg_mem_size = true; | |
8241 | ||
8242 | /* This is the catch all argument type of register types supported by | |
8243 | * check_helper_mem_access. However, we only allow when argument type is | |
8244 | * pointer to scalar, or struct composed (recursively) of scalars. When | |
8245 | * arg_mem_size is true, the pointer can be void *. | |
8246 | */ | |
8247 | if (!btf_type_is_scalar(ref_t) && !__btf_type_is_scalar_struct(env, meta->btf, ref_t, 0) && | |
8248 | (arg_mem_size ? !btf_type_is_void(ref_t) : 1)) { | |
8249 | verbose(env, "arg#%d pointer type %s %s must point to %sscalar, or struct with scalar\n", | |
8250 | argno, btf_type_str(ref_t), ref_tname, arg_mem_size ? "void, " : ""); | |
8251 | return -EINVAL; | |
8252 | } | |
8253 | return arg_mem_size ? KF_ARG_PTR_TO_MEM_SIZE : KF_ARG_PTR_TO_MEM; | |
8254 | } | |
8255 | ||
8256 | static int process_kf_arg_ptr_to_btf_id(struct bpf_verifier_env *env, | |
8257 | struct bpf_reg_state *reg, | |
8258 | const struct btf_type *ref_t, | |
8259 | const char *ref_tname, u32 ref_id, | |
8260 | struct bpf_kfunc_call_arg_meta *meta, | |
8261 | int argno) | |
8262 | { | |
8263 | const struct btf_type *reg_ref_t; | |
8264 | bool strict_type_match = false; | |
8265 | const struct btf *reg_btf; | |
8266 | const char *reg_ref_tname; | |
8267 | u32 reg_ref_id; | |
8268 | ||
3f00c523 | 8269 | if (base_type(reg->type) == PTR_TO_BTF_ID) { |
00b85860 KKD |
8270 | reg_btf = reg->btf; |
8271 | reg_ref_id = reg->btf_id; | |
8272 | } else { | |
8273 | reg_btf = btf_vmlinux; | |
8274 | reg_ref_id = *reg2btf_ids[base_type(reg->type)]; | |
8275 | } | |
8276 | ||
8277 | if (is_kfunc_trusted_args(meta) || (is_kfunc_release(meta) && reg->ref_obj_id)) | |
8278 | strict_type_match = true; | |
8279 | ||
8280 | reg_ref_t = btf_type_skip_modifiers(reg_btf, reg_ref_id, ®_ref_id); | |
8281 | reg_ref_tname = btf_name_by_offset(reg_btf, reg_ref_t->name_off); | |
8282 | if (!btf_struct_ids_match(&env->log, reg_btf, reg_ref_id, reg->off, meta->btf, ref_id, strict_type_match)) { | |
8283 | verbose(env, "kernel function %s args#%d expected pointer to %s %s but R%d has a pointer to %s %s\n", | |
8284 | meta->func_name, argno, btf_type_str(ref_t), ref_tname, argno + 1, | |
8285 | btf_type_str(reg_ref_t), reg_ref_tname); | |
8286 | return -EINVAL; | |
8287 | } | |
8288 | return 0; | |
8289 | } | |
8290 | ||
8291 | static int process_kf_arg_ptr_to_kptr(struct bpf_verifier_env *env, | |
8292 | struct bpf_reg_state *reg, | |
8293 | const struct btf_type *ref_t, | |
8294 | const char *ref_tname, | |
8295 | struct bpf_kfunc_call_arg_meta *meta, | |
8296 | int argno) | |
8297 | { | |
8298 | struct btf_field *kptr_field; | |
8299 | ||
8300 | /* check_func_arg_reg_off allows var_off for | |
8301 | * PTR_TO_MAP_VALUE, but we need fixed offset to find | |
8302 | * off_desc. | |
8303 | */ | |
8304 | if (!tnum_is_const(reg->var_off)) { | |
8305 | verbose(env, "arg#0 must have constant offset\n"); | |
8306 | return -EINVAL; | |
8307 | } | |
8308 | ||
8309 | kptr_field = btf_record_find(reg->map_ptr->record, reg->off + reg->var_off.value, BPF_KPTR); | |
8310 | if (!kptr_field || kptr_field->type != BPF_KPTR_REF) { | |
8311 | verbose(env, "arg#0 no referenced kptr at map value offset=%llu\n", | |
8312 | reg->off + reg->var_off.value); | |
8313 | return -EINVAL; | |
8314 | } | |
8315 | ||
8316 | if (!btf_struct_ids_match(&env->log, meta->btf, ref_t->type, 0, kptr_field->kptr.btf, | |
8317 | kptr_field->kptr.btf_id, true)) { | |
8318 | verbose(env, "kernel function %s args#%d expected pointer to %s %s\n", | |
8319 | meta->func_name, argno, btf_type_str(ref_t), ref_tname); | |
8320 | return -EINVAL; | |
8321 | } | |
8322 | return 0; | |
8323 | } | |
8324 | ||
534e86bc KKD |
8325 | static int ref_set_release_on_unlock(struct bpf_verifier_env *env, u32 ref_obj_id) |
8326 | { | |
8327 | struct bpf_func_state *state = cur_func(env); | |
8328 | struct bpf_reg_state *reg; | |
8329 | int i; | |
8330 | ||
8331 | /* bpf_spin_lock only allows calling list_push and list_pop, no BPF | |
8332 | * subprogs, no global functions. This means that the references would | |
8333 | * not be released inside the critical section but they may be added to | |
8334 | * the reference state, and the acquired_refs are never copied out for a | |
8335 | * different frame as BPF to BPF calls don't work in bpf_spin_lock | |
8336 | * critical sections. | |
8337 | */ | |
8338 | if (!ref_obj_id) { | |
8339 | verbose(env, "verifier internal error: ref_obj_id is zero for release_on_unlock\n"); | |
8340 | return -EFAULT; | |
8341 | } | |
8342 | for (i = 0; i < state->acquired_refs; i++) { | |
8343 | if (state->refs[i].id == ref_obj_id) { | |
8344 | if (state->refs[i].release_on_unlock) { | |
8345 | verbose(env, "verifier internal error: expected false release_on_unlock"); | |
8346 | return -EFAULT; | |
8347 | } | |
8348 | state->refs[i].release_on_unlock = true; | |
8349 | /* Now mark everyone sharing same ref_obj_id as untrusted */ | |
8350 | bpf_for_each_reg_in_vstate(env->cur_state, state, reg, ({ | |
8351 | if (reg->ref_obj_id == ref_obj_id) | |
8352 | reg->type |= PTR_UNTRUSTED; | |
8353 | })); | |
8354 | return 0; | |
8355 | } | |
8356 | } | |
8357 | verbose(env, "verifier internal error: ref state missing for ref_obj_id\n"); | |
8358 | return -EFAULT; | |
8359 | } | |
8360 | ||
8cab76ec KKD |
8361 | /* Implementation details: |
8362 | * | |
8363 | * Each register points to some region of memory, which we define as an | |
8364 | * allocation. Each allocation may embed a bpf_spin_lock which protects any | |
8365 | * special BPF objects (bpf_list_head, bpf_rb_root, etc.) part of the same | |
8366 | * allocation. The lock and the data it protects are colocated in the same | |
8367 | * memory region. | |
8368 | * | |
8369 | * Hence, everytime a register holds a pointer value pointing to such | |
8370 | * allocation, the verifier preserves a unique reg->id for it. | |
8371 | * | |
8372 | * The verifier remembers the lock 'ptr' and the lock 'id' whenever | |
8373 | * bpf_spin_lock is called. | |
8374 | * | |
8375 | * To enable this, lock state in the verifier captures two values: | |
8376 | * active_lock.ptr = Register's type specific pointer | |
8377 | * active_lock.id = A unique ID for each register pointer value | |
8378 | * | |
8379 | * Currently, PTR_TO_MAP_VALUE and PTR_TO_BTF_ID | MEM_ALLOC are the two | |
8380 | * supported register types. | |
8381 | * | |
8382 | * The active_lock.ptr in case of map values is the reg->map_ptr, and in case of | |
8383 | * allocated objects is the reg->btf pointer. | |
8384 | * | |
8385 | * The active_lock.id is non-unique for maps supporting direct_value_addr, as we | |
8386 | * can establish the provenance of the map value statically for each distinct | |
8387 | * lookup into such maps. They always contain a single map value hence unique | |
8388 | * IDs for each pseudo load pessimizes the algorithm and rejects valid programs. | |
8389 | * | |
8390 | * So, in case of global variables, they use array maps with max_entries = 1, | |
8391 | * hence their active_lock.ptr becomes map_ptr and id = 0 (since they all point | |
8392 | * into the same map value as max_entries is 1, as described above). | |
8393 | * | |
8394 | * In case of inner map lookups, the inner map pointer has same map_ptr as the | |
8395 | * outer map pointer (in verifier context), but each lookup into an inner map | |
8396 | * assigns a fresh reg->id to the lookup, so while lookups into distinct inner | |
8397 | * maps from the same outer map share the same map_ptr as active_lock.ptr, they | |
8398 | * will get different reg->id assigned to each lookup, hence different | |
8399 | * active_lock.id. | |
8400 | * | |
8401 | * In case of allocated objects, active_lock.ptr is the reg->btf, and the | |
8402 | * reg->id is a unique ID preserved after the NULL pointer check on the pointer | |
8403 | * returned from bpf_obj_new. Each allocation receives a new reg->id. | |
8404 | */ | |
8405 | static int check_reg_allocation_locked(struct bpf_verifier_env *env, struct bpf_reg_state *reg) | |
8406 | { | |
8407 | void *ptr; | |
8408 | u32 id; | |
8409 | ||
8410 | switch ((int)reg->type) { | |
8411 | case PTR_TO_MAP_VALUE: | |
8412 | ptr = reg->map_ptr; | |
8413 | break; | |
8414 | case PTR_TO_BTF_ID | MEM_ALLOC: | |
3f00c523 | 8415 | case PTR_TO_BTF_ID | MEM_ALLOC | PTR_TRUSTED: |
8cab76ec KKD |
8416 | ptr = reg->btf; |
8417 | break; | |
8418 | default: | |
8419 | verbose(env, "verifier internal error: unknown reg type for lock check\n"); | |
8420 | return -EFAULT; | |
8421 | } | |
8422 | id = reg->id; | |
8423 | ||
8424 | if (!env->cur_state->active_lock.ptr) | |
8425 | return -EINVAL; | |
8426 | if (env->cur_state->active_lock.ptr != ptr || | |
8427 | env->cur_state->active_lock.id != id) { | |
8428 | verbose(env, "held lock and object are not in the same allocation\n"); | |
8429 | return -EINVAL; | |
8430 | } | |
8431 | return 0; | |
8432 | } | |
8433 | ||
8434 | static bool is_bpf_list_api_kfunc(u32 btf_id) | |
8435 | { | |
8436 | return btf_id == special_kfunc_list[KF_bpf_list_push_front] || | |
8437 | btf_id == special_kfunc_list[KF_bpf_list_push_back] || | |
8438 | btf_id == special_kfunc_list[KF_bpf_list_pop_front] || | |
8439 | btf_id == special_kfunc_list[KF_bpf_list_pop_back]; | |
8440 | } | |
8441 | ||
8442 | static int process_kf_arg_ptr_to_list_head(struct bpf_verifier_env *env, | |
8443 | struct bpf_reg_state *reg, u32 regno, | |
8444 | struct bpf_kfunc_call_arg_meta *meta) | |
8445 | { | |
8446 | struct btf_field *field; | |
8447 | struct btf_record *rec; | |
8448 | u32 list_head_off; | |
8449 | ||
8450 | if (meta->btf != btf_vmlinux || !is_bpf_list_api_kfunc(meta->func_id)) { | |
8451 | verbose(env, "verifier internal error: bpf_list_head argument for unknown kfunc\n"); | |
8452 | return -EFAULT; | |
8453 | } | |
8454 | ||
8455 | if (!tnum_is_const(reg->var_off)) { | |
8456 | verbose(env, | |
8457 | "R%d doesn't have constant offset. bpf_list_head has to be at the constant offset\n", | |
8458 | regno); | |
8459 | return -EINVAL; | |
8460 | } | |
8461 | ||
8462 | rec = reg_btf_record(reg); | |
8463 | list_head_off = reg->off + reg->var_off.value; | |
8464 | field = btf_record_find(rec, list_head_off, BPF_LIST_HEAD); | |
8465 | if (!field) { | |
8466 | verbose(env, "bpf_list_head not found at offset=%u\n", list_head_off); | |
8467 | return -EINVAL; | |
8468 | } | |
8469 | ||
8470 | /* All functions require bpf_list_head to be protected using a bpf_spin_lock */ | |
8471 | if (check_reg_allocation_locked(env, reg)) { | |
8472 | verbose(env, "bpf_spin_lock at off=%d must be held for bpf_list_head\n", | |
8473 | rec->spin_lock_off); | |
8474 | return -EINVAL; | |
8475 | } | |
8476 | ||
8477 | if (meta->arg_list_head.field) { | |
8478 | verbose(env, "verifier internal error: repeating bpf_list_head arg\n"); | |
8479 | return -EFAULT; | |
8480 | } | |
8481 | meta->arg_list_head.field = field; | |
8482 | return 0; | |
8483 | } | |
8484 | ||
8485 | static int process_kf_arg_ptr_to_list_node(struct bpf_verifier_env *env, | |
8486 | struct bpf_reg_state *reg, u32 regno, | |
8487 | struct bpf_kfunc_call_arg_meta *meta) | |
8488 | { | |
8489 | const struct btf_type *et, *t; | |
8490 | struct btf_field *field; | |
8491 | struct btf_record *rec; | |
8492 | u32 list_node_off; | |
8493 | ||
8494 | if (meta->btf != btf_vmlinux || | |
8495 | (meta->func_id != special_kfunc_list[KF_bpf_list_push_front] && | |
8496 | meta->func_id != special_kfunc_list[KF_bpf_list_push_back])) { | |
8497 | verbose(env, "verifier internal error: bpf_list_node argument for unknown kfunc\n"); | |
8498 | return -EFAULT; | |
8499 | } | |
8500 | ||
8501 | if (!tnum_is_const(reg->var_off)) { | |
8502 | verbose(env, | |
8503 | "R%d doesn't have constant offset. bpf_list_node has to be at the constant offset\n", | |
8504 | regno); | |
8505 | return -EINVAL; | |
8506 | } | |
8507 | ||
8508 | rec = reg_btf_record(reg); | |
8509 | list_node_off = reg->off + reg->var_off.value; | |
8510 | field = btf_record_find(rec, list_node_off, BPF_LIST_NODE); | |
8511 | if (!field || field->offset != list_node_off) { | |
8512 | verbose(env, "bpf_list_node not found at offset=%u\n", list_node_off); | |
8513 | return -EINVAL; | |
8514 | } | |
8515 | ||
8516 | field = meta->arg_list_head.field; | |
8517 | ||
8518 | et = btf_type_by_id(field->list_head.btf, field->list_head.value_btf_id); | |
8519 | t = btf_type_by_id(reg->btf, reg->btf_id); | |
8520 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, 0, field->list_head.btf, | |
8521 | field->list_head.value_btf_id, true)) { | |
8522 | verbose(env, "operation on bpf_list_head expects arg#1 bpf_list_node at offset=%d " | |
8523 | "in struct %s, but arg is at offset=%d in struct %s\n", | |
8524 | field->list_head.node_offset, btf_name_by_offset(field->list_head.btf, et->name_off), | |
8525 | list_node_off, btf_name_by_offset(reg->btf, t->name_off)); | |
8526 | return -EINVAL; | |
8527 | } | |
8528 | ||
8529 | if (list_node_off != field->list_head.node_offset) { | |
8530 | verbose(env, "arg#1 offset=%d, but expected bpf_list_node at offset=%d in struct %s\n", | |
8531 | list_node_off, field->list_head.node_offset, | |
8532 | btf_name_by_offset(field->list_head.btf, et->name_off)); | |
8533 | return -EINVAL; | |
8534 | } | |
534e86bc KKD |
8535 | /* Set arg#1 for expiration after unlock */ |
8536 | return ref_set_release_on_unlock(env, reg->ref_obj_id); | |
8cab76ec KKD |
8537 | } |
8538 | ||
00b85860 KKD |
8539 | static int check_kfunc_args(struct bpf_verifier_env *env, struct bpf_kfunc_call_arg_meta *meta) |
8540 | { | |
8541 | const char *func_name = meta->func_name, *ref_tname; | |
8542 | const struct btf *btf = meta->btf; | |
8543 | const struct btf_param *args; | |
8544 | u32 i, nargs; | |
8545 | int ret; | |
8546 | ||
8547 | args = (const struct btf_param *)(meta->func_proto + 1); | |
8548 | nargs = btf_type_vlen(meta->func_proto); | |
8549 | if (nargs > MAX_BPF_FUNC_REG_ARGS) { | |
8550 | verbose(env, "Function %s has %d > %d args\n", func_name, nargs, | |
8551 | MAX_BPF_FUNC_REG_ARGS); | |
8552 | return -EINVAL; | |
8553 | } | |
8554 | ||
8555 | /* Check that BTF function arguments match actual types that the | |
8556 | * verifier sees. | |
8557 | */ | |
8558 | for (i = 0; i < nargs; i++) { | |
8559 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[i + 1]; | |
8560 | const struct btf_type *t, *ref_t, *resolve_ret; | |
8561 | enum bpf_arg_type arg_type = ARG_DONTCARE; | |
8562 | u32 regno = i + 1, ref_id, type_size; | |
8563 | bool is_ret_buf_sz = false; | |
8564 | int kf_arg_type; | |
8565 | ||
8566 | t = btf_type_skip_modifiers(btf, args[i].type, NULL); | |
958cf2e2 KKD |
8567 | |
8568 | if (is_kfunc_arg_ignore(btf, &args[i])) | |
8569 | continue; | |
8570 | ||
00b85860 KKD |
8571 | if (btf_type_is_scalar(t)) { |
8572 | if (reg->type != SCALAR_VALUE) { | |
8573 | verbose(env, "R%d is not a scalar\n", regno); | |
8574 | return -EINVAL; | |
8575 | } | |
a50388db KKD |
8576 | |
8577 | if (is_kfunc_arg_constant(meta->btf, &args[i])) { | |
8578 | if (meta->arg_constant.found) { | |
8579 | verbose(env, "verifier internal error: only one constant argument permitted\n"); | |
8580 | return -EFAULT; | |
8581 | } | |
8582 | if (!tnum_is_const(reg->var_off)) { | |
8583 | verbose(env, "R%d must be a known constant\n", regno); | |
8584 | return -EINVAL; | |
8585 | } | |
8586 | ret = mark_chain_precision(env, regno); | |
8587 | if (ret < 0) | |
8588 | return ret; | |
8589 | meta->arg_constant.found = true; | |
8590 | meta->arg_constant.value = reg->var_off.value; | |
8591 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdonly_buf_size")) { | |
00b85860 KKD |
8592 | meta->r0_rdonly = true; |
8593 | is_ret_buf_sz = true; | |
8594 | } else if (is_kfunc_arg_scalar_with_name(btf, &args[i], "rdwr_buf_size")) { | |
8595 | is_ret_buf_sz = true; | |
8596 | } | |
8597 | ||
8598 | if (is_ret_buf_sz) { | |
8599 | if (meta->r0_size) { | |
8600 | verbose(env, "2 or more rdonly/rdwr_buf_size parameters for kfunc"); | |
8601 | return -EINVAL; | |
8602 | } | |
8603 | ||
8604 | if (!tnum_is_const(reg->var_off)) { | |
8605 | verbose(env, "R%d is not a const\n", regno); | |
8606 | return -EINVAL; | |
8607 | } | |
8608 | ||
8609 | meta->r0_size = reg->var_off.value; | |
8610 | ret = mark_chain_precision(env, regno); | |
8611 | if (ret) | |
8612 | return ret; | |
8613 | } | |
8614 | continue; | |
8615 | } | |
8616 | ||
8617 | if (!btf_type_is_ptr(t)) { | |
8618 | verbose(env, "Unrecognized arg#%d type %s\n", i, btf_type_str(t)); | |
8619 | return -EINVAL; | |
8620 | } | |
8621 | ||
8622 | if (reg->ref_obj_id) { | |
8623 | if (is_kfunc_release(meta) && meta->ref_obj_id) { | |
8624 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
8625 | regno, reg->ref_obj_id, | |
8626 | meta->ref_obj_id); | |
8627 | return -EFAULT; | |
8628 | } | |
8629 | meta->ref_obj_id = reg->ref_obj_id; | |
8630 | if (is_kfunc_release(meta)) | |
8631 | meta->release_regno = regno; | |
8632 | } | |
8633 | ||
8634 | ref_t = btf_type_skip_modifiers(btf, t->type, &ref_id); | |
8635 | ref_tname = btf_name_by_offset(btf, ref_t->name_off); | |
8636 | ||
8637 | kf_arg_type = get_kfunc_ptr_arg_type(env, meta, t, ref_t, ref_tname, args, i, nargs); | |
8638 | if (kf_arg_type < 0) | |
8639 | return kf_arg_type; | |
8640 | ||
8641 | switch (kf_arg_type) { | |
ac9f0605 | 8642 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
00b85860 KKD |
8643 | case KF_ARG_PTR_TO_BTF_ID: |
8644 | if (!is_kfunc_trusted_args(meta)) | |
8645 | break; | |
3f00c523 DV |
8646 | |
8647 | if (!is_trusted_reg(reg)) { | |
8648 | verbose(env, "R%d must be referenced or trusted\n", regno); | |
00b85860 KKD |
8649 | return -EINVAL; |
8650 | } | |
8651 | fallthrough; | |
8652 | case KF_ARG_PTR_TO_CTX: | |
8653 | /* Trusted arguments have the same offset checks as release arguments */ | |
8654 | arg_type |= OBJ_RELEASE; | |
8655 | break; | |
8656 | case KF_ARG_PTR_TO_KPTR: | |
8657 | case KF_ARG_PTR_TO_DYNPTR: | |
8cab76ec KKD |
8658 | case KF_ARG_PTR_TO_LIST_HEAD: |
8659 | case KF_ARG_PTR_TO_LIST_NODE: | |
00b85860 KKD |
8660 | case KF_ARG_PTR_TO_MEM: |
8661 | case KF_ARG_PTR_TO_MEM_SIZE: | |
8662 | /* Trusted by default */ | |
8663 | break; | |
8664 | default: | |
8665 | WARN_ON_ONCE(1); | |
8666 | return -EFAULT; | |
8667 | } | |
8668 | ||
8669 | if (is_kfunc_release(meta) && reg->ref_obj_id) | |
8670 | arg_type |= OBJ_RELEASE; | |
8671 | ret = check_func_arg_reg_off(env, reg, regno, arg_type); | |
8672 | if (ret < 0) | |
8673 | return ret; | |
8674 | ||
8675 | switch (kf_arg_type) { | |
8676 | case KF_ARG_PTR_TO_CTX: | |
8677 | if (reg->type != PTR_TO_CTX) { | |
8678 | verbose(env, "arg#%d expected pointer to ctx, but got %s\n", i, btf_type_str(t)); | |
8679 | return -EINVAL; | |
8680 | } | |
fd264ca0 YS |
8681 | |
8682 | if (meta->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { | |
8683 | ret = get_kern_ctx_btf_id(&env->log, resolve_prog_type(env->prog)); | |
8684 | if (ret < 0) | |
8685 | return -EINVAL; | |
8686 | meta->ret_btf_id = ret; | |
8687 | } | |
00b85860 | 8688 | break; |
ac9f0605 KKD |
8689 | case KF_ARG_PTR_TO_ALLOC_BTF_ID: |
8690 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
8691 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
8692 | return -EINVAL; | |
8693 | } | |
8694 | if (!reg->ref_obj_id) { | |
8695 | verbose(env, "allocated object must be referenced\n"); | |
8696 | return -EINVAL; | |
8697 | } | |
8698 | if (meta->btf == btf_vmlinux && | |
8699 | meta->func_id == special_kfunc_list[KF_bpf_obj_drop_impl]) { | |
8700 | meta->arg_obj_drop.btf = reg->btf; | |
8701 | meta->arg_obj_drop.btf_id = reg->btf_id; | |
8702 | } | |
8703 | break; | |
00b85860 KKD |
8704 | case KF_ARG_PTR_TO_KPTR: |
8705 | if (reg->type != PTR_TO_MAP_VALUE) { | |
8706 | verbose(env, "arg#0 expected pointer to map value\n"); | |
8707 | return -EINVAL; | |
8708 | } | |
8709 | ret = process_kf_arg_ptr_to_kptr(env, reg, ref_t, ref_tname, meta, i); | |
8710 | if (ret < 0) | |
8711 | return ret; | |
8712 | break; | |
8713 | case KF_ARG_PTR_TO_DYNPTR: | |
8714 | if (reg->type != PTR_TO_STACK) { | |
8715 | verbose(env, "arg#%d expected pointer to stack\n", i); | |
8716 | return -EINVAL; | |
8717 | } | |
8718 | ||
8719 | if (!is_dynptr_reg_valid_init(env, reg)) { | |
8720 | verbose(env, "arg#%d pointer type %s %s must be valid and initialized\n", | |
8721 | i, btf_type_str(ref_t), ref_tname); | |
8722 | return -EINVAL; | |
8723 | } | |
8724 | ||
8725 | if (!is_dynptr_type_expected(env, reg, ARG_PTR_TO_DYNPTR | DYNPTR_TYPE_LOCAL)) { | |
8726 | verbose(env, "arg#%d pointer type %s %s points to unsupported dynamic pointer type\n", | |
8727 | i, btf_type_str(ref_t), ref_tname); | |
8728 | return -EINVAL; | |
8729 | } | |
8730 | break; | |
8cab76ec KKD |
8731 | case KF_ARG_PTR_TO_LIST_HEAD: |
8732 | if (reg->type != PTR_TO_MAP_VALUE && | |
8733 | reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
8734 | verbose(env, "arg#%d expected pointer to map value or allocated object\n", i); | |
8735 | return -EINVAL; | |
8736 | } | |
8737 | if (reg->type == (PTR_TO_BTF_ID | MEM_ALLOC) && !reg->ref_obj_id) { | |
8738 | verbose(env, "allocated object must be referenced\n"); | |
8739 | return -EINVAL; | |
8740 | } | |
8741 | ret = process_kf_arg_ptr_to_list_head(env, reg, regno, meta); | |
8742 | if (ret < 0) | |
8743 | return ret; | |
8744 | break; | |
8745 | case KF_ARG_PTR_TO_LIST_NODE: | |
8746 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC)) { | |
8747 | verbose(env, "arg#%d expected pointer to allocated object\n", i); | |
8748 | return -EINVAL; | |
8749 | } | |
8750 | if (!reg->ref_obj_id) { | |
8751 | verbose(env, "allocated object must be referenced\n"); | |
8752 | return -EINVAL; | |
8753 | } | |
8754 | ret = process_kf_arg_ptr_to_list_node(env, reg, regno, meta); | |
8755 | if (ret < 0) | |
8756 | return ret; | |
8757 | break; | |
00b85860 KKD |
8758 | case KF_ARG_PTR_TO_BTF_ID: |
8759 | /* Only base_type is checked, further checks are done here */ | |
3f00c523 DV |
8760 | if ((base_type(reg->type) != PTR_TO_BTF_ID || |
8761 | bpf_type_has_unsafe_modifiers(reg->type)) && | |
8762 | !reg2btf_ids[base_type(reg->type)]) { | |
8763 | verbose(env, "arg#%d is %s ", i, reg_type_str(env, reg->type)); | |
8764 | verbose(env, "expected %s or socket\n", | |
8765 | reg_type_str(env, base_type(reg->type) | | |
8766 | (type_flag(reg->type) & BPF_REG_TRUSTED_MODIFIERS))); | |
00b85860 KKD |
8767 | return -EINVAL; |
8768 | } | |
8769 | ret = process_kf_arg_ptr_to_btf_id(env, reg, ref_t, ref_tname, ref_id, meta, i); | |
8770 | if (ret < 0) | |
8771 | return ret; | |
8772 | break; | |
8773 | case KF_ARG_PTR_TO_MEM: | |
8774 | resolve_ret = btf_resolve_size(btf, ref_t, &type_size); | |
8775 | if (IS_ERR(resolve_ret)) { | |
8776 | verbose(env, "arg#%d reference type('%s %s') size cannot be determined: %ld\n", | |
8777 | i, btf_type_str(ref_t), ref_tname, PTR_ERR(resolve_ret)); | |
8778 | return -EINVAL; | |
8779 | } | |
8780 | ret = check_mem_reg(env, reg, regno, type_size); | |
8781 | if (ret < 0) | |
8782 | return ret; | |
8783 | break; | |
8784 | case KF_ARG_PTR_TO_MEM_SIZE: | |
8785 | ret = check_kfunc_mem_size_reg(env, ®s[regno + 1], regno + 1); | |
8786 | if (ret < 0) { | |
8787 | verbose(env, "arg#%d arg#%d memory, len pair leads to invalid memory access\n", i, i + 1); | |
8788 | return ret; | |
8789 | } | |
8790 | /* Skip next '__sz' argument */ | |
8791 | i++; | |
8792 | break; | |
8793 | } | |
8794 | } | |
8795 | ||
8796 | if (is_kfunc_release(meta) && !meta->release_regno) { | |
8797 | verbose(env, "release kernel function %s expects refcounted PTR_TO_BTF_ID\n", | |
8798 | func_name); | |
8799 | return -EINVAL; | |
8800 | } | |
8801 | ||
8802 | return 0; | |
8803 | } | |
8804 | ||
5c073f26 KKD |
8805 | static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
8806 | int *insn_idx_p) | |
e6ac2450 MKL |
8807 | { |
8808 | const struct btf_type *t, *func, *func_proto, *ptr_type; | |
8809 | struct bpf_reg_state *regs = cur_regs(env); | |
8810 | const char *func_name, *ptr_type_name; | |
00b85860 | 8811 | struct bpf_kfunc_call_arg_meta meta; |
e6ac2450 | 8812 | u32 i, nargs, func_id, ptr_type_id; |
5c073f26 | 8813 | int err, insn_idx = *insn_idx_p; |
e6ac2450 | 8814 | const struct btf_param *args; |
a35b9af4 | 8815 | const struct btf_type *ret_t; |
2357672c | 8816 | struct btf *desc_btf; |
a4703e31 | 8817 | u32 *kfunc_flags; |
e6ac2450 | 8818 | |
a5d82727 KKD |
8819 | /* skip for now, but return error when we find this in fixup_kfunc_call */ |
8820 | if (!insn->imm) | |
8821 | return 0; | |
8822 | ||
43bf0878 | 8823 | desc_btf = find_kfunc_desc_btf(env, insn->off); |
2357672c KKD |
8824 | if (IS_ERR(desc_btf)) |
8825 | return PTR_ERR(desc_btf); | |
8826 | ||
e6ac2450 | 8827 | func_id = insn->imm; |
2357672c KKD |
8828 | func = btf_type_by_id(desc_btf, func_id); |
8829 | func_name = btf_name_by_offset(desc_btf, func->name_off); | |
8830 | func_proto = btf_type_by_id(desc_btf, func->type); | |
e6ac2450 | 8831 | |
a4703e31 KKD |
8832 | kfunc_flags = btf_kfunc_id_set_contains(desc_btf, resolve_prog_type(env->prog), func_id); |
8833 | if (!kfunc_flags) { | |
e6ac2450 MKL |
8834 | verbose(env, "calling kernel function %s is not allowed\n", |
8835 | func_name); | |
8836 | return -EACCES; | |
8837 | } | |
00b85860 KKD |
8838 | |
8839 | /* Prepare kfunc call metadata */ | |
8840 | memset(&meta, 0, sizeof(meta)); | |
8841 | meta.btf = desc_btf; | |
8842 | meta.func_id = func_id; | |
8843 | meta.kfunc_flags = *kfunc_flags; | |
8844 | meta.func_proto = func_proto; | |
8845 | meta.func_name = func_name; | |
8846 | ||
8847 | if (is_kfunc_destructive(&meta) && !capable(CAP_SYS_BOOT)) { | |
8848 | verbose(env, "destructive kfunc calls require CAP_SYS_BOOT capability\n"); | |
4dd48c6f AS |
8849 | return -EACCES; |
8850 | } | |
8851 | ||
00b85860 KKD |
8852 | if (is_kfunc_sleepable(&meta) && !env->prog->aux->sleepable) { |
8853 | verbose(env, "program must be sleepable to call sleepable kfunc %s\n", func_name); | |
8854 | return -EACCES; | |
8855 | } | |
eb1f7f71 | 8856 | |
e6ac2450 | 8857 | /* Check the arguments */ |
00b85860 | 8858 | err = check_kfunc_args(env, &meta); |
5c073f26 | 8859 | if (err < 0) |
e6ac2450 | 8860 | return err; |
5c073f26 | 8861 | /* In case of release function, we get register number of refcounted |
00b85860 | 8862 | * PTR_TO_BTF_ID in bpf_kfunc_arg_meta, do the release now. |
5c073f26 | 8863 | */ |
00b85860 KKD |
8864 | if (meta.release_regno) { |
8865 | err = release_reference(env, regs[meta.release_regno].ref_obj_id); | |
5c073f26 KKD |
8866 | if (err) { |
8867 | verbose(env, "kfunc %s#%d reference has not been acquired before\n", | |
8868 | func_name, func_id); | |
8869 | return err; | |
8870 | } | |
8871 | } | |
e6ac2450 MKL |
8872 | |
8873 | for (i = 0; i < CALLER_SAVED_REGS; i++) | |
8874 | mark_reg_not_init(env, regs, caller_saved[i]); | |
8875 | ||
8876 | /* Check return type */ | |
2357672c | 8877 | t = btf_type_skip_modifiers(desc_btf, func_proto->type, NULL); |
5c073f26 | 8878 | |
00b85860 | 8879 | if (is_kfunc_acquire(&meta) && !btf_type_is_struct_ptr(meta.btf, t)) { |
958cf2e2 KKD |
8880 | /* Only exception is bpf_obj_new_impl */ |
8881 | if (meta.btf != btf_vmlinux || meta.func_id != special_kfunc_list[KF_bpf_obj_new_impl]) { | |
8882 | verbose(env, "acquire kernel function does not return PTR_TO_BTF_ID\n"); | |
8883 | return -EINVAL; | |
8884 | } | |
5c073f26 KKD |
8885 | } |
8886 | ||
e6ac2450 MKL |
8887 | if (btf_type_is_scalar(t)) { |
8888 | mark_reg_unknown(env, regs, BPF_REG_0); | |
8889 | mark_btf_func_reg_size(env, BPF_REG_0, t->size); | |
8890 | } else if (btf_type_is_ptr(t)) { | |
958cf2e2 KKD |
8891 | ptr_type = btf_type_skip_modifiers(desc_btf, t->type, &ptr_type_id); |
8892 | ||
8893 | if (meta.btf == btf_vmlinux && btf_id_set_contains(&special_kfunc_set, meta.func_id)) { | |
8894 | if (meta.func_id == special_kfunc_list[KF_bpf_obj_new_impl]) { | |
958cf2e2 KKD |
8895 | struct btf *ret_btf; |
8896 | u32 ret_btf_id; | |
8897 | ||
e181d3f1 KKD |
8898 | if (unlikely(!bpf_global_ma_set)) |
8899 | return -ENOMEM; | |
8900 | ||
958cf2e2 KKD |
8901 | if (((u64)(u32)meta.arg_constant.value) != meta.arg_constant.value) { |
8902 | verbose(env, "local type ID argument must be in range [0, U32_MAX]\n"); | |
8903 | return -EINVAL; | |
8904 | } | |
8905 | ||
8906 | ret_btf = env->prog->aux->btf; | |
8907 | ret_btf_id = meta.arg_constant.value; | |
8908 | ||
8909 | /* This may be NULL due to user not supplying a BTF */ | |
8910 | if (!ret_btf) { | |
8911 | verbose(env, "bpf_obj_new requires prog BTF\n"); | |
8912 | return -EINVAL; | |
8913 | } | |
8914 | ||
8915 | ret_t = btf_type_by_id(ret_btf, ret_btf_id); | |
8916 | if (!ret_t || !__btf_type_is_struct(ret_t)) { | |
8917 | verbose(env, "bpf_obj_new type ID argument must be of a struct\n"); | |
8918 | return -EINVAL; | |
8919 | } | |
8920 | ||
8921 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8922 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
8923 | regs[BPF_REG_0].btf = ret_btf; | |
8924 | regs[BPF_REG_0].btf_id = ret_btf_id; | |
8925 | ||
8926 | env->insn_aux_data[insn_idx].obj_new_size = ret_t->size; | |
8927 | env->insn_aux_data[insn_idx].kptr_struct_meta = | |
8928 | btf_find_struct_meta(ret_btf, ret_btf_id); | |
ac9f0605 KKD |
8929 | } else if (meta.func_id == special_kfunc_list[KF_bpf_obj_drop_impl]) { |
8930 | env->insn_aux_data[insn_idx].kptr_struct_meta = | |
8931 | btf_find_struct_meta(meta.arg_obj_drop.btf, | |
8932 | meta.arg_obj_drop.btf_id); | |
8cab76ec KKD |
8933 | } else if (meta.func_id == special_kfunc_list[KF_bpf_list_pop_front] || |
8934 | meta.func_id == special_kfunc_list[KF_bpf_list_pop_back]) { | |
8935 | struct btf_field *field = meta.arg_list_head.field; | |
8936 | ||
8937 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8938 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | MEM_ALLOC; | |
8939 | regs[BPF_REG_0].btf = field->list_head.btf; | |
8940 | regs[BPF_REG_0].btf_id = field->list_head.value_btf_id; | |
8941 | regs[BPF_REG_0].off = field->list_head.node_offset; | |
fd264ca0 YS |
8942 | } else if (meta.func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx]) { |
8943 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8944 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_TRUSTED; | |
8945 | regs[BPF_REG_0].btf = desc_btf; | |
8946 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; | |
a35b9af4 YS |
8947 | } else if (meta.func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { |
8948 | ret_t = btf_type_by_id(desc_btf, meta.arg_constant.value); | |
8949 | if (!ret_t || !btf_type_is_struct(ret_t)) { | |
8950 | verbose(env, | |
8951 | "kfunc bpf_rdonly_cast type ID argument must be of a struct\n"); | |
8952 | return -EINVAL; | |
8953 | } | |
8954 | ||
8955 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8956 | regs[BPF_REG_0].type = PTR_TO_BTF_ID | PTR_UNTRUSTED; | |
8957 | regs[BPF_REG_0].btf = desc_btf; | |
8958 | regs[BPF_REG_0].btf_id = meta.arg_constant.value; | |
958cf2e2 KKD |
8959 | } else { |
8960 | verbose(env, "kernel function %s unhandled dynamic return type\n", | |
8961 | meta.func_name); | |
8962 | return -EFAULT; | |
8963 | } | |
8964 | } else if (!__btf_type_is_struct(ptr_type)) { | |
eb1f7f71 BT |
8965 | if (!meta.r0_size) { |
8966 | ptr_type_name = btf_name_by_offset(desc_btf, | |
8967 | ptr_type->name_off); | |
8968 | verbose(env, | |
8969 | "kernel function %s returns pointer type %s %s is not supported\n", | |
8970 | func_name, | |
8971 | btf_type_str(ptr_type), | |
8972 | ptr_type_name); | |
8973 | return -EINVAL; | |
8974 | } | |
8975 | ||
8976 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8977 | regs[BPF_REG_0].type = PTR_TO_MEM; | |
8978 | regs[BPF_REG_0].mem_size = meta.r0_size; | |
8979 | ||
8980 | if (meta.r0_rdonly) | |
8981 | regs[BPF_REG_0].type |= MEM_RDONLY; | |
8982 | ||
8983 | /* Ensures we don't access the memory after a release_reference() */ | |
8984 | if (meta.ref_obj_id) | |
8985 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
8986 | } else { | |
8987 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
8988 | regs[BPF_REG_0].btf = desc_btf; | |
8989 | regs[BPF_REG_0].type = PTR_TO_BTF_ID; | |
8990 | regs[BPF_REG_0].btf_id = ptr_type_id; | |
e6ac2450 | 8991 | } |
958cf2e2 | 8992 | |
00b85860 | 8993 | if (is_kfunc_ret_null(&meta)) { |
5c073f26 KKD |
8994 | regs[BPF_REG_0].type |= PTR_MAYBE_NULL; |
8995 | /* For mark_ptr_or_null_reg, see 93c230e3f5bd6 */ | |
8996 | regs[BPF_REG_0].id = ++env->id_gen; | |
8997 | } | |
e6ac2450 | 8998 | mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *)); |
00b85860 | 8999 | if (is_kfunc_acquire(&meta)) { |
5c073f26 KKD |
9000 | int id = acquire_reference_state(env, insn_idx); |
9001 | ||
9002 | if (id < 0) | |
9003 | return id; | |
00b85860 KKD |
9004 | if (is_kfunc_ret_null(&meta)) |
9005 | regs[BPF_REG_0].id = id; | |
5c073f26 KKD |
9006 | regs[BPF_REG_0].ref_obj_id = id; |
9007 | } | |
00b85860 KKD |
9008 | if (reg_may_point_to_spin_lock(®s[BPF_REG_0]) && !regs[BPF_REG_0].id) |
9009 | regs[BPF_REG_0].id = ++env->id_gen; | |
e6ac2450 MKL |
9010 | } /* else { add_kfunc_call() ensures it is btf_type_is_void(t) } */ |
9011 | ||
9012 | nargs = btf_type_vlen(func_proto); | |
9013 | args = (const struct btf_param *)(func_proto + 1); | |
9014 | for (i = 0; i < nargs; i++) { | |
9015 | u32 regno = i + 1; | |
9016 | ||
2357672c | 9017 | t = btf_type_skip_modifiers(desc_btf, args[i].type, NULL); |
e6ac2450 MKL |
9018 | if (btf_type_is_ptr(t)) |
9019 | mark_btf_func_reg_size(env, regno, sizeof(void *)); | |
9020 | else | |
9021 | /* scalar. ensured by btf_check_kfunc_arg_match() */ | |
9022 | mark_btf_func_reg_size(env, regno, t->size); | |
9023 | } | |
9024 | ||
9025 | return 0; | |
9026 | } | |
9027 | ||
b03c9f9f EC |
9028 | static bool signed_add_overflows(s64 a, s64 b) |
9029 | { | |
9030 | /* Do the add in u64, where overflow is well-defined */ | |
9031 | s64 res = (s64)((u64)a + (u64)b); | |
9032 | ||
9033 | if (b < 0) | |
9034 | return res > a; | |
9035 | return res < a; | |
9036 | } | |
9037 | ||
bc895e8b | 9038 | static bool signed_add32_overflows(s32 a, s32 b) |
3f50f132 JF |
9039 | { |
9040 | /* Do the add in u32, where overflow is well-defined */ | |
9041 | s32 res = (s32)((u32)a + (u32)b); | |
9042 | ||
9043 | if (b < 0) | |
9044 | return res > a; | |
9045 | return res < a; | |
9046 | } | |
9047 | ||
bc895e8b | 9048 | static bool signed_sub_overflows(s64 a, s64 b) |
b03c9f9f EC |
9049 | { |
9050 | /* Do the sub in u64, where overflow is well-defined */ | |
9051 | s64 res = (s64)((u64)a - (u64)b); | |
9052 | ||
9053 | if (b < 0) | |
9054 | return res < a; | |
9055 | return res > a; | |
969bf05e AS |
9056 | } |
9057 | ||
3f50f132 JF |
9058 | static bool signed_sub32_overflows(s32 a, s32 b) |
9059 | { | |
bc895e8b | 9060 | /* Do the sub in u32, where overflow is well-defined */ |
3f50f132 JF |
9061 | s32 res = (s32)((u32)a - (u32)b); |
9062 | ||
9063 | if (b < 0) | |
9064 | return res < a; | |
9065 | return res > a; | |
9066 | } | |
9067 | ||
bb7f0f98 AS |
9068 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
9069 | const struct bpf_reg_state *reg, | |
9070 | enum bpf_reg_type type) | |
9071 | { | |
9072 | bool known = tnum_is_const(reg->var_off); | |
9073 | s64 val = reg->var_off.value; | |
9074 | s64 smin = reg->smin_value; | |
9075 | ||
9076 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
9077 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
c25b2ae1 | 9078 | reg_type_str(env, type), val); |
bb7f0f98 AS |
9079 | return false; |
9080 | } | |
9081 | ||
9082 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
9083 | verbose(env, "%s pointer offset %d is not allowed\n", | |
c25b2ae1 | 9084 | reg_type_str(env, type), reg->off); |
bb7f0f98 AS |
9085 | return false; |
9086 | } | |
9087 | ||
9088 | if (smin == S64_MIN) { | |
9089 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
c25b2ae1 | 9090 | reg_type_str(env, type)); |
bb7f0f98 AS |
9091 | return false; |
9092 | } | |
9093 | ||
9094 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
9095 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
c25b2ae1 | 9096 | smin, reg_type_str(env, type)); |
bb7f0f98 AS |
9097 | return false; |
9098 | } | |
9099 | ||
9100 | return true; | |
9101 | } | |
9102 | ||
a6aaece0 DB |
9103 | enum { |
9104 | REASON_BOUNDS = -1, | |
9105 | REASON_TYPE = -2, | |
9106 | REASON_PATHS = -3, | |
9107 | REASON_LIMIT = -4, | |
9108 | REASON_STACK = -5, | |
9109 | }; | |
9110 | ||
979d63d5 | 9111 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, |
bb01a1bb | 9112 | u32 *alu_limit, bool mask_to_left) |
979d63d5 | 9113 | { |
7fedb63a | 9114 | u32 max = 0, ptr_limit = 0; |
979d63d5 DB |
9115 | |
9116 | switch (ptr_reg->type) { | |
9117 | case PTR_TO_STACK: | |
1b1597e6 | 9118 | /* Offset 0 is out-of-bounds, but acceptable start for the |
7fedb63a DB |
9119 | * left direction, see BPF_REG_FP. Also, unknown scalar |
9120 | * offset where we would need to deal with min/max bounds is | |
9121 | * currently prohibited for unprivileged. | |
1b1597e6 PK |
9122 | */ |
9123 | max = MAX_BPF_STACK + mask_to_left; | |
7fedb63a | 9124 | ptr_limit = -(ptr_reg->var_off.value + ptr_reg->off); |
b658bbb8 | 9125 | break; |
979d63d5 | 9126 | case PTR_TO_MAP_VALUE: |
1b1597e6 | 9127 | max = ptr_reg->map_ptr->value_size; |
7fedb63a DB |
9128 | ptr_limit = (mask_to_left ? |
9129 | ptr_reg->smin_value : | |
9130 | ptr_reg->umax_value) + ptr_reg->off; | |
b658bbb8 | 9131 | break; |
979d63d5 | 9132 | default: |
a6aaece0 | 9133 | return REASON_TYPE; |
979d63d5 | 9134 | } |
b658bbb8 DB |
9135 | |
9136 | if (ptr_limit >= max) | |
a6aaece0 | 9137 | return REASON_LIMIT; |
b658bbb8 DB |
9138 | *alu_limit = ptr_limit; |
9139 | return 0; | |
979d63d5 DB |
9140 | } |
9141 | ||
d3bd7413 DB |
9142 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
9143 | const struct bpf_insn *insn) | |
9144 | { | |
2c78ee89 | 9145 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
9146 | } |
9147 | ||
9148 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
9149 | u32 alu_state, u32 alu_limit) | |
9150 | { | |
9151 | /* If we arrived here from different branches with different | |
9152 | * state or limits to sanitize, then this won't work. | |
9153 | */ | |
9154 | if (aux->alu_state && | |
9155 | (aux->alu_state != alu_state || | |
9156 | aux->alu_limit != alu_limit)) | |
a6aaece0 | 9157 | return REASON_PATHS; |
d3bd7413 | 9158 | |
e6ac5933 | 9159 | /* Corresponding fixup done in do_misc_fixups(). */ |
d3bd7413 DB |
9160 | aux->alu_state = alu_state; |
9161 | aux->alu_limit = alu_limit; | |
9162 | return 0; | |
9163 | } | |
9164 | ||
9165 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
9166 | struct bpf_insn *insn) | |
9167 | { | |
9168 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
9169 | ||
9170 | if (can_skip_alu_sanitation(env, insn)) | |
9171 | return 0; | |
9172 | ||
9173 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
9174 | } | |
9175 | ||
f5288193 DB |
9176 | static bool sanitize_needed(u8 opcode) |
9177 | { | |
9178 | return opcode == BPF_ADD || opcode == BPF_SUB; | |
9179 | } | |
9180 | ||
3d0220f6 DB |
9181 | struct bpf_sanitize_info { |
9182 | struct bpf_insn_aux_data aux; | |
bb01a1bb | 9183 | bool mask_to_left; |
3d0220f6 DB |
9184 | }; |
9185 | ||
9183671a DB |
9186 | static struct bpf_verifier_state * |
9187 | sanitize_speculative_path(struct bpf_verifier_env *env, | |
9188 | const struct bpf_insn *insn, | |
9189 | u32 next_idx, u32 curr_idx) | |
9190 | { | |
9191 | struct bpf_verifier_state *branch; | |
9192 | struct bpf_reg_state *regs; | |
9193 | ||
9194 | branch = push_stack(env, next_idx, curr_idx, true); | |
9195 | if (branch && insn) { | |
9196 | regs = branch->frame[branch->curframe]->regs; | |
9197 | if (BPF_SRC(insn->code) == BPF_K) { | |
9198 | mark_reg_unknown(env, regs, insn->dst_reg); | |
9199 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
9200 | mark_reg_unknown(env, regs, insn->dst_reg); | |
9201 | mark_reg_unknown(env, regs, insn->src_reg); | |
9202 | } | |
9203 | } | |
9204 | return branch; | |
9205 | } | |
9206 | ||
979d63d5 DB |
9207 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
9208 | struct bpf_insn *insn, | |
9209 | const struct bpf_reg_state *ptr_reg, | |
6f55b2f2 | 9210 | const struct bpf_reg_state *off_reg, |
979d63d5 | 9211 | struct bpf_reg_state *dst_reg, |
3d0220f6 | 9212 | struct bpf_sanitize_info *info, |
7fedb63a | 9213 | const bool commit_window) |
979d63d5 | 9214 | { |
3d0220f6 | 9215 | struct bpf_insn_aux_data *aux = commit_window ? cur_aux(env) : &info->aux; |
979d63d5 | 9216 | struct bpf_verifier_state *vstate = env->cur_state; |
801c6058 | 9217 | bool off_is_imm = tnum_is_const(off_reg->var_off); |
6f55b2f2 | 9218 | bool off_is_neg = off_reg->smin_value < 0; |
979d63d5 DB |
9219 | bool ptr_is_dst_reg = ptr_reg == dst_reg; |
9220 | u8 opcode = BPF_OP(insn->code); | |
9221 | u32 alu_state, alu_limit; | |
9222 | struct bpf_reg_state tmp; | |
9223 | bool ret; | |
f232326f | 9224 | int err; |
979d63d5 | 9225 | |
d3bd7413 | 9226 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
9227 | return 0; |
9228 | ||
9229 | /* We already marked aux for masking from non-speculative | |
9230 | * paths, thus we got here in the first place. We only care | |
9231 | * to explore bad access from here. | |
9232 | */ | |
9233 | if (vstate->speculative) | |
9234 | goto do_sim; | |
9235 | ||
bb01a1bb DB |
9236 | if (!commit_window) { |
9237 | if (!tnum_is_const(off_reg->var_off) && | |
9238 | (off_reg->smin_value < 0) != (off_reg->smax_value < 0)) | |
9239 | return REASON_BOUNDS; | |
9240 | ||
9241 | info->mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
9242 | (opcode == BPF_SUB && !off_is_neg); | |
9243 | } | |
9244 | ||
9245 | err = retrieve_ptr_limit(ptr_reg, &alu_limit, info->mask_to_left); | |
f232326f PK |
9246 | if (err < 0) |
9247 | return err; | |
9248 | ||
7fedb63a DB |
9249 | if (commit_window) { |
9250 | /* In commit phase we narrow the masking window based on | |
9251 | * the observed pointer move after the simulated operation. | |
9252 | */ | |
3d0220f6 DB |
9253 | alu_state = info->aux.alu_state; |
9254 | alu_limit = abs(info->aux.alu_limit - alu_limit); | |
7fedb63a DB |
9255 | } else { |
9256 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
801c6058 | 9257 | alu_state |= off_is_imm ? BPF_ALU_IMMEDIATE : 0; |
7fedb63a DB |
9258 | alu_state |= ptr_is_dst_reg ? |
9259 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
e042aa53 DB |
9260 | |
9261 | /* Limit pruning on unknown scalars to enable deep search for | |
9262 | * potential masking differences from other program paths. | |
9263 | */ | |
9264 | if (!off_is_imm) | |
9265 | env->explore_alu_limits = true; | |
7fedb63a DB |
9266 | } |
9267 | ||
f232326f PK |
9268 | err = update_alu_sanitation_state(aux, alu_state, alu_limit); |
9269 | if (err < 0) | |
9270 | return err; | |
979d63d5 | 9271 | do_sim: |
7fedb63a DB |
9272 | /* If we're in commit phase, we're done here given we already |
9273 | * pushed the truncated dst_reg into the speculative verification | |
9274 | * stack. | |
a7036191 DB |
9275 | * |
9276 | * Also, when register is a known constant, we rewrite register-based | |
9277 | * operation to immediate-based, and thus do not need masking (and as | |
9278 | * a consequence, do not need to simulate the zero-truncation either). | |
7fedb63a | 9279 | */ |
a7036191 | 9280 | if (commit_window || off_is_imm) |
7fedb63a DB |
9281 | return 0; |
9282 | ||
979d63d5 DB |
9283 | /* Simulate and find potential out-of-bounds access under |
9284 | * speculative execution from truncation as a result of | |
9285 | * masking when off was not within expected range. If off | |
9286 | * sits in dst, then we temporarily need to move ptr there | |
9287 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
9288 | * for cases where we use K-based arithmetic in one direction | |
9289 | * and truncated reg-based in the other in order to explore | |
9290 | * bad access. | |
9291 | */ | |
9292 | if (!ptr_is_dst_reg) { | |
9293 | tmp = *dst_reg; | |
9294 | *dst_reg = *ptr_reg; | |
9295 | } | |
9183671a DB |
9296 | ret = sanitize_speculative_path(env, NULL, env->insn_idx + 1, |
9297 | env->insn_idx); | |
0803278b | 9298 | if (!ptr_is_dst_reg && ret) |
979d63d5 | 9299 | *dst_reg = tmp; |
a6aaece0 DB |
9300 | return !ret ? REASON_STACK : 0; |
9301 | } | |
9302 | ||
fe9a5ca7 DB |
9303 | static void sanitize_mark_insn_seen(struct bpf_verifier_env *env) |
9304 | { | |
9305 | struct bpf_verifier_state *vstate = env->cur_state; | |
9306 | ||
9307 | /* If we simulate paths under speculation, we don't update the | |
9308 | * insn as 'seen' such that when we verify unreachable paths in | |
9309 | * the non-speculative domain, sanitize_dead_code() can still | |
9310 | * rewrite/sanitize them. | |
9311 | */ | |
9312 | if (!vstate->speculative) | |
9313 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; | |
9314 | } | |
9315 | ||
a6aaece0 DB |
9316 | static int sanitize_err(struct bpf_verifier_env *env, |
9317 | const struct bpf_insn *insn, int reason, | |
9318 | const struct bpf_reg_state *off_reg, | |
9319 | const struct bpf_reg_state *dst_reg) | |
9320 | { | |
9321 | static const char *err = "pointer arithmetic with it prohibited for !root"; | |
9322 | const char *op = BPF_OP(insn->code) == BPF_ADD ? "add" : "sub"; | |
9323 | u32 dst = insn->dst_reg, src = insn->src_reg; | |
9324 | ||
9325 | switch (reason) { | |
9326 | case REASON_BOUNDS: | |
9327 | verbose(env, "R%d has unknown scalar with mixed signed bounds, %s\n", | |
9328 | off_reg == dst_reg ? dst : src, err); | |
9329 | break; | |
9330 | case REASON_TYPE: | |
9331 | verbose(env, "R%d has pointer with unsupported alu operation, %s\n", | |
9332 | off_reg == dst_reg ? src : dst, err); | |
9333 | break; | |
9334 | case REASON_PATHS: | |
9335 | verbose(env, "R%d tried to %s from different maps, paths or scalars, %s\n", | |
9336 | dst, op, err); | |
9337 | break; | |
9338 | case REASON_LIMIT: | |
9339 | verbose(env, "R%d tried to %s beyond pointer bounds, %s\n", | |
9340 | dst, op, err); | |
9341 | break; | |
9342 | case REASON_STACK: | |
9343 | verbose(env, "R%d could not be pushed for speculative verification, %s\n", | |
9344 | dst, err); | |
9345 | break; | |
9346 | default: | |
9347 | verbose(env, "verifier internal error: unknown reason (%d)\n", | |
9348 | reason); | |
9349 | break; | |
9350 | } | |
9351 | ||
9352 | return -EACCES; | |
979d63d5 DB |
9353 | } |
9354 | ||
01f810ac AM |
9355 | /* check that stack access falls within stack limits and that 'reg' doesn't |
9356 | * have a variable offset. | |
9357 | * | |
9358 | * Variable offset is prohibited for unprivileged mode for simplicity since it | |
9359 | * requires corresponding support in Spectre masking for stack ALU. See also | |
9360 | * retrieve_ptr_limit(). | |
9361 | * | |
9362 | * | |
9363 | * 'off' includes 'reg->off'. | |
9364 | */ | |
9365 | static int check_stack_access_for_ptr_arithmetic( | |
9366 | struct bpf_verifier_env *env, | |
9367 | int regno, | |
9368 | const struct bpf_reg_state *reg, | |
9369 | int off) | |
9370 | { | |
9371 | if (!tnum_is_const(reg->var_off)) { | |
9372 | char tn_buf[48]; | |
9373 | ||
9374 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
9375 | verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n", | |
9376 | regno, tn_buf, off); | |
9377 | return -EACCES; | |
9378 | } | |
9379 | ||
9380 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
9381 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
9382 | "prohibited for !root; off=%d\n", regno, off); | |
9383 | return -EACCES; | |
9384 | } | |
9385 | ||
9386 | return 0; | |
9387 | } | |
9388 | ||
073815b7 DB |
9389 | static int sanitize_check_bounds(struct bpf_verifier_env *env, |
9390 | const struct bpf_insn *insn, | |
9391 | const struct bpf_reg_state *dst_reg) | |
9392 | { | |
9393 | u32 dst = insn->dst_reg; | |
9394 | ||
9395 | /* For unprivileged we require that resulting offset must be in bounds | |
9396 | * in order to be able to sanitize access later on. | |
9397 | */ | |
9398 | if (env->bypass_spec_v1) | |
9399 | return 0; | |
9400 | ||
9401 | switch (dst_reg->type) { | |
9402 | case PTR_TO_STACK: | |
9403 | if (check_stack_access_for_ptr_arithmetic(env, dst, dst_reg, | |
9404 | dst_reg->off + dst_reg->var_off.value)) | |
9405 | return -EACCES; | |
9406 | break; | |
9407 | case PTR_TO_MAP_VALUE: | |
61df10c7 | 9408 | if (check_map_access(env, dst, dst_reg->off, 1, false, ACCESS_HELPER)) { |
073815b7 DB |
9409 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " |
9410 | "prohibited for !root\n", dst); | |
9411 | return -EACCES; | |
9412 | } | |
9413 | break; | |
9414 | default: | |
9415 | break; | |
9416 | } | |
9417 | ||
9418 | return 0; | |
9419 | } | |
01f810ac | 9420 | |
f1174f77 | 9421 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
9422 | * Caller should also handle BPF_MOV case separately. |
9423 | * If we return -EACCES, caller may want to try again treating pointer as a | |
9424 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
9425 | */ | |
9426 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
9427 | struct bpf_insn *insn, | |
9428 | const struct bpf_reg_state *ptr_reg, | |
9429 | const struct bpf_reg_state *off_reg) | |
969bf05e | 9430 | { |
f4d7e40a AS |
9431 | struct bpf_verifier_state *vstate = env->cur_state; |
9432 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
9433 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 9434 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
9435 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
9436 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
9437 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
9438 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
3d0220f6 | 9439 | struct bpf_sanitize_info info = {}; |
969bf05e | 9440 | u8 opcode = BPF_OP(insn->code); |
24c109bb | 9441 | u32 dst = insn->dst_reg; |
979d63d5 | 9442 | int ret; |
969bf05e | 9443 | |
f1174f77 | 9444 | dst_reg = ®s[dst]; |
969bf05e | 9445 | |
6f16101e DB |
9446 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
9447 | smin_val > smax_val || umin_val > umax_val) { | |
9448 | /* Taint dst register if offset had invalid bounds derived from | |
9449 | * e.g. dead branches. | |
9450 | */ | |
f54c7898 | 9451 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 9452 | return 0; |
f1174f77 EC |
9453 | } |
9454 | ||
9455 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
9456 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
6c693541 YS |
9457 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
9458 | __mark_reg_unknown(env, dst_reg); | |
9459 | return 0; | |
9460 | } | |
9461 | ||
82abbf8d AS |
9462 | verbose(env, |
9463 | "R%d 32-bit pointer arithmetic prohibited\n", | |
9464 | dst); | |
f1174f77 | 9465 | return -EACCES; |
969bf05e AS |
9466 | } |
9467 | ||
c25b2ae1 | 9468 | if (ptr_reg->type & PTR_MAYBE_NULL) { |
aad2eeaf | 9469 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", |
c25b2ae1 | 9470 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 9471 | return -EACCES; |
c25b2ae1 HL |
9472 | } |
9473 | ||
9474 | switch (base_type(ptr_reg->type)) { | |
aad2eeaf | 9475 | case CONST_PTR_TO_MAP: |
7c696732 YS |
9476 | /* smin_val represents the known value */ |
9477 | if (known && smin_val == 0 && opcode == BPF_ADD) | |
9478 | break; | |
8731745e | 9479 | fallthrough; |
aad2eeaf | 9480 | case PTR_TO_PACKET_END: |
c64b7983 | 9481 | case PTR_TO_SOCKET: |
46f8bc92 | 9482 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 9483 | case PTR_TO_TCP_SOCK: |
fada7fdc | 9484 | case PTR_TO_XDP_SOCK: |
aad2eeaf | 9485 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
c25b2ae1 | 9486 | dst, reg_type_str(env, ptr_reg->type)); |
f1174f77 | 9487 | return -EACCES; |
aad2eeaf JS |
9488 | default: |
9489 | break; | |
f1174f77 EC |
9490 | } |
9491 | ||
9492 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
9493 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 9494 | */ |
f1174f77 EC |
9495 | dst_reg->type = ptr_reg->type; |
9496 | dst_reg->id = ptr_reg->id; | |
969bf05e | 9497 | |
bb7f0f98 AS |
9498 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
9499 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
9500 | return -EINVAL; | |
9501 | ||
3f50f132 JF |
9502 | /* pointer types do not carry 32-bit bounds at the moment. */ |
9503 | __mark_reg32_unbounded(dst_reg); | |
9504 | ||
7fedb63a DB |
9505 | if (sanitize_needed(opcode)) { |
9506 | ret = sanitize_ptr_alu(env, insn, ptr_reg, off_reg, dst_reg, | |
3d0220f6 | 9507 | &info, false); |
a6aaece0 DB |
9508 | if (ret < 0) |
9509 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
7fedb63a | 9510 | } |
a6aaece0 | 9511 | |
f1174f77 EC |
9512 | switch (opcode) { |
9513 | case BPF_ADD: | |
9514 | /* We can take a fixed offset as long as it doesn't overflow | |
9515 | * the s32 'off' field | |
969bf05e | 9516 | */ |
b03c9f9f EC |
9517 | if (known && (ptr_reg->off + smin_val == |
9518 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 9519 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
9520 | dst_reg->smin_value = smin_ptr; |
9521 | dst_reg->smax_value = smax_ptr; | |
9522 | dst_reg->umin_value = umin_ptr; | |
9523 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 9524 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 9525 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 9526 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
9527 | break; |
9528 | } | |
f1174f77 EC |
9529 | /* A new variable offset is created. Note that off_reg->off |
9530 | * == 0, since it's a scalar. | |
9531 | * dst_reg gets the pointer type and since some positive | |
9532 | * integer value was added to the pointer, give it a new 'id' | |
9533 | * if it's a PTR_TO_PACKET. | |
9534 | * this creates a new 'base' pointer, off_reg (variable) gets | |
9535 | * added into the variable offset, and we copy the fixed offset | |
9536 | * from ptr_reg. | |
969bf05e | 9537 | */ |
b03c9f9f EC |
9538 | if (signed_add_overflows(smin_ptr, smin_val) || |
9539 | signed_add_overflows(smax_ptr, smax_val)) { | |
9540 | dst_reg->smin_value = S64_MIN; | |
9541 | dst_reg->smax_value = S64_MAX; | |
9542 | } else { | |
9543 | dst_reg->smin_value = smin_ptr + smin_val; | |
9544 | dst_reg->smax_value = smax_ptr + smax_val; | |
9545 | } | |
9546 | if (umin_ptr + umin_val < umin_ptr || | |
9547 | umax_ptr + umax_val < umax_ptr) { | |
9548 | dst_reg->umin_value = 0; | |
9549 | dst_reg->umax_value = U64_MAX; | |
9550 | } else { | |
9551 | dst_reg->umin_value = umin_ptr + umin_val; | |
9552 | dst_reg->umax_value = umax_ptr + umax_val; | |
9553 | } | |
f1174f77 EC |
9554 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
9555 | dst_reg->off = ptr_reg->off; | |
0962590e | 9556 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 9557 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
9558 | dst_reg->id = ++env->id_gen; |
9559 | /* something was added to pkt_ptr, set range to zero */ | |
22dc4a0f | 9560 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
f1174f77 EC |
9561 | } |
9562 | break; | |
9563 | case BPF_SUB: | |
9564 | if (dst_reg == off_reg) { | |
9565 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
9566 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
9567 | dst); | |
f1174f77 EC |
9568 | return -EACCES; |
9569 | } | |
9570 | /* We don't allow subtraction from FP, because (according to | |
9571 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
9572 | * be able to deal with it. | |
969bf05e | 9573 | */ |
f1174f77 | 9574 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
9575 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
9576 | dst); | |
f1174f77 EC |
9577 | return -EACCES; |
9578 | } | |
b03c9f9f EC |
9579 | if (known && (ptr_reg->off - smin_val == |
9580 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 9581 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
9582 | dst_reg->smin_value = smin_ptr; |
9583 | dst_reg->smax_value = smax_ptr; | |
9584 | dst_reg->umin_value = umin_ptr; | |
9585 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
9586 | dst_reg->var_off = ptr_reg->var_off; |
9587 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 9588 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 9589 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
9590 | break; |
9591 | } | |
f1174f77 EC |
9592 | /* A new variable offset is created. If the subtrahend is known |
9593 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 9594 | */ |
b03c9f9f EC |
9595 | if (signed_sub_overflows(smin_ptr, smax_val) || |
9596 | signed_sub_overflows(smax_ptr, smin_val)) { | |
9597 | /* Overflow possible, we know nothing */ | |
9598 | dst_reg->smin_value = S64_MIN; | |
9599 | dst_reg->smax_value = S64_MAX; | |
9600 | } else { | |
9601 | dst_reg->smin_value = smin_ptr - smax_val; | |
9602 | dst_reg->smax_value = smax_ptr - smin_val; | |
9603 | } | |
9604 | if (umin_ptr < umax_val) { | |
9605 | /* Overflow possible, we know nothing */ | |
9606 | dst_reg->umin_value = 0; | |
9607 | dst_reg->umax_value = U64_MAX; | |
9608 | } else { | |
9609 | /* Cannot overflow (as long as bounds are consistent) */ | |
9610 | dst_reg->umin_value = umin_ptr - umax_val; | |
9611 | dst_reg->umax_value = umax_ptr - umin_val; | |
9612 | } | |
f1174f77 EC |
9613 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
9614 | dst_reg->off = ptr_reg->off; | |
0962590e | 9615 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 9616 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
9617 | dst_reg->id = ++env->id_gen; |
9618 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 9619 | if (smin_val < 0) |
22dc4a0f | 9620 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
43188702 | 9621 | } |
f1174f77 EC |
9622 | break; |
9623 | case BPF_AND: | |
9624 | case BPF_OR: | |
9625 | case BPF_XOR: | |
82abbf8d AS |
9626 | /* bitwise ops on pointers are troublesome, prohibit. */ |
9627 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
9628 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
9629 | return -EACCES; |
9630 | default: | |
9631 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
9632 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
9633 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 9634 | return -EACCES; |
43188702 JF |
9635 | } |
9636 | ||
bb7f0f98 AS |
9637 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
9638 | return -EINVAL; | |
3844d153 | 9639 | reg_bounds_sync(dst_reg); |
073815b7 DB |
9640 | if (sanitize_check_bounds(env, insn, dst_reg) < 0) |
9641 | return -EACCES; | |
7fedb63a DB |
9642 | if (sanitize_needed(opcode)) { |
9643 | ret = sanitize_ptr_alu(env, insn, dst_reg, off_reg, dst_reg, | |
3d0220f6 | 9644 | &info, true); |
7fedb63a DB |
9645 | if (ret < 0) |
9646 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
0d6303db DB |
9647 | } |
9648 | ||
43188702 JF |
9649 | return 0; |
9650 | } | |
9651 | ||
3f50f132 JF |
9652 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
9653 | struct bpf_reg_state *src_reg) | |
9654 | { | |
9655 | s32 smin_val = src_reg->s32_min_value; | |
9656 | s32 smax_val = src_reg->s32_max_value; | |
9657 | u32 umin_val = src_reg->u32_min_value; | |
9658 | u32 umax_val = src_reg->u32_max_value; | |
9659 | ||
9660 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
9661 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
9662 | dst_reg->s32_min_value = S32_MIN; | |
9663 | dst_reg->s32_max_value = S32_MAX; | |
9664 | } else { | |
9665 | dst_reg->s32_min_value += smin_val; | |
9666 | dst_reg->s32_max_value += smax_val; | |
9667 | } | |
9668 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
9669 | dst_reg->u32_max_value + umax_val < umax_val) { | |
9670 | dst_reg->u32_min_value = 0; | |
9671 | dst_reg->u32_max_value = U32_MAX; | |
9672 | } else { | |
9673 | dst_reg->u32_min_value += umin_val; | |
9674 | dst_reg->u32_max_value += umax_val; | |
9675 | } | |
9676 | } | |
9677 | ||
07cd2631 JF |
9678 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
9679 | struct bpf_reg_state *src_reg) | |
9680 | { | |
9681 | s64 smin_val = src_reg->smin_value; | |
9682 | s64 smax_val = src_reg->smax_value; | |
9683 | u64 umin_val = src_reg->umin_value; | |
9684 | u64 umax_val = src_reg->umax_value; | |
9685 | ||
9686 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
9687 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
9688 | dst_reg->smin_value = S64_MIN; | |
9689 | dst_reg->smax_value = S64_MAX; | |
9690 | } else { | |
9691 | dst_reg->smin_value += smin_val; | |
9692 | dst_reg->smax_value += smax_val; | |
9693 | } | |
9694 | if (dst_reg->umin_value + umin_val < umin_val || | |
9695 | dst_reg->umax_value + umax_val < umax_val) { | |
9696 | dst_reg->umin_value = 0; | |
9697 | dst_reg->umax_value = U64_MAX; | |
9698 | } else { | |
9699 | dst_reg->umin_value += umin_val; | |
9700 | dst_reg->umax_value += umax_val; | |
9701 | } | |
3f50f132 JF |
9702 | } |
9703 | ||
9704 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
9705 | struct bpf_reg_state *src_reg) | |
9706 | { | |
9707 | s32 smin_val = src_reg->s32_min_value; | |
9708 | s32 smax_val = src_reg->s32_max_value; | |
9709 | u32 umin_val = src_reg->u32_min_value; | |
9710 | u32 umax_val = src_reg->u32_max_value; | |
9711 | ||
9712 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
9713 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
9714 | /* Overflow possible, we know nothing */ | |
9715 | dst_reg->s32_min_value = S32_MIN; | |
9716 | dst_reg->s32_max_value = S32_MAX; | |
9717 | } else { | |
9718 | dst_reg->s32_min_value -= smax_val; | |
9719 | dst_reg->s32_max_value -= smin_val; | |
9720 | } | |
9721 | if (dst_reg->u32_min_value < umax_val) { | |
9722 | /* Overflow possible, we know nothing */ | |
9723 | dst_reg->u32_min_value = 0; | |
9724 | dst_reg->u32_max_value = U32_MAX; | |
9725 | } else { | |
9726 | /* Cannot overflow (as long as bounds are consistent) */ | |
9727 | dst_reg->u32_min_value -= umax_val; | |
9728 | dst_reg->u32_max_value -= umin_val; | |
9729 | } | |
07cd2631 JF |
9730 | } |
9731 | ||
9732 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
9733 | struct bpf_reg_state *src_reg) | |
9734 | { | |
9735 | s64 smin_val = src_reg->smin_value; | |
9736 | s64 smax_val = src_reg->smax_value; | |
9737 | u64 umin_val = src_reg->umin_value; | |
9738 | u64 umax_val = src_reg->umax_value; | |
9739 | ||
9740 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
9741 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
9742 | /* Overflow possible, we know nothing */ | |
9743 | dst_reg->smin_value = S64_MIN; | |
9744 | dst_reg->smax_value = S64_MAX; | |
9745 | } else { | |
9746 | dst_reg->smin_value -= smax_val; | |
9747 | dst_reg->smax_value -= smin_val; | |
9748 | } | |
9749 | if (dst_reg->umin_value < umax_val) { | |
9750 | /* Overflow possible, we know nothing */ | |
9751 | dst_reg->umin_value = 0; | |
9752 | dst_reg->umax_value = U64_MAX; | |
9753 | } else { | |
9754 | /* Cannot overflow (as long as bounds are consistent) */ | |
9755 | dst_reg->umin_value -= umax_val; | |
9756 | dst_reg->umax_value -= umin_val; | |
9757 | } | |
3f50f132 JF |
9758 | } |
9759 | ||
9760 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
9761 | struct bpf_reg_state *src_reg) | |
9762 | { | |
9763 | s32 smin_val = src_reg->s32_min_value; | |
9764 | u32 umin_val = src_reg->u32_min_value; | |
9765 | u32 umax_val = src_reg->u32_max_value; | |
9766 | ||
9767 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
9768 | /* Ain't nobody got time to multiply that sign */ | |
9769 | __mark_reg32_unbounded(dst_reg); | |
9770 | return; | |
9771 | } | |
9772 | /* Both values are positive, so we can work with unsigned and | |
9773 | * copy the result to signed (unless it exceeds S32_MAX). | |
9774 | */ | |
9775 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
9776 | /* Potential overflow, we know nothing */ | |
9777 | __mark_reg32_unbounded(dst_reg); | |
9778 | return; | |
9779 | } | |
9780 | dst_reg->u32_min_value *= umin_val; | |
9781 | dst_reg->u32_max_value *= umax_val; | |
9782 | if (dst_reg->u32_max_value > S32_MAX) { | |
9783 | /* Overflow possible, we know nothing */ | |
9784 | dst_reg->s32_min_value = S32_MIN; | |
9785 | dst_reg->s32_max_value = S32_MAX; | |
9786 | } else { | |
9787 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
9788 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
9789 | } | |
07cd2631 JF |
9790 | } |
9791 | ||
9792 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
9793 | struct bpf_reg_state *src_reg) | |
9794 | { | |
9795 | s64 smin_val = src_reg->smin_value; | |
9796 | u64 umin_val = src_reg->umin_value; | |
9797 | u64 umax_val = src_reg->umax_value; | |
9798 | ||
07cd2631 JF |
9799 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
9800 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 9801 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
9802 | return; |
9803 | } | |
9804 | /* Both values are positive, so we can work with unsigned and | |
9805 | * copy the result to signed (unless it exceeds S64_MAX). | |
9806 | */ | |
9807 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
9808 | /* Potential overflow, we know nothing */ | |
3f50f132 | 9809 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
9810 | return; |
9811 | } | |
9812 | dst_reg->umin_value *= umin_val; | |
9813 | dst_reg->umax_value *= umax_val; | |
9814 | if (dst_reg->umax_value > S64_MAX) { | |
9815 | /* Overflow possible, we know nothing */ | |
9816 | dst_reg->smin_value = S64_MIN; | |
9817 | dst_reg->smax_value = S64_MAX; | |
9818 | } else { | |
9819 | dst_reg->smin_value = dst_reg->umin_value; | |
9820 | dst_reg->smax_value = dst_reg->umax_value; | |
9821 | } | |
9822 | } | |
9823 | ||
3f50f132 JF |
9824 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
9825 | struct bpf_reg_state *src_reg) | |
9826 | { | |
9827 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
9828 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
9829 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
9830 | s32 smin_val = src_reg->s32_min_value; | |
9831 | u32 umax_val = src_reg->u32_max_value; | |
9832 | ||
049c4e13 DB |
9833 | if (src_known && dst_known) { |
9834 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 9835 | return; |
049c4e13 | 9836 | } |
3f50f132 JF |
9837 | |
9838 | /* We get our minimum from the var_off, since that's inherently | |
9839 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
9840 | */ | |
9841 | dst_reg->u32_min_value = var32_off.value; | |
9842 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
9843 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
9844 | /* Lose signed bounds when ANDing negative numbers, | |
9845 | * ain't nobody got time for that. | |
9846 | */ | |
9847 | dst_reg->s32_min_value = S32_MIN; | |
9848 | dst_reg->s32_max_value = S32_MAX; | |
9849 | } else { | |
9850 | /* ANDing two positives gives a positive, so safe to | |
9851 | * cast result into s64. | |
9852 | */ | |
9853 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
9854 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
9855 | } | |
3f50f132 JF |
9856 | } |
9857 | ||
07cd2631 JF |
9858 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
9859 | struct bpf_reg_state *src_reg) | |
9860 | { | |
3f50f132 JF |
9861 | bool src_known = tnum_is_const(src_reg->var_off); |
9862 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
9863 | s64 smin_val = src_reg->smin_value; |
9864 | u64 umax_val = src_reg->umax_value; | |
9865 | ||
3f50f132 | 9866 | if (src_known && dst_known) { |
4fbb38a3 | 9867 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
9868 | return; |
9869 | } | |
9870 | ||
07cd2631 JF |
9871 | /* We get our minimum from the var_off, since that's inherently |
9872 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
9873 | */ | |
07cd2631 JF |
9874 | dst_reg->umin_value = dst_reg->var_off.value; |
9875 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
9876 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
9877 | /* Lose signed bounds when ANDing negative numbers, | |
9878 | * ain't nobody got time for that. | |
9879 | */ | |
9880 | dst_reg->smin_value = S64_MIN; | |
9881 | dst_reg->smax_value = S64_MAX; | |
9882 | } else { | |
9883 | /* ANDing two positives gives a positive, so safe to | |
9884 | * cast result into s64. | |
9885 | */ | |
9886 | dst_reg->smin_value = dst_reg->umin_value; | |
9887 | dst_reg->smax_value = dst_reg->umax_value; | |
9888 | } | |
9889 | /* We may learn something more from the var_off */ | |
9890 | __update_reg_bounds(dst_reg); | |
9891 | } | |
9892 | ||
3f50f132 JF |
9893 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
9894 | struct bpf_reg_state *src_reg) | |
9895 | { | |
9896 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
9897 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
9898 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5b9fbeb7 DB |
9899 | s32 smin_val = src_reg->s32_min_value; |
9900 | u32 umin_val = src_reg->u32_min_value; | |
3f50f132 | 9901 | |
049c4e13 DB |
9902 | if (src_known && dst_known) { |
9903 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 9904 | return; |
049c4e13 | 9905 | } |
3f50f132 JF |
9906 | |
9907 | /* We get our maximum from the var_off, and our minimum is the | |
9908 | * maximum of the operands' minima | |
9909 | */ | |
9910 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
9911 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
9912 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
9913 | /* Lose signed bounds when ORing negative numbers, | |
9914 | * ain't nobody got time for that. | |
9915 | */ | |
9916 | dst_reg->s32_min_value = S32_MIN; | |
9917 | dst_reg->s32_max_value = S32_MAX; | |
9918 | } else { | |
9919 | /* ORing two positives gives a positive, so safe to | |
9920 | * cast result into s64. | |
9921 | */ | |
5b9fbeb7 DB |
9922 | dst_reg->s32_min_value = dst_reg->u32_min_value; |
9923 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
3f50f132 JF |
9924 | } |
9925 | } | |
9926 | ||
07cd2631 JF |
9927 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
9928 | struct bpf_reg_state *src_reg) | |
9929 | { | |
3f50f132 JF |
9930 | bool src_known = tnum_is_const(src_reg->var_off); |
9931 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
9932 | s64 smin_val = src_reg->smin_value; |
9933 | u64 umin_val = src_reg->umin_value; | |
9934 | ||
3f50f132 | 9935 | if (src_known && dst_known) { |
4fbb38a3 | 9936 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
9937 | return; |
9938 | } | |
9939 | ||
07cd2631 JF |
9940 | /* We get our maximum from the var_off, and our minimum is the |
9941 | * maximum of the operands' minima | |
9942 | */ | |
07cd2631 JF |
9943 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
9944 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
9945 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
9946 | /* Lose signed bounds when ORing negative numbers, | |
9947 | * ain't nobody got time for that. | |
9948 | */ | |
9949 | dst_reg->smin_value = S64_MIN; | |
9950 | dst_reg->smax_value = S64_MAX; | |
9951 | } else { | |
9952 | /* ORing two positives gives a positive, so safe to | |
9953 | * cast result into s64. | |
9954 | */ | |
9955 | dst_reg->smin_value = dst_reg->umin_value; | |
9956 | dst_reg->smax_value = dst_reg->umax_value; | |
9957 | } | |
9958 | /* We may learn something more from the var_off */ | |
9959 | __update_reg_bounds(dst_reg); | |
9960 | } | |
9961 | ||
2921c90d YS |
9962 | static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg, |
9963 | struct bpf_reg_state *src_reg) | |
9964 | { | |
9965 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
9966 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
9967 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
9968 | s32 smin_val = src_reg->s32_min_value; | |
9969 | ||
049c4e13 DB |
9970 | if (src_known && dst_known) { |
9971 | __mark_reg32_known(dst_reg, var32_off.value); | |
2921c90d | 9972 | return; |
049c4e13 | 9973 | } |
2921c90d YS |
9974 | |
9975 | /* We get both minimum and maximum from the var32_off. */ | |
9976 | dst_reg->u32_min_value = var32_off.value; | |
9977 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
9978 | ||
9979 | if (dst_reg->s32_min_value >= 0 && smin_val >= 0) { | |
9980 | /* XORing two positive sign numbers gives a positive, | |
9981 | * so safe to cast u32 result into s32. | |
9982 | */ | |
9983 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
9984 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
9985 | } else { | |
9986 | dst_reg->s32_min_value = S32_MIN; | |
9987 | dst_reg->s32_max_value = S32_MAX; | |
9988 | } | |
9989 | } | |
9990 | ||
9991 | static void scalar_min_max_xor(struct bpf_reg_state *dst_reg, | |
9992 | struct bpf_reg_state *src_reg) | |
9993 | { | |
9994 | bool src_known = tnum_is_const(src_reg->var_off); | |
9995 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
9996 | s64 smin_val = src_reg->smin_value; | |
9997 | ||
9998 | if (src_known && dst_known) { | |
9999 | /* dst_reg->var_off.value has been updated earlier */ | |
10000 | __mark_reg_known(dst_reg, dst_reg->var_off.value); | |
10001 | return; | |
10002 | } | |
10003 | ||
10004 | /* We get both minimum and maximum from the var_off. */ | |
10005 | dst_reg->umin_value = dst_reg->var_off.value; | |
10006 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
10007 | ||
10008 | if (dst_reg->smin_value >= 0 && smin_val >= 0) { | |
10009 | /* XORing two positive sign numbers gives a positive, | |
10010 | * so safe to cast u64 result into s64. | |
10011 | */ | |
10012 | dst_reg->smin_value = dst_reg->umin_value; | |
10013 | dst_reg->smax_value = dst_reg->umax_value; | |
10014 | } else { | |
10015 | dst_reg->smin_value = S64_MIN; | |
10016 | dst_reg->smax_value = S64_MAX; | |
10017 | } | |
10018 | ||
10019 | __update_reg_bounds(dst_reg); | |
10020 | } | |
10021 | ||
3f50f132 JF |
10022 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
10023 | u64 umin_val, u64 umax_val) | |
07cd2631 | 10024 | { |
07cd2631 JF |
10025 | /* We lose all sign bit information (except what we can pick |
10026 | * up from var_off) | |
10027 | */ | |
3f50f132 JF |
10028 | dst_reg->s32_min_value = S32_MIN; |
10029 | dst_reg->s32_max_value = S32_MAX; | |
10030 | /* If we might shift our top bit out, then we know nothing */ | |
10031 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
10032 | dst_reg->u32_min_value = 0; | |
10033 | dst_reg->u32_max_value = U32_MAX; | |
10034 | } else { | |
10035 | dst_reg->u32_min_value <<= umin_val; | |
10036 | dst_reg->u32_max_value <<= umax_val; | |
10037 | } | |
10038 | } | |
10039 | ||
10040 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
10041 | struct bpf_reg_state *src_reg) | |
10042 | { | |
10043 | u32 umax_val = src_reg->u32_max_value; | |
10044 | u32 umin_val = src_reg->u32_min_value; | |
10045 | /* u32 alu operation will zext upper bits */ | |
10046 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
10047 | ||
10048 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
10049 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
10050 | /* Not required but being careful mark reg64 bounds as unknown so | |
10051 | * that we are forced to pick them up from tnum and zext later and | |
10052 | * if some path skips this step we are still safe. | |
10053 | */ | |
10054 | __mark_reg64_unbounded(dst_reg); | |
10055 | __update_reg32_bounds(dst_reg); | |
10056 | } | |
10057 | ||
10058 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
10059 | u64 umin_val, u64 umax_val) | |
10060 | { | |
10061 | /* Special case <<32 because it is a common compiler pattern to sign | |
10062 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
10063 | * positive we know this shift will also be positive so we can track | |
10064 | * bounds correctly. Otherwise we lose all sign bit information except | |
10065 | * what we can pick up from var_off. Perhaps we can generalize this | |
10066 | * later to shifts of any length. | |
10067 | */ | |
10068 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
10069 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
10070 | else | |
10071 | dst_reg->smax_value = S64_MAX; | |
10072 | ||
10073 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
10074 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
10075 | else | |
10076 | dst_reg->smin_value = S64_MIN; | |
10077 | ||
07cd2631 JF |
10078 | /* If we might shift our top bit out, then we know nothing */ |
10079 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
10080 | dst_reg->umin_value = 0; | |
10081 | dst_reg->umax_value = U64_MAX; | |
10082 | } else { | |
10083 | dst_reg->umin_value <<= umin_val; | |
10084 | dst_reg->umax_value <<= umax_val; | |
10085 | } | |
3f50f132 JF |
10086 | } |
10087 | ||
10088 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
10089 | struct bpf_reg_state *src_reg) | |
10090 | { | |
10091 | u64 umax_val = src_reg->umax_value; | |
10092 | u64 umin_val = src_reg->umin_value; | |
10093 | ||
10094 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
10095 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
10096 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
10097 | ||
07cd2631 JF |
10098 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
10099 | /* We may learn something more from the var_off */ | |
10100 | __update_reg_bounds(dst_reg); | |
10101 | } | |
10102 | ||
3f50f132 JF |
10103 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
10104 | struct bpf_reg_state *src_reg) | |
10105 | { | |
10106 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
10107 | u32 umax_val = src_reg->u32_max_value; | |
10108 | u32 umin_val = src_reg->u32_min_value; | |
10109 | ||
10110 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
10111 | * be negative, then either: | |
10112 | * 1) src_reg might be zero, so the sign bit of the result is | |
10113 | * unknown, so we lose our signed bounds | |
10114 | * 2) it's known negative, thus the unsigned bounds capture the | |
10115 | * signed bounds | |
10116 | * 3) the signed bounds cross zero, so they tell us nothing | |
10117 | * about the result | |
10118 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 10119 | * unsigned bounds capture the signed bounds. |
3f50f132 JF |
10120 | * Thus, in all cases it suffices to blow away our signed bounds |
10121 | * and rely on inferring new ones from the unsigned bounds and | |
10122 | * var_off of the result. | |
10123 | */ | |
10124 | dst_reg->s32_min_value = S32_MIN; | |
10125 | dst_reg->s32_max_value = S32_MAX; | |
10126 | ||
10127 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
10128 | dst_reg->u32_min_value >>= umax_val; | |
10129 | dst_reg->u32_max_value >>= umin_val; | |
10130 | ||
10131 | __mark_reg64_unbounded(dst_reg); | |
10132 | __update_reg32_bounds(dst_reg); | |
10133 | } | |
10134 | ||
07cd2631 JF |
10135 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
10136 | struct bpf_reg_state *src_reg) | |
10137 | { | |
10138 | u64 umax_val = src_reg->umax_value; | |
10139 | u64 umin_val = src_reg->umin_value; | |
10140 | ||
10141 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
10142 | * be negative, then either: | |
10143 | * 1) src_reg might be zero, so the sign bit of the result is | |
10144 | * unknown, so we lose our signed bounds | |
10145 | * 2) it's known negative, thus the unsigned bounds capture the | |
10146 | * signed bounds | |
10147 | * 3) the signed bounds cross zero, so they tell us nothing | |
10148 | * about the result | |
10149 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 10150 | * unsigned bounds capture the signed bounds. |
07cd2631 JF |
10151 | * Thus, in all cases it suffices to blow away our signed bounds |
10152 | * and rely on inferring new ones from the unsigned bounds and | |
10153 | * var_off of the result. | |
10154 | */ | |
10155 | dst_reg->smin_value = S64_MIN; | |
10156 | dst_reg->smax_value = S64_MAX; | |
10157 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
10158 | dst_reg->umin_value >>= umax_val; | |
10159 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
10160 | |
10161 | /* Its not easy to operate on alu32 bounds here because it depends | |
10162 | * on bits being shifted in. Take easy way out and mark unbounded | |
10163 | * so we can recalculate later from tnum. | |
10164 | */ | |
10165 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
10166 | __update_reg_bounds(dst_reg); |
10167 | } | |
10168 | ||
3f50f132 JF |
10169 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
10170 | struct bpf_reg_state *src_reg) | |
07cd2631 | 10171 | { |
3f50f132 | 10172 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
10173 | |
10174 | /* Upon reaching here, src_known is true and | |
10175 | * umax_val is equal to umin_val. | |
10176 | */ | |
3f50f132 JF |
10177 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
10178 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 10179 | |
3f50f132 JF |
10180 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
10181 | ||
10182 | /* blow away the dst_reg umin_value/umax_value and rely on | |
10183 | * dst_reg var_off to refine the result. | |
10184 | */ | |
10185 | dst_reg->u32_min_value = 0; | |
10186 | dst_reg->u32_max_value = U32_MAX; | |
10187 | ||
10188 | __mark_reg64_unbounded(dst_reg); | |
10189 | __update_reg32_bounds(dst_reg); | |
10190 | } | |
10191 | ||
10192 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
10193 | struct bpf_reg_state *src_reg) | |
10194 | { | |
10195 | u64 umin_val = src_reg->umin_value; | |
10196 | ||
10197 | /* Upon reaching here, src_known is true and umax_val is equal | |
10198 | * to umin_val. | |
10199 | */ | |
10200 | dst_reg->smin_value >>= umin_val; | |
10201 | dst_reg->smax_value >>= umin_val; | |
10202 | ||
10203 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
10204 | |
10205 | /* blow away the dst_reg umin_value/umax_value and rely on | |
10206 | * dst_reg var_off to refine the result. | |
10207 | */ | |
10208 | dst_reg->umin_value = 0; | |
10209 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
10210 | |
10211 | /* Its not easy to operate on alu32 bounds here because it depends | |
10212 | * on bits being shifted in from upper 32-bits. Take easy way out | |
10213 | * and mark unbounded so we can recalculate later from tnum. | |
10214 | */ | |
10215 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
10216 | __update_reg_bounds(dst_reg); |
10217 | } | |
10218 | ||
468f6eaf JH |
10219 | /* WARNING: This function does calculations on 64-bit values, but the actual |
10220 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
10221 | * need extra checks in the 32-bit case. | |
10222 | */ | |
f1174f77 EC |
10223 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
10224 | struct bpf_insn *insn, | |
10225 | struct bpf_reg_state *dst_reg, | |
10226 | struct bpf_reg_state src_reg) | |
969bf05e | 10227 | { |
638f5b90 | 10228 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 10229 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 10230 | bool src_known; |
b03c9f9f EC |
10231 | s64 smin_val, smax_val; |
10232 | u64 umin_val, umax_val; | |
3f50f132 JF |
10233 | s32 s32_min_val, s32_max_val; |
10234 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 10235 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
3f50f132 | 10236 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
a6aaece0 | 10237 | int ret; |
b799207e | 10238 | |
b03c9f9f EC |
10239 | smin_val = src_reg.smin_value; |
10240 | smax_val = src_reg.smax_value; | |
10241 | umin_val = src_reg.umin_value; | |
10242 | umax_val = src_reg.umax_value; | |
f23cc643 | 10243 | |
3f50f132 JF |
10244 | s32_min_val = src_reg.s32_min_value; |
10245 | s32_max_val = src_reg.s32_max_value; | |
10246 | u32_min_val = src_reg.u32_min_value; | |
10247 | u32_max_val = src_reg.u32_max_value; | |
10248 | ||
10249 | if (alu32) { | |
10250 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
10251 | if ((src_known && |
10252 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
10253 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
10254 | /* Taint dst register if offset had invalid bounds | |
10255 | * derived from e.g. dead branches. | |
10256 | */ | |
10257 | __mark_reg_unknown(env, dst_reg); | |
10258 | return 0; | |
10259 | } | |
10260 | } else { | |
10261 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
10262 | if ((src_known && |
10263 | (smin_val != smax_val || umin_val != umax_val)) || | |
10264 | smin_val > smax_val || umin_val > umax_val) { | |
10265 | /* Taint dst register if offset had invalid bounds | |
10266 | * derived from e.g. dead branches. | |
10267 | */ | |
10268 | __mark_reg_unknown(env, dst_reg); | |
10269 | return 0; | |
10270 | } | |
6f16101e DB |
10271 | } |
10272 | ||
bb7f0f98 AS |
10273 | if (!src_known && |
10274 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 10275 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
10276 | return 0; |
10277 | } | |
10278 | ||
f5288193 DB |
10279 | if (sanitize_needed(opcode)) { |
10280 | ret = sanitize_val_alu(env, insn); | |
10281 | if (ret < 0) | |
10282 | return sanitize_err(env, insn, ret, NULL, NULL); | |
10283 | } | |
10284 | ||
3f50f132 JF |
10285 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
10286 | * There are two classes of instructions: The first class we track both | |
10287 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
10288 | * greatest amount of precision when alu operations are mixed with jmp32 | |
10289 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
10290 | * and BPF_OR. This is possible because these ops have fairly easy to | |
10291 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
10292 | * See alu32 verifier tests for examples. The second class of | |
10293 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
10294 | * with regards to tracking sign/unsigned bounds because the bits may | |
10295 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
10296 | * the reg unbounded in the subreg bound space and use the resulting | |
10297 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
10298 | */ | |
48461135 JB |
10299 | switch (opcode) { |
10300 | case BPF_ADD: | |
3f50f132 | 10301 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 10302 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 10303 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
10304 | break; |
10305 | case BPF_SUB: | |
3f50f132 | 10306 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 10307 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 10308 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
10309 | break; |
10310 | case BPF_MUL: | |
3f50f132 JF |
10311 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
10312 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 10313 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
10314 | break; |
10315 | case BPF_AND: | |
3f50f132 JF |
10316 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
10317 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 10318 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
10319 | break; |
10320 | case BPF_OR: | |
3f50f132 JF |
10321 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
10322 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 10323 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 | 10324 | break; |
2921c90d YS |
10325 | case BPF_XOR: |
10326 | dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off); | |
10327 | scalar32_min_max_xor(dst_reg, &src_reg); | |
10328 | scalar_min_max_xor(dst_reg, &src_reg); | |
10329 | break; | |
48461135 | 10330 | case BPF_LSH: |
468f6eaf JH |
10331 | if (umax_val >= insn_bitness) { |
10332 | /* Shifts greater than 31 or 63 are undefined. | |
10333 | * This includes shifts by a negative number. | |
b03c9f9f | 10334 | */ |
61bd5218 | 10335 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
10336 | break; |
10337 | } | |
3f50f132 JF |
10338 | if (alu32) |
10339 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
10340 | else | |
10341 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
10342 | break; |
10343 | case BPF_RSH: | |
468f6eaf JH |
10344 | if (umax_val >= insn_bitness) { |
10345 | /* Shifts greater than 31 or 63 are undefined. | |
10346 | * This includes shifts by a negative number. | |
b03c9f9f | 10347 | */ |
61bd5218 | 10348 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
10349 | break; |
10350 | } | |
3f50f132 JF |
10351 | if (alu32) |
10352 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
10353 | else | |
10354 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 10355 | break; |
9cbe1f5a YS |
10356 | case BPF_ARSH: |
10357 | if (umax_val >= insn_bitness) { | |
10358 | /* Shifts greater than 31 or 63 are undefined. | |
10359 | * This includes shifts by a negative number. | |
10360 | */ | |
10361 | mark_reg_unknown(env, regs, insn->dst_reg); | |
10362 | break; | |
10363 | } | |
3f50f132 JF |
10364 | if (alu32) |
10365 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
10366 | else | |
10367 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 10368 | break; |
48461135 | 10369 | default: |
61bd5218 | 10370 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
10371 | break; |
10372 | } | |
10373 | ||
3f50f132 JF |
10374 | /* ALU32 ops are zero extended into 64bit register */ |
10375 | if (alu32) | |
10376 | zext_32_to_64(dst_reg); | |
3844d153 | 10377 | reg_bounds_sync(dst_reg); |
f1174f77 EC |
10378 | return 0; |
10379 | } | |
10380 | ||
10381 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
10382 | * and var_off. | |
10383 | */ | |
10384 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
10385 | struct bpf_insn *insn) | |
10386 | { | |
f4d7e40a AS |
10387 | struct bpf_verifier_state *vstate = env->cur_state; |
10388 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
10389 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
10390 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
10391 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 10392 | int err; |
f1174f77 EC |
10393 | |
10394 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
10395 | src_reg = NULL; |
10396 | if (dst_reg->type != SCALAR_VALUE) | |
10397 | ptr_reg = dst_reg; | |
75748837 AS |
10398 | else |
10399 | /* Make sure ID is cleared otherwise dst_reg min/max could be | |
10400 | * incorrectly propagated into other registers by find_equal_scalars() | |
10401 | */ | |
10402 | dst_reg->id = 0; | |
f1174f77 EC |
10403 | if (BPF_SRC(insn->code) == BPF_X) { |
10404 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
10405 | if (src_reg->type != SCALAR_VALUE) { |
10406 | if (dst_reg->type != SCALAR_VALUE) { | |
10407 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
10408 | * an arbitrary scalar. Disallow all math except |
10409 | * pointer subtraction | |
f1174f77 | 10410 | */ |
dd066823 | 10411 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
10412 | mark_reg_unknown(env, regs, insn->dst_reg); |
10413 | return 0; | |
f1174f77 | 10414 | } |
82abbf8d AS |
10415 | verbose(env, "R%d pointer %s pointer prohibited\n", |
10416 | insn->dst_reg, | |
10417 | bpf_alu_string[opcode >> 4]); | |
10418 | return -EACCES; | |
f1174f77 EC |
10419 | } else { |
10420 | /* scalar += pointer | |
10421 | * This is legal, but we have to reverse our | |
10422 | * src/dest handling in computing the range | |
10423 | */ | |
b5dc0163 AS |
10424 | err = mark_chain_precision(env, insn->dst_reg); |
10425 | if (err) | |
10426 | return err; | |
82abbf8d AS |
10427 | return adjust_ptr_min_max_vals(env, insn, |
10428 | src_reg, dst_reg); | |
f1174f77 EC |
10429 | } |
10430 | } else if (ptr_reg) { | |
10431 | /* pointer += scalar */ | |
b5dc0163 AS |
10432 | err = mark_chain_precision(env, insn->src_reg); |
10433 | if (err) | |
10434 | return err; | |
82abbf8d AS |
10435 | return adjust_ptr_min_max_vals(env, insn, |
10436 | dst_reg, src_reg); | |
a3b666bf AN |
10437 | } else if (dst_reg->precise) { |
10438 | /* if dst_reg is precise, src_reg should be precise as well */ | |
10439 | err = mark_chain_precision(env, insn->src_reg); | |
10440 | if (err) | |
10441 | return err; | |
f1174f77 EC |
10442 | } |
10443 | } else { | |
10444 | /* Pretend the src is a reg with a known value, since we only | |
10445 | * need to be able to read from this state. | |
10446 | */ | |
10447 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 10448 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 10449 | src_reg = &off_reg; |
82abbf8d AS |
10450 | if (ptr_reg) /* pointer += K */ |
10451 | return adjust_ptr_min_max_vals(env, insn, | |
10452 | ptr_reg, src_reg); | |
f1174f77 EC |
10453 | } |
10454 | ||
10455 | /* Got here implies adding two SCALAR_VALUEs */ | |
10456 | if (WARN_ON_ONCE(ptr_reg)) { | |
0f55f9ed | 10457 | print_verifier_state(env, state, true); |
61bd5218 | 10458 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
10459 | return -EINVAL; |
10460 | } | |
10461 | if (WARN_ON(!src_reg)) { | |
0f55f9ed | 10462 | print_verifier_state(env, state, true); |
61bd5218 | 10463 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
10464 | return -EINVAL; |
10465 | } | |
10466 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
10467 | } |
10468 | ||
17a52670 | 10469 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 10470 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 10471 | { |
638f5b90 | 10472 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
10473 | u8 opcode = BPF_OP(insn->code); |
10474 | int err; | |
10475 | ||
10476 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
10477 | if (opcode == BPF_NEG) { | |
395e942d | 10478 | if (BPF_SRC(insn->code) != BPF_K || |
17a52670 AS |
10479 | insn->src_reg != BPF_REG_0 || |
10480 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 10481 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
10482 | return -EINVAL; |
10483 | } | |
10484 | } else { | |
10485 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
10486 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
10487 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 10488 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
10489 | return -EINVAL; |
10490 | } | |
10491 | } | |
10492 | ||
10493 | /* check src operand */ | |
dc503a8a | 10494 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10495 | if (err) |
10496 | return err; | |
10497 | ||
1be7f75d | 10498 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 10499 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
10500 | insn->dst_reg); |
10501 | return -EACCES; | |
10502 | } | |
10503 | ||
17a52670 | 10504 | /* check dest operand */ |
dc503a8a | 10505 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
10506 | if (err) |
10507 | return err; | |
10508 | ||
10509 | } else if (opcode == BPF_MOV) { | |
10510 | ||
10511 | if (BPF_SRC(insn->code) == BPF_X) { | |
10512 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 10513 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
10514 | return -EINVAL; |
10515 | } | |
10516 | ||
10517 | /* check src operand */ | |
dc503a8a | 10518 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10519 | if (err) |
10520 | return err; | |
10521 | } else { | |
10522 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 10523 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
10524 | return -EINVAL; |
10525 | } | |
10526 | } | |
10527 | ||
fbeb1603 AF |
10528 | /* check dest operand, mark as required later */ |
10529 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
10530 | if (err) |
10531 | return err; | |
10532 | ||
10533 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
10534 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
10535 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
10536 | ||
17a52670 AS |
10537 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
10538 | /* case: R1 = R2 | |
10539 | * copy register state to dest reg | |
10540 | */ | |
75748837 AS |
10541 | if (src_reg->type == SCALAR_VALUE && !src_reg->id) |
10542 | /* Assign src and dst registers the same ID | |
10543 | * that will be used by find_equal_scalars() | |
10544 | * to propagate min/max range. | |
10545 | */ | |
10546 | src_reg->id = ++env->id_gen; | |
e434b8cd JW |
10547 | *dst_reg = *src_reg; |
10548 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 10549 | dst_reg->subreg_def = DEF_NOT_SUBREG; |
17a52670 | 10550 | } else { |
f1174f77 | 10551 | /* R1 = (u32) R2 */ |
1be7f75d | 10552 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
10553 | verbose(env, |
10554 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
10555 | insn->src_reg); |
10556 | return -EACCES; | |
e434b8cd JW |
10557 | } else if (src_reg->type == SCALAR_VALUE) { |
10558 | *dst_reg = *src_reg; | |
75748837 AS |
10559 | /* Make sure ID is cleared otherwise |
10560 | * dst_reg min/max could be incorrectly | |
10561 | * propagated into src_reg by find_equal_scalars() | |
10562 | */ | |
10563 | dst_reg->id = 0; | |
e434b8cd | 10564 | dst_reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 10565 | dst_reg->subreg_def = env->insn_idx + 1; |
e434b8cd JW |
10566 | } else { |
10567 | mark_reg_unknown(env, regs, | |
10568 | insn->dst_reg); | |
1be7f75d | 10569 | } |
3f50f132 | 10570 | zext_32_to_64(dst_reg); |
3844d153 | 10571 | reg_bounds_sync(dst_reg); |
17a52670 AS |
10572 | } |
10573 | } else { | |
10574 | /* case: R = imm | |
10575 | * remember the value we stored into this reg | |
10576 | */ | |
fbeb1603 AF |
10577 | /* clear any state __mark_reg_known doesn't set */ |
10578 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 10579 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
10580 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
10581 | __mark_reg_known(regs + insn->dst_reg, | |
10582 | insn->imm); | |
10583 | } else { | |
10584 | __mark_reg_known(regs + insn->dst_reg, | |
10585 | (u32)insn->imm); | |
10586 | } | |
17a52670 AS |
10587 | } |
10588 | ||
10589 | } else if (opcode > BPF_END) { | |
61bd5218 | 10590 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
10591 | return -EINVAL; |
10592 | ||
10593 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
10594 | ||
17a52670 AS |
10595 | if (BPF_SRC(insn->code) == BPF_X) { |
10596 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 10597 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
10598 | return -EINVAL; |
10599 | } | |
10600 | /* check src1 operand */ | |
dc503a8a | 10601 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10602 | if (err) |
10603 | return err; | |
10604 | } else { | |
10605 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 10606 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
10607 | return -EINVAL; |
10608 | } | |
10609 | } | |
10610 | ||
10611 | /* check src2 operand */ | |
dc503a8a | 10612 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10613 | if (err) |
10614 | return err; | |
10615 | ||
10616 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
10617 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 10618 | verbose(env, "div by zero\n"); |
17a52670 AS |
10619 | return -EINVAL; |
10620 | } | |
10621 | ||
229394e8 RV |
10622 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
10623 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
10624 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
10625 | ||
10626 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 10627 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
10628 | return -EINVAL; |
10629 | } | |
10630 | } | |
10631 | ||
1a0dc1ac | 10632 | /* check dest operand */ |
dc503a8a | 10633 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
10634 | if (err) |
10635 | return err; | |
10636 | ||
f1174f77 | 10637 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
10638 | } |
10639 | ||
10640 | return 0; | |
10641 | } | |
10642 | ||
f4d7e40a | 10643 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 10644 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 10645 | enum bpf_reg_type type, |
fb2a311a | 10646 | bool range_right_open) |
969bf05e | 10647 | { |
b239da34 KKD |
10648 | struct bpf_func_state *state; |
10649 | struct bpf_reg_state *reg; | |
10650 | int new_range; | |
2d2be8ca | 10651 | |
fb2a311a DB |
10652 | if (dst_reg->off < 0 || |
10653 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
10654 | /* This doesn't give us any range */ |
10655 | return; | |
10656 | ||
b03c9f9f EC |
10657 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
10658 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
10659 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
10660 | * than pkt_end, but that's because it's also less than pkt. | |
10661 | */ | |
10662 | return; | |
10663 | ||
fb2a311a DB |
10664 | new_range = dst_reg->off; |
10665 | if (range_right_open) | |
2fa7d94a | 10666 | new_range++; |
fb2a311a DB |
10667 | |
10668 | /* Examples for register markings: | |
2d2be8ca | 10669 | * |
fb2a311a | 10670 | * pkt_data in dst register: |
2d2be8ca DB |
10671 | * |
10672 | * r2 = r3; | |
10673 | * r2 += 8; | |
10674 | * if (r2 > pkt_end) goto <handle exception> | |
10675 | * <access okay> | |
10676 | * | |
b4e432f1 DB |
10677 | * r2 = r3; |
10678 | * r2 += 8; | |
10679 | * if (r2 < pkt_end) goto <access okay> | |
10680 | * <handle exception> | |
10681 | * | |
2d2be8ca DB |
10682 | * Where: |
10683 | * r2 == dst_reg, pkt_end == src_reg | |
10684 | * r2=pkt(id=n,off=8,r=0) | |
10685 | * r3=pkt(id=n,off=0,r=0) | |
10686 | * | |
fb2a311a | 10687 | * pkt_data in src register: |
2d2be8ca DB |
10688 | * |
10689 | * r2 = r3; | |
10690 | * r2 += 8; | |
10691 | * if (pkt_end >= r2) goto <access okay> | |
10692 | * <handle exception> | |
10693 | * | |
b4e432f1 DB |
10694 | * r2 = r3; |
10695 | * r2 += 8; | |
10696 | * if (pkt_end <= r2) goto <handle exception> | |
10697 | * <access okay> | |
10698 | * | |
2d2be8ca DB |
10699 | * Where: |
10700 | * pkt_end == dst_reg, r2 == src_reg | |
10701 | * r2=pkt(id=n,off=8,r=0) | |
10702 | * r3=pkt(id=n,off=0,r=0) | |
10703 | * | |
10704 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
10705 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
10706 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
10707 | * the check. | |
969bf05e | 10708 | */ |
2d2be8ca | 10709 | |
f1174f77 EC |
10710 | /* If our ids match, then we must have the same max_value. And we |
10711 | * don't care about the other reg's fixed offset, since if it's too big | |
10712 | * the range won't allow anything. | |
10713 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
10714 | */ | |
b239da34 KKD |
10715 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
10716 | if (reg->type == type && reg->id == dst_reg->id) | |
10717 | /* keep the maximum range already checked */ | |
10718 | reg->range = max(reg->range, new_range); | |
10719 | })); | |
969bf05e AS |
10720 | } |
10721 | ||
3f50f132 | 10722 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 10723 | { |
3f50f132 JF |
10724 | struct tnum subreg = tnum_subreg(reg->var_off); |
10725 | s32 sval = (s32)val; | |
a72dafaf | 10726 | |
3f50f132 JF |
10727 | switch (opcode) { |
10728 | case BPF_JEQ: | |
10729 | if (tnum_is_const(subreg)) | |
10730 | return !!tnum_equals_const(subreg, val); | |
10731 | break; | |
10732 | case BPF_JNE: | |
10733 | if (tnum_is_const(subreg)) | |
10734 | return !tnum_equals_const(subreg, val); | |
10735 | break; | |
10736 | case BPF_JSET: | |
10737 | if ((~subreg.mask & subreg.value) & val) | |
10738 | return 1; | |
10739 | if (!((subreg.mask | subreg.value) & val)) | |
10740 | return 0; | |
10741 | break; | |
10742 | case BPF_JGT: | |
10743 | if (reg->u32_min_value > val) | |
10744 | return 1; | |
10745 | else if (reg->u32_max_value <= val) | |
10746 | return 0; | |
10747 | break; | |
10748 | case BPF_JSGT: | |
10749 | if (reg->s32_min_value > sval) | |
10750 | return 1; | |
ee114dd6 | 10751 | else if (reg->s32_max_value <= sval) |
3f50f132 JF |
10752 | return 0; |
10753 | break; | |
10754 | case BPF_JLT: | |
10755 | if (reg->u32_max_value < val) | |
10756 | return 1; | |
10757 | else if (reg->u32_min_value >= val) | |
10758 | return 0; | |
10759 | break; | |
10760 | case BPF_JSLT: | |
10761 | if (reg->s32_max_value < sval) | |
10762 | return 1; | |
10763 | else if (reg->s32_min_value >= sval) | |
10764 | return 0; | |
10765 | break; | |
10766 | case BPF_JGE: | |
10767 | if (reg->u32_min_value >= val) | |
10768 | return 1; | |
10769 | else if (reg->u32_max_value < val) | |
10770 | return 0; | |
10771 | break; | |
10772 | case BPF_JSGE: | |
10773 | if (reg->s32_min_value >= sval) | |
10774 | return 1; | |
10775 | else if (reg->s32_max_value < sval) | |
10776 | return 0; | |
10777 | break; | |
10778 | case BPF_JLE: | |
10779 | if (reg->u32_max_value <= val) | |
10780 | return 1; | |
10781 | else if (reg->u32_min_value > val) | |
10782 | return 0; | |
10783 | break; | |
10784 | case BPF_JSLE: | |
10785 | if (reg->s32_max_value <= sval) | |
10786 | return 1; | |
10787 | else if (reg->s32_min_value > sval) | |
10788 | return 0; | |
10789 | break; | |
10790 | } | |
4f7b3e82 | 10791 | |
3f50f132 JF |
10792 | return -1; |
10793 | } | |
092ed096 | 10794 | |
3f50f132 JF |
10795 | |
10796 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
10797 | { | |
10798 | s64 sval = (s64)val; | |
a72dafaf | 10799 | |
4f7b3e82 AS |
10800 | switch (opcode) { |
10801 | case BPF_JEQ: | |
10802 | if (tnum_is_const(reg->var_off)) | |
10803 | return !!tnum_equals_const(reg->var_off, val); | |
10804 | break; | |
10805 | case BPF_JNE: | |
10806 | if (tnum_is_const(reg->var_off)) | |
10807 | return !tnum_equals_const(reg->var_off, val); | |
10808 | break; | |
960ea056 JK |
10809 | case BPF_JSET: |
10810 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
10811 | return 1; | |
10812 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
10813 | return 0; | |
10814 | break; | |
4f7b3e82 AS |
10815 | case BPF_JGT: |
10816 | if (reg->umin_value > val) | |
10817 | return 1; | |
10818 | else if (reg->umax_value <= val) | |
10819 | return 0; | |
10820 | break; | |
10821 | case BPF_JSGT: | |
a72dafaf | 10822 | if (reg->smin_value > sval) |
4f7b3e82 | 10823 | return 1; |
ee114dd6 | 10824 | else if (reg->smax_value <= sval) |
4f7b3e82 AS |
10825 | return 0; |
10826 | break; | |
10827 | case BPF_JLT: | |
10828 | if (reg->umax_value < val) | |
10829 | return 1; | |
10830 | else if (reg->umin_value >= val) | |
10831 | return 0; | |
10832 | break; | |
10833 | case BPF_JSLT: | |
a72dafaf | 10834 | if (reg->smax_value < sval) |
4f7b3e82 | 10835 | return 1; |
a72dafaf | 10836 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
10837 | return 0; |
10838 | break; | |
10839 | case BPF_JGE: | |
10840 | if (reg->umin_value >= val) | |
10841 | return 1; | |
10842 | else if (reg->umax_value < val) | |
10843 | return 0; | |
10844 | break; | |
10845 | case BPF_JSGE: | |
a72dafaf | 10846 | if (reg->smin_value >= sval) |
4f7b3e82 | 10847 | return 1; |
a72dafaf | 10848 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
10849 | return 0; |
10850 | break; | |
10851 | case BPF_JLE: | |
10852 | if (reg->umax_value <= val) | |
10853 | return 1; | |
10854 | else if (reg->umin_value > val) | |
10855 | return 0; | |
10856 | break; | |
10857 | case BPF_JSLE: | |
a72dafaf | 10858 | if (reg->smax_value <= sval) |
4f7b3e82 | 10859 | return 1; |
a72dafaf | 10860 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
10861 | return 0; |
10862 | break; | |
10863 | } | |
10864 | ||
10865 | return -1; | |
10866 | } | |
10867 | ||
3f50f132 JF |
10868 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
10869 | * and return: | |
10870 | * 1 - branch will be taken and "goto target" will be executed | |
10871 | * 0 - branch will not be taken and fall-through to next insn | |
10872 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
10873 | * range [0,10] | |
604dca5e | 10874 | */ |
3f50f132 JF |
10875 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
10876 | bool is_jmp32) | |
604dca5e | 10877 | { |
cac616db JF |
10878 | if (__is_pointer_value(false, reg)) { |
10879 | if (!reg_type_not_null(reg->type)) | |
10880 | return -1; | |
10881 | ||
10882 | /* If pointer is valid tests against zero will fail so we can | |
10883 | * use this to direct branch taken. | |
10884 | */ | |
10885 | if (val != 0) | |
10886 | return -1; | |
10887 | ||
10888 | switch (opcode) { | |
10889 | case BPF_JEQ: | |
10890 | return 0; | |
10891 | case BPF_JNE: | |
10892 | return 1; | |
10893 | default: | |
10894 | return -1; | |
10895 | } | |
10896 | } | |
604dca5e | 10897 | |
3f50f132 JF |
10898 | if (is_jmp32) |
10899 | return is_branch32_taken(reg, val, opcode); | |
10900 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
10901 | } |
10902 | ||
6d94e741 AS |
10903 | static int flip_opcode(u32 opcode) |
10904 | { | |
10905 | /* How can we transform "a <op> b" into "b <op> a"? */ | |
10906 | static const u8 opcode_flip[16] = { | |
10907 | /* these stay the same */ | |
10908 | [BPF_JEQ >> 4] = BPF_JEQ, | |
10909 | [BPF_JNE >> 4] = BPF_JNE, | |
10910 | [BPF_JSET >> 4] = BPF_JSET, | |
10911 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
10912 | [BPF_JGE >> 4] = BPF_JLE, | |
10913 | [BPF_JGT >> 4] = BPF_JLT, | |
10914 | [BPF_JLE >> 4] = BPF_JGE, | |
10915 | [BPF_JLT >> 4] = BPF_JGT, | |
10916 | [BPF_JSGE >> 4] = BPF_JSLE, | |
10917 | [BPF_JSGT >> 4] = BPF_JSLT, | |
10918 | [BPF_JSLE >> 4] = BPF_JSGE, | |
10919 | [BPF_JSLT >> 4] = BPF_JSGT | |
10920 | }; | |
10921 | return opcode_flip[opcode >> 4]; | |
10922 | } | |
10923 | ||
10924 | static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg, | |
10925 | struct bpf_reg_state *src_reg, | |
10926 | u8 opcode) | |
10927 | { | |
10928 | struct bpf_reg_state *pkt; | |
10929 | ||
10930 | if (src_reg->type == PTR_TO_PACKET_END) { | |
10931 | pkt = dst_reg; | |
10932 | } else if (dst_reg->type == PTR_TO_PACKET_END) { | |
10933 | pkt = src_reg; | |
10934 | opcode = flip_opcode(opcode); | |
10935 | } else { | |
10936 | return -1; | |
10937 | } | |
10938 | ||
10939 | if (pkt->range >= 0) | |
10940 | return -1; | |
10941 | ||
10942 | switch (opcode) { | |
10943 | case BPF_JLE: | |
10944 | /* pkt <= pkt_end */ | |
10945 | fallthrough; | |
10946 | case BPF_JGT: | |
10947 | /* pkt > pkt_end */ | |
10948 | if (pkt->range == BEYOND_PKT_END) | |
10949 | /* pkt has at last one extra byte beyond pkt_end */ | |
10950 | return opcode == BPF_JGT; | |
10951 | break; | |
10952 | case BPF_JLT: | |
10953 | /* pkt < pkt_end */ | |
10954 | fallthrough; | |
10955 | case BPF_JGE: | |
10956 | /* pkt >= pkt_end */ | |
10957 | if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END) | |
10958 | return opcode == BPF_JGE; | |
10959 | break; | |
10960 | } | |
10961 | return -1; | |
10962 | } | |
10963 | ||
48461135 JB |
10964 | /* Adjusts the register min/max values in the case that the dst_reg is the |
10965 | * variable register that we are working on, and src_reg is a constant or we're | |
10966 | * simply doing a BPF_K check. | |
f1174f77 | 10967 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
10968 | */ |
10969 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
10970 | struct bpf_reg_state *false_reg, |
10971 | u64 val, u32 val32, | |
092ed096 | 10972 | u8 opcode, bool is_jmp32) |
48461135 | 10973 | { |
3f50f132 JF |
10974 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
10975 | struct tnum false_64off = false_reg->var_off; | |
10976 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
10977 | struct tnum true_64off = true_reg->var_off; | |
10978 | s64 sval = (s64)val; | |
10979 | s32 sval32 = (s32)val32; | |
a72dafaf | 10980 | |
f1174f77 EC |
10981 | /* If the dst_reg is a pointer, we can't learn anything about its |
10982 | * variable offset from the compare (unless src_reg were a pointer into | |
10983 | * the same object, but we don't bother with that. | |
10984 | * Since false_reg and true_reg have the same type by construction, we | |
10985 | * only need to check one of them for pointerness. | |
10986 | */ | |
10987 | if (__is_pointer_value(false, false_reg)) | |
10988 | return; | |
4cabc5b1 | 10989 | |
48461135 | 10990 | switch (opcode) { |
a12ca627 DB |
10991 | /* JEQ/JNE comparison doesn't change the register equivalence. |
10992 | * | |
10993 | * r1 = r2; | |
10994 | * if (r1 == 42) goto label; | |
10995 | * ... | |
10996 | * label: // here both r1 and r2 are known to be 42. | |
10997 | * | |
10998 | * Hence when marking register as known preserve it's ID. | |
10999 | */ | |
48461135 | 11000 | case BPF_JEQ: |
a12ca627 DB |
11001 | if (is_jmp32) { |
11002 | __mark_reg32_known(true_reg, val32); | |
11003 | true_32off = tnum_subreg(true_reg->var_off); | |
11004 | } else { | |
11005 | ___mark_reg_known(true_reg, val); | |
11006 | true_64off = true_reg->var_off; | |
11007 | } | |
11008 | break; | |
48461135 | 11009 | case BPF_JNE: |
a12ca627 DB |
11010 | if (is_jmp32) { |
11011 | __mark_reg32_known(false_reg, val32); | |
11012 | false_32off = tnum_subreg(false_reg->var_off); | |
11013 | } else { | |
11014 | ___mark_reg_known(false_reg, val); | |
11015 | false_64off = false_reg->var_off; | |
11016 | } | |
48461135 | 11017 | break; |
960ea056 | 11018 | case BPF_JSET: |
3f50f132 JF |
11019 | if (is_jmp32) { |
11020 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
11021 | if (is_power_of_2(val32)) | |
11022 | true_32off = tnum_or(true_32off, | |
11023 | tnum_const(val32)); | |
11024 | } else { | |
11025 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
11026 | if (is_power_of_2(val)) | |
11027 | true_64off = tnum_or(true_64off, | |
11028 | tnum_const(val)); | |
11029 | } | |
960ea056 | 11030 | break; |
48461135 | 11031 | case BPF_JGE: |
a72dafaf JW |
11032 | case BPF_JGT: |
11033 | { | |
3f50f132 JF |
11034 | if (is_jmp32) { |
11035 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
11036 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
11037 | ||
11038 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
11039 | false_umax); | |
11040 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
11041 | true_umin); | |
11042 | } else { | |
11043 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
11044 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
11045 | ||
11046 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
11047 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
11048 | } | |
b03c9f9f | 11049 | break; |
a72dafaf | 11050 | } |
48461135 | 11051 | case BPF_JSGE: |
a72dafaf JW |
11052 | case BPF_JSGT: |
11053 | { | |
3f50f132 JF |
11054 | if (is_jmp32) { |
11055 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
11056 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 11057 | |
3f50f132 JF |
11058 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
11059 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
11060 | } else { | |
11061 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
11062 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
11063 | ||
11064 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
11065 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
11066 | } | |
48461135 | 11067 | break; |
a72dafaf | 11068 | } |
b4e432f1 | 11069 | case BPF_JLE: |
a72dafaf JW |
11070 | case BPF_JLT: |
11071 | { | |
3f50f132 JF |
11072 | if (is_jmp32) { |
11073 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
11074 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
11075 | ||
11076 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
11077 | false_umin); | |
11078 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
11079 | true_umax); | |
11080 | } else { | |
11081 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
11082 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
11083 | ||
11084 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
11085 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
11086 | } | |
b4e432f1 | 11087 | break; |
a72dafaf | 11088 | } |
b4e432f1 | 11089 | case BPF_JSLE: |
a72dafaf JW |
11090 | case BPF_JSLT: |
11091 | { | |
3f50f132 JF |
11092 | if (is_jmp32) { |
11093 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
11094 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 11095 | |
3f50f132 JF |
11096 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
11097 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
11098 | } else { | |
11099 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
11100 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
11101 | ||
11102 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
11103 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
11104 | } | |
b4e432f1 | 11105 | break; |
a72dafaf | 11106 | } |
48461135 | 11107 | default: |
0fc31b10 | 11108 | return; |
48461135 JB |
11109 | } |
11110 | ||
3f50f132 JF |
11111 | if (is_jmp32) { |
11112 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
11113 | tnum_subreg(false_32off)); | |
11114 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
11115 | tnum_subreg(true_32off)); | |
11116 | __reg_combine_32_into_64(false_reg); | |
11117 | __reg_combine_32_into_64(true_reg); | |
11118 | } else { | |
11119 | false_reg->var_off = false_64off; | |
11120 | true_reg->var_off = true_64off; | |
11121 | __reg_combine_64_into_32(false_reg); | |
11122 | __reg_combine_64_into_32(true_reg); | |
11123 | } | |
48461135 JB |
11124 | } |
11125 | ||
f1174f77 EC |
11126 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
11127 | * the variable reg. | |
48461135 JB |
11128 | */ |
11129 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
11130 | struct bpf_reg_state *false_reg, |
11131 | u64 val, u32 val32, | |
092ed096 | 11132 | u8 opcode, bool is_jmp32) |
48461135 | 11133 | { |
6d94e741 | 11134 | opcode = flip_opcode(opcode); |
0fc31b10 JH |
11135 | /* This uses zero as "not present in table"; luckily the zero opcode, |
11136 | * BPF_JA, can't get here. | |
b03c9f9f | 11137 | */ |
0fc31b10 | 11138 | if (opcode) |
3f50f132 | 11139 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
11140 | } |
11141 | ||
11142 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
11143 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
11144 | struct bpf_reg_state *dst_reg) | |
11145 | { | |
b03c9f9f EC |
11146 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
11147 | dst_reg->umin_value); | |
11148 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
11149 | dst_reg->umax_value); | |
11150 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
11151 | dst_reg->smin_value); | |
11152 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
11153 | dst_reg->smax_value); | |
f1174f77 EC |
11154 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
11155 | dst_reg->var_off); | |
3844d153 DB |
11156 | reg_bounds_sync(src_reg); |
11157 | reg_bounds_sync(dst_reg); | |
f1174f77 EC |
11158 | } |
11159 | ||
11160 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
11161 | struct bpf_reg_state *true_dst, | |
11162 | struct bpf_reg_state *false_src, | |
11163 | struct bpf_reg_state *false_dst, | |
11164 | u8 opcode) | |
11165 | { | |
11166 | switch (opcode) { | |
11167 | case BPF_JEQ: | |
11168 | __reg_combine_min_max(true_src, true_dst); | |
11169 | break; | |
11170 | case BPF_JNE: | |
11171 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 11172 | break; |
4cabc5b1 | 11173 | } |
48461135 JB |
11174 | } |
11175 | ||
fd978bf7 JS |
11176 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
11177 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 11178 | bool is_null) |
57a09bf0 | 11179 | { |
c25b2ae1 | 11180 | if (type_may_be_null(reg->type) && reg->id == id && |
93c230e3 | 11181 | !WARN_ON_ONCE(!reg->id)) { |
df57f38a KKD |
11182 | /* Old offset (both fixed and variable parts) should have been |
11183 | * known-zero, because we don't allow pointer arithmetic on | |
11184 | * pointers that might be NULL. If we see this happening, don't | |
11185 | * convert the register. | |
11186 | * | |
11187 | * But in some cases, some helpers that return local kptrs | |
11188 | * advance offset for the returned pointer. In those cases, it | |
11189 | * is fine to expect to see reg->off. | |
11190 | */ | |
11191 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || !tnum_equals_const(reg->var_off, 0))) | |
11192 | return; | |
11193 | if (reg->type != (PTR_TO_BTF_ID | MEM_ALLOC | PTR_MAYBE_NULL) && WARN_ON_ONCE(reg->off)) | |
e60b0d12 | 11194 | return; |
f1174f77 EC |
11195 | if (is_null) { |
11196 | reg->type = SCALAR_VALUE; | |
1b986589 MKL |
11197 | /* We don't need id and ref_obj_id from this point |
11198 | * onwards anymore, thus we should better reset it, | |
11199 | * so that state pruning has chances to take effect. | |
11200 | */ | |
11201 | reg->id = 0; | |
11202 | reg->ref_obj_id = 0; | |
4ddb7416 DB |
11203 | |
11204 | return; | |
11205 | } | |
11206 | ||
11207 | mark_ptr_not_null_reg(reg); | |
11208 | ||
11209 | if (!reg_may_point_to_spin_lock(reg)) { | |
1b986589 | 11210 | /* For not-NULL ptr, reg->ref_obj_id will be reset |
b239da34 | 11211 | * in release_reference(). |
1b986589 MKL |
11212 | * |
11213 | * reg->id is still used by spin_lock ptr. Other | |
11214 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
11215 | */ |
11216 | reg->id = 0; | |
56f668df | 11217 | } |
57a09bf0 TG |
11218 | } |
11219 | } | |
11220 | ||
11221 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
11222 | * be folded together at some point. | |
11223 | */ | |
840b9615 JS |
11224 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
11225 | bool is_null) | |
57a09bf0 | 11226 | { |
f4d7e40a | 11227 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
b239da34 | 11228 | struct bpf_reg_state *regs = state->regs, *reg; |
1b986589 | 11229 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 11230 | u32 id = regs[regno].id; |
57a09bf0 | 11231 | |
1b986589 MKL |
11232 | if (ref_obj_id && ref_obj_id == id && is_null) |
11233 | /* regs[regno] is in the " == NULL" branch. | |
11234 | * No one could have freed the reference state before | |
11235 | * doing the NULL check. | |
11236 | */ | |
11237 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 11238 | |
b239da34 KKD |
11239 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
11240 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
11241 | })); | |
57a09bf0 TG |
11242 | } |
11243 | ||
5beca081 DB |
11244 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
11245 | struct bpf_reg_state *dst_reg, | |
11246 | struct bpf_reg_state *src_reg, | |
11247 | struct bpf_verifier_state *this_branch, | |
11248 | struct bpf_verifier_state *other_branch) | |
11249 | { | |
11250 | if (BPF_SRC(insn->code) != BPF_X) | |
11251 | return false; | |
11252 | ||
092ed096 JW |
11253 | /* Pointers are always 64-bit. */ |
11254 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
11255 | return false; | |
11256 | ||
5beca081 DB |
11257 | switch (BPF_OP(insn->code)) { |
11258 | case BPF_JGT: | |
11259 | if ((dst_reg->type == PTR_TO_PACKET && | |
11260 | src_reg->type == PTR_TO_PACKET_END) || | |
11261 | (dst_reg->type == PTR_TO_PACKET_META && | |
11262 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
11263 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
11264 | find_good_pkt_pointers(this_branch, dst_reg, | |
11265 | dst_reg->type, false); | |
6d94e741 | 11266 | mark_pkt_end(other_branch, insn->dst_reg, true); |
5beca081 DB |
11267 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
11268 | src_reg->type == PTR_TO_PACKET) || | |
11269 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
11270 | src_reg->type == PTR_TO_PACKET_META)) { | |
11271 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
11272 | find_good_pkt_pointers(other_branch, src_reg, | |
11273 | src_reg->type, true); | |
6d94e741 | 11274 | mark_pkt_end(this_branch, insn->src_reg, false); |
5beca081 DB |
11275 | } else { |
11276 | return false; | |
11277 | } | |
11278 | break; | |
11279 | case BPF_JLT: | |
11280 | if ((dst_reg->type == PTR_TO_PACKET && | |
11281 | src_reg->type == PTR_TO_PACKET_END) || | |
11282 | (dst_reg->type == PTR_TO_PACKET_META && | |
11283 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
11284 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
11285 | find_good_pkt_pointers(other_branch, dst_reg, | |
11286 | dst_reg->type, true); | |
6d94e741 | 11287 | mark_pkt_end(this_branch, insn->dst_reg, false); |
5beca081 DB |
11288 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
11289 | src_reg->type == PTR_TO_PACKET) || | |
11290 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
11291 | src_reg->type == PTR_TO_PACKET_META)) { | |
11292 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
11293 | find_good_pkt_pointers(this_branch, src_reg, | |
11294 | src_reg->type, false); | |
6d94e741 | 11295 | mark_pkt_end(other_branch, insn->src_reg, true); |
5beca081 DB |
11296 | } else { |
11297 | return false; | |
11298 | } | |
11299 | break; | |
11300 | case BPF_JGE: | |
11301 | if ((dst_reg->type == PTR_TO_PACKET && | |
11302 | src_reg->type == PTR_TO_PACKET_END) || | |
11303 | (dst_reg->type == PTR_TO_PACKET_META && | |
11304 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
11305 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
11306 | find_good_pkt_pointers(this_branch, dst_reg, | |
11307 | dst_reg->type, true); | |
6d94e741 | 11308 | mark_pkt_end(other_branch, insn->dst_reg, false); |
5beca081 DB |
11309 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
11310 | src_reg->type == PTR_TO_PACKET) || | |
11311 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
11312 | src_reg->type == PTR_TO_PACKET_META)) { | |
11313 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
11314 | find_good_pkt_pointers(other_branch, src_reg, | |
11315 | src_reg->type, false); | |
6d94e741 | 11316 | mark_pkt_end(this_branch, insn->src_reg, true); |
5beca081 DB |
11317 | } else { |
11318 | return false; | |
11319 | } | |
11320 | break; | |
11321 | case BPF_JLE: | |
11322 | if ((dst_reg->type == PTR_TO_PACKET && | |
11323 | src_reg->type == PTR_TO_PACKET_END) || | |
11324 | (dst_reg->type == PTR_TO_PACKET_META && | |
11325 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
11326 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
11327 | find_good_pkt_pointers(other_branch, dst_reg, | |
11328 | dst_reg->type, false); | |
6d94e741 | 11329 | mark_pkt_end(this_branch, insn->dst_reg, true); |
5beca081 DB |
11330 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
11331 | src_reg->type == PTR_TO_PACKET) || | |
11332 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
11333 | src_reg->type == PTR_TO_PACKET_META)) { | |
11334 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
11335 | find_good_pkt_pointers(this_branch, src_reg, | |
11336 | src_reg->type, true); | |
6d94e741 | 11337 | mark_pkt_end(other_branch, insn->src_reg, false); |
5beca081 DB |
11338 | } else { |
11339 | return false; | |
11340 | } | |
11341 | break; | |
11342 | default: | |
11343 | return false; | |
11344 | } | |
11345 | ||
11346 | return true; | |
11347 | } | |
11348 | ||
75748837 AS |
11349 | static void find_equal_scalars(struct bpf_verifier_state *vstate, |
11350 | struct bpf_reg_state *known_reg) | |
11351 | { | |
11352 | struct bpf_func_state *state; | |
11353 | struct bpf_reg_state *reg; | |
75748837 | 11354 | |
b239da34 KKD |
11355 | bpf_for_each_reg_in_vstate(vstate, state, reg, ({ |
11356 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
11357 | *reg = *known_reg; | |
11358 | })); | |
75748837 AS |
11359 | } |
11360 | ||
58e2af8b | 11361 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
11362 | struct bpf_insn *insn, int *insn_idx) |
11363 | { | |
f4d7e40a AS |
11364 | struct bpf_verifier_state *this_branch = env->cur_state; |
11365 | struct bpf_verifier_state *other_branch; | |
11366 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 11367 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
befae758 | 11368 | struct bpf_reg_state *eq_branch_regs; |
17a52670 | 11369 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 11370 | bool is_jmp32; |
fb8d251e | 11371 | int pred = -1; |
17a52670 AS |
11372 | int err; |
11373 | ||
092ed096 JW |
11374 | /* Only conditional jumps are expected to reach here. */ |
11375 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
11376 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
11377 | return -EINVAL; |
11378 | } | |
11379 | ||
11380 | if (BPF_SRC(insn->code) == BPF_X) { | |
11381 | if (insn->imm != 0) { | |
092ed096 | 11382 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
11383 | return -EINVAL; |
11384 | } | |
11385 | ||
11386 | /* check src1 operand */ | |
dc503a8a | 11387 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
11388 | if (err) |
11389 | return err; | |
1be7f75d AS |
11390 | |
11391 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 11392 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
11393 | insn->src_reg); |
11394 | return -EACCES; | |
11395 | } | |
fb8d251e | 11396 | src_reg = ®s[insn->src_reg]; |
17a52670 AS |
11397 | } else { |
11398 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 11399 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
11400 | return -EINVAL; |
11401 | } | |
11402 | } | |
11403 | ||
11404 | /* check src2 operand */ | |
dc503a8a | 11405 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
11406 | if (err) |
11407 | return err; | |
11408 | ||
1a0dc1ac | 11409 | dst_reg = ®s[insn->dst_reg]; |
092ed096 | 11410 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 11411 | |
3f50f132 JF |
11412 | if (BPF_SRC(insn->code) == BPF_K) { |
11413 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
11414 | } else if (src_reg->type == SCALAR_VALUE && | |
11415 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
11416 | pred = is_branch_taken(dst_reg, | |
11417 | tnum_subreg(src_reg->var_off).value, | |
11418 | opcode, | |
11419 | is_jmp32); | |
11420 | } else if (src_reg->type == SCALAR_VALUE && | |
11421 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
11422 | pred = is_branch_taken(dst_reg, | |
11423 | src_reg->var_off.value, | |
11424 | opcode, | |
11425 | is_jmp32); | |
6d94e741 AS |
11426 | } else if (reg_is_pkt_pointer_any(dst_reg) && |
11427 | reg_is_pkt_pointer_any(src_reg) && | |
11428 | !is_jmp32) { | |
11429 | pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode); | |
3f50f132 JF |
11430 | } |
11431 | ||
b5dc0163 | 11432 | if (pred >= 0) { |
cac616db JF |
11433 | /* If we get here with a dst_reg pointer type it is because |
11434 | * above is_branch_taken() special cased the 0 comparison. | |
11435 | */ | |
11436 | if (!__is_pointer_value(false, dst_reg)) | |
11437 | err = mark_chain_precision(env, insn->dst_reg); | |
6d94e741 AS |
11438 | if (BPF_SRC(insn->code) == BPF_X && !err && |
11439 | !__is_pointer_value(false, src_reg)) | |
b5dc0163 AS |
11440 | err = mark_chain_precision(env, insn->src_reg); |
11441 | if (err) | |
11442 | return err; | |
11443 | } | |
9183671a | 11444 | |
fb8d251e | 11445 | if (pred == 1) { |
9183671a DB |
11446 | /* Only follow the goto, ignore fall-through. If needed, push |
11447 | * the fall-through branch for simulation under speculative | |
11448 | * execution. | |
11449 | */ | |
11450 | if (!env->bypass_spec_v1 && | |
11451 | !sanitize_speculative_path(env, insn, *insn_idx + 1, | |
11452 | *insn_idx)) | |
11453 | return -EFAULT; | |
fb8d251e AS |
11454 | *insn_idx += insn->off; |
11455 | return 0; | |
11456 | } else if (pred == 0) { | |
9183671a DB |
11457 | /* Only follow the fall-through branch, since that's where the |
11458 | * program will go. If needed, push the goto branch for | |
11459 | * simulation under speculative execution. | |
fb8d251e | 11460 | */ |
9183671a DB |
11461 | if (!env->bypass_spec_v1 && |
11462 | !sanitize_speculative_path(env, insn, | |
11463 | *insn_idx + insn->off + 1, | |
11464 | *insn_idx)) | |
11465 | return -EFAULT; | |
fb8d251e | 11466 | return 0; |
17a52670 AS |
11467 | } |
11468 | ||
979d63d5 DB |
11469 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
11470 | false); | |
17a52670 AS |
11471 | if (!other_branch) |
11472 | return -EFAULT; | |
f4d7e40a | 11473 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 11474 | |
48461135 JB |
11475 | /* detect if we are comparing against a constant value so we can adjust |
11476 | * our min/max values for our dst register. | |
f1174f77 | 11477 | * this is only legit if both are scalars (or pointers to the same |
befae758 EZ |
11478 | * object, I suppose, see the PTR_MAYBE_NULL related if block below), |
11479 | * because otherwise the different base pointers mean the offsets aren't | |
f1174f77 | 11480 | * comparable. |
48461135 JB |
11481 | */ |
11482 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 11483 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 11484 | |
f1174f77 | 11485 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
11486 | src_reg->type == SCALAR_VALUE) { |
11487 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
11488 | (is_jmp32 && |
11489 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 11490 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 11491 | dst_reg, |
3f50f132 JF |
11492 | src_reg->var_off.value, |
11493 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
11494 | opcode, is_jmp32); |
11495 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
11496 | (is_jmp32 && |
11497 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 11498 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 11499 | src_reg, |
3f50f132 JF |
11500 | dst_reg->var_off.value, |
11501 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
11502 | opcode, is_jmp32); |
11503 | else if (!is_jmp32 && | |
11504 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 11505 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
11506 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
11507 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 11508 | src_reg, dst_reg, opcode); |
e688c3db AS |
11509 | if (src_reg->id && |
11510 | !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) { | |
75748837 AS |
11511 | find_equal_scalars(this_branch, src_reg); |
11512 | find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]); | |
11513 | } | |
11514 | ||
f1174f77 EC |
11515 | } |
11516 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 11517 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
11518 | dst_reg, insn->imm, (u32)insn->imm, |
11519 | opcode, is_jmp32); | |
48461135 JB |
11520 | } |
11521 | ||
e688c3db AS |
11522 | if (dst_reg->type == SCALAR_VALUE && dst_reg->id && |
11523 | !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) { | |
75748837 AS |
11524 | find_equal_scalars(this_branch, dst_reg); |
11525 | find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]); | |
11526 | } | |
11527 | ||
befae758 EZ |
11528 | /* if one pointer register is compared to another pointer |
11529 | * register check if PTR_MAYBE_NULL could be lifted. | |
11530 | * E.g. register A - maybe null | |
11531 | * register B - not null | |
11532 | * for JNE A, B, ... - A is not null in the false branch; | |
11533 | * for JEQ A, B, ... - A is not null in the true branch. | |
11534 | */ | |
11535 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_X && | |
11536 | __is_pointer_value(false, src_reg) && __is_pointer_value(false, dst_reg) && | |
11537 | type_may_be_null(src_reg->type) != type_may_be_null(dst_reg->type)) { | |
11538 | eq_branch_regs = NULL; | |
11539 | switch (opcode) { | |
11540 | case BPF_JEQ: | |
11541 | eq_branch_regs = other_branch_regs; | |
11542 | break; | |
11543 | case BPF_JNE: | |
11544 | eq_branch_regs = regs; | |
11545 | break; | |
11546 | default: | |
11547 | /* do nothing */ | |
11548 | break; | |
11549 | } | |
11550 | if (eq_branch_regs) { | |
11551 | if (type_may_be_null(src_reg->type)) | |
11552 | mark_ptr_not_null_reg(&eq_branch_regs[insn->src_reg]); | |
11553 | else | |
11554 | mark_ptr_not_null_reg(&eq_branch_regs[insn->dst_reg]); | |
11555 | } | |
11556 | } | |
11557 | ||
092ed096 JW |
11558 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
11559 | * NOTE: these optimizations below are related with pointer comparison | |
11560 | * which will never be JMP32. | |
11561 | */ | |
11562 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 11563 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
c25b2ae1 | 11564 | type_may_be_null(dst_reg->type)) { |
840b9615 | 11565 | /* Mark all identical registers in each branch as either |
57a09bf0 TG |
11566 | * safe or unknown depending R == 0 or R != 0 conditional. |
11567 | */ | |
840b9615 JS |
11568 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
11569 | opcode == BPF_JNE); | |
11570 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
11571 | opcode == BPF_JEQ); | |
5beca081 DB |
11572 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
11573 | this_branch, other_branch) && | |
11574 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
11575 | verbose(env, "R%d pointer comparison prohibited\n", |
11576 | insn->dst_reg); | |
1be7f75d | 11577 | return -EACCES; |
17a52670 | 11578 | } |
06ee7115 | 11579 | if (env->log.level & BPF_LOG_LEVEL) |
2e576648 | 11580 | print_insn_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
11581 | return 0; |
11582 | } | |
11583 | ||
17a52670 | 11584 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 11585 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 11586 | { |
d8eca5bb | 11587 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 11588 | struct bpf_reg_state *regs = cur_regs(env); |
4976b718 | 11589 | struct bpf_reg_state *dst_reg; |
d8eca5bb | 11590 | struct bpf_map *map; |
17a52670 AS |
11591 | int err; |
11592 | ||
11593 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 11594 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
11595 | return -EINVAL; |
11596 | } | |
11597 | if (insn->off != 0) { | |
61bd5218 | 11598 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
11599 | return -EINVAL; |
11600 | } | |
11601 | ||
dc503a8a | 11602 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
11603 | if (err) |
11604 | return err; | |
11605 | ||
4976b718 | 11606 | dst_reg = ®s[insn->dst_reg]; |
6b173873 | 11607 | if (insn->src_reg == 0) { |
6b173873 JK |
11608 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
11609 | ||
4976b718 | 11610 | dst_reg->type = SCALAR_VALUE; |
b03c9f9f | 11611 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 11612 | return 0; |
6b173873 | 11613 | } |
17a52670 | 11614 | |
d400a6cf DB |
11615 | /* All special src_reg cases are listed below. From this point onwards |
11616 | * we either succeed and assign a corresponding dst_reg->type after | |
11617 | * zeroing the offset, or fail and reject the program. | |
11618 | */ | |
11619 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
4976b718 | 11620 | |
d400a6cf | 11621 | if (insn->src_reg == BPF_PSEUDO_BTF_ID) { |
4976b718 | 11622 | dst_reg->type = aux->btf_var.reg_type; |
34d3a78c | 11623 | switch (base_type(dst_reg->type)) { |
4976b718 HL |
11624 | case PTR_TO_MEM: |
11625 | dst_reg->mem_size = aux->btf_var.mem_size; | |
11626 | break; | |
11627 | case PTR_TO_BTF_ID: | |
22dc4a0f | 11628 | dst_reg->btf = aux->btf_var.btf; |
4976b718 HL |
11629 | dst_reg->btf_id = aux->btf_var.btf_id; |
11630 | break; | |
11631 | default: | |
11632 | verbose(env, "bpf verifier is misconfigured\n"); | |
11633 | return -EFAULT; | |
11634 | } | |
11635 | return 0; | |
11636 | } | |
11637 | ||
69c087ba YS |
11638 | if (insn->src_reg == BPF_PSEUDO_FUNC) { |
11639 | struct bpf_prog_aux *aux = env->prog->aux; | |
3990ed4c MKL |
11640 | u32 subprogno = find_subprog(env, |
11641 | env->insn_idx + insn->imm + 1); | |
69c087ba YS |
11642 | |
11643 | if (!aux->func_info) { | |
11644 | verbose(env, "missing btf func_info\n"); | |
11645 | return -EINVAL; | |
11646 | } | |
11647 | if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) { | |
11648 | verbose(env, "callback function not static\n"); | |
11649 | return -EINVAL; | |
11650 | } | |
11651 | ||
11652 | dst_reg->type = PTR_TO_FUNC; | |
11653 | dst_reg->subprogno = subprogno; | |
11654 | return 0; | |
11655 | } | |
11656 | ||
d8eca5bb | 11657 | map = env->used_maps[aux->map_index]; |
4976b718 | 11658 | dst_reg->map_ptr = map; |
d8eca5bb | 11659 | |
387544bf AS |
11660 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE || |
11661 | insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE) { | |
4976b718 HL |
11662 | dst_reg->type = PTR_TO_MAP_VALUE; |
11663 | dst_reg->off = aux->map_off; | |
d0d78c1d KKD |
11664 | WARN_ON_ONCE(map->max_entries != 1); |
11665 | /* We want reg->id to be same (0) as map_value is not distinct */ | |
387544bf AS |
11666 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD || |
11667 | insn->src_reg == BPF_PSEUDO_MAP_IDX) { | |
4976b718 | 11668 | dst_reg->type = CONST_PTR_TO_MAP; |
d8eca5bb DB |
11669 | } else { |
11670 | verbose(env, "bpf verifier is misconfigured\n"); | |
11671 | return -EINVAL; | |
11672 | } | |
17a52670 | 11673 | |
17a52670 AS |
11674 | return 0; |
11675 | } | |
11676 | ||
96be4325 DB |
11677 | static bool may_access_skb(enum bpf_prog_type type) |
11678 | { | |
11679 | switch (type) { | |
11680 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
11681 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 11682 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
11683 | return true; |
11684 | default: | |
11685 | return false; | |
11686 | } | |
11687 | } | |
11688 | ||
ddd872bc AS |
11689 | /* verify safety of LD_ABS|LD_IND instructions: |
11690 | * - they can only appear in the programs where ctx == skb | |
11691 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
11692 | * preserve R6-R9, and store return value into R0 | |
11693 | * | |
11694 | * Implicit input: | |
11695 | * ctx == skb == R6 == CTX | |
11696 | * | |
11697 | * Explicit input: | |
11698 | * SRC == any register | |
11699 | * IMM == 32-bit immediate | |
11700 | * | |
11701 | * Output: | |
11702 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
11703 | */ | |
58e2af8b | 11704 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 11705 | { |
638f5b90 | 11706 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 11707 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 11708 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
11709 | int i, err; |
11710 | ||
7e40781c | 11711 | if (!may_access_skb(resolve_prog_type(env->prog))) { |
61bd5218 | 11712 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
11713 | return -EINVAL; |
11714 | } | |
11715 | ||
e0cea7ce DB |
11716 | if (!env->ops->gen_ld_abs) { |
11717 | verbose(env, "bpf verifier is misconfigured\n"); | |
11718 | return -EINVAL; | |
11719 | } | |
11720 | ||
ddd872bc | 11721 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 11722 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 11723 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 11724 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
11725 | return -EINVAL; |
11726 | } | |
11727 | ||
11728 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 11729 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
11730 | if (err) |
11731 | return err; | |
11732 | ||
fd978bf7 JS |
11733 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
11734 | * gen_ld_abs() may terminate the program at runtime, leading to | |
11735 | * reference leak. | |
11736 | */ | |
11737 | err = check_reference_leak(env); | |
11738 | if (err) { | |
11739 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
11740 | return err; | |
11741 | } | |
11742 | ||
d0d78c1d | 11743 | if (env->cur_state->active_lock.ptr) { |
d83525ca AS |
11744 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); |
11745 | return -EINVAL; | |
11746 | } | |
11747 | ||
6d4f151a | 11748 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
11749 | verbose(env, |
11750 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
11751 | return -EINVAL; |
11752 | } | |
11753 | ||
11754 | if (mode == BPF_IND) { | |
11755 | /* check explicit source operand */ | |
dc503a8a | 11756 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
11757 | if (err) |
11758 | return err; | |
11759 | } | |
11760 | ||
be80a1d3 | 11761 | err = check_ptr_off_reg(env, ®s[ctx_reg], ctx_reg); |
6d4f151a DB |
11762 | if (err < 0) |
11763 | return err; | |
11764 | ||
ddd872bc | 11765 | /* reset caller saved regs to unreadable */ |
dc503a8a | 11766 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 11767 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
11768 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
11769 | } | |
ddd872bc AS |
11770 | |
11771 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
11772 | * the value fetched from the packet. |
11773 | * Already marked as written above. | |
ddd872bc | 11774 | */ |
61bd5218 | 11775 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
11776 | /* ld_abs load up to 32-bit skb data. */ |
11777 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
11778 | return 0; |
11779 | } | |
11780 | ||
390ee7e2 AS |
11781 | static int check_return_code(struct bpf_verifier_env *env) |
11782 | { | |
5cf1e914 | 11783 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 11784 | const struct bpf_prog *prog = env->prog; |
390ee7e2 AS |
11785 | struct bpf_reg_state *reg; |
11786 | struct tnum range = tnum_range(0, 1); | |
7e40781c | 11787 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
27ae7997 | 11788 | int err; |
bfc6bb74 AS |
11789 | struct bpf_func_state *frame = env->cur_state->frame[0]; |
11790 | const bool is_subprog = frame->subprogno; | |
27ae7997 | 11791 | |
9e4e01df | 11792 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
d1a6edec SF |
11793 | if (!is_subprog) { |
11794 | switch (prog_type) { | |
11795 | case BPF_PROG_TYPE_LSM: | |
11796 | if (prog->expected_attach_type == BPF_LSM_CGROUP) | |
11797 | /* See below, can be 0 or 0-1 depending on hook. */ | |
11798 | break; | |
11799 | fallthrough; | |
11800 | case BPF_PROG_TYPE_STRUCT_OPS: | |
11801 | if (!prog->aux->attach_func_proto->type) | |
11802 | return 0; | |
11803 | break; | |
11804 | default: | |
11805 | break; | |
11806 | } | |
11807 | } | |
27ae7997 | 11808 | |
8fb33b60 | 11809 | /* eBPF calling convention is such that R0 is used |
27ae7997 MKL |
11810 | * to return the value from eBPF program. |
11811 | * Make sure that it's readable at this time | |
11812 | * of bpf_exit, which means that program wrote | |
11813 | * something into it earlier | |
11814 | */ | |
11815 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
11816 | if (err) | |
11817 | return err; | |
11818 | ||
11819 | if (is_pointer_value(env, BPF_REG_0)) { | |
11820 | verbose(env, "R0 leaks addr as return value\n"); | |
11821 | return -EACCES; | |
11822 | } | |
390ee7e2 | 11823 | |
f782e2c3 | 11824 | reg = cur_regs(env) + BPF_REG_0; |
bfc6bb74 AS |
11825 | |
11826 | if (frame->in_async_callback_fn) { | |
11827 | /* enforce return zero from async callbacks like timer */ | |
11828 | if (reg->type != SCALAR_VALUE) { | |
11829 | verbose(env, "In async callback the register R0 is not a known value (%s)\n", | |
c25b2ae1 | 11830 | reg_type_str(env, reg->type)); |
bfc6bb74 AS |
11831 | return -EINVAL; |
11832 | } | |
11833 | ||
11834 | if (!tnum_in(tnum_const(0), reg->var_off)) { | |
11835 | verbose_invalid_scalar(env, reg, &range, "async callback", "R0"); | |
11836 | return -EINVAL; | |
11837 | } | |
11838 | return 0; | |
11839 | } | |
11840 | ||
f782e2c3 DB |
11841 | if (is_subprog) { |
11842 | if (reg->type != SCALAR_VALUE) { | |
11843 | verbose(env, "At subprogram exit the register R0 is not a scalar value (%s)\n", | |
c25b2ae1 | 11844 | reg_type_str(env, reg->type)); |
f782e2c3 DB |
11845 | return -EINVAL; |
11846 | } | |
11847 | return 0; | |
11848 | } | |
11849 | ||
7e40781c | 11850 | switch (prog_type) { |
983695fa DB |
11851 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
11852 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 DB |
11853 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
11854 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || | |
11855 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
11856 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || | |
11857 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME) | |
983695fa | 11858 | range = tnum_range(1, 1); |
77241217 SF |
11859 | if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND || |
11860 | env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND) | |
11861 | range = tnum_range(0, 3); | |
ed4ed404 | 11862 | break; |
390ee7e2 | 11863 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 11864 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
11865 | range = tnum_range(0, 3); | |
11866 | enforce_attach_type_range = tnum_range(2, 3); | |
11867 | } | |
ed4ed404 | 11868 | break; |
390ee7e2 AS |
11869 | case BPF_PROG_TYPE_CGROUP_SOCK: |
11870 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 11871 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 11872 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 11873 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 11874 | break; |
15ab09bd AS |
11875 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
11876 | if (!env->prog->aux->attach_btf_id) | |
11877 | return 0; | |
11878 | range = tnum_const(0); | |
11879 | break; | |
15d83c4d | 11880 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
11881 | switch (env->prog->expected_attach_type) { |
11882 | case BPF_TRACE_FENTRY: | |
11883 | case BPF_TRACE_FEXIT: | |
11884 | range = tnum_const(0); | |
11885 | break; | |
11886 | case BPF_TRACE_RAW_TP: | |
11887 | case BPF_MODIFY_RETURN: | |
15d83c4d | 11888 | return 0; |
2ec0616e DB |
11889 | case BPF_TRACE_ITER: |
11890 | break; | |
e92888c7 YS |
11891 | default: |
11892 | return -ENOTSUPP; | |
11893 | } | |
15d83c4d | 11894 | break; |
e9ddbb77 JS |
11895 | case BPF_PROG_TYPE_SK_LOOKUP: |
11896 | range = tnum_range(SK_DROP, SK_PASS); | |
11897 | break; | |
69fd337a SF |
11898 | |
11899 | case BPF_PROG_TYPE_LSM: | |
11900 | if (env->prog->expected_attach_type != BPF_LSM_CGROUP) { | |
11901 | /* Regular BPF_PROG_TYPE_LSM programs can return | |
11902 | * any value. | |
11903 | */ | |
11904 | return 0; | |
11905 | } | |
11906 | if (!env->prog->aux->attach_func_proto->type) { | |
11907 | /* Make sure programs that attach to void | |
11908 | * hooks don't try to modify return value. | |
11909 | */ | |
11910 | range = tnum_range(1, 1); | |
11911 | } | |
11912 | break; | |
11913 | ||
e92888c7 YS |
11914 | case BPF_PROG_TYPE_EXT: |
11915 | /* freplace program can return anything as its return value | |
11916 | * depends on the to-be-replaced kernel func or bpf program. | |
11917 | */ | |
390ee7e2 AS |
11918 | default: |
11919 | return 0; | |
11920 | } | |
11921 | ||
390ee7e2 | 11922 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 11923 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
c25b2ae1 | 11924 | reg_type_str(env, reg->type)); |
390ee7e2 AS |
11925 | return -EINVAL; |
11926 | } | |
11927 | ||
11928 | if (!tnum_in(range, reg->var_off)) { | |
bc2591d6 | 11929 | verbose_invalid_scalar(env, reg, &range, "program exit", "R0"); |
69fd337a | 11930 | if (prog->expected_attach_type == BPF_LSM_CGROUP && |
d1a6edec | 11931 | prog_type == BPF_PROG_TYPE_LSM && |
69fd337a SF |
11932 | !prog->aux->attach_func_proto->type) |
11933 | verbose(env, "Note, BPF_LSM_CGROUP that attach to void LSM hooks can't modify return value!\n"); | |
390ee7e2 AS |
11934 | return -EINVAL; |
11935 | } | |
5cf1e914 | 11936 | |
11937 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
11938 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
11939 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
11940 | return 0; |
11941 | } | |
11942 | ||
475fb78f AS |
11943 | /* non-recursive DFS pseudo code |
11944 | * 1 procedure DFS-iterative(G,v): | |
11945 | * 2 label v as discovered | |
11946 | * 3 let S be a stack | |
11947 | * 4 S.push(v) | |
11948 | * 5 while S is not empty | |
b6d20799 | 11949 | * 6 t <- S.peek() |
475fb78f AS |
11950 | * 7 if t is what we're looking for: |
11951 | * 8 return t | |
11952 | * 9 for all edges e in G.adjacentEdges(t) do | |
11953 | * 10 if edge e is already labelled | |
11954 | * 11 continue with the next edge | |
11955 | * 12 w <- G.adjacentVertex(t,e) | |
11956 | * 13 if vertex w is not discovered and not explored | |
11957 | * 14 label e as tree-edge | |
11958 | * 15 label w as discovered | |
11959 | * 16 S.push(w) | |
11960 | * 17 continue at 5 | |
11961 | * 18 else if vertex w is discovered | |
11962 | * 19 label e as back-edge | |
11963 | * 20 else | |
11964 | * 21 // vertex w is explored | |
11965 | * 22 label e as forward- or cross-edge | |
11966 | * 23 label t as explored | |
11967 | * 24 S.pop() | |
11968 | * | |
11969 | * convention: | |
11970 | * 0x10 - discovered | |
11971 | * 0x11 - discovered and fall-through edge labelled | |
11972 | * 0x12 - discovered and fall-through and branch edges labelled | |
11973 | * 0x20 - explored | |
11974 | */ | |
11975 | ||
11976 | enum { | |
11977 | DISCOVERED = 0x10, | |
11978 | EXPLORED = 0x20, | |
11979 | FALLTHROUGH = 1, | |
11980 | BRANCH = 2, | |
11981 | }; | |
11982 | ||
dc2a4ebc AS |
11983 | static u32 state_htab_size(struct bpf_verifier_env *env) |
11984 | { | |
11985 | return env->prog->len; | |
11986 | } | |
11987 | ||
5d839021 AS |
11988 | static struct bpf_verifier_state_list **explored_state( |
11989 | struct bpf_verifier_env *env, | |
11990 | int idx) | |
11991 | { | |
dc2a4ebc AS |
11992 | struct bpf_verifier_state *cur = env->cur_state; |
11993 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
11994 | ||
11995 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
11996 | } |
11997 | ||
11998 | static void init_explored_state(struct bpf_verifier_env *env, int idx) | |
11999 | { | |
a8f500af | 12000 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 12001 | } |
f1bca824 | 12002 | |
59e2e27d WAF |
12003 | enum { |
12004 | DONE_EXPLORING = 0, | |
12005 | KEEP_EXPLORING = 1, | |
12006 | }; | |
12007 | ||
475fb78f AS |
12008 | /* t, w, e - match pseudo-code above: |
12009 | * t - index of current instruction | |
12010 | * w - next instruction | |
12011 | * e - edge | |
12012 | */ | |
2589726d AS |
12013 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env, |
12014 | bool loop_ok) | |
475fb78f | 12015 | { |
7df737e9 AS |
12016 | int *insn_stack = env->cfg.insn_stack; |
12017 | int *insn_state = env->cfg.insn_state; | |
12018 | ||
475fb78f | 12019 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
59e2e27d | 12020 | return DONE_EXPLORING; |
475fb78f AS |
12021 | |
12022 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
59e2e27d | 12023 | return DONE_EXPLORING; |
475fb78f AS |
12024 | |
12025 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 12026 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 12027 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
12028 | return -EINVAL; |
12029 | } | |
12030 | ||
f1bca824 AS |
12031 | if (e == BRANCH) |
12032 | /* mark branch target for state pruning */ | |
5d839021 | 12033 | init_explored_state(env, w); |
f1bca824 | 12034 | |
475fb78f AS |
12035 | if (insn_state[w] == 0) { |
12036 | /* tree-edge */ | |
12037 | insn_state[t] = DISCOVERED | e; | |
12038 | insn_state[w] = DISCOVERED; | |
7df737e9 | 12039 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 12040 | return -E2BIG; |
7df737e9 | 12041 | insn_stack[env->cfg.cur_stack++] = w; |
59e2e27d | 12042 | return KEEP_EXPLORING; |
475fb78f | 12043 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { |
2c78ee89 | 12044 | if (loop_ok && env->bpf_capable) |
59e2e27d | 12045 | return DONE_EXPLORING; |
d9762e84 MKL |
12046 | verbose_linfo(env, t, "%d: ", t); |
12047 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 12048 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
12049 | return -EINVAL; |
12050 | } else if (insn_state[w] == EXPLORED) { | |
12051 | /* forward- or cross-edge */ | |
12052 | insn_state[t] = DISCOVERED | e; | |
12053 | } else { | |
61bd5218 | 12054 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
12055 | return -EFAULT; |
12056 | } | |
59e2e27d WAF |
12057 | return DONE_EXPLORING; |
12058 | } | |
12059 | ||
efdb22de YS |
12060 | static int visit_func_call_insn(int t, int insn_cnt, |
12061 | struct bpf_insn *insns, | |
12062 | struct bpf_verifier_env *env, | |
12063 | bool visit_callee) | |
12064 | { | |
12065 | int ret; | |
12066 | ||
12067 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); | |
12068 | if (ret) | |
12069 | return ret; | |
12070 | ||
12071 | if (t + 1 < insn_cnt) | |
12072 | init_explored_state(env, t + 1); | |
12073 | if (visit_callee) { | |
12074 | init_explored_state(env, t); | |
86fc6ee6 AS |
12075 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env, |
12076 | /* It's ok to allow recursion from CFG point of | |
12077 | * view. __check_func_call() will do the actual | |
12078 | * check. | |
12079 | */ | |
12080 | bpf_pseudo_func(insns + t)); | |
efdb22de YS |
12081 | } |
12082 | return ret; | |
12083 | } | |
12084 | ||
59e2e27d WAF |
12085 | /* Visits the instruction at index t and returns one of the following: |
12086 | * < 0 - an error occurred | |
12087 | * DONE_EXPLORING - the instruction was fully explored | |
12088 | * KEEP_EXPLORING - there is still work to be done before it is fully explored | |
12089 | */ | |
12090 | static int visit_insn(int t, int insn_cnt, struct bpf_verifier_env *env) | |
12091 | { | |
12092 | struct bpf_insn *insns = env->prog->insnsi; | |
12093 | int ret; | |
12094 | ||
69c087ba YS |
12095 | if (bpf_pseudo_func(insns + t)) |
12096 | return visit_func_call_insn(t, insn_cnt, insns, env, true); | |
12097 | ||
59e2e27d WAF |
12098 | /* All non-branch instructions have a single fall-through edge. */ |
12099 | if (BPF_CLASS(insns[t].code) != BPF_JMP && | |
12100 | BPF_CLASS(insns[t].code) != BPF_JMP32) | |
12101 | return push_insn(t, t + 1, FALLTHROUGH, env, false); | |
12102 | ||
12103 | switch (BPF_OP(insns[t].code)) { | |
12104 | case BPF_EXIT: | |
12105 | return DONE_EXPLORING; | |
12106 | ||
12107 | case BPF_CALL: | |
bfc6bb74 AS |
12108 | if (insns[t].imm == BPF_FUNC_timer_set_callback) |
12109 | /* Mark this call insn to trigger is_state_visited() check | |
12110 | * before call itself is processed by __check_func_call(). | |
12111 | * Otherwise new async state will be pushed for further | |
12112 | * exploration. | |
12113 | */ | |
12114 | init_explored_state(env, t); | |
efdb22de YS |
12115 | return visit_func_call_insn(t, insn_cnt, insns, env, |
12116 | insns[t].src_reg == BPF_PSEUDO_CALL); | |
59e2e27d WAF |
12117 | |
12118 | case BPF_JA: | |
12119 | if (BPF_SRC(insns[t].code) != BPF_K) | |
12120 | return -EINVAL; | |
12121 | ||
12122 | /* unconditional jump with single edge */ | |
12123 | ret = push_insn(t, t + insns[t].off + 1, FALLTHROUGH, env, | |
12124 | true); | |
12125 | if (ret) | |
12126 | return ret; | |
12127 | ||
12128 | /* unconditional jmp is not a good pruning point, | |
12129 | * but it's marked, since backtracking needs | |
12130 | * to record jmp history in is_state_visited(). | |
12131 | */ | |
12132 | init_explored_state(env, t + insns[t].off + 1); | |
12133 | /* tell verifier to check for equivalent states | |
12134 | * after every call and jump | |
12135 | */ | |
12136 | if (t + 1 < insn_cnt) | |
12137 | init_explored_state(env, t + 1); | |
12138 | ||
12139 | return ret; | |
12140 | ||
12141 | default: | |
12142 | /* conditional jump with two edges */ | |
12143 | init_explored_state(env, t); | |
12144 | ret = push_insn(t, t + 1, FALLTHROUGH, env, true); | |
12145 | if (ret) | |
12146 | return ret; | |
12147 | ||
12148 | return push_insn(t, t + insns[t].off + 1, BRANCH, env, true); | |
12149 | } | |
475fb78f AS |
12150 | } |
12151 | ||
12152 | /* non-recursive depth-first-search to detect loops in BPF program | |
12153 | * loop == back-edge in directed graph | |
12154 | */ | |
58e2af8b | 12155 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f | 12156 | { |
475fb78f | 12157 | int insn_cnt = env->prog->len; |
7df737e9 | 12158 | int *insn_stack, *insn_state; |
475fb78f | 12159 | int ret = 0; |
59e2e27d | 12160 | int i; |
475fb78f | 12161 | |
7df737e9 | 12162 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
12163 | if (!insn_state) |
12164 | return -ENOMEM; | |
12165 | ||
7df737e9 | 12166 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 12167 | if (!insn_stack) { |
71dde681 | 12168 | kvfree(insn_state); |
475fb78f AS |
12169 | return -ENOMEM; |
12170 | } | |
12171 | ||
12172 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
12173 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 12174 | env->cfg.cur_stack = 1; |
475fb78f | 12175 | |
59e2e27d WAF |
12176 | while (env->cfg.cur_stack > 0) { |
12177 | int t = insn_stack[env->cfg.cur_stack - 1]; | |
475fb78f | 12178 | |
59e2e27d WAF |
12179 | ret = visit_insn(t, insn_cnt, env); |
12180 | switch (ret) { | |
12181 | case DONE_EXPLORING: | |
12182 | insn_state[t] = EXPLORED; | |
12183 | env->cfg.cur_stack--; | |
12184 | break; | |
12185 | case KEEP_EXPLORING: | |
12186 | break; | |
12187 | default: | |
12188 | if (ret > 0) { | |
12189 | verbose(env, "visit_insn internal bug\n"); | |
12190 | ret = -EFAULT; | |
475fb78f | 12191 | } |
475fb78f | 12192 | goto err_free; |
59e2e27d | 12193 | } |
475fb78f AS |
12194 | } |
12195 | ||
59e2e27d | 12196 | if (env->cfg.cur_stack < 0) { |
61bd5218 | 12197 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
12198 | ret = -EFAULT; |
12199 | goto err_free; | |
12200 | } | |
475fb78f | 12201 | |
475fb78f AS |
12202 | for (i = 0; i < insn_cnt; i++) { |
12203 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 12204 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
12205 | ret = -EINVAL; |
12206 | goto err_free; | |
12207 | } | |
12208 | } | |
12209 | ret = 0; /* cfg looks good */ | |
12210 | ||
12211 | err_free: | |
71dde681 AS |
12212 | kvfree(insn_state); |
12213 | kvfree(insn_stack); | |
7df737e9 | 12214 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
12215 | return ret; |
12216 | } | |
12217 | ||
09b28d76 AS |
12218 | static int check_abnormal_return(struct bpf_verifier_env *env) |
12219 | { | |
12220 | int i; | |
12221 | ||
12222 | for (i = 1; i < env->subprog_cnt; i++) { | |
12223 | if (env->subprog_info[i].has_ld_abs) { | |
12224 | verbose(env, "LD_ABS is not allowed in subprogs without BTF\n"); | |
12225 | return -EINVAL; | |
12226 | } | |
12227 | if (env->subprog_info[i].has_tail_call) { | |
12228 | verbose(env, "tail_call is not allowed in subprogs without BTF\n"); | |
12229 | return -EINVAL; | |
12230 | } | |
12231 | } | |
12232 | return 0; | |
12233 | } | |
12234 | ||
838e9690 YS |
12235 | /* The minimum supported BTF func info size */ |
12236 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
12237 | #define MAX_FUNCINFO_REC_SIZE 252 | |
12238 | ||
c454a46b MKL |
12239 | static int check_btf_func(struct bpf_verifier_env *env, |
12240 | const union bpf_attr *attr, | |
af2ac3e1 | 12241 | bpfptr_t uattr) |
838e9690 | 12242 | { |
09b28d76 | 12243 | const struct btf_type *type, *func_proto, *ret_type; |
d0b2818e | 12244 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 12245 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 12246 | struct bpf_func_info *krecord; |
8c1b6e69 | 12247 | struct bpf_func_info_aux *info_aux = NULL; |
c454a46b MKL |
12248 | struct bpf_prog *prog; |
12249 | const struct btf *btf; | |
af2ac3e1 | 12250 | bpfptr_t urecord; |
d0b2818e | 12251 | u32 prev_offset = 0; |
09b28d76 | 12252 | bool scalar_return; |
e7ed83d6 | 12253 | int ret = -ENOMEM; |
838e9690 YS |
12254 | |
12255 | nfuncs = attr->func_info_cnt; | |
09b28d76 AS |
12256 | if (!nfuncs) { |
12257 | if (check_abnormal_return(env)) | |
12258 | return -EINVAL; | |
838e9690 | 12259 | return 0; |
09b28d76 | 12260 | } |
838e9690 YS |
12261 | |
12262 | if (nfuncs != env->subprog_cnt) { | |
12263 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
12264 | return -EINVAL; | |
12265 | } | |
12266 | ||
12267 | urec_size = attr->func_info_rec_size; | |
12268 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
12269 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
12270 | urec_size % sizeof(u32)) { | |
12271 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
12272 | return -EINVAL; | |
12273 | } | |
12274 | ||
c454a46b MKL |
12275 | prog = env->prog; |
12276 | btf = prog->aux->btf; | |
838e9690 | 12277 | |
af2ac3e1 | 12278 | urecord = make_bpfptr(attr->func_info, uattr.is_kernel); |
838e9690 YS |
12279 | min_size = min_t(u32, krec_size, urec_size); |
12280 | ||
ba64e7d8 | 12281 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
12282 | if (!krecord) |
12283 | return -ENOMEM; | |
8c1b6e69 AS |
12284 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); |
12285 | if (!info_aux) | |
12286 | goto err_free; | |
ba64e7d8 | 12287 | |
838e9690 YS |
12288 | for (i = 0; i < nfuncs; i++) { |
12289 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
12290 | if (ret) { | |
12291 | if (ret == -E2BIG) { | |
12292 | verbose(env, "nonzero tailing record in func info"); | |
12293 | /* set the size kernel expects so loader can zero | |
12294 | * out the rest of the record. | |
12295 | */ | |
af2ac3e1 AS |
12296 | if (copy_to_bpfptr_offset(uattr, |
12297 | offsetof(union bpf_attr, func_info_rec_size), | |
12298 | &min_size, sizeof(min_size))) | |
838e9690 YS |
12299 | ret = -EFAULT; |
12300 | } | |
c454a46b | 12301 | goto err_free; |
838e9690 YS |
12302 | } |
12303 | ||
af2ac3e1 | 12304 | if (copy_from_bpfptr(&krecord[i], urecord, min_size)) { |
838e9690 | 12305 | ret = -EFAULT; |
c454a46b | 12306 | goto err_free; |
838e9690 YS |
12307 | } |
12308 | ||
d30d42e0 | 12309 | /* check insn_off */ |
09b28d76 | 12310 | ret = -EINVAL; |
838e9690 | 12311 | if (i == 0) { |
d30d42e0 | 12312 | if (krecord[i].insn_off) { |
838e9690 | 12313 | verbose(env, |
d30d42e0 MKL |
12314 | "nonzero insn_off %u for the first func info record", |
12315 | krecord[i].insn_off); | |
c454a46b | 12316 | goto err_free; |
838e9690 | 12317 | } |
d30d42e0 | 12318 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
12319 | verbose(env, |
12320 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 12321 | krecord[i].insn_off, prev_offset); |
c454a46b | 12322 | goto err_free; |
838e9690 YS |
12323 | } |
12324 | ||
d30d42e0 | 12325 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 | 12326 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
c454a46b | 12327 | goto err_free; |
838e9690 YS |
12328 | } |
12329 | ||
12330 | /* check type_id */ | |
ba64e7d8 | 12331 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 12332 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 12333 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 12334 | krecord[i].type_id); |
c454a46b | 12335 | goto err_free; |
838e9690 | 12336 | } |
51c39bb1 | 12337 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); |
09b28d76 AS |
12338 | |
12339 | func_proto = btf_type_by_id(btf, type->type); | |
12340 | if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto))) | |
12341 | /* btf_func_check() already verified it during BTF load */ | |
12342 | goto err_free; | |
12343 | ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL); | |
12344 | scalar_return = | |
6089fb32 | 12345 | btf_type_is_small_int(ret_type) || btf_is_any_enum(ret_type); |
09b28d76 AS |
12346 | if (i && !scalar_return && env->subprog_info[i].has_ld_abs) { |
12347 | verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n"); | |
12348 | goto err_free; | |
12349 | } | |
12350 | if (i && !scalar_return && env->subprog_info[i].has_tail_call) { | |
12351 | verbose(env, "tail_call is only allowed in functions that return 'int'.\n"); | |
12352 | goto err_free; | |
12353 | } | |
12354 | ||
d30d42e0 | 12355 | prev_offset = krecord[i].insn_off; |
af2ac3e1 | 12356 | bpfptr_add(&urecord, urec_size); |
838e9690 YS |
12357 | } |
12358 | ||
ba64e7d8 YS |
12359 | prog->aux->func_info = krecord; |
12360 | prog->aux->func_info_cnt = nfuncs; | |
8c1b6e69 | 12361 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
12362 | return 0; |
12363 | ||
c454a46b | 12364 | err_free: |
ba64e7d8 | 12365 | kvfree(krecord); |
8c1b6e69 | 12366 | kfree(info_aux); |
838e9690 YS |
12367 | return ret; |
12368 | } | |
12369 | ||
ba64e7d8 YS |
12370 | static void adjust_btf_func(struct bpf_verifier_env *env) |
12371 | { | |
8c1b6e69 | 12372 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
12373 | int i; |
12374 | ||
8c1b6e69 | 12375 | if (!aux->func_info) |
ba64e7d8 YS |
12376 | return; |
12377 | ||
12378 | for (i = 0; i < env->subprog_cnt; i++) | |
8c1b6e69 | 12379 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
12380 | } |
12381 | ||
1b773d00 | 12382 | #define MIN_BPF_LINEINFO_SIZE offsetofend(struct bpf_line_info, line_col) |
c454a46b MKL |
12383 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE |
12384 | ||
12385 | static int check_btf_line(struct bpf_verifier_env *env, | |
12386 | const union bpf_attr *attr, | |
af2ac3e1 | 12387 | bpfptr_t uattr) |
c454a46b MKL |
12388 | { |
12389 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
12390 | struct bpf_subprog_info *sub; | |
12391 | struct bpf_line_info *linfo; | |
12392 | struct bpf_prog *prog; | |
12393 | const struct btf *btf; | |
af2ac3e1 | 12394 | bpfptr_t ulinfo; |
c454a46b MKL |
12395 | int err; |
12396 | ||
12397 | nr_linfo = attr->line_info_cnt; | |
12398 | if (!nr_linfo) | |
12399 | return 0; | |
0e6491b5 BC |
12400 | if (nr_linfo > INT_MAX / sizeof(struct bpf_line_info)) |
12401 | return -EINVAL; | |
c454a46b MKL |
12402 | |
12403 | rec_size = attr->line_info_rec_size; | |
12404 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
12405 | rec_size > MAX_LINEINFO_REC_SIZE || | |
12406 | rec_size & (sizeof(u32) - 1)) | |
12407 | return -EINVAL; | |
12408 | ||
12409 | /* Need to zero it in case the userspace may | |
12410 | * pass in a smaller bpf_line_info object. | |
12411 | */ | |
12412 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
12413 | GFP_KERNEL | __GFP_NOWARN); | |
12414 | if (!linfo) | |
12415 | return -ENOMEM; | |
12416 | ||
12417 | prog = env->prog; | |
12418 | btf = prog->aux->btf; | |
12419 | ||
12420 | s = 0; | |
12421 | sub = env->subprog_info; | |
af2ac3e1 | 12422 | ulinfo = make_bpfptr(attr->line_info, uattr.is_kernel); |
c454a46b MKL |
12423 | expected_size = sizeof(struct bpf_line_info); |
12424 | ncopy = min_t(u32, expected_size, rec_size); | |
12425 | for (i = 0; i < nr_linfo; i++) { | |
12426 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
12427 | if (err) { | |
12428 | if (err == -E2BIG) { | |
12429 | verbose(env, "nonzero tailing record in line_info"); | |
af2ac3e1 AS |
12430 | if (copy_to_bpfptr_offset(uattr, |
12431 | offsetof(union bpf_attr, line_info_rec_size), | |
12432 | &expected_size, sizeof(expected_size))) | |
c454a46b MKL |
12433 | err = -EFAULT; |
12434 | } | |
12435 | goto err_free; | |
12436 | } | |
12437 | ||
af2ac3e1 | 12438 | if (copy_from_bpfptr(&linfo[i], ulinfo, ncopy)) { |
c454a46b MKL |
12439 | err = -EFAULT; |
12440 | goto err_free; | |
12441 | } | |
12442 | ||
12443 | /* | |
12444 | * Check insn_off to ensure | |
12445 | * 1) strictly increasing AND | |
12446 | * 2) bounded by prog->len | |
12447 | * | |
12448 | * The linfo[0].insn_off == 0 check logically falls into | |
12449 | * the later "missing bpf_line_info for func..." case | |
12450 | * because the first linfo[0].insn_off must be the | |
12451 | * first sub also and the first sub must have | |
12452 | * subprog_info[0].start == 0. | |
12453 | */ | |
12454 | if ((i && linfo[i].insn_off <= prev_offset) || | |
12455 | linfo[i].insn_off >= prog->len) { | |
12456 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
12457 | i, linfo[i].insn_off, prev_offset, | |
12458 | prog->len); | |
12459 | err = -EINVAL; | |
12460 | goto err_free; | |
12461 | } | |
12462 | ||
fdbaa0be MKL |
12463 | if (!prog->insnsi[linfo[i].insn_off].code) { |
12464 | verbose(env, | |
12465 | "Invalid insn code at line_info[%u].insn_off\n", | |
12466 | i); | |
12467 | err = -EINVAL; | |
12468 | goto err_free; | |
12469 | } | |
12470 | ||
23127b33 MKL |
12471 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
12472 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
12473 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
12474 | err = -EINVAL; | |
12475 | goto err_free; | |
12476 | } | |
12477 | ||
12478 | if (s != env->subprog_cnt) { | |
12479 | if (linfo[i].insn_off == sub[s].start) { | |
12480 | sub[s].linfo_idx = i; | |
12481 | s++; | |
12482 | } else if (sub[s].start < linfo[i].insn_off) { | |
12483 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
12484 | err = -EINVAL; | |
12485 | goto err_free; | |
12486 | } | |
12487 | } | |
12488 | ||
12489 | prev_offset = linfo[i].insn_off; | |
af2ac3e1 | 12490 | bpfptr_add(&ulinfo, rec_size); |
c454a46b MKL |
12491 | } |
12492 | ||
12493 | if (s != env->subprog_cnt) { | |
12494 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
12495 | env->subprog_cnt - s, s); | |
12496 | err = -EINVAL; | |
12497 | goto err_free; | |
12498 | } | |
12499 | ||
12500 | prog->aux->linfo = linfo; | |
12501 | prog->aux->nr_linfo = nr_linfo; | |
12502 | ||
12503 | return 0; | |
12504 | ||
12505 | err_free: | |
12506 | kvfree(linfo); | |
12507 | return err; | |
12508 | } | |
12509 | ||
fbd94c7a AS |
12510 | #define MIN_CORE_RELO_SIZE sizeof(struct bpf_core_relo) |
12511 | #define MAX_CORE_RELO_SIZE MAX_FUNCINFO_REC_SIZE | |
12512 | ||
12513 | static int check_core_relo(struct bpf_verifier_env *env, | |
12514 | const union bpf_attr *attr, | |
12515 | bpfptr_t uattr) | |
12516 | { | |
12517 | u32 i, nr_core_relo, ncopy, expected_size, rec_size; | |
12518 | struct bpf_core_relo core_relo = {}; | |
12519 | struct bpf_prog *prog = env->prog; | |
12520 | const struct btf *btf = prog->aux->btf; | |
12521 | struct bpf_core_ctx ctx = { | |
12522 | .log = &env->log, | |
12523 | .btf = btf, | |
12524 | }; | |
12525 | bpfptr_t u_core_relo; | |
12526 | int err; | |
12527 | ||
12528 | nr_core_relo = attr->core_relo_cnt; | |
12529 | if (!nr_core_relo) | |
12530 | return 0; | |
12531 | if (nr_core_relo > INT_MAX / sizeof(struct bpf_core_relo)) | |
12532 | return -EINVAL; | |
12533 | ||
12534 | rec_size = attr->core_relo_rec_size; | |
12535 | if (rec_size < MIN_CORE_RELO_SIZE || | |
12536 | rec_size > MAX_CORE_RELO_SIZE || | |
12537 | rec_size % sizeof(u32)) | |
12538 | return -EINVAL; | |
12539 | ||
12540 | u_core_relo = make_bpfptr(attr->core_relos, uattr.is_kernel); | |
12541 | expected_size = sizeof(struct bpf_core_relo); | |
12542 | ncopy = min_t(u32, expected_size, rec_size); | |
12543 | ||
12544 | /* Unlike func_info and line_info, copy and apply each CO-RE | |
12545 | * relocation record one at a time. | |
12546 | */ | |
12547 | for (i = 0; i < nr_core_relo; i++) { | |
12548 | /* future proofing when sizeof(bpf_core_relo) changes */ | |
12549 | err = bpf_check_uarg_tail_zero(u_core_relo, expected_size, rec_size); | |
12550 | if (err) { | |
12551 | if (err == -E2BIG) { | |
12552 | verbose(env, "nonzero tailing record in core_relo"); | |
12553 | if (copy_to_bpfptr_offset(uattr, | |
12554 | offsetof(union bpf_attr, core_relo_rec_size), | |
12555 | &expected_size, sizeof(expected_size))) | |
12556 | err = -EFAULT; | |
12557 | } | |
12558 | break; | |
12559 | } | |
12560 | ||
12561 | if (copy_from_bpfptr(&core_relo, u_core_relo, ncopy)) { | |
12562 | err = -EFAULT; | |
12563 | break; | |
12564 | } | |
12565 | ||
12566 | if (core_relo.insn_off % 8 || core_relo.insn_off / 8 >= prog->len) { | |
12567 | verbose(env, "Invalid core_relo[%u].insn_off:%u prog->len:%u\n", | |
12568 | i, core_relo.insn_off, prog->len); | |
12569 | err = -EINVAL; | |
12570 | break; | |
12571 | } | |
12572 | ||
12573 | err = bpf_core_apply(&ctx, &core_relo, i, | |
12574 | &prog->insnsi[core_relo.insn_off / 8]); | |
12575 | if (err) | |
12576 | break; | |
12577 | bpfptr_add(&u_core_relo, rec_size); | |
12578 | } | |
12579 | return err; | |
12580 | } | |
12581 | ||
c454a46b MKL |
12582 | static int check_btf_info(struct bpf_verifier_env *env, |
12583 | const union bpf_attr *attr, | |
af2ac3e1 | 12584 | bpfptr_t uattr) |
c454a46b MKL |
12585 | { |
12586 | struct btf *btf; | |
12587 | int err; | |
12588 | ||
09b28d76 AS |
12589 | if (!attr->func_info_cnt && !attr->line_info_cnt) { |
12590 | if (check_abnormal_return(env)) | |
12591 | return -EINVAL; | |
c454a46b | 12592 | return 0; |
09b28d76 | 12593 | } |
c454a46b MKL |
12594 | |
12595 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
12596 | if (IS_ERR(btf)) | |
12597 | return PTR_ERR(btf); | |
350a5c4d AS |
12598 | if (btf_is_kernel(btf)) { |
12599 | btf_put(btf); | |
12600 | return -EACCES; | |
12601 | } | |
c454a46b MKL |
12602 | env->prog->aux->btf = btf; |
12603 | ||
12604 | err = check_btf_func(env, attr, uattr); | |
12605 | if (err) | |
12606 | return err; | |
12607 | ||
12608 | err = check_btf_line(env, attr, uattr); | |
12609 | if (err) | |
12610 | return err; | |
12611 | ||
fbd94c7a AS |
12612 | err = check_core_relo(env, attr, uattr); |
12613 | if (err) | |
12614 | return err; | |
12615 | ||
c454a46b | 12616 | return 0; |
ba64e7d8 YS |
12617 | } |
12618 | ||
f1174f77 EC |
12619 | /* check %cur's range satisfies %old's */ |
12620 | static bool range_within(struct bpf_reg_state *old, | |
12621 | struct bpf_reg_state *cur) | |
12622 | { | |
b03c9f9f EC |
12623 | return old->umin_value <= cur->umin_value && |
12624 | old->umax_value >= cur->umax_value && | |
12625 | old->smin_value <= cur->smin_value && | |
fd675184 DB |
12626 | old->smax_value >= cur->smax_value && |
12627 | old->u32_min_value <= cur->u32_min_value && | |
12628 | old->u32_max_value >= cur->u32_max_value && | |
12629 | old->s32_min_value <= cur->s32_min_value && | |
12630 | old->s32_max_value >= cur->s32_max_value; | |
f1174f77 EC |
12631 | } |
12632 | ||
f1174f77 EC |
12633 | /* If in the old state two registers had the same id, then they need to have |
12634 | * the same id in the new state as well. But that id could be different from | |
12635 | * the old state, so we need to track the mapping from old to new ids. | |
12636 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
12637 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
12638 | * regs with a different old id could still have new id 9, we don't care about | |
12639 | * that. | |
12640 | * So we look through our idmap to see if this old id has been seen before. If | |
12641 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 12642 | */ |
c9e73e3d | 12643 | static bool check_ids(u32 old_id, u32 cur_id, struct bpf_id_pair *idmap) |
969bf05e | 12644 | { |
f1174f77 | 12645 | unsigned int i; |
969bf05e | 12646 | |
c9e73e3d | 12647 | for (i = 0; i < BPF_ID_MAP_SIZE; i++) { |
f1174f77 EC |
12648 | if (!idmap[i].old) { |
12649 | /* Reached an empty slot; haven't seen this id before */ | |
12650 | idmap[i].old = old_id; | |
12651 | idmap[i].cur = cur_id; | |
12652 | return true; | |
12653 | } | |
12654 | if (idmap[i].old == old_id) | |
12655 | return idmap[i].cur == cur_id; | |
12656 | } | |
12657 | /* We ran out of idmap slots, which should be impossible */ | |
12658 | WARN_ON_ONCE(1); | |
12659 | return false; | |
12660 | } | |
12661 | ||
9242b5f5 AS |
12662 | static void clean_func_state(struct bpf_verifier_env *env, |
12663 | struct bpf_func_state *st) | |
12664 | { | |
12665 | enum bpf_reg_liveness live; | |
12666 | int i, j; | |
12667 | ||
12668 | for (i = 0; i < BPF_REG_FP; i++) { | |
12669 | live = st->regs[i].live; | |
12670 | /* liveness must not touch this register anymore */ | |
12671 | st->regs[i].live |= REG_LIVE_DONE; | |
12672 | if (!(live & REG_LIVE_READ)) | |
12673 | /* since the register is unused, clear its state | |
12674 | * to make further comparison simpler | |
12675 | */ | |
f54c7898 | 12676 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
12677 | } |
12678 | ||
12679 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
12680 | live = st->stack[i].spilled_ptr.live; | |
12681 | /* liveness must not touch this stack slot anymore */ | |
12682 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
12683 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 12684 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
12685 | for (j = 0; j < BPF_REG_SIZE; j++) |
12686 | st->stack[i].slot_type[j] = STACK_INVALID; | |
12687 | } | |
12688 | } | |
12689 | } | |
12690 | ||
12691 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
12692 | struct bpf_verifier_state *st) | |
12693 | { | |
12694 | int i; | |
12695 | ||
12696 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
12697 | /* all regs in this state in all frames were already marked */ | |
12698 | return; | |
12699 | ||
12700 | for (i = 0; i <= st->curframe; i++) | |
12701 | clean_func_state(env, st->frame[i]); | |
12702 | } | |
12703 | ||
12704 | /* the parentage chains form a tree. | |
12705 | * the verifier states are added to state lists at given insn and | |
12706 | * pushed into state stack for future exploration. | |
12707 | * when the verifier reaches bpf_exit insn some of the verifer states | |
12708 | * stored in the state lists have their final liveness state already, | |
12709 | * but a lot of states will get revised from liveness point of view when | |
12710 | * the verifier explores other branches. | |
12711 | * Example: | |
12712 | * 1: r0 = 1 | |
12713 | * 2: if r1 == 100 goto pc+1 | |
12714 | * 3: r0 = 2 | |
12715 | * 4: exit | |
12716 | * when the verifier reaches exit insn the register r0 in the state list of | |
12717 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
12718 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
12719 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
12720 | * | |
12721 | * Since the verifier pushes the branch states as it sees them while exploring | |
12722 | * the program the condition of walking the branch instruction for the second | |
12723 | * time means that all states below this branch were already explored and | |
8fb33b60 | 12724 | * their final liveness marks are already propagated. |
9242b5f5 AS |
12725 | * Hence when the verifier completes the search of state list in is_state_visited() |
12726 | * we can call this clean_live_states() function to mark all liveness states | |
12727 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
12728 | * will not be used. | |
12729 | * This function also clears the registers and stack for states that !READ | |
12730 | * to simplify state merging. | |
12731 | * | |
12732 | * Important note here that walking the same branch instruction in the callee | |
12733 | * doesn't meant that the states are DONE. The verifier has to compare | |
12734 | * the callsites | |
12735 | */ | |
12736 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
12737 | struct bpf_verifier_state *cur) | |
12738 | { | |
12739 | struct bpf_verifier_state_list *sl; | |
12740 | int i; | |
12741 | ||
5d839021 | 12742 | sl = *explored_state(env, insn); |
a8f500af | 12743 | while (sl) { |
2589726d AS |
12744 | if (sl->state.branches) |
12745 | goto next; | |
dc2a4ebc AS |
12746 | if (sl->state.insn_idx != insn || |
12747 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
12748 | goto next; |
12749 | for (i = 0; i <= cur->curframe; i++) | |
12750 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
12751 | goto next; | |
12752 | clean_verifier_state(env, &sl->state); | |
12753 | next: | |
12754 | sl = sl->next; | |
12755 | } | |
12756 | } | |
12757 | ||
f1174f77 | 12758 | /* Returns true if (rold safe implies rcur safe) */ |
e042aa53 DB |
12759 | static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold, |
12760 | struct bpf_reg_state *rcur, struct bpf_id_pair *idmap) | |
f1174f77 | 12761 | { |
f4d7e40a AS |
12762 | bool equal; |
12763 | ||
dc503a8a EC |
12764 | if (!(rold->live & REG_LIVE_READ)) |
12765 | /* explored state didn't use this */ | |
12766 | return true; | |
12767 | ||
679c782d | 12768 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
12769 | |
12770 | if (rold->type == PTR_TO_STACK) | |
12771 | /* two stack pointers are equal only if they're pointing to | |
12772 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
12773 | */ | |
12774 | return equal && rold->frameno == rcur->frameno; | |
12775 | ||
12776 | if (equal) | |
969bf05e AS |
12777 | return true; |
12778 | ||
f1174f77 EC |
12779 | if (rold->type == NOT_INIT) |
12780 | /* explored state can't have used this */ | |
969bf05e | 12781 | return true; |
f1174f77 EC |
12782 | if (rcur->type == NOT_INIT) |
12783 | return false; | |
c25b2ae1 | 12784 | switch (base_type(rold->type)) { |
f1174f77 | 12785 | case SCALAR_VALUE: |
e042aa53 DB |
12786 | if (env->explore_alu_limits) |
12787 | return false; | |
f1174f77 | 12788 | if (rcur->type == SCALAR_VALUE) { |
f63181b6 | 12789 | if (!rold->precise) |
b5dc0163 | 12790 | return true; |
f1174f77 EC |
12791 | /* new val must satisfy old val knowledge */ |
12792 | return range_within(rold, rcur) && | |
12793 | tnum_in(rold->var_off, rcur->var_off); | |
12794 | } else { | |
179d1c56 JH |
12795 | /* We're trying to use a pointer in place of a scalar. |
12796 | * Even if the scalar was unbounded, this could lead to | |
12797 | * pointer leaks because scalars are allowed to leak | |
12798 | * while pointers are not. We could make this safe in | |
12799 | * special cases if root is calling us, but it's | |
12800 | * probably not worth the hassle. | |
f1174f77 | 12801 | */ |
179d1c56 | 12802 | return false; |
f1174f77 | 12803 | } |
69c087ba | 12804 | case PTR_TO_MAP_KEY: |
f1174f77 | 12805 | case PTR_TO_MAP_VALUE: |
c25b2ae1 HL |
12806 | /* a PTR_TO_MAP_VALUE could be safe to use as a |
12807 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
12808 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
12809 | * checked, doing so could have affected others with the same | |
12810 | * id, and we can't check for that because we lost the id when | |
12811 | * we converted to a PTR_TO_MAP_VALUE. | |
12812 | */ | |
12813 | if (type_may_be_null(rold->type)) { | |
12814 | if (!type_may_be_null(rcur->type)) | |
12815 | return false; | |
12816 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
12817 | return false; | |
12818 | /* Check our ids match any regs they're supposed to */ | |
12819 | return check_ids(rold->id, rcur->id, idmap); | |
12820 | } | |
12821 | ||
1b688a19 EC |
12822 | /* If the new min/max/var_off satisfy the old ones and |
12823 | * everything else matches, we are OK. | |
d83525ca AS |
12824 | * 'id' is not compared, since it's only used for maps with |
12825 | * bpf_spin_lock inside map element and in such cases if | |
12826 | * the rest of the prog is valid for one map element then | |
12827 | * it's valid for all map elements regardless of the key | |
12828 | * used in bpf_map_lookup() | |
1b688a19 EC |
12829 | */ |
12830 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
12831 | range_within(rold, rcur) && | |
12832 | tnum_in(rold->var_off, rcur->var_off); | |
de8f3a83 | 12833 | case PTR_TO_PACKET_META: |
f1174f77 | 12834 | case PTR_TO_PACKET: |
de8f3a83 | 12835 | if (rcur->type != rold->type) |
f1174f77 EC |
12836 | return false; |
12837 | /* We must have at least as much range as the old ptr | |
12838 | * did, so that any accesses which were safe before are | |
12839 | * still safe. This is true even if old range < old off, | |
12840 | * since someone could have accessed through (ptr - k), or | |
12841 | * even done ptr -= k in a register, to get a safe access. | |
12842 | */ | |
12843 | if (rold->range > rcur->range) | |
12844 | return false; | |
12845 | /* If the offsets don't match, we can't trust our alignment; | |
12846 | * nor can we be sure that we won't fall out of range. | |
12847 | */ | |
12848 | if (rold->off != rcur->off) | |
12849 | return false; | |
12850 | /* id relations must be preserved */ | |
12851 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
12852 | return false; | |
12853 | /* new val must satisfy old val knowledge */ | |
12854 | return range_within(rold, rcur) && | |
12855 | tnum_in(rold->var_off, rcur->var_off); | |
12856 | case PTR_TO_CTX: | |
12857 | case CONST_PTR_TO_MAP: | |
f1174f77 | 12858 | case PTR_TO_PACKET_END: |
d58e468b | 12859 | case PTR_TO_FLOW_KEYS: |
c64b7983 | 12860 | case PTR_TO_SOCKET: |
46f8bc92 | 12861 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 12862 | case PTR_TO_TCP_SOCK: |
fada7fdc | 12863 | case PTR_TO_XDP_SOCK: |
f1174f77 EC |
12864 | /* Only valid matches are exact, which memcmp() above |
12865 | * would have accepted | |
12866 | */ | |
12867 | default: | |
12868 | /* Don't know what's going on, just say it's not safe */ | |
12869 | return false; | |
12870 | } | |
969bf05e | 12871 | |
f1174f77 EC |
12872 | /* Shouldn't get here; if we do, say it's not safe */ |
12873 | WARN_ON_ONCE(1); | |
969bf05e AS |
12874 | return false; |
12875 | } | |
12876 | ||
e042aa53 DB |
12877 | static bool stacksafe(struct bpf_verifier_env *env, struct bpf_func_state *old, |
12878 | struct bpf_func_state *cur, struct bpf_id_pair *idmap) | |
638f5b90 AS |
12879 | { |
12880 | int i, spi; | |
12881 | ||
638f5b90 AS |
12882 | /* walk slots of the explored stack and ignore any additional |
12883 | * slots in the current stack, since explored(safe) state | |
12884 | * didn't use them | |
12885 | */ | |
12886 | for (i = 0; i < old->allocated_stack; i++) { | |
12887 | spi = i / BPF_REG_SIZE; | |
12888 | ||
b233920c AS |
12889 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
12890 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 12891 | /* explored state didn't use this */ |
fd05e57b | 12892 | continue; |
b233920c | 12893 | } |
cc2b14d5 | 12894 | |
638f5b90 AS |
12895 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
12896 | continue; | |
19e2dbb7 AS |
12897 | |
12898 | /* explored stack has more populated slots than current stack | |
12899 | * and these slots were used | |
12900 | */ | |
12901 | if (i >= cur->allocated_stack) | |
12902 | return false; | |
12903 | ||
cc2b14d5 AS |
12904 | /* if old state was safe with misc data in the stack |
12905 | * it will be safe with zero-initialized stack. | |
12906 | * The opposite is not true | |
12907 | */ | |
12908 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
12909 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
12910 | continue; | |
638f5b90 AS |
12911 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
12912 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
12913 | /* Ex: old explored (safe) state has STACK_SPILL in | |
b8c1a309 | 12914 | * this stack slot, but current has STACK_MISC -> |
638f5b90 AS |
12915 | * this verifier states are not equivalent, |
12916 | * return false to continue verification of this path | |
12917 | */ | |
12918 | return false; | |
27113c59 | 12919 | if (i % BPF_REG_SIZE != BPF_REG_SIZE - 1) |
638f5b90 | 12920 | continue; |
27113c59 | 12921 | if (!is_spilled_reg(&old->stack[spi])) |
638f5b90 | 12922 | continue; |
e042aa53 DB |
12923 | if (!regsafe(env, &old->stack[spi].spilled_ptr, |
12924 | &cur->stack[spi].spilled_ptr, idmap)) | |
638f5b90 AS |
12925 | /* when explored and current stack slot are both storing |
12926 | * spilled registers, check that stored pointers types | |
12927 | * are the same as well. | |
12928 | * Ex: explored safe path could have stored | |
12929 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
12930 | * but current path has stored: | |
12931 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
12932 | * such verifier states are not equivalent. | |
12933 | * return false to continue verification of this path | |
12934 | */ | |
12935 | return false; | |
12936 | } | |
12937 | return true; | |
12938 | } | |
12939 | ||
fd978bf7 JS |
12940 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
12941 | { | |
12942 | if (old->acquired_refs != cur->acquired_refs) | |
12943 | return false; | |
12944 | return !memcmp(old->refs, cur->refs, | |
12945 | sizeof(*old->refs) * old->acquired_refs); | |
12946 | } | |
12947 | ||
f1bca824 AS |
12948 | /* compare two verifier states |
12949 | * | |
12950 | * all states stored in state_list are known to be valid, since | |
12951 | * verifier reached 'bpf_exit' instruction through them | |
12952 | * | |
12953 | * this function is called when verifier exploring different branches of | |
12954 | * execution popped from the state stack. If it sees an old state that has | |
12955 | * more strict register state and more strict stack state then this execution | |
12956 | * branch doesn't need to be explored further, since verifier already | |
12957 | * concluded that more strict state leads to valid finish. | |
12958 | * | |
12959 | * Therefore two states are equivalent if register state is more conservative | |
12960 | * and explored stack state is more conservative than the current one. | |
12961 | * Example: | |
12962 | * explored current | |
12963 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
12964 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
12965 | * | |
12966 | * In other words if current stack state (one being explored) has more | |
12967 | * valid slots than old one that already passed validation, it means | |
12968 | * the verifier can stop exploring and conclude that current state is valid too | |
12969 | * | |
12970 | * Similarly with registers. If explored state has register type as invalid | |
12971 | * whereas register type in current state is meaningful, it means that | |
12972 | * the current state will reach 'bpf_exit' instruction safely | |
12973 | */ | |
c9e73e3d | 12974 | static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_state *old, |
f4d7e40a | 12975 | struct bpf_func_state *cur) |
f1bca824 AS |
12976 | { |
12977 | int i; | |
12978 | ||
c9e73e3d LB |
12979 | memset(env->idmap_scratch, 0, sizeof(env->idmap_scratch)); |
12980 | for (i = 0; i < MAX_BPF_REG; i++) | |
e042aa53 DB |
12981 | if (!regsafe(env, &old->regs[i], &cur->regs[i], |
12982 | env->idmap_scratch)) | |
c9e73e3d | 12983 | return false; |
f1bca824 | 12984 | |
e042aa53 | 12985 | if (!stacksafe(env, old, cur, env->idmap_scratch)) |
c9e73e3d | 12986 | return false; |
fd978bf7 JS |
12987 | |
12988 | if (!refsafe(old, cur)) | |
c9e73e3d LB |
12989 | return false; |
12990 | ||
12991 | return true; | |
f1bca824 AS |
12992 | } |
12993 | ||
f4d7e40a AS |
12994 | static bool states_equal(struct bpf_verifier_env *env, |
12995 | struct bpf_verifier_state *old, | |
12996 | struct bpf_verifier_state *cur) | |
12997 | { | |
12998 | int i; | |
12999 | ||
13000 | if (old->curframe != cur->curframe) | |
13001 | return false; | |
13002 | ||
979d63d5 DB |
13003 | /* Verification state from speculative execution simulation |
13004 | * must never prune a non-speculative execution one. | |
13005 | */ | |
13006 | if (old->speculative && !cur->speculative) | |
13007 | return false; | |
13008 | ||
d0d78c1d KKD |
13009 | if (old->active_lock.ptr != cur->active_lock.ptr || |
13010 | old->active_lock.id != cur->active_lock.id) | |
d83525ca AS |
13011 | return false; |
13012 | ||
f4d7e40a AS |
13013 | /* for states to be equal callsites have to be the same |
13014 | * and all frame states need to be equivalent | |
13015 | */ | |
13016 | for (i = 0; i <= old->curframe; i++) { | |
13017 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
13018 | return false; | |
c9e73e3d | 13019 | if (!func_states_equal(env, old->frame[i], cur->frame[i])) |
f4d7e40a AS |
13020 | return false; |
13021 | } | |
13022 | return true; | |
13023 | } | |
13024 | ||
5327ed3d JW |
13025 | /* Return 0 if no propagation happened. Return negative error code if error |
13026 | * happened. Otherwise, return the propagated bit. | |
13027 | */ | |
55e7f3b5 JW |
13028 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
13029 | struct bpf_reg_state *reg, | |
13030 | struct bpf_reg_state *parent_reg) | |
13031 | { | |
5327ed3d JW |
13032 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
13033 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
13034 | int err; |
13035 | ||
5327ed3d JW |
13036 | /* When comes here, read flags of PARENT_REG or REG could be any of |
13037 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
13038 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
13039 | */ | |
13040 | if (parent_flag == REG_LIVE_READ64 || | |
13041 | /* Or if there is no read flag from REG. */ | |
13042 | !flag || | |
13043 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
13044 | parent_flag == flag) | |
55e7f3b5 JW |
13045 | return 0; |
13046 | ||
5327ed3d | 13047 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
13048 | if (err) |
13049 | return err; | |
13050 | ||
5327ed3d | 13051 | return flag; |
55e7f3b5 JW |
13052 | } |
13053 | ||
8e9cd9ce | 13054 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
13055 | * straight-line code between a state and its parent. When we arrive at an |
13056 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
13057 | * code, so read marks in the state must propagate to the parent regardless | |
13058 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 13059 | * in mark_reg_read() is for. |
8e9cd9ce | 13060 | */ |
f4d7e40a AS |
13061 | static int propagate_liveness(struct bpf_verifier_env *env, |
13062 | const struct bpf_verifier_state *vstate, | |
13063 | struct bpf_verifier_state *vparent) | |
dc503a8a | 13064 | { |
3f8cafa4 | 13065 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 13066 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 13067 | int i, frame, err = 0; |
dc503a8a | 13068 | |
f4d7e40a AS |
13069 | if (vparent->curframe != vstate->curframe) { |
13070 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
13071 | vparent->curframe, vstate->curframe); | |
13072 | return -EFAULT; | |
13073 | } | |
dc503a8a EC |
13074 | /* Propagate read liveness of registers... */ |
13075 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 13076 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
13077 | parent = vparent->frame[frame]; |
13078 | state = vstate->frame[frame]; | |
13079 | parent_reg = parent->regs; | |
13080 | state_reg = state->regs; | |
83d16312 JK |
13081 | /* We don't need to worry about FP liveness, it's read-only */ |
13082 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
13083 | err = propagate_liveness_reg(env, &state_reg[i], |
13084 | &parent_reg[i]); | |
5327ed3d | 13085 | if (err < 0) |
3f8cafa4 | 13086 | return err; |
5327ed3d JW |
13087 | if (err == REG_LIVE_READ64) |
13088 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 13089 | } |
f4d7e40a | 13090 | |
1b04aee7 | 13091 | /* Propagate stack slots. */ |
f4d7e40a AS |
13092 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
13093 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
13094 | parent_reg = &parent->stack[i].spilled_ptr; |
13095 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
13096 | err = propagate_liveness_reg(env, state_reg, |
13097 | parent_reg); | |
5327ed3d | 13098 | if (err < 0) |
3f8cafa4 | 13099 | return err; |
dc503a8a EC |
13100 | } |
13101 | } | |
5327ed3d | 13102 | return 0; |
dc503a8a EC |
13103 | } |
13104 | ||
a3ce685d AS |
13105 | /* find precise scalars in the previous equivalent state and |
13106 | * propagate them into the current state | |
13107 | */ | |
13108 | static int propagate_precision(struct bpf_verifier_env *env, | |
13109 | const struct bpf_verifier_state *old) | |
13110 | { | |
13111 | struct bpf_reg_state *state_reg; | |
13112 | struct bpf_func_state *state; | |
529409ea | 13113 | int i, err = 0, fr; |
a3ce685d | 13114 | |
529409ea AN |
13115 | for (fr = old->curframe; fr >= 0; fr--) { |
13116 | state = old->frame[fr]; | |
13117 | state_reg = state->regs; | |
13118 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { | |
13119 | if (state_reg->type != SCALAR_VALUE || | |
13120 | !state_reg->precise) | |
13121 | continue; | |
13122 | if (env->log.level & BPF_LOG_LEVEL2) | |
13123 | verbose(env, "frame %d: propagating r%d\n", i, fr); | |
13124 | err = mark_chain_precision_frame(env, fr, i); | |
13125 | if (err < 0) | |
13126 | return err; | |
13127 | } | |
a3ce685d | 13128 | |
529409ea AN |
13129 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
13130 | if (!is_spilled_reg(&state->stack[i])) | |
13131 | continue; | |
13132 | state_reg = &state->stack[i].spilled_ptr; | |
13133 | if (state_reg->type != SCALAR_VALUE || | |
13134 | !state_reg->precise) | |
13135 | continue; | |
13136 | if (env->log.level & BPF_LOG_LEVEL2) | |
13137 | verbose(env, "frame %d: propagating fp%d\n", | |
13138 | (-i - 1) * BPF_REG_SIZE, fr); | |
13139 | err = mark_chain_precision_stack_frame(env, fr, i); | |
13140 | if (err < 0) | |
13141 | return err; | |
13142 | } | |
a3ce685d AS |
13143 | } |
13144 | return 0; | |
13145 | } | |
13146 | ||
2589726d AS |
13147 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
13148 | struct bpf_verifier_state *cur) | |
13149 | { | |
13150 | struct bpf_func_state *fold, *fcur; | |
13151 | int i, fr = cur->curframe; | |
13152 | ||
13153 | if (old->curframe != fr) | |
13154 | return false; | |
13155 | ||
13156 | fold = old->frame[fr]; | |
13157 | fcur = cur->frame[fr]; | |
13158 | for (i = 0; i < MAX_BPF_REG; i++) | |
13159 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
13160 | offsetof(struct bpf_reg_state, parent))) | |
13161 | return false; | |
13162 | return true; | |
13163 | } | |
13164 | ||
13165 | ||
58e2af8b | 13166 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 13167 | { |
58e2af8b | 13168 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 13169 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 13170 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 13171 | int i, j, err, states_cnt = 0; |
10d274e8 | 13172 | bool add_new_state = env->test_state_freq ? true : false; |
f1bca824 | 13173 | |
b5dc0163 | 13174 | cur->last_insn_idx = env->prev_insn_idx; |
a8f500af | 13175 | if (!env->insn_aux_data[insn_idx].prune_point) |
f1bca824 AS |
13176 | /* this 'insn_idx' instruction wasn't marked, so we will not |
13177 | * be doing state search here | |
13178 | */ | |
13179 | return 0; | |
13180 | ||
2589726d AS |
13181 | /* bpf progs typically have pruning point every 4 instructions |
13182 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
13183 | * Do not add new state for future pruning if the verifier hasn't seen | |
13184 | * at least 2 jumps and at least 8 instructions. | |
13185 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
13186 | * In tests that amounts to up to 50% reduction into total verifier | |
13187 | * memory consumption and 20% verifier time speedup. | |
13188 | */ | |
13189 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
13190 | env->insn_processed - env->prev_insn_processed >= 8) | |
13191 | add_new_state = true; | |
13192 | ||
a8f500af AS |
13193 | pprev = explored_state(env, insn_idx); |
13194 | sl = *pprev; | |
13195 | ||
9242b5f5 AS |
13196 | clean_live_states(env, insn_idx, cur); |
13197 | ||
a8f500af | 13198 | while (sl) { |
dc2a4ebc AS |
13199 | states_cnt++; |
13200 | if (sl->state.insn_idx != insn_idx) | |
13201 | goto next; | |
bfc6bb74 | 13202 | |
2589726d | 13203 | if (sl->state.branches) { |
bfc6bb74 AS |
13204 | struct bpf_func_state *frame = sl->state.frame[sl->state.curframe]; |
13205 | ||
13206 | if (frame->in_async_callback_fn && | |
13207 | frame->async_entry_cnt != cur->frame[cur->curframe]->async_entry_cnt) { | |
13208 | /* Different async_entry_cnt means that the verifier is | |
13209 | * processing another entry into async callback. | |
13210 | * Seeing the same state is not an indication of infinite | |
13211 | * loop or infinite recursion. | |
13212 | * But finding the same state doesn't mean that it's safe | |
13213 | * to stop processing the current state. The previous state | |
13214 | * hasn't yet reached bpf_exit, since state.branches > 0. | |
13215 | * Checking in_async_callback_fn alone is not enough either. | |
13216 | * Since the verifier still needs to catch infinite loops | |
13217 | * inside async callbacks. | |
13218 | */ | |
13219 | } else if (states_maybe_looping(&sl->state, cur) && | |
13220 | states_equal(env, &sl->state, cur)) { | |
2589726d AS |
13221 | verbose_linfo(env, insn_idx, "; "); |
13222 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
13223 | return -EINVAL; | |
13224 | } | |
13225 | /* if the verifier is processing a loop, avoid adding new state | |
13226 | * too often, since different loop iterations have distinct | |
13227 | * states and may not help future pruning. | |
13228 | * This threshold shouldn't be too low to make sure that | |
13229 | * a loop with large bound will be rejected quickly. | |
13230 | * The most abusive loop will be: | |
13231 | * r1 += 1 | |
13232 | * if r1 < 1000000 goto pc-2 | |
13233 | * 1M insn_procssed limit / 100 == 10k peak states. | |
13234 | * This threshold shouldn't be too high either, since states | |
13235 | * at the end of the loop are likely to be useful in pruning. | |
13236 | */ | |
13237 | if (env->jmps_processed - env->prev_jmps_processed < 20 && | |
13238 | env->insn_processed - env->prev_insn_processed < 100) | |
13239 | add_new_state = false; | |
13240 | goto miss; | |
13241 | } | |
638f5b90 | 13242 | if (states_equal(env, &sl->state, cur)) { |
9f4686c4 | 13243 | sl->hit_cnt++; |
f1bca824 | 13244 | /* reached equivalent register/stack state, |
dc503a8a EC |
13245 | * prune the search. |
13246 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
13247 | * If we have any write marks in env->cur_state, they |
13248 | * will prevent corresponding reads in the continuation | |
13249 | * from reaching our parent (an explored_state). Our | |
13250 | * own state will get the read marks recorded, but | |
13251 | * they'll be immediately forgotten as we're pruning | |
13252 | * this state and will pop a new one. | |
f1bca824 | 13253 | */ |
f4d7e40a | 13254 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
13255 | |
13256 | /* if previous state reached the exit with precision and | |
13257 | * current state is equivalent to it (except precsion marks) | |
13258 | * the precision needs to be propagated back in | |
13259 | * the current state. | |
13260 | */ | |
13261 | err = err ? : push_jmp_history(env, cur); | |
13262 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
13263 | if (err) |
13264 | return err; | |
f1bca824 | 13265 | return 1; |
dc503a8a | 13266 | } |
2589726d AS |
13267 | miss: |
13268 | /* when new state is not going to be added do not increase miss count. | |
13269 | * Otherwise several loop iterations will remove the state | |
13270 | * recorded earlier. The goal of these heuristics is to have | |
13271 | * states from some iterations of the loop (some in the beginning | |
13272 | * and some at the end) to help pruning. | |
13273 | */ | |
13274 | if (add_new_state) | |
13275 | sl->miss_cnt++; | |
9f4686c4 AS |
13276 | /* heuristic to determine whether this state is beneficial |
13277 | * to keep checking from state equivalence point of view. | |
13278 | * Higher numbers increase max_states_per_insn and verification time, | |
13279 | * but do not meaningfully decrease insn_processed. | |
13280 | */ | |
13281 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
13282 | /* the state is unlikely to be useful. Remove it to | |
13283 | * speed up verification | |
13284 | */ | |
13285 | *pprev = sl->next; | |
13286 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
2589726d AS |
13287 | u32 br = sl->state.branches; |
13288 | ||
13289 | WARN_ONCE(br, | |
13290 | "BUG live_done but branches_to_explore %d\n", | |
13291 | br); | |
9f4686c4 AS |
13292 | free_verifier_state(&sl->state, false); |
13293 | kfree(sl); | |
13294 | env->peak_states--; | |
13295 | } else { | |
13296 | /* cannot free this state, since parentage chain may | |
13297 | * walk it later. Add it for free_list instead to | |
13298 | * be freed at the end of verification | |
13299 | */ | |
13300 | sl->next = env->free_list; | |
13301 | env->free_list = sl; | |
13302 | } | |
13303 | sl = *pprev; | |
13304 | continue; | |
13305 | } | |
dc2a4ebc | 13306 | next: |
9f4686c4 AS |
13307 | pprev = &sl->next; |
13308 | sl = *pprev; | |
f1bca824 AS |
13309 | } |
13310 | ||
06ee7115 AS |
13311 | if (env->max_states_per_insn < states_cnt) |
13312 | env->max_states_per_insn = states_cnt; | |
13313 | ||
2c78ee89 | 13314 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
b5dc0163 | 13315 | return push_jmp_history(env, cur); |
ceefbc96 | 13316 | |
2589726d | 13317 | if (!add_new_state) |
b5dc0163 | 13318 | return push_jmp_history(env, cur); |
ceefbc96 | 13319 | |
2589726d AS |
13320 | /* There were no equivalent states, remember the current one. |
13321 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 13322 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 13323 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 13324 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
13325 | * again on the way to bpf_exit. |
13326 | * When looping the sl->state.branches will be > 0 and this state | |
13327 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 13328 | */ |
638f5b90 | 13329 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
13330 | if (!new_sl) |
13331 | return -ENOMEM; | |
06ee7115 AS |
13332 | env->total_states++; |
13333 | env->peak_states++; | |
2589726d AS |
13334 | env->prev_jmps_processed = env->jmps_processed; |
13335 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 | 13336 | |
7a830b53 AN |
13337 | /* forget precise markings we inherited, see __mark_chain_precision */ |
13338 | if (env->bpf_capable) | |
13339 | mark_all_scalars_imprecise(env, cur); | |
13340 | ||
f1bca824 | 13341 | /* add new state to the head of linked list */ |
679c782d EC |
13342 | new = &new_sl->state; |
13343 | err = copy_verifier_state(new, cur); | |
1969db47 | 13344 | if (err) { |
679c782d | 13345 | free_verifier_state(new, false); |
1969db47 AS |
13346 | kfree(new_sl); |
13347 | return err; | |
13348 | } | |
dc2a4ebc | 13349 | new->insn_idx = insn_idx; |
2589726d AS |
13350 | WARN_ONCE(new->branches != 1, |
13351 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 13352 | |
2589726d | 13353 | cur->parent = new; |
b5dc0163 AS |
13354 | cur->first_insn_idx = insn_idx; |
13355 | clear_jmp_history(cur); | |
5d839021 AS |
13356 | new_sl->next = *explored_state(env, insn_idx); |
13357 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
13358 | /* connect new state to parentage chain. Current frame needs all |
13359 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
13360 | * to the stack implicitly by JITs) so in callers' frames connect just | |
13361 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
13362 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
13363 | * from callee with its full parentage chain, anyway. | |
13364 | */ | |
8e9cd9ce EC |
13365 | /* clear write marks in current state: the writes we did are not writes |
13366 | * our child did, so they don't screen off its reads from us. | |
13367 | * (There are no read marks in current state, because reads always mark | |
13368 | * their parent and current state never has children yet. Only | |
13369 | * explored_states can get read marks.) | |
13370 | */ | |
eea1c227 AS |
13371 | for (j = 0; j <= cur->curframe; j++) { |
13372 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
13373 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
13374 | for (i = 0; i < BPF_REG_FP; i++) | |
13375 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
13376 | } | |
f4d7e40a AS |
13377 | |
13378 | /* all stack frames are accessible from callee, clear them all */ | |
13379 | for (j = 0; j <= cur->curframe; j++) { | |
13380 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 13381 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 13382 | |
679c782d | 13383 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 13384 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
13385 | frame->stack[i].spilled_ptr.parent = |
13386 | &newframe->stack[i].spilled_ptr; | |
13387 | } | |
f4d7e40a | 13388 | } |
f1bca824 AS |
13389 | return 0; |
13390 | } | |
13391 | ||
c64b7983 JS |
13392 | /* Return true if it's OK to have the same insn return a different type. */ |
13393 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
13394 | { | |
c25b2ae1 | 13395 | switch (base_type(type)) { |
c64b7983 JS |
13396 | case PTR_TO_CTX: |
13397 | case PTR_TO_SOCKET: | |
46f8bc92 | 13398 | case PTR_TO_SOCK_COMMON: |
655a51e5 | 13399 | case PTR_TO_TCP_SOCK: |
fada7fdc | 13400 | case PTR_TO_XDP_SOCK: |
2a02759e | 13401 | case PTR_TO_BTF_ID: |
c64b7983 JS |
13402 | return false; |
13403 | default: | |
13404 | return true; | |
13405 | } | |
13406 | } | |
13407 | ||
13408 | /* If an instruction was previously used with particular pointer types, then we | |
13409 | * need to be careful to avoid cases such as the below, where it may be ok | |
13410 | * for one branch accessing the pointer, but not ok for the other branch: | |
13411 | * | |
13412 | * R1 = sock_ptr | |
13413 | * goto X; | |
13414 | * ... | |
13415 | * R1 = some_other_valid_ptr; | |
13416 | * goto X; | |
13417 | * ... | |
13418 | * R2 = *(u32 *)(R1 + 0); | |
13419 | */ | |
13420 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
13421 | { | |
13422 | return src != prev && (!reg_type_mismatch_ok(src) || | |
13423 | !reg_type_mismatch_ok(prev)); | |
13424 | } | |
13425 | ||
58e2af8b | 13426 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 13427 | { |
6f8a57cc | 13428 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 13429 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 13430 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 13431 | struct bpf_reg_state *regs; |
06ee7115 | 13432 | int insn_cnt = env->prog->len; |
17a52670 | 13433 | bool do_print_state = false; |
b5dc0163 | 13434 | int prev_insn_idx = -1; |
17a52670 | 13435 | |
17a52670 AS |
13436 | for (;;) { |
13437 | struct bpf_insn *insn; | |
13438 | u8 class; | |
13439 | int err; | |
13440 | ||
b5dc0163 | 13441 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 13442 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 13443 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 13444 | env->insn_idx, insn_cnt); |
17a52670 AS |
13445 | return -EFAULT; |
13446 | } | |
13447 | ||
c08435ec | 13448 | insn = &insns[env->insn_idx]; |
17a52670 AS |
13449 | class = BPF_CLASS(insn->code); |
13450 | ||
06ee7115 | 13451 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
13452 | verbose(env, |
13453 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 13454 | env->insn_processed); |
17a52670 AS |
13455 | return -E2BIG; |
13456 | } | |
13457 | ||
c08435ec | 13458 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
13459 | if (err < 0) |
13460 | return err; | |
13461 | if (err == 1) { | |
13462 | /* found equivalent state, can prune the search */ | |
06ee7115 | 13463 | if (env->log.level & BPF_LOG_LEVEL) { |
f1bca824 | 13464 | if (do_print_state) |
979d63d5 DB |
13465 | verbose(env, "\nfrom %d to %d%s: safe\n", |
13466 | env->prev_insn_idx, env->insn_idx, | |
13467 | env->cur_state->speculative ? | |
13468 | " (speculative execution)" : ""); | |
f1bca824 | 13469 | else |
c08435ec | 13470 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
13471 | } |
13472 | goto process_bpf_exit; | |
13473 | } | |
13474 | ||
c3494801 AS |
13475 | if (signal_pending(current)) |
13476 | return -EAGAIN; | |
13477 | ||
3c2ce60b DB |
13478 | if (need_resched()) |
13479 | cond_resched(); | |
13480 | ||
2e576648 CL |
13481 | if (env->log.level & BPF_LOG_LEVEL2 && do_print_state) { |
13482 | verbose(env, "\nfrom %d to %d%s:", | |
13483 | env->prev_insn_idx, env->insn_idx, | |
13484 | env->cur_state->speculative ? | |
13485 | " (speculative execution)" : ""); | |
13486 | print_verifier_state(env, state->frame[state->curframe], true); | |
17a52670 AS |
13487 | do_print_state = false; |
13488 | } | |
13489 | ||
06ee7115 | 13490 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 | 13491 | const struct bpf_insn_cbs cbs = { |
e6ac2450 | 13492 | .cb_call = disasm_kfunc_name, |
7105e828 | 13493 | .cb_print = verbose, |
abe08840 | 13494 | .private_data = env, |
7105e828 DB |
13495 | }; |
13496 | ||
2e576648 CL |
13497 | if (verifier_state_scratched(env)) |
13498 | print_insn_state(env, state->frame[state->curframe]); | |
13499 | ||
c08435ec | 13500 | verbose_linfo(env, env->insn_idx, "; "); |
2e576648 | 13501 | env->prev_log_len = env->log.len_used; |
c08435ec | 13502 | verbose(env, "%d: ", env->insn_idx); |
abe08840 | 13503 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
2e576648 CL |
13504 | env->prev_insn_print_len = env->log.len_used - env->prev_log_len; |
13505 | env->prev_log_len = env->log.len_used; | |
17a52670 AS |
13506 | } |
13507 | ||
cae1927c | 13508 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
13509 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
13510 | env->prev_insn_idx); | |
cae1927c JK |
13511 | if (err) |
13512 | return err; | |
13513 | } | |
13a27dfc | 13514 | |
638f5b90 | 13515 | regs = cur_regs(env); |
fe9a5ca7 | 13516 | sanitize_mark_insn_seen(env); |
b5dc0163 | 13517 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 13518 | |
17a52670 | 13519 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 13520 | err = check_alu_op(env, insn); |
17a52670 AS |
13521 | if (err) |
13522 | return err; | |
13523 | ||
13524 | } else if (class == BPF_LDX) { | |
3df126f3 | 13525 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
13526 | |
13527 | /* check for reserved fields is already done */ | |
13528 | ||
17a52670 | 13529 | /* check src operand */ |
dc503a8a | 13530 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
13531 | if (err) |
13532 | return err; | |
13533 | ||
dc503a8a | 13534 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
13535 | if (err) |
13536 | return err; | |
13537 | ||
725f9dcd AS |
13538 | src_reg_type = regs[insn->src_reg].type; |
13539 | ||
17a52670 AS |
13540 | /* check that memory (src_reg + off) is readable, |
13541 | * the state of dst_reg will be updated by this func | |
13542 | */ | |
c08435ec DB |
13543 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
13544 | insn->off, BPF_SIZE(insn->code), | |
13545 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
13546 | if (err) |
13547 | return err; | |
13548 | ||
c08435ec | 13549 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
13550 | |
13551 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
13552 | /* saw a valid insn |
13553 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 13554 | * save type to validate intersecting paths |
9bac3d6d | 13555 | */ |
3df126f3 | 13556 | *prev_src_type = src_reg_type; |
9bac3d6d | 13557 | |
c64b7983 | 13558 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
13559 | /* ABuser program is trying to use the same insn |
13560 | * dst_reg = *(u32*) (src_reg + off) | |
13561 | * with different pointer types: | |
13562 | * src_reg == ctx in one branch and | |
13563 | * src_reg == stack|map in some other branch. | |
13564 | * Reject it. | |
13565 | */ | |
61bd5218 | 13566 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
13567 | return -EINVAL; |
13568 | } | |
13569 | ||
17a52670 | 13570 | } else if (class == BPF_STX) { |
3df126f3 | 13571 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 13572 | |
91c960b0 BJ |
13573 | if (BPF_MODE(insn->code) == BPF_ATOMIC) { |
13574 | err = check_atomic(env, env->insn_idx, insn); | |
17a52670 AS |
13575 | if (err) |
13576 | return err; | |
c08435ec | 13577 | env->insn_idx++; |
17a52670 AS |
13578 | continue; |
13579 | } | |
13580 | ||
5ca419f2 BJ |
13581 | if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) { |
13582 | verbose(env, "BPF_STX uses reserved fields\n"); | |
13583 | return -EINVAL; | |
13584 | } | |
13585 | ||
17a52670 | 13586 | /* check src1 operand */ |
dc503a8a | 13587 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
13588 | if (err) |
13589 | return err; | |
13590 | /* check src2 operand */ | |
dc503a8a | 13591 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
13592 | if (err) |
13593 | return err; | |
13594 | ||
d691f9e8 AS |
13595 | dst_reg_type = regs[insn->dst_reg].type; |
13596 | ||
17a52670 | 13597 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
13598 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
13599 | insn->off, BPF_SIZE(insn->code), | |
13600 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
13601 | if (err) |
13602 | return err; | |
13603 | ||
c08435ec | 13604 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
13605 | |
13606 | if (*prev_dst_type == NOT_INIT) { | |
13607 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 13608 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 13609 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
13610 | return -EINVAL; |
13611 | } | |
13612 | ||
17a52670 AS |
13613 | } else if (class == BPF_ST) { |
13614 | if (BPF_MODE(insn->code) != BPF_MEM || | |
13615 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 13616 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
13617 | return -EINVAL; |
13618 | } | |
13619 | /* check src operand */ | |
dc503a8a | 13620 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
13621 | if (err) |
13622 | return err; | |
13623 | ||
f37a8cb8 | 13624 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 13625 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f | 13626 | insn->dst_reg, |
c25b2ae1 | 13627 | reg_type_str(env, reg_state(env, insn->dst_reg)->type)); |
f37a8cb8 DB |
13628 | return -EACCES; |
13629 | } | |
13630 | ||
17a52670 | 13631 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
13632 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
13633 | insn->off, BPF_SIZE(insn->code), | |
13634 | BPF_WRITE, -1, false); | |
17a52670 AS |
13635 | if (err) |
13636 | return err; | |
13637 | ||
092ed096 | 13638 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
13639 | u8 opcode = BPF_OP(insn->code); |
13640 | ||
2589726d | 13641 | env->jmps_processed++; |
17a52670 AS |
13642 | if (opcode == BPF_CALL) { |
13643 | if (BPF_SRC(insn->code) != BPF_K || | |
2357672c KKD |
13644 | (insn->src_reg != BPF_PSEUDO_KFUNC_CALL |
13645 | && insn->off != 0) || | |
f4d7e40a | 13646 | (insn->src_reg != BPF_REG_0 && |
e6ac2450 MKL |
13647 | insn->src_reg != BPF_PSEUDO_CALL && |
13648 | insn->src_reg != BPF_PSEUDO_KFUNC_CALL) || | |
092ed096 JW |
13649 | insn->dst_reg != BPF_REG_0 || |
13650 | class == BPF_JMP32) { | |
61bd5218 | 13651 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
13652 | return -EINVAL; |
13653 | } | |
13654 | ||
8cab76ec KKD |
13655 | if (env->cur_state->active_lock.ptr) { |
13656 | if ((insn->src_reg == BPF_REG_0 && insn->imm != BPF_FUNC_spin_unlock) || | |
13657 | (insn->src_reg == BPF_PSEUDO_CALL) || | |
13658 | (insn->src_reg == BPF_PSEUDO_KFUNC_CALL && | |
13659 | (insn->off != 0 || !is_bpf_list_api_kfunc(insn->imm)))) { | |
13660 | verbose(env, "function calls are not allowed while holding a lock\n"); | |
13661 | return -EINVAL; | |
13662 | } | |
d83525ca | 13663 | } |
f4d7e40a | 13664 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 13665 | err = check_func_call(env, insn, &env->insn_idx); |
e6ac2450 | 13666 | else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) |
5c073f26 | 13667 | err = check_kfunc_call(env, insn, &env->insn_idx); |
f4d7e40a | 13668 | else |
69c087ba | 13669 | err = check_helper_call(env, insn, &env->insn_idx); |
17a52670 AS |
13670 | if (err) |
13671 | return err; | |
17a52670 AS |
13672 | } else if (opcode == BPF_JA) { |
13673 | if (BPF_SRC(insn->code) != BPF_K || | |
13674 | insn->imm != 0 || | |
13675 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
13676 | insn->dst_reg != BPF_REG_0 || |
13677 | class == BPF_JMP32) { | |
61bd5218 | 13678 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
13679 | return -EINVAL; |
13680 | } | |
13681 | ||
c08435ec | 13682 | env->insn_idx += insn->off + 1; |
17a52670 AS |
13683 | continue; |
13684 | ||
13685 | } else if (opcode == BPF_EXIT) { | |
13686 | if (BPF_SRC(insn->code) != BPF_K || | |
13687 | insn->imm != 0 || | |
13688 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
13689 | insn->dst_reg != BPF_REG_0 || |
13690 | class == BPF_JMP32) { | |
61bd5218 | 13691 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
13692 | return -EINVAL; |
13693 | } | |
13694 | ||
d0d78c1d | 13695 | if (env->cur_state->active_lock.ptr) { |
d83525ca AS |
13696 | verbose(env, "bpf_spin_unlock is missing\n"); |
13697 | return -EINVAL; | |
13698 | } | |
13699 | ||
9d9d00ac KKD |
13700 | /* We must do check_reference_leak here before |
13701 | * prepare_func_exit to handle the case when | |
13702 | * state->curframe > 0, it may be a callback | |
13703 | * function, for which reference_state must | |
13704 | * match caller reference state when it exits. | |
13705 | */ | |
13706 | err = check_reference_leak(env); | |
13707 | if (err) | |
13708 | return err; | |
13709 | ||
f4d7e40a AS |
13710 | if (state->curframe) { |
13711 | /* exit from nested function */ | |
c08435ec | 13712 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
13713 | if (err) |
13714 | return err; | |
13715 | do_print_state = true; | |
13716 | continue; | |
13717 | } | |
13718 | ||
390ee7e2 AS |
13719 | err = check_return_code(env); |
13720 | if (err) | |
13721 | return err; | |
f1bca824 | 13722 | process_bpf_exit: |
0f55f9ed | 13723 | mark_verifier_state_scratched(env); |
2589726d | 13724 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 13725 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 13726 | &env->insn_idx, pop_log); |
638f5b90 AS |
13727 | if (err < 0) { |
13728 | if (err != -ENOENT) | |
13729 | return err; | |
17a52670 AS |
13730 | break; |
13731 | } else { | |
13732 | do_print_state = true; | |
13733 | continue; | |
13734 | } | |
13735 | } else { | |
c08435ec | 13736 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
13737 | if (err) |
13738 | return err; | |
13739 | } | |
13740 | } else if (class == BPF_LD) { | |
13741 | u8 mode = BPF_MODE(insn->code); | |
13742 | ||
13743 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
13744 | err = check_ld_abs(env, insn); |
13745 | if (err) | |
13746 | return err; | |
13747 | ||
17a52670 AS |
13748 | } else if (mode == BPF_IMM) { |
13749 | err = check_ld_imm(env, insn); | |
13750 | if (err) | |
13751 | return err; | |
13752 | ||
c08435ec | 13753 | env->insn_idx++; |
fe9a5ca7 | 13754 | sanitize_mark_insn_seen(env); |
17a52670 | 13755 | } else { |
61bd5218 | 13756 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
13757 | return -EINVAL; |
13758 | } | |
13759 | } else { | |
61bd5218 | 13760 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
13761 | return -EINVAL; |
13762 | } | |
13763 | ||
c08435ec | 13764 | env->insn_idx++; |
17a52670 AS |
13765 | } |
13766 | ||
13767 | return 0; | |
13768 | } | |
13769 | ||
541c3bad AN |
13770 | static int find_btf_percpu_datasec(struct btf *btf) |
13771 | { | |
13772 | const struct btf_type *t; | |
13773 | const char *tname; | |
13774 | int i, n; | |
13775 | ||
13776 | /* | |
13777 | * Both vmlinux and module each have their own ".data..percpu" | |
13778 | * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF | |
13779 | * types to look at only module's own BTF types. | |
13780 | */ | |
13781 | n = btf_nr_types(btf); | |
13782 | if (btf_is_module(btf)) | |
13783 | i = btf_nr_types(btf_vmlinux); | |
13784 | else | |
13785 | i = 1; | |
13786 | ||
13787 | for(; i < n; i++) { | |
13788 | t = btf_type_by_id(btf, i); | |
13789 | if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC) | |
13790 | continue; | |
13791 | ||
13792 | tname = btf_name_by_offset(btf, t->name_off); | |
13793 | if (!strcmp(tname, ".data..percpu")) | |
13794 | return i; | |
13795 | } | |
13796 | ||
13797 | return -ENOENT; | |
13798 | } | |
13799 | ||
4976b718 HL |
13800 | /* replace pseudo btf_id with kernel symbol address */ |
13801 | static int check_pseudo_btf_id(struct bpf_verifier_env *env, | |
13802 | struct bpf_insn *insn, | |
13803 | struct bpf_insn_aux_data *aux) | |
13804 | { | |
eaa6bcb7 HL |
13805 | const struct btf_var_secinfo *vsi; |
13806 | const struct btf_type *datasec; | |
541c3bad | 13807 | struct btf_mod_pair *btf_mod; |
4976b718 HL |
13808 | const struct btf_type *t; |
13809 | const char *sym_name; | |
eaa6bcb7 | 13810 | bool percpu = false; |
f16e6313 | 13811 | u32 type, id = insn->imm; |
541c3bad | 13812 | struct btf *btf; |
f16e6313 | 13813 | s32 datasec_id; |
4976b718 | 13814 | u64 addr; |
541c3bad | 13815 | int i, btf_fd, err; |
4976b718 | 13816 | |
541c3bad AN |
13817 | btf_fd = insn[1].imm; |
13818 | if (btf_fd) { | |
13819 | btf = btf_get_by_fd(btf_fd); | |
13820 | if (IS_ERR(btf)) { | |
13821 | verbose(env, "invalid module BTF object FD specified.\n"); | |
13822 | return -EINVAL; | |
13823 | } | |
13824 | } else { | |
13825 | if (!btf_vmlinux) { | |
13826 | verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n"); | |
13827 | return -EINVAL; | |
13828 | } | |
13829 | btf = btf_vmlinux; | |
13830 | btf_get(btf); | |
4976b718 HL |
13831 | } |
13832 | ||
541c3bad | 13833 | t = btf_type_by_id(btf, id); |
4976b718 HL |
13834 | if (!t) { |
13835 | verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id); | |
541c3bad AN |
13836 | err = -ENOENT; |
13837 | goto err_put; | |
4976b718 HL |
13838 | } |
13839 | ||
13840 | if (!btf_type_is_var(t)) { | |
541c3bad AN |
13841 | verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id); |
13842 | err = -EINVAL; | |
13843 | goto err_put; | |
4976b718 HL |
13844 | } |
13845 | ||
541c3bad | 13846 | sym_name = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
13847 | addr = kallsyms_lookup_name(sym_name); |
13848 | if (!addr) { | |
13849 | verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n", | |
13850 | sym_name); | |
541c3bad AN |
13851 | err = -ENOENT; |
13852 | goto err_put; | |
4976b718 HL |
13853 | } |
13854 | ||
541c3bad | 13855 | datasec_id = find_btf_percpu_datasec(btf); |
eaa6bcb7 | 13856 | if (datasec_id > 0) { |
541c3bad | 13857 | datasec = btf_type_by_id(btf, datasec_id); |
eaa6bcb7 HL |
13858 | for_each_vsi(i, datasec, vsi) { |
13859 | if (vsi->type == id) { | |
13860 | percpu = true; | |
13861 | break; | |
13862 | } | |
13863 | } | |
13864 | } | |
13865 | ||
4976b718 HL |
13866 | insn[0].imm = (u32)addr; |
13867 | insn[1].imm = addr >> 32; | |
13868 | ||
13869 | type = t->type; | |
541c3bad | 13870 | t = btf_type_skip_modifiers(btf, type, NULL); |
eaa6bcb7 | 13871 | if (percpu) { |
5844101a | 13872 | aux->btf_var.reg_type = PTR_TO_BTF_ID | MEM_PERCPU; |
541c3bad | 13873 | aux->btf_var.btf = btf; |
eaa6bcb7 HL |
13874 | aux->btf_var.btf_id = type; |
13875 | } else if (!btf_type_is_struct(t)) { | |
4976b718 HL |
13876 | const struct btf_type *ret; |
13877 | const char *tname; | |
13878 | u32 tsize; | |
13879 | ||
13880 | /* resolve the type size of ksym. */ | |
541c3bad | 13881 | ret = btf_resolve_size(btf, t, &tsize); |
4976b718 | 13882 | if (IS_ERR(ret)) { |
541c3bad | 13883 | tname = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
13884 | verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n", |
13885 | tname, PTR_ERR(ret)); | |
541c3bad AN |
13886 | err = -EINVAL; |
13887 | goto err_put; | |
4976b718 | 13888 | } |
34d3a78c | 13889 | aux->btf_var.reg_type = PTR_TO_MEM | MEM_RDONLY; |
4976b718 HL |
13890 | aux->btf_var.mem_size = tsize; |
13891 | } else { | |
13892 | aux->btf_var.reg_type = PTR_TO_BTF_ID; | |
541c3bad | 13893 | aux->btf_var.btf = btf; |
4976b718 HL |
13894 | aux->btf_var.btf_id = type; |
13895 | } | |
541c3bad AN |
13896 | |
13897 | /* check whether we recorded this BTF (and maybe module) already */ | |
13898 | for (i = 0; i < env->used_btf_cnt; i++) { | |
13899 | if (env->used_btfs[i].btf == btf) { | |
13900 | btf_put(btf); | |
13901 | return 0; | |
13902 | } | |
13903 | } | |
13904 | ||
13905 | if (env->used_btf_cnt >= MAX_USED_BTFS) { | |
13906 | err = -E2BIG; | |
13907 | goto err_put; | |
13908 | } | |
13909 | ||
13910 | btf_mod = &env->used_btfs[env->used_btf_cnt]; | |
13911 | btf_mod->btf = btf; | |
13912 | btf_mod->module = NULL; | |
13913 | ||
13914 | /* if we reference variables from kernel module, bump its refcount */ | |
13915 | if (btf_is_module(btf)) { | |
13916 | btf_mod->module = btf_try_get_module(btf); | |
13917 | if (!btf_mod->module) { | |
13918 | err = -ENXIO; | |
13919 | goto err_put; | |
13920 | } | |
13921 | } | |
13922 | ||
13923 | env->used_btf_cnt++; | |
13924 | ||
4976b718 | 13925 | return 0; |
541c3bad AN |
13926 | err_put: |
13927 | btf_put(btf); | |
13928 | return err; | |
4976b718 HL |
13929 | } |
13930 | ||
d83525ca AS |
13931 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
13932 | { | |
13933 | switch (type) { | |
13934 | case BPF_PROG_TYPE_KPROBE: | |
13935 | case BPF_PROG_TYPE_TRACEPOINT: | |
13936 | case BPF_PROG_TYPE_PERF_EVENT: | |
13937 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
5002615a | 13938 | case BPF_PROG_TYPE_RAW_TRACEPOINT_WRITABLE: |
d83525ca AS |
13939 | return true; |
13940 | default: | |
13941 | return false; | |
13942 | } | |
13943 | } | |
13944 | ||
61bd5218 JK |
13945 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
13946 | struct bpf_map *map, | |
fdc15d38 AS |
13947 | struct bpf_prog *prog) |
13948 | ||
13949 | { | |
7e40781c | 13950 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
a3884572 | 13951 | |
f0c5941f KKD |
13952 | if (btf_record_has_field(map->record, BPF_LIST_HEAD)) { |
13953 | if (is_tracing_prog_type(prog_type)) { | |
13954 | verbose(env, "tracing progs cannot use bpf_list_head yet\n"); | |
13955 | return -EINVAL; | |
13956 | } | |
13957 | } | |
13958 | ||
db559117 | 13959 | if (btf_record_has_field(map->record, BPF_SPIN_LOCK)) { |
9e7a4d98 KS |
13960 | if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) { |
13961 | verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n"); | |
13962 | return -EINVAL; | |
13963 | } | |
13964 | ||
13965 | if (is_tracing_prog_type(prog_type)) { | |
13966 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
13967 | return -EINVAL; | |
13968 | } | |
13969 | ||
13970 | if (prog->aux->sleepable) { | |
13971 | verbose(env, "sleepable progs cannot use bpf_spin_lock yet\n"); | |
13972 | return -EINVAL; | |
13973 | } | |
d83525ca AS |
13974 | } |
13975 | ||
db559117 | 13976 | if (btf_record_has_field(map->record, BPF_TIMER)) { |
5e0bc308 DB |
13977 | if (is_tracing_prog_type(prog_type)) { |
13978 | verbose(env, "tracing progs cannot use bpf_timer yet\n"); | |
13979 | return -EINVAL; | |
13980 | } | |
13981 | } | |
13982 | ||
a3884572 | 13983 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && |
09728266 | 13984 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
13985 | verbose(env, "offload device mismatch between prog and map\n"); |
13986 | return -EINVAL; | |
13987 | } | |
13988 | ||
85d33df3 MKL |
13989 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
13990 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
13991 | return -EINVAL; | |
13992 | } | |
13993 | ||
1e6c62a8 AS |
13994 | if (prog->aux->sleepable) |
13995 | switch (map->map_type) { | |
13996 | case BPF_MAP_TYPE_HASH: | |
13997 | case BPF_MAP_TYPE_LRU_HASH: | |
13998 | case BPF_MAP_TYPE_ARRAY: | |
638e4b82 AS |
13999 | case BPF_MAP_TYPE_PERCPU_HASH: |
14000 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
14001 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
14002 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
14003 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
ba90c2cc | 14004 | case BPF_MAP_TYPE_RINGBUF: |
583c1f42 | 14005 | case BPF_MAP_TYPE_USER_RINGBUF: |
0fe4b381 KS |
14006 | case BPF_MAP_TYPE_INODE_STORAGE: |
14007 | case BPF_MAP_TYPE_SK_STORAGE: | |
14008 | case BPF_MAP_TYPE_TASK_STORAGE: | |
ba90c2cc | 14009 | break; |
1e6c62a8 AS |
14010 | default: |
14011 | verbose(env, | |
ba90c2cc | 14012 | "Sleepable programs can only use array, hash, and ringbuf maps\n"); |
1e6c62a8 AS |
14013 | return -EINVAL; |
14014 | } | |
14015 | ||
fdc15d38 AS |
14016 | return 0; |
14017 | } | |
14018 | ||
b741f163 RG |
14019 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
14020 | { | |
14021 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
14022 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
14023 | } | |
14024 | ||
4976b718 HL |
14025 | /* find and rewrite pseudo imm in ld_imm64 instructions: |
14026 | * | |
14027 | * 1. if it accesses map FD, replace it with actual map pointer. | |
14028 | * 2. if it accesses btf_id of a VAR, replace it with pointer to the var. | |
14029 | * | |
14030 | * NOTE: btf_vmlinux is required for converting pseudo btf_id. | |
0246e64d | 14031 | */ |
4976b718 | 14032 | static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env) |
0246e64d AS |
14033 | { |
14034 | struct bpf_insn *insn = env->prog->insnsi; | |
14035 | int insn_cnt = env->prog->len; | |
fdc15d38 | 14036 | int i, j, err; |
0246e64d | 14037 | |
f1f7714e | 14038 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
14039 | if (err) |
14040 | return err; | |
14041 | ||
0246e64d | 14042 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 14043 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 14044 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 14045 | verbose(env, "BPF_LDX uses reserved fields\n"); |
d691f9e8 AS |
14046 | return -EINVAL; |
14047 | } | |
14048 | ||
0246e64d | 14049 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 14050 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
14051 | struct bpf_map *map; |
14052 | struct fd f; | |
d8eca5bb | 14053 | u64 addr; |
387544bf | 14054 | u32 fd; |
0246e64d AS |
14055 | |
14056 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
14057 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
14058 | insn[1].off != 0) { | |
61bd5218 | 14059 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
14060 | return -EINVAL; |
14061 | } | |
14062 | ||
d8eca5bb | 14063 | if (insn[0].src_reg == 0) |
0246e64d AS |
14064 | /* valid generic load 64-bit imm */ |
14065 | goto next_insn; | |
14066 | ||
4976b718 HL |
14067 | if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) { |
14068 | aux = &env->insn_aux_data[i]; | |
14069 | err = check_pseudo_btf_id(env, insn, aux); | |
14070 | if (err) | |
14071 | return err; | |
14072 | goto next_insn; | |
14073 | } | |
14074 | ||
69c087ba YS |
14075 | if (insn[0].src_reg == BPF_PSEUDO_FUNC) { |
14076 | aux = &env->insn_aux_data[i]; | |
14077 | aux->ptr_type = PTR_TO_FUNC; | |
14078 | goto next_insn; | |
14079 | } | |
14080 | ||
d8eca5bb DB |
14081 | /* In final convert_pseudo_ld_imm64() step, this is |
14082 | * converted into regular 64-bit imm load insn. | |
14083 | */ | |
387544bf AS |
14084 | switch (insn[0].src_reg) { |
14085 | case BPF_PSEUDO_MAP_VALUE: | |
14086 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
14087 | break; | |
14088 | case BPF_PSEUDO_MAP_FD: | |
14089 | case BPF_PSEUDO_MAP_IDX: | |
14090 | if (insn[1].imm == 0) | |
14091 | break; | |
14092 | fallthrough; | |
14093 | default: | |
14094 | verbose(env, "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
14095 | return -EINVAL; |
14096 | } | |
14097 | ||
387544bf AS |
14098 | switch (insn[0].src_reg) { |
14099 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
14100 | case BPF_PSEUDO_MAP_IDX: | |
14101 | if (bpfptr_is_null(env->fd_array)) { | |
14102 | verbose(env, "fd_idx without fd_array is invalid\n"); | |
14103 | return -EPROTO; | |
14104 | } | |
14105 | if (copy_from_bpfptr_offset(&fd, env->fd_array, | |
14106 | insn[0].imm * sizeof(fd), | |
14107 | sizeof(fd))) | |
14108 | return -EFAULT; | |
14109 | break; | |
14110 | default: | |
14111 | fd = insn[0].imm; | |
14112 | break; | |
14113 | } | |
14114 | ||
14115 | f = fdget(fd); | |
c2101297 | 14116 | map = __bpf_map_get(f); |
0246e64d | 14117 | if (IS_ERR(map)) { |
61bd5218 | 14118 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 14119 | insn[0].imm); |
0246e64d AS |
14120 | return PTR_ERR(map); |
14121 | } | |
14122 | ||
61bd5218 | 14123 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
14124 | if (err) { |
14125 | fdput(f); | |
14126 | return err; | |
14127 | } | |
14128 | ||
d8eca5bb | 14129 | aux = &env->insn_aux_data[i]; |
387544bf AS |
14130 | if (insn[0].src_reg == BPF_PSEUDO_MAP_FD || |
14131 | insn[0].src_reg == BPF_PSEUDO_MAP_IDX) { | |
d8eca5bb DB |
14132 | addr = (unsigned long)map; |
14133 | } else { | |
14134 | u32 off = insn[1].imm; | |
14135 | ||
14136 | if (off >= BPF_MAX_VAR_OFF) { | |
14137 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
14138 | fdput(f); | |
14139 | return -EINVAL; | |
14140 | } | |
14141 | ||
14142 | if (!map->ops->map_direct_value_addr) { | |
14143 | verbose(env, "no direct value access support for this map type\n"); | |
14144 | fdput(f); | |
14145 | return -EINVAL; | |
14146 | } | |
14147 | ||
14148 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
14149 | if (err) { | |
14150 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
14151 | map->value_size, off); | |
14152 | fdput(f); | |
14153 | return err; | |
14154 | } | |
14155 | ||
14156 | aux->map_off = off; | |
14157 | addr += off; | |
14158 | } | |
14159 | ||
14160 | insn[0].imm = (u32)addr; | |
14161 | insn[1].imm = addr >> 32; | |
0246e64d AS |
14162 | |
14163 | /* check whether we recorded this map already */ | |
d8eca5bb | 14164 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 14165 | if (env->used_maps[j] == map) { |
d8eca5bb | 14166 | aux->map_index = j; |
0246e64d AS |
14167 | fdput(f); |
14168 | goto next_insn; | |
14169 | } | |
d8eca5bb | 14170 | } |
0246e64d AS |
14171 | |
14172 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
14173 | fdput(f); | |
14174 | return -E2BIG; | |
14175 | } | |
14176 | ||
0246e64d AS |
14177 | /* hold the map. If the program is rejected by verifier, |
14178 | * the map will be released by release_maps() or it | |
14179 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 14180 | * and all maps are released in free_used_maps() |
0246e64d | 14181 | */ |
1e0bd5a0 | 14182 | bpf_map_inc(map); |
d8eca5bb DB |
14183 | |
14184 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
14185 | env->used_maps[env->used_map_cnt++] = map; |
14186 | ||
b741f163 | 14187 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 14188 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 14189 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
14190 | fdput(f); |
14191 | return -EBUSY; | |
14192 | } | |
14193 | ||
0246e64d AS |
14194 | fdput(f); |
14195 | next_insn: | |
14196 | insn++; | |
14197 | i++; | |
5e581dad DB |
14198 | continue; |
14199 | } | |
14200 | ||
14201 | /* Basic sanity check before we invest more work here. */ | |
14202 | if (!bpf_opcode_in_insntable(insn->code)) { | |
14203 | verbose(env, "unknown opcode %02x\n", insn->code); | |
14204 | return -EINVAL; | |
0246e64d AS |
14205 | } |
14206 | } | |
14207 | ||
14208 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
14209 | * 'struct bpf_map *' into a register instead of user map_fd. | |
14210 | * These pointers will be used later by verifier to validate map access. | |
14211 | */ | |
14212 | return 0; | |
14213 | } | |
14214 | ||
14215 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 14216 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 14217 | { |
a2ea0746 DB |
14218 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
14219 | env->used_map_cnt); | |
0246e64d AS |
14220 | } |
14221 | ||
541c3bad AN |
14222 | /* drop refcnt of maps used by the rejected program */ |
14223 | static void release_btfs(struct bpf_verifier_env *env) | |
14224 | { | |
14225 | __bpf_free_used_btfs(env->prog->aux, env->used_btfs, | |
14226 | env->used_btf_cnt); | |
14227 | } | |
14228 | ||
0246e64d | 14229 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ |
58e2af8b | 14230 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
14231 | { |
14232 | struct bpf_insn *insn = env->prog->insnsi; | |
14233 | int insn_cnt = env->prog->len; | |
14234 | int i; | |
14235 | ||
69c087ba YS |
14236 | for (i = 0; i < insn_cnt; i++, insn++) { |
14237 | if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) | |
14238 | continue; | |
14239 | if (insn->src_reg == BPF_PSEUDO_FUNC) | |
14240 | continue; | |
14241 | insn->src_reg = 0; | |
14242 | } | |
0246e64d AS |
14243 | } |
14244 | ||
8041902d AS |
14245 | /* single env->prog->insni[off] instruction was replaced with the range |
14246 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
14247 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
14248 | */ | |
75f0fc7b HF |
14249 | static void adjust_insn_aux_data(struct bpf_verifier_env *env, |
14250 | struct bpf_insn_aux_data *new_data, | |
14251 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d | 14252 | { |
75f0fc7b | 14253 | struct bpf_insn_aux_data *old_data = env->insn_aux_data; |
b325fbca | 14254 | struct bpf_insn *insn = new_prog->insnsi; |
d203b0fd | 14255 | u32 old_seen = old_data[off].seen; |
b325fbca | 14256 | u32 prog_len; |
c131187d | 14257 | int i; |
8041902d | 14258 | |
b325fbca JW |
14259 | /* aux info at OFF always needs adjustment, no matter fast path |
14260 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
14261 | * original insn at old prog. | |
14262 | */ | |
14263 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
14264 | ||
8041902d | 14265 | if (cnt == 1) |
75f0fc7b | 14266 | return; |
b325fbca | 14267 | prog_len = new_prog->len; |
75f0fc7b | 14268 | |
8041902d AS |
14269 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); |
14270 | memcpy(new_data + off + cnt - 1, old_data + off, | |
14271 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 14272 | for (i = off; i < off + cnt - 1; i++) { |
d203b0fd DB |
14273 | /* Expand insni[off]'s seen count to the patched range. */ |
14274 | new_data[i].seen = old_seen; | |
b325fbca JW |
14275 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
14276 | } | |
8041902d AS |
14277 | env->insn_aux_data = new_data; |
14278 | vfree(old_data); | |
8041902d AS |
14279 | } |
14280 | ||
cc8b0b92 AS |
14281 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
14282 | { | |
14283 | int i; | |
14284 | ||
14285 | if (len == 1) | |
14286 | return; | |
4cb3d99c JW |
14287 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
14288 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 14289 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 14290 | continue; |
9c8105bd | 14291 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
14292 | } |
14293 | } | |
14294 | ||
7506d211 | 14295 | static void adjust_poke_descs(struct bpf_prog *prog, u32 off, u32 len) |
a748c697 MF |
14296 | { |
14297 | struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab; | |
14298 | int i, sz = prog->aux->size_poke_tab; | |
14299 | struct bpf_jit_poke_descriptor *desc; | |
14300 | ||
14301 | for (i = 0; i < sz; i++) { | |
14302 | desc = &tab[i]; | |
7506d211 JF |
14303 | if (desc->insn_idx <= off) |
14304 | continue; | |
a748c697 MF |
14305 | desc->insn_idx += len - 1; |
14306 | } | |
14307 | } | |
14308 | ||
8041902d AS |
14309 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
14310 | const struct bpf_insn *patch, u32 len) | |
14311 | { | |
14312 | struct bpf_prog *new_prog; | |
75f0fc7b HF |
14313 | struct bpf_insn_aux_data *new_data = NULL; |
14314 | ||
14315 | if (len > 1) { | |
14316 | new_data = vzalloc(array_size(env->prog->len + len - 1, | |
14317 | sizeof(struct bpf_insn_aux_data))); | |
14318 | if (!new_data) | |
14319 | return NULL; | |
14320 | } | |
8041902d AS |
14321 | |
14322 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
14323 | if (IS_ERR(new_prog)) { |
14324 | if (PTR_ERR(new_prog) == -ERANGE) | |
14325 | verbose(env, | |
14326 | "insn %d cannot be patched due to 16-bit range\n", | |
14327 | env->insn_aux_data[off].orig_idx); | |
75f0fc7b | 14328 | vfree(new_data); |
8041902d | 14329 | return NULL; |
4f73379e | 14330 | } |
75f0fc7b | 14331 | adjust_insn_aux_data(env, new_data, new_prog, off, len); |
cc8b0b92 | 14332 | adjust_subprog_starts(env, off, len); |
7506d211 | 14333 | adjust_poke_descs(new_prog, off, len); |
8041902d AS |
14334 | return new_prog; |
14335 | } | |
14336 | ||
52875a04 JK |
14337 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
14338 | u32 off, u32 cnt) | |
14339 | { | |
14340 | int i, j; | |
14341 | ||
14342 | /* find first prog starting at or after off (first to remove) */ | |
14343 | for (i = 0; i < env->subprog_cnt; i++) | |
14344 | if (env->subprog_info[i].start >= off) | |
14345 | break; | |
14346 | /* find first prog starting at or after off + cnt (first to stay) */ | |
14347 | for (j = i; j < env->subprog_cnt; j++) | |
14348 | if (env->subprog_info[j].start >= off + cnt) | |
14349 | break; | |
14350 | /* if j doesn't start exactly at off + cnt, we are just removing | |
14351 | * the front of previous prog | |
14352 | */ | |
14353 | if (env->subprog_info[j].start != off + cnt) | |
14354 | j--; | |
14355 | ||
14356 | if (j > i) { | |
14357 | struct bpf_prog_aux *aux = env->prog->aux; | |
14358 | int move; | |
14359 | ||
14360 | /* move fake 'exit' subprog as well */ | |
14361 | move = env->subprog_cnt + 1 - j; | |
14362 | ||
14363 | memmove(env->subprog_info + i, | |
14364 | env->subprog_info + j, | |
14365 | sizeof(*env->subprog_info) * move); | |
14366 | env->subprog_cnt -= j - i; | |
14367 | ||
14368 | /* remove func_info */ | |
14369 | if (aux->func_info) { | |
14370 | move = aux->func_info_cnt - j; | |
14371 | ||
14372 | memmove(aux->func_info + i, | |
14373 | aux->func_info + j, | |
14374 | sizeof(*aux->func_info) * move); | |
14375 | aux->func_info_cnt -= j - i; | |
14376 | /* func_info->insn_off is set after all code rewrites, | |
14377 | * in adjust_btf_func() - no need to adjust | |
14378 | */ | |
14379 | } | |
14380 | } else { | |
14381 | /* convert i from "first prog to remove" to "first to adjust" */ | |
14382 | if (env->subprog_info[i].start == off) | |
14383 | i++; | |
14384 | } | |
14385 | ||
14386 | /* update fake 'exit' subprog as well */ | |
14387 | for (; i <= env->subprog_cnt; i++) | |
14388 | env->subprog_info[i].start -= cnt; | |
14389 | ||
14390 | return 0; | |
14391 | } | |
14392 | ||
14393 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
14394 | u32 cnt) | |
14395 | { | |
14396 | struct bpf_prog *prog = env->prog; | |
14397 | u32 i, l_off, l_cnt, nr_linfo; | |
14398 | struct bpf_line_info *linfo; | |
14399 | ||
14400 | nr_linfo = prog->aux->nr_linfo; | |
14401 | if (!nr_linfo) | |
14402 | return 0; | |
14403 | ||
14404 | linfo = prog->aux->linfo; | |
14405 | ||
14406 | /* find first line info to remove, count lines to be removed */ | |
14407 | for (i = 0; i < nr_linfo; i++) | |
14408 | if (linfo[i].insn_off >= off) | |
14409 | break; | |
14410 | ||
14411 | l_off = i; | |
14412 | l_cnt = 0; | |
14413 | for (; i < nr_linfo; i++) | |
14414 | if (linfo[i].insn_off < off + cnt) | |
14415 | l_cnt++; | |
14416 | else | |
14417 | break; | |
14418 | ||
14419 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
14420 | * last removed linfo. prog is already modified, so prog->len == off | |
14421 | * means no live instructions after (tail of the program was removed). | |
14422 | */ | |
14423 | if (prog->len != off && l_cnt && | |
14424 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
14425 | l_cnt--; | |
14426 | linfo[--i].insn_off = off + cnt; | |
14427 | } | |
14428 | ||
14429 | /* remove the line info which refer to the removed instructions */ | |
14430 | if (l_cnt) { | |
14431 | memmove(linfo + l_off, linfo + i, | |
14432 | sizeof(*linfo) * (nr_linfo - i)); | |
14433 | ||
14434 | prog->aux->nr_linfo -= l_cnt; | |
14435 | nr_linfo = prog->aux->nr_linfo; | |
14436 | } | |
14437 | ||
14438 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
14439 | for (i = l_off; i < nr_linfo; i++) | |
14440 | linfo[i].insn_off -= cnt; | |
14441 | ||
14442 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
14443 | for (i = 0; i <= env->subprog_cnt; i++) | |
14444 | if (env->subprog_info[i].linfo_idx > l_off) { | |
14445 | /* program may have started in the removed region but | |
14446 | * may not be fully removed | |
14447 | */ | |
14448 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
14449 | env->subprog_info[i].linfo_idx -= l_cnt; | |
14450 | else | |
14451 | env->subprog_info[i].linfo_idx = l_off; | |
14452 | } | |
14453 | ||
14454 | return 0; | |
14455 | } | |
14456 | ||
14457 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
14458 | { | |
14459 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
14460 | unsigned int orig_prog_len = env->prog->len; | |
14461 | int err; | |
14462 | ||
08ca90af JK |
14463 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
14464 | bpf_prog_offload_remove_insns(env, off, cnt); | |
14465 | ||
52875a04 JK |
14466 | err = bpf_remove_insns(env->prog, off, cnt); |
14467 | if (err) | |
14468 | return err; | |
14469 | ||
14470 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
14471 | if (err) | |
14472 | return err; | |
14473 | ||
14474 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
14475 | if (err) | |
14476 | return err; | |
14477 | ||
14478 | memmove(aux_data + off, aux_data + off + cnt, | |
14479 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
14480 | ||
14481 | return 0; | |
14482 | } | |
14483 | ||
2a5418a1 DB |
14484 | /* The verifier does more data flow analysis than llvm and will not |
14485 | * explore branches that are dead at run time. Malicious programs can | |
14486 | * have dead code too. Therefore replace all dead at-run-time code | |
14487 | * with 'ja -1'. | |
14488 | * | |
14489 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
14490 | * program and through another bug we would manage to jump there, then | |
14491 | * we'd execute beyond program memory otherwise. Returning exception | |
14492 | * code also wouldn't work since we can have subprogs where the dead | |
14493 | * code could be located. | |
c131187d AS |
14494 | */ |
14495 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
14496 | { | |
14497 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 14498 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
14499 | struct bpf_insn *insn = env->prog->insnsi; |
14500 | const int insn_cnt = env->prog->len; | |
14501 | int i; | |
14502 | ||
14503 | for (i = 0; i < insn_cnt; i++) { | |
14504 | if (aux_data[i].seen) | |
14505 | continue; | |
2a5418a1 | 14506 | memcpy(insn + i, &trap, sizeof(trap)); |
45c709f8 | 14507 | aux_data[i].zext_dst = false; |
c131187d AS |
14508 | } |
14509 | } | |
14510 | ||
e2ae4ca2 JK |
14511 | static bool insn_is_cond_jump(u8 code) |
14512 | { | |
14513 | u8 op; | |
14514 | ||
092ed096 JW |
14515 | if (BPF_CLASS(code) == BPF_JMP32) |
14516 | return true; | |
14517 | ||
e2ae4ca2 JK |
14518 | if (BPF_CLASS(code) != BPF_JMP) |
14519 | return false; | |
14520 | ||
14521 | op = BPF_OP(code); | |
14522 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
14523 | } | |
14524 | ||
14525 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
14526 | { | |
14527 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
14528 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
14529 | struct bpf_insn *insn = env->prog->insnsi; | |
14530 | const int insn_cnt = env->prog->len; | |
14531 | int i; | |
14532 | ||
14533 | for (i = 0; i < insn_cnt; i++, insn++) { | |
14534 | if (!insn_is_cond_jump(insn->code)) | |
14535 | continue; | |
14536 | ||
14537 | if (!aux_data[i + 1].seen) | |
14538 | ja.off = insn->off; | |
14539 | else if (!aux_data[i + 1 + insn->off].seen) | |
14540 | ja.off = 0; | |
14541 | else | |
14542 | continue; | |
14543 | ||
08ca90af JK |
14544 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
14545 | bpf_prog_offload_replace_insn(env, i, &ja); | |
14546 | ||
e2ae4ca2 JK |
14547 | memcpy(insn, &ja, sizeof(ja)); |
14548 | } | |
14549 | } | |
14550 | ||
52875a04 JK |
14551 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
14552 | { | |
14553 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
14554 | int insn_cnt = env->prog->len; | |
14555 | int i, err; | |
14556 | ||
14557 | for (i = 0; i < insn_cnt; i++) { | |
14558 | int j; | |
14559 | ||
14560 | j = 0; | |
14561 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
14562 | j++; | |
14563 | if (!j) | |
14564 | continue; | |
14565 | ||
14566 | err = verifier_remove_insns(env, i, j); | |
14567 | if (err) | |
14568 | return err; | |
14569 | insn_cnt = env->prog->len; | |
14570 | } | |
14571 | ||
14572 | return 0; | |
14573 | } | |
14574 | ||
a1b14abc JK |
14575 | static int opt_remove_nops(struct bpf_verifier_env *env) |
14576 | { | |
14577 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
14578 | struct bpf_insn *insn = env->prog->insnsi; | |
14579 | int insn_cnt = env->prog->len; | |
14580 | int i, err; | |
14581 | ||
14582 | for (i = 0; i < insn_cnt; i++) { | |
14583 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
14584 | continue; | |
14585 | ||
14586 | err = verifier_remove_insns(env, i, 1); | |
14587 | if (err) | |
14588 | return err; | |
14589 | insn_cnt--; | |
14590 | i--; | |
14591 | } | |
14592 | ||
14593 | return 0; | |
14594 | } | |
14595 | ||
d6c2308c JW |
14596 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
14597 | const union bpf_attr *attr) | |
a4b1d3c1 | 14598 | { |
d6c2308c | 14599 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 14600 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 14601 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 14602 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 14603 | struct bpf_prog *new_prog; |
d6c2308c | 14604 | bool rnd_hi32; |
a4b1d3c1 | 14605 | |
d6c2308c | 14606 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 14607 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
14608 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
14609 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
14610 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
14611 | for (i = 0; i < len; i++) { |
14612 | int adj_idx = i + delta; | |
14613 | struct bpf_insn insn; | |
83a28819 | 14614 | int load_reg; |
a4b1d3c1 | 14615 | |
d6c2308c | 14616 | insn = insns[adj_idx]; |
83a28819 | 14617 | load_reg = insn_def_regno(&insn); |
d6c2308c JW |
14618 | if (!aux[adj_idx].zext_dst) { |
14619 | u8 code, class; | |
14620 | u32 imm_rnd; | |
14621 | ||
14622 | if (!rnd_hi32) | |
14623 | continue; | |
14624 | ||
14625 | code = insn.code; | |
14626 | class = BPF_CLASS(code); | |
83a28819 | 14627 | if (load_reg == -1) |
d6c2308c JW |
14628 | continue; |
14629 | ||
14630 | /* NOTE: arg "reg" (the fourth one) is only used for | |
83a28819 IL |
14631 | * BPF_STX + SRC_OP, so it is safe to pass NULL |
14632 | * here. | |
d6c2308c | 14633 | */ |
83a28819 | 14634 | if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) { |
d6c2308c JW |
14635 | if (class == BPF_LD && |
14636 | BPF_MODE(code) == BPF_IMM) | |
14637 | i++; | |
14638 | continue; | |
14639 | } | |
14640 | ||
14641 | /* ctx load could be transformed into wider load. */ | |
14642 | if (class == BPF_LDX && | |
14643 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
14644 | continue; | |
14645 | ||
a251c17a | 14646 | imm_rnd = get_random_u32(); |
d6c2308c JW |
14647 | rnd_hi32_patch[0] = insn; |
14648 | rnd_hi32_patch[1].imm = imm_rnd; | |
83a28819 | 14649 | rnd_hi32_patch[3].dst_reg = load_reg; |
d6c2308c JW |
14650 | patch = rnd_hi32_patch; |
14651 | patch_len = 4; | |
14652 | goto apply_patch_buffer; | |
14653 | } | |
14654 | ||
39491867 BJ |
14655 | /* Add in an zero-extend instruction if a) the JIT has requested |
14656 | * it or b) it's a CMPXCHG. | |
14657 | * | |
14658 | * The latter is because: BPF_CMPXCHG always loads a value into | |
14659 | * R0, therefore always zero-extends. However some archs' | |
14660 | * equivalent instruction only does this load when the | |
14661 | * comparison is successful. This detail of CMPXCHG is | |
14662 | * orthogonal to the general zero-extension behaviour of the | |
14663 | * CPU, so it's treated independently of bpf_jit_needs_zext. | |
14664 | */ | |
14665 | if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn)) | |
a4b1d3c1 JW |
14666 | continue; |
14667 | ||
83a28819 IL |
14668 | if (WARN_ON(load_reg == -1)) { |
14669 | verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n"); | |
14670 | return -EFAULT; | |
b2e37a71 IL |
14671 | } |
14672 | ||
a4b1d3c1 | 14673 | zext_patch[0] = insn; |
b2e37a71 IL |
14674 | zext_patch[1].dst_reg = load_reg; |
14675 | zext_patch[1].src_reg = load_reg; | |
d6c2308c JW |
14676 | patch = zext_patch; |
14677 | patch_len = 2; | |
14678 | apply_patch_buffer: | |
14679 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
14680 | if (!new_prog) |
14681 | return -ENOMEM; | |
14682 | env->prog = new_prog; | |
14683 | insns = new_prog->insnsi; | |
14684 | aux = env->insn_aux_data; | |
d6c2308c | 14685 | delta += patch_len - 1; |
a4b1d3c1 JW |
14686 | } |
14687 | ||
14688 | return 0; | |
14689 | } | |
14690 | ||
c64b7983 JS |
14691 | /* convert load instructions that access fields of a context type into a |
14692 | * sequence of instructions that access fields of the underlying structure: | |
14693 | * struct __sk_buff -> struct sk_buff | |
14694 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 14695 | */ |
58e2af8b | 14696 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 14697 | { |
00176a34 | 14698 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 14699 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 14700 | const int insn_cnt = env->prog->len; |
36bbef52 | 14701 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 14702 | u32 target_size, size_default, off; |
9bac3d6d | 14703 | struct bpf_prog *new_prog; |
d691f9e8 | 14704 | enum bpf_access_type type; |
f96da094 | 14705 | bool is_narrower_load; |
9bac3d6d | 14706 | |
b09928b9 DB |
14707 | if (ops->gen_prologue || env->seen_direct_write) { |
14708 | if (!ops->gen_prologue) { | |
14709 | verbose(env, "bpf verifier is misconfigured\n"); | |
14710 | return -EINVAL; | |
14711 | } | |
36bbef52 DB |
14712 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
14713 | env->prog); | |
14714 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 14715 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
14716 | return -EINVAL; |
14717 | } else if (cnt) { | |
8041902d | 14718 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
14719 | if (!new_prog) |
14720 | return -ENOMEM; | |
8041902d | 14721 | |
36bbef52 | 14722 | env->prog = new_prog; |
3df126f3 | 14723 | delta += cnt - 1; |
36bbef52 DB |
14724 | } |
14725 | } | |
14726 | ||
c64b7983 | 14727 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
14728 | return 0; |
14729 | ||
3df126f3 | 14730 | insn = env->prog->insnsi + delta; |
36bbef52 | 14731 | |
9bac3d6d | 14732 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 | 14733 | bpf_convert_ctx_access_t convert_ctx_access; |
2039f26f | 14734 | bool ctx_access; |
c64b7983 | 14735 | |
62c7989b DB |
14736 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
14737 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
14738 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
2039f26f | 14739 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) { |
d691f9e8 | 14740 | type = BPF_READ; |
2039f26f DB |
14741 | ctx_access = true; |
14742 | } else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || | |
14743 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
14744 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
14745 | insn->code == (BPF_STX | BPF_MEM | BPF_DW) || | |
14746 | insn->code == (BPF_ST | BPF_MEM | BPF_B) || | |
14747 | insn->code == (BPF_ST | BPF_MEM | BPF_H) || | |
14748 | insn->code == (BPF_ST | BPF_MEM | BPF_W) || | |
14749 | insn->code == (BPF_ST | BPF_MEM | BPF_DW)) { | |
d691f9e8 | 14750 | type = BPF_WRITE; |
2039f26f DB |
14751 | ctx_access = BPF_CLASS(insn->code) == BPF_STX; |
14752 | } else { | |
9bac3d6d | 14753 | continue; |
2039f26f | 14754 | } |
9bac3d6d | 14755 | |
af86ca4e | 14756 | if (type == BPF_WRITE && |
2039f26f | 14757 | env->insn_aux_data[i + delta].sanitize_stack_spill) { |
af86ca4e | 14758 | struct bpf_insn patch[] = { |
af86ca4e | 14759 | *insn, |
2039f26f | 14760 | BPF_ST_NOSPEC(), |
af86ca4e AS |
14761 | }; |
14762 | ||
14763 | cnt = ARRAY_SIZE(patch); | |
14764 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
14765 | if (!new_prog) | |
14766 | return -ENOMEM; | |
14767 | ||
14768 | delta += cnt - 1; | |
14769 | env->prog = new_prog; | |
14770 | insn = new_prog->insnsi + i + delta; | |
14771 | continue; | |
14772 | } | |
14773 | ||
2039f26f DB |
14774 | if (!ctx_access) |
14775 | continue; | |
14776 | ||
6efe152d | 14777 | switch ((int)env->insn_aux_data[i + delta].ptr_type) { |
c64b7983 JS |
14778 | case PTR_TO_CTX: |
14779 | if (!ops->convert_ctx_access) | |
14780 | continue; | |
14781 | convert_ctx_access = ops->convert_ctx_access; | |
14782 | break; | |
14783 | case PTR_TO_SOCKET: | |
46f8bc92 | 14784 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
14785 | convert_ctx_access = bpf_sock_convert_ctx_access; |
14786 | break; | |
655a51e5 MKL |
14787 | case PTR_TO_TCP_SOCK: |
14788 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
14789 | break; | |
fada7fdc JL |
14790 | case PTR_TO_XDP_SOCK: |
14791 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
14792 | break; | |
2a02759e | 14793 | case PTR_TO_BTF_ID: |
6efe152d | 14794 | case PTR_TO_BTF_ID | PTR_UNTRUSTED: |
3f00c523 | 14795 | case PTR_TO_BTF_ID | PTR_TRUSTED: |
282de143 KKD |
14796 | /* PTR_TO_BTF_ID | MEM_ALLOC always has a valid lifetime, unlike |
14797 | * PTR_TO_BTF_ID, and an active ref_obj_id, but the same cannot | |
14798 | * be said once it is marked PTR_UNTRUSTED, hence we must handle | |
14799 | * any faults for loads into such types. BPF_WRITE is disallowed | |
14800 | * for this case. | |
14801 | */ | |
14802 | case PTR_TO_BTF_ID | MEM_ALLOC | PTR_UNTRUSTED: | |
3f00c523 DV |
14803 | case PTR_TO_BTF_ID | PTR_UNTRUSTED | PTR_TRUSTED: |
14804 | case PTR_TO_BTF_ID | PTR_UNTRUSTED | MEM_ALLOC | PTR_TRUSTED: | |
27ae7997 MKL |
14805 | if (type == BPF_READ) { |
14806 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
14807 | BPF_SIZE((insn)->code); | |
14808 | env->prog->aux->num_exentries++; | |
2a02759e | 14809 | } |
2a02759e | 14810 | continue; |
c64b7983 | 14811 | default: |
9bac3d6d | 14812 | continue; |
c64b7983 | 14813 | } |
9bac3d6d | 14814 | |
31fd8581 | 14815 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 14816 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
14817 | |
14818 | /* If the read access is a narrower load of the field, | |
14819 | * convert to a 4/8-byte load, to minimum program type specific | |
14820 | * convert_ctx_access changes. If conversion is successful, | |
14821 | * we will apply proper mask to the result. | |
14822 | */ | |
f96da094 | 14823 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
14824 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
14825 | off = insn->off; | |
31fd8581 | 14826 | if (is_narrower_load) { |
f96da094 DB |
14827 | u8 size_code; |
14828 | ||
14829 | if (type == BPF_WRITE) { | |
61bd5218 | 14830 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
14831 | return -EINVAL; |
14832 | } | |
31fd8581 | 14833 | |
f96da094 | 14834 | size_code = BPF_H; |
31fd8581 YS |
14835 | if (ctx_field_size == 4) |
14836 | size_code = BPF_W; | |
14837 | else if (ctx_field_size == 8) | |
14838 | size_code = BPF_DW; | |
f96da094 | 14839 | |
bc23105c | 14840 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
14841 | insn->code = BPF_LDX | BPF_MEM | size_code; |
14842 | } | |
f96da094 DB |
14843 | |
14844 | target_size = 0; | |
c64b7983 JS |
14845 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
14846 | &target_size); | |
f96da094 DB |
14847 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
14848 | (ctx_field_size && !target_size)) { | |
61bd5218 | 14849 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
14850 | return -EINVAL; |
14851 | } | |
f96da094 DB |
14852 | |
14853 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
14854 | u8 shift = bpf_ctx_narrow_access_offset( |
14855 | off, size, size_default) * 8; | |
d7af7e49 AI |
14856 | if (shift && cnt + 1 >= ARRAY_SIZE(insn_buf)) { |
14857 | verbose(env, "bpf verifier narrow ctx load misconfigured\n"); | |
14858 | return -EINVAL; | |
14859 | } | |
46f53a65 AI |
14860 | if (ctx_field_size <= 4) { |
14861 | if (shift) | |
14862 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
14863 | insn->dst_reg, | |
14864 | shift); | |
31fd8581 | 14865 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 14866 | (1 << size * 8) - 1); |
46f53a65 AI |
14867 | } else { |
14868 | if (shift) | |
14869 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
14870 | insn->dst_reg, | |
14871 | shift); | |
31fd8581 | 14872 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 14873 | (1ULL << size * 8) - 1); |
46f53a65 | 14874 | } |
31fd8581 | 14875 | } |
9bac3d6d | 14876 | |
8041902d | 14877 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
14878 | if (!new_prog) |
14879 | return -ENOMEM; | |
14880 | ||
3df126f3 | 14881 | delta += cnt - 1; |
9bac3d6d AS |
14882 | |
14883 | /* keep walking new program and skip insns we just inserted */ | |
14884 | env->prog = new_prog; | |
3df126f3 | 14885 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
14886 | } |
14887 | ||
14888 | return 0; | |
14889 | } | |
14890 | ||
1c2a088a AS |
14891 | static int jit_subprogs(struct bpf_verifier_env *env) |
14892 | { | |
14893 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
14894 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
a748c697 | 14895 | struct bpf_map *map_ptr; |
7105e828 | 14896 | struct bpf_insn *insn; |
1c2a088a | 14897 | void *old_bpf_func; |
c4c0bdc0 | 14898 | int err, num_exentries; |
1c2a088a | 14899 | |
f910cefa | 14900 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
14901 | return 0; |
14902 | ||
7105e828 | 14903 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
3990ed4c | 14904 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn)) |
69c087ba | 14905 | continue; |
69c087ba | 14906 | |
c7a89784 DB |
14907 | /* Upon error here we cannot fall back to interpreter but |
14908 | * need a hard reject of the program. Thus -EFAULT is | |
14909 | * propagated in any case. | |
14910 | */ | |
1c2a088a AS |
14911 | subprog = find_subprog(env, i + insn->imm + 1); |
14912 | if (subprog < 0) { | |
14913 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
14914 | i + insn->imm + 1); | |
14915 | return -EFAULT; | |
14916 | } | |
14917 | /* temporarily remember subprog id inside insn instead of | |
14918 | * aux_data, since next loop will split up all insns into funcs | |
14919 | */ | |
f910cefa | 14920 | insn->off = subprog; |
1c2a088a AS |
14921 | /* remember original imm in case JIT fails and fallback |
14922 | * to interpreter will be needed | |
14923 | */ | |
14924 | env->insn_aux_data[i].call_imm = insn->imm; | |
14925 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
14926 | insn->imm = 1; | |
3990ed4c MKL |
14927 | if (bpf_pseudo_func(insn)) |
14928 | /* jit (e.g. x86_64) may emit fewer instructions | |
14929 | * if it learns a u32 imm is the same as a u64 imm. | |
14930 | * Force a non zero here. | |
14931 | */ | |
14932 | insn[1].imm = 1; | |
1c2a088a AS |
14933 | } |
14934 | ||
c454a46b MKL |
14935 | err = bpf_prog_alloc_jited_linfo(prog); |
14936 | if (err) | |
14937 | goto out_undo_insn; | |
14938 | ||
14939 | err = -ENOMEM; | |
6396bb22 | 14940 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 14941 | if (!func) |
c7a89784 | 14942 | goto out_undo_insn; |
1c2a088a | 14943 | |
f910cefa | 14944 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 14945 | subprog_start = subprog_end; |
4cb3d99c | 14946 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
14947 | |
14948 | len = subprog_end - subprog_start; | |
fb7dd8bc | 14949 | /* bpf_prog_run() doesn't call subprogs directly, |
492ecee8 AS |
14950 | * hence main prog stats include the runtime of subprogs. |
14951 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
700d4796 | 14952 | * func[i]->stats will never be accessed and stays NULL |
492ecee8 AS |
14953 | */ |
14954 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
14955 | if (!func[i]) |
14956 | goto out_free; | |
14957 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
14958 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 14959 | func[i]->type = prog->type; |
1c2a088a | 14960 | func[i]->len = len; |
4f74d809 DB |
14961 | if (bpf_prog_calc_tag(func[i])) |
14962 | goto out_free; | |
1c2a088a | 14963 | func[i]->is_func = 1; |
ba64e7d8 | 14964 | func[i]->aux->func_idx = i; |
f263a814 | 14965 | /* Below members will be freed only at prog->aux */ |
ba64e7d8 YS |
14966 | func[i]->aux->btf = prog->aux->btf; |
14967 | func[i]->aux->func_info = prog->aux->func_info; | |
9c7c48d6 | 14968 | func[i]->aux->func_info_cnt = prog->aux->func_info_cnt; |
f263a814 JF |
14969 | func[i]->aux->poke_tab = prog->aux->poke_tab; |
14970 | func[i]->aux->size_poke_tab = prog->aux->size_poke_tab; | |
ba64e7d8 | 14971 | |
a748c697 | 14972 | for (j = 0; j < prog->aux->size_poke_tab; j++) { |
f263a814 | 14973 | struct bpf_jit_poke_descriptor *poke; |
a748c697 | 14974 | |
f263a814 JF |
14975 | poke = &prog->aux->poke_tab[j]; |
14976 | if (poke->insn_idx < subprog_end && | |
14977 | poke->insn_idx >= subprog_start) | |
14978 | poke->aux = func[i]->aux; | |
a748c697 MF |
14979 | } |
14980 | ||
1c2a088a | 14981 | func[i]->aux->name[0] = 'F'; |
9c8105bd | 14982 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 14983 | func[i]->jit_requested = 1; |
d2a3b7c5 | 14984 | func[i]->blinding_requested = prog->blinding_requested; |
e6ac2450 | 14985 | func[i]->aux->kfunc_tab = prog->aux->kfunc_tab; |
2357672c | 14986 | func[i]->aux->kfunc_btf_tab = prog->aux->kfunc_btf_tab; |
c454a46b MKL |
14987 | func[i]->aux->linfo = prog->aux->linfo; |
14988 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
14989 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
14990 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
c4c0bdc0 YS |
14991 | num_exentries = 0; |
14992 | insn = func[i]->insnsi; | |
14993 | for (j = 0; j < func[i]->len; j++, insn++) { | |
14994 | if (BPF_CLASS(insn->code) == BPF_LDX && | |
14995 | BPF_MODE(insn->code) == BPF_PROBE_MEM) | |
14996 | num_exentries++; | |
14997 | } | |
14998 | func[i]->aux->num_exentries = num_exentries; | |
ebf7d1f5 | 14999 | func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable; |
1c2a088a AS |
15000 | func[i] = bpf_int_jit_compile(func[i]); |
15001 | if (!func[i]->jited) { | |
15002 | err = -ENOTSUPP; | |
15003 | goto out_free; | |
15004 | } | |
15005 | cond_resched(); | |
15006 | } | |
a748c697 | 15007 | |
1c2a088a AS |
15008 | /* at this point all bpf functions were successfully JITed |
15009 | * now populate all bpf_calls with correct addresses and | |
15010 | * run last pass of JIT | |
15011 | */ | |
f910cefa | 15012 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
15013 | insn = func[i]->insnsi; |
15014 | for (j = 0; j < func[i]->len; j++, insn++) { | |
69c087ba | 15015 | if (bpf_pseudo_func(insn)) { |
3990ed4c | 15016 | subprog = insn->off; |
69c087ba YS |
15017 | insn[0].imm = (u32)(long)func[subprog]->bpf_func; |
15018 | insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32; | |
15019 | continue; | |
15020 | } | |
23a2d70c | 15021 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
15022 | continue; |
15023 | subprog = insn->off; | |
3d717fad | 15024 | insn->imm = BPF_CALL_IMM(func[subprog]->bpf_func); |
1c2a088a | 15025 | } |
2162fed4 SD |
15026 | |
15027 | /* we use the aux data to keep a list of the start addresses | |
15028 | * of the JITed images for each function in the program | |
15029 | * | |
15030 | * for some architectures, such as powerpc64, the imm field | |
15031 | * might not be large enough to hold the offset of the start | |
15032 | * address of the callee's JITed image from __bpf_call_base | |
15033 | * | |
15034 | * in such cases, we can lookup the start address of a callee | |
15035 | * by using its subprog id, available from the off field of | |
15036 | * the call instruction, as an index for this list | |
15037 | */ | |
15038 | func[i]->aux->func = func; | |
15039 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 15040 | } |
f910cefa | 15041 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
15042 | old_bpf_func = func[i]->bpf_func; |
15043 | tmp = bpf_int_jit_compile(func[i]); | |
15044 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
15045 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 15046 | err = -ENOTSUPP; |
1c2a088a AS |
15047 | goto out_free; |
15048 | } | |
15049 | cond_resched(); | |
15050 | } | |
15051 | ||
15052 | /* finally lock prog and jit images for all functions and | |
15053 | * populate kallsysm | |
15054 | */ | |
f910cefa | 15055 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
15056 | bpf_prog_lock_ro(func[i]); |
15057 | bpf_prog_kallsyms_add(func[i]); | |
15058 | } | |
7105e828 DB |
15059 | |
15060 | /* Last step: make now unused interpreter insns from main | |
15061 | * prog consistent for later dump requests, so they can | |
15062 | * later look the same as if they were interpreted only. | |
15063 | */ | |
15064 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
69c087ba YS |
15065 | if (bpf_pseudo_func(insn)) { |
15066 | insn[0].imm = env->insn_aux_data[i].call_imm; | |
3990ed4c MKL |
15067 | insn[1].imm = insn->off; |
15068 | insn->off = 0; | |
69c087ba YS |
15069 | continue; |
15070 | } | |
23a2d70c | 15071 | if (!bpf_pseudo_call(insn)) |
7105e828 DB |
15072 | continue; |
15073 | insn->off = env->insn_aux_data[i].call_imm; | |
15074 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 15075 | insn->imm = subprog; |
7105e828 DB |
15076 | } |
15077 | ||
1c2a088a AS |
15078 | prog->jited = 1; |
15079 | prog->bpf_func = func[0]->bpf_func; | |
d00c6473 | 15080 | prog->jited_len = func[0]->jited_len; |
1c2a088a | 15081 | prog->aux->func = func; |
f910cefa | 15082 | prog->aux->func_cnt = env->subprog_cnt; |
e16301fb | 15083 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
15084 | return 0; |
15085 | out_free: | |
f263a814 JF |
15086 | /* We failed JIT'ing, so at this point we need to unregister poke |
15087 | * descriptors from subprogs, so that kernel is not attempting to | |
15088 | * patch it anymore as we're freeing the subprog JIT memory. | |
15089 | */ | |
15090 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
15091 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
15092 | map_ptr->ops->map_poke_untrack(map_ptr, prog->aux); | |
15093 | } | |
15094 | /* At this point we're guaranteed that poke descriptors are not | |
15095 | * live anymore. We can just unlink its descriptor table as it's | |
15096 | * released with the main prog. | |
15097 | */ | |
a748c697 MF |
15098 | for (i = 0; i < env->subprog_cnt; i++) { |
15099 | if (!func[i]) | |
15100 | continue; | |
f263a814 | 15101 | func[i]->aux->poke_tab = NULL; |
a748c697 MF |
15102 | bpf_jit_free(func[i]); |
15103 | } | |
1c2a088a | 15104 | kfree(func); |
c7a89784 | 15105 | out_undo_insn: |
1c2a088a AS |
15106 | /* cleanup main prog to be interpreted */ |
15107 | prog->jit_requested = 0; | |
d2a3b7c5 | 15108 | prog->blinding_requested = 0; |
1c2a088a | 15109 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
23a2d70c | 15110 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
15111 | continue; |
15112 | insn->off = 0; | |
15113 | insn->imm = env->insn_aux_data[i].call_imm; | |
15114 | } | |
e16301fb | 15115 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
15116 | return err; |
15117 | } | |
15118 | ||
1ea47e01 AS |
15119 | static int fixup_call_args(struct bpf_verifier_env *env) |
15120 | { | |
19d28fbd | 15121 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
15122 | struct bpf_prog *prog = env->prog; |
15123 | struct bpf_insn *insn = prog->insnsi; | |
e6ac2450 | 15124 | bool has_kfunc_call = bpf_prog_has_kfunc_call(prog); |
1ea47e01 | 15125 | int i, depth; |
19d28fbd | 15126 | #endif |
e4052d06 | 15127 | int err = 0; |
1ea47e01 | 15128 | |
e4052d06 QM |
15129 | if (env->prog->jit_requested && |
15130 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
15131 | err = jit_subprogs(env); |
15132 | if (err == 0) | |
1c2a088a | 15133 | return 0; |
c7a89784 DB |
15134 | if (err == -EFAULT) |
15135 | return err; | |
19d28fbd DM |
15136 | } |
15137 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
e6ac2450 MKL |
15138 | if (has_kfunc_call) { |
15139 | verbose(env, "calling kernel functions are not allowed in non-JITed programs\n"); | |
15140 | return -EINVAL; | |
15141 | } | |
e411901c MF |
15142 | if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) { |
15143 | /* When JIT fails the progs with bpf2bpf calls and tail_calls | |
15144 | * have to be rejected, since interpreter doesn't support them yet. | |
15145 | */ | |
15146 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
15147 | return -EINVAL; | |
15148 | } | |
1ea47e01 | 15149 | for (i = 0; i < prog->len; i++, insn++) { |
69c087ba YS |
15150 | if (bpf_pseudo_func(insn)) { |
15151 | /* When JIT fails the progs with callback calls | |
15152 | * have to be rejected, since interpreter doesn't support them yet. | |
15153 | */ | |
15154 | verbose(env, "callbacks are not allowed in non-JITed programs\n"); | |
15155 | return -EINVAL; | |
15156 | } | |
15157 | ||
23a2d70c | 15158 | if (!bpf_pseudo_call(insn)) |
1ea47e01 AS |
15159 | continue; |
15160 | depth = get_callee_stack_depth(env, insn, i); | |
15161 | if (depth < 0) | |
15162 | return depth; | |
15163 | bpf_patch_call_args(insn, depth); | |
15164 | } | |
19d28fbd DM |
15165 | err = 0; |
15166 | #endif | |
15167 | return err; | |
1ea47e01 AS |
15168 | } |
15169 | ||
958cf2e2 KKD |
15170 | static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
15171 | struct bpf_insn *insn_buf, int insn_idx, int *cnt) | |
e6ac2450 MKL |
15172 | { |
15173 | const struct bpf_kfunc_desc *desc; | |
15174 | ||
a5d82727 KKD |
15175 | if (!insn->imm) { |
15176 | verbose(env, "invalid kernel function call not eliminated in verifier pass\n"); | |
15177 | return -EINVAL; | |
15178 | } | |
15179 | ||
e6ac2450 MKL |
15180 | /* insn->imm has the btf func_id. Replace it with |
15181 | * an address (relative to __bpf_base_call). | |
15182 | */ | |
2357672c | 15183 | desc = find_kfunc_desc(env->prog, insn->imm, insn->off); |
e6ac2450 MKL |
15184 | if (!desc) { |
15185 | verbose(env, "verifier internal error: kernel function descriptor not found for func_id %u\n", | |
15186 | insn->imm); | |
15187 | return -EFAULT; | |
15188 | } | |
15189 | ||
958cf2e2 | 15190 | *cnt = 0; |
e6ac2450 | 15191 | insn->imm = desc->imm; |
958cf2e2 KKD |
15192 | if (insn->off) |
15193 | return 0; | |
15194 | if (desc->func_id == special_kfunc_list[KF_bpf_obj_new_impl]) { | |
15195 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; | |
15196 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
15197 | u64 obj_new_size = env->insn_aux_data[insn_idx].obj_new_size; | |
e6ac2450 | 15198 | |
958cf2e2 KKD |
15199 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_1, obj_new_size); |
15200 | insn_buf[1] = addr[0]; | |
15201 | insn_buf[2] = addr[1]; | |
15202 | insn_buf[3] = *insn; | |
15203 | *cnt = 4; | |
ac9f0605 KKD |
15204 | } else if (desc->func_id == special_kfunc_list[KF_bpf_obj_drop_impl]) { |
15205 | struct btf_struct_meta *kptr_struct_meta = env->insn_aux_data[insn_idx].kptr_struct_meta; | |
15206 | struct bpf_insn addr[2] = { BPF_LD_IMM64(BPF_REG_2, (long)kptr_struct_meta) }; | |
15207 | ||
15208 | insn_buf[0] = addr[0]; | |
15209 | insn_buf[1] = addr[1]; | |
15210 | insn_buf[2] = *insn; | |
15211 | *cnt = 3; | |
a35b9af4 YS |
15212 | } else if (desc->func_id == special_kfunc_list[KF_bpf_cast_to_kern_ctx] || |
15213 | desc->func_id == special_kfunc_list[KF_bpf_rdonly_cast]) { | |
fd264ca0 YS |
15214 | insn_buf[0] = BPF_MOV64_REG(BPF_REG_0, BPF_REG_1); |
15215 | *cnt = 1; | |
958cf2e2 | 15216 | } |
e6ac2450 MKL |
15217 | return 0; |
15218 | } | |
15219 | ||
e6ac5933 BJ |
15220 | /* Do various post-verification rewrites in a single program pass. |
15221 | * These rewrites simplify JIT and interpreter implementations. | |
e245c5c6 | 15222 | */ |
e6ac5933 | 15223 | static int do_misc_fixups(struct bpf_verifier_env *env) |
e245c5c6 | 15224 | { |
79741b3b | 15225 | struct bpf_prog *prog = env->prog; |
f92c1e18 | 15226 | enum bpf_attach_type eatype = prog->expected_attach_type; |
9b99edca | 15227 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
79741b3b | 15228 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 15229 | const struct bpf_func_proto *fn; |
79741b3b | 15230 | const int insn_cnt = prog->len; |
09772d92 | 15231 | const struct bpf_map_ops *ops; |
c93552c4 | 15232 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
15233 | struct bpf_insn insn_buf[16]; |
15234 | struct bpf_prog *new_prog; | |
15235 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 15236 | int i, ret, cnt, delta = 0; |
e245c5c6 | 15237 | |
79741b3b | 15238 | for (i = 0; i < insn_cnt; i++, insn++) { |
e6ac5933 | 15239 | /* Make divide-by-zero exceptions impossible. */ |
f6b1b3bf DB |
15240 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
15241 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
15242 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 15243 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf | 15244 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
e88b2c6e DB |
15245 | bool isdiv = BPF_OP(insn->code) == BPF_DIV; |
15246 | struct bpf_insn *patchlet; | |
15247 | struct bpf_insn chk_and_div[] = { | |
9b00f1b7 | 15248 | /* [R,W]x div 0 -> 0 */ |
e88b2c6e DB |
15249 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
15250 | BPF_JNE | BPF_K, insn->src_reg, | |
15251 | 0, 2, 0), | |
f6b1b3bf DB |
15252 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), |
15253 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
15254 | *insn, | |
15255 | }; | |
e88b2c6e | 15256 | struct bpf_insn chk_and_mod[] = { |
9b00f1b7 | 15257 | /* [R,W]x mod 0 -> [R,W]x */ |
e88b2c6e DB |
15258 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
15259 | BPF_JEQ | BPF_K, insn->src_reg, | |
9b00f1b7 | 15260 | 0, 1 + (is64 ? 0 : 1), 0), |
f6b1b3bf | 15261 | *insn, |
9b00f1b7 DB |
15262 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), |
15263 | BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), | |
f6b1b3bf | 15264 | }; |
f6b1b3bf | 15265 | |
e88b2c6e DB |
15266 | patchlet = isdiv ? chk_and_div : chk_and_mod; |
15267 | cnt = isdiv ? ARRAY_SIZE(chk_and_div) : | |
9b00f1b7 | 15268 | ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0); |
f6b1b3bf DB |
15269 | |
15270 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
15271 | if (!new_prog) |
15272 | return -ENOMEM; | |
15273 | ||
15274 | delta += cnt - 1; | |
15275 | env->prog = prog = new_prog; | |
15276 | insn = new_prog->insnsi + i + delta; | |
15277 | continue; | |
15278 | } | |
15279 | ||
e6ac5933 | 15280 | /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */ |
e0cea7ce DB |
15281 | if (BPF_CLASS(insn->code) == BPF_LD && |
15282 | (BPF_MODE(insn->code) == BPF_ABS || | |
15283 | BPF_MODE(insn->code) == BPF_IND)) { | |
15284 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
15285 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
15286 | verbose(env, "bpf verifier is misconfigured\n"); | |
15287 | return -EINVAL; | |
15288 | } | |
15289 | ||
15290 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15291 | if (!new_prog) | |
15292 | return -ENOMEM; | |
15293 | ||
15294 | delta += cnt - 1; | |
15295 | env->prog = prog = new_prog; | |
15296 | insn = new_prog->insnsi + i + delta; | |
15297 | continue; | |
15298 | } | |
15299 | ||
e6ac5933 | 15300 | /* Rewrite pointer arithmetic to mitigate speculation attacks. */ |
979d63d5 DB |
15301 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
15302 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
15303 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
15304 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
979d63d5 | 15305 | struct bpf_insn *patch = &insn_buf[0]; |
801c6058 | 15306 | bool issrc, isneg, isimm; |
979d63d5 DB |
15307 | u32 off_reg; |
15308 | ||
15309 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
15310 | if (!aux->alu_state || |
15311 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
15312 | continue; |
15313 | ||
15314 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
15315 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
15316 | BPF_ALU_SANITIZE_SRC; | |
801c6058 | 15317 | isimm = aux->alu_state & BPF_ALU_IMMEDIATE; |
979d63d5 DB |
15318 | |
15319 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
801c6058 DB |
15320 | if (isimm) { |
15321 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
15322 | } else { | |
15323 | if (isneg) | |
15324 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
15325 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
15326 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
15327 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
15328 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
15329 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
15330 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, off_reg); | |
15331 | } | |
b9b34ddb DB |
15332 | if (!issrc) |
15333 | *patch++ = BPF_MOV64_REG(insn->dst_reg, insn->src_reg); | |
15334 | insn->src_reg = BPF_REG_AX; | |
979d63d5 DB |
15335 | if (isneg) |
15336 | insn->code = insn->code == code_add ? | |
15337 | code_sub : code_add; | |
15338 | *patch++ = *insn; | |
801c6058 | 15339 | if (issrc && isneg && !isimm) |
979d63d5 DB |
15340 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); |
15341 | cnt = patch - insn_buf; | |
15342 | ||
15343 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15344 | if (!new_prog) | |
15345 | return -ENOMEM; | |
15346 | ||
15347 | delta += cnt - 1; | |
15348 | env->prog = prog = new_prog; | |
15349 | insn = new_prog->insnsi + i + delta; | |
15350 | continue; | |
15351 | } | |
15352 | ||
79741b3b AS |
15353 | if (insn->code != (BPF_JMP | BPF_CALL)) |
15354 | continue; | |
cc8b0b92 AS |
15355 | if (insn->src_reg == BPF_PSEUDO_CALL) |
15356 | continue; | |
e6ac2450 | 15357 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
958cf2e2 | 15358 | ret = fixup_kfunc_call(env, insn, insn_buf, i + delta, &cnt); |
e6ac2450 MKL |
15359 | if (ret) |
15360 | return ret; | |
958cf2e2 KKD |
15361 | if (cnt == 0) |
15362 | continue; | |
15363 | ||
15364 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15365 | if (!new_prog) | |
15366 | return -ENOMEM; | |
15367 | ||
15368 | delta += cnt - 1; | |
15369 | env->prog = prog = new_prog; | |
15370 | insn = new_prog->insnsi + i + delta; | |
e6ac2450 MKL |
15371 | continue; |
15372 | } | |
e245c5c6 | 15373 | |
79741b3b AS |
15374 | if (insn->imm == BPF_FUNC_get_route_realm) |
15375 | prog->dst_needed = 1; | |
15376 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
15377 | bpf_user_rnd_init_once(); | |
9802d865 JB |
15378 | if (insn->imm == BPF_FUNC_override_return) |
15379 | prog->kprobe_override = 1; | |
79741b3b | 15380 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
15381 | /* If we tail call into other programs, we |
15382 | * cannot make any assumptions since they can | |
15383 | * be replaced dynamically during runtime in | |
15384 | * the program array. | |
15385 | */ | |
15386 | prog->cb_access = 1; | |
e411901c MF |
15387 | if (!allow_tail_call_in_subprogs(env)) |
15388 | prog->aux->stack_depth = MAX_BPF_STACK; | |
15389 | prog->aux->max_pkt_offset = MAX_PACKET_OFF; | |
7b9f6da1 | 15390 | |
79741b3b | 15391 | /* mark bpf_tail_call as different opcode to avoid |
8fb33b60 | 15392 | * conditional branch in the interpreter for every normal |
79741b3b AS |
15393 | * call and to prevent accidental JITing by JIT compiler |
15394 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 15395 | */ |
79741b3b | 15396 | insn->imm = 0; |
71189fa9 | 15397 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 15398 | |
c93552c4 | 15399 | aux = &env->insn_aux_data[i + delta]; |
d2a3b7c5 | 15400 | if (env->bpf_capable && !prog->blinding_requested && |
cc52d914 | 15401 | prog->jit_requested && |
d2e4c1e6 DB |
15402 | !bpf_map_key_poisoned(aux) && |
15403 | !bpf_map_ptr_poisoned(aux) && | |
15404 | !bpf_map_ptr_unpriv(aux)) { | |
15405 | struct bpf_jit_poke_descriptor desc = { | |
15406 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
15407 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
15408 | .tail_call.key = bpf_map_key_immediate(aux), | |
a748c697 | 15409 | .insn_idx = i + delta, |
d2e4c1e6 DB |
15410 | }; |
15411 | ||
15412 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
15413 | if (ret < 0) { | |
15414 | verbose(env, "adding tail call poke descriptor failed\n"); | |
15415 | return ret; | |
15416 | } | |
15417 | ||
15418 | insn->imm = ret + 1; | |
15419 | continue; | |
15420 | } | |
15421 | ||
c93552c4 DB |
15422 | if (!bpf_map_ptr_unpriv(aux)) |
15423 | continue; | |
15424 | ||
b2157399 AS |
15425 | /* instead of changing every JIT dealing with tail_call |
15426 | * emit two extra insns: | |
15427 | * if (index >= max_entries) goto out; | |
15428 | * index &= array->index_mask; | |
15429 | * to avoid out-of-bounds cpu speculation | |
15430 | */ | |
c93552c4 | 15431 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 15432 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
15433 | return -EINVAL; |
15434 | } | |
c93552c4 | 15435 | |
d2e4c1e6 | 15436 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
15437 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
15438 | map_ptr->max_entries, 2); | |
15439 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
15440 | container_of(map_ptr, | |
15441 | struct bpf_array, | |
15442 | map)->index_mask); | |
15443 | insn_buf[2] = *insn; | |
15444 | cnt = 3; | |
15445 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15446 | if (!new_prog) | |
15447 | return -ENOMEM; | |
15448 | ||
15449 | delta += cnt - 1; | |
15450 | env->prog = prog = new_prog; | |
15451 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
15452 | continue; |
15453 | } | |
e245c5c6 | 15454 | |
b00628b1 AS |
15455 | if (insn->imm == BPF_FUNC_timer_set_callback) { |
15456 | /* The verifier will process callback_fn as many times as necessary | |
15457 | * with different maps and the register states prepared by | |
15458 | * set_timer_callback_state will be accurate. | |
15459 | * | |
15460 | * The following use case is valid: | |
15461 | * map1 is shared by prog1, prog2, prog3. | |
15462 | * prog1 calls bpf_timer_init for some map1 elements | |
15463 | * prog2 calls bpf_timer_set_callback for some map1 elements. | |
15464 | * Those that were not bpf_timer_init-ed will return -EINVAL. | |
15465 | * prog3 calls bpf_timer_start for some map1 elements. | |
15466 | * Those that were not both bpf_timer_init-ed and | |
15467 | * bpf_timer_set_callback-ed will return -EINVAL. | |
15468 | */ | |
15469 | struct bpf_insn ld_addrs[2] = { | |
15470 | BPF_LD_IMM64(BPF_REG_3, (long)prog->aux), | |
15471 | }; | |
15472 | ||
15473 | insn_buf[0] = ld_addrs[0]; | |
15474 | insn_buf[1] = ld_addrs[1]; | |
15475 | insn_buf[2] = *insn; | |
15476 | cnt = 3; | |
15477 | ||
15478 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15479 | if (!new_prog) | |
15480 | return -ENOMEM; | |
15481 | ||
15482 | delta += cnt - 1; | |
15483 | env->prog = prog = new_prog; | |
15484 | insn = new_prog->insnsi + i + delta; | |
15485 | goto patch_call_imm; | |
15486 | } | |
15487 | ||
b00fa38a JK |
15488 | if (insn->imm == BPF_FUNC_task_storage_get || |
15489 | insn->imm == BPF_FUNC_sk_storage_get || | |
c4bcfb38 YS |
15490 | insn->imm == BPF_FUNC_inode_storage_get || |
15491 | insn->imm == BPF_FUNC_cgrp_storage_get) { | |
b00fa38a | 15492 | if (env->prog->aux->sleepable) |
d56c9fe6 | 15493 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_KERNEL); |
b00fa38a | 15494 | else |
d56c9fe6 | 15495 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_5, (__force __s32)GFP_ATOMIC); |
b00fa38a JK |
15496 | insn_buf[1] = *insn; |
15497 | cnt = 2; | |
15498 | ||
15499 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15500 | if (!new_prog) | |
15501 | return -ENOMEM; | |
15502 | ||
15503 | delta += cnt - 1; | |
15504 | env->prog = prog = new_prog; | |
15505 | insn = new_prog->insnsi + i + delta; | |
15506 | goto patch_call_imm; | |
15507 | } | |
15508 | ||
89c63074 | 15509 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
15510 | * and other inlining handlers are currently limited to 64 bit |
15511 | * only. | |
89c63074 | 15512 | */ |
60b58afc | 15513 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
15514 | (insn->imm == BPF_FUNC_map_lookup_elem || |
15515 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
15516 | insn->imm == BPF_FUNC_map_delete_elem || |
15517 | insn->imm == BPF_FUNC_map_push_elem || | |
15518 | insn->imm == BPF_FUNC_map_pop_elem || | |
e6a4750f | 15519 | insn->imm == BPF_FUNC_map_peek_elem || |
0640c77c | 15520 | insn->imm == BPF_FUNC_redirect_map || |
07343110 FZ |
15521 | insn->imm == BPF_FUNC_for_each_map_elem || |
15522 | insn->imm == BPF_FUNC_map_lookup_percpu_elem)) { | |
c93552c4 DB |
15523 | aux = &env->insn_aux_data[i + delta]; |
15524 | if (bpf_map_ptr_poisoned(aux)) | |
15525 | goto patch_call_imm; | |
15526 | ||
d2e4c1e6 | 15527 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
15528 | ops = map_ptr->ops; |
15529 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
15530 | ops->map_gen_lookup) { | |
15531 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
4a8f87e6 DB |
15532 | if (cnt == -EOPNOTSUPP) |
15533 | goto patch_map_ops_generic; | |
15534 | if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
09772d92 DB |
15535 | verbose(env, "bpf verifier is misconfigured\n"); |
15536 | return -EINVAL; | |
15537 | } | |
81ed18ab | 15538 | |
09772d92 DB |
15539 | new_prog = bpf_patch_insn_data(env, i + delta, |
15540 | insn_buf, cnt); | |
15541 | if (!new_prog) | |
15542 | return -ENOMEM; | |
81ed18ab | 15543 | |
09772d92 DB |
15544 | delta += cnt - 1; |
15545 | env->prog = prog = new_prog; | |
15546 | insn = new_prog->insnsi + i + delta; | |
15547 | continue; | |
15548 | } | |
81ed18ab | 15549 | |
09772d92 DB |
15550 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
15551 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
15552 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
15553 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
15554 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
15555 | (int (*)(struct bpf_map *map, void *key, void *value, | |
15556 | u64 flags))NULL)); | |
84430d42 DB |
15557 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
15558 | (int (*)(struct bpf_map *map, void *value, | |
15559 | u64 flags))NULL)); | |
15560 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
15561 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
15562 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
15563 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
e6a4750f | 15564 | BUILD_BUG_ON(!__same_type(ops->map_redirect, |
32637e33 | 15565 | (int (*)(struct bpf_map *map, u64 index, u64 flags))NULL)); |
0640c77c AI |
15566 | BUILD_BUG_ON(!__same_type(ops->map_for_each_callback, |
15567 | (int (*)(struct bpf_map *map, | |
15568 | bpf_callback_t callback_fn, | |
15569 | void *callback_ctx, | |
15570 | u64 flags))NULL)); | |
07343110 FZ |
15571 | BUILD_BUG_ON(!__same_type(ops->map_lookup_percpu_elem, |
15572 | (void *(*)(struct bpf_map *map, void *key, u32 cpu))NULL)); | |
e6a4750f | 15573 | |
4a8f87e6 | 15574 | patch_map_ops_generic: |
09772d92 DB |
15575 | switch (insn->imm) { |
15576 | case BPF_FUNC_map_lookup_elem: | |
3d717fad | 15577 | insn->imm = BPF_CALL_IMM(ops->map_lookup_elem); |
09772d92 DB |
15578 | continue; |
15579 | case BPF_FUNC_map_update_elem: | |
3d717fad | 15580 | insn->imm = BPF_CALL_IMM(ops->map_update_elem); |
09772d92 DB |
15581 | continue; |
15582 | case BPF_FUNC_map_delete_elem: | |
3d717fad | 15583 | insn->imm = BPF_CALL_IMM(ops->map_delete_elem); |
09772d92 | 15584 | continue; |
84430d42 | 15585 | case BPF_FUNC_map_push_elem: |
3d717fad | 15586 | insn->imm = BPF_CALL_IMM(ops->map_push_elem); |
84430d42 DB |
15587 | continue; |
15588 | case BPF_FUNC_map_pop_elem: | |
3d717fad | 15589 | insn->imm = BPF_CALL_IMM(ops->map_pop_elem); |
84430d42 DB |
15590 | continue; |
15591 | case BPF_FUNC_map_peek_elem: | |
3d717fad | 15592 | insn->imm = BPF_CALL_IMM(ops->map_peek_elem); |
84430d42 | 15593 | continue; |
e6a4750f | 15594 | case BPF_FUNC_redirect_map: |
3d717fad | 15595 | insn->imm = BPF_CALL_IMM(ops->map_redirect); |
e6a4750f | 15596 | continue; |
0640c77c AI |
15597 | case BPF_FUNC_for_each_map_elem: |
15598 | insn->imm = BPF_CALL_IMM(ops->map_for_each_callback); | |
e6a4750f | 15599 | continue; |
07343110 FZ |
15600 | case BPF_FUNC_map_lookup_percpu_elem: |
15601 | insn->imm = BPF_CALL_IMM(ops->map_lookup_percpu_elem); | |
15602 | continue; | |
09772d92 | 15603 | } |
81ed18ab | 15604 | |
09772d92 | 15605 | goto patch_call_imm; |
81ed18ab AS |
15606 | } |
15607 | ||
e6ac5933 | 15608 | /* Implement bpf_jiffies64 inline. */ |
5576b991 MKL |
15609 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
15610 | insn->imm == BPF_FUNC_jiffies64) { | |
15611 | struct bpf_insn ld_jiffies_addr[2] = { | |
15612 | BPF_LD_IMM64(BPF_REG_0, | |
15613 | (unsigned long)&jiffies), | |
15614 | }; | |
15615 | ||
15616 | insn_buf[0] = ld_jiffies_addr[0]; | |
15617 | insn_buf[1] = ld_jiffies_addr[1]; | |
15618 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
15619 | BPF_REG_0, 0); | |
15620 | cnt = 3; | |
15621 | ||
15622 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
15623 | cnt); | |
15624 | if (!new_prog) | |
15625 | return -ENOMEM; | |
15626 | ||
15627 | delta += cnt - 1; | |
15628 | env->prog = prog = new_prog; | |
15629 | insn = new_prog->insnsi + i + delta; | |
15630 | continue; | |
15631 | } | |
15632 | ||
f92c1e18 JO |
15633 | /* Implement bpf_get_func_arg inline. */ |
15634 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
15635 | insn->imm == BPF_FUNC_get_func_arg) { | |
15636 | /* Load nr_args from ctx - 8 */ | |
15637 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
15638 | insn_buf[1] = BPF_JMP32_REG(BPF_JGE, BPF_REG_2, BPF_REG_0, 6); | |
15639 | insn_buf[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_2, 3); | |
15640 | insn_buf[3] = BPF_ALU64_REG(BPF_ADD, BPF_REG_2, BPF_REG_1); | |
15641 | insn_buf[4] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_2, 0); | |
15642 | insn_buf[5] = BPF_STX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
15643 | insn_buf[6] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
15644 | insn_buf[7] = BPF_JMP_A(1); | |
15645 | insn_buf[8] = BPF_MOV64_IMM(BPF_REG_0, -EINVAL); | |
15646 | cnt = 9; | |
15647 | ||
15648 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15649 | if (!new_prog) | |
15650 | return -ENOMEM; | |
15651 | ||
15652 | delta += cnt - 1; | |
15653 | env->prog = prog = new_prog; | |
15654 | insn = new_prog->insnsi + i + delta; | |
15655 | continue; | |
15656 | } | |
15657 | ||
15658 | /* Implement bpf_get_func_ret inline. */ | |
15659 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
15660 | insn->imm == BPF_FUNC_get_func_ret) { | |
15661 | if (eatype == BPF_TRACE_FEXIT || | |
15662 | eatype == BPF_MODIFY_RETURN) { | |
15663 | /* Load nr_args from ctx - 8 */ | |
15664 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
15665 | insn_buf[1] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_0, 3); | |
15666 | insn_buf[2] = BPF_ALU64_REG(BPF_ADD, BPF_REG_0, BPF_REG_1); | |
15667 | insn_buf[3] = BPF_LDX_MEM(BPF_DW, BPF_REG_3, BPF_REG_0, 0); | |
15668 | insn_buf[4] = BPF_STX_MEM(BPF_DW, BPF_REG_2, BPF_REG_3, 0); | |
15669 | insn_buf[5] = BPF_MOV64_IMM(BPF_REG_0, 0); | |
15670 | cnt = 6; | |
15671 | } else { | |
15672 | insn_buf[0] = BPF_MOV64_IMM(BPF_REG_0, -EOPNOTSUPP); | |
15673 | cnt = 1; | |
15674 | } | |
15675 | ||
15676 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
15677 | if (!new_prog) | |
15678 | return -ENOMEM; | |
15679 | ||
15680 | delta += cnt - 1; | |
15681 | env->prog = prog = new_prog; | |
15682 | insn = new_prog->insnsi + i + delta; | |
15683 | continue; | |
15684 | } | |
15685 | ||
15686 | /* Implement get_func_arg_cnt inline. */ | |
15687 | if (prog_type == BPF_PROG_TYPE_TRACING && | |
15688 | insn->imm == BPF_FUNC_get_func_arg_cnt) { | |
15689 | /* Load nr_args from ctx - 8 */ | |
15690 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8); | |
15691 | ||
15692 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
15693 | if (!new_prog) | |
15694 | return -ENOMEM; | |
15695 | ||
15696 | env->prog = prog = new_prog; | |
15697 | insn = new_prog->insnsi + i + delta; | |
15698 | continue; | |
15699 | } | |
15700 | ||
f705ec76 | 15701 | /* Implement bpf_get_func_ip inline. */ |
9b99edca JO |
15702 | if (prog_type == BPF_PROG_TYPE_TRACING && |
15703 | insn->imm == BPF_FUNC_get_func_ip) { | |
f92c1e18 JO |
15704 | /* Load IP address from ctx - 16 */ |
15705 | insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -16); | |
9b99edca JO |
15706 | |
15707 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, 1); | |
15708 | if (!new_prog) | |
15709 | return -ENOMEM; | |
15710 | ||
15711 | env->prog = prog = new_prog; | |
15712 | insn = new_prog->insnsi + i + delta; | |
15713 | continue; | |
15714 | } | |
15715 | ||
81ed18ab | 15716 | patch_call_imm: |
5e43f899 | 15717 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
15718 | /* all functions that have prototype and verifier allowed |
15719 | * programs to call them, must be real in-kernel functions | |
15720 | */ | |
15721 | if (!fn->func) { | |
61bd5218 JK |
15722 | verbose(env, |
15723 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
15724 | func_id_name(insn->imm), insn->imm); |
15725 | return -EFAULT; | |
e245c5c6 | 15726 | } |
79741b3b | 15727 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 15728 | } |
e245c5c6 | 15729 | |
d2e4c1e6 DB |
15730 | /* Since poke tab is now finalized, publish aux to tracker. */ |
15731 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
15732 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
15733 | if (!map_ptr->ops->map_poke_track || | |
15734 | !map_ptr->ops->map_poke_untrack || | |
15735 | !map_ptr->ops->map_poke_run) { | |
15736 | verbose(env, "bpf verifier is misconfigured\n"); | |
15737 | return -EINVAL; | |
15738 | } | |
15739 | ||
15740 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
15741 | if (ret < 0) { | |
15742 | verbose(env, "tracking tail call prog failed\n"); | |
15743 | return ret; | |
15744 | } | |
15745 | } | |
15746 | ||
e6ac2450 MKL |
15747 | sort_kfunc_descs_by_imm(env->prog); |
15748 | ||
79741b3b AS |
15749 | return 0; |
15750 | } | |
e245c5c6 | 15751 | |
1ade2371 EZ |
15752 | static struct bpf_prog *inline_bpf_loop(struct bpf_verifier_env *env, |
15753 | int position, | |
15754 | s32 stack_base, | |
15755 | u32 callback_subprogno, | |
15756 | u32 *cnt) | |
15757 | { | |
15758 | s32 r6_offset = stack_base + 0 * BPF_REG_SIZE; | |
15759 | s32 r7_offset = stack_base + 1 * BPF_REG_SIZE; | |
15760 | s32 r8_offset = stack_base + 2 * BPF_REG_SIZE; | |
15761 | int reg_loop_max = BPF_REG_6; | |
15762 | int reg_loop_cnt = BPF_REG_7; | |
15763 | int reg_loop_ctx = BPF_REG_8; | |
15764 | ||
15765 | struct bpf_prog *new_prog; | |
15766 | u32 callback_start; | |
15767 | u32 call_insn_offset; | |
15768 | s32 callback_offset; | |
15769 | ||
15770 | /* This represents an inlined version of bpf_iter.c:bpf_loop, | |
15771 | * be careful to modify this code in sync. | |
15772 | */ | |
15773 | struct bpf_insn insn_buf[] = { | |
15774 | /* Return error and jump to the end of the patch if | |
15775 | * expected number of iterations is too big. | |
15776 | */ | |
15777 | BPF_JMP_IMM(BPF_JLE, BPF_REG_1, BPF_MAX_LOOPS, 2), | |
15778 | BPF_MOV32_IMM(BPF_REG_0, -E2BIG), | |
15779 | BPF_JMP_IMM(BPF_JA, 0, 0, 16), | |
15780 | /* spill R6, R7, R8 to use these as loop vars */ | |
15781 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_6, r6_offset), | |
15782 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_7, r7_offset), | |
15783 | BPF_STX_MEM(BPF_DW, BPF_REG_10, BPF_REG_8, r8_offset), | |
15784 | /* initialize loop vars */ | |
15785 | BPF_MOV64_REG(reg_loop_max, BPF_REG_1), | |
15786 | BPF_MOV32_IMM(reg_loop_cnt, 0), | |
15787 | BPF_MOV64_REG(reg_loop_ctx, BPF_REG_3), | |
15788 | /* loop header, | |
15789 | * if reg_loop_cnt >= reg_loop_max skip the loop body | |
15790 | */ | |
15791 | BPF_JMP_REG(BPF_JGE, reg_loop_cnt, reg_loop_max, 5), | |
15792 | /* callback call, | |
15793 | * correct callback offset would be set after patching | |
15794 | */ | |
15795 | BPF_MOV64_REG(BPF_REG_1, reg_loop_cnt), | |
15796 | BPF_MOV64_REG(BPF_REG_2, reg_loop_ctx), | |
15797 | BPF_CALL_REL(0), | |
15798 | /* increment loop counter */ | |
15799 | BPF_ALU64_IMM(BPF_ADD, reg_loop_cnt, 1), | |
15800 | /* jump to loop header if callback returned 0 */ | |
15801 | BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, -6), | |
15802 | /* return value of bpf_loop, | |
15803 | * set R0 to the number of iterations | |
15804 | */ | |
15805 | BPF_MOV64_REG(BPF_REG_0, reg_loop_cnt), | |
15806 | /* restore original values of R6, R7, R8 */ | |
15807 | BPF_LDX_MEM(BPF_DW, BPF_REG_6, BPF_REG_10, r6_offset), | |
15808 | BPF_LDX_MEM(BPF_DW, BPF_REG_7, BPF_REG_10, r7_offset), | |
15809 | BPF_LDX_MEM(BPF_DW, BPF_REG_8, BPF_REG_10, r8_offset), | |
15810 | }; | |
15811 | ||
15812 | *cnt = ARRAY_SIZE(insn_buf); | |
15813 | new_prog = bpf_patch_insn_data(env, position, insn_buf, *cnt); | |
15814 | if (!new_prog) | |
15815 | return new_prog; | |
15816 | ||
15817 | /* callback start is known only after patching */ | |
15818 | callback_start = env->subprog_info[callback_subprogno].start; | |
15819 | /* Note: insn_buf[12] is an offset of BPF_CALL_REL instruction */ | |
15820 | call_insn_offset = position + 12; | |
15821 | callback_offset = callback_start - call_insn_offset - 1; | |
fb4e3b33 | 15822 | new_prog->insnsi[call_insn_offset].imm = callback_offset; |
1ade2371 EZ |
15823 | |
15824 | return new_prog; | |
15825 | } | |
15826 | ||
15827 | static bool is_bpf_loop_call(struct bpf_insn *insn) | |
15828 | { | |
15829 | return insn->code == (BPF_JMP | BPF_CALL) && | |
15830 | insn->src_reg == 0 && | |
15831 | insn->imm == BPF_FUNC_loop; | |
15832 | } | |
15833 | ||
15834 | /* For all sub-programs in the program (including main) check | |
15835 | * insn_aux_data to see if there are bpf_loop calls that require | |
15836 | * inlining. If such calls are found the calls are replaced with a | |
15837 | * sequence of instructions produced by `inline_bpf_loop` function and | |
15838 | * subprog stack_depth is increased by the size of 3 registers. | |
15839 | * This stack space is used to spill values of the R6, R7, R8. These | |
15840 | * registers are used to store the loop bound, counter and context | |
15841 | * variables. | |
15842 | */ | |
15843 | static int optimize_bpf_loop(struct bpf_verifier_env *env) | |
15844 | { | |
15845 | struct bpf_subprog_info *subprogs = env->subprog_info; | |
15846 | int i, cur_subprog = 0, cnt, delta = 0; | |
15847 | struct bpf_insn *insn = env->prog->insnsi; | |
15848 | int insn_cnt = env->prog->len; | |
15849 | u16 stack_depth = subprogs[cur_subprog].stack_depth; | |
15850 | u16 stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
15851 | u16 stack_depth_extra = 0; | |
15852 | ||
15853 | for (i = 0; i < insn_cnt; i++, insn++) { | |
15854 | struct bpf_loop_inline_state *inline_state = | |
15855 | &env->insn_aux_data[i + delta].loop_inline_state; | |
15856 | ||
15857 | if (is_bpf_loop_call(insn) && inline_state->fit_for_inline) { | |
15858 | struct bpf_prog *new_prog; | |
15859 | ||
15860 | stack_depth_extra = BPF_REG_SIZE * 3 + stack_depth_roundup; | |
15861 | new_prog = inline_bpf_loop(env, | |
15862 | i + delta, | |
15863 | -(stack_depth + stack_depth_extra), | |
15864 | inline_state->callback_subprogno, | |
15865 | &cnt); | |
15866 | if (!new_prog) | |
15867 | return -ENOMEM; | |
15868 | ||
15869 | delta += cnt - 1; | |
15870 | env->prog = new_prog; | |
15871 | insn = new_prog->insnsi + i + delta; | |
15872 | } | |
15873 | ||
15874 | if (subprogs[cur_subprog + 1].start == i + delta + 1) { | |
15875 | subprogs[cur_subprog].stack_depth += stack_depth_extra; | |
15876 | cur_subprog++; | |
15877 | stack_depth = subprogs[cur_subprog].stack_depth; | |
15878 | stack_depth_roundup = round_up(stack_depth, 8) - stack_depth; | |
15879 | stack_depth_extra = 0; | |
15880 | } | |
15881 | } | |
15882 | ||
15883 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
15884 | ||
15885 | return 0; | |
15886 | } | |
15887 | ||
58e2af8b | 15888 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 15889 | { |
58e2af8b | 15890 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
15891 | int i; |
15892 | ||
9f4686c4 AS |
15893 | sl = env->free_list; |
15894 | while (sl) { | |
15895 | sln = sl->next; | |
15896 | free_verifier_state(&sl->state, false); | |
15897 | kfree(sl); | |
15898 | sl = sln; | |
15899 | } | |
51c39bb1 | 15900 | env->free_list = NULL; |
9f4686c4 | 15901 | |
f1bca824 AS |
15902 | if (!env->explored_states) |
15903 | return; | |
15904 | ||
dc2a4ebc | 15905 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
15906 | sl = env->explored_states[i]; |
15907 | ||
a8f500af AS |
15908 | while (sl) { |
15909 | sln = sl->next; | |
15910 | free_verifier_state(&sl->state, false); | |
15911 | kfree(sl); | |
15912 | sl = sln; | |
15913 | } | |
51c39bb1 | 15914 | env->explored_states[i] = NULL; |
f1bca824 | 15915 | } |
51c39bb1 | 15916 | } |
f1bca824 | 15917 | |
51c39bb1 AS |
15918 | static int do_check_common(struct bpf_verifier_env *env, int subprog) |
15919 | { | |
6f8a57cc | 15920 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
15921 | struct bpf_verifier_state *state; |
15922 | struct bpf_reg_state *regs; | |
15923 | int ret, i; | |
15924 | ||
15925 | env->prev_linfo = NULL; | |
15926 | env->pass_cnt++; | |
15927 | ||
15928 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
15929 | if (!state) | |
15930 | return -ENOMEM; | |
15931 | state->curframe = 0; | |
15932 | state->speculative = false; | |
15933 | state->branches = 1; | |
15934 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
15935 | if (!state->frame[0]) { | |
15936 | kfree(state); | |
15937 | return -ENOMEM; | |
15938 | } | |
15939 | env->cur_state = state; | |
15940 | init_func_state(env, state->frame[0], | |
15941 | BPF_MAIN_FUNC /* callsite */, | |
15942 | 0 /* frameno */, | |
15943 | subprog); | |
be2ef816 AN |
15944 | state->first_insn_idx = env->subprog_info[subprog].start; |
15945 | state->last_insn_idx = -1; | |
51c39bb1 AS |
15946 | |
15947 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 15948 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
51c39bb1 AS |
15949 | ret = btf_prepare_func_args(env, subprog, regs); |
15950 | if (ret) | |
15951 | goto out; | |
15952 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
15953 | if (regs[i].type == PTR_TO_CTX) | |
15954 | mark_reg_known_zero(env, regs, i); | |
15955 | else if (regs[i].type == SCALAR_VALUE) | |
15956 | mark_reg_unknown(env, regs, i); | |
cf9f2f8d | 15957 | else if (base_type(regs[i].type) == PTR_TO_MEM) { |
e5069b9c DB |
15958 | const u32 mem_size = regs[i].mem_size; |
15959 | ||
15960 | mark_reg_known_zero(env, regs, i); | |
15961 | regs[i].mem_size = mem_size; | |
15962 | regs[i].id = ++env->id_gen; | |
15963 | } | |
51c39bb1 AS |
15964 | } |
15965 | } else { | |
15966 | /* 1st arg to a function */ | |
15967 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
15968 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
34747c41 | 15969 | ret = btf_check_subprog_arg_match(env, subprog, regs); |
51c39bb1 AS |
15970 | if (ret == -EFAULT) |
15971 | /* unlikely verifier bug. abort. | |
15972 | * ret == 0 and ret < 0 are sadly acceptable for | |
15973 | * main() function due to backward compatibility. | |
15974 | * Like socket filter program may be written as: | |
15975 | * int bpf_prog(struct pt_regs *ctx) | |
15976 | * and never dereference that ctx in the program. | |
15977 | * 'struct pt_regs' is a type mismatch for socket | |
15978 | * filter that should be using 'struct __sk_buff'. | |
15979 | */ | |
15980 | goto out; | |
15981 | } | |
15982 | ||
15983 | ret = do_check(env); | |
15984 | out: | |
f59bbfc2 AS |
15985 | /* check for NULL is necessary, since cur_state can be freed inside |
15986 | * do_check() under memory pressure. | |
15987 | */ | |
15988 | if (env->cur_state) { | |
15989 | free_verifier_state(env->cur_state, true); | |
15990 | env->cur_state = NULL; | |
15991 | } | |
6f8a57cc AN |
15992 | while (!pop_stack(env, NULL, NULL, false)); |
15993 | if (!ret && pop_log) | |
15994 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 | 15995 | free_states(env); |
51c39bb1 AS |
15996 | return ret; |
15997 | } | |
15998 | ||
15999 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
16000 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
16001 | * Consider: | |
16002 | * int bar(int); | |
16003 | * int foo(int f) | |
16004 | * { | |
16005 | * return bar(f); | |
16006 | * } | |
16007 | * int bar(int b) | |
16008 | * { | |
16009 | * ... | |
16010 | * } | |
16011 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
16012 | * will be assumed that bar() already verified successfully and call to bar() | |
16013 | * from foo() will be checked for type match only. Later bar() will be verified | |
16014 | * independently to check that it's safe for R1=any_scalar_value. | |
16015 | */ | |
16016 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
16017 | { | |
16018 | struct bpf_prog_aux *aux = env->prog->aux; | |
16019 | int i, ret; | |
16020 | ||
16021 | if (!aux->func_info) | |
16022 | return 0; | |
16023 | ||
16024 | for (i = 1; i < env->subprog_cnt; i++) { | |
16025 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
16026 | continue; | |
16027 | env->insn_idx = env->subprog_info[i].start; | |
16028 | WARN_ON_ONCE(env->insn_idx == 0); | |
16029 | ret = do_check_common(env, i); | |
16030 | if (ret) { | |
16031 | return ret; | |
16032 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
16033 | verbose(env, | |
16034 | "Func#%d is safe for any args that match its prototype\n", | |
16035 | i); | |
16036 | } | |
16037 | } | |
16038 | return 0; | |
16039 | } | |
16040 | ||
16041 | static int do_check_main(struct bpf_verifier_env *env) | |
16042 | { | |
16043 | int ret; | |
16044 | ||
16045 | env->insn_idx = 0; | |
16046 | ret = do_check_common(env, 0); | |
16047 | if (!ret) | |
16048 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
16049 | return ret; | |
16050 | } | |
16051 | ||
16052 | ||
06ee7115 AS |
16053 | static void print_verification_stats(struct bpf_verifier_env *env) |
16054 | { | |
16055 | int i; | |
16056 | ||
16057 | if (env->log.level & BPF_LOG_STATS) { | |
16058 | verbose(env, "verification time %lld usec\n", | |
16059 | div_u64(env->verification_time, 1000)); | |
16060 | verbose(env, "stack depth "); | |
16061 | for (i = 0; i < env->subprog_cnt; i++) { | |
16062 | u32 depth = env->subprog_info[i].stack_depth; | |
16063 | ||
16064 | verbose(env, "%d", depth); | |
16065 | if (i + 1 < env->subprog_cnt) | |
16066 | verbose(env, "+"); | |
16067 | } | |
16068 | verbose(env, "\n"); | |
16069 | } | |
16070 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
16071 | "total_states %d peak_states %d mark_read %d\n", | |
16072 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
16073 | env->max_states_per_insn, env->total_states, | |
16074 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
16075 | } |
16076 | ||
27ae7997 MKL |
16077 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
16078 | { | |
16079 | const struct btf_type *t, *func_proto; | |
16080 | const struct bpf_struct_ops *st_ops; | |
16081 | const struct btf_member *member; | |
16082 | struct bpf_prog *prog = env->prog; | |
16083 | u32 btf_id, member_idx; | |
16084 | const char *mname; | |
16085 | ||
12aa8a94 THJ |
16086 | if (!prog->gpl_compatible) { |
16087 | verbose(env, "struct ops programs must have a GPL compatible license\n"); | |
16088 | return -EINVAL; | |
16089 | } | |
16090 | ||
27ae7997 MKL |
16091 | btf_id = prog->aux->attach_btf_id; |
16092 | st_ops = bpf_struct_ops_find(btf_id); | |
16093 | if (!st_ops) { | |
16094 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
16095 | btf_id); | |
16096 | return -ENOTSUPP; | |
16097 | } | |
16098 | ||
16099 | t = st_ops->type; | |
16100 | member_idx = prog->expected_attach_type; | |
16101 | if (member_idx >= btf_type_vlen(t)) { | |
16102 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
16103 | member_idx, st_ops->name); | |
16104 | return -EINVAL; | |
16105 | } | |
16106 | ||
16107 | member = &btf_type_member(t)[member_idx]; | |
16108 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
16109 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
16110 | NULL); | |
16111 | if (!func_proto) { | |
16112 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
16113 | mname, member_idx, st_ops->name); | |
16114 | return -EINVAL; | |
16115 | } | |
16116 | ||
16117 | if (st_ops->check_member) { | |
16118 | int err = st_ops->check_member(t, member); | |
16119 | ||
16120 | if (err) { | |
16121 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
16122 | mname, st_ops->name); | |
16123 | return err; | |
16124 | } | |
16125 | } | |
16126 | ||
16127 | prog->aux->attach_func_proto = func_proto; | |
16128 | prog->aux->attach_func_name = mname; | |
16129 | env->ops = st_ops->verifier_ops; | |
16130 | ||
16131 | return 0; | |
16132 | } | |
6ba43b76 KS |
16133 | #define SECURITY_PREFIX "security_" |
16134 | ||
f7b12b6f | 16135 | static int check_attach_modify_return(unsigned long addr, const char *func_name) |
6ba43b76 | 16136 | { |
69191754 | 16137 | if (within_error_injection_list(addr) || |
f7b12b6f | 16138 | !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1)) |
6ba43b76 | 16139 | return 0; |
6ba43b76 | 16140 | |
6ba43b76 KS |
16141 | return -EINVAL; |
16142 | } | |
27ae7997 | 16143 | |
1e6c62a8 AS |
16144 | /* list of non-sleepable functions that are otherwise on |
16145 | * ALLOW_ERROR_INJECTION list | |
16146 | */ | |
16147 | BTF_SET_START(btf_non_sleepable_error_inject) | |
16148 | /* Three functions below can be called from sleepable and non-sleepable context. | |
16149 | * Assume non-sleepable from bpf safety point of view. | |
16150 | */ | |
9dd3d069 | 16151 | BTF_ID(func, __filemap_add_folio) |
1e6c62a8 AS |
16152 | BTF_ID(func, should_fail_alloc_page) |
16153 | BTF_ID(func, should_failslab) | |
16154 | BTF_SET_END(btf_non_sleepable_error_inject) | |
16155 | ||
16156 | static int check_non_sleepable_error_inject(u32 btf_id) | |
16157 | { | |
16158 | return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id); | |
16159 | } | |
16160 | ||
f7b12b6f THJ |
16161 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
16162 | const struct bpf_prog *prog, | |
16163 | const struct bpf_prog *tgt_prog, | |
16164 | u32 btf_id, | |
16165 | struct bpf_attach_target_info *tgt_info) | |
38207291 | 16166 | { |
be8704ff | 16167 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
f1b9509c | 16168 | const char prefix[] = "btf_trace_"; |
5b92a28a | 16169 | int ret = 0, subprog = -1, i; |
38207291 | 16170 | const struct btf_type *t; |
5b92a28a | 16171 | bool conservative = true; |
38207291 | 16172 | const char *tname; |
5b92a28a | 16173 | struct btf *btf; |
f7b12b6f | 16174 | long addr = 0; |
38207291 | 16175 | |
f1b9509c | 16176 | if (!btf_id) { |
efc68158 | 16177 | bpf_log(log, "Tracing programs must provide btf_id\n"); |
f1b9509c AS |
16178 | return -EINVAL; |
16179 | } | |
22dc4a0f | 16180 | btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf; |
5b92a28a | 16181 | if (!btf) { |
efc68158 | 16182 | bpf_log(log, |
5b92a28a AS |
16183 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); |
16184 | return -EINVAL; | |
16185 | } | |
16186 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c | 16187 | if (!t) { |
efc68158 | 16188 | bpf_log(log, "attach_btf_id %u is invalid\n", btf_id); |
f1b9509c AS |
16189 | return -EINVAL; |
16190 | } | |
5b92a28a | 16191 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c | 16192 | if (!tname) { |
efc68158 | 16193 | bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id); |
f1b9509c AS |
16194 | return -EINVAL; |
16195 | } | |
5b92a28a AS |
16196 | if (tgt_prog) { |
16197 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
16198 | ||
16199 | for (i = 0; i < aux->func_info_cnt; i++) | |
16200 | if (aux->func_info[i].type_id == btf_id) { | |
16201 | subprog = i; | |
16202 | break; | |
16203 | } | |
16204 | if (subprog == -1) { | |
efc68158 | 16205 | bpf_log(log, "Subprog %s doesn't exist\n", tname); |
5b92a28a AS |
16206 | return -EINVAL; |
16207 | } | |
16208 | conservative = aux->func_info_aux[subprog].unreliable; | |
be8704ff AS |
16209 | if (prog_extension) { |
16210 | if (conservative) { | |
efc68158 | 16211 | bpf_log(log, |
be8704ff AS |
16212 | "Cannot replace static functions\n"); |
16213 | return -EINVAL; | |
16214 | } | |
16215 | if (!prog->jit_requested) { | |
efc68158 | 16216 | bpf_log(log, |
be8704ff AS |
16217 | "Extension programs should be JITed\n"); |
16218 | return -EINVAL; | |
16219 | } | |
be8704ff AS |
16220 | } |
16221 | if (!tgt_prog->jited) { | |
efc68158 | 16222 | bpf_log(log, "Can attach to only JITed progs\n"); |
be8704ff AS |
16223 | return -EINVAL; |
16224 | } | |
16225 | if (tgt_prog->type == prog->type) { | |
16226 | /* Cannot fentry/fexit another fentry/fexit program. | |
16227 | * Cannot attach program extension to another extension. | |
16228 | * It's ok to attach fentry/fexit to extension program. | |
16229 | */ | |
efc68158 | 16230 | bpf_log(log, "Cannot recursively attach\n"); |
be8704ff AS |
16231 | return -EINVAL; |
16232 | } | |
16233 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
16234 | prog_extension && | |
16235 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
16236 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
16237 | /* Program extensions can extend all program types | |
16238 | * except fentry/fexit. The reason is the following. | |
16239 | * The fentry/fexit programs are used for performance | |
16240 | * analysis, stats and can be attached to any program | |
16241 | * type except themselves. When extension program is | |
16242 | * replacing XDP function it is necessary to allow | |
16243 | * performance analysis of all functions. Both original | |
16244 | * XDP program and its program extension. Hence | |
16245 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
16246 | * allowed. If extending of fentry/fexit was allowed it | |
16247 | * would be possible to create long call chain | |
16248 | * fentry->extension->fentry->extension beyond | |
16249 | * reasonable stack size. Hence extending fentry is not | |
16250 | * allowed. | |
16251 | */ | |
efc68158 | 16252 | bpf_log(log, "Cannot extend fentry/fexit\n"); |
be8704ff AS |
16253 | return -EINVAL; |
16254 | } | |
5b92a28a | 16255 | } else { |
be8704ff | 16256 | if (prog_extension) { |
efc68158 | 16257 | bpf_log(log, "Cannot replace kernel functions\n"); |
be8704ff AS |
16258 | return -EINVAL; |
16259 | } | |
5b92a28a | 16260 | } |
f1b9509c AS |
16261 | |
16262 | switch (prog->expected_attach_type) { | |
16263 | case BPF_TRACE_RAW_TP: | |
5b92a28a | 16264 | if (tgt_prog) { |
efc68158 | 16265 | bpf_log(log, |
5b92a28a AS |
16266 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); |
16267 | return -EINVAL; | |
16268 | } | |
38207291 | 16269 | if (!btf_type_is_typedef(t)) { |
efc68158 | 16270 | bpf_log(log, "attach_btf_id %u is not a typedef\n", |
38207291 MKL |
16271 | btf_id); |
16272 | return -EINVAL; | |
16273 | } | |
f1b9509c | 16274 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
efc68158 | 16275 | bpf_log(log, "attach_btf_id %u points to wrong type name %s\n", |
38207291 MKL |
16276 | btf_id, tname); |
16277 | return -EINVAL; | |
16278 | } | |
16279 | tname += sizeof(prefix) - 1; | |
5b92a28a | 16280 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
16281 | if (!btf_type_is_ptr(t)) |
16282 | /* should never happen in valid vmlinux build */ | |
16283 | return -EINVAL; | |
5b92a28a | 16284 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
16285 | if (!btf_type_is_func_proto(t)) |
16286 | /* should never happen in valid vmlinux build */ | |
16287 | return -EINVAL; | |
16288 | ||
f7b12b6f | 16289 | break; |
15d83c4d YS |
16290 | case BPF_TRACE_ITER: |
16291 | if (!btf_type_is_func(t)) { | |
efc68158 | 16292 | bpf_log(log, "attach_btf_id %u is not a function\n", |
15d83c4d YS |
16293 | btf_id); |
16294 | return -EINVAL; | |
16295 | } | |
16296 | t = btf_type_by_id(btf, t->type); | |
16297 | if (!btf_type_is_func_proto(t)) | |
16298 | return -EINVAL; | |
f7b12b6f THJ |
16299 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); |
16300 | if (ret) | |
16301 | return ret; | |
16302 | break; | |
be8704ff AS |
16303 | default: |
16304 | if (!prog_extension) | |
16305 | return -EINVAL; | |
df561f66 | 16306 | fallthrough; |
ae240823 | 16307 | case BPF_MODIFY_RETURN: |
9e4e01df | 16308 | case BPF_LSM_MAC: |
69fd337a | 16309 | case BPF_LSM_CGROUP: |
fec56f58 AS |
16310 | case BPF_TRACE_FENTRY: |
16311 | case BPF_TRACE_FEXIT: | |
16312 | if (!btf_type_is_func(t)) { | |
efc68158 | 16313 | bpf_log(log, "attach_btf_id %u is not a function\n", |
fec56f58 AS |
16314 | btf_id); |
16315 | return -EINVAL; | |
16316 | } | |
be8704ff | 16317 | if (prog_extension && |
efc68158 | 16318 | btf_check_type_match(log, prog, btf, t)) |
be8704ff | 16319 | return -EINVAL; |
5b92a28a | 16320 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
16321 | if (!btf_type_is_func_proto(t)) |
16322 | return -EINVAL; | |
f7b12b6f | 16323 | |
4a1e7c0c THJ |
16324 | if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) && |
16325 | (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type || | |
16326 | prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type)) | |
16327 | return -EINVAL; | |
16328 | ||
f7b12b6f | 16329 | if (tgt_prog && conservative) |
5b92a28a | 16330 | t = NULL; |
f7b12b6f THJ |
16331 | |
16332 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); | |
fec56f58 | 16333 | if (ret < 0) |
f7b12b6f THJ |
16334 | return ret; |
16335 | ||
5b92a28a | 16336 | if (tgt_prog) { |
e9eeec58 YS |
16337 | if (subprog == 0) |
16338 | addr = (long) tgt_prog->bpf_func; | |
16339 | else | |
16340 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a AS |
16341 | } else { |
16342 | addr = kallsyms_lookup_name(tname); | |
16343 | if (!addr) { | |
efc68158 | 16344 | bpf_log(log, |
5b92a28a AS |
16345 | "The address of function %s cannot be found\n", |
16346 | tname); | |
f7b12b6f | 16347 | return -ENOENT; |
5b92a28a | 16348 | } |
fec56f58 | 16349 | } |
18644cec | 16350 | |
1e6c62a8 AS |
16351 | if (prog->aux->sleepable) { |
16352 | ret = -EINVAL; | |
16353 | switch (prog->type) { | |
16354 | case BPF_PROG_TYPE_TRACING: | |
16355 | /* fentry/fexit/fmod_ret progs can be sleepable only if they are | |
16356 | * attached to ALLOW_ERROR_INJECTION and are not in denylist. | |
16357 | */ | |
16358 | if (!check_non_sleepable_error_inject(btf_id) && | |
16359 | within_error_injection_list(addr)) | |
16360 | ret = 0; | |
16361 | break; | |
16362 | case BPF_PROG_TYPE_LSM: | |
16363 | /* LSM progs check that they are attached to bpf_lsm_*() funcs. | |
16364 | * Only some of them are sleepable. | |
16365 | */ | |
423f1610 | 16366 | if (bpf_lsm_is_sleepable_hook(btf_id)) |
1e6c62a8 AS |
16367 | ret = 0; |
16368 | break; | |
16369 | default: | |
16370 | break; | |
16371 | } | |
f7b12b6f THJ |
16372 | if (ret) { |
16373 | bpf_log(log, "%s is not sleepable\n", tname); | |
16374 | return ret; | |
16375 | } | |
1e6c62a8 | 16376 | } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) { |
1af9270e | 16377 | if (tgt_prog) { |
efc68158 | 16378 | bpf_log(log, "can't modify return codes of BPF programs\n"); |
f7b12b6f THJ |
16379 | return -EINVAL; |
16380 | } | |
16381 | ret = check_attach_modify_return(addr, tname); | |
16382 | if (ret) { | |
16383 | bpf_log(log, "%s() is not modifiable\n", tname); | |
16384 | return ret; | |
1af9270e | 16385 | } |
18644cec | 16386 | } |
f7b12b6f THJ |
16387 | |
16388 | break; | |
16389 | } | |
16390 | tgt_info->tgt_addr = addr; | |
16391 | tgt_info->tgt_name = tname; | |
16392 | tgt_info->tgt_type = t; | |
16393 | return 0; | |
16394 | } | |
16395 | ||
35e3815f JO |
16396 | BTF_SET_START(btf_id_deny) |
16397 | BTF_ID_UNUSED | |
16398 | #ifdef CONFIG_SMP | |
16399 | BTF_ID(func, migrate_disable) | |
16400 | BTF_ID(func, migrate_enable) | |
16401 | #endif | |
16402 | #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU | |
16403 | BTF_ID(func, rcu_read_unlock_strict) | |
16404 | #endif | |
16405 | BTF_SET_END(btf_id_deny) | |
16406 | ||
f7b12b6f THJ |
16407 | static int check_attach_btf_id(struct bpf_verifier_env *env) |
16408 | { | |
16409 | struct bpf_prog *prog = env->prog; | |
3aac1ead | 16410 | struct bpf_prog *tgt_prog = prog->aux->dst_prog; |
f7b12b6f THJ |
16411 | struct bpf_attach_target_info tgt_info = {}; |
16412 | u32 btf_id = prog->aux->attach_btf_id; | |
16413 | struct bpf_trampoline *tr; | |
16414 | int ret; | |
16415 | u64 key; | |
16416 | ||
79a7f8bd AS |
16417 | if (prog->type == BPF_PROG_TYPE_SYSCALL) { |
16418 | if (prog->aux->sleepable) | |
16419 | /* attach_btf_id checked to be zero already */ | |
16420 | return 0; | |
16421 | verbose(env, "Syscall programs can only be sleepable\n"); | |
16422 | return -EINVAL; | |
16423 | } | |
16424 | ||
f7b12b6f | 16425 | if (prog->aux->sleepable && prog->type != BPF_PROG_TYPE_TRACING && |
64ad7556 DK |
16426 | prog->type != BPF_PROG_TYPE_LSM && prog->type != BPF_PROG_TYPE_KPROBE) { |
16427 | verbose(env, "Only fentry/fexit/fmod_ret, lsm, and kprobe/uprobe programs can be sleepable\n"); | |
f7b12b6f THJ |
16428 | return -EINVAL; |
16429 | } | |
16430 | ||
16431 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) | |
16432 | return check_struct_ops_btf_id(env); | |
16433 | ||
16434 | if (prog->type != BPF_PROG_TYPE_TRACING && | |
16435 | prog->type != BPF_PROG_TYPE_LSM && | |
16436 | prog->type != BPF_PROG_TYPE_EXT) | |
16437 | return 0; | |
16438 | ||
16439 | ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info); | |
16440 | if (ret) | |
fec56f58 | 16441 | return ret; |
f7b12b6f THJ |
16442 | |
16443 | if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) { | |
3aac1ead THJ |
16444 | /* to make freplace equivalent to their targets, they need to |
16445 | * inherit env->ops and expected_attach_type for the rest of the | |
16446 | * verification | |
16447 | */ | |
f7b12b6f THJ |
16448 | env->ops = bpf_verifier_ops[tgt_prog->type]; |
16449 | prog->expected_attach_type = tgt_prog->expected_attach_type; | |
16450 | } | |
16451 | ||
16452 | /* store info about the attachment target that will be used later */ | |
16453 | prog->aux->attach_func_proto = tgt_info.tgt_type; | |
16454 | prog->aux->attach_func_name = tgt_info.tgt_name; | |
16455 | ||
4a1e7c0c THJ |
16456 | if (tgt_prog) { |
16457 | prog->aux->saved_dst_prog_type = tgt_prog->type; | |
16458 | prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type; | |
16459 | } | |
16460 | ||
f7b12b6f THJ |
16461 | if (prog->expected_attach_type == BPF_TRACE_RAW_TP) { |
16462 | prog->aux->attach_btf_trace = true; | |
16463 | return 0; | |
16464 | } else if (prog->expected_attach_type == BPF_TRACE_ITER) { | |
16465 | if (!bpf_iter_prog_supported(prog)) | |
16466 | return -EINVAL; | |
16467 | return 0; | |
16468 | } | |
16469 | ||
16470 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
16471 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
16472 | if (ret < 0) | |
16473 | return ret; | |
35e3815f JO |
16474 | } else if (prog->type == BPF_PROG_TYPE_TRACING && |
16475 | btf_id_set_contains(&btf_id_deny, btf_id)) { | |
16476 | return -EINVAL; | |
38207291 | 16477 | } |
f7b12b6f | 16478 | |
22dc4a0f | 16479 | key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id); |
f7b12b6f THJ |
16480 | tr = bpf_trampoline_get(key, &tgt_info); |
16481 | if (!tr) | |
16482 | return -ENOMEM; | |
16483 | ||
3aac1ead | 16484 | prog->aux->dst_trampoline = tr; |
f7b12b6f | 16485 | return 0; |
38207291 MKL |
16486 | } |
16487 | ||
76654e67 AM |
16488 | struct btf *bpf_get_btf_vmlinux(void) |
16489 | { | |
16490 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { | |
16491 | mutex_lock(&bpf_verifier_lock); | |
16492 | if (!btf_vmlinux) | |
16493 | btf_vmlinux = btf_parse_vmlinux(); | |
16494 | mutex_unlock(&bpf_verifier_lock); | |
16495 | } | |
16496 | return btf_vmlinux; | |
16497 | } | |
16498 | ||
af2ac3e1 | 16499 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr) |
51580e79 | 16500 | { |
06ee7115 | 16501 | u64 start_time = ktime_get_ns(); |
58e2af8b | 16502 | struct bpf_verifier_env *env; |
b9193c1b | 16503 | struct bpf_verifier_log *log; |
9e4c24e7 | 16504 | int i, len, ret = -EINVAL; |
e2ae4ca2 | 16505 | bool is_priv; |
51580e79 | 16506 | |
eba0c929 AB |
16507 | /* no program is valid */ |
16508 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
16509 | return -EINVAL; | |
16510 | ||
58e2af8b | 16511 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
16512 | * allocate/free it every time bpf_check() is called |
16513 | */ | |
58e2af8b | 16514 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
16515 | if (!env) |
16516 | return -ENOMEM; | |
61bd5218 | 16517 | log = &env->log; |
cbd35700 | 16518 | |
9e4c24e7 | 16519 | len = (*prog)->len; |
fad953ce | 16520 | env->insn_aux_data = |
9e4c24e7 | 16521 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
16522 | ret = -ENOMEM; |
16523 | if (!env->insn_aux_data) | |
16524 | goto err_free_env; | |
9e4c24e7 JK |
16525 | for (i = 0; i < len; i++) |
16526 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 16527 | env->prog = *prog; |
00176a34 | 16528 | env->ops = bpf_verifier_ops[env->prog->type]; |
387544bf | 16529 | env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel); |
2c78ee89 | 16530 | is_priv = bpf_capable(); |
0246e64d | 16531 | |
76654e67 | 16532 | bpf_get_btf_vmlinux(); |
8580ac94 | 16533 | |
cbd35700 | 16534 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
16535 | if (!is_priv) |
16536 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 AS |
16537 | |
16538 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
16539 | /* user requested verbose verifier output | |
16540 | * and supplied buffer to store the verification trace | |
16541 | */ | |
e7bf8249 JK |
16542 | log->level = attr->log_level; |
16543 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
16544 | log->len_total = attr->log_size; | |
cbd35700 | 16545 | |
e7bf8249 | 16546 | /* log attributes have to be sane */ |
866de407 HT |
16547 | if (!bpf_verifier_log_attr_valid(log)) { |
16548 | ret = -EINVAL; | |
3df126f3 | 16549 | goto err_unlock; |
866de407 | 16550 | } |
cbd35700 | 16551 | } |
1ad2f583 | 16552 | |
0f55f9ed CL |
16553 | mark_verifier_state_clean(env); |
16554 | ||
8580ac94 AS |
16555 | if (IS_ERR(btf_vmlinux)) { |
16556 | /* Either gcc or pahole or kernel are broken. */ | |
16557 | verbose(env, "in-kernel BTF is malformed\n"); | |
16558 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 16559 | goto skip_full_check; |
8580ac94 AS |
16560 | } |
16561 | ||
1ad2f583 DB |
16562 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
16563 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 16564 | env->strict_alignment = true; |
e9ee9efc DM |
16565 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
16566 | env->strict_alignment = false; | |
cbd35700 | 16567 | |
2c78ee89 | 16568 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
01f810ac | 16569 | env->allow_uninit_stack = bpf_allow_uninit_stack(); |
41c48f3a | 16570 | env->allow_ptr_to_map_access = bpf_allow_ptr_to_map_access(); |
2c78ee89 AS |
16571 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); |
16572 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
16573 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 16574 | |
10d274e8 AS |
16575 | if (is_priv) |
16576 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
16577 | ||
dc2a4ebc | 16578 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 16579 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
16580 | GFP_USER); |
16581 | ret = -ENOMEM; | |
16582 | if (!env->explored_states) | |
16583 | goto skip_full_check; | |
16584 | ||
e6ac2450 MKL |
16585 | ret = add_subprog_and_kfunc(env); |
16586 | if (ret < 0) | |
16587 | goto skip_full_check; | |
16588 | ||
d9762e84 | 16589 | ret = check_subprogs(env); |
475fb78f AS |
16590 | if (ret < 0) |
16591 | goto skip_full_check; | |
16592 | ||
c454a46b | 16593 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
16594 | if (ret < 0) |
16595 | goto skip_full_check; | |
16596 | ||
be8704ff AS |
16597 | ret = check_attach_btf_id(env); |
16598 | if (ret) | |
16599 | goto skip_full_check; | |
16600 | ||
4976b718 HL |
16601 | ret = resolve_pseudo_ldimm64(env); |
16602 | if (ret < 0) | |
16603 | goto skip_full_check; | |
16604 | ||
ceb11679 YZ |
16605 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
16606 | ret = bpf_prog_offload_verifier_prep(env->prog); | |
16607 | if (ret) | |
16608 | goto skip_full_check; | |
16609 | } | |
16610 | ||
d9762e84 MKL |
16611 | ret = check_cfg(env); |
16612 | if (ret < 0) | |
16613 | goto skip_full_check; | |
16614 | ||
51c39bb1 AS |
16615 | ret = do_check_subprogs(env); |
16616 | ret = ret ?: do_check_main(env); | |
cbd35700 | 16617 | |
c941ce9c QM |
16618 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
16619 | ret = bpf_prog_offload_finalize(env); | |
16620 | ||
0246e64d | 16621 | skip_full_check: |
51c39bb1 | 16622 | kvfree(env->explored_states); |
0246e64d | 16623 | |
c131187d | 16624 | if (ret == 0) |
9b38c405 | 16625 | ret = check_max_stack_depth(env); |
c131187d | 16626 | |
9b38c405 | 16627 | /* instruction rewrites happen after this point */ |
1ade2371 EZ |
16628 | if (ret == 0) |
16629 | ret = optimize_bpf_loop(env); | |
16630 | ||
e2ae4ca2 JK |
16631 | if (is_priv) { |
16632 | if (ret == 0) | |
16633 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
16634 | if (ret == 0) |
16635 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
16636 | if (ret == 0) |
16637 | ret = opt_remove_nops(env); | |
52875a04 JK |
16638 | } else { |
16639 | if (ret == 0) | |
16640 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
16641 | } |
16642 | ||
9bac3d6d AS |
16643 | if (ret == 0) |
16644 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
16645 | ret = convert_ctx_accesses(env); | |
16646 | ||
e245c5c6 | 16647 | if (ret == 0) |
e6ac5933 | 16648 | ret = do_misc_fixups(env); |
e245c5c6 | 16649 | |
a4b1d3c1 JW |
16650 | /* do 32-bit optimization after insn patching has done so those patched |
16651 | * insns could be handled correctly. | |
16652 | */ | |
d6c2308c JW |
16653 | if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) { |
16654 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); | |
16655 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
16656 | : false; | |
a4b1d3c1 JW |
16657 | } |
16658 | ||
1ea47e01 AS |
16659 | if (ret == 0) |
16660 | ret = fixup_call_args(env); | |
16661 | ||
06ee7115 AS |
16662 | env->verification_time = ktime_get_ns() - start_time; |
16663 | print_verification_stats(env); | |
aba64c7d | 16664 | env->prog->aux->verified_insns = env->insn_processed; |
06ee7115 | 16665 | |
a2a7d570 | 16666 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 16667 | ret = -ENOSPC; |
a2a7d570 | 16668 | if (log->level && !log->ubuf) { |
cbd35700 | 16669 | ret = -EFAULT; |
a2a7d570 | 16670 | goto err_release_maps; |
cbd35700 AS |
16671 | } |
16672 | ||
541c3bad AN |
16673 | if (ret) |
16674 | goto err_release_maps; | |
16675 | ||
16676 | if (env->used_map_cnt) { | |
0246e64d | 16677 | /* if program passed verifier, update used_maps in bpf_prog_info */ |
9bac3d6d AS |
16678 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
16679 | sizeof(env->used_maps[0]), | |
16680 | GFP_KERNEL); | |
0246e64d | 16681 | |
9bac3d6d | 16682 | if (!env->prog->aux->used_maps) { |
0246e64d | 16683 | ret = -ENOMEM; |
a2a7d570 | 16684 | goto err_release_maps; |
0246e64d AS |
16685 | } |
16686 | ||
9bac3d6d | 16687 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 16688 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 16689 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
541c3bad AN |
16690 | } |
16691 | if (env->used_btf_cnt) { | |
16692 | /* if program passed verifier, update used_btfs in bpf_prog_aux */ | |
16693 | env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt, | |
16694 | sizeof(env->used_btfs[0]), | |
16695 | GFP_KERNEL); | |
16696 | if (!env->prog->aux->used_btfs) { | |
16697 | ret = -ENOMEM; | |
16698 | goto err_release_maps; | |
16699 | } | |
0246e64d | 16700 | |
541c3bad AN |
16701 | memcpy(env->prog->aux->used_btfs, env->used_btfs, |
16702 | sizeof(env->used_btfs[0]) * env->used_btf_cnt); | |
16703 | env->prog->aux->used_btf_cnt = env->used_btf_cnt; | |
16704 | } | |
16705 | if (env->used_map_cnt || env->used_btf_cnt) { | |
0246e64d AS |
16706 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic |
16707 | * bpf_ld_imm64 instructions | |
16708 | */ | |
16709 | convert_pseudo_ld_imm64(env); | |
16710 | } | |
cbd35700 | 16711 | |
541c3bad | 16712 | adjust_btf_func(env); |
ba64e7d8 | 16713 | |
a2a7d570 | 16714 | err_release_maps: |
9bac3d6d | 16715 | if (!env->prog->aux->used_maps) |
0246e64d | 16716 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 16717 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
16718 | */ |
16719 | release_maps(env); | |
541c3bad AN |
16720 | if (!env->prog->aux->used_btfs) |
16721 | release_btfs(env); | |
03f87c0b THJ |
16722 | |
16723 | /* extension progs temporarily inherit the attach_type of their targets | |
16724 | for verification purposes, so set it back to zero before returning | |
16725 | */ | |
16726 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
16727 | env->prog->expected_attach_type = 0; | |
16728 | ||
9bac3d6d | 16729 | *prog = env->prog; |
3df126f3 | 16730 | err_unlock: |
45a73c17 AS |
16731 | if (!is_priv) |
16732 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
16733 | vfree(env->insn_aux_data); |
16734 | err_free_env: | |
16735 | kfree(env); | |
51580e79 AS |
16736 | return ret; |
16737 | } |