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> |
51580e79 AS |
7 | #include <linux/kernel.h> |
8 | #include <linux/types.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/bpf.h> | |
838e9690 | 11 | #include <linux/btf.h> |
58e2af8b | 12 | #include <linux/bpf_verifier.h> |
51580e79 AS |
13 | #include <linux/filter.h> |
14 | #include <net/netlink.h> | |
15 | #include <linux/file.h> | |
16 | #include <linux/vmalloc.h> | |
ebb676da | 17 | #include <linux/stringify.h> |
cc8b0b92 AS |
18 | #include <linux/bsearch.h> |
19 | #include <linux/sort.h> | |
c195651e | 20 | #include <linux/perf_event.h> |
d9762e84 | 21 | #include <linux/ctype.h> |
6ba43b76 | 22 | #include <linux/error-injection.h> |
9e4e01df | 23 | #include <linux/bpf_lsm.h> |
1e6c62a8 | 24 | #include <linux/btf_ids.h> |
51580e79 | 25 | |
f4ac7e0b JK |
26 | #include "disasm.h" |
27 | ||
00176a34 | 28 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
91cc1a99 | 29 | #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ |
00176a34 JK |
30 | [_id] = & _name ## _verifier_ops, |
31 | #define BPF_MAP_TYPE(_id, _ops) | |
f2e10bff | 32 | #define BPF_LINK_TYPE(_id, _name) |
00176a34 JK |
33 | #include <linux/bpf_types.h> |
34 | #undef BPF_PROG_TYPE | |
35 | #undef BPF_MAP_TYPE | |
f2e10bff | 36 | #undef BPF_LINK_TYPE |
00176a34 JK |
37 | }; |
38 | ||
51580e79 AS |
39 | /* bpf_check() is a static code analyzer that walks eBPF program |
40 | * instruction by instruction and updates register/stack state. | |
41 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
42 | * | |
43 | * The first pass is depth-first-search to check that the program is a DAG. | |
44 | * It rejects the following programs: | |
45 | * - larger than BPF_MAXINSNS insns | |
46 | * - if loop is present (detected via back-edge) | |
47 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
48 | * - out of bounds or malformed jumps | |
49 | * The second pass is all possible path descent from the 1st insn. | |
8fb33b60 | 50 | * Since it's analyzing all paths through the program, the length of the |
eba38a96 | 51 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
52 | * insn is less then 4K, but there are too many branches that change stack/regs. |
53 | * Number of 'branches to be analyzed' is limited to 1k | |
54 | * | |
55 | * On entry to each instruction, each register has a type, and the instruction | |
56 | * changes the types of the registers depending on instruction semantics. | |
57 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
58 | * copied to R1. | |
59 | * | |
60 | * All registers are 64-bit. | |
61 | * R0 - return register | |
62 | * R1-R5 argument passing registers | |
63 | * R6-R9 callee saved registers | |
64 | * R10 - frame pointer read-only | |
65 | * | |
66 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
67 | * and has type PTR_TO_CTX. | |
68 | * | |
69 | * Verifier tracks arithmetic operations on pointers in case: | |
70 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
71 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
72 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
73 | * and 2nd arithmetic instruction is pattern matched to recognize | |
74 | * that it wants to construct a pointer to some element within stack. | |
75 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
76 | * (and -20 constant is saved for further stack bounds checking). | |
77 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
78 | * | |
f1174f77 | 79 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 80 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 81 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
82 | * |
83 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
84 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
85 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
86 | * |
87 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
88 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
89 | * | |
90 | * registers used to pass values to function calls are checked against | |
91 | * function argument constraints. | |
92 | * | |
93 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
94 | * It means that the register type passed to this function must be | |
95 | * PTR_TO_STACK and it will be used inside the function as | |
96 | * 'pointer to map element key' | |
97 | * | |
98 | * For example the argument constraints for bpf_map_lookup_elem(): | |
99 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
100 | * .arg1_type = ARG_CONST_MAP_PTR, | |
101 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
102 | * | |
103 | * ret_type says that this function returns 'pointer to map elem value or null' | |
104 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
105 | * 2nd argument should be a pointer to stack, which will be used inside | |
106 | * the helper function as a pointer to map element key. | |
107 | * | |
108 | * On the kernel side the helper function looks like: | |
109 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
110 | * { | |
111 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
112 | * void *key = (void *) (unsigned long) r2; | |
113 | * void *value; | |
114 | * | |
115 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
116 | * [key, key + map->key_size) bytes are valid and were initialized on | |
117 | * the stack of eBPF program. | |
118 | * } | |
119 | * | |
120 | * Corresponding eBPF program may look like: | |
121 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
122 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
123 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
124 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
125 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
126 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
127 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
128 | * | |
129 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
130 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
131 | * and were initialized prior to this call. | |
132 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
133 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
134 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
8fb33b60 | 135 | * returns either pointer to map value or NULL. |
51580e79 AS |
136 | * |
137 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
138 | * insn, the register holding that pointer in the true branch changes state to | |
139 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
140 | * branch. See check_cond_jmp_op(). | |
141 | * | |
142 | * After the call R0 is set to return type of the function and registers R1-R5 | |
143 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
144 | * |
145 | * The following reference types represent a potential reference to a kernel | |
146 | * resource which, after first being allocated, must be checked and freed by | |
147 | * the BPF program: | |
148 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
149 | * | |
150 | * When the verifier sees a helper call return a reference type, it allocates a | |
151 | * pointer id for the reference and stores it in the current function state. | |
152 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
153 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
154 | * passes through a NULL-check conditional. For the branch wherein the state is | |
155 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
156 | * |
157 | * For each helper function that allocates a reference, such as | |
158 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
159 | * bpf_sk_release(). When a reference type passes into the release function, | |
160 | * the verifier also releases the reference. If any unchecked or unreleased | |
161 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
162 | */ |
163 | ||
17a52670 | 164 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 165 | struct bpf_verifier_stack_elem { |
17a52670 AS |
166 | /* verifer state is 'st' |
167 | * before processing instruction 'insn_idx' | |
168 | * and after processing instruction 'prev_insn_idx' | |
169 | */ | |
58e2af8b | 170 | struct bpf_verifier_state st; |
17a52670 AS |
171 | int insn_idx; |
172 | int prev_insn_idx; | |
58e2af8b | 173 | struct bpf_verifier_stack_elem *next; |
6f8a57cc AN |
174 | /* length of verifier log at the time this state was pushed on stack */ |
175 | u32 log_pos; | |
cbd35700 AS |
176 | }; |
177 | ||
b285fcb7 | 178 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 179 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 180 | |
d2e4c1e6 DB |
181 | #define BPF_MAP_KEY_POISON (1ULL << 63) |
182 | #define BPF_MAP_KEY_SEEN (1ULL << 62) | |
183 | ||
c93552c4 DB |
184 | #define BPF_MAP_PTR_UNPRIV 1UL |
185 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
186 | POISON_POINTER_DELTA)) | |
187 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
188 | ||
189 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) | |
190 | { | |
d2e4c1e6 | 191 | return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON; |
c93552c4 DB |
192 | } |
193 | ||
194 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
195 | { | |
d2e4c1e6 | 196 | return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV; |
c93552c4 DB |
197 | } |
198 | ||
199 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
200 | const struct bpf_map *map, bool unpriv) | |
201 | { | |
202 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
203 | unpriv |= bpf_map_ptr_unpriv(aux); | |
d2e4c1e6 DB |
204 | aux->map_ptr_state = (unsigned long)map | |
205 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
206 | } | |
207 | ||
208 | static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux) | |
209 | { | |
210 | return aux->map_key_state & BPF_MAP_KEY_POISON; | |
211 | } | |
212 | ||
213 | static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux) | |
214 | { | |
215 | return !(aux->map_key_state & BPF_MAP_KEY_SEEN); | |
216 | } | |
217 | ||
218 | static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux) | |
219 | { | |
220 | return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON); | |
221 | } | |
222 | ||
223 | static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state) | |
224 | { | |
225 | bool poisoned = bpf_map_key_poisoned(aux); | |
226 | ||
227 | aux->map_key_state = state | BPF_MAP_KEY_SEEN | | |
228 | (poisoned ? BPF_MAP_KEY_POISON : 0ULL); | |
c93552c4 | 229 | } |
fad73a1a | 230 | |
23a2d70c YS |
231 | static bool bpf_pseudo_call(const struct bpf_insn *insn) |
232 | { | |
233 | return insn->code == (BPF_JMP | BPF_CALL) && | |
234 | insn->src_reg == BPF_PSEUDO_CALL; | |
235 | } | |
236 | ||
e6ac2450 MKL |
237 | static bool bpf_pseudo_kfunc_call(const struct bpf_insn *insn) |
238 | { | |
239 | return insn->code == (BPF_JMP | BPF_CALL) && | |
240 | insn->src_reg == BPF_PSEUDO_KFUNC_CALL; | |
241 | } | |
242 | ||
69c087ba YS |
243 | static bool bpf_pseudo_func(const struct bpf_insn *insn) |
244 | { | |
245 | return insn->code == (BPF_LD | BPF_IMM | BPF_DW) && | |
246 | insn->src_reg == BPF_PSEUDO_FUNC; | |
247 | } | |
248 | ||
33ff9823 DB |
249 | struct bpf_call_arg_meta { |
250 | struct bpf_map *map_ptr; | |
435faee1 | 251 | bool raw_mode; |
36bbef52 | 252 | bool pkt_access; |
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; |
d83525ca | 258 | int func_id; |
22dc4a0f | 259 | struct btf *btf; |
eaa6bcb7 | 260 | u32 btf_id; |
22dc4a0f | 261 | struct btf *ret_btf; |
eaa6bcb7 | 262 | u32 ret_btf_id; |
69c087ba | 263 | u32 subprogno; |
33ff9823 DB |
264 | }; |
265 | ||
8580ac94 AS |
266 | struct btf *btf_vmlinux; |
267 | ||
cbd35700 AS |
268 | static DEFINE_MUTEX(bpf_verifier_lock); |
269 | ||
d9762e84 MKL |
270 | static const struct bpf_line_info * |
271 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
272 | { | |
273 | const struct bpf_line_info *linfo; | |
274 | const struct bpf_prog *prog; | |
275 | u32 i, nr_linfo; | |
276 | ||
277 | prog = env->prog; | |
278 | nr_linfo = prog->aux->nr_linfo; | |
279 | ||
280 | if (!nr_linfo || insn_off >= prog->len) | |
281 | return NULL; | |
282 | ||
283 | linfo = prog->aux->linfo; | |
284 | for (i = 1; i < nr_linfo; i++) | |
285 | if (insn_off < linfo[i].insn_off) | |
286 | break; | |
287 | ||
288 | return &linfo[i - 1]; | |
289 | } | |
290 | ||
77d2e05a MKL |
291 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
292 | va_list args) | |
cbd35700 | 293 | { |
a2a7d570 | 294 | unsigned int n; |
cbd35700 | 295 | |
a2a7d570 | 296 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
297 | |
298 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
299 | "verifier log line truncated - local buffer too short\n"); | |
300 | ||
301 | n = min(log->len_total - log->len_used - 1, n); | |
302 | log->kbuf[n] = '\0'; | |
303 | ||
8580ac94 AS |
304 | if (log->level == BPF_LOG_KERNEL) { |
305 | pr_err("BPF:%s\n", log->kbuf); | |
306 | return; | |
307 | } | |
a2a7d570 JK |
308 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) |
309 | log->len_used += n; | |
310 | else | |
311 | log->ubuf = NULL; | |
cbd35700 | 312 | } |
abe08840 | 313 | |
6f8a57cc AN |
314 | static void bpf_vlog_reset(struct bpf_verifier_log *log, u32 new_pos) |
315 | { | |
316 | char zero = 0; | |
317 | ||
318 | if (!bpf_verifier_log_needed(log)) | |
319 | return; | |
320 | ||
321 | log->len_used = new_pos; | |
322 | if (put_user(zero, log->ubuf + new_pos)) | |
323 | log->ubuf = NULL; | |
324 | } | |
325 | ||
abe08840 JO |
326 | /* log_level controls verbosity level of eBPF verifier. |
327 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
328 | * so the user can figure out what's wrong with the program | |
430e68d1 | 329 | */ |
abe08840 JO |
330 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
331 | const char *fmt, ...) | |
332 | { | |
333 | va_list args; | |
334 | ||
77d2e05a MKL |
335 | if (!bpf_verifier_log_needed(&env->log)) |
336 | return; | |
337 | ||
abe08840 | 338 | va_start(args, fmt); |
77d2e05a | 339 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
340 | va_end(args); |
341 | } | |
342 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
343 | ||
344 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
345 | { | |
77d2e05a | 346 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
347 | va_list args; |
348 | ||
77d2e05a MKL |
349 | if (!bpf_verifier_log_needed(&env->log)) |
350 | return; | |
351 | ||
abe08840 | 352 | va_start(args, fmt); |
77d2e05a | 353 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
354 | va_end(args); |
355 | } | |
cbd35700 | 356 | |
9e15db66 AS |
357 | __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, |
358 | const char *fmt, ...) | |
359 | { | |
360 | va_list args; | |
361 | ||
362 | if (!bpf_verifier_log_needed(log)) | |
363 | return; | |
364 | ||
365 | va_start(args, fmt); | |
366 | bpf_verifier_vlog(log, fmt, args); | |
367 | va_end(args); | |
368 | } | |
369 | ||
d9762e84 MKL |
370 | static const char *ltrim(const char *s) |
371 | { | |
372 | while (isspace(*s)) | |
373 | s++; | |
374 | ||
375 | return s; | |
376 | } | |
377 | ||
378 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
379 | u32 insn_off, | |
380 | const char *prefix_fmt, ...) | |
381 | { | |
382 | const struct bpf_line_info *linfo; | |
383 | ||
384 | if (!bpf_verifier_log_needed(&env->log)) | |
385 | return; | |
386 | ||
387 | linfo = find_linfo(env, insn_off); | |
388 | if (!linfo || linfo == env->prev_linfo) | |
389 | return; | |
390 | ||
391 | if (prefix_fmt) { | |
392 | va_list args; | |
393 | ||
394 | va_start(args, prefix_fmt); | |
395 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
396 | va_end(args); | |
397 | } | |
398 | ||
399 | verbose(env, "%s\n", | |
400 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
401 | linfo->line_off))); | |
402 | ||
403 | env->prev_linfo = linfo; | |
404 | } | |
405 | ||
bc2591d6 YS |
406 | static void verbose_invalid_scalar(struct bpf_verifier_env *env, |
407 | struct bpf_reg_state *reg, | |
408 | struct tnum *range, const char *ctx, | |
409 | const char *reg_name) | |
410 | { | |
411 | char tn_buf[48]; | |
412 | ||
413 | verbose(env, "At %s the register %s ", ctx, reg_name); | |
414 | if (!tnum_is_unknown(reg->var_off)) { | |
415 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
416 | verbose(env, "has value %s", tn_buf); | |
417 | } else { | |
418 | verbose(env, "has unknown scalar value"); | |
419 | } | |
420 | tnum_strn(tn_buf, sizeof(tn_buf), *range); | |
421 | verbose(env, " should have been in %s\n", tn_buf); | |
422 | } | |
423 | ||
de8f3a83 DB |
424 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
425 | { | |
426 | return type == PTR_TO_PACKET || | |
427 | type == PTR_TO_PACKET_META; | |
428 | } | |
429 | ||
46f8bc92 MKL |
430 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
431 | { | |
432 | return type == PTR_TO_SOCKET || | |
655a51e5 | 433 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
434 | type == PTR_TO_TCP_SOCK || |
435 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
436 | } |
437 | ||
cac616db JF |
438 | static bool reg_type_not_null(enum bpf_reg_type type) |
439 | { | |
440 | return type == PTR_TO_SOCKET || | |
441 | type == PTR_TO_TCP_SOCK || | |
442 | type == PTR_TO_MAP_VALUE || | |
69c087ba | 443 | type == PTR_TO_MAP_KEY || |
01c66c48 | 444 | type == PTR_TO_SOCK_COMMON; |
cac616db JF |
445 | } |
446 | ||
840b9615 JS |
447 | static bool reg_type_may_be_null(enum bpf_reg_type type) |
448 | { | |
fd978bf7 | 449 | return type == PTR_TO_MAP_VALUE_OR_NULL || |
46f8bc92 | 450 | type == PTR_TO_SOCKET_OR_NULL || |
655a51e5 | 451 | type == PTR_TO_SOCK_COMMON_OR_NULL || |
b121b341 | 452 | type == PTR_TO_TCP_SOCK_OR_NULL || |
457f4436 | 453 | type == PTR_TO_BTF_ID_OR_NULL || |
afbf21dc YS |
454 | type == PTR_TO_MEM_OR_NULL || |
455 | type == PTR_TO_RDONLY_BUF_OR_NULL || | |
456 | type == PTR_TO_RDWR_BUF_OR_NULL; | |
fd978bf7 JS |
457 | } |
458 | ||
d83525ca AS |
459 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
460 | { | |
461 | return reg->type == PTR_TO_MAP_VALUE && | |
462 | map_value_has_spin_lock(reg->map_ptr); | |
463 | } | |
464 | ||
cba368c1 MKL |
465 | static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type) |
466 | { | |
467 | return type == PTR_TO_SOCKET || | |
468 | type == PTR_TO_SOCKET_OR_NULL || | |
469 | type == PTR_TO_TCP_SOCK || | |
457f4436 AN |
470 | type == PTR_TO_TCP_SOCK_OR_NULL || |
471 | type == PTR_TO_MEM || | |
472 | type == PTR_TO_MEM_OR_NULL; | |
cba368c1 MKL |
473 | } |
474 | ||
1b986589 | 475 | static bool arg_type_may_be_refcounted(enum bpf_arg_type type) |
fd978bf7 | 476 | { |
1b986589 | 477 | return type == ARG_PTR_TO_SOCK_COMMON; |
fd978bf7 JS |
478 | } |
479 | ||
fd1b0d60 LB |
480 | static bool arg_type_may_be_null(enum bpf_arg_type type) |
481 | { | |
482 | return type == ARG_PTR_TO_MAP_VALUE_OR_NULL || | |
483 | type == ARG_PTR_TO_MEM_OR_NULL || | |
484 | type == ARG_PTR_TO_CTX_OR_NULL || | |
485 | type == ARG_PTR_TO_SOCKET_OR_NULL || | |
69c087ba YS |
486 | type == ARG_PTR_TO_ALLOC_MEM_OR_NULL || |
487 | type == ARG_PTR_TO_STACK_OR_NULL; | |
fd1b0d60 LB |
488 | } |
489 | ||
fd978bf7 JS |
490 | /* Determine whether the function releases some resources allocated by another |
491 | * function call. The first reference type argument will be assumed to be | |
492 | * released by release_reference(). | |
493 | */ | |
494 | static bool is_release_function(enum bpf_func_id func_id) | |
495 | { | |
457f4436 AN |
496 | return func_id == BPF_FUNC_sk_release || |
497 | func_id == BPF_FUNC_ringbuf_submit || | |
498 | func_id == BPF_FUNC_ringbuf_discard; | |
840b9615 JS |
499 | } |
500 | ||
64d85290 | 501 | static bool may_be_acquire_function(enum bpf_func_id func_id) |
46f8bc92 MKL |
502 | { |
503 | return func_id == BPF_FUNC_sk_lookup_tcp || | |
edbf8c01 | 504 | func_id == BPF_FUNC_sk_lookup_udp || |
64d85290 | 505 | func_id == BPF_FUNC_skc_lookup_tcp || |
457f4436 AN |
506 | func_id == BPF_FUNC_map_lookup_elem || |
507 | func_id == BPF_FUNC_ringbuf_reserve; | |
64d85290 JS |
508 | } |
509 | ||
510 | static bool is_acquire_function(enum bpf_func_id func_id, | |
511 | const struct bpf_map *map) | |
512 | { | |
513 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
514 | ||
515 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
516 | func_id == BPF_FUNC_sk_lookup_udp || | |
457f4436 AN |
517 | func_id == BPF_FUNC_skc_lookup_tcp || |
518 | func_id == BPF_FUNC_ringbuf_reserve) | |
64d85290 JS |
519 | return true; |
520 | ||
521 | if (func_id == BPF_FUNC_map_lookup_elem && | |
522 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
523 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
524 | return true; | |
525 | ||
526 | return false; | |
46f8bc92 MKL |
527 | } |
528 | ||
1b986589 MKL |
529 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
530 | { | |
531 | return func_id == BPF_FUNC_tcp_sock || | |
1df8f55a MKL |
532 | func_id == BPF_FUNC_sk_fullsock || |
533 | func_id == BPF_FUNC_skc_to_tcp_sock || | |
534 | func_id == BPF_FUNC_skc_to_tcp6_sock || | |
535 | func_id == BPF_FUNC_skc_to_udp6_sock || | |
536 | func_id == BPF_FUNC_skc_to_tcp_timewait_sock || | |
537 | func_id == BPF_FUNC_skc_to_tcp_request_sock; | |
1b986589 MKL |
538 | } |
539 | ||
39491867 BJ |
540 | static bool is_cmpxchg_insn(const struct bpf_insn *insn) |
541 | { | |
542 | return BPF_CLASS(insn->code) == BPF_STX && | |
543 | BPF_MODE(insn->code) == BPF_ATOMIC && | |
544 | insn->imm == BPF_CMPXCHG; | |
545 | } | |
546 | ||
17a52670 AS |
547 | /* string representation of 'enum bpf_reg_type' */ |
548 | static const char * const reg_type_str[] = { | |
549 | [NOT_INIT] = "?", | |
f1174f77 | 550 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
551 | [PTR_TO_CTX] = "ctx", |
552 | [CONST_PTR_TO_MAP] = "map_ptr", | |
553 | [PTR_TO_MAP_VALUE] = "map_value", | |
554 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 555 | [PTR_TO_STACK] = "fp", |
969bf05e | 556 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 557 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 558 | [PTR_TO_PACKET_END] = "pkt_end", |
d58e468b | 559 | [PTR_TO_FLOW_KEYS] = "flow_keys", |
c64b7983 JS |
560 | [PTR_TO_SOCKET] = "sock", |
561 | [PTR_TO_SOCKET_OR_NULL] = "sock_or_null", | |
46f8bc92 MKL |
562 | [PTR_TO_SOCK_COMMON] = "sock_common", |
563 | [PTR_TO_SOCK_COMMON_OR_NULL] = "sock_common_or_null", | |
655a51e5 MKL |
564 | [PTR_TO_TCP_SOCK] = "tcp_sock", |
565 | [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", | |
9df1c28b | 566 | [PTR_TO_TP_BUFFER] = "tp_buffer", |
fada7fdc | 567 | [PTR_TO_XDP_SOCK] = "xdp_sock", |
9e15db66 | 568 | [PTR_TO_BTF_ID] = "ptr_", |
b121b341 | 569 | [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", |
eaa6bcb7 | 570 | [PTR_TO_PERCPU_BTF_ID] = "percpu_ptr_", |
457f4436 AN |
571 | [PTR_TO_MEM] = "mem", |
572 | [PTR_TO_MEM_OR_NULL] = "mem_or_null", | |
afbf21dc YS |
573 | [PTR_TO_RDONLY_BUF] = "rdonly_buf", |
574 | [PTR_TO_RDONLY_BUF_OR_NULL] = "rdonly_buf_or_null", | |
575 | [PTR_TO_RDWR_BUF] = "rdwr_buf", | |
576 | [PTR_TO_RDWR_BUF_OR_NULL] = "rdwr_buf_or_null", | |
69c087ba YS |
577 | [PTR_TO_FUNC] = "func", |
578 | [PTR_TO_MAP_KEY] = "map_key", | |
17a52670 AS |
579 | }; |
580 | ||
8efea21d EC |
581 | static char slot_type_char[] = { |
582 | [STACK_INVALID] = '?', | |
583 | [STACK_SPILL] = 'r', | |
584 | [STACK_MISC] = 'm', | |
585 | [STACK_ZERO] = '0', | |
586 | }; | |
587 | ||
4e92024a AS |
588 | static void print_liveness(struct bpf_verifier_env *env, |
589 | enum bpf_reg_liveness live) | |
590 | { | |
9242b5f5 | 591 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
592 | verbose(env, "_"); |
593 | if (live & REG_LIVE_READ) | |
594 | verbose(env, "r"); | |
595 | if (live & REG_LIVE_WRITTEN) | |
596 | verbose(env, "w"); | |
9242b5f5 AS |
597 | if (live & REG_LIVE_DONE) |
598 | verbose(env, "D"); | |
4e92024a AS |
599 | } |
600 | ||
f4d7e40a AS |
601 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
602 | const struct bpf_reg_state *reg) | |
603 | { | |
604 | struct bpf_verifier_state *cur = env->cur_state; | |
605 | ||
606 | return cur->frame[reg->frameno]; | |
607 | } | |
608 | ||
22dc4a0f | 609 | static const char *kernel_type_name(const struct btf* btf, u32 id) |
9e15db66 | 610 | { |
22dc4a0f | 611 | return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off); |
9e15db66 AS |
612 | } |
613 | ||
61bd5218 | 614 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 615 | const struct bpf_func_state *state) |
17a52670 | 616 | { |
f4d7e40a | 617 | const struct bpf_reg_state *reg; |
17a52670 AS |
618 | enum bpf_reg_type t; |
619 | int i; | |
620 | ||
f4d7e40a AS |
621 | if (state->frameno) |
622 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 623 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
624 | reg = &state->regs[i]; |
625 | t = reg->type; | |
17a52670 AS |
626 | if (t == NOT_INIT) |
627 | continue; | |
4e92024a AS |
628 | verbose(env, " R%d", i); |
629 | print_liveness(env, reg->live); | |
630 | verbose(env, "=%s", reg_type_str[t]); | |
b5dc0163 AS |
631 | if (t == SCALAR_VALUE && reg->precise) |
632 | verbose(env, "P"); | |
f1174f77 EC |
633 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
634 | tnum_is_const(reg->var_off)) { | |
635 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 636 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 637 | } else { |
eaa6bcb7 HL |
638 | if (t == PTR_TO_BTF_ID || |
639 | t == PTR_TO_BTF_ID_OR_NULL || | |
640 | t == PTR_TO_PERCPU_BTF_ID) | |
22dc4a0f | 641 | verbose(env, "%s", kernel_type_name(reg->btf, reg->btf_id)); |
cba368c1 MKL |
642 | verbose(env, "(id=%d", reg->id); |
643 | if (reg_type_may_be_refcounted_or_null(t)) | |
644 | verbose(env, ",ref_obj_id=%d", reg->ref_obj_id); | |
f1174f77 | 645 | if (t != SCALAR_VALUE) |
61bd5218 | 646 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 647 | if (type_is_pkt_pointer(t)) |
61bd5218 | 648 | verbose(env, ",r=%d", reg->range); |
f1174f77 | 649 | else if (t == CONST_PTR_TO_MAP || |
69c087ba | 650 | t == PTR_TO_MAP_KEY || |
f1174f77 EC |
651 | t == PTR_TO_MAP_VALUE || |
652 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 653 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
654 | reg->map_ptr->key_size, |
655 | reg->map_ptr->value_size); | |
7d1238f2 EC |
656 | if (tnum_is_const(reg->var_off)) { |
657 | /* Typically an immediate SCALAR_VALUE, but | |
658 | * could be a pointer whose offset is too big | |
659 | * for reg->off | |
660 | */ | |
61bd5218 | 661 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
662 | } else { |
663 | if (reg->smin_value != reg->umin_value && | |
664 | reg->smin_value != S64_MIN) | |
61bd5218 | 665 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
666 | (long long)reg->smin_value); |
667 | if (reg->smax_value != reg->umax_value && | |
668 | reg->smax_value != S64_MAX) | |
61bd5218 | 669 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
670 | (long long)reg->smax_value); |
671 | if (reg->umin_value != 0) | |
61bd5218 | 672 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
673 | (unsigned long long)reg->umin_value); |
674 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 675 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
676 | (unsigned long long)reg->umax_value); |
677 | if (!tnum_is_unknown(reg->var_off)) { | |
678 | char tn_buf[48]; | |
f1174f77 | 679 | |
7d1238f2 | 680 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 681 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 682 | } |
3f50f132 JF |
683 | if (reg->s32_min_value != reg->smin_value && |
684 | reg->s32_min_value != S32_MIN) | |
685 | verbose(env, ",s32_min_value=%d", | |
686 | (int)(reg->s32_min_value)); | |
687 | if (reg->s32_max_value != reg->smax_value && | |
688 | reg->s32_max_value != S32_MAX) | |
689 | verbose(env, ",s32_max_value=%d", | |
690 | (int)(reg->s32_max_value)); | |
691 | if (reg->u32_min_value != reg->umin_value && | |
692 | reg->u32_min_value != U32_MIN) | |
693 | verbose(env, ",u32_min_value=%d", | |
694 | (int)(reg->u32_min_value)); | |
695 | if (reg->u32_max_value != reg->umax_value && | |
696 | reg->u32_max_value != U32_MAX) | |
697 | verbose(env, ",u32_max_value=%d", | |
698 | (int)(reg->u32_max_value)); | |
f1174f77 | 699 | } |
61bd5218 | 700 | verbose(env, ")"); |
f1174f77 | 701 | } |
17a52670 | 702 | } |
638f5b90 | 703 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
704 | char types_buf[BPF_REG_SIZE + 1]; |
705 | bool valid = false; | |
706 | int j; | |
707 | ||
708 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
709 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
710 | valid = true; | |
711 | types_buf[j] = slot_type_char[ | |
712 | state->stack[i].slot_type[j]]; | |
713 | } | |
714 | types_buf[BPF_REG_SIZE] = 0; | |
715 | if (!valid) | |
716 | continue; | |
717 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
718 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
b5dc0163 AS |
719 | if (state->stack[i].slot_type[0] == STACK_SPILL) { |
720 | reg = &state->stack[i].spilled_ptr; | |
721 | t = reg->type; | |
722 | verbose(env, "=%s", reg_type_str[t]); | |
723 | if (t == SCALAR_VALUE && reg->precise) | |
724 | verbose(env, "P"); | |
725 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
726 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
727 | } else { | |
8efea21d | 728 | verbose(env, "=%s", types_buf); |
b5dc0163 | 729 | } |
17a52670 | 730 | } |
fd978bf7 JS |
731 | if (state->acquired_refs && state->refs[0].id) { |
732 | verbose(env, " refs=%d", state->refs[0].id); | |
733 | for (i = 1; i < state->acquired_refs; i++) | |
734 | if (state->refs[i].id) | |
735 | verbose(env, ",%d", state->refs[i].id); | |
736 | } | |
61bd5218 | 737 | verbose(env, "\n"); |
17a52670 AS |
738 | } |
739 | ||
c69431aa LB |
740 | /* copy array src of length n * size bytes to dst. dst is reallocated if it's too |
741 | * small to hold src. This is different from krealloc since we don't want to preserve | |
742 | * the contents of dst. | |
743 | * | |
744 | * Leaves dst untouched if src is NULL or length is zero. Returns NULL if memory could | |
745 | * not be allocated. | |
638f5b90 | 746 | */ |
c69431aa | 747 | static void *copy_array(void *dst, const void *src, size_t n, size_t size, gfp_t flags) |
638f5b90 | 748 | { |
c69431aa LB |
749 | size_t bytes; |
750 | ||
751 | if (ZERO_OR_NULL_PTR(src)) | |
752 | goto out; | |
753 | ||
754 | if (unlikely(check_mul_overflow(n, size, &bytes))) | |
755 | return NULL; | |
756 | ||
757 | if (ksize(dst) < bytes) { | |
758 | kfree(dst); | |
759 | dst = kmalloc_track_caller(bytes, flags); | |
760 | if (!dst) | |
761 | return NULL; | |
762 | } | |
763 | ||
764 | memcpy(dst, src, bytes); | |
765 | out: | |
766 | return dst ? dst : ZERO_SIZE_PTR; | |
767 | } | |
768 | ||
769 | /* resize an array from old_n items to new_n items. the array is reallocated if it's too | |
770 | * small to hold new_n items. new items are zeroed out if the array grows. | |
771 | * | |
772 | * Contrary to krealloc_array, does not free arr if new_n is zero. | |
773 | */ | |
774 | static void *realloc_array(void *arr, size_t old_n, size_t new_n, size_t size) | |
775 | { | |
776 | if (!new_n || old_n == new_n) | |
777 | goto out; | |
778 | ||
779 | arr = krealloc_array(arr, new_n, size, GFP_KERNEL); | |
780 | if (!arr) | |
781 | return NULL; | |
782 | ||
783 | if (new_n > old_n) | |
784 | memset(arr + old_n * size, 0, (new_n - old_n) * size); | |
785 | ||
786 | out: | |
787 | return arr ? arr : ZERO_SIZE_PTR; | |
788 | } | |
789 | ||
790 | static int copy_reference_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
791 | { | |
792 | dst->refs = copy_array(dst->refs, src->refs, src->acquired_refs, | |
793 | sizeof(struct bpf_reference_state), GFP_KERNEL); | |
794 | if (!dst->refs) | |
795 | return -ENOMEM; | |
796 | ||
797 | dst->acquired_refs = src->acquired_refs; | |
798 | return 0; | |
799 | } | |
800 | ||
801 | static int copy_stack_state(struct bpf_func_state *dst, const struct bpf_func_state *src) | |
802 | { | |
803 | size_t n = src->allocated_stack / BPF_REG_SIZE; | |
804 | ||
805 | dst->stack = copy_array(dst->stack, src->stack, n, sizeof(struct bpf_stack_state), | |
806 | GFP_KERNEL); | |
807 | if (!dst->stack) | |
808 | return -ENOMEM; | |
809 | ||
810 | dst->allocated_stack = src->allocated_stack; | |
811 | return 0; | |
812 | } | |
813 | ||
814 | static int resize_reference_state(struct bpf_func_state *state, size_t n) | |
815 | { | |
816 | state->refs = realloc_array(state->refs, state->acquired_refs, n, | |
817 | sizeof(struct bpf_reference_state)); | |
818 | if (!state->refs) | |
819 | return -ENOMEM; | |
820 | ||
821 | state->acquired_refs = n; | |
822 | return 0; | |
823 | } | |
824 | ||
825 | static int grow_stack_state(struct bpf_func_state *state, int size) | |
826 | { | |
827 | size_t old_n = state->allocated_stack / BPF_REG_SIZE, n = size / BPF_REG_SIZE; | |
828 | ||
829 | if (old_n >= n) | |
830 | return 0; | |
831 | ||
832 | state->stack = realloc_array(state->stack, old_n, n, sizeof(struct bpf_stack_state)); | |
833 | if (!state->stack) | |
834 | return -ENOMEM; | |
835 | ||
836 | state->allocated_stack = size; | |
837 | return 0; | |
fd978bf7 JS |
838 | } |
839 | ||
840 | /* Acquire a pointer id from the env and update the state->refs to include | |
841 | * this new pointer reference. | |
842 | * On success, returns a valid pointer id to associate with the register | |
843 | * On failure, returns a negative errno. | |
638f5b90 | 844 | */ |
fd978bf7 | 845 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 846 | { |
fd978bf7 JS |
847 | struct bpf_func_state *state = cur_func(env); |
848 | int new_ofs = state->acquired_refs; | |
849 | int id, err; | |
850 | ||
c69431aa | 851 | err = resize_reference_state(state, state->acquired_refs + 1); |
fd978bf7 JS |
852 | if (err) |
853 | return err; | |
854 | id = ++env->id_gen; | |
855 | state->refs[new_ofs].id = id; | |
856 | state->refs[new_ofs].insn_idx = insn_idx; | |
638f5b90 | 857 | |
fd978bf7 JS |
858 | return id; |
859 | } | |
860 | ||
861 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 862 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
863 | { |
864 | int i, last_idx; | |
865 | ||
fd978bf7 JS |
866 | last_idx = state->acquired_refs - 1; |
867 | for (i = 0; i < state->acquired_refs; i++) { | |
868 | if (state->refs[i].id == ptr_id) { | |
869 | if (last_idx && i != last_idx) | |
870 | memcpy(&state->refs[i], &state->refs[last_idx], | |
871 | sizeof(*state->refs)); | |
872 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
873 | state->acquired_refs--; | |
638f5b90 | 874 | return 0; |
638f5b90 | 875 | } |
638f5b90 | 876 | } |
46f8bc92 | 877 | return -EINVAL; |
fd978bf7 JS |
878 | } |
879 | ||
f4d7e40a AS |
880 | static void free_func_state(struct bpf_func_state *state) |
881 | { | |
5896351e AS |
882 | if (!state) |
883 | return; | |
fd978bf7 | 884 | kfree(state->refs); |
f4d7e40a AS |
885 | kfree(state->stack); |
886 | kfree(state); | |
887 | } | |
888 | ||
b5dc0163 AS |
889 | static void clear_jmp_history(struct bpf_verifier_state *state) |
890 | { | |
891 | kfree(state->jmp_history); | |
892 | state->jmp_history = NULL; | |
893 | state->jmp_history_cnt = 0; | |
894 | } | |
895 | ||
1969db47 AS |
896 | static void free_verifier_state(struct bpf_verifier_state *state, |
897 | bool free_self) | |
638f5b90 | 898 | { |
f4d7e40a AS |
899 | int i; |
900 | ||
901 | for (i = 0; i <= state->curframe; i++) { | |
902 | free_func_state(state->frame[i]); | |
903 | state->frame[i] = NULL; | |
904 | } | |
b5dc0163 | 905 | clear_jmp_history(state); |
1969db47 AS |
906 | if (free_self) |
907 | kfree(state); | |
638f5b90 AS |
908 | } |
909 | ||
910 | /* copy verifier state from src to dst growing dst stack space | |
911 | * when necessary to accommodate larger src stack | |
912 | */ | |
f4d7e40a AS |
913 | static int copy_func_state(struct bpf_func_state *dst, |
914 | const struct bpf_func_state *src) | |
638f5b90 AS |
915 | { |
916 | int err; | |
917 | ||
fd978bf7 JS |
918 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); |
919 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
920 | if (err) |
921 | return err; | |
638f5b90 AS |
922 | return copy_stack_state(dst, src); |
923 | } | |
924 | ||
f4d7e40a AS |
925 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
926 | const struct bpf_verifier_state *src) | |
927 | { | |
928 | struct bpf_func_state *dst; | |
929 | int i, err; | |
930 | ||
06ab6a50 LB |
931 | dst_state->jmp_history = copy_array(dst_state->jmp_history, src->jmp_history, |
932 | src->jmp_history_cnt, sizeof(struct bpf_idx_pair), | |
933 | GFP_USER); | |
934 | if (!dst_state->jmp_history) | |
935 | return -ENOMEM; | |
b5dc0163 AS |
936 | dst_state->jmp_history_cnt = src->jmp_history_cnt; |
937 | ||
f4d7e40a AS |
938 | /* if dst has more stack frames then src frame, free them */ |
939 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
940 | free_func_state(dst_state->frame[i]); | |
941 | dst_state->frame[i] = NULL; | |
942 | } | |
979d63d5 | 943 | dst_state->speculative = src->speculative; |
f4d7e40a | 944 | dst_state->curframe = src->curframe; |
d83525ca | 945 | dst_state->active_spin_lock = src->active_spin_lock; |
2589726d AS |
946 | dst_state->branches = src->branches; |
947 | dst_state->parent = src->parent; | |
b5dc0163 AS |
948 | dst_state->first_insn_idx = src->first_insn_idx; |
949 | dst_state->last_insn_idx = src->last_insn_idx; | |
f4d7e40a AS |
950 | for (i = 0; i <= src->curframe; i++) { |
951 | dst = dst_state->frame[i]; | |
952 | if (!dst) { | |
953 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
954 | if (!dst) | |
955 | return -ENOMEM; | |
956 | dst_state->frame[i] = dst; | |
957 | } | |
958 | err = copy_func_state(dst, src->frame[i]); | |
959 | if (err) | |
960 | return err; | |
961 | } | |
962 | return 0; | |
963 | } | |
964 | ||
2589726d AS |
965 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
966 | { | |
967 | while (st) { | |
968 | u32 br = --st->branches; | |
969 | ||
970 | /* WARN_ON(br > 1) technically makes sense here, | |
971 | * but see comment in push_stack(), hence: | |
972 | */ | |
973 | WARN_ONCE((int)br < 0, | |
974 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
975 | br); | |
976 | if (br) | |
977 | break; | |
978 | st = st->parent; | |
979 | } | |
980 | } | |
981 | ||
638f5b90 | 982 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 983 | int *insn_idx, bool pop_log) |
638f5b90 AS |
984 | { |
985 | struct bpf_verifier_state *cur = env->cur_state; | |
986 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
987 | int err; | |
17a52670 AS |
988 | |
989 | if (env->head == NULL) | |
638f5b90 | 990 | return -ENOENT; |
17a52670 | 991 | |
638f5b90 AS |
992 | if (cur) { |
993 | err = copy_verifier_state(cur, &head->st); | |
994 | if (err) | |
995 | return err; | |
996 | } | |
6f8a57cc AN |
997 | if (pop_log) |
998 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
999 | if (insn_idx) |
1000 | *insn_idx = head->insn_idx; | |
17a52670 | 1001 | if (prev_insn_idx) |
638f5b90 AS |
1002 | *prev_insn_idx = head->prev_insn_idx; |
1003 | elem = head->next; | |
1969db47 | 1004 | free_verifier_state(&head->st, false); |
638f5b90 | 1005 | kfree(head); |
17a52670 AS |
1006 | env->head = elem; |
1007 | env->stack_size--; | |
638f5b90 | 1008 | return 0; |
17a52670 AS |
1009 | } |
1010 | ||
58e2af8b | 1011 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
1012 | int insn_idx, int prev_insn_idx, |
1013 | bool speculative) | |
17a52670 | 1014 | { |
638f5b90 | 1015 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 1016 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 1017 | int err; |
17a52670 | 1018 | |
638f5b90 | 1019 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
1020 | if (!elem) |
1021 | goto err; | |
1022 | ||
17a52670 AS |
1023 | elem->insn_idx = insn_idx; |
1024 | elem->prev_insn_idx = prev_insn_idx; | |
1025 | elem->next = env->head; | |
6f8a57cc | 1026 | elem->log_pos = env->log.len_used; |
17a52670 AS |
1027 | env->head = elem; |
1028 | env->stack_size++; | |
1969db47 AS |
1029 | err = copy_verifier_state(&elem->st, cur); |
1030 | if (err) | |
1031 | goto err; | |
979d63d5 | 1032 | elem->st.speculative |= speculative; |
b285fcb7 AS |
1033 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
1034 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
1035 | env->stack_size); | |
17a52670 AS |
1036 | goto err; |
1037 | } | |
2589726d AS |
1038 | if (elem->st.parent) { |
1039 | ++elem->st.parent->branches; | |
1040 | /* WARN_ON(branches > 2) technically makes sense here, | |
1041 | * but | |
1042 | * 1. speculative states will bump 'branches' for non-branch | |
1043 | * instructions | |
1044 | * 2. is_state_visited() heuristics may decide not to create | |
1045 | * a new state for a sequence of branches and all such current | |
1046 | * and cloned states will be pointing to a single parent state | |
1047 | * which might have large 'branches' count. | |
1048 | */ | |
1049 | } | |
17a52670 AS |
1050 | return &elem->st; |
1051 | err: | |
5896351e AS |
1052 | free_verifier_state(env->cur_state, true); |
1053 | env->cur_state = NULL; | |
17a52670 | 1054 | /* pop all elements and return */ |
6f8a57cc | 1055 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
1056 | return NULL; |
1057 | } | |
1058 | ||
1059 | #define CALLER_SAVED_REGS 6 | |
1060 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
1061 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
1062 | }; | |
1063 | ||
f54c7898 DB |
1064 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1065 | struct bpf_reg_state *reg); | |
f1174f77 | 1066 | |
e688c3db AS |
1067 | /* This helper doesn't clear reg->id */ |
1068 | static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
b03c9f9f | 1069 | { |
b03c9f9f EC |
1070 | reg->var_off = tnum_const(imm); |
1071 | reg->smin_value = (s64)imm; | |
1072 | reg->smax_value = (s64)imm; | |
1073 | reg->umin_value = imm; | |
1074 | reg->umax_value = imm; | |
3f50f132 JF |
1075 | |
1076 | reg->s32_min_value = (s32)imm; | |
1077 | reg->s32_max_value = (s32)imm; | |
1078 | reg->u32_min_value = (u32)imm; | |
1079 | reg->u32_max_value = (u32)imm; | |
1080 | } | |
1081 | ||
e688c3db AS |
1082 | /* Mark the unknown part of a register (variable offset or scalar value) as |
1083 | * known to have the value @imm. | |
1084 | */ | |
1085 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
1086 | { | |
1087 | /* Clear id, off, and union(map_ptr, range) */ | |
1088 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
1089 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
1090 | ___mark_reg_known(reg, imm); | |
1091 | } | |
1092 | ||
3f50f132 JF |
1093 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) |
1094 | { | |
1095 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
1096 | reg->s32_min_value = (s32)imm; | |
1097 | reg->s32_max_value = (s32)imm; | |
1098 | reg->u32_min_value = (u32)imm; | |
1099 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
1100 | } |
1101 | ||
f1174f77 EC |
1102 | /* Mark the 'variable offset' part of a register as zero. This should be |
1103 | * used only on registers holding a pointer type. | |
1104 | */ | |
1105 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 1106 | { |
b03c9f9f | 1107 | __mark_reg_known(reg, 0); |
f1174f77 | 1108 | } |
a9789ef9 | 1109 | |
cc2b14d5 AS |
1110 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
1111 | { | |
1112 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
1113 | reg->type = SCALAR_VALUE; |
1114 | } | |
1115 | ||
61bd5218 JK |
1116 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
1117 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1118 | { |
1119 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1120 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
1121 | /* Something bad happened, let's kill all regs */ |
1122 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 1123 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1124 | return; |
1125 | } | |
1126 | __mark_reg_known_zero(regs + regno); | |
1127 | } | |
1128 | ||
4ddb7416 DB |
1129 | static void mark_ptr_not_null_reg(struct bpf_reg_state *reg) |
1130 | { | |
1131 | switch (reg->type) { | |
1132 | case PTR_TO_MAP_VALUE_OR_NULL: { | |
1133 | const struct bpf_map *map = reg->map_ptr; | |
1134 | ||
1135 | if (map->inner_map_meta) { | |
1136 | reg->type = CONST_PTR_TO_MAP; | |
1137 | reg->map_ptr = map->inner_map_meta; | |
1138 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { | |
1139 | reg->type = PTR_TO_XDP_SOCK; | |
1140 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || | |
1141 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
1142 | reg->type = PTR_TO_SOCKET; | |
1143 | } else { | |
1144 | reg->type = PTR_TO_MAP_VALUE; | |
1145 | } | |
1146 | break; | |
1147 | } | |
1148 | case PTR_TO_SOCKET_OR_NULL: | |
1149 | reg->type = PTR_TO_SOCKET; | |
1150 | break; | |
1151 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
1152 | reg->type = PTR_TO_SOCK_COMMON; | |
1153 | break; | |
1154 | case PTR_TO_TCP_SOCK_OR_NULL: | |
1155 | reg->type = PTR_TO_TCP_SOCK; | |
1156 | break; | |
1157 | case PTR_TO_BTF_ID_OR_NULL: | |
1158 | reg->type = PTR_TO_BTF_ID; | |
1159 | break; | |
1160 | case PTR_TO_MEM_OR_NULL: | |
1161 | reg->type = PTR_TO_MEM; | |
1162 | break; | |
1163 | case PTR_TO_RDONLY_BUF_OR_NULL: | |
1164 | reg->type = PTR_TO_RDONLY_BUF; | |
1165 | break; | |
1166 | case PTR_TO_RDWR_BUF_OR_NULL: | |
1167 | reg->type = PTR_TO_RDWR_BUF; | |
1168 | break; | |
1169 | default: | |
33ccec5f | 1170 | WARN_ONCE(1, "unknown nullable register type"); |
4ddb7416 DB |
1171 | } |
1172 | } | |
1173 | ||
de8f3a83 DB |
1174 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
1175 | { | |
1176 | return type_is_pkt_pointer(reg->type); | |
1177 | } | |
1178 | ||
1179 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
1180 | { | |
1181 | return reg_is_pkt_pointer(reg) || | |
1182 | reg->type == PTR_TO_PACKET_END; | |
1183 | } | |
1184 | ||
1185 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
1186 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
1187 | enum bpf_reg_type which) | |
1188 | { | |
1189 | /* The register can already have a range from prior markings. | |
1190 | * This is fine as long as it hasn't been advanced from its | |
1191 | * origin. | |
1192 | */ | |
1193 | return reg->type == which && | |
1194 | reg->id == 0 && | |
1195 | reg->off == 0 && | |
1196 | tnum_equals_const(reg->var_off, 0); | |
1197 | } | |
1198 | ||
3f50f132 JF |
1199 | /* Reset the min/max bounds of a register */ |
1200 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
1201 | { | |
1202 | reg->smin_value = S64_MIN; | |
1203 | reg->smax_value = S64_MAX; | |
1204 | reg->umin_value = 0; | |
1205 | reg->umax_value = U64_MAX; | |
1206 | ||
1207 | reg->s32_min_value = S32_MIN; | |
1208 | reg->s32_max_value = S32_MAX; | |
1209 | reg->u32_min_value = 0; | |
1210 | reg->u32_max_value = U32_MAX; | |
1211 | } | |
1212 | ||
1213 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
1214 | { | |
1215 | reg->smin_value = S64_MIN; | |
1216 | reg->smax_value = S64_MAX; | |
1217 | reg->umin_value = 0; | |
1218 | reg->umax_value = U64_MAX; | |
1219 | } | |
1220 | ||
1221 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
1222 | { | |
1223 | reg->s32_min_value = S32_MIN; | |
1224 | reg->s32_max_value = S32_MAX; | |
1225 | reg->u32_min_value = 0; | |
1226 | reg->u32_max_value = U32_MAX; | |
1227 | } | |
1228 | ||
1229 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
1230 | { | |
1231 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
1232 | ||
1233 | /* min signed is max(sign bit) | min(other bits) */ | |
1234 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
1235 | var32_off.value | (var32_off.mask & S32_MIN)); | |
1236 | /* max signed is min(sign bit) | max(other bits) */ | |
1237 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
1238 | var32_off.value | (var32_off.mask & S32_MAX)); | |
1239 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
1240 | reg->u32_max_value = min(reg->u32_max_value, | |
1241 | (u32)(var32_off.value | var32_off.mask)); | |
1242 | } | |
1243 | ||
1244 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1245 | { |
1246 | /* min signed is max(sign bit) | min(other bits) */ | |
1247 | reg->smin_value = max_t(s64, reg->smin_value, | |
1248 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
1249 | /* max signed is min(sign bit) | max(other bits) */ | |
1250 | reg->smax_value = min_t(s64, reg->smax_value, | |
1251 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
1252 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
1253 | reg->umax_value = min(reg->umax_value, | |
1254 | reg->var_off.value | reg->var_off.mask); | |
1255 | } | |
1256 | ||
3f50f132 JF |
1257 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
1258 | { | |
1259 | __update_reg32_bounds(reg); | |
1260 | __update_reg64_bounds(reg); | |
1261 | } | |
1262 | ||
b03c9f9f | 1263 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
1264 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
1265 | { | |
1266 | /* Learn sign from signed bounds. | |
1267 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1268 | * are the same, so combine. This works even in the negative case, e.g. | |
1269 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1270 | */ | |
1271 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
1272 | reg->s32_min_value = reg->u32_min_value = | |
1273 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1274 | reg->s32_max_value = reg->u32_max_value = | |
1275 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1276 | return; | |
1277 | } | |
1278 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1279 | * boundary, so we must be careful. | |
1280 | */ | |
1281 | if ((s32)reg->u32_max_value >= 0) { | |
1282 | /* Positive. We can't learn anything from the smin, but smax | |
1283 | * is positive, hence safe. | |
1284 | */ | |
1285 | reg->s32_min_value = reg->u32_min_value; | |
1286 | reg->s32_max_value = reg->u32_max_value = | |
1287 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1288 | } else if ((s32)reg->u32_min_value < 0) { | |
1289 | /* Negative. We can't learn anything from the smax, but smin | |
1290 | * is negative, hence safe. | |
1291 | */ | |
1292 | reg->s32_min_value = reg->u32_min_value = | |
1293 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1294 | reg->s32_max_value = reg->u32_max_value; | |
1295 | } | |
1296 | } | |
1297 | ||
1298 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1299 | { |
1300 | /* Learn sign from signed bounds. | |
1301 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1302 | * are the same, so combine. This works even in the negative case, e.g. | |
1303 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1304 | */ | |
1305 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
1306 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1307 | reg->umin_value); | |
1308 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1309 | reg->umax_value); | |
1310 | return; | |
1311 | } | |
1312 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1313 | * boundary, so we must be careful. | |
1314 | */ | |
1315 | if ((s64)reg->umax_value >= 0) { | |
1316 | /* Positive. We can't learn anything from the smin, but smax | |
1317 | * is positive, hence safe. | |
1318 | */ | |
1319 | reg->smin_value = reg->umin_value; | |
1320 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1321 | reg->umax_value); | |
1322 | } else if ((s64)reg->umin_value < 0) { | |
1323 | /* Negative. We can't learn anything from the smax, but smin | |
1324 | * is negative, hence safe. | |
1325 | */ | |
1326 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1327 | reg->umin_value); | |
1328 | reg->smax_value = reg->umax_value; | |
1329 | } | |
1330 | } | |
1331 | ||
3f50f132 JF |
1332 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
1333 | { | |
1334 | __reg32_deduce_bounds(reg); | |
1335 | __reg64_deduce_bounds(reg); | |
1336 | } | |
1337 | ||
b03c9f9f EC |
1338 | /* Attempts to improve var_off based on unsigned min/max information */ |
1339 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
1340 | { | |
3f50f132 JF |
1341 | struct tnum var64_off = tnum_intersect(reg->var_off, |
1342 | tnum_range(reg->umin_value, | |
1343 | reg->umax_value)); | |
1344 | struct tnum var32_off = tnum_intersect(tnum_subreg(reg->var_off), | |
1345 | tnum_range(reg->u32_min_value, | |
1346 | reg->u32_max_value)); | |
1347 | ||
1348 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
1349 | } |
1350 | ||
3f50f132 | 1351 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 1352 | { |
3f50f132 JF |
1353 | reg->umin_value = reg->u32_min_value; |
1354 | reg->umax_value = reg->u32_max_value; | |
1355 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds | |
1356 | * but must be positive otherwise set to worse case bounds | |
1357 | * and refine later from tnum. | |
1358 | */ | |
3a71dc36 | 1359 | if (reg->s32_min_value >= 0 && reg->s32_max_value >= 0) |
3f50f132 JF |
1360 | reg->smax_value = reg->s32_max_value; |
1361 | else | |
1362 | reg->smax_value = U32_MAX; | |
3a71dc36 JF |
1363 | if (reg->s32_min_value >= 0) |
1364 | reg->smin_value = reg->s32_min_value; | |
1365 | else | |
1366 | reg->smin_value = 0; | |
3f50f132 JF |
1367 | } |
1368 | ||
1369 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
1370 | { | |
1371 | /* special case when 64-bit register has upper 32-bit register | |
1372 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
1373 | * allowing us to use 32-bit bounds directly, | |
1374 | */ | |
1375 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
1376 | __reg_assign_32_into_64(reg); | |
1377 | } else { | |
1378 | /* Otherwise the best we can do is push lower 32bit known and | |
1379 | * unknown bits into register (var_off set from jmp logic) | |
1380 | * then learn as much as possible from the 64-bit tnum | |
1381 | * known and unknown bits. The previous smin/smax bounds are | |
1382 | * invalid here because of jmp32 compare so mark them unknown | |
1383 | * so they do not impact tnum bounds calculation. | |
1384 | */ | |
1385 | __mark_reg64_unbounded(reg); | |
1386 | __update_reg_bounds(reg); | |
1387 | } | |
1388 | ||
1389 | /* Intersecting with the old var_off might have improved our bounds | |
1390 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1391 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1392 | */ | |
1393 | __reg_deduce_bounds(reg); | |
1394 | __reg_bound_offset(reg); | |
1395 | __update_reg_bounds(reg); | |
1396 | } | |
1397 | ||
1398 | static bool __reg64_bound_s32(s64 a) | |
1399 | { | |
b0270958 | 1400 | return a > S32_MIN && a < S32_MAX; |
3f50f132 JF |
1401 | } |
1402 | ||
1403 | static bool __reg64_bound_u32(u64 a) | |
1404 | { | |
10bf4e83 | 1405 | return a > U32_MIN && a < U32_MAX; |
3f50f132 JF |
1406 | } |
1407 | ||
1408 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
1409 | { | |
1410 | __mark_reg32_unbounded(reg); | |
1411 | ||
b0270958 | 1412 | if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) { |
3f50f132 | 1413 | reg->s32_min_value = (s32)reg->smin_value; |
3f50f132 | 1414 | reg->s32_max_value = (s32)reg->smax_value; |
b0270958 | 1415 | } |
10bf4e83 | 1416 | if (__reg64_bound_u32(reg->umin_value) && __reg64_bound_u32(reg->umax_value)) { |
3f50f132 | 1417 | reg->u32_min_value = (u32)reg->umin_value; |
3f50f132 | 1418 | reg->u32_max_value = (u32)reg->umax_value; |
10bf4e83 | 1419 | } |
3f50f132 JF |
1420 | |
1421 | /* Intersecting with the old var_off might have improved our bounds | |
1422 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1423 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1424 | */ | |
1425 | __reg_deduce_bounds(reg); | |
1426 | __reg_bound_offset(reg); | |
1427 | __update_reg_bounds(reg); | |
b03c9f9f EC |
1428 | } |
1429 | ||
f1174f77 | 1430 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
1431 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1432 | struct bpf_reg_state *reg) | |
f1174f77 | 1433 | { |
a9c676bc AS |
1434 | /* |
1435 | * Clear type, id, off, and union(map_ptr, range) and | |
1436 | * padding between 'type' and union | |
1437 | */ | |
1438 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 1439 | reg->type = SCALAR_VALUE; |
f1174f77 | 1440 | reg->var_off = tnum_unknown; |
f4d7e40a | 1441 | reg->frameno = 0; |
2c78ee89 | 1442 | reg->precise = env->subprog_cnt > 1 || !env->bpf_capable; |
b03c9f9f | 1443 | __mark_reg_unbounded(reg); |
f1174f77 EC |
1444 | } |
1445 | ||
61bd5218 JK |
1446 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
1447 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1448 | { |
1449 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1450 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
1451 | /* Something bad happened, let's kill all regs except FP */ |
1452 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1453 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1454 | return; |
1455 | } | |
f54c7898 | 1456 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
1457 | } |
1458 | ||
f54c7898 DB |
1459 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1460 | struct bpf_reg_state *reg) | |
f1174f77 | 1461 | { |
f54c7898 | 1462 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
1463 | reg->type = NOT_INIT; |
1464 | } | |
1465 | ||
61bd5218 JK |
1466 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
1467 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1468 | { |
1469 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1470 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
1471 | /* Something bad happened, let's kill all regs except FP */ |
1472 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1473 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1474 | return; |
1475 | } | |
f54c7898 | 1476 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
1477 | } |
1478 | ||
41c48f3a AI |
1479 | static void mark_btf_ld_reg(struct bpf_verifier_env *env, |
1480 | struct bpf_reg_state *regs, u32 regno, | |
22dc4a0f AN |
1481 | enum bpf_reg_type reg_type, |
1482 | struct btf *btf, u32 btf_id) | |
41c48f3a AI |
1483 | { |
1484 | if (reg_type == SCALAR_VALUE) { | |
1485 | mark_reg_unknown(env, regs, regno); | |
1486 | return; | |
1487 | } | |
1488 | mark_reg_known_zero(env, regs, regno); | |
1489 | regs[regno].type = PTR_TO_BTF_ID; | |
22dc4a0f | 1490 | regs[regno].btf = btf; |
41c48f3a AI |
1491 | regs[regno].btf_id = btf_id; |
1492 | } | |
1493 | ||
5327ed3d | 1494 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 1495 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 1496 | struct bpf_func_state *state) |
17a52670 | 1497 | { |
f4d7e40a | 1498 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
1499 | int i; |
1500 | ||
dc503a8a | 1501 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 1502 | mark_reg_not_init(env, regs, i); |
dc503a8a | 1503 | regs[i].live = REG_LIVE_NONE; |
679c782d | 1504 | regs[i].parent = NULL; |
5327ed3d | 1505 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 1506 | } |
17a52670 AS |
1507 | |
1508 | /* frame pointer */ | |
f1174f77 | 1509 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 1510 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 1511 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
1512 | } |
1513 | ||
f4d7e40a AS |
1514 | #define BPF_MAIN_FUNC (-1) |
1515 | static void init_func_state(struct bpf_verifier_env *env, | |
1516 | struct bpf_func_state *state, | |
1517 | int callsite, int frameno, int subprogno) | |
1518 | { | |
1519 | state->callsite = callsite; | |
1520 | state->frameno = frameno; | |
1521 | state->subprogno = subprogno; | |
1522 | init_reg_state(env, state); | |
1523 | } | |
1524 | ||
17a52670 AS |
1525 | enum reg_arg_type { |
1526 | SRC_OP, /* register is used as source operand */ | |
1527 | DST_OP, /* register is used as destination operand */ | |
1528 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1529 | }; | |
1530 | ||
cc8b0b92 AS |
1531 | static int cmp_subprogs(const void *a, const void *b) |
1532 | { | |
9c8105bd JW |
1533 | return ((struct bpf_subprog_info *)a)->start - |
1534 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1535 | } |
1536 | ||
1537 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1538 | { | |
9c8105bd | 1539 | struct bpf_subprog_info *p; |
cc8b0b92 | 1540 | |
9c8105bd JW |
1541 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1542 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1543 | if (!p) |
1544 | return -ENOENT; | |
9c8105bd | 1545 | return p - env->subprog_info; |
cc8b0b92 AS |
1546 | |
1547 | } | |
1548 | ||
1549 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1550 | { | |
1551 | int insn_cnt = env->prog->len; | |
1552 | int ret; | |
1553 | ||
1554 | if (off >= insn_cnt || off < 0) { | |
1555 | verbose(env, "call to invalid destination\n"); | |
1556 | return -EINVAL; | |
1557 | } | |
1558 | ret = find_subprog(env, off); | |
1559 | if (ret >= 0) | |
282a0f46 | 1560 | return ret; |
4cb3d99c | 1561 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1562 | verbose(env, "too many subprograms\n"); |
1563 | return -E2BIG; | |
1564 | } | |
e6ac2450 | 1565 | /* determine subprog starts. The end is one before the next starts */ |
9c8105bd JW |
1566 | env->subprog_info[env->subprog_cnt++].start = off; |
1567 | sort(env->subprog_info, env->subprog_cnt, | |
1568 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
282a0f46 | 1569 | return env->subprog_cnt - 1; |
cc8b0b92 AS |
1570 | } |
1571 | ||
e6ac2450 MKL |
1572 | struct bpf_kfunc_desc { |
1573 | struct btf_func_model func_model; | |
1574 | u32 func_id; | |
1575 | s32 imm; | |
1576 | }; | |
1577 | ||
1578 | #define MAX_KFUNC_DESCS 256 | |
1579 | struct bpf_kfunc_desc_tab { | |
1580 | struct bpf_kfunc_desc descs[MAX_KFUNC_DESCS]; | |
1581 | u32 nr_descs; | |
1582 | }; | |
1583 | ||
1584 | static int kfunc_desc_cmp_by_id(const void *a, const void *b) | |
1585 | { | |
1586 | const struct bpf_kfunc_desc *d0 = a; | |
1587 | const struct bpf_kfunc_desc *d1 = b; | |
1588 | ||
1589 | /* func_id is not greater than BTF_MAX_TYPE */ | |
1590 | return d0->func_id - d1->func_id; | |
1591 | } | |
1592 | ||
1593 | static const struct bpf_kfunc_desc * | |
1594 | find_kfunc_desc(const struct bpf_prog *prog, u32 func_id) | |
1595 | { | |
1596 | struct bpf_kfunc_desc desc = { | |
1597 | .func_id = func_id, | |
1598 | }; | |
1599 | struct bpf_kfunc_desc_tab *tab; | |
1600 | ||
1601 | tab = prog->aux->kfunc_tab; | |
1602 | return bsearch(&desc, tab->descs, tab->nr_descs, | |
1603 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_id); | |
1604 | } | |
1605 | ||
1606 | static int add_kfunc_call(struct bpf_verifier_env *env, u32 func_id) | |
1607 | { | |
1608 | const struct btf_type *func, *func_proto; | |
1609 | struct bpf_kfunc_desc_tab *tab; | |
1610 | struct bpf_prog_aux *prog_aux; | |
1611 | struct bpf_kfunc_desc *desc; | |
1612 | const char *func_name; | |
1613 | unsigned long addr; | |
1614 | int err; | |
1615 | ||
1616 | prog_aux = env->prog->aux; | |
1617 | tab = prog_aux->kfunc_tab; | |
1618 | if (!tab) { | |
1619 | if (!btf_vmlinux) { | |
1620 | verbose(env, "calling kernel function is not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
1621 | return -ENOTSUPP; | |
1622 | } | |
1623 | ||
1624 | if (!env->prog->jit_requested) { | |
1625 | verbose(env, "JIT is required for calling kernel function\n"); | |
1626 | return -ENOTSUPP; | |
1627 | } | |
1628 | ||
1629 | if (!bpf_jit_supports_kfunc_call()) { | |
1630 | verbose(env, "JIT does not support calling kernel function\n"); | |
1631 | return -ENOTSUPP; | |
1632 | } | |
1633 | ||
1634 | if (!env->prog->gpl_compatible) { | |
1635 | verbose(env, "cannot call kernel function from non-GPL compatible program\n"); | |
1636 | return -EINVAL; | |
1637 | } | |
1638 | ||
1639 | tab = kzalloc(sizeof(*tab), GFP_KERNEL); | |
1640 | if (!tab) | |
1641 | return -ENOMEM; | |
1642 | prog_aux->kfunc_tab = tab; | |
1643 | } | |
1644 | ||
1645 | if (find_kfunc_desc(env->prog, func_id)) | |
1646 | return 0; | |
1647 | ||
1648 | if (tab->nr_descs == MAX_KFUNC_DESCS) { | |
1649 | verbose(env, "too many different kernel function calls\n"); | |
1650 | return -E2BIG; | |
1651 | } | |
1652 | ||
1653 | func = btf_type_by_id(btf_vmlinux, func_id); | |
1654 | if (!func || !btf_type_is_func(func)) { | |
1655 | verbose(env, "kernel btf_id %u is not a function\n", | |
1656 | func_id); | |
1657 | return -EINVAL; | |
1658 | } | |
1659 | func_proto = btf_type_by_id(btf_vmlinux, func->type); | |
1660 | if (!func_proto || !btf_type_is_func_proto(func_proto)) { | |
1661 | verbose(env, "kernel function btf_id %u does not have a valid func_proto\n", | |
1662 | func_id); | |
1663 | return -EINVAL; | |
1664 | } | |
1665 | ||
1666 | func_name = btf_name_by_offset(btf_vmlinux, func->name_off); | |
1667 | addr = kallsyms_lookup_name(func_name); | |
1668 | if (!addr) { | |
1669 | verbose(env, "cannot find address for kernel function %s\n", | |
1670 | func_name); | |
1671 | return -EINVAL; | |
1672 | } | |
1673 | ||
1674 | desc = &tab->descs[tab->nr_descs++]; | |
1675 | desc->func_id = func_id; | |
1676 | desc->imm = BPF_CAST_CALL(addr) - __bpf_call_base; | |
1677 | err = btf_distill_func_proto(&env->log, btf_vmlinux, | |
1678 | func_proto, func_name, | |
1679 | &desc->func_model); | |
1680 | if (!err) | |
1681 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
1682 | kfunc_desc_cmp_by_id, NULL); | |
1683 | return err; | |
1684 | } | |
1685 | ||
1686 | static int kfunc_desc_cmp_by_imm(const void *a, const void *b) | |
1687 | { | |
1688 | const struct bpf_kfunc_desc *d0 = a; | |
1689 | const struct bpf_kfunc_desc *d1 = b; | |
1690 | ||
1691 | if (d0->imm > d1->imm) | |
1692 | return 1; | |
1693 | else if (d0->imm < d1->imm) | |
1694 | return -1; | |
1695 | return 0; | |
1696 | } | |
1697 | ||
1698 | static void sort_kfunc_descs_by_imm(struct bpf_prog *prog) | |
1699 | { | |
1700 | struct bpf_kfunc_desc_tab *tab; | |
1701 | ||
1702 | tab = prog->aux->kfunc_tab; | |
1703 | if (!tab) | |
1704 | return; | |
1705 | ||
1706 | sort(tab->descs, tab->nr_descs, sizeof(tab->descs[0]), | |
1707 | kfunc_desc_cmp_by_imm, NULL); | |
1708 | } | |
1709 | ||
1710 | bool bpf_prog_has_kfunc_call(const struct bpf_prog *prog) | |
1711 | { | |
1712 | return !!prog->aux->kfunc_tab; | |
1713 | } | |
1714 | ||
1715 | const struct btf_func_model * | |
1716 | bpf_jit_find_kfunc_model(const struct bpf_prog *prog, | |
1717 | const struct bpf_insn *insn) | |
1718 | { | |
1719 | const struct bpf_kfunc_desc desc = { | |
1720 | .imm = insn->imm, | |
1721 | }; | |
1722 | const struct bpf_kfunc_desc *res; | |
1723 | struct bpf_kfunc_desc_tab *tab; | |
1724 | ||
1725 | tab = prog->aux->kfunc_tab; | |
1726 | res = bsearch(&desc, tab->descs, tab->nr_descs, | |
1727 | sizeof(tab->descs[0]), kfunc_desc_cmp_by_imm); | |
1728 | ||
1729 | return res ? &res->func_model : NULL; | |
1730 | } | |
1731 | ||
1732 | static int add_subprog_and_kfunc(struct bpf_verifier_env *env) | |
cc8b0b92 | 1733 | { |
9c8105bd | 1734 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 | 1735 | struct bpf_insn *insn = env->prog->insnsi; |
e6ac2450 | 1736 | int i, ret, insn_cnt = env->prog->len; |
cc8b0b92 | 1737 | |
f910cefa JW |
1738 | /* Add entry function. */ |
1739 | ret = add_subprog(env, 0); | |
e6ac2450 | 1740 | if (ret) |
f910cefa JW |
1741 | return ret; |
1742 | ||
e6ac2450 MKL |
1743 | for (i = 0; i < insn_cnt; i++, insn++) { |
1744 | if (!bpf_pseudo_func(insn) && !bpf_pseudo_call(insn) && | |
1745 | !bpf_pseudo_kfunc_call(insn)) | |
cc8b0b92 | 1746 | continue; |
e6ac2450 | 1747 | |
2c78ee89 | 1748 | if (!env->bpf_capable) { |
e6ac2450 | 1749 | verbose(env, "loading/calling other bpf or kernel functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); |
cc8b0b92 AS |
1750 | return -EPERM; |
1751 | } | |
e6ac2450 MKL |
1752 | |
1753 | if (bpf_pseudo_func(insn)) { | |
1754 | ret = add_subprog(env, i + insn->imm + 1); | |
1755 | if (ret >= 0) | |
1756 | /* remember subprog */ | |
1757 | insn[1].imm = ret; | |
1758 | } else if (bpf_pseudo_call(insn)) { | |
1759 | ret = add_subprog(env, i + insn->imm + 1); | |
1760 | } else { | |
1761 | ret = add_kfunc_call(env, insn->imm); | |
1762 | } | |
1763 | ||
cc8b0b92 AS |
1764 | if (ret < 0) |
1765 | return ret; | |
1766 | } | |
1767 | ||
4cb3d99c JW |
1768 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
1769 | * logic. 'subprog_cnt' should not be increased. | |
1770 | */ | |
1771 | subprog[env->subprog_cnt].start = insn_cnt; | |
1772 | ||
06ee7115 | 1773 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 1774 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 1775 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 | 1776 | |
e6ac2450 MKL |
1777 | return 0; |
1778 | } | |
1779 | ||
1780 | static int check_subprogs(struct bpf_verifier_env *env) | |
1781 | { | |
1782 | int i, subprog_start, subprog_end, off, cur_subprog = 0; | |
1783 | struct bpf_subprog_info *subprog = env->subprog_info; | |
1784 | struct bpf_insn *insn = env->prog->insnsi; | |
1785 | int insn_cnt = env->prog->len; | |
1786 | ||
cc8b0b92 | 1787 | /* now check that all jumps are within the same subprog */ |
4cb3d99c JW |
1788 | subprog_start = subprog[cur_subprog].start; |
1789 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
1790 | for (i = 0; i < insn_cnt; i++) { |
1791 | u8 code = insn[i].code; | |
1792 | ||
7f6e4312 MF |
1793 | if (code == (BPF_JMP | BPF_CALL) && |
1794 | insn[i].imm == BPF_FUNC_tail_call && | |
1795 | insn[i].src_reg != BPF_PSEUDO_CALL) | |
1796 | subprog[cur_subprog].has_tail_call = true; | |
09b28d76 AS |
1797 | if (BPF_CLASS(code) == BPF_LD && |
1798 | (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND)) | |
1799 | subprog[cur_subprog].has_ld_abs = true; | |
092ed096 | 1800 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
1801 | goto next; |
1802 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
1803 | goto next; | |
1804 | off = i + insn[i].off + 1; | |
1805 | if (off < subprog_start || off >= subprog_end) { | |
1806 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
1807 | return -EINVAL; | |
1808 | } | |
1809 | next: | |
1810 | if (i == subprog_end - 1) { | |
1811 | /* to avoid fall-through from one subprog into another | |
1812 | * the last insn of the subprog should be either exit | |
1813 | * or unconditional jump back | |
1814 | */ | |
1815 | if (code != (BPF_JMP | BPF_EXIT) && | |
1816 | code != (BPF_JMP | BPF_JA)) { | |
1817 | verbose(env, "last insn is not an exit or jmp\n"); | |
1818 | return -EINVAL; | |
1819 | } | |
1820 | subprog_start = subprog_end; | |
4cb3d99c JW |
1821 | cur_subprog++; |
1822 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 1823 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
1824 | } |
1825 | } | |
1826 | return 0; | |
1827 | } | |
1828 | ||
679c782d EC |
1829 | /* Parentage chain of this register (or stack slot) should take care of all |
1830 | * issues like callee-saved registers, stack slot allocation time, etc. | |
1831 | */ | |
f4d7e40a | 1832 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 1833 | const struct bpf_reg_state *state, |
5327ed3d | 1834 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
1835 | { |
1836 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 1837 | int cnt = 0; |
dc503a8a EC |
1838 | |
1839 | while (parent) { | |
1840 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 1841 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 1842 | break; |
9242b5f5 AS |
1843 | if (parent->live & REG_LIVE_DONE) { |
1844 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
1845 | reg_type_str[parent->type], | |
1846 | parent->var_off.value, parent->off); | |
1847 | return -EFAULT; | |
1848 | } | |
5327ed3d JW |
1849 | /* The first condition is more likely to be true than the |
1850 | * second, checked it first. | |
1851 | */ | |
1852 | if ((parent->live & REG_LIVE_READ) == flag || | |
1853 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
1854 | /* The parentage chain never changes and |
1855 | * this parent was already marked as LIVE_READ. | |
1856 | * There is no need to keep walking the chain again and | |
1857 | * keep re-marking all parents as LIVE_READ. | |
1858 | * This case happens when the same register is read | |
1859 | * multiple times without writes into it in-between. | |
5327ed3d JW |
1860 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
1861 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
1862 | */ |
1863 | break; | |
dc503a8a | 1864 | /* ... then we depend on parent's value */ |
5327ed3d JW |
1865 | parent->live |= flag; |
1866 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
1867 | if (flag == REG_LIVE_READ64) | |
1868 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
1869 | state = parent; |
1870 | parent = state->parent; | |
f4d7e40a | 1871 | writes = true; |
06ee7115 | 1872 | cnt++; |
dc503a8a | 1873 | } |
06ee7115 AS |
1874 | |
1875 | if (env->longest_mark_read_walk < cnt) | |
1876 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 1877 | return 0; |
dc503a8a EC |
1878 | } |
1879 | ||
5327ed3d JW |
1880 | /* This function is supposed to be used by the following 32-bit optimization |
1881 | * code only. It returns TRUE if the source or destination register operates | |
1882 | * on 64-bit, otherwise return FALSE. | |
1883 | */ | |
1884 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
1885 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
1886 | { | |
1887 | u8 code, class, op; | |
1888 | ||
1889 | code = insn->code; | |
1890 | class = BPF_CLASS(code); | |
1891 | op = BPF_OP(code); | |
1892 | if (class == BPF_JMP) { | |
1893 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
1894 | * conservatively. | |
1895 | */ | |
1896 | if (op == BPF_EXIT) | |
1897 | return true; | |
1898 | if (op == BPF_CALL) { | |
1899 | /* BPF to BPF call will reach here because of marking | |
1900 | * caller saved clobber with DST_OP_NO_MARK for which we | |
1901 | * don't care the register def because they are anyway | |
1902 | * marked as NOT_INIT already. | |
1903 | */ | |
1904 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
1905 | return false; | |
1906 | /* Helper call will reach here because of arg type | |
1907 | * check, conservatively return TRUE. | |
1908 | */ | |
1909 | if (t == SRC_OP) | |
1910 | return true; | |
1911 | ||
1912 | return false; | |
1913 | } | |
1914 | } | |
1915 | ||
1916 | if (class == BPF_ALU64 || class == BPF_JMP || | |
1917 | /* BPF_END always use BPF_ALU class. */ | |
1918 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) | |
1919 | return true; | |
1920 | ||
1921 | if (class == BPF_ALU || class == BPF_JMP32) | |
1922 | return false; | |
1923 | ||
1924 | if (class == BPF_LDX) { | |
1925 | if (t != SRC_OP) | |
1926 | return BPF_SIZE(code) == BPF_DW; | |
1927 | /* LDX source must be ptr. */ | |
1928 | return true; | |
1929 | } | |
1930 | ||
1931 | if (class == BPF_STX) { | |
83a28819 IL |
1932 | /* BPF_STX (including atomic variants) has multiple source |
1933 | * operands, one of which is a ptr. Check whether the caller is | |
1934 | * asking about it. | |
1935 | */ | |
1936 | if (t == SRC_OP && reg->type != SCALAR_VALUE) | |
5327ed3d JW |
1937 | return true; |
1938 | return BPF_SIZE(code) == BPF_DW; | |
1939 | } | |
1940 | ||
1941 | if (class == BPF_LD) { | |
1942 | u8 mode = BPF_MODE(code); | |
1943 | ||
1944 | /* LD_IMM64 */ | |
1945 | if (mode == BPF_IMM) | |
1946 | return true; | |
1947 | ||
1948 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
1949 | if (t != SRC_OP) | |
1950 | return false; | |
1951 | ||
1952 | /* Implicit ctx ptr. */ | |
1953 | if (regno == BPF_REG_6) | |
1954 | return true; | |
1955 | ||
1956 | /* Explicit source could be any width. */ | |
1957 | return true; | |
1958 | } | |
1959 | ||
1960 | if (class == BPF_ST) | |
1961 | /* The only source register for BPF_ST is a ptr. */ | |
1962 | return true; | |
1963 | ||
1964 | /* Conservatively return true at default. */ | |
1965 | return true; | |
1966 | } | |
1967 | ||
83a28819 IL |
1968 | /* Return the regno defined by the insn, or -1. */ |
1969 | static int insn_def_regno(const struct bpf_insn *insn) | |
b325fbca | 1970 | { |
83a28819 IL |
1971 | switch (BPF_CLASS(insn->code)) { |
1972 | case BPF_JMP: | |
1973 | case BPF_JMP32: | |
1974 | case BPF_ST: | |
1975 | return -1; | |
1976 | case BPF_STX: | |
1977 | if (BPF_MODE(insn->code) == BPF_ATOMIC && | |
1978 | (insn->imm & BPF_FETCH)) { | |
1979 | if (insn->imm == BPF_CMPXCHG) | |
1980 | return BPF_REG_0; | |
1981 | else | |
1982 | return insn->src_reg; | |
1983 | } else { | |
1984 | return -1; | |
1985 | } | |
1986 | default: | |
1987 | return insn->dst_reg; | |
1988 | } | |
b325fbca JW |
1989 | } |
1990 | ||
1991 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
1992 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
1993 | { | |
83a28819 IL |
1994 | int dst_reg = insn_def_regno(insn); |
1995 | ||
1996 | if (dst_reg == -1) | |
b325fbca JW |
1997 | return false; |
1998 | ||
83a28819 | 1999 | return !is_reg64(env, insn, dst_reg, NULL, DST_OP); |
b325fbca JW |
2000 | } |
2001 | ||
5327ed3d JW |
2002 | static void mark_insn_zext(struct bpf_verifier_env *env, |
2003 | struct bpf_reg_state *reg) | |
2004 | { | |
2005 | s32 def_idx = reg->subreg_def; | |
2006 | ||
2007 | if (def_idx == DEF_NOT_SUBREG) | |
2008 | return; | |
2009 | ||
2010 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
2011 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
2012 | reg->subreg_def = DEF_NOT_SUBREG; | |
2013 | } | |
2014 | ||
dc503a8a | 2015 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
2016 | enum reg_arg_type t) |
2017 | { | |
f4d7e40a AS |
2018 | struct bpf_verifier_state *vstate = env->cur_state; |
2019 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 2020 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 2021 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 2022 | bool rw64; |
dc503a8a | 2023 | |
17a52670 | 2024 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 2025 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
2026 | return -EINVAL; |
2027 | } | |
2028 | ||
c342dc10 | 2029 | reg = ®s[regno]; |
5327ed3d | 2030 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
2031 | if (t == SRC_OP) { |
2032 | /* check whether register used as source operand can be read */ | |
c342dc10 | 2033 | if (reg->type == NOT_INIT) { |
61bd5218 | 2034 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
2035 | return -EACCES; |
2036 | } | |
679c782d | 2037 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
2038 | if (regno == BPF_REG_FP) |
2039 | return 0; | |
2040 | ||
5327ed3d JW |
2041 | if (rw64) |
2042 | mark_insn_zext(env, reg); | |
2043 | ||
2044 | return mark_reg_read(env, reg, reg->parent, | |
2045 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
2046 | } else { |
2047 | /* check whether register used as dest operand can be written to */ | |
2048 | if (regno == BPF_REG_FP) { | |
61bd5218 | 2049 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
2050 | return -EACCES; |
2051 | } | |
c342dc10 | 2052 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 2053 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 2054 | if (t == DST_OP) |
61bd5218 | 2055 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
2056 | } |
2057 | return 0; | |
2058 | } | |
2059 | ||
b5dc0163 AS |
2060 | /* for any branch, call, exit record the history of jmps in the given state */ |
2061 | static int push_jmp_history(struct bpf_verifier_env *env, | |
2062 | struct bpf_verifier_state *cur) | |
2063 | { | |
2064 | u32 cnt = cur->jmp_history_cnt; | |
2065 | struct bpf_idx_pair *p; | |
2066 | ||
2067 | cnt++; | |
2068 | p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER); | |
2069 | if (!p) | |
2070 | return -ENOMEM; | |
2071 | p[cnt - 1].idx = env->insn_idx; | |
2072 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
2073 | cur->jmp_history = p; | |
2074 | cur->jmp_history_cnt = cnt; | |
2075 | return 0; | |
2076 | } | |
2077 | ||
2078 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
2079 | * history then previous instruction came from straight line execution. | |
2080 | */ | |
2081 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
2082 | u32 *history) | |
2083 | { | |
2084 | u32 cnt = *history; | |
2085 | ||
2086 | if (cnt && st->jmp_history[cnt - 1].idx == i) { | |
2087 | i = st->jmp_history[cnt - 1].prev_idx; | |
2088 | (*history)--; | |
2089 | } else { | |
2090 | i--; | |
2091 | } | |
2092 | return i; | |
2093 | } | |
2094 | ||
e6ac2450 MKL |
2095 | static const char *disasm_kfunc_name(void *data, const struct bpf_insn *insn) |
2096 | { | |
2097 | const struct btf_type *func; | |
2098 | ||
2099 | if (insn->src_reg != BPF_PSEUDO_KFUNC_CALL) | |
2100 | return NULL; | |
2101 | ||
2102 | func = btf_type_by_id(btf_vmlinux, insn->imm); | |
2103 | return btf_name_by_offset(btf_vmlinux, func->name_off); | |
2104 | } | |
2105 | ||
b5dc0163 AS |
2106 | /* For given verifier state backtrack_insn() is called from the last insn to |
2107 | * the first insn. Its purpose is to compute a bitmask of registers and | |
2108 | * stack slots that needs precision in the parent verifier state. | |
2109 | */ | |
2110 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, | |
2111 | u32 *reg_mask, u64 *stack_mask) | |
2112 | { | |
2113 | const struct bpf_insn_cbs cbs = { | |
e6ac2450 | 2114 | .cb_call = disasm_kfunc_name, |
b5dc0163 AS |
2115 | .cb_print = verbose, |
2116 | .private_data = env, | |
2117 | }; | |
2118 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
2119 | u8 class = BPF_CLASS(insn->code); | |
2120 | u8 opcode = BPF_OP(insn->code); | |
2121 | u8 mode = BPF_MODE(insn->code); | |
2122 | u32 dreg = 1u << insn->dst_reg; | |
2123 | u32 sreg = 1u << insn->src_reg; | |
2124 | u32 spi; | |
2125 | ||
2126 | if (insn->code == 0) | |
2127 | return 0; | |
2128 | if (env->log.level & BPF_LOG_LEVEL) { | |
2129 | verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask); | |
2130 | verbose(env, "%d: ", idx); | |
2131 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
2132 | } | |
2133 | ||
2134 | if (class == BPF_ALU || class == BPF_ALU64) { | |
2135 | if (!(*reg_mask & dreg)) | |
2136 | return 0; | |
2137 | if (opcode == BPF_MOV) { | |
2138 | if (BPF_SRC(insn->code) == BPF_X) { | |
2139 | /* dreg = sreg | |
2140 | * dreg needs precision after this insn | |
2141 | * sreg needs precision before this insn | |
2142 | */ | |
2143 | *reg_mask &= ~dreg; | |
2144 | *reg_mask |= sreg; | |
2145 | } else { | |
2146 | /* dreg = K | |
2147 | * dreg needs precision after this insn. | |
2148 | * Corresponding register is already marked | |
2149 | * as precise=true in this verifier state. | |
2150 | * No further markings in parent are necessary | |
2151 | */ | |
2152 | *reg_mask &= ~dreg; | |
2153 | } | |
2154 | } else { | |
2155 | if (BPF_SRC(insn->code) == BPF_X) { | |
2156 | /* dreg += sreg | |
2157 | * both dreg and sreg need precision | |
2158 | * before this insn | |
2159 | */ | |
2160 | *reg_mask |= sreg; | |
2161 | } /* else dreg += K | |
2162 | * dreg still needs precision before this insn | |
2163 | */ | |
2164 | } | |
2165 | } else if (class == BPF_LDX) { | |
2166 | if (!(*reg_mask & dreg)) | |
2167 | return 0; | |
2168 | *reg_mask &= ~dreg; | |
2169 | ||
2170 | /* scalars can only be spilled into stack w/o losing precision. | |
2171 | * Load from any other memory can be zero extended. | |
2172 | * The desire to keep that precision is already indicated | |
2173 | * by 'precise' mark in corresponding register of this state. | |
2174 | * No further tracking necessary. | |
2175 | */ | |
2176 | if (insn->src_reg != BPF_REG_FP) | |
2177 | return 0; | |
2178 | if (BPF_SIZE(insn->code) != BPF_DW) | |
2179 | return 0; | |
2180 | ||
2181 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
2182 | * that [fp - off] slot contains scalar that needs to be | |
2183 | * tracked with precision | |
2184 | */ | |
2185 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
2186 | if (spi >= 64) { | |
2187 | verbose(env, "BUG spi %d\n", spi); | |
2188 | WARN_ONCE(1, "verifier backtracking bug"); | |
2189 | return -EFAULT; | |
2190 | } | |
2191 | *stack_mask |= 1ull << spi; | |
b3b50f05 | 2192 | } else if (class == BPF_STX || class == BPF_ST) { |
b5dc0163 | 2193 | if (*reg_mask & dreg) |
b3b50f05 | 2194 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
2195 | * to access memory. It means backtracking |
2196 | * encountered a case of pointer subtraction. | |
2197 | */ | |
2198 | return -ENOTSUPP; | |
2199 | /* scalars can only be spilled into stack */ | |
2200 | if (insn->dst_reg != BPF_REG_FP) | |
2201 | return 0; | |
2202 | if (BPF_SIZE(insn->code) != BPF_DW) | |
2203 | return 0; | |
2204 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
2205 | if (spi >= 64) { | |
2206 | verbose(env, "BUG spi %d\n", spi); | |
2207 | WARN_ONCE(1, "verifier backtracking bug"); | |
2208 | return -EFAULT; | |
2209 | } | |
2210 | if (!(*stack_mask & (1ull << spi))) | |
2211 | return 0; | |
2212 | *stack_mask &= ~(1ull << spi); | |
b3b50f05 AN |
2213 | if (class == BPF_STX) |
2214 | *reg_mask |= sreg; | |
b5dc0163 AS |
2215 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
2216 | if (opcode == BPF_CALL) { | |
2217 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
2218 | return -ENOTSUPP; | |
2219 | /* regular helper call sets R0 */ | |
2220 | *reg_mask &= ~1; | |
2221 | if (*reg_mask & 0x3f) { | |
2222 | /* if backtracing was looking for registers R1-R5 | |
2223 | * they should have been found already. | |
2224 | */ | |
2225 | verbose(env, "BUG regs %x\n", *reg_mask); | |
2226 | WARN_ONCE(1, "verifier backtracking bug"); | |
2227 | return -EFAULT; | |
2228 | } | |
2229 | } else if (opcode == BPF_EXIT) { | |
2230 | return -ENOTSUPP; | |
2231 | } | |
2232 | } else if (class == BPF_LD) { | |
2233 | if (!(*reg_mask & dreg)) | |
2234 | return 0; | |
2235 | *reg_mask &= ~dreg; | |
2236 | /* It's ld_imm64 or ld_abs or ld_ind. | |
2237 | * For ld_imm64 no further tracking of precision | |
2238 | * into parent is necessary | |
2239 | */ | |
2240 | if (mode == BPF_IND || mode == BPF_ABS) | |
2241 | /* to be analyzed */ | |
2242 | return -ENOTSUPP; | |
b5dc0163 AS |
2243 | } |
2244 | return 0; | |
2245 | } | |
2246 | ||
2247 | /* the scalar precision tracking algorithm: | |
2248 | * . at the start all registers have precise=false. | |
2249 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
2250 | * . once precise value of the scalar register is used in: | |
2251 | * . ptr + scalar alu | |
2252 | * . if (scalar cond K|scalar) | |
2253 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
2254 | * backtrack through the verifier states and mark all registers and | |
2255 | * stack slots with spilled constants that these scalar regisers | |
2256 | * should be precise. | |
2257 | * . during state pruning two registers (or spilled stack slots) | |
2258 | * are equivalent if both are not precise. | |
2259 | * | |
2260 | * Note the verifier cannot simply walk register parentage chain, | |
2261 | * since many different registers and stack slots could have been | |
2262 | * used to compute single precise scalar. | |
2263 | * | |
2264 | * The approach of starting with precise=true for all registers and then | |
2265 | * backtrack to mark a register as not precise when the verifier detects | |
2266 | * that program doesn't care about specific value (e.g., when helper | |
2267 | * takes register as ARG_ANYTHING parameter) is not safe. | |
2268 | * | |
2269 | * It's ok to walk single parentage chain of the verifier states. | |
2270 | * It's possible that this backtracking will go all the way till 1st insn. | |
2271 | * All other branches will be explored for needing precision later. | |
2272 | * | |
2273 | * The backtracking needs to deal with cases like: | |
2274 | * 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) | |
2275 | * r9 -= r8 | |
2276 | * r5 = r9 | |
2277 | * if r5 > 0x79f goto pc+7 | |
2278 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
2279 | * r5 += 1 | |
2280 | * ... | |
2281 | * call bpf_perf_event_output#25 | |
2282 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
2283 | * | |
2284 | * and this case: | |
2285 | * r6 = 1 | |
2286 | * call foo // uses callee's r6 inside to compute r0 | |
2287 | * r0 += r6 | |
2288 | * if r0 == 0 goto | |
2289 | * | |
2290 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
2291 | * | |
2292 | * Also if parent's curframe > frame where backtracking started, | |
2293 | * the verifier need to mark registers in both frames, otherwise callees | |
2294 | * may incorrectly prune callers. This is similar to | |
2295 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
2296 | * | |
2297 | * For now backtracking falls back into conservative marking. | |
2298 | */ | |
2299 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
2300 | struct bpf_verifier_state *st) | |
2301 | { | |
2302 | struct bpf_func_state *func; | |
2303 | struct bpf_reg_state *reg; | |
2304 | int i, j; | |
2305 | ||
2306 | /* big hammer: mark all scalars precise in this path. | |
2307 | * pop_stack may still get !precise scalars. | |
2308 | */ | |
2309 | for (; st; st = st->parent) | |
2310 | for (i = 0; i <= st->curframe; i++) { | |
2311 | func = st->frame[i]; | |
2312 | for (j = 0; j < BPF_REG_FP; j++) { | |
2313 | reg = &func->regs[j]; | |
2314 | if (reg->type != SCALAR_VALUE) | |
2315 | continue; | |
2316 | reg->precise = true; | |
2317 | } | |
2318 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
2319 | if (func->stack[j].slot_type[0] != STACK_SPILL) | |
2320 | continue; | |
2321 | reg = &func->stack[j].spilled_ptr; | |
2322 | if (reg->type != SCALAR_VALUE) | |
2323 | continue; | |
2324 | reg->precise = true; | |
2325 | } | |
2326 | } | |
2327 | } | |
2328 | ||
a3ce685d AS |
2329 | static int __mark_chain_precision(struct bpf_verifier_env *env, int regno, |
2330 | int spi) | |
b5dc0163 AS |
2331 | { |
2332 | struct bpf_verifier_state *st = env->cur_state; | |
2333 | int first_idx = st->first_insn_idx; | |
2334 | int last_idx = env->insn_idx; | |
2335 | struct bpf_func_state *func; | |
2336 | struct bpf_reg_state *reg; | |
a3ce685d AS |
2337 | u32 reg_mask = regno >= 0 ? 1u << regno : 0; |
2338 | u64 stack_mask = spi >= 0 ? 1ull << spi : 0; | |
b5dc0163 | 2339 | bool skip_first = true; |
a3ce685d | 2340 | bool new_marks = false; |
b5dc0163 AS |
2341 | int i, err; |
2342 | ||
2c78ee89 | 2343 | if (!env->bpf_capable) |
b5dc0163 AS |
2344 | return 0; |
2345 | ||
2346 | func = st->frame[st->curframe]; | |
a3ce685d AS |
2347 | if (regno >= 0) { |
2348 | reg = &func->regs[regno]; | |
2349 | if (reg->type != SCALAR_VALUE) { | |
2350 | WARN_ONCE(1, "backtracing misuse"); | |
2351 | return -EFAULT; | |
2352 | } | |
2353 | if (!reg->precise) | |
2354 | new_marks = true; | |
2355 | else | |
2356 | reg_mask = 0; | |
2357 | reg->precise = true; | |
b5dc0163 | 2358 | } |
b5dc0163 | 2359 | |
a3ce685d AS |
2360 | while (spi >= 0) { |
2361 | if (func->stack[spi].slot_type[0] != STACK_SPILL) { | |
2362 | stack_mask = 0; | |
2363 | break; | |
2364 | } | |
2365 | reg = &func->stack[spi].spilled_ptr; | |
2366 | if (reg->type != SCALAR_VALUE) { | |
2367 | stack_mask = 0; | |
2368 | break; | |
2369 | } | |
2370 | if (!reg->precise) | |
2371 | new_marks = true; | |
2372 | else | |
2373 | stack_mask = 0; | |
2374 | reg->precise = true; | |
2375 | break; | |
2376 | } | |
2377 | ||
2378 | if (!new_marks) | |
2379 | return 0; | |
2380 | if (!reg_mask && !stack_mask) | |
2381 | return 0; | |
b5dc0163 AS |
2382 | for (;;) { |
2383 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
2384 | u32 history = st->jmp_history_cnt; |
2385 | ||
2386 | if (env->log.level & BPF_LOG_LEVEL) | |
2387 | verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx); | |
2388 | for (i = last_idx;;) { | |
2389 | if (skip_first) { | |
2390 | err = 0; | |
2391 | skip_first = false; | |
2392 | } else { | |
2393 | err = backtrack_insn(env, i, ®_mask, &stack_mask); | |
2394 | } | |
2395 | if (err == -ENOTSUPP) { | |
2396 | mark_all_scalars_precise(env, st); | |
2397 | return 0; | |
2398 | } else if (err) { | |
2399 | return err; | |
2400 | } | |
2401 | if (!reg_mask && !stack_mask) | |
2402 | /* Found assignment(s) into tracked register in this state. | |
2403 | * Since this state is already marked, just return. | |
2404 | * Nothing to be tracked further in the parent state. | |
2405 | */ | |
2406 | return 0; | |
2407 | if (i == first_idx) | |
2408 | break; | |
2409 | i = get_prev_insn_idx(st, i, &history); | |
2410 | if (i >= env->prog->len) { | |
2411 | /* This can happen if backtracking reached insn 0 | |
2412 | * and there are still reg_mask or stack_mask | |
2413 | * to backtrack. | |
2414 | * It means the backtracking missed the spot where | |
2415 | * particular register was initialized with a constant. | |
2416 | */ | |
2417 | verbose(env, "BUG backtracking idx %d\n", i); | |
2418 | WARN_ONCE(1, "verifier backtracking bug"); | |
2419 | return -EFAULT; | |
2420 | } | |
2421 | } | |
2422 | st = st->parent; | |
2423 | if (!st) | |
2424 | break; | |
2425 | ||
a3ce685d | 2426 | new_marks = false; |
b5dc0163 AS |
2427 | func = st->frame[st->curframe]; |
2428 | bitmap_from_u64(mask, reg_mask); | |
2429 | for_each_set_bit(i, mask, 32) { | |
2430 | reg = &func->regs[i]; | |
a3ce685d AS |
2431 | if (reg->type != SCALAR_VALUE) { |
2432 | reg_mask &= ~(1u << i); | |
b5dc0163 | 2433 | continue; |
a3ce685d | 2434 | } |
b5dc0163 AS |
2435 | if (!reg->precise) |
2436 | new_marks = true; | |
2437 | reg->precise = true; | |
2438 | } | |
2439 | ||
2440 | bitmap_from_u64(mask, stack_mask); | |
2441 | for_each_set_bit(i, mask, 64) { | |
2442 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
2339cd6c AS |
2443 | /* the sequence of instructions: |
2444 | * 2: (bf) r3 = r10 | |
2445 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
2446 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
2447 | * doesn't contain jmps. It's backtracked | |
2448 | * as a single block. | |
2449 | * During backtracking insn 3 is not recognized as | |
2450 | * stack access, so at the end of backtracking | |
2451 | * stack slot fp-8 is still marked in stack_mask. | |
2452 | * However the parent state may not have accessed | |
2453 | * fp-8 and it's "unallocated" stack space. | |
2454 | * In such case fallback to conservative. | |
b5dc0163 | 2455 | */ |
2339cd6c AS |
2456 | mark_all_scalars_precise(env, st); |
2457 | return 0; | |
b5dc0163 AS |
2458 | } |
2459 | ||
a3ce685d AS |
2460 | if (func->stack[i].slot_type[0] != STACK_SPILL) { |
2461 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2462 | continue; |
a3ce685d | 2463 | } |
b5dc0163 | 2464 | reg = &func->stack[i].spilled_ptr; |
a3ce685d AS |
2465 | if (reg->type != SCALAR_VALUE) { |
2466 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2467 | continue; |
a3ce685d | 2468 | } |
b5dc0163 AS |
2469 | if (!reg->precise) |
2470 | new_marks = true; | |
2471 | reg->precise = true; | |
2472 | } | |
2473 | if (env->log.level & BPF_LOG_LEVEL) { | |
2474 | print_verifier_state(env, func); | |
2475 | verbose(env, "parent %s regs=%x stack=%llx marks\n", | |
2476 | new_marks ? "didn't have" : "already had", | |
2477 | reg_mask, stack_mask); | |
2478 | } | |
2479 | ||
a3ce685d AS |
2480 | if (!reg_mask && !stack_mask) |
2481 | break; | |
b5dc0163 AS |
2482 | if (!new_marks) |
2483 | break; | |
2484 | ||
2485 | last_idx = st->last_insn_idx; | |
2486 | first_idx = st->first_insn_idx; | |
2487 | } | |
2488 | return 0; | |
2489 | } | |
2490 | ||
a3ce685d AS |
2491 | static int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
2492 | { | |
2493 | return __mark_chain_precision(env, regno, -1); | |
2494 | } | |
2495 | ||
2496 | static int mark_chain_precision_stack(struct bpf_verifier_env *env, int spi) | |
2497 | { | |
2498 | return __mark_chain_precision(env, -1, spi); | |
2499 | } | |
b5dc0163 | 2500 | |
1be7f75d AS |
2501 | static bool is_spillable_regtype(enum bpf_reg_type type) |
2502 | { | |
2503 | switch (type) { | |
2504 | case PTR_TO_MAP_VALUE: | |
2505 | case PTR_TO_MAP_VALUE_OR_NULL: | |
2506 | case PTR_TO_STACK: | |
2507 | case PTR_TO_CTX: | |
969bf05e | 2508 | case PTR_TO_PACKET: |
de8f3a83 | 2509 | case PTR_TO_PACKET_META: |
969bf05e | 2510 | case PTR_TO_PACKET_END: |
d58e468b | 2511 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 2512 | case CONST_PTR_TO_MAP: |
c64b7983 JS |
2513 | case PTR_TO_SOCKET: |
2514 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
2515 | case PTR_TO_SOCK_COMMON: |
2516 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
2517 | case PTR_TO_TCP_SOCK: |
2518 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 2519 | case PTR_TO_XDP_SOCK: |
65726b5b | 2520 | case PTR_TO_BTF_ID: |
b121b341 | 2521 | case PTR_TO_BTF_ID_OR_NULL: |
afbf21dc YS |
2522 | case PTR_TO_RDONLY_BUF: |
2523 | case PTR_TO_RDONLY_BUF_OR_NULL: | |
2524 | case PTR_TO_RDWR_BUF: | |
2525 | case PTR_TO_RDWR_BUF_OR_NULL: | |
eaa6bcb7 | 2526 | case PTR_TO_PERCPU_BTF_ID: |
744ea4e3 GR |
2527 | case PTR_TO_MEM: |
2528 | case PTR_TO_MEM_OR_NULL: | |
69c087ba YS |
2529 | case PTR_TO_FUNC: |
2530 | case PTR_TO_MAP_KEY: | |
1be7f75d AS |
2531 | return true; |
2532 | default: | |
2533 | return false; | |
2534 | } | |
2535 | } | |
2536 | ||
cc2b14d5 AS |
2537 | /* Does this register contain a constant zero? */ |
2538 | static bool register_is_null(struct bpf_reg_state *reg) | |
2539 | { | |
2540 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
2541 | } | |
2542 | ||
f7cf25b2 AS |
2543 | static bool register_is_const(struct bpf_reg_state *reg) |
2544 | { | |
2545 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
2546 | } | |
2547 | ||
5689d49b YS |
2548 | static bool __is_scalar_unbounded(struct bpf_reg_state *reg) |
2549 | { | |
2550 | return tnum_is_unknown(reg->var_off) && | |
2551 | reg->smin_value == S64_MIN && reg->smax_value == S64_MAX && | |
2552 | reg->umin_value == 0 && reg->umax_value == U64_MAX && | |
2553 | reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX && | |
2554 | reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX; | |
2555 | } | |
2556 | ||
2557 | static bool register_is_bounded(struct bpf_reg_state *reg) | |
2558 | { | |
2559 | return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg); | |
2560 | } | |
2561 | ||
6e7e63cb JH |
2562 | static bool __is_pointer_value(bool allow_ptr_leaks, |
2563 | const struct bpf_reg_state *reg) | |
2564 | { | |
2565 | if (allow_ptr_leaks) | |
2566 | return false; | |
2567 | ||
2568 | return reg->type != SCALAR_VALUE; | |
2569 | } | |
2570 | ||
f7cf25b2 AS |
2571 | static void save_register_state(struct bpf_func_state *state, |
2572 | int spi, struct bpf_reg_state *reg) | |
2573 | { | |
2574 | int i; | |
2575 | ||
2576 | state->stack[spi].spilled_ptr = *reg; | |
2577 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2578 | ||
2579 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2580 | state->stack[spi].slot_type[i] = STACK_SPILL; | |
2581 | } | |
2582 | ||
01f810ac | 2583 | /* check_stack_{read,write}_fixed_off functions track spill/fill of registers, |
17a52670 AS |
2584 | * stack boundary and alignment are checked in check_mem_access() |
2585 | */ | |
01f810ac AM |
2586 | static int check_stack_write_fixed_off(struct bpf_verifier_env *env, |
2587 | /* stack frame we're writing to */ | |
2588 | struct bpf_func_state *state, | |
2589 | int off, int size, int value_regno, | |
2590 | int insn_idx) | |
17a52670 | 2591 | { |
f4d7e40a | 2592 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 2593 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
b5dc0163 | 2594 | u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg; |
f7cf25b2 | 2595 | struct bpf_reg_state *reg = NULL; |
638f5b90 | 2596 | |
c69431aa | 2597 | err = grow_stack_state(state, round_up(slot + 1, BPF_REG_SIZE)); |
638f5b90 AS |
2598 | if (err) |
2599 | return err; | |
9c399760 AS |
2600 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
2601 | * so it's aligned access and [off, off + size) are within stack limits | |
2602 | */ | |
638f5b90 AS |
2603 | if (!env->allow_ptr_leaks && |
2604 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
2605 | size != BPF_REG_SIZE) { | |
2606 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
2607 | return -EACCES; | |
2608 | } | |
17a52670 | 2609 | |
f4d7e40a | 2610 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
2611 | if (value_regno >= 0) |
2612 | reg = &cur->regs[value_regno]; | |
17a52670 | 2613 | |
5689d49b | 2614 | if (reg && size == BPF_REG_SIZE && register_is_bounded(reg) && |
2c78ee89 | 2615 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
2616 | if (dst_reg != BPF_REG_FP) { |
2617 | /* The backtracking logic can only recognize explicit | |
2618 | * stack slot address like [fp - 8]. Other spill of | |
8fb33b60 | 2619 | * scalar via different register has to be conservative. |
b5dc0163 AS |
2620 | * Backtrack from here and mark all registers as precise |
2621 | * that contributed into 'reg' being a constant. | |
2622 | */ | |
2623 | err = mark_chain_precision(env, value_regno); | |
2624 | if (err) | |
2625 | return err; | |
2626 | } | |
f7cf25b2 AS |
2627 | save_register_state(state, spi, reg); |
2628 | } else if (reg && is_spillable_regtype(reg->type)) { | |
17a52670 | 2629 | /* register containing pointer is being spilled into stack */ |
9c399760 | 2630 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 2631 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 2632 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
2633 | return -EACCES; |
2634 | } | |
2635 | ||
f7cf25b2 | 2636 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
2637 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
2638 | return -EINVAL; | |
2639 | } | |
2640 | ||
2c78ee89 | 2641 | if (!env->bypass_spec_v4) { |
f7cf25b2 | 2642 | bool sanitize = false; |
17a52670 | 2643 | |
f7cf25b2 AS |
2644 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
2645 | register_is_const(&state->stack[spi].spilled_ptr)) | |
2646 | sanitize = true; | |
2647 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2648 | if (state->stack[spi].slot_type[i] == STACK_MISC) { | |
2649 | sanitize = true; | |
2650 | break; | |
2651 | } | |
2652 | if (sanitize) { | |
af86ca4e AS |
2653 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; |
2654 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
2655 | ||
2656 | /* detected reuse of integer stack slot with a pointer | |
2657 | * which means either llvm is reusing stack slot or | |
2658 | * an attacker is trying to exploit CVE-2018-3639 | |
2659 | * (speculative store bypass) | |
2660 | * Have to sanitize that slot with preemptive | |
2661 | * store of zero. | |
2662 | */ | |
2663 | if (*poff && *poff != soff) { | |
2664 | /* disallow programs where single insn stores | |
2665 | * into two different stack slots, since verifier | |
2666 | * cannot sanitize them | |
2667 | */ | |
2668 | verbose(env, | |
2669 | "insn %d cannot access two stack slots fp%d and fp%d", | |
2670 | insn_idx, *poff, soff); | |
2671 | return -EINVAL; | |
2672 | } | |
2673 | *poff = soff; | |
2674 | } | |
af86ca4e | 2675 | } |
f7cf25b2 | 2676 | save_register_state(state, spi, reg); |
9c399760 | 2677 | } else { |
cc2b14d5 AS |
2678 | u8 type = STACK_MISC; |
2679 | ||
679c782d EC |
2680 | /* regular write of data into stack destroys any spilled ptr */ |
2681 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d JW |
2682 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
2683 | if (state->stack[spi].slot_type[0] == STACK_SPILL) | |
2684 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2685 | state->stack[spi].slot_type[i] = STACK_MISC; | |
9c399760 | 2686 | |
cc2b14d5 AS |
2687 | /* only mark the slot as written if all 8 bytes were written |
2688 | * otherwise read propagation may incorrectly stop too soon | |
2689 | * when stack slots are partially written. | |
2690 | * This heuristic means that read propagation will be | |
2691 | * conservative, since it will add reg_live_read marks | |
2692 | * to stack slots all the way to first state when programs | |
2693 | * writes+reads less than 8 bytes | |
2694 | */ | |
2695 | if (size == BPF_REG_SIZE) | |
2696 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2697 | ||
2698 | /* when we zero initialize stack slots mark them as such */ | |
b5dc0163 AS |
2699 | if (reg && register_is_null(reg)) { |
2700 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
2701 | err = mark_chain_precision(env, value_regno); | |
2702 | if (err) | |
2703 | return err; | |
cc2b14d5 | 2704 | type = STACK_ZERO; |
b5dc0163 | 2705 | } |
cc2b14d5 | 2706 | |
0bae2d4d | 2707 | /* Mark slots affected by this stack write. */ |
9c399760 | 2708 | for (i = 0; i < size; i++) |
638f5b90 | 2709 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 2710 | type; |
17a52670 AS |
2711 | } |
2712 | return 0; | |
2713 | } | |
2714 | ||
01f810ac AM |
2715 | /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is |
2716 | * known to contain a variable offset. | |
2717 | * This function checks whether the write is permitted and conservatively | |
2718 | * tracks the effects of the write, considering that each stack slot in the | |
2719 | * dynamic range is potentially written to. | |
2720 | * | |
2721 | * 'off' includes 'regno->off'. | |
2722 | * 'value_regno' can be -1, meaning that an unknown value is being written to | |
2723 | * the stack. | |
2724 | * | |
2725 | * Spilled pointers in range are not marked as written because we don't know | |
2726 | * what's going to be actually written. This means that read propagation for | |
2727 | * future reads cannot be terminated by this write. | |
2728 | * | |
2729 | * For privileged programs, uninitialized stack slots are considered | |
2730 | * initialized by this write (even though we don't know exactly what offsets | |
2731 | * are going to be written to). The idea is that we don't want the verifier to | |
2732 | * reject future reads that access slots written to through variable offsets. | |
2733 | */ | |
2734 | static int check_stack_write_var_off(struct bpf_verifier_env *env, | |
2735 | /* func where register points to */ | |
2736 | struct bpf_func_state *state, | |
2737 | int ptr_regno, int off, int size, | |
2738 | int value_regno, int insn_idx) | |
2739 | { | |
2740 | struct bpf_func_state *cur; /* state of the current function */ | |
2741 | int min_off, max_off; | |
2742 | int i, err; | |
2743 | struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL; | |
2744 | bool writing_zero = false; | |
2745 | /* set if the fact that we're writing a zero is used to let any | |
2746 | * stack slots remain STACK_ZERO | |
2747 | */ | |
2748 | bool zero_used = false; | |
2749 | ||
2750 | cur = env->cur_state->frame[env->cur_state->curframe]; | |
2751 | ptr_reg = &cur->regs[ptr_regno]; | |
2752 | min_off = ptr_reg->smin_value + off; | |
2753 | max_off = ptr_reg->smax_value + off + size; | |
2754 | if (value_regno >= 0) | |
2755 | value_reg = &cur->regs[value_regno]; | |
2756 | if (value_reg && register_is_null(value_reg)) | |
2757 | writing_zero = true; | |
2758 | ||
c69431aa | 2759 | err = grow_stack_state(state, round_up(-min_off, BPF_REG_SIZE)); |
01f810ac AM |
2760 | if (err) |
2761 | return err; | |
2762 | ||
2763 | ||
2764 | /* Variable offset writes destroy any spilled pointers in range. */ | |
2765 | for (i = min_off; i < max_off; i++) { | |
2766 | u8 new_type, *stype; | |
2767 | int slot, spi; | |
2768 | ||
2769 | slot = -i - 1; | |
2770 | spi = slot / BPF_REG_SIZE; | |
2771 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
2772 | ||
2773 | if (!env->allow_ptr_leaks | |
2774 | && *stype != NOT_INIT | |
2775 | && *stype != SCALAR_VALUE) { | |
2776 | /* Reject the write if there's are spilled pointers in | |
2777 | * range. If we didn't reject here, the ptr status | |
2778 | * would be erased below (even though not all slots are | |
2779 | * actually overwritten), possibly opening the door to | |
2780 | * leaks. | |
2781 | */ | |
2782 | verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d", | |
2783 | insn_idx, i); | |
2784 | return -EINVAL; | |
2785 | } | |
2786 | ||
2787 | /* Erase all spilled pointers. */ | |
2788 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
2789 | ||
2790 | /* Update the slot type. */ | |
2791 | new_type = STACK_MISC; | |
2792 | if (writing_zero && *stype == STACK_ZERO) { | |
2793 | new_type = STACK_ZERO; | |
2794 | zero_used = true; | |
2795 | } | |
2796 | /* If the slot is STACK_INVALID, we check whether it's OK to | |
2797 | * pretend that it will be initialized by this write. The slot | |
2798 | * might not actually be written to, and so if we mark it as | |
2799 | * initialized future reads might leak uninitialized memory. | |
2800 | * For privileged programs, we will accept such reads to slots | |
2801 | * that may or may not be written because, if we're reject | |
2802 | * them, the error would be too confusing. | |
2803 | */ | |
2804 | if (*stype == STACK_INVALID && !env->allow_uninit_stack) { | |
2805 | verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d", | |
2806 | insn_idx, i); | |
2807 | return -EINVAL; | |
2808 | } | |
2809 | *stype = new_type; | |
2810 | } | |
2811 | if (zero_used) { | |
2812 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
2813 | err = mark_chain_precision(env, value_regno); | |
2814 | if (err) | |
2815 | return err; | |
2816 | } | |
2817 | return 0; | |
2818 | } | |
2819 | ||
2820 | /* When register 'dst_regno' is assigned some values from stack[min_off, | |
2821 | * max_off), we set the register's type according to the types of the | |
2822 | * respective stack slots. If all the stack values are known to be zeros, then | |
2823 | * so is the destination reg. Otherwise, the register is considered to be | |
2824 | * SCALAR. This function does not deal with register filling; the caller must | |
2825 | * ensure that all spilled registers in the stack range have been marked as | |
2826 | * read. | |
2827 | */ | |
2828 | static void mark_reg_stack_read(struct bpf_verifier_env *env, | |
2829 | /* func where src register points to */ | |
2830 | struct bpf_func_state *ptr_state, | |
2831 | int min_off, int max_off, int dst_regno) | |
2832 | { | |
2833 | struct bpf_verifier_state *vstate = env->cur_state; | |
2834 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
2835 | int i, slot, spi; | |
2836 | u8 *stype; | |
2837 | int zeros = 0; | |
2838 | ||
2839 | for (i = min_off; i < max_off; i++) { | |
2840 | slot = -i - 1; | |
2841 | spi = slot / BPF_REG_SIZE; | |
2842 | stype = ptr_state->stack[spi].slot_type; | |
2843 | if (stype[slot % BPF_REG_SIZE] != STACK_ZERO) | |
2844 | break; | |
2845 | zeros++; | |
2846 | } | |
2847 | if (zeros == max_off - min_off) { | |
2848 | /* any access_size read into register is zero extended, | |
2849 | * so the whole register == const_zero | |
2850 | */ | |
2851 | __mark_reg_const_zero(&state->regs[dst_regno]); | |
2852 | /* backtracking doesn't support STACK_ZERO yet, | |
2853 | * so mark it precise here, so that later | |
2854 | * backtracking can stop here. | |
2855 | * Backtracking may not need this if this register | |
2856 | * doesn't participate in pointer adjustment. | |
2857 | * Forward propagation of precise flag is not | |
2858 | * necessary either. This mark is only to stop | |
2859 | * backtracking. Any register that contributed | |
2860 | * to const 0 was marked precise before spill. | |
2861 | */ | |
2862 | state->regs[dst_regno].precise = true; | |
2863 | } else { | |
2864 | /* have read misc data from the stack */ | |
2865 | mark_reg_unknown(env, state->regs, dst_regno); | |
2866 | } | |
2867 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
2868 | } | |
2869 | ||
2870 | /* Read the stack at 'off' and put the results into the register indicated by | |
2871 | * 'dst_regno'. It handles reg filling if the addressed stack slot is a | |
2872 | * spilled reg. | |
2873 | * | |
2874 | * 'dst_regno' can be -1, meaning that the read value is not going to a | |
2875 | * register. | |
2876 | * | |
2877 | * The access is assumed to be within the current stack bounds. | |
2878 | */ | |
2879 | static int check_stack_read_fixed_off(struct bpf_verifier_env *env, | |
2880 | /* func where src register points to */ | |
2881 | struct bpf_func_state *reg_state, | |
2882 | int off, int size, int dst_regno) | |
17a52670 | 2883 | { |
f4d7e40a AS |
2884 | struct bpf_verifier_state *vstate = env->cur_state; |
2885 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 2886 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 2887 | struct bpf_reg_state *reg; |
638f5b90 | 2888 | u8 *stype; |
17a52670 | 2889 | |
f4d7e40a | 2890 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 2891 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 2892 | |
638f5b90 | 2893 | if (stype[0] == STACK_SPILL) { |
9c399760 | 2894 | if (size != BPF_REG_SIZE) { |
f7cf25b2 AS |
2895 | if (reg->type != SCALAR_VALUE) { |
2896 | verbose_linfo(env, env->insn_idx, "; "); | |
2897 | verbose(env, "invalid size of register fill\n"); | |
2898 | return -EACCES; | |
2899 | } | |
01f810ac AM |
2900 | if (dst_regno >= 0) { |
2901 | mark_reg_unknown(env, state->regs, dst_regno); | |
2902 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
f7cf25b2 AS |
2903 | } |
2904 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
2905 | return 0; | |
17a52670 | 2906 | } |
9c399760 | 2907 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 2908 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 2909 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
2910 | return -EACCES; |
2911 | } | |
2912 | } | |
2913 | ||
01f810ac | 2914 | if (dst_regno >= 0) { |
17a52670 | 2915 | /* restore register state from stack */ |
01f810ac | 2916 | state->regs[dst_regno] = *reg; |
2f18f62e AS |
2917 | /* mark reg as written since spilled pointer state likely |
2918 | * has its liveness marks cleared by is_state_visited() | |
2919 | * which resets stack/reg liveness for state transitions | |
2920 | */ | |
01f810ac | 2921 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
6e7e63cb | 2922 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
01f810ac | 2923 | /* If dst_regno==-1, the caller is asking us whether |
6e7e63cb JH |
2924 | * it is acceptable to use this value as a SCALAR_VALUE |
2925 | * (e.g. for XADD). | |
2926 | * We must not allow unprivileged callers to do that | |
2927 | * with spilled pointers. | |
2928 | */ | |
2929 | verbose(env, "leaking pointer from stack off %d\n", | |
2930 | off); | |
2931 | return -EACCES; | |
dc503a8a | 2932 | } |
f7cf25b2 | 2933 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 | 2934 | } else { |
01f810ac | 2935 | u8 type; |
cc2b14d5 | 2936 | |
17a52670 | 2937 | for (i = 0; i < size; i++) { |
01f810ac AM |
2938 | type = stype[(slot - i) % BPF_REG_SIZE]; |
2939 | if (type == STACK_MISC) | |
cc2b14d5 | 2940 | continue; |
01f810ac | 2941 | if (type == STACK_ZERO) |
cc2b14d5 | 2942 | continue; |
cc2b14d5 AS |
2943 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
2944 | off, i, size); | |
2945 | return -EACCES; | |
2946 | } | |
f7cf25b2 | 2947 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
01f810ac AM |
2948 | if (dst_regno >= 0) |
2949 | mark_reg_stack_read(env, reg_state, off, off + size, dst_regno); | |
17a52670 | 2950 | } |
f7cf25b2 | 2951 | return 0; |
17a52670 AS |
2952 | } |
2953 | ||
01f810ac AM |
2954 | enum stack_access_src { |
2955 | ACCESS_DIRECT = 1, /* the access is performed by an instruction */ | |
2956 | ACCESS_HELPER = 2, /* the access is performed by a helper */ | |
2957 | }; | |
2958 | ||
2959 | static int check_stack_range_initialized(struct bpf_verifier_env *env, | |
2960 | int regno, int off, int access_size, | |
2961 | bool zero_size_allowed, | |
2962 | enum stack_access_src type, | |
2963 | struct bpf_call_arg_meta *meta); | |
2964 | ||
2965 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) | |
2966 | { | |
2967 | return cur_regs(env) + regno; | |
2968 | } | |
2969 | ||
2970 | /* Read the stack at 'ptr_regno + off' and put the result into the register | |
2971 | * 'dst_regno'. | |
2972 | * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'), | |
2973 | * but not its variable offset. | |
2974 | * 'size' is assumed to be <= reg size and the access is assumed to be aligned. | |
2975 | * | |
2976 | * As opposed to check_stack_read_fixed_off, this function doesn't deal with | |
2977 | * filling registers (i.e. reads of spilled register cannot be detected when | |
2978 | * the offset is not fixed). We conservatively mark 'dst_regno' as containing | |
2979 | * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable | |
2980 | * offset; for a fixed offset check_stack_read_fixed_off should be used | |
2981 | * instead. | |
2982 | */ | |
2983 | static int check_stack_read_var_off(struct bpf_verifier_env *env, | |
2984 | int ptr_regno, int off, int size, int dst_regno) | |
e4298d25 | 2985 | { |
01f810ac AM |
2986 | /* The state of the source register. */ |
2987 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
2988 | struct bpf_func_state *ptr_state = func(env, reg); | |
2989 | int err; | |
2990 | int min_off, max_off; | |
2991 | ||
2992 | /* Note that we pass a NULL meta, so raw access will not be permitted. | |
e4298d25 | 2993 | */ |
01f810ac AM |
2994 | err = check_stack_range_initialized(env, ptr_regno, off, size, |
2995 | false, ACCESS_DIRECT, NULL); | |
2996 | if (err) | |
2997 | return err; | |
2998 | ||
2999 | min_off = reg->smin_value + off; | |
3000 | max_off = reg->smax_value + off; | |
3001 | mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno); | |
3002 | return 0; | |
3003 | } | |
3004 | ||
3005 | /* check_stack_read dispatches to check_stack_read_fixed_off or | |
3006 | * check_stack_read_var_off. | |
3007 | * | |
3008 | * The caller must ensure that the offset falls within the allocated stack | |
3009 | * bounds. | |
3010 | * | |
3011 | * 'dst_regno' is a register which will receive the value from the stack. It | |
3012 | * can be -1, meaning that the read value is not going to a register. | |
3013 | */ | |
3014 | static int check_stack_read(struct bpf_verifier_env *env, | |
3015 | int ptr_regno, int off, int size, | |
3016 | int dst_regno) | |
3017 | { | |
3018 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
3019 | struct bpf_func_state *state = func(env, reg); | |
3020 | int err; | |
3021 | /* Some accesses are only permitted with a static offset. */ | |
3022 | bool var_off = !tnum_is_const(reg->var_off); | |
3023 | ||
3024 | /* The offset is required to be static when reads don't go to a | |
3025 | * register, in order to not leak pointers (see | |
3026 | * check_stack_read_fixed_off). | |
3027 | */ | |
3028 | if (dst_regno < 0 && var_off) { | |
e4298d25 DB |
3029 | char tn_buf[48]; |
3030 | ||
3031 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac | 3032 | verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n", |
e4298d25 DB |
3033 | tn_buf, off, size); |
3034 | return -EACCES; | |
3035 | } | |
01f810ac AM |
3036 | /* Variable offset is prohibited for unprivileged mode for simplicity |
3037 | * since it requires corresponding support in Spectre masking for stack | |
3038 | * ALU. See also retrieve_ptr_limit(). | |
3039 | */ | |
3040 | if (!env->bypass_spec_v1 && var_off) { | |
3041 | char tn_buf[48]; | |
e4298d25 | 3042 | |
01f810ac AM |
3043 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
3044 | verbose(env, "R%d variable offset stack access prohibited for !root, var_off=%s\n", | |
3045 | ptr_regno, tn_buf); | |
e4298d25 DB |
3046 | return -EACCES; |
3047 | } | |
3048 | ||
01f810ac AM |
3049 | if (!var_off) { |
3050 | off += reg->var_off.value; | |
3051 | err = check_stack_read_fixed_off(env, state, off, size, | |
3052 | dst_regno); | |
3053 | } else { | |
3054 | /* Variable offset stack reads need more conservative handling | |
3055 | * than fixed offset ones. Note that dst_regno >= 0 on this | |
3056 | * branch. | |
3057 | */ | |
3058 | err = check_stack_read_var_off(env, ptr_regno, off, size, | |
3059 | dst_regno); | |
3060 | } | |
3061 | return err; | |
3062 | } | |
3063 | ||
3064 | ||
3065 | /* check_stack_write dispatches to check_stack_write_fixed_off or | |
3066 | * check_stack_write_var_off. | |
3067 | * | |
3068 | * 'ptr_regno' is the register used as a pointer into the stack. | |
3069 | * 'off' includes 'ptr_regno->off', but not its variable offset (if any). | |
3070 | * 'value_regno' is the register whose value we're writing to the stack. It can | |
3071 | * be -1, meaning that we're not writing from a register. | |
3072 | * | |
3073 | * The caller must ensure that the offset falls within the maximum stack size. | |
3074 | */ | |
3075 | static int check_stack_write(struct bpf_verifier_env *env, | |
3076 | int ptr_regno, int off, int size, | |
3077 | int value_regno, int insn_idx) | |
3078 | { | |
3079 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
3080 | struct bpf_func_state *state = func(env, reg); | |
3081 | int err; | |
3082 | ||
3083 | if (tnum_is_const(reg->var_off)) { | |
3084 | off += reg->var_off.value; | |
3085 | err = check_stack_write_fixed_off(env, state, off, size, | |
3086 | value_regno, insn_idx); | |
3087 | } else { | |
3088 | /* Variable offset stack reads need more conservative handling | |
3089 | * than fixed offset ones. | |
3090 | */ | |
3091 | err = check_stack_write_var_off(env, state, | |
3092 | ptr_regno, off, size, | |
3093 | value_regno, insn_idx); | |
3094 | } | |
3095 | return err; | |
e4298d25 DB |
3096 | } |
3097 | ||
591fe988 DB |
3098 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
3099 | int off, int size, enum bpf_access_type type) | |
3100 | { | |
3101 | struct bpf_reg_state *regs = cur_regs(env); | |
3102 | struct bpf_map *map = regs[regno].map_ptr; | |
3103 | u32 cap = bpf_map_flags_to_cap(map); | |
3104 | ||
3105 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
3106 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
3107 | map->value_size, off, size); | |
3108 | return -EACCES; | |
3109 | } | |
3110 | ||
3111 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
3112 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
3113 | map->value_size, off, size); | |
3114 | return -EACCES; | |
3115 | } | |
3116 | ||
3117 | return 0; | |
3118 | } | |
3119 | ||
457f4436 AN |
3120 | /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */ |
3121 | static int __check_mem_access(struct bpf_verifier_env *env, int regno, | |
3122 | int off, int size, u32 mem_size, | |
3123 | bool zero_size_allowed) | |
17a52670 | 3124 | { |
457f4436 AN |
3125 | bool size_ok = size > 0 || (size == 0 && zero_size_allowed); |
3126 | struct bpf_reg_state *reg; | |
3127 | ||
3128 | if (off >= 0 && size_ok && (u64)off + size <= mem_size) | |
3129 | return 0; | |
17a52670 | 3130 | |
457f4436 AN |
3131 | reg = &cur_regs(env)[regno]; |
3132 | switch (reg->type) { | |
69c087ba YS |
3133 | case PTR_TO_MAP_KEY: |
3134 | verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n", | |
3135 | mem_size, off, size); | |
3136 | break; | |
457f4436 | 3137 | case PTR_TO_MAP_VALUE: |
61bd5218 | 3138 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
457f4436 AN |
3139 | mem_size, off, size); |
3140 | break; | |
3141 | case PTR_TO_PACKET: | |
3142 | case PTR_TO_PACKET_META: | |
3143 | case PTR_TO_PACKET_END: | |
3144 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", | |
3145 | off, size, regno, reg->id, off, mem_size); | |
3146 | break; | |
3147 | case PTR_TO_MEM: | |
3148 | default: | |
3149 | verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n", | |
3150 | mem_size, off, size); | |
17a52670 | 3151 | } |
457f4436 AN |
3152 | |
3153 | return -EACCES; | |
17a52670 AS |
3154 | } |
3155 | ||
457f4436 AN |
3156 | /* check read/write into a memory region with possible variable offset */ |
3157 | static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno, | |
3158 | int off, int size, u32 mem_size, | |
3159 | bool zero_size_allowed) | |
dbcfe5f7 | 3160 | { |
f4d7e40a AS |
3161 | struct bpf_verifier_state *vstate = env->cur_state; |
3162 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
3163 | struct bpf_reg_state *reg = &state->regs[regno]; |
3164 | int err; | |
3165 | ||
457f4436 | 3166 | /* We may have adjusted the register pointing to memory region, so we |
f1174f77 EC |
3167 | * need to try adding each of min_value and max_value to off |
3168 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 3169 | */ |
06ee7115 | 3170 | if (env->log.level & BPF_LOG_LEVEL) |
61bd5218 | 3171 | print_verifier_state(env, state); |
b7137c4e | 3172 | |
dbcfe5f7 GB |
3173 | /* The minimum value is only important with signed |
3174 | * comparisons where we can't assume the floor of a | |
3175 | * value is 0. If we are using signed variables for our | |
3176 | * index'es we need to make sure that whatever we use | |
3177 | * will have a set floor within our range. | |
3178 | */ | |
b7137c4e DB |
3179 | if (reg->smin_value < 0 && |
3180 | (reg->smin_value == S64_MIN || | |
3181 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
3182 | reg->smin_value + off < 0)) { | |
61bd5218 | 3183 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
3184 | regno); |
3185 | return -EACCES; | |
3186 | } | |
457f4436 AN |
3187 | err = __check_mem_access(env, regno, reg->smin_value + off, size, |
3188 | mem_size, zero_size_allowed); | |
dbcfe5f7 | 3189 | if (err) { |
457f4436 | 3190 | verbose(env, "R%d min value is outside of the allowed memory range\n", |
61bd5218 | 3191 | regno); |
dbcfe5f7 GB |
3192 | return err; |
3193 | } | |
3194 | ||
b03c9f9f EC |
3195 | /* If we haven't set a max value then we need to bail since we can't be |
3196 | * sure we won't do bad things. | |
3197 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 3198 | */ |
b03c9f9f | 3199 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
457f4436 | 3200 | verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n", |
dbcfe5f7 GB |
3201 | regno); |
3202 | return -EACCES; | |
3203 | } | |
457f4436 AN |
3204 | err = __check_mem_access(env, regno, reg->umax_value + off, size, |
3205 | mem_size, zero_size_allowed); | |
3206 | if (err) { | |
3207 | verbose(env, "R%d max value is outside of the allowed memory range\n", | |
61bd5218 | 3208 | regno); |
457f4436 AN |
3209 | return err; |
3210 | } | |
3211 | ||
3212 | return 0; | |
3213 | } | |
d83525ca | 3214 | |
457f4436 AN |
3215 | /* check read/write into a map element with possible variable offset */ |
3216 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
3217 | int off, int size, bool zero_size_allowed) | |
3218 | { | |
3219 | struct bpf_verifier_state *vstate = env->cur_state; | |
3220 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3221 | struct bpf_reg_state *reg = &state->regs[regno]; | |
3222 | struct bpf_map *map = reg->map_ptr; | |
3223 | int err; | |
3224 | ||
3225 | err = check_mem_region_access(env, regno, off, size, map->value_size, | |
3226 | zero_size_allowed); | |
3227 | if (err) | |
3228 | return err; | |
3229 | ||
3230 | if (map_value_has_spin_lock(map)) { | |
3231 | u32 lock = map->spin_lock_off; | |
d83525ca AS |
3232 | |
3233 | /* if any part of struct bpf_spin_lock can be touched by | |
3234 | * load/store reject this program. | |
3235 | * To check that [x1, x2) overlaps with [y1, y2) | |
3236 | * it is sufficient to check x1 < y2 && y1 < x2. | |
3237 | */ | |
3238 | if (reg->smin_value + off < lock + sizeof(struct bpf_spin_lock) && | |
3239 | lock < reg->umax_value + off + size) { | |
3240 | verbose(env, "bpf_spin_lock cannot be accessed directly by load/store\n"); | |
3241 | return -EACCES; | |
3242 | } | |
3243 | } | |
f1174f77 | 3244 | return err; |
dbcfe5f7 GB |
3245 | } |
3246 | ||
969bf05e AS |
3247 | #define MAX_PACKET_OFF 0xffff |
3248 | ||
7e40781c UP |
3249 | static enum bpf_prog_type resolve_prog_type(struct bpf_prog *prog) |
3250 | { | |
3aac1ead | 3251 | return prog->aux->dst_prog ? prog->aux->dst_prog->type : prog->type; |
7e40781c UP |
3252 | } |
3253 | ||
58e2af8b | 3254 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
3255 | const struct bpf_call_arg_meta *meta, |
3256 | enum bpf_access_type t) | |
4acf6c0b | 3257 | { |
7e40781c UP |
3258 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
3259 | ||
3260 | switch (prog_type) { | |
5d66fa7d | 3261 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
3262 | case BPF_PROG_TYPE_LWT_IN: |
3263 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 3264 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 3265 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 3266 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 3267 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
3268 | if (t == BPF_WRITE) |
3269 | return false; | |
8731745e | 3270 | fallthrough; |
5d66fa7d DB |
3271 | |
3272 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
3273 | case BPF_PROG_TYPE_SCHED_CLS: |
3274 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 3275 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 3276 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 3277 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 3278 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
3279 | if (meta) |
3280 | return meta->pkt_access; | |
3281 | ||
3282 | env->seen_direct_write = true; | |
4acf6c0b | 3283 | return true; |
0d01da6a SF |
3284 | |
3285 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
3286 | if (t == BPF_WRITE) | |
3287 | env->seen_direct_write = true; | |
3288 | ||
3289 | return true; | |
3290 | ||
4acf6c0b BB |
3291 | default: |
3292 | return false; | |
3293 | } | |
3294 | } | |
3295 | ||
f1174f77 | 3296 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 3297 | int size, bool zero_size_allowed) |
f1174f77 | 3298 | { |
638f5b90 | 3299 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
3300 | struct bpf_reg_state *reg = ®s[regno]; |
3301 | int err; | |
3302 | ||
3303 | /* We may have added a variable offset to the packet pointer; but any | |
3304 | * reg->range we have comes after that. We are only checking the fixed | |
3305 | * offset. | |
3306 | */ | |
3307 | ||
3308 | /* We don't allow negative numbers, because we aren't tracking enough | |
3309 | * detail to prove they're safe. | |
3310 | */ | |
b03c9f9f | 3311 | if (reg->smin_value < 0) { |
61bd5218 | 3312 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
3313 | regno); |
3314 | return -EACCES; | |
3315 | } | |
6d94e741 AS |
3316 | |
3317 | err = reg->range < 0 ? -EINVAL : | |
3318 | __check_mem_access(env, regno, off, size, reg->range, | |
457f4436 | 3319 | zero_size_allowed); |
f1174f77 | 3320 | if (err) { |
61bd5218 | 3321 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
3322 | return err; |
3323 | } | |
e647815a | 3324 | |
457f4436 | 3325 | /* __check_mem_access has made sure "off + size - 1" is within u16. |
e647815a JW |
3326 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, |
3327 | * otherwise find_good_pkt_pointers would have refused to set range info | |
457f4436 | 3328 | * that __check_mem_access would have rejected this pkt access. |
e647815a JW |
3329 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. |
3330 | */ | |
3331 | env->prog->aux->max_pkt_offset = | |
3332 | max_t(u32, env->prog->aux->max_pkt_offset, | |
3333 | off + reg->umax_value + size - 1); | |
3334 | ||
f1174f77 EC |
3335 | return err; |
3336 | } | |
3337 | ||
3338 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 3339 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 | 3340 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
22dc4a0f | 3341 | struct btf **btf, u32 *btf_id) |
17a52670 | 3342 | { |
f96da094 DB |
3343 | struct bpf_insn_access_aux info = { |
3344 | .reg_type = *reg_type, | |
9e15db66 | 3345 | .log = &env->log, |
f96da094 | 3346 | }; |
31fd8581 | 3347 | |
4f9218aa | 3348 | if (env->ops->is_valid_access && |
5e43f899 | 3349 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
3350 | /* A non zero info.ctx_field_size indicates that this field is a |
3351 | * candidate for later verifier transformation to load the whole | |
3352 | * field and then apply a mask when accessed with a narrower | |
3353 | * access than actual ctx access size. A zero info.ctx_field_size | |
3354 | * will only allow for whole field access and rejects any other | |
3355 | * type of narrower access. | |
31fd8581 | 3356 | */ |
23994631 | 3357 | *reg_type = info.reg_type; |
31fd8581 | 3358 | |
22dc4a0f AN |
3359 | if (*reg_type == PTR_TO_BTF_ID || *reg_type == PTR_TO_BTF_ID_OR_NULL) { |
3360 | *btf = info.btf; | |
9e15db66 | 3361 | *btf_id = info.btf_id; |
22dc4a0f | 3362 | } else { |
9e15db66 | 3363 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
22dc4a0f | 3364 | } |
32bbe007 AS |
3365 | /* remember the offset of last byte accessed in ctx */ |
3366 | if (env->prog->aux->max_ctx_offset < off + size) | |
3367 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 3368 | return 0; |
32bbe007 | 3369 | } |
17a52670 | 3370 | |
61bd5218 | 3371 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
3372 | return -EACCES; |
3373 | } | |
3374 | ||
d58e468b PP |
3375 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
3376 | int size) | |
3377 | { | |
3378 | if (size < 0 || off < 0 || | |
3379 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
3380 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
3381 | off, size); | |
3382 | return -EACCES; | |
3383 | } | |
3384 | return 0; | |
3385 | } | |
3386 | ||
5f456649 MKL |
3387 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
3388 | u32 regno, int off, int size, | |
3389 | enum bpf_access_type t) | |
c64b7983 JS |
3390 | { |
3391 | struct bpf_reg_state *regs = cur_regs(env); | |
3392 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 3393 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 3394 | bool valid; |
c64b7983 JS |
3395 | |
3396 | if (reg->smin_value < 0) { | |
3397 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
3398 | regno); | |
3399 | return -EACCES; | |
3400 | } | |
3401 | ||
46f8bc92 MKL |
3402 | switch (reg->type) { |
3403 | case PTR_TO_SOCK_COMMON: | |
3404 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
3405 | break; | |
3406 | case PTR_TO_SOCKET: | |
3407 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
3408 | break; | |
655a51e5 MKL |
3409 | case PTR_TO_TCP_SOCK: |
3410 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
3411 | break; | |
fada7fdc JL |
3412 | case PTR_TO_XDP_SOCK: |
3413 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
3414 | break; | |
46f8bc92 MKL |
3415 | default: |
3416 | valid = false; | |
c64b7983 JS |
3417 | } |
3418 | ||
5f456649 | 3419 | |
46f8bc92 MKL |
3420 | if (valid) { |
3421 | env->insn_aux_data[insn_idx].ctx_field_size = | |
3422 | info.ctx_field_size; | |
3423 | return 0; | |
3424 | } | |
3425 | ||
3426 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
3427 | regno, reg_type_str[reg->type], off, size); | |
3428 | ||
3429 | return -EACCES; | |
c64b7983 JS |
3430 | } |
3431 | ||
4cabc5b1 DB |
3432 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
3433 | { | |
2a159c6f | 3434 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
3435 | } |
3436 | ||
f37a8cb8 DB |
3437 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
3438 | { | |
2a159c6f | 3439 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 3440 | |
46f8bc92 MKL |
3441 | return reg->type == PTR_TO_CTX; |
3442 | } | |
3443 | ||
3444 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
3445 | { | |
3446 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
3447 | ||
3448 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
3449 | } |
3450 | ||
ca369602 DB |
3451 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
3452 | { | |
2a159c6f | 3453 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
3454 | |
3455 | return type_is_pkt_pointer(reg->type); | |
3456 | } | |
3457 | ||
4b5defde DB |
3458 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
3459 | { | |
3460 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
3461 | ||
3462 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
3463 | return reg->type == PTR_TO_FLOW_KEYS; | |
3464 | } | |
3465 | ||
61bd5218 JK |
3466 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
3467 | const struct bpf_reg_state *reg, | |
d1174416 | 3468 | int off, int size, bool strict) |
969bf05e | 3469 | { |
f1174f77 | 3470 | struct tnum reg_off; |
e07b98d9 | 3471 | int ip_align; |
d1174416 DM |
3472 | |
3473 | /* Byte size accesses are always allowed. */ | |
3474 | if (!strict || size == 1) | |
3475 | return 0; | |
3476 | ||
e4eda884 DM |
3477 | /* For platforms that do not have a Kconfig enabling |
3478 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
3479 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
3480 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
3481 | * to this code only in strict mode where we want to emulate | |
3482 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
3483 | * unconditional IP align value of '2'. | |
e07b98d9 | 3484 | */ |
e4eda884 | 3485 | ip_align = 2; |
f1174f77 EC |
3486 | |
3487 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
3488 | if (!tnum_is_aligned(reg_off, size)) { | |
3489 | char tn_buf[48]; | |
3490 | ||
3491 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
3492 | verbose(env, |
3493 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 3494 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
3495 | return -EACCES; |
3496 | } | |
79adffcd | 3497 | |
969bf05e AS |
3498 | return 0; |
3499 | } | |
3500 | ||
61bd5218 JK |
3501 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
3502 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
3503 | const char *pointer_desc, |
3504 | int off, int size, bool strict) | |
79adffcd | 3505 | { |
f1174f77 EC |
3506 | struct tnum reg_off; |
3507 | ||
3508 | /* Byte size accesses are always allowed. */ | |
3509 | if (!strict || size == 1) | |
3510 | return 0; | |
3511 | ||
3512 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
3513 | if (!tnum_is_aligned(reg_off, size)) { | |
3514 | char tn_buf[48]; | |
3515 | ||
3516 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3517 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 3518 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
3519 | return -EACCES; |
3520 | } | |
3521 | ||
969bf05e AS |
3522 | return 0; |
3523 | } | |
3524 | ||
e07b98d9 | 3525 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
3526 | const struct bpf_reg_state *reg, int off, |
3527 | int size, bool strict_alignment_once) | |
79adffcd | 3528 | { |
ca369602 | 3529 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 3530 | const char *pointer_desc = ""; |
d1174416 | 3531 | |
79adffcd DB |
3532 | switch (reg->type) { |
3533 | case PTR_TO_PACKET: | |
de8f3a83 DB |
3534 | case PTR_TO_PACKET_META: |
3535 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
3536 | * right in front, treat it the very same way. | |
3537 | */ | |
61bd5218 | 3538 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
3539 | case PTR_TO_FLOW_KEYS: |
3540 | pointer_desc = "flow keys "; | |
3541 | break; | |
69c087ba YS |
3542 | case PTR_TO_MAP_KEY: |
3543 | pointer_desc = "key "; | |
3544 | break; | |
f1174f77 EC |
3545 | case PTR_TO_MAP_VALUE: |
3546 | pointer_desc = "value "; | |
3547 | break; | |
3548 | case PTR_TO_CTX: | |
3549 | pointer_desc = "context "; | |
3550 | break; | |
3551 | case PTR_TO_STACK: | |
3552 | pointer_desc = "stack "; | |
01f810ac AM |
3553 | /* The stack spill tracking logic in check_stack_write_fixed_off() |
3554 | * and check_stack_read_fixed_off() relies on stack accesses being | |
a5ec6ae1 JH |
3555 | * aligned. |
3556 | */ | |
3557 | strict = true; | |
f1174f77 | 3558 | break; |
c64b7983 JS |
3559 | case PTR_TO_SOCKET: |
3560 | pointer_desc = "sock "; | |
3561 | break; | |
46f8bc92 MKL |
3562 | case PTR_TO_SOCK_COMMON: |
3563 | pointer_desc = "sock_common "; | |
3564 | break; | |
655a51e5 MKL |
3565 | case PTR_TO_TCP_SOCK: |
3566 | pointer_desc = "tcp_sock "; | |
3567 | break; | |
fada7fdc JL |
3568 | case PTR_TO_XDP_SOCK: |
3569 | pointer_desc = "xdp_sock "; | |
3570 | break; | |
79adffcd | 3571 | default: |
f1174f77 | 3572 | break; |
79adffcd | 3573 | } |
61bd5218 JK |
3574 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
3575 | strict); | |
79adffcd DB |
3576 | } |
3577 | ||
f4d7e40a AS |
3578 | static int update_stack_depth(struct bpf_verifier_env *env, |
3579 | const struct bpf_func_state *func, | |
3580 | int off) | |
3581 | { | |
9c8105bd | 3582 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
3583 | |
3584 | if (stack >= -off) | |
3585 | return 0; | |
3586 | ||
3587 | /* update known max for given subprogram */ | |
9c8105bd | 3588 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
3589 | return 0; |
3590 | } | |
f4d7e40a | 3591 | |
70a87ffe AS |
3592 | /* starting from main bpf function walk all instructions of the function |
3593 | * and recursively walk all callees that given function can call. | |
3594 | * Ignore jump and exit insns. | |
3595 | * Since recursion is prevented by check_cfg() this algorithm | |
3596 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
3597 | */ | |
3598 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
3599 | { | |
9c8105bd JW |
3600 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
3601 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 3602 | struct bpf_insn *insn = env->prog->insnsi; |
ebf7d1f5 | 3603 | bool tail_call_reachable = false; |
70a87ffe AS |
3604 | int ret_insn[MAX_CALL_FRAMES]; |
3605 | int ret_prog[MAX_CALL_FRAMES]; | |
ebf7d1f5 | 3606 | int j; |
f4d7e40a | 3607 | |
70a87ffe | 3608 | process_func: |
7f6e4312 MF |
3609 | /* protect against potential stack overflow that might happen when |
3610 | * bpf2bpf calls get combined with tailcalls. Limit the caller's stack | |
3611 | * depth for such case down to 256 so that the worst case scenario | |
3612 | * would result in 8k stack size (32 which is tailcall limit * 256 = | |
3613 | * 8k). | |
3614 | * | |
3615 | * To get the idea what might happen, see an example: | |
3616 | * func1 -> sub rsp, 128 | |
3617 | * subfunc1 -> sub rsp, 256 | |
3618 | * tailcall1 -> add rsp, 256 | |
3619 | * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320) | |
3620 | * subfunc2 -> sub rsp, 64 | |
3621 | * subfunc22 -> sub rsp, 128 | |
3622 | * tailcall2 -> add rsp, 128 | |
3623 | * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416) | |
3624 | * | |
3625 | * tailcall will unwind the current stack frame but it will not get rid | |
3626 | * of caller's stack as shown on the example above. | |
3627 | */ | |
3628 | if (idx && subprog[idx].has_tail_call && depth >= 256) { | |
3629 | verbose(env, | |
3630 | "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n", | |
3631 | depth); | |
3632 | return -EACCES; | |
3633 | } | |
70a87ffe AS |
3634 | /* round up to 32-bytes, since this is granularity |
3635 | * of interpreter stack size | |
3636 | */ | |
9c8105bd | 3637 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 3638 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 3639 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 3640 | frame + 1, depth); |
f4d7e40a AS |
3641 | return -EACCES; |
3642 | } | |
70a87ffe | 3643 | continue_func: |
4cb3d99c | 3644 | subprog_end = subprog[idx + 1].start; |
70a87ffe | 3645 | for (; i < subprog_end; i++) { |
69c087ba | 3646 | if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i)) |
70a87ffe AS |
3647 | continue; |
3648 | /* remember insn and function to return to */ | |
3649 | ret_insn[frame] = i + 1; | |
9c8105bd | 3650 | ret_prog[frame] = idx; |
70a87ffe AS |
3651 | |
3652 | /* find the callee */ | |
3653 | i = i + insn[i].imm + 1; | |
9c8105bd JW |
3654 | idx = find_subprog(env, i); |
3655 | if (idx < 0) { | |
70a87ffe AS |
3656 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
3657 | i); | |
3658 | return -EFAULT; | |
3659 | } | |
ebf7d1f5 MF |
3660 | |
3661 | if (subprog[idx].has_tail_call) | |
3662 | tail_call_reachable = true; | |
3663 | ||
70a87ffe AS |
3664 | frame++; |
3665 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
3666 | verbose(env, "the call stack of %d frames is too deep !\n", |
3667 | frame); | |
3668 | return -E2BIG; | |
70a87ffe AS |
3669 | } |
3670 | goto process_func; | |
3671 | } | |
ebf7d1f5 MF |
3672 | /* if tail call got detected across bpf2bpf calls then mark each of the |
3673 | * currently present subprog frames as tail call reachable subprogs; | |
3674 | * this info will be utilized by JIT so that we will be preserving the | |
3675 | * tail call counter throughout bpf2bpf calls combined with tailcalls | |
3676 | */ | |
3677 | if (tail_call_reachable) | |
3678 | for (j = 0; j < frame; j++) | |
3679 | subprog[ret_prog[j]].tail_call_reachable = true; | |
5dd0a6b8 DB |
3680 | if (subprog[0].tail_call_reachable) |
3681 | env->prog->aux->tail_call_reachable = true; | |
ebf7d1f5 | 3682 | |
70a87ffe AS |
3683 | /* end of for() loop means the last insn of the 'subprog' |
3684 | * was reached. Doesn't matter whether it was JA or EXIT | |
3685 | */ | |
3686 | if (frame == 0) | |
3687 | return 0; | |
9c8105bd | 3688 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
3689 | frame--; |
3690 | i = ret_insn[frame]; | |
9c8105bd | 3691 | idx = ret_prog[frame]; |
70a87ffe | 3692 | goto continue_func; |
f4d7e40a AS |
3693 | } |
3694 | ||
19d28fbd | 3695 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
3696 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
3697 | const struct bpf_insn *insn, int idx) | |
3698 | { | |
3699 | int start = idx + insn->imm + 1, subprog; | |
3700 | ||
3701 | subprog = find_subprog(env, start); | |
3702 | if (subprog < 0) { | |
3703 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
3704 | start); | |
3705 | return -EFAULT; | |
3706 | } | |
9c8105bd | 3707 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 3708 | } |
19d28fbd | 3709 | #endif |
1ea47e01 | 3710 | |
51c39bb1 AS |
3711 | int check_ctx_reg(struct bpf_verifier_env *env, |
3712 | const struct bpf_reg_state *reg, int regno) | |
58990d1f DB |
3713 | { |
3714 | /* Access to ctx or passing it to a helper is only allowed in | |
3715 | * its original, unmodified form. | |
3716 | */ | |
3717 | ||
3718 | if (reg->off) { | |
3719 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
3720 | regno, reg->off); | |
3721 | return -EACCES; | |
3722 | } | |
3723 | ||
3724 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3725 | char tn_buf[48]; | |
3726 | ||
3727 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3728 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
3729 | return -EACCES; | |
3730 | } | |
3731 | ||
3732 | return 0; | |
3733 | } | |
3734 | ||
afbf21dc YS |
3735 | static int __check_buffer_access(struct bpf_verifier_env *env, |
3736 | const char *buf_info, | |
3737 | const struct bpf_reg_state *reg, | |
3738 | int regno, int off, int size) | |
9df1c28b MM |
3739 | { |
3740 | if (off < 0) { | |
3741 | verbose(env, | |
4fc00b79 | 3742 | "R%d invalid %s buffer access: off=%d, size=%d\n", |
afbf21dc | 3743 | regno, buf_info, off, size); |
9df1c28b MM |
3744 | return -EACCES; |
3745 | } | |
3746 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3747 | char tn_buf[48]; | |
3748 | ||
3749 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3750 | verbose(env, | |
4fc00b79 | 3751 | "R%d invalid variable buffer offset: off=%d, var_off=%s\n", |
9df1c28b MM |
3752 | regno, off, tn_buf); |
3753 | return -EACCES; | |
3754 | } | |
afbf21dc YS |
3755 | |
3756 | return 0; | |
3757 | } | |
3758 | ||
3759 | static int check_tp_buffer_access(struct bpf_verifier_env *env, | |
3760 | const struct bpf_reg_state *reg, | |
3761 | int regno, int off, int size) | |
3762 | { | |
3763 | int err; | |
3764 | ||
3765 | err = __check_buffer_access(env, "tracepoint", reg, regno, off, size); | |
3766 | if (err) | |
3767 | return err; | |
3768 | ||
9df1c28b MM |
3769 | if (off + size > env->prog->aux->max_tp_access) |
3770 | env->prog->aux->max_tp_access = off + size; | |
3771 | ||
3772 | return 0; | |
3773 | } | |
3774 | ||
afbf21dc YS |
3775 | static int check_buffer_access(struct bpf_verifier_env *env, |
3776 | const struct bpf_reg_state *reg, | |
3777 | int regno, int off, int size, | |
3778 | bool zero_size_allowed, | |
3779 | const char *buf_info, | |
3780 | u32 *max_access) | |
3781 | { | |
3782 | int err; | |
3783 | ||
3784 | err = __check_buffer_access(env, buf_info, reg, regno, off, size); | |
3785 | if (err) | |
3786 | return err; | |
3787 | ||
3788 | if (off + size > *max_access) | |
3789 | *max_access = off + size; | |
3790 | ||
3791 | return 0; | |
3792 | } | |
3793 | ||
3f50f132 JF |
3794 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
3795 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
3796 | { | |
3797 | reg->var_off = tnum_subreg(reg->var_off); | |
3798 | __reg_assign_32_into_64(reg); | |
3799 | } | |
9df1c28b | 3800 | |
0c17d1d2 JH |
3801 | /* truncate register to smaller size (in bytes) |
3802 | * must be called with size < BPF_REG_SIZE | |
3803 | */ | |
3804 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
3805 | { | |
3806 | u64 mask; | |
3807 | ||
3808 | /* clear high bits in bit representation */ | |
3809 | reg->var_off = tnum_cast(reg->var_off, size); | |
3810 | ||
3811 | /* fix arithmetic bounds */ | |
3812 | mask = ((u64)1 << (size * 8)) - 1; | |
3813 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
3814 | reg->umin_value &= mask; | |
3815 | reg->umax_value &= mask; | |
3816 | } else { | |
3817 | reg->umin_value = 0; | |
3818 | reg->umax_value = mask; | |
3819 | } | |
3820 | reg->smin_value = reg->umin_value; | |
3821 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
3822 | |
3823 | /* If size is smaller than 32bit register the 32bit register | |
3824 | * values are also truncated so we push 64-bit bounds into | |
3825 | * 32-bit bounds. Above were truncated < 32-bits already. | |
3826 | */ | |
3827 | if (size >= 4) | |
3828 | return; | |
3829 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
3830 | } |
3831 | ||
a23740ec AN |
3832 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
3833 | { | |
3834 | return (map->map_flags & BPF_F_RDONLY_PROG) && map->frozen; | |
3835 | } | |
3836 | ||
3837 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val) | |
3838 | { | |
3839 | void *ptr; | |
3840 | u64 addr; | |
3841 | int err; | |
3842 | ||
3843 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
3844 | if (err) | |
3845 | return err; | |
2dedd7d2 | 3846 | ptr = (void *)(long)addr + off; |
a23740ec AN |
3847 | |
3848 | switch (size) { | |
3849 | case sizeof(u8): | |
3850 | *val = (u64)*(u8 *)ptr; | |
3851 | break; | |
3852 | case sizeof(u16): | |
3853 | *val = (u64)*(u16 *)ptr; | |
3854 | break; | |
3855 | case sizeof(u32): | |
3856 | *val = (u64)*(u32 *)ptr; | |
3857 | break; | |
3858 | case sizeof(u64): | |
3859 | *val = *(u64 *)ptr; | |
3860 | break; | |
3861 | default: | |
3862 | return -EINVAL; | |
3863 | } | |
3864 | return 0; | |
3865 | } | |
3866 | ||
9e15db66 AS |
3867 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
3868 | struct bpf_reg_state *regs, | |
3869 | int regno, int off, int size, | |
3870 | enum bpf_access_type atype, | |
3871 | int value_regno) | |
3872 | { | |
3873 | struct bpf_reg_state *reg = regs + regno; | |
22dc4a0f AN |
3874 | const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id); |
3875 | const char *tname = btf_name_by_offset(reg->btf, t->name_off); | |
9e15db66 AS |
3876 | u32 btf_id; |
3877 | int ret; | |
3878 | ||
9e15db66 AS |
3879 | if (off < 0) { |
3880 | verbose(env, | |
3881 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
3882 | regno, tname, off); | |
3883 | return -EACCES; | |
3884 | } | |
3885 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3886 | char tn_buf[48]; | |
3887 | ||
3888 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3889 | verbose(env, | |
3890 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
3891 | regno, tname, off, tn_buf); | |
3892 | return -EACCES; | |
3893 | } | |
3894 | ||
27ae7997 | 3895 | if (env->ops->btf_struct_access) { |
22dc4a0f AN |
3896 | ret = env->ops->btf_struct_access(&env->log, reg->btf, t, |
3897 | off, size, atype, &btf_id); | |
27ae7997 MKL |
3898 | } else { |
3899 | if (atype != BPF_READ) { | |
3900 | verbose(env, "only read is supported\n"); | |
3901 | return -EACCES; | |
3902 | } | |
3903 | ||
22dc4a0f AN |
3904 | ret = btf_struct_access(&env->log, reg->btf, t, off, size, |
3905 | atype, &btf_id); | |
27ae7997 MKL |
3906 | } |
3907 | ||
9e15db66 AS |
3908 | if (ret < 0) |
3909 | return ret; | |
3910 | ||
41c48f3a | 3911 | if (atype == BPF_READ && value_regno >= 0) |
22dc4a0f | 3912 | mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id); |
41c48f3a AI |
3913 | |
3914 | return 0; | |
3915 | } | |
3916 | ||
3917 | static int check_ptr_to_map_access(struct bpf_verifier_env *env, | |
3918 | struct bpf_reg_state *regs, | |
3919 | int regno, int off, int size, | |
3920 | enum bpf_access_type atype, | |
3921 | int value_regno) | |
3922 | { | |
3923 | struct bpf_reg_state *reg = regs + regno; | |
3924 | struct bpf_map *map = reg->map_ptr; | |
3925 | const struct btf_type *t; | |
3926 | const char *tname; | |
3927 | u32 btf_id; | |
3928 | int ret; | |
3929 | ||
3930 | if (!btf_vmlinux) { | |
3931 | verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
3932 | return -ENOTSUPP; | |
3933 | } | |
3934 | ||
3935 | if (!map->ops->map_btf_id || !*map->ops->map_btf_id) { | |
3936 | verbose(env, "map_ptr access not supported for map type %d\n", | |
3937 | map->map_type); | |
3938 | return -ENOTSUPP; | |
3939 | } | |
3940 | ||
3941 | t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id); | |
3942 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
3943 | ||
3944 | if (!env->allow_ptr_to_map_access) { | |
3945 | verbose(env, | |
3946 | "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", | |
3947 | tname); | |
3948 | return -EPERM; | |
9e15db66 | 3949 | } |
27ae7997 | 3950 | |
41c48f3a AI |
3951 | if (off < 0) { |
3952 | verbose(env, "R%d is %s invalid negative access: off=%d\n", | |
3953 | regno, tname, off); | |
3954 | return -EACCES; | |
3955 | } | |
3956 | ||
3957 | if (atype != BPF_READ) { | |
3958 | verbose(env, "only read from %s is supported\n", tname); | |
3959 | return -EACCES; | |
3960 | } | |
3961 | ||
22dc4a0f | 3962 | ret = btf_struct_access(&env->log, btf_vmlinux, t, off, size, atype, &btf_id); |
41c48f3a AI |
3963 | if (ret < 0) |
3964 | return ret; | |
3965 | ||
3966 | if (value_regno >= 0) | |
22dc4a0f | 3967 | mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id); |
41c48f3a | 3968 | |
9e15db66 AS |
3969 | return 0; |
3970 | } | |
3971 | ||
01f810ac AM |
3972 | /* Check that the stack access at the given offset is within bounds. The |
3973 | * maximum valid offset is -1. | |
3974 | * | |
3975 | * The minimum valid offset is -MAX_BPF_STACK for writes, and | |
3976 | * -state->allocated_stack for reads. | |
3977 | */ | |
3978 | static int check_stack_slot_within_bounds(int off, | |
3979 | struct bpf_func_state *state, | |
3980 | enum bpf_access_type t) | |
3981 | { | |
3982 | int min_valid_off; | |
3983 | ||
3984 | if (t == BPF_WRITE) | |
3985 | min_valid_off = -MAX_BPF_STACK; | |
3986 | else | |
3987 | min_valid_off = -state->allocated_stack; | |
3988 | ||
3989 | if (off < min_valid_off || off > -1) | |
3990 | return -EACCES; | |
3991 | return 0; | |
3992 | } | |
3993 | ||
3994 | /* Check that the stack access at 'regno + off' falls within the maximum stack | |
3995 | * bounds. | |
3996 | * | |
3997 | * 'off' includes `regno->offset`, but not its dynamic part (if any). | |
3998 | */ | |
3999 | static int check_stack_access_within_bounds( | |
4000 | struct bpf_verifier_env *env, | |
4001 | int regno, int off, int access_size, | |
4002 | enum stack_access_src src, enum bpf_access_type type) | |
4003 | { | |
4004 | struct bpf_reg_state *regs = cur_regs(env); | |
4005 | struct bpf_reg_state *reg = regs + regno; | |
4006 | struct bpf_func_state *state = func(env, reg); | |
4007 | int min_off, max_off; | |
4008 | int err; | |
4009 | char *err_extra; | |
4010 | ||
4011 | if (src == ACCESS_HELPER) | |
4012 | /* We don't know if helpers are reading or writing (or both). */ | |
4013 | err_extra = " indirect access to"; | |
4014 | else if (type == BPF_READ) | |
4015 | err_extra = " read from"; | |
4016 | else | |
4017 | err_extra = " write to"; | |
4018 | ||
4019 | if (tnum_is_const(reg->var_off)) { | |
4020 | min_off = reg->var_off.value + off; | |
4021 | if (access_size > 0) | |
4022 | max_off = min_off + access_size - 1; | |
4023 | else | |
4024 | max_off = min_off; | |
4025 | } else { | |
4026 | if (reg->smax_value >= BPF_MAX_VAR_OFF || | |
4027 | reg->smin_value <= -BPF_MAX_VAR_OFF) { | |
4028 | verbose(env, "invalid unbounded variable-offset%s stack R%d\n", | |
4029 | err_extra, regno); | |
4030 | return -EACCES; | |
4031 | } | |
4032 | min_off = reg->smin_value + off; | |
4033 | if (access_size > 0) | |
4034 | max_off = reg->smax_value + off + access_size - 1; | |
4035 | else | |
4036 | max_off = min_off; | |
4037 | } | |
4038 | ||
4039 | err = check_stack_slot_within_bounds(min_off, state, type); | |
4040 | if (!err) | |
4041 | err = check_stack_slot_within_bounds(max_off, state, type); | |
4042 | ||
4043 | if (err) { | |
4044 | if (tnum_is_const(reg->var_off)) { | |
4045 | verbose(env, "invalid%s stack R%d off=%d size=%d\n", | |
4046 | err_extra, regno, off, access_size); | |
4047 | } else { | |
4048 | char tn_buf[48]; | |
4049 | ||
4050 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
4051 | verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n", | |
4052 | err_extra, regno, tn_buf, access_size); | |
4053 | } | |
4054 | } | |
4055 | return err; | |
4056 | } | |
41c48f3a | 4057 | |
17a52670 AS |
4058 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
4059 | * if t==write, value_regno is a register which value is stored into memory | |
4060 | * if t==read, value_regno is a register which will receive the value from memory | |
4061 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
4062 | * if t==read && value_regno==-1, don't care what we read from memory | |
4063 | */ | |
ca369602 DB |
4064 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
4065 | int off, int bpf_size, enum bpf_access_type t, | |
4066 | int value_regno, bool strict_alignment_once) | |
17a52670 | 4067 | { |
638f5b90 AS |
4068 | struct bpf_reg_state *regs = cur_regs(env); |
4069 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 4070 | struct bpf_func_state *state; |
17a52670 AS |
4071 | int size, err = 0; |
4072 | ||
4073 | size = bpf_size_to_bytes(bpf_size); | |
4074 | if (size < 0) | |
4075 | return size; | |
4076 | ||
f1174f77 | 4077 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 4078 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
4079 | if (err) |
4080 | return err; | |
17a52670 | 4081 | |
f1174f77 EC |
4082 | /* for access checks, reg->off is just part of off */ |
4083 | off += reg->off; | |
4084 | ||
69c087ba YS |
4085 | if (reg->type == PTR_TO_MAP_KEY) { |
4086 | if (t == BPF_WRITE) { | |
4087 | verbose(env, "write to change key R%d not allowed\n", regno); | |
4088 | return -EACCES; | |
4089 | } | |
4090 | ||
4091 | err = check_mem_region_access(env, regno, off, size, | |
4092 | reg->map_ptr->key_size, false); | |
4093 | if (err) | |
4094 | return err; | |
4095 | if (value_regno >= 0) | |
4096 | mark_reg_unknown(env, regs, value_regno); | |
4097 | } else if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
4098 | if (t == BPF_WRITE && value_regno >= 0 && |
4099 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 4100 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
4101 | return -EACCES; |
4102 | } | |
591fe988 DB |
4103 | err = check_map_access_type(env, regno, off, size, t); |
4104 | if (err) | |
4105 | return err; | |
9fd29c08 | 4106 | err = check_map_access(env, regno, off, size, false); |
a23740ec AN |
4107 | if (!err && t == BPF_READ && value_regno >= 0) { |
4108 | struct bpf_map *map = reg->map_ptr; | |
4109 | ||
4110 | /* if map is read-only, track its contents as scalars */ | |
4111 | if (tnum_is_const(reg->var_off) && | |
4112 | bpf_map_is_rdonly(map) && | |
4113 | map->ops->map_direct_value_addr) { | |
4114 | int map_off = off + reg->var_off.value; | |
4115 | u64 val = 0; | |
4116 | ||
4117 | err = bpf_map_direct_read(map, map_off, size, | |
4118 | &val); | |
4119 | if (err) | |
4120 | return err; | |
4121 | ||
4122 | regs[value_regno].type = SCALAR_VALUE; | |
4123 | __mark_reg_known(®s[value_regno], val); | |
4124 | } else { | |
4125 | mark_reg_unknown(env, regs, value_regno); | |
4126 | } | |
4127 | } | |
457f4436 AN |
4128 | } else if (reg->type == PTR_TO_MEM) { |
4129 | if (t == BPF_WRITE && value_regno >= 0 && | |
4130 | is_pointer_value(env, value_regno)) { | |
4131 | verbose(env, "R%d leaks addr into mem\n", value_regno); | |
4132 | return -EACCES; | |
4133 | } | |
4134 | err = check_mem_region_access(env, regno, off, size, | |
4135 | reg->mem_size, false); | |
4136 | if (!err && t == BPF_READ && value_regno >= 0) | |
4137 | mark_reg_unknown(env, regs, value_regno); | |
1a0dc1ac | 4138 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 4139 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
22dc4a0f | 4140 | struct btf *btf = NULL; |
9e15db66 | 4141 | u32 btf_id = 0; |
19de99f7 | 4142 | |
1be7f75d AS |
4143 | if (t == BPF_WRITE && value_regno >= 0 && |
4144 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 4145 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
4146 | return -EACCES; |
4147 | } | |
f1174f77 | 4148 | |
58990d1f DB |
4149 | err = check_ctx_reg(env, reg, regno); |
4150 | if (err < 0) | |
4151 | return err; | |
4152 | ||
22dc4a0f | 4153 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf, &btf_id); |
9e15db66 AS |
4154 | if (err) |
4155 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 4156 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 4157 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
4158 | * PTR_TO_PACKET[_META,_END]. In the latter |
4159 | * case, we know the offset is zero. | |
f1174f77 | 4160 | */ |
46f8bc92 | 4161 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 4162 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 4163 | } else { |
638f5b90 | 4164 | mark_reg_known_zero(env, regs, |
61bd5218 | 4165 | value_regno); |
46f8bc92 MKL |
4166 | if (reg_type_may_be_null(reg_type)) |
4167 | regs[value_regno].id = ++env->id_gen; | |
5327ed3d JW |
4168 | /* A load of ctx field could have different |
4169 | * actual load size with the one encoded in the | |
4170 | * insn. When the dst is PTR, it is for sure not | |
4171 | * a sub-register. | |
4172 | */ | |
4173 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
b121b341 | 4174 | if (reg_type == PTR_TO_BTF_ID || |
22dc4a0f AN |
4175 | reg_type == PTR_TO_BTF_ID_OR_NULL) { |
4176 | regs[value_regno].btf = btf; | |
9e15db66 | 4177 | regs[value_regno].btf_id = btf_id; |
22dc4a0f | 4178 | } |
46f8bc92 | 4179 | } |
638f5b90 | 4180 | regs[value_regno].type = reg_type; |
969bf05e | 4181 | } |
17a52670 | 4182 | |
f1174f77 | 4183 | } else if (reg->type == PTR_TO_STACK) { |
01f810ac AM |
4184 | /* Basic bounds checks. */ |
4185 | err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t); | |
e4298d25 DB |
4186 | if (err) |
4187 | return err; | |
8726679a | 4188 | |
f4d7e40a AS |
4189 | state = func(env, reg); |
4190 | err = update_stack_depth(env, state, off); | |
4191 | if (err) | |
4192 | return err; | |
8726679a | 4193 | |
01f810ac AM |
4194 | if (t == BPF_READ) |
4195 | err = check_stack_read(env, regno, off, size, | |
61bd5218 | 4196 | value_regno); |
01f810ac AM |
4197 | else |
4198 | err = check_stack_write(env, regno, off, size, | |
4199 | value_regno, insn_idx); | |
de8f3a83 | 4200 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 4201 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 4202 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
4203 | return -EACCES; |
4204 | } | |
4acf6c0b BB |
4205 | if (t == BPF_WRITE && value_regno >= 0 && |
4206 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
4207 | verbose(env, "R%d leaks addr into packet\n", |
4208 | value_regno); | |
4acf6c0b BB |
4209 | return -EACCES; |
4210 | } | |
9fd29c08 | 4211 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 4212 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 4213 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
4214 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
4215 | if (t == BPF_WRITE && value_regno >= 0 && | |
4216 | is_pointer_value(env, value_regno)) { | |
4217 | verbose(env, "R%d leaks addr into flow keys\n", | |
4218 | value_regno); | |
4219 | return -EACCES; | |
4220 | } | |
4221 | ||
4222 | err = check_flow_keys_access(env, off, size); | |
4223 | if (!err && t == BPF_READ && value_regno >= 0) | |
4224 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 4225 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 4226 | if (t == BPF_WRITE) { |
46f8bc92 MKL |
4227 | verbose(env, "R%d cannot write into %s\n", |
4228 | regno, reg_type_str[reg->type]); | |
c64b7983 JS |
4229 | return -EACCES; |
4230 | } | |
5f456649 | 4231 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
4232 | if (!err && value_regno >= 0) |
4233 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
4234 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
4235 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
4236 | if (!err && t == BPF_READ && value_regno >= 0) | |
4237 | mark_reg_unknown(env, regs, value_regno); | |
9e15db66 AS |
4238 | } else if (reg->type == PTR_TO_BTF_ID) { |
4239 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, | |
4240 | value_regno); | |
41c48f3a AI |
4241 | } else if (reg->type == CONST_PTR_TO_MAP) { |
4242 | err = check_ptr_to_map_access(env, regs, regno, off, size, t, | |
4243 | value_regno); | |
afbf21dc YS |
4244 | } else if (reg->type == PTR_TO_RDONLY_BUF) { |
4245 | if (t == BPF_WRITE) { | |
4246 | verbose(env, "R%d cannot write into %s\n", | |
4247 | regno, reg_type_str[reg->type]); | |
4248 | return -EACCES; | |
4249 | } | |
f6dfbe31 CIK |
4250 | err = check_buffer_access(env, reg, regno, off, size, false, |
4251 | "rdonly", | |
afbf21dc YS |
4252 | &env->prog->aux->max_rdonly_access); |
4253 | if (!err && value_regno >= 0) | |
4254 | mark_reg_unknown(env, regs, value_regno); | |
4255 | } else if (reg->type == PTR_TO_RDWR_BUF) { | |
f6dfbe31 CIK |
4256 | err = check_buffer_access(env, reg, regno, off, size, false, |
4257 | "rdwr", | |
afbf21dc YS |
4258 | &env->prog->aux->max_rdwr_access); |
4259 | if (!err && t == BPF_READ && value_regno >= 0) | |
4260 | mark_reg_unknown(env, regs, value_regno); | |
17a52670 | 4261 | } else { |
61bd5218 JK |
4262 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
4263 | reg_type_str[reg->type]); | |
17a52670 AS |
4264 | return -EACCES; |
4265 | } | |
969bf05e | 4266 | |
f1174f77 | 4267 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 4268 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 4269 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 4270 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 4271 | } |
17a52670 AS |
4272 | return err; |
4273 | } | |
4274 | ||
91c960b0 | 4275 | static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 4276 | { |
5ffa2550 | 4277 | int load_reg; |
17a52670 AS |
4278 | int err; |
4279 | ||
5ca419f2 BJ |
4280 | switch (insn->imm) { |
4281 | case BPF_ADD: | |
4282 | case BPF_ADD | BPF_FETCH: | |
981f94c3 BJ |
4283 | case BPF_AND: |
4284 | case BPF_AND | BPF_FETCH: | |
4285 | case BPF_OR: | |
4286 | case BPF_OR | BPF_FETCH: | |
4287 | case BPF_XOR: | |
4288 | case BPF_XOR | BPF_FETCH: | |
5ffa2550 BJ |
4289 | case BPF_XCHG: |
4290 | case BPF_CMPXCHG: | |
5ca419f2 BJ |
4291 | break; |
4292 | default: | |
91c960b0 BJ |
4293 | verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm); |
4294 | return -EINVAL; | |
4295 | } | |
4296 | ||
4297 | if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) { | |
4298 | verbose(env, "invalid atomic operand size\n"); | |
17a52670 AS |
4299 | return -EINVAL; |
4300 | } | |
4301 | ||
4302 | /* check src1 operand */ | |
dc503a8a | 4303 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4304 | if (err) |
4305 | return err; | |
4306 | ||
4307 | /* check src2 operand */ | |
dc503a8a | 4308 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4309 | if (err) |
4310 | return err; | |
4311 | ||
5ffa2550 BJ |
4312 | if (insn->imm == BPF_CMPXCHG) { |
4313 | /* Check comparison of R0 with memory location */ | |
4314 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
4315 | if (err) | |
4316 | return err; | |
4317 | } | |
4318 | ||
6bdf6abc | 4319 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 4320 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
4321 | return -EACCES; |
4322 | } | |
4323 | ||
ca369602 | 4324 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 4325 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
4326 | is_flow_key_reg(env, insn->dst_reg) || |
4327 | is_sk_reg(env, insn->dst_reg)) { | |
91c960b0 | 4328 | verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n", |
2a159c6f DB |
4329 | insn->dst_reg, |
4330 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
4331 | return -EACCES; |
4332 | } | |
4333 | ||
37086bfd BJ |
4334 | if (insn->imm & BPF_FETCH) { |
4335 | if (insn->imm == BPF_CMPXCHG) | |
4336 | load_reg = BPF_REG_0; | |
4337 | else | |
4338 | load_reg = insn->src_reg; | |
4339 | ||
4340 | /* check and record load of old value */ | |
4341 | err = check_reg_arg(env, load_reg, DST_OP); | |
4342 | if (err) | |
4343 | return err; | |
4344 | } else { | |
4345 | /* This instruction accesses a memory location but doesn't | |
4346 | * actually load it into a register. | |
4347 | */ | |
4348 | load_reg = -1; | |
4349 | } | |
4350 | ||
91c960b0 | 4351 | /* check whether we can read the memory */ |
31fd8581 | 4352 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
37086bfd | 4353 | BPF_SIZE(insn->code), BPF_READ, load_reg, true); |
17a52670 AS |
4354 | if (err) |
4355 | return err; | |
4356 | ||
91c960b0 | 4357 | /* check whether we can write into the same memory */ |
5ca419f2 BJ |
4358 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
4359 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); | |
4360 | if (err) | |
4361 | return err; | |
4362 | ||
5ca419f2 | 4363 | return 0; |
17a52670 AS |
4364 | } |
4365 | ||
01f810ac AM |
4366 | /* When register 'regno' is used to read the stack (either directly or through |
4367 | * a helper function) make sure that it's within stack boundary and, depending | |
4368 | * on the access type, that all elements of the stack are initialized. | |
4369 | * | |
4370 | * 'off' includes 'regno->off', but not its dynamic part (if any). | |
4371 | * | |
4372 | * All registers that have been spilled on the stack in the slots within the | |
4373 | * read offsets are marked as read. | |
4374 | */ | |
4375 | static int check_stack_range_initialized( | |
4376 | struct bpf_verifier_env *env, int regno, int off, | |
4377 | int access_size, bool zero_size_allowed, | |
4378 | enum stack_access_src type, struct bpf_call_arg_meta *meta) | |
2011fccf AI |
4379 | { |
4380 | struct bpf_reg_state *reg = reg_state(env, regno); | |
01f810ac AM |
4381 | struct bpf_func_state *state = func(env, reg); |
4382 | int err, min_off, max_off, i, j, slot, spi; | |
4383 | char *err_extra = type == ACCESS_HELPER ? " indirect" : ""; | |
4384 | enum bpf_access_type bounds_check_type; | |
4385 | /* Some accesses can write anything into the stack, others are | |
4386 | * read-only. | |
4387 | */ | |
4388 | bool clobber = false; | |
2011fccf | 4389 | |
01f810ac AM |
4390 | if (access_size == 0 && !zero_size_allowed) { |
4391 | verbose(env, "invalid zero-sized read\n"); | |
2011fccf AI |
4392 | return -EACCES; |
4393 | } | |
2011fccf | 4394 | |
01f810ac AM |
4395 | if (type == ACCESS_HELPER) { |
4396 | /* The bounds checks for writes are more permissive than for | |
4397 | * reads. However, if raw_mode is not set, we'll do extra | |
4398 | * checks below. | |
4399 | */ | |
4400 | bounds_check_type = BPF_WRITE; | |
4401 | clobber = true; | |
4402 | } else { | |
4403 | bounds_check_type = BPF_READ; | |
4404 | } | |
4405 | err = check_stack_access_within_bounds(env, regno, off, access_size, | |
4406 | type, bounds_check_type); | |
4407 | if (err) | |
4408 | return err; | |
4409 | ||
17a52670 | 4410 | |
2011fccf | 4411 | if (tnum_is_const(reg->var_off)) { |
01f810ac | 4412 | min_off = max_off = reg->var_off.value + off; |
2011fccf | 4413 | } else { |
088ec26d AI |
4414 | /* Variable offset is prohibited for unprivileged mode for |
4415 | * simplicity since it requires corresponding support in | |
4416 | * Spectre masking for stack ALU. | |
4417 | * See also retrieve_ptr_limit(). | |
4418 | */ | |
2c78ee89 | 4419 | if (!env->bypass_spec_v1) { |
088ec26d | 4420 | char tn_buf[48]; |
f1174f77 | 4421 | |
088ec26d | 4422 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
01f810ac AM |
4423 | verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n", |
4424 | regno, err_extra, tn_buf); | |
088ec26d AI |
4425 | return -EACCES; |
4426 | } | |
f2bcd05e AI |
4427 | /* Only initialized buffer on stack is allowed to be accessed |
4428 | * with variable offset. With uninitialized buffer it's hard to | |
4429 | * guarantee that whole memory is marked as initialized on | |
4430 | * helper return since specific bounds are unknown what may | |
4431 | * cause uninitialized stack leaking. | |
4432 | */ | |
4433 | if (meta && meta->raw_mode) | |
4434 | meta = NULL; | |
4435 | ||
01f810ac AM |
4436 | min_off = reg->smin_value + off; |
4437 | max_off = reg->smax_value + off; | |
17a52670 AS |
4438 | } |
4439 | ||
435faee1 DB |
4440 | if (meta && meta->raw_mode) { |
4441 | meta->access_size = access_size; | |
4442 | meta->regno = regno; | |
4443 | return 0; | |
4444 | } | |
4445 | ||
2011fccf | 4446 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
4447 | u8 *stype; |
4448 | ||
2011fccf | 4449 | slot = -i - 1; |
638f5b90 | 4450 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
4451 | if (state->allocated_stack <= slot) |
4452 | goto err; | |
4453 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
4454 | if (*stype == STACK_MISC) | |
4455 | goto mark; | |
4456 | if (*stype == STACK_ZERO) { | |
01f810ac AM |
4457 | if (clobber) { |
4458 | /* helper can write anything into the stack */ | |
4459 | *stype = STACK_MISC; | |
4460 | } | |
cc2b14d5 | 4461 | goto mark; |
17a52670 | 4462 | } |
1d68f22b YS |
4463 | |
4464 | if (state->stack[spi].slot_type[0] == STACK_SPILL && | |
4465 | state->stack[spi].spilled_ptr.type == PTR_TO_BTF_ID) | |
4466 | goto mark; | |
4467 | ||
f7cf25b2 | 4468 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
cd17d38f YS |
4469 | (state->stack[spi].spilled_ptr.type == SCALAR_VALUE || |
4470 | env->allow_ptr_leaks)) { | |
01f810ac AM |
4471 | if (clobber) { |
4472 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); | |
4473 | for (j = 0; j < BPF_REG_SIZE; j++) | |
4474 | state->stack[spi].slot_type[j] = STACK_MISC; | |
4475 | } | |
f7cf25b2 AS |
4476 | goto mark; |
4477 | } | |
4478 | ||
cc2b14d5 | 4479 | err: |
2011fccf | 4480 | if (tnum_is_const(reg->var_off)) { |
01f810ac AM |
4481 | verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n", |
4482 | err_extra, regno, min_off, i - min_off, access_size); | |
2011fccf AI |
4483 | } else { |
4484 | char tn_buf[48]; | |
4485 | ||
4486 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac AM |
4487 | verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n", |
4488 | err_extra, regno, tn_buf, i - min_off, access_size); | |
2011fccf | 4489 | } |
cc2b14d5 AS |
4490 | return -EACCES; |
4491 | mark: | |
4492 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
4493 | * the whole slot to be marked as 'read' | |
4494 | */ | |
679c782d | 4495 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
4496 | state->stack[spi].spilled_ptr.parent, |
4497 | REG_LIVE_READ64); | |
17a52670 | 4498 | } |
2011fccf | 4499 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
4500 | } |
4501 | ||
06c1c049 GB |
4502 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
4503 | int access_size, bool zero_size_allowed, | |
4504 | struct bpf_call_arg_meta *meta) | |
4505 | { | |
638f5b90 | 4506 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 4507 | |
f1174f77 | 4508 | switch (reg->type) { |
06c1c049 | 4509 | case PTR_TO_PACKET: |
de8f3a83 | 4510 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
4511 | return check_packet_access(env, regno, reg->off, access_size, |
4512 | zero_size_allowed); | |
69c087ba YS |
4513 | case PTR_TO_MAP_KEY: |
4514 | return check_mem_region_access(env, regno, reg->off, access_size, | |
4515 | reg->map_ptr->key_size, false); | |
06c1c049 | 4516 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
4517 | if (check_map_access_type(env, regno, reg->off, access_size, |
4518 | meta && meta->raw_mode ? BPF_WRITE : | |
4519 | BPF_READ)) | |
4520 | return -EACCES; | |
9fd29c08 YS |
4521 | return check_map_access(env, regno, reg->off, access_size, |
4522 | zero_size_allowed); | |
457f4436 AN |
4523 | case PTR_TO_MEM: |
4524 | return check_mem_region_access(env, regno, reg->off, | |
4525 | access_size, reg->mem_size, | |
4526 | zero_size_allowed); | |
afbf21dc YS |
4527 | case PTR_TO_RDONLY_BUF: |
4528 | if (meta && meta->raw_mode) | |
4529 | return -EACCES; | |
4530 | return check_buffer_access(env, reg, regno, reg->off, | |
4531 | access_size, zero_size_allowed, | |
4532 | "rdonly", | |
4533 | &env->prog->aux->max_rdonly_access); | |
4534 | case PTR_TO_RDWR_BUF: | |
4535 | return check_buffer_access(env, reg, regno, reg->off, | |
4536 | access_size, zero_size_allowed, | |
4537 | "rdwr", | |
4538 | &env->prog->aux->max_rdwr_access); | |
0d004c02 | 4539 | case PTR_TO_STACK: |
01f810ac AM |
4540 | return check_stack_range_initialized( |
4541 | env, | |
4542 | regno, reg->off, access_size, | |
4543 | zero_size_allowed, ACCESS_HELPER, meta); | |
0d004c02 LB |
4544 | default: /* scalar_value or invalid ptr */ |
4545 | /* Allow zero-byte read from NULL, regardless of pointer type */ | |
4546 | if (zero_size_allowed && access_size == 0 && | |
4547 | register_is_null(reg)) | |
4548 | return 0; | |
4549 | ||
4550 | verbose(env, "R%d type=%s expected=%s\n", regno, | |
4551 | reg_type_str[reg->type], | |
4552 | reg_type_str[PTR_TO_STACK]); | |
4553 | return -EACCES; | |
06c1c049 GB |
4554 | } |
4555 | } | |
4556 | ||
e5069b9c DB |
4557 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
4558 | u32 regno, u32 mem_size) | |
4559 | { | |
4560 | if (register_is_null(reg)) | |
4561 | return 0; | |
4562 | ||
4563 | if (reg_type_may_be_null(reg->type)) { | |
4564 | /* Assuming that the register contains a value check if the memory | |
4565 | * access is safe. Temporarily save and restore the register's state as | |
4566 | * the conversion shouldn't be visible to a caller. | |
4567 | */ | |
4568 | const struct bpf_reg_state saved_reg = *reg; | |
4569 | int rv; | |
4570 | ||
4571 | mark_ptr_not_null_reg(reg); | |
4572 | rv = check_helper_mem_access(env, regno, mem_size, true, NULL); | |
4573 | *reg = saved_reg; | |
4574 | return rv; | |
4575 | } | |
4576 | ||
4577 | return check_helper_mem_access(env, regno, mem_size, true, NULL); | |
4578 | } | |
4579 | ||
d83525ca AS |
4580 | /* Implementation details: |
4581 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL | |
4582 | * Two bpf_map_lookups (even with the same key) will have different reg->id. | |
4583 | * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after | |
4584 | * value_or_null->value transition, since the verifier only cares about | |
4585 | * the range of access to valid map value pointer and doesn't care about actual | |
4586 | * address of the map element. | |
4587 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps | |
4588 | * reg->id > 0 after value_or_null->value transition. By doing so | |
4589 | * two bpf_map_lookups will be considered two different pointers that | |
4590 | * point to different bpf_spin_locks. | |
4591 | * The verifier allows taking only one bpf_spin_lock at a time to avoid | |
4592 | * dead-locks. | |
4593 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
4594 | * reg_is_refcounted() logic. The verifier needs to remember only | |
4595 | * one spin_lock instead of array of acquired_refs. | |
4596 | * cur_state->active_spin_lock remembers which map value element got locked | |
4597 | * and clears it after bpf_spin_unlock. | |
4598 | */ | |
4599 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
4600 | bool is_lock) | |
4601 | { | |
4602 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
4603 | struct bpf_verifier_state *cur = env->cur_state; | |
4604 | bool is_const = tnum_is_const(reg->var_off); | |
4605 | struct bpf_map *map = reg->map_ptr; | |
4606 | u64 val = reg->var_off.value; | |
4607 | ||
d83525ca AS |
4608 | if (!is_const) { |
4609 | verbose(env, | |
4610 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
4611 | regno); | |
4612 | return -EINVAL; | |
4613 | } | |
4614 | if (!map->btf) { | |
4615 | verbose(env, | |
4616 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
4617 | map->name); | |
4618 | return -EINVAL; | |
4619 | } | |
4620 | if (!map_value_has_spin_lock(map)) { | |
4621 | if (map->spin_lock_off == -E2BIG) | |
4622 | verbose(env, | |
4623 | "map '%s' has more than one 'struct bpf_spin_lock'\n", | |
4624 | map->name); | |
4625 | else if (map->spin_lock_off == -ENOENT) | |
4626 | verbose(env, | |
4627 | "map '%s' doesn't have 'struct bpf_spin_lock'\n", | |
4628 | map->name); | |
4629 | else | |
4630 | verbose(env, | |
4631 | "map '%s' is not a struct type or bpf_spin_lock is mangled\n", | |
4632 | map->name); | |
4633 | return -EINVAL; | |
4634 | } | |
4635 | if (map->spin_lock_off != val + reg->off) { | |
4636 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock'\n", | |
4637 | val + reg->off); | |
4638 | return -EINVAL; | |
4639 | } | |
4640 | if (is_lock) { | |
4641 | if (cur->active_spin_lock) { | |
4642 | verbose(env, | |
4643 | "Locking two bpf_spin_locks are not allowed\n"); | |
4644 | return -EINVAL; | |
4645 | } | |
4646 | cur->active_spin_lock = reg->id; | |
4647 | } else { | |
4648 | if (!cur->active_spin_lock) { | |
4649 | verbose(env, "bpf_spin_unlock without taking a lock\n"); | |
4650 | return -EINVAL; | |
4651 | } | |
4652 | if (cur->active_spin_lock != reg->id) { | |
4653 | verbose(env, "bpf_spin_unlock of different lock\n"); | |
4654 | return -EINVAL; | |
4655 | } | |
4656 | cur->active_spin_lock = 0; | |
4657 | } | |
4658 | return 0; | |
4659 | } | |
4660 | ||
90133415 DB |
4661 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
4662 | { | |
4663 | return type == ARG_PTR_TO_MEM || | |
4664 | type == ARG_PTR_TO_MEM_OR_NULL || | |
4665 | type == ARG_PTR_TO_UNINIT_MEM; | |
4666 | } | |
4667 | ||
4668 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
4669 | { | |
4670 | return type == ARG_CONST_SIZE || | |
4671 | type == ARG_CONST_SIZE_OR_ZERO; | |
4672 | } | |
4673 | ||
457f4436 AN |
4674 | static bool arg_type_is_alloc_size(enum bpf_arg_type type) |
4675 | { | |
4676 | return type == ARG_CONST_ALLOC_SIZE_OR_ZERO; | |
4677 | } | |
4678 | ||
57c3bb72 AI |
4679 | static bool arg_type_is_int_ptr(enum bpf_arg_type type) |
4680 | { | |
4681 | return type == ARG_PTR_TO_INT || | |
4682 | type == ARG_PTR_TO_LONG; | |
4683 | } | |
4684 | ||
4685 | static int int_ptr_type_to_size(enum bpf_arg_type type) | |
4686 | { | |
4687 | if (type == ARG_PTR_TO_INT) | |
4688 | return sizeof(u32); | |
4689 | else if (type == ARG_PTR_TO_LONG) | |
4690 | return sizeof(u64); | |
4691 | ||
4692 | return -EINVAL; | |
4693 | } | |
4694 | ||
912f442c LB |
4695 | static int resolve_map_arg_type(struct bpf_verifier_env *env, |
4696 | const struct bpf_call_arg_meta *meta, | |
4697 | enum bpf_arg_type *arg_type) | |
4698 | { | |
4699 | if (!meta->map_ptr) { | |
4700 | /* kernel subsystem misconfigured verifier */ | |
4701 | verbose(env, "invalid map_ptr to access map->type\n"); | |
4702 | return -EACCES; | |
4703 | } | |
4704 | ||
4705 | switch (meta->map_ptr->map_type) { | |
4706 | case BPF_MAP_TYPE_SOCKMAP: | |
4707 | case BPF_MAP_TYPE_SOCKHASH: | |
4708 | if (*arg_type == ARG_PTR_TO_MAP_VALUE) { | |
6550f2dd | 4709 | *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON; |
912f442c LB |
4710 | } else { |
4711 | verbose(env, "invalid arg_type for sockmap/sockhash\n"); | |
4712 | return -EINVAL; | |
4713 | } | |
4714 | break; | |
4715 | ||
4716 | default: | |
4717 | break; | |
4718 | } | |
4719 | return 0; | |
4720 | } | |
4721 | ||
f79e7ea5 LB |
4722 | struct bpf_reg_types { |
4723 | const enum bpf_reg_type types[10]; | |
1df8f55a | 4724 | u32 *btf_id; |
f79e7ea5 LB |
4725 | }; |
4726 | ||
4727 | static const struct bpf_reg_types map_key_value_types = { | |
4728 | .types = { | |
4729 | PTR_TO_STACK, | |
4730 | PTR_TO_PACKET, | |
4731 | PTR_TO_PACKET_META, | |
69c087ba | 4732 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4733 | PTR_TO_MAP_VALUE, |
4734 | }, | |
4735 | }; | |
4736 | ||
4737 | static const struct bpf_reg_types sock_types = { | |
4738 | .types = { | |
4739 | PTR_TO_SOCK_COMMON, | |
4740 | PTR_TO_SOCKET, | |
4741 | PTR_TO_TCP_SOCK, | |
4742 | PTR_TO_XDP_SOCK, | |
4743 | }, | |
4744 | }; | |
4745 | ||
49a2a4d4 | 4746 | #ifdef CONFIG_NET |
1df8f55a MKL |
4747 | static const struct bpf_reg_types btf_id_sock_common_types = { |
4748 | .types = { | |
4749 | PTR_TO_SOCK_COMMON, | |
4750 | PTR_TO_SOCKET, | |
4751 | PTR_TO_TCP_SOCK, | |
4752 | PTR_TO_XDP_SOCK, | |
4753 | PTR_TO_BTF_ID, | |
4754 | }, | |
4755 | .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
4756 | }; | |
49a2a4d4 | 4757 | #endif |
1df8f55a | 4758 | |
f79e7ea5 LB |
4759 | static const struct bpf_reg_types mem_types = { |
4760 | .types = { | |
4761 | PTR_TO_STACK, | |
4762 | PTR_TO_PACKET, | |
4763 | PTR_TO_PACKET_META, | |
69c087ba | 4764 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4765 | PTR_TO_MAP_VALUE, |
4766 | PTR_TO_MEM, | |
4767 | PTR_TO_RDONLY_BUF, | |
4768 | PTR_TO_RDWR_BUF, | |
4769 | }, | |
4770 | }; | |
4771 | ||
4772 | static const struct bpf_reg_types int_ptr_types = { | |
4773 | .types = { | |
4774 | PTR_TO_STACK, | |
4775 | PTR_TO_PACKET, | |
4776 | PTR_TO_PACKET_META, | |
69c087ba | 4777 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4778 | PTR_TO_MAP_VALUE, |
4779 | }, | |
4780 | }; | |
4781 | ||
4782 | static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } }; | |
4783 | static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } }; | |
4784 | static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } }; | |
4785 | static const struct bpf_reg_types alloc_mem_types = { .types = { PTR_TO_MEM } }; | |
4786 | static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } }; | |
4787 | static const struct bpf_reg_types btf_ptr_types = { .types = { PTR_TO_BTF_ID } }; | |
4788 | static const struct bpf_reg_types spin_lock_types = { .types = { PTR_TO_MAP_VALUE } }; | |
eaa6bcb7 | 4789 | static const struct bpf_reg_types percpu_btf_ptr_types = { .types = { PTR_TO_PERCPU_BTF_ID } }; |
69c087ba YS |
4790 | static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } }; |
4791 | static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } }; | |
fff13c4b | 4792 | static const struct bpf_reg_types const_str_ptr_types = { .types = { PTR_TO_MAP_VALUE } }; |
f79e7ea5 | 4793 | |
0789e13b | 4794 | static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = { |
f79e7ea5 LB |
4795 | [ARG_PTR_TO_MAP_KEY] = &map_key_value_types, |
4796 | [ARG_PTR_TO_MAP_VALUE] = &map_key_value_types, | |
4797 | [ARG_PTR_TO_UNINIT_MAP_VALUE] = &map_key_value_types, | |
4798 | [ARG_PTR_TO_MAP_VALUE_OR_NULL] = &map_key_value_types, | |
4799 | [ARG_CONST_SIZE] = &scalar_types, | |
4800 | [ARG_CONST_SIZE_OR_ZERO] = &scalar_types, | |
4801 | [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types, | |
4802 | [ARG_CONST_MAP_PTR] = &const_map_ptr_types, | |
4803 | [ARG_PTR_TO_CTX] = &context_types, | |
4804 | [ARG_PTR_TO_CTX_OR_NULL] = &context_types, | |
4805 | [ARG_PTR_TO_SOCK_COMMON] = &sock_types, | |
49a2a4d4 | 4806 | #ifdef CONFIG_NET |
1df8f55a | 4807 | [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types, |
49a2a4d4 | 4808 | #endif |
f79e7ea5 LB |
4809 | [ARG_PTR_TO_SOCKET] = &fullsock_types, |
4810 | [ARG_PTR_TO_SOCKET_OR_NULL] = &fullsock_types, | |
4811 | [ARG_PTR_TO_BTF_ID] = &btf_ptr_types, | |
4812 | [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types, | |
4813 | [ARG_PTR_TO_MEM] = &mem_types, | |
4814 | [ARG_PTR_TO_MEM_OR_NULL] = &mem_types, | |
4815 | [ARG_PTR_TO_UNINIT_MEM] = &mem_types, | |
4816 | [ARG_PTR_TO_ALLOC_MEM] = &alloc_mem_types, | |
4817 | [ARG_PTR_TO_ALLOC_MEM_OR_NULL] = &alloc_mem_types, | |
4818 | [ARG_PTR_TO_INT] = &int_ptr_types, | |
4819 | [ARG_PTR_TO_LONG] = &int_ptr_types, | |
eaa6bcb7 | 4820 | [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types, |
69c087ba YS |
4821 | [ARG_PTR_TO_FUNC] = &func_ptr_types, |
4822 | [ARG_PTR_TO_STACK_OR_NULL] = &stack_ptr_types, | |
fff13c4b | 4823 | [ARG_PTR_TO_CONST_STR] = &const_str_ptr_types, |
f79e7ea5 LB |
4824 | }; |
4825 | ||
4826 | static int check_reg_type(struct bpf_verifier_env *env, u32 regno, | |
a968d5e2 MKL |
4827 | enum bpf_arg_type arg_type, |
4828 | const u32 *arg_btf_id) | |
f79e7ea5 LB |
4829 | { |
4830 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
4831 | enum bpf_reg_type expected, type = reg->type; | |
a968d5e2 | 4832 | const struct bpf_reg_types *compatible; |
f79e7ea5 LB |
4833 | int i, j; |
4834 | ||
a968d5e2 MKL |
4835 | compatible = compatible_reg_types[arg_type]; |
4836 | if (!compatible) { | |
4837 | verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type); | |
4838 | return -EFAULT; | |
4839 | } | |
4840 | ||
f79e7ea5 LB |
4841 | for (i = 0; i < ARRAY_SIZE(compatible->types); i++) { |
4842 | expected = compatible->types[i]; | |
4843 | if (expected == NOT_INIT) | |
4844 | break; | |
4845 | ||
4846 | if (type == expected) | |
a968d5e2 | 4847 | goto found; |
f79e7ea5 LB |
4848 | } |
4849 | ||
4850 | verbose(env, "R%d type=%s expected=", regno, reg_type_str[type]); | |
4851 | for (j = 0; j + 1 < i; j++) | |
4852 | verbose(env, "%s, ", reg_type_str[compatible->types[j]]); | |
4853 | verbose(env, "%s\n", reg_type_str[compatible->types[j]]); | |
4854 | return -EACCES; | |
a968d5e2 MKL |
4855 | |
4856 | found: | |
4857 | if (type == PTR_TO_BTF_ID) { | |
1df8f55a MKL |
4858 | if (!arg_btf_id) { |
4859 | if (!compatible->btf_id) { | |
4860 | verbose(env, "verifier internal error: missing arg compatible BTF ID\n"); | |
4861 | return -EFAULT; | |
4862 | } | |
4863 | arg_btf_id = compatible->btf_id; | |
4864 | } | |
4865 | ||
22dc4a0f AN |
4866 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, |
4867 | btf_vmlinux, *arg_btf_id)) { | |
a968d5e2 | 4868 | verbose(env, "R%d is of type %s but %s is expected\n", |
22dc4a0f AN |
4869 | regno, kernel_type_name(reg->btf, reg->btf_id), |
4870 | kernel_type_name(btf_vmlinux, *arg_btf_id)); | |
a968d5e2 MKL |
4871 | return -EACCES; |
4872 | } | |
4873 | ||
4874 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
4875 | verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n", | |
4876 | regno); | |
4877 | return -EACCES; | |
4878 | } | |
4879 | } | |
4880 | ||
4881 | return 0; | |
f79e7ea5 LB |
4882 | } |
4883 | ||
af7ec138 YS |
4884 | static int check_func_arg(struct bpf_verifier_env *env, u32 arg, |
4885 | struct bpf_call_arg_meta *meta, | |
4886 | const struct bpf_func_proto *fn) | |
17a52670 | 4887 | { |
af7ec138 | 4888 | u32 regno = BPF_REG_1 + arg; |
638f5b90 | 4889 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
af7ec138 | 4890 | enum bpf_arg_type arg_type = fn->arg_type[arg]; |
f79e7ea5 | 4891 | enum bpf_reg_type type = reg->type; |
17a52670 AS |
4892 | int err = 0; |
4893 | ||
80f1d68c | 4894 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
4895 | return 0; |
4896 | ||
dc503a8a EC |
4897 | err = check_reg_arg(env, regno, SRC_OP); |
4898 | if (err) | |
4899 | return err; | |
17a52670 | 4900 | |
1be7f75d AS |
4901 | if (arg_type == ARG_ANYTHING) { |
4902 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
4903 | verbose(env, "R%d leaks addr into helper function\n", |
4904 | regno); | |
1be7f75d AS |
4905 | return -EACCES; |
4906 | } | |
80f1d68c | 4907 | return 0; |
1be7f75d | 4908 | } |
80f1d68c | 4909 | |
de8f3a83 | 4910 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 4911 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 4912 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
4913 | return -EACCES; |
4914 | } | |
4915 | ||
912f442c LB |
4916 | if (arg_type == ARG_PTR_TO_MAP_VALUE || |
4917 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE || | |
4918 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) { | |
4919 | err = resolve_map_arg_type(env, meta, &arg_type); | |
4920 | if (err) | |
4921 | return err; | |
4922 | } | |
4923 | ||
fd1b0d60 LB |
4924 | if (register_is_null(reg) && arg_type_may_be_null(arg_type)) |
4925 | /* A NULL register has a SCALAR_VALUE type, so skip | |
4926 | * type checking. | |
4927 | */ | |
4928 | goto skip_type_check; | |
4929 | ||
a968d5e2 | 4930 | err = check_reg_type(env, regno, arg_type, fn->arg_btf_id[arg]); |
f79e7ea5 LB |
4931 | if (err) |
4932 | return err; | |
4933 | ||
a968d5e2 | 4934 | if (type == PTR_TO_CTX) { |
feec7040 LB |
4935 | err = check_ctx_reg(env, reg, regno); |
4936 | if (err < 0) | |
4937 | return err; | |
d7b9454a LB |
4938 | } |
4939 | ||
fd1b0d60 | 4940 | skip_type_check: |
02f7c958 | 4941 | if (reg->ref_obj_id) { |
457f4436 AN |
4942 | if (meta->ref_obj_id) { |
4943 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
4944 | regno, reg->ref_obj_id, | |
4945 | meta->ref_obj_id); | |
4946 | return -EFAULT; | |
4947 | } | |
4948 | meta->ref_obj_id = reg->ref_obj_id; | |
17a52670 AS |
4949 | } |
4950 | ||
17a52670 AS |
4951 | if (arg_type == ARG_CONST_MAP_PTR) { |
4952 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 4953 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
4954 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
4955 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
4956 | * check that [key, key + map->key_size) are within | |
4957 | * stack limits and initialized | |
4958 | */ | |
33ff9823 | 4959 | if (!meta->map_ptr) { |
17a52670 AS |
4960 | /* in function declaration map_ptr must come before |
4961 | * map_key, so that it's verified and known before | |
4962 | * we have to check map_key here. Otherwise it means | |
4963 | * that kernel subsystem misconfigured verifier | |
4964 | */ | |
61bd5218 | 4965 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
4966 | return -EACCES; |
4967 | } | |
d71962f3 PC |
4968 | err = check_helper_mem_access(env, regno, |
4969 | meta->map_ptr->key_size, false, | |
4970 | NULL); | |
2ea864c5 | 4971 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
4972 | (arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL && |
4973 | !register_is_null(reg)) || | |
2ea864c5 | 4974 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { |
17a52670 AS |
4975 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
4976 | * check [value, value + map->value_size) validity | |
4977 | */ | |
33ff9823 | 4978 | if (!meta->map_ptr) { |
17a52670 | 4979 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 4980 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
4981 | return -EACCES; |
4982 | } | |
2ea864c5 | 4983 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE); |
d71962f3 PC |
4984 | err = check_helper_mem_access(env, regno, |
4985 | meta->map_ptr->value_size, false, | |
2ea864c5 | 4986 | meta); |
eaa6bcb7 HL |
4987 | } else if (arg_type == ARG_PTR_TO_PERCPU_BTF_ID) { |
4988 | if (!reg->btf_id) { | |
4989 | verbose(env, "Helper has invalid btf_id in R%d\n", regno); | |
4990 | return -EACCES; | |
4991 | } | |
22dc4a0f | 4992 | meta->ret_btf = reg->btf; |
eaa6bcb7 | 4993 | meta->ret_btf_id = reg->btf_id; |
c18f0b6a LB |
4994 | } else if (arg_type == ARG_PTR_TO_SPIN_LOCK) { |
4995 | if (meta->func_id == BPF_FUNC_spin_lock) { | |
4996 | if (process_spin_lock(env, regno, true)) | |
4997 | return -EACCES; | |
4998 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { | |
4999 | if (process_spin_lock(env, regno, false)) | |
5000 | return -EACCES; | |
5001 | } else { | |
5002 | verbose(env, "verifier internal error\n"); | |
5003 | return -EFAULT; | |
5004 | } | |
69c087ba YS |
5005 | } else if (arg_type == ARG_PTR_TO_FUNC) { |
5006 | meta->subprogno = reg->subprogno; | |
a2bbe7cc LB |
5007 | } else if (arg_type_is_mem_ptr(arg_type)) { |
5008 | /* The access to this pointer is only checked when we hit the | |
5009 | * next is_mem_size argument below. | |
5010 | */ | |
5011 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MEM); | |
90133415 | 5012 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 5013 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 5014 | |
10060503 JF |
5015 | /* This is used to refine r0 return value bounds for helpers |
5016 | * that enforce this value as an upper bound on return values. | |
5017 | * See do_refine_retval_range() for helpers that can refine | |
5018 | * the return value. C type of helper is u32 so we pull register | |
5019 | * bound from umax_value however, if negative verifier errors | |
5020 | * out. Only upper bounds can be learned because retval is an | |
5021 | * int type and negative retvals are allowed. | |
849fa506 | 5022 | */ |
10060503 | 5023 | meta->msize_max_value = reg->umax_value; |
849fa506 | 5024 | |
f1174f77 EC |
5025 | /* The register is SCALAR_VALUE; the access check |
5026 | * happens using its boundaries. | |
06c1c049 | 5027 | */ |
f1174f77 | 5028 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
5029 | /* For unprivileged variable accesses, disable raw |
5030 | * mode so that the program is required to | |
5031 | * initialize all the memory that the helper could | |
5032 | * just partially fill up. | |
5033 | */ | |
5034 | meta = NULL; | |
5035 | ||
b03c9f9f | 5036 | if (reg->smin_value < 0) { |
61bd5218 | 5037 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
5038 | regno); |
5039 | return -EACCES; | |
5040 | } | |
06c1c049 | 5041 | |
b03c9f9f | 5042 | if (reg->umin_value == 0) { |
f1174f77 EC |
5043 | err = check_helper_mem_access(env, regno - 1, 0, |
5044 | zero_size_allowed, | |
5045 | meta); | |
06c1c049 GB |
5046 | if (err) |
5047 | return err; | |
06c1c049 | 5048 | } |
f1174f77 | 5049 | |
b03c9f9f | 5050 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 5051 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
5052 | regno); |
5053 | return -EACCES; | |
5054 | } | |
5055 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 5056 | reg->umax_value, |
f1174f77 | 5057 | zero_size_allowed, meta); |
b5dc0163 AS |
5058 | if (!err) |
5059 | err = mark_chain_precision(env, regno); | |
457f4436 AN |
5060 | } else if (arg_type_is_alloc_size(arg_type)) { |
5061 | if (!tnum_is_const(reg->var_off)) { | |
28a8add6 | 5062 | verbose(env, "R%d is not a known constant'\n", |
457f4436 AN |
5063 | regno); |
5064 | return -EACCES; | |
5065 | } | |
5066 | meta->mem_size = reg->var_off.value; | |
57c3bb72 AI |
5067 | } else if (arg_type_is_int_ptr(arg_type)) { |
5068 | int size = int_ptr_type_to_size(arg_type); | |
5069 | ||
5070 | err = check_helper_mem_access(env, regno, size, false, meta); | |
5071 | if (err) | |
5072 | return err; | |
5073 | err = check_ptr_alignment(env, reg, 0, size, true); | |
fff13c4b FR |
5074 | } else if (arg_type == ARG_PTR_TO_CONST_STR) { |
5075 | struct bpf_map *map = reg->map_ptr; | |
5076 | int map_off; | |
5077 | u64 map_addr; | |
5078 | char *str_ptr; | |
5079 | ||
a8fad73e | 5080 | if (!bpf_map_is_rdonly(map)) { |
fff13c4b FR |
5081 | verbose(env, "R%d does not point to a readonly map'\n", regno); |
5082 | return -EACCES; | |
5083 | } | |
5084 | ||
5085 | if (!tnum_is_const(reg->var_off)) { | |
5086 | verbose(env, "R%d is not a constant address'\n", regno); | |
5087 | return -EACCES; | |
5088 | } | |
5089 | ||
5090 | if (!map->ops->map_direct_value_addr) { | |
5091 | verbose(env, "no direct value access support for this map type\n"); | |
5092 | return -EACCES; | |
5093 | } | |
5094 | ||
5095 | err = check_map_access(env, regno, reg->off, | |
5096 | map->value_size - reg->off, false); | |
5097 | if (err) | |
5098 | return err; | |
5099 | ||
5100 | map_off = reg->off + reg->var_off.value; | |
5101 | err = map->ops->map_direct_value_addr(map, &map_addr, map_off); | |
5102 | if (err) { | |
5103 | verbose(env, "direct value access on string failed\n"); | |
5104 | return err; | |
5105 | } | |
5106 | ||
5107 | str_ptr = (char *)(long)(map_addr); | |
5108 | if (!strnchr(str_ptr + map_off, map->value_size - map_off, 0)) { | |
5109 | verbose(env, "string is not zero-terminated\n"); | |
5110 | return -EINVAL; | |
5111 | } | |
17a52670 AS |
5112 | } |
5113 | ||
5114 | return err; | |
5115 | } | |
5116 | ||
0126240f LB |
5117 | static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) |
5118 | { | |
5119 | enum bpf_attach_type eatype = env->prog->expected_attach_type; | |
7e40781c | 5120 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
0126240f LB |
5121 | |
5122 | if (func_id != BPF_FUNC_map_update_elem) | |
5123 | return false; | |
5124 | ||
5125 | /* It's not possible to get access to a locked struct sock in these | |
5126 | * contexts, so updating is safe. | |
5127 | */ | |
5128 | switch (type) { | |
5129 | case BPF_PROG_TYPE_TRACING: | |
5130 | if (eatype == BPF_TRACE_ITER) | |
5131 | return true; | |
5132 | break; | |
5133 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
5134 | case BPF_PROG_TYPE_SCHED_CLS: | |
5135 | case BPF_PROG_TYPE_SCHED_ACT: | |
5136 | case BPF_PROG_TYPE_XDP: | |
5137 | case BPF_PROG_TYPE_SK_REUSEPORT: | |
5138 | case BPF_PROG_TYPE_FLOW_DISSECTOR: | |
5139 | case BPF_PROG_TYPE_SK_LOOKUP: | |
5140 | return true; | |
5141 | default: | |
5142 | break; | |
5143 | } | |
5144 | ||
5145 | verbose(env, "cannot update sockmap in this context\n"); | |
5146 | return false; | |
5147 | } | |
5148 | ||
e411901c MF |
5149 | static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env) |
5150 | { | |
5151 | return env->prog->jit_requested && IS_ENABLED(CONFIG_X86_64); | |
5152 | } | |
5153 | ||
61bd5218 JK |
5154 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
5155 | struct bpf_map *map, int func_id) | |
35578d79 | 5156 | { |
35578d79 KX |
5157 | if (!map) |
5158 | return 0; | |
5159 | ||
6aff67c8 AS |
5160 | /* We need a two way check, first is from map perspective ... */ |
5161 | switch (map->map_type) { | |
5162 | case BPF_MAP_TYPE_PROG_ARRAY: | |
5163 | if (func_id != BPF_FUNC_tail_call) | |
5164 | goto error; | |
5165 | break; | |
5166 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
5167 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 5168 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 5169 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
5170 | func_id != BPF_FUNC_perf_event_read_value && |
5171 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
5172 | goto error; |
5173 | break; | |
457f4436 AN |
5174 | case BPF_MAP_TYPE_RINGBUF: |
5175 | if (func_id != BPF_FUNC_ringbuf_output && | |
5176 | func_id != BPF_FUNC_ringbuf_reserve && | |
5177 | func_id != BPF_FUNC_ringbuf_submit && | |
5178 | func_id != BPF_FUNC_ringbuf_discard && | |
5179 | func_id != BPF_FUNC_ringbuf_query) | |
5180 | goto error; | |
5181 | break; | |
6aff67c8 AS |
5182 | case BPF_MAP_TYPE_STACK_TRACE: |
5183 | if (func_id != BPF_FUNC_get_stackid) | |
5184 | goto error; | |
5185 | break; | |
4ed8ec52 | 5186 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 5187 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 5188 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
5189 | goto error; |
5190 | break; | |
cd339431 | 5191 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 5192 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
5193 | if (func_id != BPF_FUNC_get_local_storage) |
5194 | goto error; | |
5195 | break; | |
546ac1ff | 5196 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 5197 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
5198 | if (func_id != BPF_FUNC_redirect_map && |
5199 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
5200 | goto error; |
5201 | break; | |
fbfc504a BT |
5202 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
5203 | * appear. | |
5204 | */ | |
6710e112 JDB |
5205 | case BPF_MAP_TYPE_CPUMAP: |
5206 | if (func_id != BPF_FUNC_redirect_map) | |
5207 | goto error; | |
5208 | break; | |
fada7fdc JL |
5209 | case BPF_MAP_TYPE_XSKMAP: |
5210 | if (func_id != BPF_FUNC_redirect_map && | |
5211 | func_id != BPF_FUNC_map_lookup_elem) | |
5212 | goto error; | |
5213 | break; | |
56f668df | 5214 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 5215 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
5216 | if (func_id != BPF_FUNC_map_lookup_elem) |
5217 | goto error; | |
16a43625 | 5218 | break; |
174a79ff JF |
5219 | case BPF_MAP_TYPE_SOCKMAP: |
5220 | if (func_id != BPF_FUNC_sk_redirect_map && | |
5221 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 5222 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 5223 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 | 5224 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
5225 | func_id != BPF_FUNC_map_lookup_elem && |
5226 | !may_update_sockmap(env, func_id)) | |
174a79ff JF |
5227 | goto error; |
5228 | break; | |
81110384 JF |
5229 | case BPF_MAP_TYPE_SOCKHASH: |
5230 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
5231 | func_id != BPF_FUNC_sock_hash_update && | |
5232 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 5233 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 | 5234 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
5235 | func_id != BPF_FUNC_map_lookup_elem && |
5236 | !may_update_sockmap(env, func_id)) | |
81110384 JF |
5237 | goto error; |
5238 | break; | |
2dbb9b9e MKL |
5239 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
5240 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
5241 | goto error; | |
5242 | break; | |
f1a2e44a MV |
5243 | case BPF_MAP_TYPE_QUEUE: |
5244 | case BPF_MAP_TYPE_STACK: | |
5245 | if (func_id != BPF_FUNC_map_peek_elem && | |
5246 | func_id != BPF_FUNC_map_pop_elem && | |
5247 | func_id != BPF_FUNC_map_push_elem) | |
5248 | goto error; | |
5249 | break; | |
6ac99e8f MKL |
5250 | case BPF_MAP_TYPE_SK_STORAGE: |
5251 | if (func_id != BPF_FUNC_sk_storage_get && | |
5252 | func_id != BPF_FUNC_sk_storage_delete) | |
5253 | goto error; | |
5254 | break; | |
8ea63684 KS |
5255 | case BPF_MAP_TYPE_INODE_STORAGE: |
5256 | if (func_id != BPF_FUNC_inode_storage_get && | |
5257 | func_id != BPF_FUNC_inode_storage_delete) | |
5258 | goto error; | |
5259 | break; | |
4cf1bc1f KS |
5260 | case BPF_MAP_TYPE_TASK_STORAGE: |
5261 | if (func_id != BPF_FUNC_task_storage_get && | |
5262 | func_id != BPF_FUNC_task_storage_delete) | |
5263 | goto error; | |
5264 | break; | |
6aff67c8 AS |
5265 | default: |
5266 | break; | |
5267 | } | |
5268 | ||
5269 | /* ... and second from the function itself. */ | |
5270 | switch (func_id) { | |
5271 | case BPF_FUNC_tail_call: | |
5272 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
5273 | goto error; | |
e411901c MF |
5274 | if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) { |
5275 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
f4d7e40a AS |
5276 | return -EINVAL; |
5277 | } | |
6aff67c8 AS |
5278 | break; |
5279 | case BPF_FUNC_perf_event_read: | |
5280 | case BPF_FUNC_perf_event_output: | |
908432ca | 5281 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 5282 | case BPF_FUNC_skb_output: |
d831ee84 | 5283 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
5284 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
5285 | goto error; | |
5286 | break; | |
5287 | case BPF_FUNC_get_stackid: | |
5288 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
5289 | goto error; | |
5290 | break; | |
60d20f91 | 5291 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 5292 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
5293 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
5294 | goto error; | |
5295 | break; | |
97f91a7c | 5296 | case BPF_FUNC_redirect_map: |
9c270af3 | 5297 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 5298 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
5299 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
5300 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
5301 | goto error; |
5302 | break; | |
174a79ff | 5303 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 5304 | case BPF_FUNC_msg_redirect_map: |
81110384 | 5305 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
5306 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
5307 | goto error; | |
5308 | break; | |
81110384 JF |
5309 | case BPF_FUNC_sk_redirect_hash: |
5310 | case BPF_FUNC_msg_redirect_hash: | |
5311 | case BPF_FUNC_sock_hash_update: | |
5312 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
5313 | goto error; |
5314 | break; | |
cd339431 | 5315 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
5316 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
5317 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
5318 | goto error; |
5319 | break; | |
2dbb9b9e | 5320 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
5321 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
5322 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
5323 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
5324 | goto error; |
5325 | break; | |
f1a2e44a MV |
5326 | case BPF_FUNC_map_peek_elem: |
5327 | case BPF_FUNC_map_pop_elem: | |
5328 | case BPF_FUNC_map_push_elem: | |
5329 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
5330 | map->map_type != BPF_MAP_TYPE_STACK) | |
5331 | goto error; | |
5332 | break; | |
6ac99e8f MKL |
5333 | case BPF_FUNC_sk_storage_get: |
5334 | case BPF_FUNC_sk_storage_delete: | |
5335 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
5336 | goto error; | |
5337 | break; | |
8ea63684 KS |
5338 | case BPF_FUNC_inode_storage_get: |
5339 | case BPF_FUNC_inode_storage_delete: | |
5340 | if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE) | |
5341 | goto error; | |
5342 | break; | |
4cf1bc1f KS |
5343 | case BPF_FUNC_task_storage_get: |
5344 | case BPF_FUNC_task_storage_delete: | |
5345 | if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) | |
5346 | goto error; | |
5347 | break; | |
6aff67c8 AS |
5348 | default: |
5349 | break; | |
35578d79 KX |
5350 | } |
5351 | ||
5352 | return 0; | |
6aff67c8 | 5353 | error: |
61bd5218 | 5354 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 5355 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 5356 | return -EINVAL; |
35578d79 KX |
5357 | } |
5358 | ||
90133415 | 5359 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
5360 | { |
5361 | int count = 0; | |
5362 | ||
39f19ebb | 5363 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5364 | count++; |
39f19ebb | 5365 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5366 | count++; |
39f19ebb | 5367 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5368 | count++; |
39f19ebb | 5369 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5370 | count++; |
39f19ebb | 5371 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
5372 | count++; |
5373 | ||
90133415 DB |
5374 | /* We only support one arg being in raw mode at the moment, |
5375 | * which is sufficient for the helper functions we have | |
5376 | * right now. | |
5377 | */ | |
5378 | return count <= 1; | |
5379 | } | |
5380 | ||
5381 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
5382 | enum bpf_arg_type arg_next) | |
5383 | { | |
5384 | return (arg_type_is_mem_ptr(arg_curr) && | |
5385 | !arg_type_is_mem_size(arg_next)) || | |
5386 | (!arg_type_is_mem_ptr(arg_curr) && | |
5387 | arg_type_is_mem_size(arg_next)); | |
5388 | } | |
5389 | ||
5390 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
5391 | { | |
5392 | /* bpf_xxx(..., buf, len) call will access 'len' | |
5393 | * bytes from memory 'buf'. Both arg types need | |
5394 | * to be paired, so make sure there's no buggy | |
5395 | * helper function specification. | |
5396 | */ | |
5397 | if (arg_type_is_mem_size(fn->arg1_type) || | |
5398 | arg_type_is_mem_ptr(fn->arg5_type) || | |
5399 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
5400 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
5401 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
5402 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
5403 | return false; | |
5404 | ||
5405 | return true; | |
5406 | } | |
5407 | ||
1b986589 | 5408 | static bool check_refcount_ok(const struct bpf_func_proto *fn, int func_id) |
fd978bf7 JS |
5409 | { |
5410 | int count = 0; | |
5411 | ||
1b986589 | 5412 | if (arg_type_may_be_refcounted(fn->arg1_type)) |
fd978bf7 | 5413 | count++; |
1b986589 | 5414 | if (arg_type_may_be_refcounted(fn->arg2_type)) |
fd978bf7 | 5415 | count++; |
1b986589 | 5416 | if (arg_type_may_be_refcounted(fn->arg3_type)) |
fd978bf7 | 5417 | count++; |
1b986589 | 5418 | if (arg_type_may_be_refcounted(fn->arg4_type)) |
fd978bf7 | 5419 | count++; |
1b986589 | 5420 | if (arg_type_may_be_refcounted(fn->arg5_type)) |
fd978bf7 JS |
5421 | count++; |
5422 | ||
1b986589 MKL |
5423 | /* A reference acquiring function cannot acquire |
5424 | * another refcounted ptr. | |
5425 | */ | |
64d85290 | 5426 | if (may_be_acquire_function(func_id) && count) |
1b986589 MKL |
5427 | return false; |
5428 | ||
fd978bf7 JS |
5429 | /* We only support one arg being unreferenced at the moment, |
5430 | * which is sufficient for the helper functions we have right now. | |
5431 | */ | |
5432 | return count <= 1; | |
5433 | } | |
5434 | ||
9436ef6e LB |
5435 | static bool check_btf_id_ok(const struct bpf_func_proto *fn) |
5436 | { | |
5437 | int i; | |
5438 | ||
1df8f55a | 5439 | for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) { |
9436ef6e LB |
5440 | if (fn->arg_type[i] == ARG_PTR_TO_BTF_ID && !fn->arg_btf_id[i]) |
5441 | return false; | |
5442 | ||
1df8f55a MKL |
5443 | if (fn->arg_type[i] != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i]) |
5444 | return false; | |
5445 | } | |
5446 | ||
9436ef6e LB |
5447 | return true; |
5448 | } | |
5449 | ||
1b986589 | 5450 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
5451 | { |
5452 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 5453 | check_arg_pair_ok(fn) && |
9436ef6e | 5454 | check_btf_id_ok(fn) && |
1b986589 | 5455 | check_refcount_ok(fn, func_id) ? 0 : -EINVAL; |
435faee1 DB |
5456 | } |
5457 | ||
de8f3a83 DB |
5458 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
5459 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 5460 | */ |
f4d7e40a AS |
5461 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
5462 | struct bpf_func_state *state) | |
969bf05e | 5463 | { |
58e2af8b | 5464 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
5465 | int i; |
5466 | ||
5467 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 5468 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 5469 | mark_reg_unknown(env, regs, i); |
969bf05e | 5470 | |
f3709f69 JS |
5471 | bpf_for_each_spilled_reg(i, state, reg) { |
5472 | if (!reg) | |
969bf05e | 5473 | continue; |
de8f3a83 | 5474 | if (reg_is_pkt_pointer_any(reg)) |
f54c7898 | 5475 | __mark_reg_unknown(env, reg); |
969bf05e AS |
5476 | } |
5477 | } | |
5478 | ||
f4d7e40a AS |
5479 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
5480 | { | |
5481 | struct bpf_verifier_state *vstate = env->cur_state; | |
5482 | int i; | |
5483 | ||
5484 | for (i = 0; i <= vstate->curframe; i++) | |
5485 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
5486 | } | |
5487 | ||
6d94e741 AS |
5488 | enum { |
5489 | AT_PKT_END = -1, | |
5490 | BEYOND_PKT_END = -2, | |
5491 | }; | |
5492 | ||
5493 | static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open) | |
5494 | { | |
5495 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5496 | struct bpf_reg_state *reg = &state->regs[regn]; | |
5497 | ||
5498 | if (reg->type != PTR_TO_PACKET) | |
5499 | /* PTR_TO_PACKET_META is not supported yet */ | |
5500 | return; | |
5501 | ||
5502 | /* The 'reg' is pkt > pkt_end or pkt >= pkt_end. | |
5503 | * How far beyond pkt_end it goes is unknown. | |
5504 | * if (!range_open) it's the case of pkt >= pkt_end | |
5505 | * if (range_open) it's the case of pkt > pkt_end | |
5506 | * hence this pointer is at least 1 byte bigger than pkt_end | |
5507 | */ | |
5508 | if (range_open) | |
5509 | reg->range = BEYOND_PKT_END; | |
5510 | else | |
5511 | reg->range = AT_PKT_END; | |
5512 | } | |
5513 | ||
fd978bf7 | 5514 | static void release_reg_references(struct bpf_verifier_env *env, |
1b986589 MKL |
5515 | struct bpf_func_state *state, |
5516 | int ref_obj_id) | |
fd978bf7 JS |
5517 | { |
5518 | struct bpf_reg_state *regs = state->regs, *reg; | |
5519 | int i; | |
5520 | ||
5521 | for (i = 0; i < MAX_BPF_REG; i++) | |
1b986589 | 5522 | if (regs[i].ref_obj_id == ref_obj_id) |
fd978bf7 JS |
5523 | mark_reg_unknown(env, regs, i); |
5524 | ||
5525 | bpf_for_each_spilled_reg(i, state, reg) { | |
5526 | if (!reg) | |
5527 | continue; | |
1b986589 | 5528 | if (reg->ref_obj_id == ref_obj_id) |
f54c7898 | 5529 | __mark_reg_unknown(env, reg); |
fd978bf7 JS |
5530 | } |
5531 | } | |
5532 | ||
5533 | /* The pointer with the specified id has released its reference to kernel | |
5534 | * resources. Identify all copies of the same pointer and clear the reference. | |
5535 | */ | |
5536 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 5537 | int ref_obj_id) |
fd978bf7 JS |
5538 | { |
5539 | struct bpf_verifier_state *vstate = env->cur_state; | |
1b986589 | 5540 | int err; |
fd978bf7 JS |
5541 | int i; |
5542 | ||
1b986589 MKL |
5543 | err = release_reference_state(cur_func(env), ref_obj_id); |
5544 | if (err) | |
5545 | return err; | |
5546 | ||
fd978bf7 | 5547 | for (i = 0; i <= vstate->curframe; i++) |
1b986589 | 5548 | release_reg_references(env, vstate->frame[i], ref_obj_id); |
fd978bf7 | 5549 | |
1b986589 | 5550 | return 0; |
fd978bf7 JS |
5551 | } |
5552 | ||
51c39bb1 AS |
5553 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
5554 | struct bpf_reg_state *regs) | |
5555 | { | |
5556 | int i; | |
5557 | ||
5558 | /* after the call registers r0 - r5 were scratched */ | |
5559 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
5560 | mark_reg_not_init(env, regs, caller_saved[i]); | |
5561 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
5562 | } | |
5563 | } | |
5564 | ||
14351375 YS |
5565 | typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env, |
5566 | struct bpf_func_state *caller, | |
5567 | struct bpf_func_state *callee, | |
5568 | int insn_idx); | |
5569 | ||
5570 | static int __check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
5571 | int *insn_idx, int subprog, | |
5572 | set_callee_state_fn set_callee_state_cb) | |
f4d7e40a AS |
5573 | { |
5574 | struct bpf_verifier_state *state = env->cur_state; | |
51c39bb1 | 5575 | struct bpf_func_info_aux *func_info_aux; |
f4d7e40a | 5576 | struct bpf_func_state *caller, *callee; |
14351375 | 5577 | int err; |
51c39bb1 | 5578 | bool is_global = false; |
f4d7e40a | 5579 | |
aada9ce6 | 5580 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 5581 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 5582 | state->curframe + 2); |
f4d7e40a AS |
5583 | return -E2BIG; |
5584 | } | |
5585 | ||
f4d7e40a AS |
5586 | caller = state->frame[state->curframe]; |
5587 | if (state->frame[state->curframe + 1]) { | |
5588 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
5589 | state->curframe + 1); | |
5590 | return -EFAULT; | |
5591 | } | |
5592 | ||
51c39bb1 AS |
5593 | func_info_aux = env->prog->aux->func_info_aux; |
5594 | if (func_info_aux) | |
5595 | is_global = func_info_aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
34747c41 | 5596 | err = btf_check_subprog_arg_match(env, subprog, caller->regs); |
51c39bb1 AS |
5597 | if (err == -EFAULT) |
5598 | return err; | |
5599 | if (is_global) { | |
5600 | if (err) { | |
5601 | verbose(env, "Caller passes invalid args into func#%d\n", | |
5602 | subprog); | |
5603 | return err; | |
5604 | } else { | |
5605 | if (env->log.level & BPF_LOG_LEVEL) | |
5606 | verbose(env, | |
5607 | "Func#%d is global and valid. Skipping.\n", | |
5608 | subprog); | |
5609 | clear_caller_saved_regs(env, caller->regs); | |
5610 | ||
45159b27 | 5611 | /* All global functions return a 64-bit SCALAR_VALUE */ |
51c39bb1 | 5612 | mark_reg_unknown(env, caller->regs, BPF_REG_0); |
45159b27 | 5613 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; |
51c39bb1 AS |
5614 | |
5615 | /* continue with next insn after call */ | |
5616 | return 0; | |
5617 | } | |
5618 | } | |
5619 | ||
f4d7e40a AS |
5620 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); |
5621 | if (!callee) | |
5622 | return -ENOMEM; | |
5623 | state->frame[state->curframe + 1] = callee; | |
5624 | ||
5625 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
5626 | * into its own stack before reading from it. | |
5627 | * callee can read/write into caller's stack | |
5628 | */ | |
5629 | init_func_state(env, callee, | |
5630 | /* remember the callsite, it will be used by bpf_exit */ | |
5631 | *insn_idx /* callsite */, | |
5632 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 5633 | subprog /* subprog number within this prog */); |
f4d7e40a | 5634 | |
fd978bf7 | 5635 | /* Transfer references to the callee */ |
c69431aa | 5636 | err = copy_reference_state(callee, caller); |
fd978bf7 JS |
5637 | if (err) |
5638 | return err; | |
5639 | ||
14351375 YS |
5640 | err = set_callee_state_cb(env, caller, callee, *insn_idx); |
5641 | if (err) | |
5642 | return err; | |
f4d7e40a | 5643 | |
51c39bb1 | 5644 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a AS |
5645 | |
5646 | /* only increment it after check_reg_arg() finished */ | |
5647 | state->curframe++; | |
5648 | ||
5649 | /* and go analyze first insn of the callee */ | |
14351375 | 5650 | *insn_idx = env->subprog_info[subprog].start - 1; |
f4d7e40a | 5651 | |
06ee7115 | 5652 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
5653 | verbose(env, "caller:\n"); |
5654 | print_verifier_state(env, caller); | |
5655 | verbose(env, "callee:\n"); | |
5656 | print_verifier_state(env, callee); | |
5657 | } | |
5658 | return 0; | |
5659 | } | |
5660 | ||
314ee05e YS |
5661 | int map_set_for_each_callback_args(struct bpf_verifier_env *env, |
5662 | struct bpf_func_state *caller, | |
5663 | struct bpf_func_state *callee) | |
5664 | { | |
5665 | /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn, | |
5666 | * void *callback_ctx, u64 flags); | |
5667 | * callback_fn(struct bpf_map *map, void *key, void *value, | |
5668 | * void *callback_ctx); | |
5669 | */ | |
5670 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
5671 | ||
5672 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
5673 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
5674 | callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
5675 | ||
5676 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
5677 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
5678 | callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
5679 | ||
5680 | /* pointer to stack or null */ | |
5681 | callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3]; | |
5682 | ||
5683 | /* unused */ | |
5684 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
5685 | return 0; | |
5686 | } | |
5687 | ||
14351375 YS |
5688 | static int set_callee_state(struct bpf_verifier_env *env, |
5689 | struct bpf_func_state *caller, | |
5690 | struct bpf_func_state *callee, int insn_idx) | |
5691 | { | |
5692 | int i; | |
5693 | ||
5694 | /* copy r1 - r5 args that callee can access. The copy includes parent | |
5695 | * pointers, which connects us up to the liveness chain | |
5696 | */ | |
5697 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
5698 | callee->regs[i] = caller->regs[i]; | |
5699 | return 0; | |
5700 | } | |
5701 | ||
5702 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
5703 | int *insn_idx) | |
5704 | { | |
5705 | int subprog, target_insn; | |
5706 | ||
5707 | target_insn = *insn_idx + insn->imm + 1; | |
5708 | subprog = find_subprog(env, target_insn); | |
5709 | if (subprog < 0) { | |
5710 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
5711 | target_insn); | |
5712 | return -EFAULT; | |
5713 | } | |
5714 | ||
5715 | return __check_func_call(env, insn, insn_idx, subprog, set_callee_state); | |
5716 | } | |
5717 | ||
69c087ba YS |
5718 | static int set_map_elem_callback_state(struct bpf_verifier_env *env, |
5719 | struct bpf_func_state *caller, | |
5720 | struct bpf_func_state *callee, | |
5721 | int insn_idx) | |
5722 | { | |
5723 | struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx]; | |
5724 | struct bpf_map *map; | |
5725 | int err; | |
5726 | ||
5727 | if (bpf_map_ptr_poisoned(insn_aux)) { | |
5728 | verbose(env, "tail_call abusing map_ptr\n"); | |
5729 | return -EINVAL; | |
5730 | } | |
5731 | ||
5732 | map = BPF_MAP_PTR(insn_aux->map_ptr_state); | |
5733 | if (!map->ops->map_set_for_each_callback_args || | |
5734 | !map->ops->map_for_each_callback) { | |
5735 | verbose(env, "callback function not allowed for map\n"); | |
5736 | return -ENOTSUPP; | |
5737 | } | |
5738 | ||
5739 | err = map->ops->map_set_for_each_callback_args(env, caller, callee); | |
5740 | if (err) | |
5741 | return err; | |
5742 | ||
5743 | callee->in_callback_fn = true; | |
5744 | return 0; | |
5745 | } | |
5746 | ||
f4d7e40a AS |
5747 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) |
5748 | { | |
5749 | struct bpf_verifier_state *state = env->cur_state; | |
5750 | struct bpf_func_state *caller, *callee; | |
5751 | struct bpf_reg_state *r0; | |
fd978bf7 | 5752 | int err; |
f4d7e40a AS |
5753 | |
5754 | callee = state->frame[state->curframe]; | |
5755 | r0 = &callee->regs[BPF_REG_0]; | |
5756 | if (r0->type == PTR_TO_STACK) { | |
5757 | /* technically it's ok to return caller's stack pointer | |
5758 | * (or caller's caller's pointer) back to the caller, | |
5759 | * since these pointers are valid. Only current stack | |
5760 | * pointer will be invalid as soon as function exits, | |
5761 | * but let's be conservative | |
5762 | */ | |
5763 | verbose(env, "cannot return stack pointer to the caller\n"); | |
5764 | return -EINVAL; | |
5765 | } | |
5766 | ||
5767 | state->curframe--; | |
5768 | caller = state->frame[state->curframe]; | |
69c087ba YS |
5769 | if (callee->in_callback_fn) { |
5770 | /* enforce R0 return value range [0, 1]. */ | |
5771 | struct tnum range = tnum_range(0, 1); | |
5772 | ||
5773 | if (r0->type != SCALAR_VALUE) { | |
5774 | verbose(env, "R0 not a scalar value\n"); | |
5775 | return -EACCES; | |
5776 | } | |
5777 | if (!tnum_in(range, r0->var_off)) { | |
5778 | verbose_invalid_scalar(env, r0, &range, "callback return", "R0"); | |
5779 | return -EINVAL; | |
5780 | } | |
5781 | } else { | |
5782 | /* return to the caller whatever r0 had in the callee */ | |
5783 | caller->regs[BPF_REG_0] = *r0; | |
5784 | } | |
f4d7e40a | 5785 | |
fd978bf7 | 5786 | /* Transfer references to the caller */ |
c69431aa | 5787 | err = copy_reference_state(caller, callee); |
fd978bf7 JS |
5788 | if (err) |
5789 | return err; | |
5790 | ||
f4d7e40a | 5791 | *insn_idx = callee->callsite + 1; |
06ee7115 | 5792 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
5793 | verbose(env, "returning from callee:\n"); |
5794 | print_verifier_state(env, callee); | |
5795 | verbose(env, "to caller at %d:\n", *insn_idx); | |
5796 | print_verifier_state(env, caller); | |
5797 | } | |
5798 | /* clear everything in the callee */ | |
5799 | free_func_state(callee); | |
5800 | state->frame[state->curframe + 1] = NULL; | |
5801 | return 0; | |
5802 | } | |
5803 | ||
849fa506 YS |
5804 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
5805 | int func_id, | |
5806 | struct bpf_call_arg_meta *meta) | |
5807 | { | |
5808 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
5809 | ||
5810 | if (ret_type != RET_INTEGER || | |
5811 | (func_id != BPF_FUNC_get_stack && | |
fd0b88f7 | 5812 | func_id != BPF_FUNC_get_task_stack && |
47cc0ed5 DB |
5813 | func_id != BPF_FUNC_probe_read_str && |
5814 | func_id != BPF_FUNC_probe_read_kernel_str && | |
5815 | func_id != BPF_FUNC_probe_read_user_str)) | |
849fa506 YS |
5816 | return; |
5817 | ||
10060503 | 5818 | ret_reg->smax_value = meta->msize_max_value; |
fa123ac0 | 5819 | ret_reg->s32_max_value = meta->msize_max_value; |
b0270958 AS |
5820 | ret_reg->smin_value = -MAX_ERRNO; |
5821 | ret_reg->s32_min_value = -MAX_ERRNO; | |
849fa506 YS |
5822 | __reg_deduce_bounds(ret_reg); |
5823 | __reg_bound_offset(ret_reg); | |
10060503 | 5824 | __update_reg_bounds(ret_reg); |
849fa506 YS |
5825 | } |
5826 | ||
c93552c4 DB |
5827 | static int |
5828 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
5829 | int func_id, int insn_idx) | |
5830 | { | |
5831 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 5832 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
5833 | |
5834 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
5835 | func_id != BPF_FUNC_map_lookup_elem && |
5836 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
5837 | func_id != BPF_FUNC_map_delete_elem && |
5838 | func_id != BPF_FUNC_map_push_elem && | |
5839 | func_id != BPF_FUNC_map_pop_elem && | |
69c087ba | 5840 | func_id != BPF_FUNC_map_peek_elem && |
e6a4750f BT |
5841 | func_id != BPF_FUNC_for_each_map_elem && |
5842 | func_id != BPF_FUNC_redirect_map) | |
c93552c4 | 5843 | return 0; |
09772d92 | 5844 | |
591fe988 | 5845 | if (map == NULL) { |
c93552c4 DB |
5846 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
5847 | return -EINVAL; | |
5848 | } | |
5849 | ||
591fe988 DB |
5850 | /* In case of read-only, some additional restrictions |
5851 | * need to be applied in order to prevent altering the | |
5852 | * state of the map from program side. | |
5853 | */ | |
5854 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
5855 | (func_id == BPF_FUNC_map_delete_elem || | |
5856 | func_id == BPF_FUNC_map_update_elem || | |
5857 | func_id == BPF_FUNC_map_push_elem || | |
5858 | func_id == BPF_FUNC_map_pop_elem)) { | |
5859 | verbose(env, "write into map forbidden\n"); | |
5860 | return -EACCES; | |
5861 | } | |
5862 | ||
d2e4c1e6 | 5863 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 5864 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 5865 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 5866 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 5867 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 5868 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
5869 | return 0; |
5870 | } | |
5871 | ||
d2e4c1e6 DB |
5872 | static int |
5873 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
5874 | int func_id, int insn_idx) | |
5875 | { | |
5876 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
5877 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
5878 | struct bpf_map *map = meta->map_ptr; | |
5879 | struct tnum range; | |
5880 | u64 val; | |
cc52d914 | 5881 | int err; |
d2e4c1e6 DB |
5882 | |
5883 | if (func_id != BPF_FUNC_tail_call) | |
5884 | return 0; | |
5885 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
5886 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
5887 | return -EINVAL; | |
5888 | } | |
5889 | ||
5890 | range = tnum_range(0, map->max_entries - 1); | |
5891 | reg = ®s[BPF_REG_3]; | |
5892 | ||
5893 | if (!register_is_const(reg) || !tnum_in(range, reg->var_off)) { | |
5894 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
5895 | return 0; | |
5896 | } | |
5897 | ||
cc52d914 DB |
5898 | err = mark_chain_precision(env, BPF_REG_3); |
5899 | if (err) | |
5900 | return err; | |
5901 | ||
d2e4c1e6 DB |
5902 | val = reg->var_off.value; |
5903 | if (bpf_map_key_unseen(aux)) | |
5904 | bpf_map_key_store(aux, val); | |
5905 | else if (!bpf_map_key_poisoned(aux) && | |
5906 | bpf_map_key_immediate(aux) != val) | |
5907 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
5908 | return 0; | |
5909 | } | |
5910 | ||
fd978bf7 JS |
5911 | static int check_reference_leak(struct bpf_verifier_env *env) |
5912 | { | |
5913 | struct bpf_func_state *state = cur_func(env); | |
5914 | int i; | |
5915 | ||
5916 | for (i = 0; i < state->acquired_refs; i++) { | |
5917 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", | |
5918 | state->refs[i].id, state->refs[i].insn_idx); | |
5919 | } | |
5920 | return state->acquired_refs ? -EINVAL : 0; | |
5921 | } | |
5922 | ||
7b15523a FR |
5923 | static int check_bpf_snprintf_call(struct bpf_verifier_env *env, |
5924 | struct bpf_reg_state *regs) | |
5925 | { | |
5926 | struct bpf_reg_state *fmt_reg = ®s[BPF_REG_3]; | |
5927 | struct bpf_reg_state *data_len_reg = ®s[BPF_REG_5]; | |
5928 | struct bpf_map *fmt_map = fmt_reg->map_ptr; | |
5929 | int err, fmt_map_off, num_args; | |
5930 | u64 fmt_addr; | |
5931 | char *fmt; | |
5932 | ||
5933 | /* data must be an array of u64 */ | |
5934 | if (data_len_reg->var_off.value % 8) | |
5935 | return -EINVAL; | |
5936 | num_args = data_len_reg->var_off.value / 8; | |
5937 | ||
5938 | /* fmt being ARG_PTR_TO_CONST_STR guarantees that var_off is const | |
5939 | * and map_direct_value_addr is set. | |
5940 | */ | |
5941 | fmt_map_off = fmt_reg->off + fmt_reg->var_off.value; | |
5942 | err = fmt_map->ops->map_direct_value_addr(fmt_map, &fmt_addr, | |
5943 | fmt_map_off); | |
8e8ee109 FR |
5944 | if (err) { |
5945 | verbose(env, "verifier bug\n"); | |
5946 | return -EFAULT; | |
5947 | } | |
7b15523a FR |
5948 | fmt = (char *)(long)fmt_addr + fmt_map_off; |
5949 | ||
5950 | /* We are also guaranteed that fmt+fmt_map_off is NULL terminated, we | |
5951 | * can focus on validating the format specifiers. | |
5952 | */ | |
48cac3f4 | 5953 | err = bpf_bprintf_prepare(fmt, UINT_MAX, NULL, NULL, num_args); |
7b15523a FR |
5954 | if (err < 0) |
5955 | verbose(env, "Invalid format string\n"); | |
5956 | ||
5957 | return err; | |
5958 | } | |
5959 | ||
69c087ba YS |
5960 | static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
5961 | int *insn_idx_p) | |
17a52670 | 5962 | { |
17a52670 | 5963 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 5964 | struct bpf_reg_state *regs; |
33ff9823 | 5965 | struct bpf_call_arg_meta meta; |
69c087ba | 5966 | int insn_idx = *insn_idx_p; |
969bf05e | 5967 | bool changes_data; |
69c087ba | 5968 | int i, err, func_id; |
17a52670 AS |
5969 | |
5970 | /* find function prototype */ | |
69c087ba | 5971 | func_id = insn->imm; |
17a52670 | 5972 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { |
61bd5218 JK |
5973 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
5974 | func_id); | |
17a52670 AS |
5975 | return -EINVAL; |
5976 | } | |
5977 | ||
00176a34 | 5978 | if (env->ops->get_func_proto) |
5e43f899 | 5979 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 5980 | if (!fn) { |
61bd5218 JK |
5981 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
5982 | func_id); | |
17a52670 AS |
5983 | return -EINVAL; |
5984 | } | |
5985 | ||
5986 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 5987 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 5988 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
5989 | return -EINVAL; |
5990 | } | |
5991 | ||
eae2e83e JO |
5992 | if (fn->allowed && !fn->allowed(env->prog)) { |
5993 | verbose(env, "helper call is not allowed in probe\n"); | |
5994 | return -EINVAL; | |
5995 | } | |
5996 | ||
04514d13 | 5997 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 5998 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
5999 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
6000 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
6001 | func_id_name(func_id), func_id); | |
6002 | return -EINVAL; | |
6003 | } | |
969bf05e | 6004 | |
33ff9823 | 6005 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 6006 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 6007 | |
1b986589 | 6008 | err = check_func_proto(fn, func_id); |
435faee1 | 6009 | if (err) { |
61bd5218 | 6010 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 6011 | func_id_name(func_id), func_id); |
435faee1 DB |
6012 | return err; |
6013 | } | |
6014 | ||
d83525ca | 6015 | meta.func_id = func_id; |
17a52670 | 6016 | /* check args */ |
523a4cf4 | 6017 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { |
af7ec138 | 6018 | err = check_func_arg(env, i, &meta, fn); |
a7658e1a AS |
6019 | if (err) |
6020 | return err; | |
6021 | } | |
17a52670 | 6022 | |
c93552c4 DB |
6023 | err = record_func_map(env, &meta, func_id, insn_idx); |
6024 | if (err) | |
6025 | return err; | |
6026 | ||
d2e4c1e6 DB |
6027 | err = record_func_key(env, &meta, func_id, insn_idx); |
6028 | if (err) | |
6029 | return err; | |
6030 | ||
435faee1 DB |
6031 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
6032 | * is inferred from register state. | |
6033 | */ | |
6034 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
6035 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
6036 | BPF_WRITE, -1, false); | |
435faee1 DB |
6037 | if (err) |
6038 | return err; | |
6039 | } | |
6040 | ||
fd978bf7 JS |
6041 | if (func_id == BPF_FUNC_tail_call) { |
6042 | err = check_reference_leak(env); | |
6043 | if (err) { | |
6044 | verbose(env, "tail_call would lead to reference leak\n"); | |
6045 | return err; | |
6046 | } | |
6047 | } else if (is_release_function(func_id)) { | |
1b986589 | 6048 | err = release_reference(env, meta.ref_obj_id); |
46f8bc92 MKL |
6049 | if (err) { |
6050 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
6051 | func_id_name(func_id), func_id); | |
fd978bf7 | 6052 | return err; |
46f8bc92 | 6053 | } |
fd978bf7 JS |
6054 | } |
6055 | ||
638f5b90 | 6056 | regs = cur_regs(env); |
cd339431 RG |
6057 | |
6058 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
6059 | * this is required because get_local_storage() can't return an error. | |
6060 | */ | |
6061 | if (func_id == BPF_FUNC_get_local_storage && | |
6062 | !register_is_null(®s[BPF_REG_2])) { | |
6063 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
6064 | return -EINVAL; | |
6065 | } | |
6066 | ||
69c087ba YS |
6067 | if (func_id == BPF_FUNC_for_each_map_elem) { |
6068 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, | |
6069 | set_map_elem_callback_state); | |
6070 | if (err < 0) | |
6071 | return -EINVAL; | |
6072 | } | |
6073 | ||
7b15523a FR |
6074 | if (func_id == BPF_FUNC_snprintf) { |
6075 | err = check_bpf_snprintf_call(env, regs); | |
6076 | if (err < 0) | |
6077 | return err; | |
6078 | } | |
6079 | ||
17a52670 | 6080 | /* reset caller saved regs */ |
dc503a8a | 6081 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 6082 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
6083 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
6084 | } | |
17a52670 | 6085 | |
5327ed3d JW |
6086 | /* helper call returns 64-bit value. */ |
6087 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
6088 | ||
dc503a8a | 6089 | /* update return register (already marked as written above) */ |
17a52670 | 6090 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 6091 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 6092 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
6093 | } else if (fn->ret_type == RET_VOID) { |
6094 | regs[BPF_REG_0].type = NOT_INIT; | |
3e6a4b3e RG |
6095 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL || |
6096 | fn->ret_type == RET_PTR_TO_MAP_VALUE) { | |
f1174f77 | 6097 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 6098 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
6099 | /* remember map_ptr, so that check_map_access() |
6100 | * can check 'value_size' boundary of memory access | |
6101 | * to map element returned from bpf_map_lookup_elem() | |
6102 | */ | |
33ff9823 | 6103 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
6104 | verbose(env, |
6105 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
6106 | return -EINVAL; |
6107 | } | |
33ff9823 | 6108 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
4d31f301 DB |
6109 | if (fn->ret_type == RET_PTR_TO_MAP_VALUE) { |
6110 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE; | |
e16d2f1a AS |
6111 | if (map_value_has_spin_lock(meta.map_ptr)) |
6112 | regs[BPF_REG_0].id = ++env->id_gen; | |
4d31f301 DB |
6113 | } else { |
6114 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
4d31f301 | 6115 | } |
c64b7983 JS |
6116 | } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) { |
6117 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
6118 | regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL; | |
85a51f8c LB |
6119 | } else if (fn->ret_type == RET_PTR_TO_SOCK_COMMON_OR_NULL) { |
6120 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
6121 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON_OR_NULL; | |
655a51e5 MKL |
6122 | } else if (fn->ret_type == RET_PTR_TO_TCP_SOCK_OR_NULL) { |
6123 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
6124 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK_OR_NULL; | |
457f4436 AN |
6125 | } else if (fn->ret_type == RET_PTR_TO_ALLOC_MEM_OR_NULL) { |
6126 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
6127 | regs[BPF_REG_0].type = PTR_TO_MEM_OR_NULL; | |
457f4436 | 6128 | regs[BPF_REG_0].mem_size = meta.mem_size; |
63d9b80d HL |
6129 | } else if (fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL || |
6130 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID) { | |
eaa6bcb7 HL |
6131 | const struct btf_type *t; |
6132 | ||
6133 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
22dc4a0f | 6134 | t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL); |
eaa6bcb7 HL |
6135 | if (!btf_type_is_struct(t)) { |
6136 | u32 tsize; | |
6137 | const struct btf_type *ret; | |
6138 | const char *tname; | |
6139 | ||
6140 | /* resolve the type size of ksym. */ | |
22dc4a0f | 6141 | ret = btf_resolve_size(meta.ret_btf, t, &tsize); |
eaa6bcb7 | 6142 | if (IS_ERR(ret)) { |
22dc4a0f | 6143 | tname = btf_name_by_offset(meta.ret_btf, t->name_off); |
eaa6bcb7 HL |
6144 | verbose(env, "unable to resolve the size of type '%s': %ld\n", |
6145 | tname, PTR_ERR(ret)); | |
6146 | return -EINVAL; | |
6147 | } | |
63d9b80d HL |
6148 | regs[BPF_REG_0].type = |
6149 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ? | |
6150 | PTR_TO_MEM : PTR_TO_MEM_OR_NULL; | |
eaa6bcb7 HL |
6151 | regs[BPF_REG_0].mem_size = tsize; |
6152 | } else { | |
63d9b80d HL |
6153 | regs[BPF_REG_0].type = |
6154 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ? | |
6155 | PTR_TO_BTF_ID : PTR_TO_BTF_ID_OR_NULL; | |
22dc4a0f | 6156 | regs[BPF_REG_0].btf = meta.ret_btf; |
eaa6bcb7 HL |
6157 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; |
6158 | } | |
3ca1032a KS |
6159 | } else if (fn->ret_type == RET_PTR_TO_BTF_ID_OR_NULL || |
6160 | fn->ret_type == RET_PTR_TO_BTF_ID) { | |
af7ec138 YS |
6161 | int ret_btf_id; |
6162 | ||
6163 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
3ca1032a KS |
6164 | regs[BPF_REG_0].type = fn->ret_type == RET_PTR_TO_BTF_ID ? |
6165 | PTR_TO_BTF_ID : | |
6166 | PTR_TO_BTF_ID_OR_NULL; | |
af7ec138 YS |
6167 | ret_btf_id = *fn->ret_btf_id; |
6168 | if (ret_btf_id == 0) { | |
6169 | verbose(env, "invalid return type %d of func %s#%d\n", | |
6170 | fn->ret_type, func_id_name(func_id), func_id); | |
6171 | return -EINVAL; | |
6172 | } | |
22dc4a0f AN |
6173 | /* current BPF helper definitions are only coming from |
6174 | * built-in code with type IDs from vmlinux BTF | |
6175 | */ | |
6176 | regs[BPF_REG_0].btf = btf_vmlinux; | |
af7ec138 | 6177 | regs[BPF_REG_0].btf_id = ret_btf_id; |
17a52670 | 6178 | } else { |
61bd5218 | 6179 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 6180 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
6181 | return -EINVAL; |
6182 | } | |
04fd61ab | 6183 | |
93c230e3 MKL |
6184 | if (reg_type_may_be_null(regs[BPF_REG_0].type)) |
6185 | regs[BPF_REG_0].id = ++env->id_gen; | |
6186 | ||
0f3adc28 | 6187 | if (is_ptr_cast_function(func_id)) { |
1b986589 MKL |
6188 | /* For release_reference() */ |
6189 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 6190 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
6191 | int id = acquire_reference_state(env, insn_idx); |
6192 | ||
6193 | if (id < 0) | |
6194 | return id; | |
6195 | /* For mark_ptr_or_null_reg() */ | |
6196 | regs[BPF_REG_0].id = id; | |
6197 | /* For release_reference() */ | |
6198 | regs[BPF_REG_0].ref_obj_id = id; | |
6199 | } | |
1b986589 | 6200 | |
849fa506 YS |
6201 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
6202 | ||
61bd5218 | 6203 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
6204 | if (err) |
6205 | return err; | |
04fd61ab | 6206 | |
fa28dcb8 SL |
6207 | if ((func_id == BPF_FUNC_get_stack || |
6208 | func_id == BPF_FUNC_get_task_stack) && | |
6209 | !env->prog->has_callchain_buf) { | |
c195651e YS |
6210 | const char *err_str; |
6211 | ||
6212 | #ifdef CONFIG_PERF_EVENTS | |
6213 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
6214 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
6215 | #else | |
6216 | err = -ENOTSUPP; | |
6217 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
6218 | #endif | |
6219 | if (err) { | |
6220 | verbose(env, err_str, func_id_name(func_id), func_id); | |
6221 | return err; | |
6222 | } | |
6223 | ||
6224 | env->prog->has_callchain_buf = true; | |
6225 | } | |
6226 | ||
5d99cb2c SL |
6227 | if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack) |
6228 | env->prog->call_get_stack = true; | |
6229 | ||
969bf05e AS |
6230 | if (changes_data) |
6231 | clear_all_pkt_pointers(env); | |
6232 | return 0; | |
6233 | } | |
6234 | ||
e6ac2450 MKL |
6235 | /* mark_btf_func_reg_size() is used when the reg size is determined by |
6236 | * the BTF func_proto's return value size and argument. | |
6237 | */ | |
6238 | static void mark_btf_func_reg_size(struct bpf_verifier_env *env, u32 regno, | |
6239 | size_t reg_size) | |
6240 | { | |
6241 | struct bpf_reg_state *reg = &cur_regs(env)[regno]; | |
6242 | ||
6243 | if (regno == BPF_REG_0) { | |
6244 | /* Function return value */ | |
6245 | reg->live |= REG_LIVE_WRITTEN; | |
6246 | reg->subreg_def = reg_size == sizeof(u64) ? | |
6247 | DEF_NOT_SUBREG : env->insn_idx + 1; | |
6248 | } else { | |
6249 | /* Function argument */ | |
6250 | if (reg_size == sizeof(u64)) { | |
6251 | mark_insn_zext(env, reg); | |
6252 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
6253 | } else { | |
6254 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ32); | |
6255 | } | |
6256 | } | |
6257 | } | |
6258 | ||
6259 | static int check_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
6260 | { | |
6261 | const struct btf_type *t, *func, *func_proto, *ptr_type; | |
6262 | struct bpf_reg_state *regs = cur_regs(env); | |
6263 | const char *func_name, *ptr_type_name; | |
6264 | u32 i, nargs, func_id, ptr_type_id; | |
6265 | const struct btf_param *args; | |
6266 | int err; | |
6267 | ||
6268 | func_id = insn->imm; | |
6269 | func = btf_type_by_id(btf_vmlinux, func_id); | |
6270 | func_name = btf_name_by_offset(btf_vmlinux, func->name_off); | |
6271 | func_proto = btf_type_by_id(btf_vmlinux, func->type); | |
6272 | ||
6273 | if (!env->ops->check_kfunc_call || | |
6274 | !env->ops->check_kfunc_call(func_id)) { | |
6275 | verbose(env, "calling kernel function %s is not allowed\n", | |
6276 | func_name); | |
6277 | return -EACCES; | |
6278 | } | |
6279 | ||
6280 | /* Check the arguments */ | |
6281 | err = btf_check_kfunc_arg_match(env, btf_vmlinux, func_id, regs); | |
6282 | if (err) | |
6283 | return err; | |
6284 | ||
6285 | for (i = 0; i < CALLER_SAVED_REGS; i++) | |
6286 | mark_reg_not_init(env, regs, caller_saved[i]); | |
6287 | ||
6288 | /* Check return type */ | |
6289 | t = btf_type_skip_modifiers(btf_vmlinux, func_proto->type, NULL); | |
6290 | if (btf_type_is_scalar(t)) { | |
6291 | mark_reg_unknown(env, regs, BPF_REG_0); | |
6292 | mark_btf_func_reg_size(env, BPF_REG_0, t->size); | |
6293 | } else if (btf_type_is_ptr(t)) { | |
6294 | ptr_type = btf_type_skip_modifiers(btf_vmlinux, t->type, | |
6295 | &ptr_type_id); | |
6296 | if (!btf_type_is_struct(ptr_type)) { | |
6297 | ptr_type_name = btf_name_by_offset(btf_vmlinux, | |
6298 | ptr_type->name_off); | |
6299 | verbose(env, "kernel function %s returns pointer type %s %s is not supported\n", | |
6300 | func_name, btf_type_str(ptr_type), | |
6301 | ptr_type_name); | |
6302 | return -EINVAL; | |
6303 | } | |
6304 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
6305 | regs[BPF_REG_0].btf = btf_vmlinux; | |
6306 | regs[BPF_REG_0].type = PTR_TO_BTF_ID; | |
6307 | regs[BPF_REG_0].btf_id = ptr_type_id; | |
6308 | mark_btf_func_reg_size(env, BPF_REG_0, sizeof(void *)); | |
6309 | } /* else { add_kfunc_call() ensures it is btf_type_is_void(t) } */ | |
6310 | ||
6311 | nargs = btf_type_vlen(func_proto); | |
6312 | args = (const struct btf_param *)(func_proto + 1); | |
6313 | for (i = 0; i < nargs; i++) { | |
6314 | u32 regno = i + 1; | |
6315 | ||
6316 | t = btf_type_skip_modifiers(btf_vmlinux, args[i].type, NULL); | |
6317 | if (btf_type_is_ptr(t)) | |
6318 | mark_btf_func_reg_size(env, regno, sizeof(void *)); | |
6319 | else | |
6320 | /* scalar. ensured by btf_check_kfunc_arg_match() */ | |
6321 | mark_btf_func_reg_size(env, regno, t->size); | |
6322 | } | |
6323 | ||
6324 | return 0; | |
6325 | } | |
6326 | ||
b03c9f9f EC |
6327 | static bool signed_add_overflows(s64 a, s64 b) |
6328 | { | |
6329 | /* Do the add in u64, where overflow is well-defined */ | |
6330 | s64 res = (s64)((u64)a + (u64)b); | |
6331 | ||
6332 | if (b < 0) | |
6333 | return res > a; | |
6334 | return res < a; | |
6335 | } | |
6336 | ||
bc895e8b | 6337 | static bool signed_add32_overflows(s32 a, s32 b) |
3f50f132 JF |
6338 | { |
6339 | /* Do the add in u32, where overflow is well-defined */ | |
6340 | s32 res = (s32)((u32)a + (u32)b); | |
6341 | ||
6342 | if (b < 0) | |
6343 | return res > a; | |
6344 | return res < a; | |
6345 | } | |
6346 | ||
bc895e8b | 6347 | static bool signed_sub_overflows(s64 a, s64 b) |
b03c9f9f EC |
6348 | { |
6349 | /* Do the sub in u64, where overflow is well-defined */ | |
6350 | s64 res = (s64)((u64)a - (u64)b); | |
6351 | ||
6352 | if (b < 0) | |
6353 | return res < a; | |
6354 | return res > a; | |
969bf05e AS |
6355 | } |
6356 | ||
3f50f132 JF |
6357 | static bool signed_sub32_overflows(s32 a, s32 b) |
6358 | { | |
bc895e8b | 6359 | /* Do the sub in u32, where overflow is well-defined */ |
3f50f132 JF |
6360 | s32 res = (s32)((u32)a - (u32)b); |
6361 | ||
6362 | if (b < 0) | |
6363 | return res < a; | |
6364 | return res > a; | |
6365 | } | |
6366 | ||
bb7f0f98 AS |
6367 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
6368 | const struct bpf_reg_state *reg, | |
6369 | enum bpf_reg_type type) | |
6370 | { | |
6371 | bool known = tnum_is_const(reg->var_off); | |
6372 | s64 val = reg->var_off.value; | |
6373 | s64 smin = reg->smin_value; | |
6374 | ||
6375 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
6376 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
6377 | reg_type_str[type], val); | |
6378 | return false; | |
6379 | } | |
6380 | ||
6381 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
6382 | verbose(env, "%s pointer offset %d is not allowed\n", | |
6383 | reg_type_str[type], reg->off); | |
6384 | return false; | |
6385 | } | |
6386 | ||
6387 | if (smin == S64_MIN) { | |
6388 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
6389 | reg_type_str[type]); | |
6390 | return false; | |
6391 | } | |
6392 | ||
6393 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
6394 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
6395 | smin, reg_type_str[type]); | |
6396 | return false; | |
6397 | } | |
6398 | ||
6399 | return true; | |
6400 | } | |
6401 | ||
979d63d5 DB |
6402 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
6403 | { | |
6404 | return &env->insn_aux_data[env->insn_idx]; | |
6405 | } | |
6406 | ||
a6aaece0 DB |
6407 | enum { |
6408 | REASON_BOUNDS = -1, | |
6409 | REASON_TYPE = -2, | |
6410 | REASON_PATHS = -3, | |
6411 | REASON_LIMIT = -4, | |
6412 | REASON_STACK = -5, | |
6413 | }; | |
6414 | ||
979d63d5 | 6415 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, |
bb01a1bb | 6416 | u32 *alu_limit, bool mask_to_left) |
979d63d5 | 6417 | { |
7fedb63a | 6418 | u32 max = 0, ptr_limit = 0; |
979d63d5 DB |
6419 | |
6420 | switch (ptr_reg->type) { | |
6421 | case PTR_TO_STACK: | |
1b1597e6 | 6422 | /* Offset 0 is out-of-bounds, but acceptable start for the |
7fedb63a DB |
6423 | * left direction, see BPF_REG_FP. Also, unknown scalar |
6424 | * offset where we would need to deal with min/max bounds is | |
6425 | * currently prohibited for unprivileged. | |
1b1597e6 PK |
6426 | */ |
6427 | max = MAX_BPF_STACK + mask_to_left; | |
7fedb63a | 6428 | ptr_limit = -(ptr_reg->var_off.value + ptr_reg->off); |
b658bbb8 | 6429 | break; |
979d63d5 | 6430 | case PTR_TO_MAP_VALUE: |
1b1597e6 | 6431 | max = ptr_reg->map_ptr->value_size; |
7fedb63a DB |
6432 | ptr_limit = (mask_to_left ? |
6433 | ptr_reg->smin_value : | |
6434 | ptr_reg->umax_value) + ptr_reg->off; | |
b658bbb8 | 6435 | break; |
979d63d5 | 6436 | default: |
a6aaece0 | 6437 | return REASON_TYPE; |
979d63d5 | 6438 | } |
b658bbb8 DB |
6439 | |
6440 | if (ptr_limit >= max) | |
a6aaece0 | 6441 | return REASON_LIMIT; |
b658bbb8 DB |
6442 | *alu_limit = ptr_limit; |
6443 | return 0; | |
979d63d5 DB |
6444 | } |
6445 | ||
d3bd7413 DB |
6446 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
6447 | const struct bpf_insn *insn) | |
6448 | { | |
2c78ee89 | 6449 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
6450 | } |
6451 | ||
6452 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
6453 | u32 alu_state, u32 alu_limit) | |
6454 | { | |
6455 | /* If we arrived here from different branches with different | |
6456 | * state or limits to sanitize, then this won't work. | |
6457 | */ | |
6458 | if (aux->alu_state && | |
6459 | (aux->alu_state != alu_state || | |
6460 | aux->alu_limit != alu_limit)) | |
a6aaece0 | 6461 | return REASON_PATHS; |
d3bd7413 | 6462 | |
e6ac5933 | 6463 | /* Corresponding fixup done in do_misc_fixups(). */ |
d3bd7413 DB |
6464 | aux->alu_state = alu_state; |
6465 | aux->alu_limit = alu_limit; | |
6466 | return 0; | |
6467 | } | |
6468 | ||
6469 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
6470 | struct bpf_insn *insn) | |
6471 | { | |
6472 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
6473 | ||
6474 | if (can_skip_alu_sanitation(env, insn)) | |
6475 | return 0; | |
6476 | ||
6477 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
6478 | } | |
6479 | ||
f5288193 DB |
6480 | static bool sanitize_needed(u8 opcode) |
6481 | { | |
6482 | return opcode == BPF_ADD || opcode == BPF_SUB; | |
6483 | } | |
6484 | ||
3d0220f6 DB |
6485 | struct bpf_sanitize_info { |
6486 | struct bpf_insn_aux_data aux; | |
bb01a1bb | 6487 | bool mask_to_left; |
3d0220f6 DB |
6488 | }; |
6489 | ||
9183671a DB |
6490 | static struct bpf_verifier_state * |
6491 | sanitize_speculative_path(struct bpf_verifier_env *env, | |
6492 | const struct bpf_insn *insn, | |
6493 | u32 next_idx, u32 curr_idx) | |
6494 | { | |
6495 | struct bpf_verifier_state *branch; | |
6496 | struct bpf_reg_state *regs; | |
6497 | ||
6498 | branch = push_stack(env, next_idx, curr_idx, true); | |
6499 | if (branch && insn) { | |
6500 | regs = branch->frame[branch->curframe]->regs; | |
6501 | if (BPF_SRC(insn->code) == BPF_K) { | |
6502 | mark_reg_unknown(env, regs, insn->dst_reg); | |
6503 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
6504 | mark_reg_unknown(env, regs, insn->dst_reg); | |
6505 | mark_reg_unknown(env, regs, insn->src_reg); | |
6506 | } | |
6507 | } | |
6508 | return branch; | |
6509 | } | |
6510 | ||
979d63d5 DB |
6511 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
6512 | struct bpf_insn *insn, | |
6513 | const struct bpf_reg_state *ptr_reg, | |
6f55b2f2 | 6514 | const struct bpf_reg_state *off_reg, |
979d63d5 | 6515 | struct bpf_reg_state *dst_reg, |
3d0220f6 | 6516 | struct bpf_sanitize_info *info, |
7fedb63a | 6517 | const bool commit_window) |
979d63d5 | 6518 | { |
3d0220f6 | 6519 | struct bpf_insn_aux_data *aux = commit_window ? cur_aux(env) : &info->aux; |
979d63d5 | 6520 | struct bpf_verifier_state *vstate = env->cur_state; |
801c6058 | 6521 | bool off_is_imm = tnum_is_const(off_reg->var_off); |
6f55b2f2 | 6522 | bool off_is_neg = off_reg->smin_value < 0; |
979d63d5 DB |
6523 | bool ptr_is_dst_reg = ptr_reg == dst_reg; |
6524 | u8 opcode = BPF_OP(insn->code); | |
6525 | u32 alu_state, alu_limit; | |
6526 | struct bpf_reg_state tmp; | |
6527 | bool ret; | |
f232326f | 6528 | int err; |
979d63d5 | 6529 | |
d3bd7413 | 6530 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
6531 | return 0; |
6532 | ||
6533 | /* We already marked aux for masking from non-speculative | |
6534 | * paths, thus we got here in the first place. We only care | |
6535 | * to explore bad access from here. | |
6536 | */ | |
6537 | if (vstate->speculative) | |
6538 | goto do_sim; | |
6539 | ||
bb01a1bb DB |
6540 | if (!commit_window) { |
6541 | if (!tnum_is_const(off_reg->var_off) && | |
6542 | (off_reg->smin_value < 0) != (off_reg->smax_value < 0)) | |
6543 | return REASON_BOUNDS; | |
6544 | ||
6545 | info->mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
6546 | (opcode == BPF_SUB && !off_is_neg); | |
6547 | } | |
6548 | ||
6549 | err = retrieve_ptr_limit(ptr_reg, &alu_limit, info->mask_to_left); | |
f232326f PK |
6550 | if (err < 0) |
6551 | return err; | |
6552 | ||
7fedb63a DB |
6553 | if (commit_window) { |
6554 | /* In commit phase we narrow the masking window based on | |
6555 | * the observed pointer move after the simulated operation. | |
6556 | */ | |
3d0220f6 DB |
6557 | alu_state = info->aux.alu_state; |
6558 | alu_limit = abs(info->aux.alu_limit - alu_limit); | |
7fedb63a DB |
6559 | } else { |
6560 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
801c6058 | 6561 | alu_state |= off_is_imm ? BPF_ALU_IMMEDIATE : 0; |
7fedb63a DB |
6562 | alu_state |= ptr_is_dst_reg ? |
6563 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
e042aa53 DB |
6564 | |
6565 | /* Limit pruning on unknown scalars to enable deep search for | |
6566 | * potential masking differences from other program paths. | |
6567 | */ | |
6568 | if (!off_is_imm) | |
6569 | env->explore_alu_limits = true; | |
7fedb63a DB |
6570 | } |
6571 | ||
f232326f PK |
6572 | err = update_alu_sanitation_state(aux, alu_state, alu_limit); |
6573 | if (err < 0) | |
6574 | return err; | |
979d63d5 | 6575 | do_sim: |
7fedb63a DB |
6576 | /* If we're in commit phase, we're done here given we already |
6577 | * pushed the truncated dst_reg into the speculative verification | |
6578 | * stack. | |
a7036191 DB |
6579 | * |
6580 | * Also, when register is a known constant, we rewrite register-based | |
6581 | * operation to immediate-based, and thus do not need masking (and as | |
6582 | * a consequence, do not need to simulate the zero-truncation either). | |
7fedb63a | 6583 | */ |
a7036191 | 6584 | if (commit_window || off_is_imm) |
7fedb63a DB |
6585 | return 0; |
6586 | ||
979d63d5 DB |
6587 | /* Simulate and find potential out-of-bounds access under |
6588 | * speculative execution from truncation as a result of | |
6589 | * masking when off was not within expected range. If off | |
6590 | * sits in dst, then we temporarily need to move ptr there | |
6591 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
6592 | * for cases where we use K-based arithmetic in one direction | |
6593 | * and truncated reg-based in the other in order to explore | |
6594 | * bad access. | |
6595 | */ | |
6596 | if (!ptr_is_dst_reg) { | |
6597 | tmp = *dst_reg; | |
6598 | *dst_reg = *ptr_reg; | |
6599 | } | |
9183671a DB |
6600 | ret = sanitize_speculative_path(env, NULL, env->insn_idx + 1, |
6601 | env->insn_idx); | |
0803278b | 6602 | if (!ptr_is_dst_reg && ret) |
979d63d5 | 6603 | *dst_reg = tmp; |
a6aaece0 DB |
6604 | return !ret ? REASON_STACK : 0; |
6605 | } | |
6606 | ||
fe9a5ca7 DB |
6607 | static void sanitize_mark_insn_seen(struct bpf_verifier_env *env) |
6608 | { | |
6609 | struct bpf_verifier_state *vstate = env->cur_state; | |
6610 | ||
6611 | /* If we simulate paths under speculation, we don't update the | |
6612 | * insn as 'seen' such that when we verify unreachable paths in | |
6613 | * the non-speculative domain, sanitize_dead_code() can still | |
6614 | * rewrite/sanitize them. | |
6615 | */ | |
6616 | if (!vstate->speculative) | |
6617 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; | |
6618 | } | |
6619 | ||
a6aaece0 DB |
6620 | static int sanitize_err(struct bpf_verifier_env *env, |
6621 | const struct bpf_insn *insn, int reason, | |
6622 | const struct bpf_reg_state *off_reg, | |
6623 | const struct bpf_reg_state *dst_reg) | |
6624 | { | |
6625 | static const char *err = "pointer arithmetic with it prohibited for !root"; | |
6626 | const char *op = BPF_OP(insn->code) == BPF_ADD ? "add" : "sub"; | |
6627 | u32 dst = insn->dst_reg, src = insn->src_reg; | |
6628 | ||
6629 | switch (reason) { | |
6630 | case REASON_BOUNDS: | |
6631 | verbose(env, "R%d has unknown scalar with mixed signed bounds, %s\n", | |
6632 | off_reg == dst_reg ? dst : src, err); | |
6633 | break; | |
6634 | case REASON_TYPE: | |
6635 | verbose(env, "R%d has pointer with unsupported alu operation, %s\n", | |
6636 | off_reg == dst_reg ? src : dst, err); | |
6637 | break; | |
6638 | case REASON_PATHS: | |
6639 | verbose(env, "R%d tried to %s from different maps, paths or scalars, %s\n", | |
6640 | dst, op, err); | |
6641 | break; | |
6642 | case REASON_LIMIT: | |
6643 | verbose(env, "R%d tried to %s beyond pointer bounds, %s\n", | |
6644 | dst, op, err); | |
6645 | break; | |
6646 | case REASON_STACK: | |
6647 | verbose(env, "R%d could not be pushed for speculative verification, %s\n", | |
6648 | dst, err); | |
6649 | break; | |
6650 | default: | |
6651 | verbose(env, "verifier internal error: unknown reason (%d)\n", | |
6652 | reason); | |
6653 | break; | |
6654 | } | |
6655 | ||
6656 | return -EACCES; | |
979d63d5 DB |
6657 | } |
6658 | ||
01f810ac AM |
6659 | /* check that stack access falls within stack limits and that 'reg' doesn't |
6660 | * have a variable offset. | |
6661 | * | |
6662 | * Variable offset is prohibited for unprivileged mode for simplicity since it | |
6663 | * requires corresponding support in Spectre masking for stack ALU. See also | |
6664 | * retrieve_ptr_limit(). | |
6665 | * | |
6666 | * | |
6667 | * 'off' includes 'reg->off'. | |
6668 | */ | |
6669 | static int check_stack_access_for_ptr_arithmetic( | |
6670 | struct bpf_verifier_env *env, | |
6671 | int regno, | |
6672 | const struct bpf_reg_state *reg, | |
6673 | int off) | |
6674 | { | |
6675 | if (!tnum_is_const(reg->var_off)) { | |
6676 | char tn_buf[48]; | |
6677 | ||
6678 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6679 | verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n", | |
6680 | regno, tn_buf, off); | |
6681 | return -EACCES; | |
6682 | } | |
6683 | ||
6684 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
6685 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
6686 | "prohibited for !root; off=%d\n", regno, off); | |
6687 | return -EACCES; | |
6688 | } | |
6689 | ||
6690 | return 0; | |
6691 | } | |
6692 | ||
073815b7 DB |
6693 | static int sanitize_check_bounds(struct bpf_verifier_env *env, |
6694 | const struct bpf_insn *insn, | |
6695 | const struct bpf_reg_state *dst_reg) | |
6696 | { | |
6697 | u32 dst = insn->dst_reg; | |
6698 | ||
6699 | /* For unprivileged we require that resulting offset must be in bounds | |
6700 | * in order to be able to sanitize access later on. | |
6701 | */ | |
6702 | if (env->bypass_spec_v1) | |
6703 | return 0; | |
6704 | ||
6705 | switch (dst_reg->type) { | |
6706 | case PTR_TO_STACK: | |
6707 | if (check_stack_access_for_ptr_arithmetic(env, dst, dst_reg, | |
6708 | dst_reg->off + dst_reg->var_off.value)) | |
6709 | return -EACCES; | |
6710 | break; | |
6711 | case PTR_TO_MAP_VALUE: | |
6712 | if (check_map_access(env, dst, dst_reg->off, 1, false)) { | |
6713 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
6714 | "prohibited for !root\n", dst); | |
6715 | return -EACCES; | |
6716 | } | |
6717 | break; | |
6718 | default: | |
6719 | break; | |
6720 | } | |
6721 | ||
6722 | return 0; | |
6723 | } | |
01f810ac | 6724 | |
f1174f77 | 6725 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
6726 | * Caller should also handle BPF_MOV case separately. |
6727 | * If we return -EACCES, caller may want to try again treating pointer as a | |
6728 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
6729 | */ | |
6730 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
6731 | struct bpf_insn *insn, | |
6732 | const struct bpf_reg_state *ptr_reg, | |
6733 | const struct bpf_reg_state *off_reg) | |
969bf05e | 6734 | { |
f4d7e40a AS |
6735 | struct bpf_verifier_state *vstate = env->cur_state; |
6736 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
6737 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 6738 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
6739 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
6740 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
6741 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
6742 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
3d0220f6 | 6743 | struct bpf_sanitize_info info = {}; |
969bf05e | 6744 | u8 opcode = BPF_OP(insn->code); |
24c109bb | 6745 | u32 dst = insn->dst_reg; |
979d63d5 | 6746 | int ret; |
969bf05e | 6747 | |
f1174f77 | 6748 | dst_reg = ®s[dst]; |
969bf05e | 6749 | |
6f16101e DB |
6750 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
6751 | smin_val > smax_val || umin_val > umax_val) { | |
6752 | /* Taint dst register if offset had invalid bounds derived from | |
6753 | * e.g. dead branches. | |
6754 | */ | |
f54c7898 | 6755 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 6756 | return 0; |
f1174f77 EC |
6757 | } |
6758 | ||
6759 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
6760 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
6c693541 YS |
6761 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
6762 | __mark_reg_unknown(env, dst_reg); | |
6763 | return 0; | |
6764 | } | |
6765 | ||
82abbf8d AS |
6766 | verbose(env, |
6767 | "R%d 32-bit pointer arithmetic prohibited\n", | |
6768 | dst); | |
f1174f77 | 6769 | return -EACCES; |
969bf05e AS |
6770 | } |
6771 | ||
aad2eeaf JS |
6772 | switch (ptr_reg->type) { |
6773 | case PTR_TO_MAP_VALUE_OR_NULL: | |
6774 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", | |
6775 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 6776 | return -EACCES; |
aad2eeaf | 6777 | case CONST_PTR_TO_MAP: |
7c696732 YS |
6778 | /* smin_val represents the known value */ |
6779 | if (known && smin_val == 0 && opcode == BPF_ADD) | |
6780 | break; | |
8731745e | 6781 | fallthrough; |
aad2eeaf | 6782 | case PTR_TO_PACKET_END: |
c64b7983 JS |
6783 | case PTR_TO_SOCKET: |
6784 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
6785 | case PTR_TO_SOCK_COMMON: |
6786 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
6787 | case PTR_TO_TCP_SOCK: |
6788 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 6789 | case PTR_TO_XDP_SOCK: |
aad2eeaf JS |
6790 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
6791 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 6792 | return -EACCES; |
aad2eeaf JS |
6793 | default: |
6794 | break; | |
f1174f77 EC |
6795 | } |
6796 | ||
6797 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
6798 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 6799 | */ |
f1174f77 EC |
6800 | dst_reg->type = ptr_reg->type; |
6801 | dst_reg->id = ptr_reg->id; | |
969bf05e | 6802 | |
bb7f0f98 AS |
6803 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
6804 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
6805 | return -EINVAL; | |
6806 | ||
3f50f132 JF |
6807 | /* pointer types do not carry 32-bit bounds at the moment. */ |
6808 | __mark_reg32_unbounded(dst_reg); | |
6809 | ||
7fedb63a DB |
6810 | if (sanitize_needed(opcode)) { |
6811 | ret = sanitize_ptr_alu(env, insn, ptr_reg, off_reg, dst_reg, | |
3d0220f6 | 6812 | &info, false); |
a6aaece0 DB |
6813 | if (ret < 0) |
6814 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
7fedb63a | 6815 | } |
a6aaece0 | 6816 | |
f1174f77 EC |
6817 | switch (opcode) { |
6818 | case BPF_ADD: | |
6819 | /* We can take a fixed offset as long as it doesn't overflow | |
6820 | * the s32 'off' field | |
969bf05e | 6821 | */ |
b03c9f9f EC |
6822 | if (known && (ptr_reg->off + smin_val == |
6823 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 6824 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
6825 | dst_reg->smin_value = smin_ptr; |
6826 | dst_reg->smax_value = smax_ptr; | |
6827 | dst_reg->umin_value = umin_ptr; | |
6828 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 6829 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 6830 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 6831 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
6832 | break; |
6833 | } | |
f1174f77 EC |
6834 | /* A new variable offset is created. Note that off_reg->off |
6835 | * == 0, since it's a scalar. | |
6836 | * dst_reg gets the pointer type and since some positive | |
6837 | * integer value was added to the pointer, give it a new 'id' | |
6838 | * if it's a PTR_TO_PACKET. | |
6839 | * this creates a new 'base' pointer, off_reg (variable) gets | |
6840 | * added into the variable offset, and we copy the fixed offset | |
6841 | * from ptr_reg. | |
969bf05e | 6842 | */ |
b03c9f9f EC |
6843 | if (signed_add_overflows(smin_ptr, smin_val) || |
6844 | signed_add_overflows(smax_ptr, smax_val)) { | |
6845 | dst_reg->smin_value = S64_MIN; | |
6846 | dst_reg->smax_value = S64_MAX; | |
6847 | } else { | |
6848 | dst_reg->smin_value = smin_ptr + smin_val; | |
6849 | dst_reg->smax_value = smax_ptr + smax_val; | |
6850 | } | |
6851 | if (umin_ptr + umin_val < umin_ptr || | |
6852 | umax_ptr + umax_val < umax_ptr) { | |
6853 | dst_reg->umin_value = 0; | |
6854 | dst_reg->umax_value = U64_MAX; | |
6855 | } else { | |
6856 | dst_reg->umin_value = umin_ptr + umin_val; | |
6857 | dst_reg->umax_value = umax_ptr + umax_val; | |
6858 | } | |
f1174f77 EC |
6859 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
6860 | dst_reg->off = ptr_reg->off; | |
0962590e | 6861 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 6862 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
6863 | dst_reg->id = ++env->id_gen; |
6864 | /* something was added to pkt_ptr, set range to zero */ | |
22dc4a0f | 6865 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
f1174f77 EC |
6866 | } |
6867 | break; | |
6868 | case BPF_SUB: | |
6869 | if (dst_reg == off_reg) { | |
6870 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
6871 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
6872 | dst); | |
f1174f77 EC |
6873 | return -EACCES; |
6874 | } | |
6875 | /* We don't allow subtraction from FP, because (according to | |
6876 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
6877 | * be able to deal with it. | |
969bf05e | 6878 | */ |
f1174f77 | 6879 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
6880 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
6881 | dst); | |
f1174f77 EC |
6882 | return -EACCES; |
6883 | } | |
b03c9f9f EC |
6884 | if (known && (ptr_reg->off - smin_val == |
6885 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 6886 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
6887 | dst_reg->smin_value = smin_ptr; |
6888 | dst_reg->smax_value = smax_ptr; | |
6889 | dst_reg->umin_value = umin_ptr; | |
6890 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
6891 | dst_reg->var_off = ptr_reg->var_off; |
6892 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 6893 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 6894 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
6895 | break; |
6896 | } | |
f1174f77 EC |
6897 | /* A new variable offset is created. If the subtrahend is known |
6898 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 6899 | */ |
b03c9f9f EC |
6900 | if (signed_sub_overflows(smin_ptr, smax_val) || |
6901 | signed_sub_overflows(smax_ptr, smin_val)) { | |
6902 | /* Overflow possible, we know nothing */ | |
6903 | dst_reg->smin_value = S64_MIN; | |
6904 | dst_reg->smax_value = S64_MAX; | |
6905 | } else { | |
6906 | dst_reg->smin_value = smin_ptr - smax_val; | |
6907 | dst_reg->smax_value = smax_ptr - smin_val; | |
6908 | } | |
6909 | if (umin_ptr < umax_val) { | |
6910 | /* Overflow possible, we know nothing */ | |
6911 | dst_reg->umin_value = 0; | |
6912 | dst_reg->umax_value = U64_MAX; | |
6913 | } else { | |
6914 | /* Cannot overflow (as long as bounds are consistent) */ | |
6915 | dst_reg->umin_value = umin_ptr - umax_val; | |
6916 | dst_reg->umax_value = umax_ptr - umin_val; | |
6917 | } | |
f1174f77 EC |
6918 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
6919 | dst_reg->off = ptr_reg->off; | |
0962590e | 6920 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 6921 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
6922 | dst_reg->id = ++env->id_gen; |
6923 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 6924 | if (smin_val < 0) |
22dc4a0f | 6925 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
43188702 | 6926 | } |
f1174f77 EC |
6927 | break; |
6928 | case BPF_AND: | |
6929 | case BPF_OR: | |
6930 | case BPF_XOR: | |
82abbf8d AS |
6931 | /* bitwise ops on pointers are troublesome, prohibit. */ |
6932 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
6933 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
6934 | return -EACCES; |
6935 | default: | |
6936 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
6937 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
6938 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 6939 | return -EACCES; |
43188702 JF |
6940 | } |
6941 | ||
bb7f0f98 AS |
6942 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
6943 | return -EINVAL; | |
6944 | ||
b03c9f9f EC |
6945 | __update_reg_bounds(dst_reg); |
6946 | __reg_deduce_bounds(dst_reg); | |
6947 | __reg_bound_offset(dst_reg); | |
0d6303db | 6948 | |
073815b7 DB |
6949 | if (sanitize_check_bounds(env, insn, dst_reg) < 0) |
6950 | return -EACCES; | |
7fedb63a DB |
6951 | if (sanitize_needed(opcode)) { |
6952 | ret = sanitize_ptr_alu(env, insn, dst_reg, off_reg, dst_reg, | |
3d0220f6 | 6953 | &info, true); |
7fedb63a DB |
6954 | if (ret < 0) |
6955 | return sanitize_err(env, insn, ret, off_reg, dst_reg); | |
0d6303db DB |
6956 | } |
6957 | ||
43188702 JF |
6958 | return 0; |
6959 | } | |
6960 | ||
3f50f132 JF |
6961 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
6962 | struct bpf_reg_state *src_reg) | |
6963 | { | |
6964 | s32 smin_val = src_reg->s32_min_value; | |
6965 | s32 smax_val = src_reg->s32_max_value; | |
6966 | u32 umin_val = src_reg->u32_min_value; | |
6967 | u32 umax_val = src_reg->u32_max_value; | |
6968 | ||
6969 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
6970 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
6971 | dst_reg->s32_min_value = S32_MIN; | |
6972 | dst_reg->s32_max_value = S32_MAX; | |
6973 | } else { | |
6974 | dst_reg->s32_min_value += smin_val; | |
6975 | dst_reg->s32_max_value += smax_val; | |
6976 | } | |
6977 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
6978 | dst_reg->u32_max_value + umax_val < umax_val) { | |
6979 | dst_reg->u32_min_value = 0; | |
6980 | dst_reg->u32_max_value = U32_MAX; | |
6981 | } else { | |
6982 | dst_reg->u32_min_value += umin_val; | |
6983 | dst_reg->u32_max_value += umax_val; | |
6984 | } | |
6985 | } | |
6986 | ||
07cd2631 JF |
6987 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
6988 | struct bpf_reg_state *src_reg) | |
6989 | { | |
6990 | s64 smin_val = src_reg->smin_value; | |
6991 | s64 smax_val = src_reg->smax_value; | |
6992 | u64 umin_val = src_reg->umin_value; | |
6993 | u64 umax_val = src_reg->umax_value; | |
6994 | ||
6995 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
6996 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
6997 | dst_reg->smin_value = S64_MIN; | |
6998 | dst_reg->smax_value = S64_MAX; | |
6999 | } else { | |
7000 | dst_reg->smin_value += smin_val; | |
7001 | dst_reg->smax_value += smax_val; | |
7002 | } | |
7003 | if (dst_reg->umin_value + umin_val < umin_val || | |
7004 | dst_reg->umax_value + umax_val < umax_val) { | |
7005 | dst_reg->umin_value = 0; | |
7006 | dst_reg->umax_value = U64_MAX; | |
7007 | } else { | |
7008 | dst_reg->umin_value += umin_val; | |
7009 | dst_reg->umax_value += umax_val; | |
7010 | } | |
3f50f132 JF |
7011 | } |
7012 | ||
7013 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
7014 | struct bpf_reg_state *src_reg) | |
7015 | { | |
7016 | s32 smin_val = src_reg->s32_min_value; | |
7017 | s32 smax_val = src_reg->s32_max_value; | |
7018 | u32 umin_val = src_reg->u32_min_value; | |
7019 | u32 umax_val = src_reg->u32_max_value; | |
7020 | ||
7021 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
7022 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
7023 | /* Overflow possible, we know nothing */ | |
7024 | dst_reg->s32_min_value = S32_MIN; | |
7025 | dst_reg->s32_max_value = S32_MAX; | |
7026 | } else { | |
7027 | dst_reg->s32_min_value -= smax_val; | |
7028 | dst_reg->s32_max_value -= smin_val; | |
7029 | } | |
7030 | if (dst_reg->u32_min_value < umax_val) { | |
7031 | /* Overflow possible, we know nothing */ | |
7032 | dst_reg->u32_min_value = 0; | |
7033 | dst_reg->u32_max_value = U32_MAX; | |
7034 | } else { | |
7035 | /* Cannot overflow (as long as bounds are consistent) */ | |
7036 | dst_reg->u32_min_value -= umax_val; | |
7037 | dst_reg->u32_max_value -= umin_val; | |
7038 | } | |
07cd2631 JF |
7039 | } |
7040 | ||
7041 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
7042 | struct bpf_reg_state *src_reg) | |
7043 | { | |
7044 | s64 smin_val = src_reg->smin_value; | |
7045 | s64 smax_val = src_reg->smax_value; | |
7046 | u64 umin_val = src_reg->umin_value; | |
7047 | u64 umax_val = src_reg->umax_value; | |
7048 | ||
7049 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
7050 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
7051 | /* Overflow possible, we know nothing */ | |
7052 | dst_reg->smin_value = S64_MIN; | |
7053 | dst_reg->smax_value = S64_MAX; | |
7054 | } else { | |
7055 | dst_reg->smin_value -= smax_val; | |
7056 | dst_reg->smax_value -= smin_val; | |
7057 | } | |
7058 | if (dst_reg->umin_value < umax_val) { | |
7059 | /* Overflow possible, we know nothing */ | |
7060 | dst_reg->umin_value = 0; | |
7061 | dst_reg->umax_value = U64_MAX; | |
7062 | } else { | |
7063 | /* Cannot overflow (as long as bounds are consistent) */ | |
7064 | dst_reg->umin_value -= umax_val; | |
7065 | dst_reg->umax_value -= umin_val; | |
7066 | } | |
3f50f132 JF |
7067 | } |
7068 | ||
7069 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
7070 | struct bpf_reg_state *src_reg) | |
7071 | { | |
7072 | s32 smin_val = src_reg->s32_min_value; | |
7073 | u32 umin_val = src_reg->u32_min_value; | |
7074 | u32 umax_val = src_reg->u32_max_value; | |
7075 | ||
7076 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
7077 | /* Ain't nobody got time to multiply that sign */ | |
7078 | __mark_reg32_unbounded(dst_reg); | |
7079 | return; | |
7080 | } | |
7081 | /* Both values are positive, so we can work with unsigned and | |
7082 | * copy the result to signed (unless it exceeds S32_MAX). | |
7083 | */ | |
7084 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
7085 | /* Potential overflow, we know nothing */ | |
7086 | __mark_reg32_unbounded(dst_reg); | |
7087 | return; | |
7088 | } | |
7089 | dst_reg->u32_min_value *= umin_val; | |
7090 | dst_reg->u32_max_value *= umax_val; | |
7091 | if (dst_reg->u32_max_value > S32_MAX) { | |
7092 | /* Overflow possible, we know nothing */ | |
7093 | dst_reg->s32_min_value = S32_MIN; | |
7094 | dst_reg->s32_max_value = S32_MAX; | |
7095 | } else { | |
7096 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
7097 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
7098 | } | |
07cd2631 JF |
7099 | } |
7100 | ||
7101 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
7102 | struct bpf_reg_state *src_reg) | |
7103 | { | |
7104 | s64 smin_val = src_reg->smin_value; | |
7105 | u64 umin_val = src_reg->umin_value; | |
7106 | u64 umax_val = src_reg->umax_value; | |
7107 | ||
07cd2631 JF |
7108 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
7109 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 7110 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
7111 | return; |
7112 | } | |
7113 | /* Both values are positive, so we can work with unsigned and | |
7114 | * copy the result to signed (unless it exceeds S64_MAX). | |
7115 | */ | |
7116 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
7117 | /* Potential overflow, we know nothing */ | |
3f50f132 | 7118 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
7119 | return; |
7120 | } | |
7121 | dst_reg->umin_value *= umin_val; | |
7122 | dst_reg->umax_value *= umax_val; | |
7123 | if (dst_reg->umax_value > S64_MAX) { | |
7124 | /* Overflow possible, we know nothing */ | |
7125 | dst_reg->smin_value = S64_MIN; | |
7126 | dst_reg->smax_value = S64_MAX; | |
7127 | } else { | |
7128 | dst_reg->smin_value = dst_reg->umin_value; | |
7129 | dst_reg->smax_value = dst_reg->umax_value; | |
7130 | } | |
7131 | } | |
7132 | ||
3f50f132 JF |
7133 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
7134 | struct bpf_reg_state *src_reg) | |
7135 | { | |
7136 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
7137 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
7138 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
7139 | s32 smin_val = src_reg->s32_min_value; | |
7140 | u32 umax_val = src_reg->u32_max_value; | |
7141 | ||
049c4e13 DB |
7142 | if (src_known && dst_known) { |
7143 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 7144 | return; |
049c4e13 | 7145 | } |
3f50f132 JF |
7146 | |
7147 | /* We get our minimum from the var_off, since that's inherently | |
7148 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
7149 | */ | |
7150 | dst_reg->u32_min_value = var32_off.value; | |
7151 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
7152 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
7153 | /* Lose signed bounds when ANDing negative numbers, | |
7154 | * ain't nobody got time for that. | |
7155 | */ | |
7156 | dst_reg->s32_min_value = S32_MIN; | |
7157 | dst_reg->s32_max_value = S32_MAX; | |
7158 | } else { | |
7159 | /* ANDing two positives gives a positive, so safe to | |
7160 | * cast result into s64. | |
7161 | */ | |
7162 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
7163 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
7164 | } | |
3f50f132 JF |
7165 | } |
7166 | ||
07cd2631 JF |
7167 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
7168 | struct bpf_reg_state *src_reg) | |
7169 | { | |
3f50f132 JF |
7170 | bool src_known = tnum_is_const(src_reg->var_off); |
7171 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
7172 | s64 smin_val = src_reg->smin_value; |
7173 | u64 umax_val = src_reg->umax_value; | |
7174 | ||
3f50f132 | 7175 | if (src_known && dst_known) { |
4fbb38a3 | 7176 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
7177 | return; |
7178 | } | |
7179 | ||
07cd2631 JF |
7180 | /* We get our minimum from the var_off, since that's inherently |
7181 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
7182 | */ | |
07cd2631 JF |
7183 | dst_reg->umin_value = dst_reg->var_off.value; |
7184 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
7185 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
7186 | /* Lose signed bounds when ANDing negative numbers, | |
7187 | * ain't nobody got time for that. | |
7188 | */ | |
7189 | dst_reg->smin_value = S64_MIN; | |
7190 | dst_reg->smax_value = S64_MAX; | |
7191 | } else { | |
7192 | /* ANDing two positives gives a positive, so safe to | |
7193 | * cast result into s64. | |
7194 | */ | |
7195 | dst_reg->smin_value = dst_reg->umin_value; | |
7196 | dst_reg->smax_value = dst_reg->umax_value; | |
7197 | } | |
7198 | /* We may learn something more from the var_off */ | |
7199 | __update_reg_bounds(dst_reg); | |
7200 | } | |
7201 | ||
3f50f132 JF |
7202 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
7203 | struct bpf_reg_state *src_reg) | |
7204 | { | |
7205 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
7206 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
7207 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5b9fbeb7 DB |
7208 | s32 smin_val = src_reg->s32_min_value; |
7209 | u32 umin_val = src_reg->u32_min_value; | |
3f50f132 | 7210 | |
049c4e13 DB |
7211 | if (src_known && dst_known) { |
7212 | __mark_reg32_known(dst_reg, var32_off.value); | |
3f50f132 | 7213 | return; |
049c4e13 | 7214 | } |
3f50f132 JF |
7215 | |
7216 | /* We get our maximum from the var_off, and our minimum is the | |
7217 | * maximum of the operands' minima | |
7218 | */ | |
7219 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
7220 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
7221 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
7222 | /* Lose signed bounds when ORing negative numbers, | |
7223 | * ain't nobody got time for that. | |
7224 | */ | |
7225 | dst_reg->s32_min_value = S32_MIN; | |
7226 | dst_reg->s32_max_value = S32_MAX; | |
7227 | } else { | |
7228 | /* ORing two positives gives a positive, so safe to | |
7229 | * cast result into s64. | |
7230 | */ | |
5b9fbeb7 DB |
7231 | dst_reg->s32_min_value = dst_reg->u32_min_value; |
7232 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
3f50f132 JF |
7233 | } |
7234 | } | |
7235 | ||
07cd2631 JF |
7236 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
7237 | struct bpf_reg_state *src_reg) | |
7238 | { | |
3f50f132 JF |
7239 | bool src_known = tnum_is_const(src_reg->var_off); |
7240 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
7241 | s64 smin_val = src_reg->smin_value; |
7242 | u64 umin_val = src_reg->umin_value; | |
7243 | ||
3f50f132 | 7244 | if (src_known && dst_known) { |
4fbb38a3 | 7245 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
7246 | return; |
7247 | } | |
7248 | ||
07cd2631 JF |
7249 | /* We get our maximum from the var_off, and our minimum is the |
7250 | * maximum of the operands' minima | |
7251 | */ | |
07cd2631 JF |
7252 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
7253 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
7254 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
7255 | /* Lose signed bounds when ORing negative numbers, | |
7256 | * ain't nobody got time for that. | |
7257 | */ | |
7258 | dst_reg->smin_value = S64_MIN; | |
7259 | dst_reg->smax_value = S64_MAX; | |
7260 | } else { | |
7261 | /* ORing two positives gives a positive, so safe to | |
7262 | * cast result into s64. | |
7263 | */ | |
7264 | dst_reg->smin_value = dst_reg->umin_value; | |
7265 | dst_reg->smax_value = dst_reg->umax_value; | |
7266 | } | |
7267 | /* We may learn something more from the var_off */ | |
7268 | __update_reg_bounds(dst_reg); | |
7269 | } | |
7270 | ||
2921c90d YS |
7271 | static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg, |
7272 | struct bpf_reg_state *src_reg) | |
7273 | { | |
7274 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
7275 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
7276 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
7277 | s32 smin_val = src_reg->s32_min_value; | |
7278 | ||
049c4e13 DB |
7279 | if (src_known && dst_known) { |
7280 | __mark_reg32_known(dst_reg, var32_off.value); | |
2921c90d | 7281 | return; |
049c4e13 | 7282 | } |
2921c90d YS |
7283 | |
7284 | /* We get both minimum and maximum from the var32_off. */ | |
7285 | dst_reg->u32_min_value = var32_off.value; | |
7286 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
7287 | ||
7288 | if (dst_reg->s32_min_value >= 0 && smin_val >= 0) { | |
7289 | /* XORing two positive sign numbers gives a positive, | |
7290 | * so safe to cast u32 result into s32. | |
7291 | */ | |
7292 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
7293 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
7294 | } else { | |
7295 | dst_reg->s32_min_value = S32_MIN; | |
7296 | dst_reg->s32_max_value = S32_MAX; | |
7297 | } | |
7298 | } | |
7299 | ||
7300 | static void scalar_min_max_xor(struct bpf_reg_state *dst_reg, | |
7301 | struct bpf_reg_state *src_reg) | |
7302 | { | |
7303 | bool src_known = tnum_is_const(src_reg->var_off); | |
7304 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
7305 | s64 smin_val = src_reg->smin_value; | |
7306 | ||
7307 | if (src_known && dst_known) { | |
7308 | /* dst_reg->var_off.value has been updated earlier */ | |
7309 | __mark_reg_known(dst_reg, dst_reg->var_off.value); | |
7310 | return; | |
7311 | } | |
7312 | ||
7313 | /* We get both minimum and maximum from the var_off. */ | |
7314 | dst_reg->umin_value = dst_reg->var_off.value; | |
7315 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
7316 | ||
7317 | if (dst_reg->smin_value >= 0 && smin_val >= 0) { | |
7318 | /* XORing two positive sign numbers gives a positive, | |
7319 | * so safe to cast u64 result into s64. | |
7320 | */ | |
7321 | dst_reg->smin_value = dst_reg->umin_value; | |
7322 | dst_reg->smax_value = dst_reg->umax_value; | |
7323 | } else { | |
7324 | dst_reg->smin_value = S64_MIN; | |
7325 | dst_reg->smax_value = S64_MAX; | |
7326 | } | |
7327 | ||
7328 | __update_reg_bounds(dst_reg); | |
7329 | } | |
7330 | ||
3f50f132 JF |
7331 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
7332 | u64 umin_val, u64 umax_val) | |
07cd2631 | 7333 | { |
07cd2631 JF |
7334 | /* We lose all sign bit information (except what we can pick |
7335 | * up from var_off) | |
7336 | */ | |
3f50f132 JF |
7337 | dst_reg->s32_min_value = S32_MIN; |
7338 | dst_reg->s32_max_value = S32_MAX; | |
7339 | /* If we might shift our top bit out, then we know nothing */ | |
7340 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
7341 | dst_reg->u32_min_value = 0; | |
7342 | dst_reg->u32_max_value = U32_MAX; | |
7343 | } else { | |
7344 | dst_reg->u32_min_value <<= umin_val; | |
7345 | dst_reg->u32_max_value <<= umax_val; | |
7346 | } | |
7347 | } | |
7348 | ||
7349 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
7350 | struct bpf_reg_state *src_reg) | |
7351 | { | |
7352 | u32 umax_val = src_reg->u32_max_value; | |
7353 | u32 umin_val = src_reg->u32_min_value; | |
7354 | /* u32 alu operation will zext upper bits */ | |
7355 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
7356 | ||
7357 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
7358 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
7359 | /* Not required but being careful mark reg64 bounds as unknown so | |
7360 | * that we are forced to pick them up from tnum and zext later and | |
7361 | * if some path skips this step we are still safe. | |
7362 | */ | |
7363 | __mark_reg64_unbounded(dst_reg); | |
7364 | __update_reg32_bounds(dst_reg); | |
7365 | } | |
7366 | ||
7367 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
7368 | u64 umin_val, u64 umax_val) | |
7369 | { | |
7370 | /* Special case <<32 because it is a common compiler pattern to sign | |
7371 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
7372 | * positive we know this shift will also be positive so we can track | |
7373 | * bounds correctly. Otherwise we lose all sign bit information except | |
7374 | * what we can pick up from var_off. Perhaps we can generalize this | |
7375 | * later to shifts of any length. | |
7376 | */ | |
7377 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
7378 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
7379 | else | |
7380 | dst_reg->smax_value = S64_MAX; | |
7381 | ||
7382 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
7383 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
7384 | else | |
7385 | dst_reg->smin_value = S64_MIN; | |
7386 | ||
07cd2631 JF |
7387 | /* If we might shift our top bit out, then we know nothing */ |
7388 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
7389 | dst_reg->umin_value = 0; | |
7390 | dst_reg->umax_value = U64_MAX; | |
7391 | } else { | |
7392 | dst_reg->umin_value <<= umin_val; | |
7393 | dst_reg->umax_value <<= umax_val; | |
7394 | } | |
3f50f132 JF |
7395 | } |
7396 | ||
7397 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
7398 | struct bpf_reg_state *src_reg) | |
7399 | { | |
7400 | u64 umax_val = src_reg->umax_value; | |
7401 | u64 umin_val = src_reg->umin_value; | |
7402 | ||
7403 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
7404 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
7405 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
7406 | ||
07cd2631 JF |
7407 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
7408 | /* We may learn something more from the var_off */ | |
7409 | __update_reg_bounds(dst_reg); | |
7410 | } | |
7411 | ||
3f50f132 JF |
7412 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
7413 | struct bpf_reg_state *src_reg) | |
7414 | { | |
7415 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
7416 | u32 umax_val = src_reg->u32_max_value; | |
7417 | u32 umin_val = src_reg->u32_min_value; | |
7418 | ||
7419 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
7420 | * be negative, then either: | |
7421 | * 1) src_reg might be zero, so the sign bit of the result is | |
7422 | * unknown, so we lose our signed bounds | |
7423 | * 2) it's known negative, thus the unsigned bounds capture the | |
7424 | * signed bounds | |
7425 | * 3) the signed bounds cross zero, so they tell us nothing | |
7426 | * about the result | |
7427 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 7428 | * unsigned bounds capture the signed bounds. |
3f50f132 JF |
7429 | * Thus, in all cases it suffices to blow away our signed bounds |
7430 | * and rely on inferring new ones from the unsigned bounds and | |
7431 | * var_off of the result. | |
7432 | */ | |
7433 | dst_reg->s32_min_value = S32_MIN; | |
7434 | dst_reg->s32_max_value = S32_MAX; | |
7435 | ||
7436 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
7437 | dst_reg->u32_min_value >>= umax_val; | |
7438 | dst_reg->u32_max_value >>= umin_val; | |
7439 | ||
7440 | __mark_reg64_unbounded(dst_reg); | |
7441 | __update_reg32_bounds(dst_reg); | |
7442 | } | |
7443 | ||
07cd2631 JF |
7444 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
7445 | struct bpf_reg_state *src_reg) | |
7446 | { | |
7447 | u64 umax_val = src_reg->umax_value; | |
7448 | u64 umin_val = src_reg->umin_value; | |
7449 | ||
7450 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
7451 | * be negative, then either: | |
7452 | * 1) src_reg might be zero, so the sign bit of the result is | |
7453 | * unknown, so we lose our signed bounds | |
7454 | * 2) it's known negative, thus the unsigned bounds capture the | |
7455 | * signed bounds | |
7456 | * 3) the signed bounds cross zero, so they tell us nothing | |
7457 | * about the result | |
7458 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 7459 | * unsigned bounds capture the signed bounds. |
07cd2631 JF |
7460 | * Thus, in all cases it suffices to blow away our signed bounds |
7461 | * and rely on inferring new ones from the unsigned bounds and | |
7462 | * var_off of the result. | |
7463 | */ | |
7464 | dst_reg->smin_value = S64_MIN; | |
7465 | dst_reg->smax_value = S64_MAX; | |
7466 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
7467 | dst_reg->umin_value >>= umax_val; | |
7468 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
7469 | |
7470 | /* Its not easy to operate on alu32 bounds here because it depends | |
7471 | * on bits being shifted in. Take easy way out and mark unbounded | |
7472 | * so we can recalculate later from tnum. | |
7473 | */ | |
7474 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
7475 | __update_reg_bounds(dst_reg); |
7476 | } | |
7477 | ||
3f50f132 JF |
7478 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
7479 | struct bpf_reg_state *src_reg) | |
07cd2631 | 7480 | { |
3f50f132 | 7481 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
7482 | |
7483 | /* Upon reaching here, src_known is true and | |
7484 | * umax_val is equal to umin_val. | |
7485 | */ | |
3f50f132 JF |
7486 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
7487 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 7488 | |
3f50f132 JF |
7489 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
7490 | ||
7491 | /* blow away the dst_reg umin_value/umax_value and rely on | |
7492 | * dst_reg var_off to refine the result. | |
7493 | */ | |
7494 | dst_reg->u32_min_value = 0; | |
7495 | dst_reg->u32_max_value = U32_MAX; | |
7496 | ||
7497 | __mark_reg64_unbounded(dst_reg); | |
7498 | __update_reg32_bounds(dst_reg); | |
7499 | } | |
7500 | ||
7501 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
7502 | struct bpf_reg_state *src_reg) | |
7503 | { | |
7504 | u64 umin_val = src_reg->umin_value; | |
7505 | ||
7506 | /* Upon reaching here, src_known is true and umax_val is equal | |
7507 | * to umin_val. | |
7508 | */ | |
7509 | dst_reg->smin_value >>= umin_val; | |
7510 | dst_reg->smax_value >>= umin_val; | |
7511 | ||
7512 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
7513 | |
7514 | /* blow away the dst_reg umin_value/umax_value and rely on | |
7515 | * dst_reg var_off to refine the result. | |
7516 | */ | |
7517 | dst_reg->umin_value = 0; | |
7518 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
7519 | |
7520 | /* Its not easy to operate on alu32 bounds here because it depends | |
7521 | * on bits being shifted in from upper 32-bits. Take easy way out | |
7522 | * and mark unbounded so we can recalculate later from tnum. | |
7523 | */ | |
7524 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
7525 | __update_reg_bounds(dst_reg); |
7526 | } | |
7527 | ||
468f6eaf JH |
7528 | /* WARNING: This function does calculations on 64-bit values, but the actual |
7529 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
7530 | * need extra checks in the 32-bit case. | |
7531 | */ | |
f1174f77 EC |
7532 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
7533 | struct bpf_insn *insn, | |
7534 | struct bpf_reg_state *dst_reg, | |
7535 | struct bpf_reg_state src_reg) | |
969bf05e | 7536 | { |
638f5b90 | 7537 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 7538 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 7539 | bool src_known; |
b03c9f9f EC |
7540 | s64 smin_val, smax_val; |
7541 | u64 umin_val, umax_val; | |
3f50f132 JF |
7542 | s32 s32_min_val, s32_max_val; |
7543 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 7544 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
3f50f132 | 7545 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
a6aaece0 | 7546 | int ret; |
b799207e | 7547 | |
b03c9f9f EC |
7548 | smin_val = src_reg.smin_value; |
7549 | smax_val = src_reg.smax_value; | |
7550 | umin_val = src_reg.umin_value; | |
7551 | umax_val = src_reg.umax_value; | |
f23cc643 | 7552 | |
3f50f132 JF |
7553 | s32_min_val = src_reg.s32_min_value; |
7554 | s32_max_val = src_reg.s32_max_value; | |
7555 | u32_min_val = src_reg.u32_min_value; | |
7556 | u32_max_val = src_reg.u32_max_value; | |
7557 | ||
7558 | if (alu32) { | |
7559 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
7560 | if ((src_known && |
7561 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
7562 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
7563 | /* Taint dst register if offset had invalid bounds | |
7564 | * derived from e.g. dead branches. | |
7565 | */ | |
7566 | __mark_reg_unknown(env, dst_reg); | |
7567 | return 0; | |
7568 | } | |
7569 | } else { | |
7570 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
7571 | if ((src_known && |
7572 | (smin_val != smax_val || umin_val != umax_val)) || | |
7573 | smin_val > smax_val || umin_val > umax_val) { | |
7574 | /* Taint dst register if offset had invalid bounds | |
7575 | * derived from e.g. dead branches. | |
7576 | */ | |
7577 | __mark_reg_unknown(env, dst_reg); | |
7578 | return 0; | |
7579 | } | |
6f16101e DB |
7580 | } |
7581 | ||
bb7f0f98 AS |
7582 | if (!src_known && |
7583 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 7584 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
7585 | return 0; |
7586 | } | |
7587 | ||
f5288193 DB |
7588 | if (sanitize_needed(opcode)) { |
7589 | ret = sanitize_val_alu(env, insn); | |
7590 | if (ret < 0) | |
7591 | return sanitize_err(env, insn, ret, NULL, NULL); | |
7592 | } | |
7593 | ||
3f50f132 JF |
7594 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
7595 | * There are two classes of instructions: The first class we track both | |
7596 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
7597 | * greatest amount of precision when alu operations are mixed with jmp32 | |
7598 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
7599 | * and BPF_OR. This is possible because these ops have fairly easy to | |
7600 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
7601 | * See alu32 verifier tests for examples. The second class of | |
7602 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
7603 | * with regards to tracking sign/unsigned bounds because the bits may | |
7604 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
7605 | * the reg unbounded in the subreg bound space and use the resulting | |
7606 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
7607 | */ | |
48461135 JB |
7608 | switch (opcode) { |
7609 | case BPF_ADD: | |
3f50f132 | 7610 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 7611 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 7612 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
7613 | break; |
7614 | case BPF_SUB: | |
3f50f132 | 7615 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 7616 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 7617 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
7618 | break; |
7619 | case BPF_MUL: | |
3f50f132 JF |
7620 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
7621 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 7622 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
7623 | break; |
7624 | case BPF_AND: | |
3f50f132 JF |
7625 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
7626 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 7627 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
7628 | break; |
7629 | case BPF_OR: | |
3f50f132 JF |
7630 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
7631 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 7632 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 | 7633 | break; |
2921c90d YS |
7634 | case BPF_XOR: |
7635 | dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off); | |
7636 | scalar32_min_max_xor(dst_reg, &src_reg); | |
7637 | scalar_min_max_xor(dst_reg, &src_reg); | |
7638 | break; | |
48461135 | 7639 | case BPF_LSH: |
468f6eaf JH |
7640 | if (umax_val >= insn_bitness) { |
7641 | /* Shifts greater than 31 or 63 are undefined. | |
7642 | * This includes shifts by a negative number. | |
b03c9f9f | 7643 | */ |
61bd5218 | 7644 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
7645 | break; |
7646 | } | |
3f50f132 JF |
7647 | if (alu32) |
7648 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
7649 | else | |
7650 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
7651 | break; |
7652 | case BPF_RSH: | |
468f6eaf JH |
7653 | if (umax_val >= insn_bitness) { |
7654 | /* Shifts greater than 31 or 63 are undefined. | |
7655 | * This includes shifts by a negative number. | |
b03c9f9f | 7656 | */ |
61bd5218 | 7657 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
7658 | break; |
7659 | } | |
3f50f132 JF |
7660 | if (alu32) |
7661 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
7662 | else | |
7663 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 7664 | break; |
9cbe1f5a YS |
7665 | case BPF_ARSH: |
7666 | if (umax_val >= insn_bitness) { | |
7667 | /* Shifts greater than 31 or 63 are undefined. | |
7668 | * This includes shifts by a negative number. | |
7669 | */ | |
7670 | mark_reg_unknown(env, regs, insn->dst_reg); | |
7671 | break; | |
7672 | } | |
3f50f132 JF |
7673 | if (alu32) |
7674 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
7675 | else | |
7676 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 7677 | break; |
48461135 | 7678 | default: |
61bd5218 | 7679 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
7680 | break; |
7681 | } | |
7682 | ||
3f50f132 JF |
7683 | /* ALU32 ops are zero extended into 64bit register */ |
7684 | if (alu32) | |
7685 | zext_32_to_64(dst_reg); | |
468f6eaf | 7686 | |
294f2fc6 | 7687 | __update_reg_bounds(dst_reg); |
b03c9f9f EC |
7688 | __reg_deduce_bounds(dst_reg); |
7689 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
7690 | return 0; |
7691 | } | |
7692 | ||
7693 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
7694 | * and var_off. | |
7695 | */ | |
7696 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
7697 | struct bpf_insn *insn) | |
7698 | { | |
f4d7e40a AS |
7699 | struct bpf_verifier_state *vstate = env->cur_state; |
7700 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
7701 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
7702 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
7703 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 7704 | int err; |
f1174f77 EC |
7705 | |
7706 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
7707 | src_reg = NULL; |
7708 | if (dst_reg->type != SCALAR_VALUE) | |
7709 | ptr_reg = dst_reg; | |
75748837 AS |
7710 | else |
7711 | /* Make sure ID is cleared otherwise dst_reg min/max could be | |
7712 | * incorrectly propagated into other registers by find_equal_scalars() | |
7713 | */ | |
7714 | dst_reg->id = 0; | |
f1174f77 EC |
7715 | if (BPF_SRC(insn->code) == BPF_X) { |
7716 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
7717 | if (src_reg->type != SCALAR_VALUE) { |
7718 | if (dst_reg->type != SCALAR_VALUE) { | |
7719 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
7720 | * an arbitrary scalar. Disallow all math except |
7721 | * pointer subtraction | |
f1174f77 | 7722 | */ |
dd066823 | 7723 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
7724 | mark_reg_unknown(env, regs, insn->dst_reg); |
7725 | return 0; | |
f1174f77 | 7726 | } |
82abbf8d AS |
7727 | verbose(env, "R%d pointer %s pointer prohibited\n", |
7728 | insn->dst_reg, | |
7729 | bpf_alu_string[opcode >> 4]); | |
7730 | return -EACCES; | |
f1174f77 EC |
7731 | } else { |
7732 | /* scalar += pointer | |
7733 | * This is legal, but we have to reverse our | |
7734 | * src/dest handling in computing the range | |
7735 | */ | |
b5dc0163 AS |
7736 | err = mark_chain_precision(env, insn->dst_reg); |
7737 | if (err) | |
7738 | return err; | |
82abbf8d AS |
7739 | return adjust_ptr_min_max_vals(env, insn, |
7740 | src_reg, dst_reg); | |
f1174f77 EC |
7741 | } |
7742 | } else if (ptr_reg) { | |
7743 | /* pointer += scalar */ | |
b5dc0163 AS |
7744 | err = mark_chain_precision(env, insn->src_reg); |
7745 | if (err) | |
7746 | return err; | |
82abbf8d AS |
7747 | return adjust_ptr_min_max_vals(env, insn, |
7748 | dst_reg, src_reg); | |
f1174f77 EC |
7749 | } |
7750 | } else { | |
7751 | /* Pretend the src is a reg with a known value, since we only | |
7752 | * need to be able to read from this state. | |
7753 | */ | |
7754 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 7755 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 7756 | src_reg = &off_reg; |
82abbf8d AS |
7757 | if (ptr_reg) /* pointer += K */ |
7758 | return adjust_ptr_min_max_vals(env, insn, | |
7759 | ptr_reg, src_reg); | |
f1174f77 EC |
7760 | } |
7761 | ||
7762 | /* Got here implies adding two SCALAR_VALUEs */ | |
7763 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 7764 | print_verifier_state(env, state); |
61bd5218 | 7765 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
7766 | return -EINVAL; |
7767 | } | |
7768 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 7769 | print_verifier_state(env, state); |
61bd5218 | 7770 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
7771 | return -EINVAL; |
7772 | } | |
7773 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
7774 | } |
7775 | ||
17a52670 | 7776 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 7777 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 7778 | { |
638f5b90 | 7779 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
7780 | u8 opcode = BPF_OP(insn->code); |
7781 | int err; | |
7782 | ||
7783 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
7784 | if (opcode == BPF_NEG) { | |
7785 | if (BPF_SRC(insn->code) != 0 || | |
7786 | insn->src_reg != BPF_REG_0 || | |
7787 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 7788 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
7789 | return -EINVAL; |
7790 | } | |
7791 | } else { | |
7792 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
7793 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
7794 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 7795 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
7796 | return -EINVAL; |
7797 | } | |
7798 | } | |
7799 | ||
7800 | /* check src operand */ | |
dc503a8a | 7801 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
7802 | if (err) |
7803 | return err; | |
7804 | ||
1be7f75d | 7805 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 7806 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
7807 | insn->dst_reg); |
7808 | return -EACCES; | |
7809 | } | |
7810 | ||
17a52670 | 7811 | /* check dest operand */ |
dc503a8a | 7812 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
7813 | if (err) |
7814 | return err; | |
7815 | ||
7816 | } else if (opcode == BPF_MOV) { | |
7817 | ||
7818 | if (BPF_SRC(insn->code) == BPF_X) { | |
7819 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 7820 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
7821 | return -EINVAL; |
7822 | } | |
7823 | ||
7824 | /* check src operand */ | |
dc503a8a | 7825 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
7826 | if (err) |
7827 | return err; | |
7828 | } else { | |
7829 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 7830 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
7831 | return -EINVAL; |
7832 | } | |
7833 | } | |
7834 | ||
fbeb1603 AF |
7835 | /* check dest operand, mark as required later */ |
7836 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
7837 | if (err) |
7838 | return err; | |
7839 | ||
7840 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
7841 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
7842 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
7843 | ||
17a52670 AS |
7844 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
7845 | /* case: R1 = R2 | |
7846 | * copy register state to dest reg | |
7847 | */ | |
75748837 AS |
7848 | if (src_reg->type == SCALAR_VALUE && !src_reg->id) |
7849 | /* Assign src and dst registers the same ID | |
7850 | * that will be used by find_equal_scalars() | |
7851 | * to propagate min/max range. | |
7852 | */ | |
7853 | src_reg->id = ++env->id_gen; | |
e434b8cd JW |
7854 | *dst_reg = *src_reg; |
7855 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 7856 | dst_reg->subreg_def = DEF_NOT_SUBREG; |
17a52670 | 7857 | } else { |
f1174f77 | 7858 | /* R1 = (u32) R2 */ |
1be7f75d | 7859 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
7860 | verbose(env, |
7861 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
7862 | insn->src_reg); |
7863 | return -EACCES; | |
e434b8cd JW |
7864 | } else if (src_reg->type == SCALAR_VALUE) { |
7865 | *dst_reg = *src_reg; | |
75748837 AS |
7866 | /* Make sure ID is cleared otherwise |
7867 | * dst_reg min/max could be incorrectly | |
7868 | * propagated into src_reg by find_equal_scalars() | |
7869 | */ | |
7870 | dst_reg->id = 0; | |
e434b8cd | 7871 | dst_reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 7872 | dst_reg->subreg_def = env->insn_idx + 1; |
e434b8cd JW |
7873 | } else { |
7874 | mark_reg_unknown(env, regs, | |
7875 | insn->dst_reg); | |
1be7f75d | 7876 | } |
3f50f132 | 7877 | zext_32_to_64(dst_reg); |
17a52670 AS |
7878 | } |
7879 | } else { | |
7880 | /* case: R = imm | |
7881 | * remember the value we stored into this reg | |
7882 | */ | |
fbeb1603 AF |
7883 | /* clear any state __mark_reg_known doesn't set */ |
7884 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 7885 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
7886 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
7887 | __mark_reg_known(regs + insn->dst_reg, | |
7888 | insn->imm); | |
7889 | } else { | |
7890 | __mark_reg_known(regs + insn->dst_reg, | |
7891 | (u32)insn->imm); | |
7892 | } | |
17a52670 AS |
7893 | } |
7894 | ||
7895 | } else if (opcode > BPF_END) { | |
61bd5218 | 7896 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
7897 | return -EINVAL; |
7898 | ||
7899 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
7900 | ||
17a52670 AS |
7901 | if (BPF_SRC(insn->code) == BPF_X) { |
7902 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 7903 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
7904 | return -EINVAL; |
7905 | } | |
7906 | /* check src1 operand */ | |
dc503a8a | 7907 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
7908 | if (err) |
7909 | return err; | |
7910 | } else { | |
7911 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 7912 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
7913 | return -EINVAL; |
7914 | } | |
7915 | } | |
7916 | ||
7917 | /* check src2 operand */ | |
dc503a8a | 7918 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
7919 | if (err) |
7920 | return err; | |
7921 | ||
7922 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
7923 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 7924 | verbose(env, "div by zero\n"); |
17a52670 AS |
7925 | return -EINVAL; |
7926 | } | |
7927 | ||
229394e8 RV |
7928 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
7929 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
7930 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
7931 | ||
7932 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 7933 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
7934 | return -EINVAL; |
7935 | } | |
7936 | } | |
7937 | ||
1a0dc1ac | 7938 | /* check dest operand */ |
dc503a8a | 7939 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
7940 | if (err) |
7941 | return err; | |
7942 | ||
f1174f77 | 7943 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
7944 | } |
7945 | ||
7946 | return 0; | |
7947 | } | |
7948 | ||
c6a9efa1 PC |
7949 | static void __find_good_pkt_pointers(struct bpf_func_state *state, |
7950 | struct bpf_reg_state *dst_reg, | |
6d94e741 | 7951 | enum bpf_reg_type type, int new_range) |
c6a9efa1 PC |
7952 | { |
7953 | struct bpf_reg_state *reg; | |
7954 | int i; | |
7955 | ||
7956 | for (i = 0; i < MAX_BPF_REG; i++) { | |
7957 | reg = &state->regs[i]; | |
7958 | if (reg->type == type && reg->id == dst_reg->id) | |
7959 | /* keep the maximum range already checked */ | |
7960 | reg->range = max(reg->range, new_range); | |
7961 | } | |
7962 | ||
7963 | bpf_for_each_spilled_reg(i, state, reg) { | |
7964 | if (!reg) | |
7965 | continue; | |
7966 | if (reg->type == type && reg->id == dst_reg->id) | |
7967 | reg->range = max(reg->range, new_range); | |
7968 | } | |
7969 | } | |
7970 | ||
f4d7e40a | 7971 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 7972 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 7973 | enum bpf_reg_type type, |
fb2a311a | 7974 | bool range_right_open) |
969bf05e | 7975 | { |
6d94e741 | 7976 | int new_range, i; |
2d2be8ca | 7977 | |
fb2a311a DB |
7978 | if (dst_reg->off < 0 || |
7979 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
7980 | /* This doesn't give us any range */ |
7981 | return; | |
7982 | ||
b03c9f9f EC |
7983 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
7984 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
7985 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
7986 | * than pkt_end, but that's because it's also less than pkt. | |
7987 | */ | |
7988 | return; | |
7989 | ||
fb2a311a DB |
7990 | new_range = dst_reg->off; |
7991 | if (range_right_open) | |
7992 | new_range--; | |
7993 | ||
7994 | /* Examples for register markings: | |
2d2be8ca | 7995 | * |
fb2a311a | 7996 | * pkt_data in dst register: |
2d2be8ca DB |
7997 | * |
7998 | * r2 = r3; | |
7999 | * r2 += 8; | |
8000 | * if (r2 > pkt_end) goto <handle exception> | |
8001 | * <access okay> | |
8002 | * | |
b4e432f1 DB |
8003 | * r2 = r3; |
8004 | * r2 += 8; | |
8005 | * if (r2 < pkt_end) goto <access okay> | |
8006 | * <handle exception> | |
8007 | * | |
2d2be8ca DB |
8008 | * Where: |
8009 | * r2 == dst_reg, pkt_end == src_reg | |
8010 | * r2=pkt(id=n,off=8,r=0) | |
8011 | * r3=pkt(id=n,off=0,r=0) | |
8012 | * | |
fb2a311a | 8013 | * pkt_data in src register: |
2d2be8ca DB |
8014 | * |
8015 | * r2 = r3; | |
8016 | * r2 += 8; | |
8017 | * if (pkt_end >= r2) goto <access okay> | |
8018 | * <handle exception> | |
8019 | * | |
b4e432f1 DB |
8020 | * r2 = r3; |
8021 | * r2 += 8; | |
8022 | * if (pkt_end <= r2) goto <handle exception> | |
8023 | * <access okay> | |
8024 | * | |
2d2be8ca DB |
8025 | * Where: |
8026 | * pkt_end == dst_reg, r2 == src_reg | |
8027 | * r2=pkt(id=n,off=8,r=0) | |
8028 | * r3=pkt(id=n,off=0,r=0) | |
8029 | * | |
8030 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
8031 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
8032 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
8033 | * the check. | |
969bf05e | 8034 | */ |
2d2be8ca | 8035 | |
f1174f77 EC |
8036 | /* If our ids match, then we must have the same max_value. And we |
8037 | * don't care about the other reg's fixed offset, since if it's too big | |
8038 | * the range won't allow anything. | |
8039 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
8040 | */ | |
c6a9efa1 PC |
8041 | for (i = 0; i <= vstate->curframe; i++) |
8042 | __find_good_pkt_pointers(vstate->frame[i], dst_reg, type, | |
8043 | new_range); | |
969bf05e AS |
8044 | } |
8045 | ||
3f50f132 | 8046 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 8047 | { |
3f50f132 JF |
8048 | struct tnum subreg = tnum_subreg(reg->var_off); |
8049 | s32 sval = (s32)val; | |
a72dafaf | 8050 | |
3f50f132 JF |
8051 | switch (opcode) { |
8052 | case BPF_JEQ: | |
8053 | if (tnum_is_const(subreg)) | |
8054 | return !!tnum_equals_const(subreg, val); | |
8055 | break; | |
8056 | case BPF_JNE: | |
8057 | if (tnum_is_const(subreg)) | |
8058 | return !tnum_equals_const(subreg, val); | |
8059 | break; | |
8060 | case BPF_JSET: | |
8061 | if ((~subreg.mask & subreg.value) & val) | |
8062 | return 1; | |
8063 | if (!((subreg.mask | subreg.value) & val)) | |
8064 | return 0; | |
8065 | break; | |
8066 | case BPF_JGT: | |
8067 | if (reg->u32_min_value > val) | |
8068 | return 1; | |
8069 | else if (reg->u32_max_value <= val) | |
8070 | return 0; | |
8071 | break; | |
8072 | case BPF_JSGT: | |
8073 | if (reg->s32_min_value > sval) | |
8074 | return 1; | |
ee114dd6 | 8075 | else if (reg->s32_max_value <= sval) |
3f50f132 JF |
8076 | return 0; |
8077 | break; | |
8078 | case BPF_JLT: | |
8079 | if (reg->u32_max_value < val) | |
8080 | return 1; | |
8081 | else if (reg->u32_min_value >= val) | |
8082 | return 0; | |
8083 | break; | |
8084 | case BPF_JSLT: | |
8085 | if (reg->s32_max_value < sval) | |
8086 | return 1; | |
8087 | else if (reg->s32_min_value >= sval) | |
8088 | return 0; | |
8089 | break; | |
8090 | case BPF_JGE: | |
8091 | if (reg->u32_min_value >= val) | |
8092 | return 1; | |
8093 | else if (reg->u32_max_value < val) | |
8094 | return 0; | |
8095 | break; | |
8096 | case BPF_JSGE: | |
8097 | if (reg->s32_min_value >= sval) | |
8098 | return 1; | |
8099 | else if (reg->s32_max_value < sval) | |
8100 | return 0; | |
8101 | break; | |
8102 | case BPF_JLE: | |
8103 | if (reg->u32_max_value <= val) | |
8104 | return 1; | |
8105 | else if (reg->u32_min_value > val) | |
8106 | return 0; | |
8107 | break; | |
8108 | case BPF_JSLE: | |
8109 | if (reg->s32_max_value <= sval) | |
8110 | return 1; | |
8111 | else if (reg->s32_min_value > sval) | |
8112 | return 0; | |
8113 | break; | |
8114 | } | |
4f7b3e82 | 8115 | |
3f50f132 JF |
8116 | return -1; |
8117 | } | |
092ed096 | 8118 | |
3f50f132 JF |
8119 | |
8120 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
8121 | { | |
8122 | s64 sval = (s64)val; | |
a72dafaf | 8123 | |
4f7b3e82 AS |
8124 | switch (opcode) { |
8125 | case BPF_JEQ: | |
8126 | if (tnum_is_const(reg->var_off)) | |
8127 | return !!tnum_equals_const(reg->var_off, val); | |
8128 | break; | |
8129 | case BPF_JNE: | |
8130 | if (tnum_is_const(reg->var_off)) | |
8131 | return !tnum_equals_const(reg->var_off, val); | |
8132 | break; | |
960ea056 JK |
8133 | case BPF_JSET: |
8134 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
8135 | return 1; | |
8136 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
8137 | return 0; | |
8138 | break; | |
4f7b3e82 AS |
8139 | case BPF_JGT: |
8140 | if (reg->umin_value > val) | |
8141 | return 1; | |
8142 | else if (reg->umax_value <= val) | |
8143 | return 0; | |
8144 | break; | |
8145 | case BPF_JSGT: | |
a72dafaf | 8146 | if (reg->smin_value > sval) |
4f7b3e82 | 8147 | return 1; |
ee114dd6 | 8148 | else if (reg->smax_value <= sval) |
4f7b3e82 AS |
8149 | return 0; |
8150 | break; | |
8151 | case BPF_JLT: | |
8152 | if (reg->umax_value < val) | |
8153 | return 1; | |
8154 | else if (reg->umin_value >= val) | |
8155 | return 0; | |
8156 | break; | |
8157 | case BPF_JSLT: | |
a72dafaf | 8158 | if (reg->smax_value < sval) |
4f7b3e82 | 8159 | return 1; |
a72dafaf | 8160 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
8161 | return 0; |
8162 | break; | |
8163 | case BPF_JGE: | |
8164 | if (reg->umin_value >= val) | |
8165 | return 1; | |
8166 | else if (reg->umax_value < val) | |
8167 | return 0; | |
8168 | break; | |
8169 | case BPF_JSGE: | |
a72dafaf | 8170 | if (reg->smin_value >= sval) |
4f7b3e82 | 8171 | return 1; |
a72dafaf | 8172 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
8173 | return 0; |
8174 | break; | |
8175 | case BPF_JLE: | |
8176 | if (reg->umax_value <= val) | |
8177 | return 1; | |
8178 | else if (reg->umin_value > val) | |
8179 | return 0; | |
8180 | break; | |
8181 | case BPF_JSLE: | |
a72dafaf | 8182 | if (reg->smax_value <= sval) |
4f7b3e82 | 8183 | return 1; |
a72dafaf | 8184 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
8185 | return 0; |
8186 | break; | |
8187 | } | |
8188 | ||
8189 | return -1; | |
8190 | } | |
8191 | ||
3f50f132 JF |
8192 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
8193 | * and return: | |
8194 | * 1 - branch will be taken and "goto target" will be executed | |
8195 | * 0 - branch will not be taken and fall-through to next insn | |
8196 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
8197 | * range [0,10] | |
604dca5e | 8198 | */ |
3f50f132 JF |
8199 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
8200 | bool is_jmp32) | |
604dca5e | 8201 | { |
cac616db JF |
8202 | if (__is_pointer_value(false, reg)) { |
8203 | if (!reg_type_not_null(reg->type)) | |
8204 | return -1; | |
8205 | ||
8206 | /* If pointer is valid tests against zero will fail so we can | |
8207 | * use this to direct branch taken. | |
8208 | */ | |
8209 | if (val != 0) | |
8210 | return -1; | |
8211 | ||
8212 | switch (opcode) { | |
8213 | case BPF_JEQ: | |
8214 | return 0; | |
8215 | case BPF_JNE: | |
8216 | return 1; | |
8217 | default: | |
8218 | return -1; | |
8219 | } | |
8220 | } | |
604dca5e | 8221 | |
3f50f132 JF |
8222 | if (is_jmp32) |
8223 | return is_branch32_taken(reg, val, opcode); | |
8224 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
8225 | } |
8226 | ||
6d94e741 AS |
8227 | static int flip_opcode(u32 opcode) |
8228 | { | |
8229 | /* How can we transform "a <op> b" into "b <op> a"? */ | |
8230 | static const u8 opcode_flip[16] = { | |
8231 | /* these stay the same */ | |
8232 | [BPF_JEQ >> 4] = BPF_JEQ, | |
8233 | [BPF_JNE >> 4] = BPF_JNE, | |
8234 | [BPF_JSET >> 4] = BPF_JSET, | |
8235 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
8236 | [BPF_JGE >> 4] = BPF_JLE, | |
8237 | [BPF_JGT >> 4] = BPF_JLT, | |
8238 | [BPF_JLE >> 4] = BPF_JGE, | |
8239 | [BPF_JLT >> 4] = BPF_JGT, | |
8240 | [BPF_JSGE >> 4] = BPF_JSLE, | |
8241 | [BPF_JSGT >> 4] = BPF_JSLT, | |
8242 | [BPF_JSLE >> 4] = BPF_JSGE, | |
8243 | [BPF_JSLT >> 4] = BPF_JSGT | |
8244 | }; | |
8245 | return opcode_flip[opcode >> 4]; | |
8246 | } | |
8247 | ||
8248 | static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg, | |
8249 | struct bpf_reg_state *src_reg, | |
8250 | u8 opcode) | |
8251 | { | |
8252 | struct bpf_reg_state *pkt; | |
8253 | ||
8254 | if (src_reg->type == PTR_TO_PACKET_END) { | |
8255 | pkt = dst_reg; | |
8256 | } else if (dst_reg->type == PTR_TO_PACKET_END) { | |
8257 | pkt = src_reg; | |
8258 | opcode = flip_opcode(opcode); | |
8259 | } else { | |
8260 | return -1; | |
8261 | } | |
8262 | ||
8263 | if (pkt->range >= 0) | |
8264 | return -1; | |
8265 | ||
8266 | switch (opcode) { | |
8267 | case BPF_JLE: | |
8268 | /* pkt <= pkt_end */ | |
8269 | fallthrough; | |
8270 | case BPF_JGT: | |
8271 | /* pkt > pkt_end */ | |
8272 | if (pkt->range == BEYOND_PKT_END) | |
8273 | /* pkt has at last one extra byte beyond pkt_end */ | |
8274 | return opcode == BPF_JGT; | |
8275 | break; | |
8276 | case BPF_JLT: | |
8277 | /* pkt < pkt_end */ | |
8278 | fallthrough; | |
8279 | case BPF_JGE: | |
8280 | /* pkt >= pkt_end */ | |
8281 | if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END) | |
8282 | return opcode == BPF_JGE; | |
8283 | break; | |
8284 | } | |
8285 | return -1; | |
8286 | } | |
8287 | ||
48461135 JB |
8288 | /* Adjusts the register min/max values in the case that the dst_reg is the |
8289 | * variable register that we are working on, and src_reg is a constant or we're | |
8290 | * simply doing a BPF_K check. | |
f1174f77 | 8291 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
8292 | */ |
8293 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
8294 | struct bpf_reg_state *false_reg, |
8295 | u64 val, u32 val32, | |
092ed096 | 8296 | u8 opcode, bool is_jmp32) |
48461135 | 8297 | { |
3f50f132 JF |
8298 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
8299 | struct tnum false_64off = false_reg->var_off; | |
8300 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
8301 | struct tnum true_64off = true_reg->var_off; | |
8302 | s64 sval = (s64)val; | |
8303 | s32 sval32 = (s32)val32; | |
a72dafaf | 8304 | |
f1174f77 EC |
8305 | /* If the dst_reg is a pointer, we can't learn anything about its |
8306 | * variable offset from the compare (unless src_reg were a pointer into | |
8307 | * the same object, but we don't bother with that. | |
8308 | * Since false_reg and true_reg have the same type by construction, we | |
8309 | * only need to check one of them for pointerness. | |
8310 | */ | |
8311 | if (__is_pointer_value(false, false_reg)) | |
8312 | return; | |
4cabc5b1 | 8313 | |
48461135 JB |
8314 | switch (opcode) { |
8315 | case BPF_JEQ: | |
48461135 | 8316 | case BPF_JNE: |
a72dafaf JW |
8317 | { |
8318 | struct bpf_reg_state *reg = | |
8319 | opcode == BPF_JEQ ? true_reg : false_reg; | |
8320 | ||
e688c3db AS |
8321 | /* JEQ/JNE comparison doesn't change the register equivalence. |
8322 | * r1 = r2; | |
8323 | * if (r1 == 42) goto label; | |
8324 | * ... | |
8325 | * label: // here both r1 and r2 are known to be 42. | |
8326 | * | |
8327 | * Hence when marking register as known preserve it's ID. | |
48461135 | 8328 | */ |
3f50f132 JF |
8329 | if (is_jmp32) |
8330 | __mark_reg32_known(reg, val32); | |
8331 | else | |
e688c3db | 8332 | ___mark_reg_known(reg, val); |
48461135 | 8333 | break; |
a72dafaf | 8334 | } |
960ea056 | 8335 | case BPF_JSET: |
3f50f132 JF |
8336 | if (is_jmp32) { |
8337 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
8338 | if (is_power_of_2(val32)) | |
8339 | true_32off = tnum_or(true_32off, | |
8340 | tnum_const(val32)); | |
8341 | } else { | |
8342 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
8343 | if (is_power_of_2(val)) | |
8344 | true_64off = tnum_or(true_64off, | |
8345 | tnum_const(val)); | |
8346 | } | |
960ea056 | 8347 | break; |
48461135 | 8348 | case BPF_JGE: |
a72dafaf JW |
8349 | case BPF_JGT: |
8350 | { | |
3f50f132 JF |
8351 | if (is_jmp32) { |
8352 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
8353 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
8354 | ||
8355 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
8356 | false_umax); | |
8357 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
8358 | true_umin); | |
8359 | } else { | |
8360 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
8361 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
8362 | ||
8363 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
8364 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
8365 | } | |
b03c9f9f | 8366 | break; |
a72dafaf | 8367 | } |
48461135 | 8368 | case BPF_JSGE: |
a72dafaf JW |
8369 | case BPF_JSGT: |
8370 | { | |
3f50f132 JF |
8371 | if (is_jmp32) { |
8372 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
8373 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 8374 | |
3f50f132 JF |
8375 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
8376 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
8377 | } else { | |
8378 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
8379 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
8380 | ||
8381 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
8382 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
8383 | } | |
48461135 | 8384 | break; |
a72dafaf | 8385 | } |
b4e432f1 | 8386 | case BPF_JLE: |
a72dafaf JW |
8387 | case BPF_JLT: |
8388 | { | |
3f50f132 JF |
8389 | if (is_jmp32) { |
8390 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
8391 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
8392 | ||
8393 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
8394 | false_umin); | |
8395 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
8396 | true_umax); | |
8397 | } else { | |
8398 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
8399 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
8400 | ||
8401 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
8402 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
8403 | } | |
b4e432f1 | 8404 | break; |
a72dafaf | 8405 | } |
b4e432f1 | 8406 | case BPF_JSLE: |
a72dafaf JW |
8407 | case BPF_JSLT: |
8408 | { | |
3f50f132 JF |
8409 | if (is_jmp32) { |
8410 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
8411 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 8412 | |
3f50f132 JF |
8413 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
8414 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
8415 | } else { | |
8416 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
8417 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
8418 | ||
8419 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
8420 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
8421 | } | |
b4e432f1 | 8422 | break; |
a72dafaf | 8423 | } |
48461135 | 8424 | default: |
0fc31b10 | 8425 | return; |
48461135 JB |
8426 | } |
8427 | ||
3f50f132 JF |
8428 | if (is_jmp32) { |
8429 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
8430 | tnum_subreg(false_32off)); | |
8431 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
8432 | tnum_subreg(true_32off)); | |
8433 | __reg_combine_32_into_64(false_reg); | |
8434 | __reg_combine_32_into_64(true_reg); | |
8435 | } else { | |
8436 | false_reg->var_off = false_64off; | |
8437 | true_reg->var_off = true_64off; | |
8438 | __reg_combine_64_into_32(false_reg); | |
8439 | __reg_combine_64_into_32(true_reg); | |
8440 | } | |
48461135 JB |
8441 | } |
8442 | ||
f1174f77 EC |
8443 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
8444 | * the variable reg. | |
48461135 JB |
8445 | */ |
8446 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
8447 | struct bpf_reg_state *false_reg, |
8448 | u64 val, u32 val32, | |
092ed096 | 8449 | u8 opcode, bool is_jmp32) |
48461135 | 8450 | { |
6d94e741 | 8451 | opcode = flip_opcode(opcode); |
0fc31b10 JH |
8452 | /* This uses zero as "not present in table"; luckily the zero opcode, |
8453 | * BPF_JA, can't get here. | |
b03c9f9f | 8454 | */ |
0fc31b10 | 8455 | if (opcode) |
3f50f132 | 8456 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
8457 | } |
8458 | ||
8459 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
8460 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
8461 | struct bpf_reg_state *dst_reg) | |
8462 | { | |
b03c9f9f EC |
8463 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
8464 | dst_reg->umin_value); | |
8465 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
8466 | dst_reg->umax_value); | |
8467 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
8468 | dst_reg->smin_value); | |
8469 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
8470 | dst_reg->smax_value); | |
f1174f77 EC |
8471 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
8472 | dst_reg->var_off); | |
b03c9f9f EC |
8473 | /* We might have learned new bounds from the var_off. */ |
8474 | __update_reg_bounds(src_reg); | |
8475 | __update_reg_bounds(dst_reg); | |
8476 | /* We might have learned something about the sign bit. */ | |
8477 | __reg_deduce_bounds(src_reg); | |
8478 | __reg_deduce_bounds(dst_reg); | |
8479 | /* We might have learned some bits from the bounds. */ | |
8480 | __reg_bound_offset(src_reg); | |
8481 | __reg_bound_offset(dst_reg); | |
8482 | /* Intersecting with the old var_off might have improved our bounds | |
8483 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
8484 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
8485 | */ | |
8486 | __update_reg_bounds(src_reg); | |
8487 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
8488 | } |
8489 | ||
8490 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
8491 | struct bpf_reg_state *true_dst, | |
8492 | struct bpf_reg_state *false_src, | |
8493 | struct bpf_reg_state *false_dst, | |
8494 | u8 opcode) | |
8495 | { | |
8496 | switch (opcode) { | |
8497 | case BPF_JEQ: | |
8498 | __reg_combine_min_max(true_src, true_dst); | |
8499 | break; | |
8500 | case BPF_JNE: | |
8501 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 8502 | break; |
4cabc5b1 | 8503 | } |
48461135 JB |
8504 | } |
8505 | ||
fd978bf7 JS |
8506 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
8507 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 8508 | bool is_null) |
57a09bf0 | 8509 | { |
93c230e3 MKL |
8510 | if (reg_type_may_be_null(reg->type) && reg->id == id && |
8511 | !WARN_ON_ONCE(!reg->id)) { | |
f1174f77 EC |
8512 | /* Old offset (both fixed and variable parts) should |
8513 | * have been known-zero, because we don't allow pointer | |
8514 | * arithmetic on pointers that might be NULL. | |
8515 | */ | |
b03c9f9f EC |
8516 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
8517 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 8518 | reg->off)) { |
b03c9f9f EC |
8519 | __mark_reg_known_zero(reg); |
8520 | reg->off = 0; | |
f1174f77 EC |
8521 | } |
8522 | if (is_null) { | |
8523 | reg->type = SCALAR_VALUE; | |
1b986589 MKL |
8524 | /* We don't need id and ref_obj_id from this point |
8525 | * onwards anymore, thus we should better reset it, | |
8526 | * so that state pruning has chances to take effect. | |
8527 | */ | |
8528 | reg->id = 0; | |
8529 | reg->ref_obj_id = 0; | |
4ddb7416 DB |
8530 | |
8531 | return; | |
8532 | } | |
8533 | ||
8534 | mark_ptr_not_null_reg(reg); | |
8535 | ||
8536 | if (!reg_may_point_to_spin_lock(reg)) { | |
1b986589 MKL |
8537 | /* For not-NULL ptr, reg->ref_obj_id will be reset |
8538 | * in release_reg_references(). | |
8539 | * | |
8540 | * reg->id is still used by spin_lock ptr. Other | |
8541 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
8542 | */ |
8543 | reg->id = 0; | |
56f668df | 8544 | } |
57a09bf0 TG |
8545 | } |
8546 | } | |
8547 | ||
c6a9efa1 PC |
8548 | static void __mark_ptr_or_null_regs(struct bpf_func_state *state, u32 id, |
8549 | bool is_null) | |
8550 | { | |
8551 | struct bpf_reg_state *reg; | |
8552 | int i; | |
8553 | ||
8554 | for (i = 0; i < MAX_BPF_REG; i++) | |
8555 | mark_ptr_or_null_reg(state, &state->regs[i], id, is_null); | |
8556 | ||
8557 | bpf_for_each_spilled_reg(i, state, reg) { | |
8558 | if (!reg) | |
8559 | continue; | |
8560 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
8561 | } | |
8562 | } | |
8563 | ||
57a09bf0 TG |
8564 | /* The logic is similar to find_good_pkt_pointers(), both could eventually |
8565 | * be folded together at some point. | |
8566 | */ | |
840b9615 JS |
8567 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
8568 | bool is_null) | |
57a09bf0 | 8569 | { |
f4d7e40a | 8570 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
c6a9efa1 | 8571 | struct bpf_reg_state *regs = state->regs; |
1b986589 | 8572 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 8573 | u32 id = regs[regno].id; |
c6a9efa1 | 8574 | int i; |
57a09bf0 | 8575 | |
1b986589 MKL |
8576 | if (ref_obj_id && ref_obj_id == id && is_null) |
8577 | /* regs[regno] is in the " == NULL" branch. | |
8578 | * No one could have freed the reference state before | |
8579 | * doing the NULL check. | |
8580 | */ | |
8581 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 8582 | |
c6a9efa1 PC |
8583 | for (i = 0; i <= vstate->curframe; i++) |
8584 | __mark_ptr_or_null_regs(vstate->frame[i], id, is_null); | |
57a09bf0 TG |
8585 | } |
8586 | ||
5beca081 DB |
8587 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
8588 | struct bpf_reg_state *dst_reg, | |
8589 | struct bpf_reg_state *src_reg, | |
8590 | struct bpf_verifier_state *this_branch, | |
8591 | struct bpf_verifier_state *other_branch) | |
8592 | { | |
8593 | if (BPF_SRC(insn->code) != BPF_X) | |
8594 | return false; | |
8595 | ||
092ed096 JW |
8596 | /* Pointers are always 64-bit. */ |
8597 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
8598 | return false; | |
8599 | ||
5beca081 DB |
8600 | switch (BPF_OP(insn->code)) { |
8601 | case BPF_JGT: | |
8602 | if ((dst_reg->type == PTR_TO_PACKET && | |
8603 | src_reg->type == PTR_TO_PACKET_END) || | |
8604 | (dst_reg->type == PTR_TO_PACKET_META && | |
8605 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8606 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
8607 | find_good_pkt_pointers(this_branch, dst_reg, | |
8608 | dst_reg->type, false); | |
6d94e741 | 8609 | mark_pkt_end(other_branch, insn->dst_reg, true); |
5beca081 DB |
8610 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8611 | src_reg->type == PTR_TO_PACKET) || | |
8612 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8613 | src_reg->type == PTR_TO_PACKET_META)) { | |
8614 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
8615 | find_good_pkt_pointers(other_branch, src_reg, | |
8616 | src_reg->type, true); | |
6d94e741 | 8617 | mark_pkt_end(this_branch, insn->src_reg, false); |
5beca081 DB |
8618 | } else { |
8619 | return false; | |
8620 | } | |
8621 | break; | |
8622 | case BPF_JLT: | |
8623 | if ((dst_reg->type == PTR_TO_PACKET && | |
8624 | src_reg->type == PTR_TO_PACKET_END) || | |
8625 | (dst_reg->type == PTR_TO_PACKET_META && | |
8626 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8627 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
8628 | find_good_pkt_pointers(other_branch, dst_reg, | |
8629 | dst_reg->type, true); | |
6d94e741 | 8630 | mark_pkt_end(this_branch, insn->dst_reg, false); |
5beca081 DB |
8631 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8632 | src_reg->type == PTR_TO_PACKET) || | |
8633 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8634 | src_reg->type == PTR_TO_PACKET_META)) { | |
8635 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
8636 | find_good_pkt_pointers(this_branch, src_reg, | |
8637 | src_reg->type, false); | |
6d94e741 | 8638 | mark_pkt_end(other_branch, insn->src_reg, true); |
5beca081 DB |
8639 | } else { |
8640 | return false; | |
8641 | } | |
8642 | break; | |
8643 | case BPF_JGE: | |
8644 | if ((dst_reg->type == PTR_TO_PACKET && | |
8645 | src_reg->type == PTR_TO_PACKET_END) || | |
8646 | (dst_reg->type == PTR_TO_PACKET_META && | |
8647 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8648 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
8649 | find_good_pkt_pointers(this_branch, dst_reg, | |
8650 | dst_reg->type, true); | |
6d94e741 | 8651 | mark_pkt_end(other_branch, insn->dst_reg, false); |
5beca081 DB |
8652 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8653 | src_reg->type == PTR_TO_PACKET) || | |
8654 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8655 | src_reg->type == PTR_TO_PACKET_META)) { | |
8656 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
8657 | find_good_pkt_pointers(other_branch, src_reg, | |
8658 | src_reg->type, false); | |
6d94e741 | 8659 | mark_pkt_end(this_branch, insn->src_reg, true); |
5beca081 DB |
8660 | } else { |
8661 | return false; | |
8662 | } | |
8663 | break; | |
8664 | case BPF_JLE: | |
8665 | if ((dst_reg->type == PTR_TO_PACKET && | |
8666 | src_reg->type == PTR_TO_PACKET_END) || | |
8667 | (dst_reg->type == PTR_TO_PACKET_META && | |
8668 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8669 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
8670 | find_good_pkt_pointers(other_branch, dst_reg, | |
8671 | dst_reg->type, false); | |
6d94e741 | 8672 | mark_pkt_end(this_branch, insn->dst_reg, true); |
5beca081 DB |
8673 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8674 | src_reg->type == PTR_TO_PACKET) || | |
8675 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8676 | src_reg->type == PTR_TO_PACKET_META)) { | |
8677 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
8678 | find_good_pkt_pointers(this_branch, src_reg, | |
8679 | src_reg->type, true); | |
6d94e741 | 8680 | mark_pkt_end(other_branch, insn->src_reg, false); |
5beca081 DB |
8681 | } else { |
8682 | return false; | |
8683 | } | |
8684 | break; | |
8685 | default: | |
8686 | return false; | |
8687 | } | |
8688 | ||
8689 | return true; | |
8690 | } | |
8691 | ||
75748837 AS |
8692 | static void find_equal_scalars(struct bpf_verifier_state *vstate, |
8693 | struct bpf_reg_state *known_reg) | |
8694 | { | |
8695 | struct bpf_func_state *state; | |
8696 | struct bpf_reg_state *reg; | |
8697 | int i, j; | |
8698 | ||
8699 | for (i = 0; i <= vstate->curframe; i++) { | |
8700 | state = vstate->frame[i]; | |
8701 | for (j = 0; j < MAX_BPF_REG; j++) { | |
8702 | reg = &state->regs[j]; | |
8703 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
8704 | *reg = *known_reg; | |
8705 | } | |
8706 | ||
8707 | bpf_for_each_spilled_reg(j, state, reg) { | |
8708 | if (!reg) | |
8709 | continue; | |
8710 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
8711 | *reg = *known_reg; | |
8712 | } | |
8713 | } | |
8714 | } | |
8715 | ||
58e2af8b | 8716 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
8717 | struct bpf_insn *insn, int *insn_idx) |
8718 | { | |
f4d7e40a AS |
8719 | struct bpf_verifier_state *this_branch = env->cur_state; |
8720 | struct bpf_verifier_state *other_branch; | |
8721 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 8722 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
17a52670 | 8723 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 8724 | bool is_jmp32; |
fb8d251e | 8725 | int pred = -1; |
17a52670 AS |
8726 | int err; |
8727 | ||
092ed096 JW |
8728 | /* Only conditional jumps are expected to reach here. */ |
8729 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
8730 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
8731 | return -EINVAL; |
8732 | } | |
8733 | ||
8734 | if (BPF_SRC(insn->code) == BPF_X) { | |
8735 | if (insn->imm != 0) { | |
092ed096 | 8736 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
8737 | return -EINVAL; |
8738 | } | |
8739 | ||
8740 | /* check src1 operand */ | |
dc503a8a | 8741 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
8742 | if (err) |
8743 | return err; | |
1be7f75d AS |
8744 | |
8745 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 8746 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
8747 | insn->src_reg); |
8748 | return -EACCES; | |
8749 | } | |
fb8d251e | 8750 | src_reg = ®s[insn->src_reg]; |
17a52670 AS |
8751 | } else { |
8752 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 8753 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
8754 | return -EINVAL; |
8755 | } | |
8756 | } | |
8757 | ||
8758 | /* check src2 operand */ | |
dc503a8a | 8759 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
8760 | if (err) |
8761 | return err; | |
8762 | ||
1a0dc1ac | 8763 | dst_reg = ®s[insn->dst_reg]; |
092ed096 | 8764 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 8765 | |
3f50f132 JF |
8766 | if (BPF_SRC(insn->code) == BPF_K) { |
8767 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
8768 | } else if (src_reg->type == SCALAR_VALUE && | |
8769 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
8770 | pred = is_branch_taken(dst_reg, | |
8771 | tnum_subreg(src_reg->var_off).value, | |
8772 | opcode, | |
8773 | is_jmp32); | |
8774 | } else if (src_reg->type == SCALAR_VALUE && | |
8775 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
8776 | pred = is_branch_taken(dst_reg, | |
8777 | src_reg->var_off.value, | |
8778 | opcode, | |
8779 | is_jmp32); | |
6d94e741 AS |
8780 | } else if (reg_is_pkt_pointer_any(dst_reg) && |
8781 | reg_is_pkt_pointer_any(src_reg) && | |
8782 | !is_jmp32) { | |
8783 | pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode); | |
3f50f132 JF |
8784 | } |
8785 | ||
b5dc0163 | 8786 | if (pred >= 0) { |
cac616db JF |
8787 | /* If we get here with a dst_reg pointer type it is because |
8788 | * above is_branch_taken() special cased the 0 comparison. | |
8789 | */ | |
8790 | if (!__is_pointer_value(false, dst_reg)) | |
8791 | err = mark_chain_precision(env, insn->dst_reg); | |
6d94e741 AS |
8792 | if (BPF_SRC(insn->code) == BPF_X && !err && |
8793 | !__is_pointer_value(false, src_reg)) | |
b5dc0163 AS |
8794 | err = mark_chain_precision(env, insn->src_reg); |
8795 | if (err) | |
8796 | return err; | |
8797 | } | |
9183671a | 8798 | |
fb8d251e | 8799 | if (pred == 1) { |
9183671a DB |
8800 | /* Only follow the goto, ignore fall-through. If needed, push |
8801 | * the fall-through branch for simulation under speculative | |
8802 | * execution. | |
8803 | */ | |
8804 | if (!env->bypass_spec_v1 && | |
8805 | !sanitize_speculative_path(env, insn, *insn_idx + 1, | |
8806 | *insn_idx)) | |
8807 | return -EFAULT; | |
fb8d251e AS |
8808 | *insn_idx += insn->off; |
8809 | return 0; | |
8810 | } else if (pred == 0) { | |
9183671a DB |
8811 | /* Only follow the fall-through branch, since that's where the |
8812 | * program will go. If needed, push the goto branch for | |
8813 | * simulation under speculative execution. | |
fb8d251e | 8814 | */ |
9183671a DB |
8815 | if (!env->bypass_spec_v1 && |
8816 | !sanitize_speculative_path(env, insn, | |
8817 | *insn_idx + insn->off + 1, | |
8818 | *insn_idx)) | |
8819 | return -EFAULT; | |
fb8d251e | 8820 | return 0; |
17a52670 AS |
8821 | } |
8822 | ||
979d63d5 DB |
8823 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
8824 | false); | |
17a52670 AS |
8825 | if (!other_branch) |
8826 | return -EFAULT; | |
f4d7e40a | 8827 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 8828 | |
48461135 JB |
8829 | /* detect if we are comparing against a constant value so we can adjust |
8830 | * our min/max values for our dst register. | |
f1174f77 EC |
8831 | * this is only legit if both are scalars (or pointers to the same |
8832 | * object, I suppose, but we don't support that right now), because | |
8833 | * otherwise the different base pointers mean the offsets aren't | |
8834 | * comparable. | |
48461135 JB |
8835 | */ |
8836 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 8837 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 8838 | |
f1174f77 | 8839 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
8840 | src_reg->type == SCALAR_VALUE) { |
8841 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
8842 | (is_jmp32 && |
8843 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 8844 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 8845 | dst_reg, |
3f50f132 JF |
8846 | src_reg->var_off.value, |
8847 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
8848 | opcode, is_jmp32); |
8849 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
8850 | (is_jmp32 && |
8851 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 8852 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 8853 | src_reg, |
3f50f132 JF |
8854 | dst_reg->var_off.value, |
8855 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
8856 | opcode, is_jmp32); |
8857 | else if (!is_jmp32 && | |
8858 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 8859 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
8860 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
8861 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 8862 | src_reg, dst_reg, opcode); |
e688c3db AS |
8863 | if (src_reg->id && |
8864 | !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) { | |
75748837 AS |
8865 | find_equal_scalars(this_branch, src_reg); |
8866 | find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]); | |
8867 | } | |
8868 | ||
f1174f77 EC |
8869 | } |
8870 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 8871 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
8872 | dst_reg, insn->imm, (u32)insn->imm, |
8873 | opcode, is_jmp32); | |
48461135 JB |
8874 | } |
8875 | ||
e688c3db AS |
8876 | if (dst_reg->type == SCALAR_VALUE && dst_reg->id && |
8877 | !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) { | |
75748837 AS |
8878 | find_equal_scalars(this_branch, dst_reg); |
8879 | find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]); | |
8880 | } | |
8881 | ||
092ed096 JW |
8882 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
8883 | * NOTE: these optimizations below are related with pointer comparison | |
8884 | * which will never be JMP32. | |
8885 | */ | |
8886 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 8887 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
840b9615 JS |
8888 | reg_type_may_be_null(dst_reg->type)) { |
8889 | /* Mark all identical registers in each branch as either | |
57a09bf0 TG |
8890 | * safe or unknown depending R == 0 or R != 0 conditional. |
8891 | */ | |
840b9615 JS |
8892 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
8893 | opcode == BPF_JNE); | |
8894 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
8895 | opcode == BPF_JEQ); | |
5beca081 DB |
8896 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
8897 | this_branch, other_branch) && | |
8898 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
8899 | verbose(env, "R%d pointer comparison prohibited\n", |
8900 | insn->dst_reg); | |
1be7f75d | 8901 | return -EACCES; |
17a52670 | 8902 | } |
06ee7115 | 8903 | if (env->log.level & BPF_LOG_LEVEL) |
f4d7e40a | 8904 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
8905 | return 0; |
8906 | } | |
8907 | ||
17a52670 | 8908 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 8909 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 8910 | { |
d8eca5bb | 8911 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 8912 | struct bpf_reg_state *regs = cur_regs(env); |
4976b718 | 8913 | struct bpf_reg_state *dst_reg; |
d8eca5bb | 8914 | struct bpf_map *map; |
17a52670 AS |
8915 | int err; |
8916 | ||
8917 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 8918 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
8919 | return -EINVAL; |
8920 | } | |
8921 | if (insn->off != 0) { | |
61bd5218 | 8922 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
8923 | return -EINVAL; |
8924 | } | |
8925 | ||
dc503a8a | 8926 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
8927 | if (err) |
8928 | return err; | |
8929 | ||
4976b718 | 8930 | dst_reg = ®s[insn->dst_reg]; |
6b173873 | 8931 | if (insn->src_reg == 0) { |
6b173873 JK |
8932 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
8933 | ||
4976b718 | 8934 | dst_reg->type = SCALAR_VALUE; |
b03c9f9f | 8935 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 8936 | return 0; |
6b173873 | 8937 | } |
17a52670 | 8938 | |
4976b718 HL |
8939 | if (insn->src_reg == BPF_PSEUDO_BTF_ID) { |
8940 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
8941 | ||
8942 | dst_reg->type = aux->btf_var.reg_type; | |
8943 | switch (dst_reg->type) { | |
8944 | case PTR_TO_MEM: | |
8945 | dst_reg->mem_size = aux->btf_var.mem_size; | |
8946 | break; | |
8947 | case PTR_TO_BTF_ID: | |
eaa6bcb7 | 8948 | case PTR_TO_PERCPU_BTF_ID: |
22dc4a0f | 8949 | dst_reg->btf = aux->btf_var.btf; |
4976b718 HL |
8950 | dst_reg->btf_id = aux->btf_var.btf_id; |
8951 | break; | |
8952 | default: | |
8953 | verbose(env, "bpf verifier is misconfigured\n"); | |
8954 | return -EFAULT; | |
8955 | } | |
8956 | return 0; | |
8957 | } | |
8958 | ||
69c087ba YS |
8959 | if (insn->src_reg == BPF_PSEUDO_FUNC) { |
8960 | struct bpf_prog_aux *aux = env->prog->aux; | |
8961 | u32 subprogno = insn[1].imm; | |
8962 | ||
8963 | if (!aux->func_info) { | |
8964 | verbose(env, "missing btf func_info\n"); | |
8965 | return -EINVAL; | |
8966 | } | |
8967 | if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) { | |
8968 | verbose(env, "callback function not static\n"); | |
8969 | return -EINVAL; | |
8970 | } | |
8971 | ||
8972 | dst_reg->type = PTR_TO_FUNC; | |
8973 | dst_reg->subprogno = subprogno; | |
8974 | return 0; | |
8975 | } | |
8976 | ||
d8eca5bb DB |
8977 | map = env->used_maps[aux->map_index]; |
8978 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
4976b718 | 8979 | dst_reg->map_ptr = map; |
d8eca5bb | 8980 | |
387544bf AS |
8981 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE || |
8982 | insn->src_reg == BPF_PSEUDO_MAP_IDX_VALUE) { | |
4976b718 HL |
8983 | dst_reg->type = PTR_TO_MAP_VALUE; |
8984 | dst_reg->off = aux->map_off; | |
d8eca5bb | 8985 | if (map_value_has_spin_lock(map)) |
4976b718 | 8986 | dst_reg->id = ++env->id_gen; |
387544bf AS |
8987 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD || |
8988 | insn->src_reg == BPF_PSEUDO_MAP_IDX) { | |
4976b718 | 8989 | dst_reg->type = CONST_PTR_TO_MAP; |
d8eca5bb DB |
8990 | } else { |
8991 | verbose(env, "bpf verifier is misconfigured\n"); | |
8992 | return -EINVAL; | |
8993 | } | |
17a52670 | 8994 | |
17a52670 AS |
8995 | return 0; |
8996 | } | |
8997 | ||
96be4325 DB |
8998 | static bool may_access_skb(enum bpf_prog_type type) |
8999 | { | |
9000 | switch (type) { | |
9001 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
9002 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 9003 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
9004 | return true; |
9005 | default: | |
9006 | return false; | |
9007 | } | |
9008 | } | |
9009 | ||
ddd872bc AS |
9010 | /* verify safety of LD_ABS|LD_IND instructions: |
9011 | * - they can only appear in the programs where ctx == skb | |
9012 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
9013 | * preserve R6-R9, and store return value into R0 | |
9014 | * | |
9015 | * Implicit input: | |
9016 | * ctx == skb == R6 == CTX | |
9017 | * | |
9018 | * Explicit input: | |
9019 | * SRC == any register | |
9020 | * IMM == 32-bit immediate | |
9021 | * | |
9022 | * Output: | |
9023 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
9024 | */ | |
58e2af8b | 9025 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 9026 | { |
638f5b90 | 9027 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 9028 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 9029 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
9030 | int i, err; |
9031 | ||
7e40781c | 9032 | if (!may_access_skb(resolve_prog_type(env->prog))) { |
61bd5218 | 9033 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
9034 | return -EINVAL; |
9035 | } | |
9036 | ||
e0cea7ce DB |
9037 | if (!env->ops->gen_ld_abs) { |
9038 | verbose(env, "bpf verifier is misconfigured\n"); | |
9039 | return -EINVAL; | |
9040 | } | |
9041 | ||
ddd872bc | 9042 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 9043 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 9044 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 9045 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
9046 | return -EINVAL; |
9047 | } | |
9048 | ||
9049 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 9050 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
9051 | if (err) |
9052 | return err; | |
9053 | ||
fd978bf7 JS |
9054 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
9055 | * gen_ld_abs() may terminate the program at runtime, leading to | |
9056 | * reference leak. | |
9057 | */ | |
9058 | err = check_reference_leak(env); | |
9059 | if (err) { | |
9060 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
9061 | return err; | |
9062 | } | |
9063 | ||
d83525ca AS |
9064 | if (env->cur_state->active_spin_lock) { |
9065 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); | |
9066 | return -EINVAL; | |
9067 | } | |
9068 | ||
6d4f151a | 9069 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
9070 | verbose(env, |
9071 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
9072 | return -EINVAL; |
9073 | } | |
9074 | ||
9075 | if (mode == BPF_IND) { | |
9076 | /* check explicit source operand */ | |
dc503a8a | 9077 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
9078 | if (err) |
9079 | return err; | |
9080 | } | |
9081 | ||
6d4f151a DB |
9082 | err = check_ctx_reg(env, ®s[ctx_reg], ctx_reg); |
9083 | if (err < 0) | |
9084 | return err; | |
9085 | ||
ddd872bc | 9086 | /* reset caller saved regs to unreadable */ |
dc503a8a | 9087 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 9088 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
9089 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
9090 | } | |
ddd872bc AS |
9091 | |
9092 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
9093 | * the value fetched from the packet. |
9094 | * Already marked as written above. | |
ddd872bc | 9095 | */ |
61bd5218 | 9096 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
9097 | /* ld_abs load up to 32-bit skb data. */ |
9098 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
9099 | return 0; |
9100 | } | |
9101 | ||
390ee7e2 AS |
9102 | static int check_return_code(struct bpf_verifier_env *env) |
9103 | { | |
5cf1e914 | 9104 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 9105 | const struct bpf_prog *prog = env->prog; |
390ee7e2 AS |
9106 | struct bpf_reg_state *reg; |
9107 | struct tnum range = tnum_range(0, 1); | |
7e40781c | 9108 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
27ae7997 | 9109 | int err; |
f782e2c3 | 9110 | const bool is_subprog = env->cur_state->frame[0]->subprogno; |
27ae7997 | 9111 | |
9e4e01df | 9112 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
f782e2c3 DB |
9113 | if (!is_subprog && |
9114 | (prog_type == BPF_PROG_TYPE_STRUCT_OPS || | |
7e40781c | 9115 | prog_type == BPF_PROG_TYPE_LSM) && |
27ae7997 MKL |
9116 | !prog->aux->attach_func_proto->type) |
9117 | return 0; | |
9118 | ||
8fb33b60 | 9119 | /* eBPF calling convention is such that R0 is used |
27ae7997 MKL |
9120 | * to return the value from eBPF program. |
9121 | * Make sure that it's readable at this time | |
9122 | * of bpf_exit, which means that program wrote | |
9123 | * something into it earlier | |
9124 | */ | |
9125 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
9126 | if (err) | |
9127 | return err; | |
9128 | ||
9129 | if (is_pointer_value(env, BPF_REG_0)) { | |
9130 | verbose(env, "R0 leaks addr as return value\n"); | |
9131 | return -EACCES; | |
9132 | } | |
390ee7e2 | 9133 | |
f782e2c3 DB |
9134 | reg = cur_regs(env) + BPF_REG_0; |
9135 | if (is_subprog) { | |
9136 | if (reg->type != SCALAR_VALUE) { | |
9137 | verbose(env, "At subprogram exit the register R0 is not a scalar value (%s)\n", | |
9138 | reg_type_str[reg->type]); | |
9139 | return -EINVAL; | |
9140 | } | |
9141 | return 0; | |
9142 | } | |
9143 | ||
7e40781c | 9144 | switch (prog_type) { |
983695fa DB |
9145 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
9146 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 DB |
9147 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
9148 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || | |
9149 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
9150 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || | |
9151 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME) | |
983695fa | 9152 | range = tnum_range(1, 1); |
77241217 SF |
9153 | if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND || |
9154 | env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND) | |
9155 | range = tnum_range(0, 3); | |
ed4ed404 | 9156 | break; |
390ee7e2 | 9157 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 9158 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
9159 | range = tnum_range(0, 3); | |
9160 | enforce_attach_type_range = tnum_range(2, 3); | |
9161 | } | |
ed4ed404 | 9162 | break; |
390ee7e2 AS |
9163 | case BPF_PROG_TYPE_CGROUP_SOCK: |
9164 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 9165 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 9166 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 9167 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 9168 | break; |
15ab09bd AS |
9169 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
9170 | if (!env->prog->aux->attach_btf_id) | |
9171 | return 0; | |
9172 | range = tnum_const(0); | |
9173 | break; | |
15d83c4d | 9174 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
9175 | switch (env->prog->expected_attach_type) { |
9176 | case BPF_TRACE_FENTRY: | |
9177 | case BPF_TRACE_FEXIT: | |
9178 | range = tnum_const(0); | |
9179 | break; | |
9180 | case BPF_TRACE_RAW_TP: | |
9181 | case BPF_MODIFY_RETURN: | |
15d83c4d | 9182 | return 0; |
2ec0616e DB |
9183 | case BPF_TRACE_ITER: |
9184 | break; | |
e92888c7 YS |
9185 | default: |
9186 | return -ENOTSUPP; | |
9187 | } | |
15d83c4d | 9188 | break; |
e9ddbb77 JS |
9189 | case BPF_PROG_TYPE_SK_LOOKUP: |
9190 | range = tnum_range(SK_DROP, SK_PASS); | |
9191 | break; | |
e92888c7 YS |
9192 | case BPF_PROG_TYPE_EXT: |
9193 | /* freplace program can return anything as its return value | |
9194 | * depends on the to-be-replaced kernel func or bpf program. | |
9195 | */ | |
390ee7e2 AS |
9196 | default: |
9197 | return 0; | |
9198 | } | |
9199 | ||
390ee7e2 | 9200 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 9201 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
9202 | reg_type_str[reg->type]); |
9203 | return -EINVAL; | |
9204 | } | |
9205 | ||
9206 | if (!tnum_in(range, reg->var_off)) { | |
bc2591d6 | 9207 | verbose_invalid_scalar(env, reg, &range, "program exit", "R0"); |
390ee7e2 AS |
9208 | return -EINVAL; |
9209 | } | |
5cf1e914 | 9210 | |
9211 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
9212 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
9213 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
9214 | return 0; |
9215 | } | |
9216 | ||
475fb78f AS |
9217 | /* non-recursive DFS pseudo code |
9218 | * 1 procedure DFS-iterative(G,v): | |
9219 | * 2 label v as discovered | |
9220 | * 3 let S be a stack | |
9221 | * 4 S.push(v) | |
9222 | * 5 while S is not empty | |
9223 | * 6 t <- S.pop() | |
9224 | * 7 if t is what we're looking for: | |
9225 | * 8 return t | |
9226 | * 9 for all edges e in G.adjacentEdges(t) do | |
9227 | * 10 if edge e is already labelled | |
9228 | * 11 continue with the next edge | |
9229 | * 12 w <- G.adjacentVertex(t,e) | |
9230 | * 13 if vertex w is not discovered and not explored | |
9231 | * 14 label e as tree-edge | |
9232 | * 15 label w as discovered | |
9233 | * 16 S.push(w) | |
9234 | * 17 continue at 5 | |
9235 | * 18 else if vertex w is discovered | |
9236 | * 19 label e as back-edge | |
9237 | * 20 else | |
9238 | * 21 // vertex w is explored | |
9239 | * 22 label e as forward- or cross-edge | |
9240 | * 23 label t as explored | |
9241 | * 24 S.pop() | |
9242 | * | |
9243 | * convention: | |
9244 | * 0x10 - discovered | |
9245 | * 0x11 - discovered and fall-through edge labelled | |
9246 | * 0x12 - discovered and fall-through and branch edges labelled | |
9247 | * 0x20 - explored | |
9248 | */ | |
9249 | ||
9250 | enum { | |
9251 | DISCOVERED = 0x10, | |
9252 | EXPLORED = 0x20, | |
9253 | FALLTHROUGH = 1, | |
9254 | BRANCH = 2, | |
9255 | }; | |
9256 | ||
dc2a4ebc AS |
9257 | static u32 state_htab_size(struct bpf_verifier_env *env) |
9258 | { | |
9259 | return env->prog->len; | |
9260 | } | |
9261 | ||
5d839021 AS |
9262 | static struct bpf_verifier_state_list **explored_state( |
9263 | struct bpf_verifier_env *env, | |
9264 | int idx) | |
9265 | { | |
dc2a4ebc AS |
9266 | struct bpf_verifier_state *cur = env->cur_state; |
9267 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
9268 | ||
9269 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
9270 | } |
9271 | ||
9272 | static void init_explored_state(struct bpf_verifier_env *env, int idx) | |
9273 | { | |
a8f500af | 9274 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 9275 | } |
f1bca824 | 9276 | |
59e2e27d WAF |
9277 | enum { |
9278 | DONE_EXPLORING = 0, | |
9279 | KEEP_EXPLORING = 1, | |
9280 | }; | |
9281 | ||
475fb78f AS |
9282 | /* t, w, e - match pseudo-code above: |
9283 | * t - index of current instruction | |
9284 | * w - next instruction | |
9285 | * e - edge | |
9286 | */ | |
2589726d AS |
9287 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env, |
9288 | bool loop_ok) | |
475fb78f | 9289 | { |
7df737e9 AS |
9290 | int *insn_stack = env->cfg.insn_stack; |
9291 | int *insn_state = env->cfg.insn_state; | |
9292 | ||
475fb78f | 9293 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
59e2e27d | 9294 | return DONE_EXPLORING; |
475fb78f AS |
9295 | |
9296 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
59e2e27d | 9297 | return DONE_EXPLORING; |
475fb78f AS |
9298 | |
9299 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 9300 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 9301 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
9302 | return -EINVAL; |
9303 | } | |
9304 | ||
f1bca824 AS |
9305 | if (e == BRANCH) |
9306 | /* mark branch target for state pruning */ | |
5d839021 | 9307 | init_explored_state(env, w); |
f1bca824 | 9308 | |
475fb78f AS |
9309 | if (insn_state[w] == 0) { |
9310 | /* tree-edge */ | |
9311 | insn_state[t] = DISCOVERED | e; | |
9312 | insn_state[w] = DISCOVERED; | |
7df737e9 | 9313 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 9314 | return -E2BIG; |
7df737e9 | 9315 | insn_stack[env->cfg.cur_stack++] = w; |
59e2e27d | 9316 | return KEEP_EXPLORING; |
475fb78f | 9317 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { |
2c78ee89 | 9318 | if (loop_ok && env->bpf_capable) |
59e2e27d | 9319 | return DONE_EXPLORING; |
d9762e84 MKL |
9320 | verbose_linfo(env, t, "%d: ", t); |
9321 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 9322 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
9323 | return -EINVAL; |
9324 | } else if (insn_state[w] == EXPLORED) { | |
9325 | /* forward- or cross-edge */ | |
9326 | insn_state[t] = DISCOVERED | e; | |
9327 | } else { | |
61bd5218 | 9328 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
9329 | return -EFAULT; |
9330 | } | |
59e2e27d WAF |
9331 | return DONE_EXPLORING; |
9332 | } | |
9333 | ||
efdb22de YS |
9334 | static int visit_func_call_insn(int t, int insn_cnt, |
9335 | struct bpf_insn *insns, | |
9336 | struct bpf_verifier_env *env, | |
9337 | bool visit_callee) | |
9338 | { | |
9339 | int ret; | |
9340 | ||
9341 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); | |
9342 | if (ret) | |
9343 | return ret; | |
9344 | ||
9345 | if (t + 1 < insn_cnt) | |
9346 | init_explored_state(env, t + 1); | |
9347 | if (visit_callee) { | |
9348 | init_explored_state(env, t); | |
9349 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, | |
9350 | env, false); | |
9351 | } | |
9352 | return ret; | |
9353 | } | |
9354 | ||
59e2e27d WAF |
9355 | /* Visits the instruction at index t and returns one of the following: |
9356 | * < 0 - an error occurred | |
9357 | * DONE_EXPLORING - the instruction was fully explored | |
9358 | * KEEP_EXPLORING - there is still work to be done before it is fully explored | |
9359 | */ | |
9360 | static int visit_insn(int t, int insn_cnt, struct bpf_verifier_env *env) | |
9361 | { | |
9362 | struct bpf_insn *insns = env->prog->insnsi; | |
9363 | int ret; | |
9364 | ||
69c087ba YS |
9365 | if (bpf_pseudo_func(insns + t)) |
9366 | return visit_func_call_insn(t, insn_cnt, insns, env, true); | |
9367 | ||
59e2e27d WAF |
9368 | /* All non-branch instructions have a single fall-through edge. */ |
9369 | if (BPF_CLASS(insns[t].code) != BPF_JMP && | |
9370 | BPF_CLASS(insns[t].code) != BPF_JMP32) | |
9371 | return push_insn(t, t + 1, FALLTHROUGH, env, false); | |
9372 | ||
9373 | switch (BPF_OP(insns[t].code)) { | |
9374 | case BPF_EXIT: | |
9375 | return DONE_EXPLORING; | |
9376 | ||
9377 | case BPF_CALL: | |
efdb22de YS |
9378 | return visit_func_call_insn(t, insn_cnt, insns, env, |
9379 | insns[t].src_reg == BPF_PSEUDO_CALL); | |
59e2e27d WAF |
9380 | |
9381 | case BPF_JA: | |
9382 | if (BPF_SRC(insns[t].code) != BPF_K) | |
9383 | return -EINVAL; | |
9384 | ||
9385 | /* unconditional jump with single edge */ | |
9386 | ret = push_insn(t, t + insns[t].off + 1, FALLTHROUGH, env, | |
9387 | true); | |
9388 | if (ret) | |
9389 | return ret; | |
9390 | ||
9391 | /* unconditional jmp is not a good pruning point, | |
9392 | * but it's marked, since backtracking needs | |
9393 | * to record jmp history in is_state_visited(). | |
9394 | */ | |
9395 | init_explored_state(env, t + insns[t].off + 1); | |
9396 | /* tell verifier to check for equivalent states | |
9397 | * after every call and jump | |
9398 | */ | |
9399 | if (t + 1 < insn_cnt) | |
9400 | init_explored_state(env, t + 1); | |
9401 | ||
9402 | return ret; | |
9403 | ||
9404 | default: | |
9405 | /* conditional jump with two edges */ | |
9406 | init_explored_state(env, t); | |
9407 | ret = push_insn(t, t + 1, FALLTHROUGH, env, true); | |
9408 | if (ret) | |
9409 | return ret; | |
9410 | ||
9411 | return push_insn(t, t + insns[t].off + 1, BRANCH, env, true); | |
9412 | } | |
475fb78f AS |
9413 | } |
9414 | ||
9415 | /* non-recursive depth-first-search to detect loops in BPF program | |
9416 | * loop == back-edge in directed graph | |
9417 | */ | |
58e2af8b | 9418 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f | 9419 | { |
475fb78f | 9420 | int insn_cnt = env->prog->len; |
7df737e9 | 9421 | int *insn_stack, *insn_state; |
475fb78f | 9422 | int ret = 0; |
59e2e27d | 9423 | int i; |
475fb78f | 9424 | |
7df737e9 | 9425 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
9426 | if (!insn_state) |
9427 | return -ENOMEM; | |
9428 | ||
7df737e9 | 9429 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 9430 | if (!insn_stack) { |
71dde681 | 9431 | kvfree(insn_state); |
475fb78f AS |
9432 | return -ENOMEM; |
9433 | } | |
9434 | ||
9435 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
9436 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 9437 | env->cfg.cur_stack = 1; |
475fb78f | 9438 | |
59e2e27d WAF |
9439 | while (env->cfg.cur_stack > 0) { |
9440 | int t = insn_stack[env->cfg.cur_stack - 1]; | |
475fb78f | 9441 | |
59e2e27d WAF |
9442 | ret = visit_insn(t, insn_cnt, env); |
9443 | switch (ret) { | |
9444 | case DONE_EXPLORING: | |
9445 | insn_state[t] = EXPLORED; | |
9446 | env->cfg.cur_stack--; | |
9447 | break; | |
9448 | case KEEP_EXPLORING: | |
9449 | break; | |
9450 | default: | |
9451 | if (ret > 0) { | |
9452 | verbose(env, "visit_insn internal bug\n"); | |
9453 | ret = -EFAULT; | |
475fb78f | 9454 | } |
475fb78f | 9455 | goto err_free; |
59e2e27d | 9456 | } |
475fb78f AS |
9457 | } |
9458 | ||
59e2e27d | 9459 | if (env->cfg.cur_stack < 0) { |
61bd5218 | 9460 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
9461 | ret = -EFAULT; |
9462 | goto err_free; | |
9463 | } | |
475fb78f | 9464 | |
475fb78f AS |
9465 | for (i = 0; i < insn_cnt; i++) { |
9466 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 9467 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
9468 | ret = -EINVAL; |
9469 | goto err_free; | |
9470 | } | |
9471 | } | |
9472 | ret = 0; /* cfg looks good */ | |
9473 | ||
9474 | err_free: | |
71dde681 AS |
9475 | kvfree(insn_state); |
9476 | kvfree(insn_stack); | |
7df737e9 | 9477 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
9478 | return ret; |
9479 | } | |
9480 | ||
09b28d76 AS |
9481 | static int check_abnormal_return(struct bpf_verifier_env *env) |
9482 | { | |
9483 | int i; | |
9484 | ||
9485 | for (i = 1; i < env->subprog_cnt; i++) { | |
9486 | if (env->subprog_info[i].has_ld_abs) { | |
9487 | verbose(env, "LD_ABS is not allowed in subprogs without BTF\n"); | |
9488 | return -EINVAL; | |
9489 | } | |
9490 | if (env->subprog_info[i].has_tail_call) { | |
9491 | verbose(env, "tail_call is not allowed in subprogs without BTF\n"); | |
9492 | return -EINVAL; | |
9493 | } | |
9494 | } | |
9495 | return 0; | |
9496 | } | |
9497 | ||
838e9690 YS |
9498 | /* The minimum supported BTF func info size */ |
9499 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
9500 | #define MAX_FUNCINFO_REC_SIZE 252 | |
9501 | ||
c454a46b MKL |
9502 | static int check_btf_func(struct bpf_verifier_env *env, |
9503 | const union bpf_attr *attr, | |
af2ac3e1 | 9504 | bpfptr_t uattr) |
838e9690 | 9505 | { |
09b28d76 | 9506 | const struct btf_type *type, *func_proto, *ret_type; |
d0b2818e | 9507 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 9508 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 9509 | struct bpf_func_info *krecord; |
8c1b6e69 | 9510 | struct bpf_func_info_aux *info_aux = NULL; |
c454a46b MKL |
9511 | struct bpf_prog *prog; |
9512 | const struct btf *btf; | |
af2ac3e1 | 9513 | bpfptr_t urecord; |
d0b2818e | 9514 | u32 prev_offset = 0; |
09b28d76 | 9515 | bool scalar_return; |
e7ed83d6 | 9516 | int ret = -ENOMEM; |
838e9690 YS |
9517 | |
9518 | nfuncs = attr->func_info_cnt; | |
09b28d76 AS |
9519 | if (!nfuncs) { |
9520 | if (check_abnormal_return(env)) | |
9521 | return -EINVAL; | |
838e9690 | 9522 | return 0; |
09b28d76 | 9523 | } |
838e9690 YS |
9524 | |
9525 | if (nfuncs != env->subprog_cnt) { | |
9526 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
9527 | return -EINVAL; | |
9528 | } | |
9529 | ||
9530 | urec_size = attr->func_info_rec_size; | |
9531 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
9532 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
9533 | urec_size % sizeof(u32)) { | |
9534 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
9535 | return -EINVAL; | |
9536 | } | |
9537 | ||
c454a46b MKL |
9538 | prog = env->prog; |
9539 | btf = prog->aux->btf; | |
838e9690 | 9540 | |
af2ac3e1 | 9541 | urecord = make_bpfptr(attr->func_info, uattr.is_kernel); |
838e9690 YS |
9542 | min_size = min_t(u32, krec_size, urec_size); |
9543 | ||
ba64e7d8 | 9544 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
9545 | if (!krecord) |
9546 | return -ENOMEM; | |
8c1b6e69 AS |
9547 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); |
9548 | if (!info_aux) | |
9549 | goto err_free; | |
ba64e7d8 | 9550 | |
838e9690 YS |
9551 | for (i = 0; i < nfuncs; i++) { |
9552 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
9553 | if (ret) { | |
9554 | if (ret == -E2BIG) { | |
9555 | verbose(env, "nonzero tailing record in func info"); | |
9556 | /* set the size kernel expects so loader can zero | |
9557 | * out the rest of the record. | |
9558 | */ | |
af2ac3e1 AS |
9559 | if (copy_to_bpfptr_offset(uattr, |
9560 | offsetof(union bpf_attr, func_info_rec_size), | |
9561 | &min_size, sizeof(min_size))) | |
838e9690 YS |
9562 | ret = -EFAULT; |
9563 | } | |
c454a46b | 9564 | goto err_free; |
838e9690 YS |
9565 | } |
9566 | ||
af2ac3e1 | 9567 | if (copy_from_bpfptr(&krecord[i], urecord, min_size)) { |
838e9690 | 9568 | ret = -EFAULT; |
c454a46b | 9569 | goto err_free; |
838e9690 YS |
9570 | } |
9571 | ||
d30d42e0 | 9572 | /* check insn_off */ |
09b28d76 | 9573 | ret = -EINVAL; |
838e9690 | 9574 | if (i == 0) { |
d30d42e0 | 9575 | if (krecord[i].insn_off) { |
838e9690 | 9576 | verbose(env, |
d30d42e0 MKL |
9577 | "nonzero insn_off %u for the first func info record", |
9578 | krecord[i].insn_off); | |
c454a46b | 9579 | goto err_free; |
838e9690 | 9580 | } |
d30d42e0 | 9581 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
9582 | verbose(env, |
9583 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 9584 | krecord[i].insn_off, prev_offset); |
c454a46b | 9585 | goto err_free; |
838e9690 YS |
9586 | } |
9587 | ||
d30d42e0 | 9588 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 | 9589 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
c454a46b | 9590 | goto err_free; |
838e9690 YS |
9591 | } |
9592 | ||
9593 | /* check type_id */ | |
ba64e7d8 | 9594 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 9595 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 9596 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 9597 | krecord[i].type_id); |
c454a46b | 9598 | goto err_free; |
838e9690 | 9599 | } |
51c39bb1 | 9600 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); |
09b28d76 AS |
9601 | |
9602 | func_proto = btf_type_by_id(btf, type->type); | |
9603 | if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto))) | |
9604 | /* btf_func_check() already verified it during BTF load */ | |
9605 | goto err_free; | |
9606 | ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL); | |
9607 | scalar_return = | |
9608 | btf_type_is_small_int(ret_type) || btf_type_is_enum(ret_type); | |
9609 | if (i && !scalar_return && env->subprog_info[i].has_ld_abs) { | |
9610 | verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n"); | |
9611 | goto err_free; | |
9612 | } | |
9613 | if (i && !scalar_return && env->subprog_info[i].has_tail_call) { | |
9614 | verbose(env, "tail_call is only allowed in functions that return 'int'.\n"); | |
9615 | goto err_free; | |
9616 | } | |
9617 | ||
d30d42e0 | 9618 | prev_offset = krecord[i].insn_off; |
af2ac3e1 | 9619 | bpfptr_add(&urecord, urec_size); |
838e9690 YS |
9620 | } |
9621 | ||
ba64e7d8 YS |
9622 | prog->aux->func_info = krecord; |
9623 | prog->aux->func_info_cnt = nfuncs; | |
8c1b6e69 | 9624 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
9625 | return 0; |
9626 | ||
c454a46b | 9627 | err_free: |
ba64e7d8 | 9628 | kvfree(krecord); |
8c1b6e69 | 9629 | kfree(info_aux); |
838e9690 YS |
9630 | return ret; |
9631 | } | |
9632 | ||
ba64e7d8 YS |
9633 | static void adjust_btf_func(struct bpf_verifier_env *env) |
9634 | { | |
8c1b6e69 | 9635 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
9636 | int i; |
9637 | ||
8c1b6e69 | 9638 | if (!aux->func_info) |
ba64e7d8 YS |
9639 | return; |
9640 | ||
9641 | for (i = 0; i < env->subprog_cnt; i++) | |
8c1b6e69 | 9642 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
9643 | } |
9644 | ||
c454a46b MKL |
9645 | #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \ |
9646 | sizeof(((struct bpf_line_info *)(0))->line_col)) | |
9647 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE | |
9648 | ||
9649 | static int check_btf_line(struct bpf_verifier_env *env, | |
9650 | const union bpf_attr *attr, | |
af2ac3e1 | 9651 | bpfptr_t uattr) |
c454a46b MKL |
9652 | { |
9653 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
9654 | struct bpf_subprog_info *sub; | |
9655 | struct bpf_line_info *linfo; | |
9656 | struct bpf_prog *prog; | |
9657 | const struct btf *btf; | |
af2ac3e1 | 9658 | bpfptr_t ulinfo; |
c454a46b MKL |
9659 | int err; |
9660 | ||
9661 | nr_linfo = attr->line_info_cnt; | |
9662 | if (!nr_linfo) | |
9663 | return 0; | |
9664 | ||
9665 | rec_size = attr->line_info_rec_size; | |
9666 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
9667 | rec_size > MAX_LINEINFO_REC_SIZE || | |
9668 | rec_size & (sizeof(u32) - 1)) | |
9669 | return -EINVAL; | |
9670 | ||
9671 | /* Need to zero it in case the userspace may | |
9672 | * pass in a smaller bpf_line_info object. | |
9673 | */ | |
9674 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
9675 | GFP_KERNEL | __GFP_NOWARN); | |
9676 | if (!linfo) | |
9677 | return -ENOMEM; | |
9678 | ||
9679 | prog = env->prog; | |
9680 | btf = prog->aux->btf; | |
9681 | ||
9682 | s = 0; | |
9683 | sub = env->subprog_info; | |
af2ac3e1 | 9684 | ulinfo = make_bpfptr(attr->line_info, uattr.is_kernel); |
c454a46b MKL |
9685 | expected_size = sizeof(struct bpf_line_info); |
9686 | ncopy = min_t(u32, expected_size, rec_size); | |
9687 | for (i = 0; i < nr_linfo; i++) { | |
9688 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
9689 | if (err) { | |
9690 | if (err == -E2BIG) { | |
9691 | verbose(env, "nonzero tailing record in line_info"); | |
af2ac3e1 AS |
9692 | if (copy_to_bpfptr_offset(uattr, |
9693 | offsetof(union bpf_attr, line_info_rec_size), | |
9694 | &expected_size, sizeof(expected_size))) | |
c454a46b MKL |
9695 | err = -EFAULT; |
9696 | } | |
9697 | goto err_free; | |
9698 | } | |
9699 | ||
af2ac3e1 | 9700 | if (copy_from_bpfptr(&linfo[i], ulinfo, ncopy)) { |
c454a46b MKL |
9701 | err = -EFAULT; |
9702 | goto err_free; | |
9703 | } | |
9704 | ||
9705 | /* | |
9706 | * Check insn_off to ensure | |
9707 | * 1) strictly increasing AND | |
9708 | * 2) bounded by prog->len | |
9709 | * | |
9710 | * The linfo[0].insn_off == 0 check logically falls into | |
9711 | * the later "missing bpf_line_info for func..." case | |
9712 | * because the first linfo[0].insn_off must be the | |
9713 | * first sub also and the first sub must have | |
9714 | * subprog_info[0].start == 0. | |
9715 | */ | |
9716 | if ((i && linfo[i].insn_off <= prev_offset) || | |
9717 | linfo[i].insn_off >= prog->len) { | |
9718 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
9719 | i, linfo[i].insn_off, prev_offset, | |
9720 | prog->len); | |
9721 | err = -EINVAL; | |
9722 | goto err_free; | |
9723 | } | |
9724 | ||
fdbaa0be MKL |
9725 | if (!prog->insnsi[linfo[i].insn_off].code) { |
9726 | verbose(env, | |
9727 | "Invalid insn code at line_info[%u].insn_off\n", | |
9728 | i); | |
9729 | err = -EINVAL; | |
9730 | goto err_free; | |
9731 | } | |
9732 | ||
23127b33 MKL |
9733 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
9734 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
9735 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
9736 | err = -EINVAL; | |
9737 | goto err_free; | |
9738 | } | |
9739 | ||
9740 | if (s != env->subprog_cnt) { | |
9741 | if (linfo[i].insn_off == sub[s].start) { | |
9742 | sub[s].linfo_idx = i; | |
9743 | s++; | |
9744 | } else if (sub[s].start < linfo[i].insn_off) { | |
9745 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
9746 | err = -EINVAL; | |
9747 | goto err_free; | |
9748 | } | |
9749 | } | |
9750 | ||
9751 | prev_offset = linfo[i].insn_off; | |
af2ac3e1 | 9752 | bpfptr_add(&ulinfo, rec_size); |
c454a46b MKL |
9753 | } |
9754 | ||
9755 | if (s != env->subprog_cnt) { | |
9756 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
9757 | env->subprog_cnt - s, s); | |
9758 | err = -EINVAL; | |
9759 | goto err_free; | |
9760 | } | |
9761 | ||
9762 | prog->aux->linfo = linfo; | |
9763 | prog->aux->nr_linfo = nr_linfo; | |
9764 | ||
9765 | return 0; | |
9766 | ||
9767 | err_free: | |
9768 | kvfree(linfo); | |
9769 | return err; | |
9770 | } | |
9771 | ||
9772 | static int check_btf_info(struct bpf_verifier_env *env, | |
9773 | const union bpf_attr *attr, | |
af2ac3e1 | 9774 | bpfptr_t uattr) |
c454a46b MKL |
9775 | { |
9776 | struct btf *btf; | |
9777 | int err; | |
9778 | ||
09b28d76 AS |
9779 | if (!attr->func_info_cnt && !attr->line_info_cnt) { |
9780 | if (check_abnormal_return(env)) | |
9781 | return -EINVAL; | |
c454a46b | 9782 | return 0; |
09b28d76 | 9783 | } |
c454a46b MKL |
9784 | |
9785 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
9786 | if (IS_ERR(btf)) | |
9787 | return PTR_ERR(btf); | |
350a5c4d AS |
9788 | if (btf_is_kernel(btf)) { |
9789 | btf_put(btf); | |
9790 | return -EACCES; | |
9791 | } | |
c454a46b MKL |
9792 | env->prog->aux->btf = btf; |
9793 | ||
9794 | err = check_btf_func(env, attr, uattr); | |
9795 | if (err) | |
9796 | return err; | |
9797 | ||
9798 | err = check_btf_line(env, attr, uattr); | |
9799 | if (err) | |
9800 | return err; | |
9801 | ||
9802 | return 0; | |
ba64e7d8 YS |
9803 | } |
9804 | ||
f1174f77 EC |
9805 | /* check %cur's range satisfies %old's */ |
9806 | static bool range_within(struct bpf_reg_state *old, | |
9807 | struct bpf_reg_state *cur) | |
9808 | { | |
b03c9f9f EC |
9809 | return old->umin_value <= cur->umin_value && |
9810 | old->umax_value >= cur->umax_value && | |
9811 | old->smin_value <= cur->smin_value && | |
fd675184 DB |
9812 | old->smax_value >= cur->smax_value && |
9813 | old->u32_min_value <= cur->u32_min_value && | |
9814 | old->u32_max_value >= cur->u32_max_value && | |
9815 | old->s32_min_value <= cur->s32_min_value && | |
9816 | old->s32_max_value >= cur->s32_max_value; | |
f1174f77 EC |
9817 | } |
9818 | ||
f1174f77 EC |
9819 | /* If in the old state two registers had the same id, then they need to have |
9820 | * the same id in the new state as well. But that id could be different from | |
9821 | * the old state, so we need to track the mapping from old to new ids. | |
9822 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
9823 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
9824 | * regs with a different old id could still have new id 9, we don't care about | |
9825 | * that. | |
9826 | * So we look through our idmap to see if this old id has been seen before. If | |
9827 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 9828 | */ |
c9e73e3d | 9829 | static bool check_ids(u32 old_id, u32 cur_id, struct bpf_id_pair *idmap) |
969bf05e | 9830 | { |
f1174f77 | 9831 | unsigned int i; |
969bf05e | 9832 | |
c9e73e3d | 9833 | for (i = 0; i < BPF_ID_MAP_SIZE; i++) { |
f1174f77 EC |
9834 | if (!idmap[i].old) { |
9835 | /* Reached an empty slot; haven't seen this id before */ | |
9836 | idmap[i].old = old_id; | |
9837 | idmap[i].cur = cur_id; | |
9838 | return true; | |
9839 | } | |
9840 | if (idmap[i].old == old_id) | |
9841 | return idmap[i].cur == cur_id; | |
9842 | } | |
9843 | /* We ran out of idmap slots, which should be impossible */ | |
9844 | WARN_ON_ONCE(1); | |
9845 | return false; | |
9846 | } | |
9847 | ||
9242b5f5 AS |
9848 | static void clean_func_state(struct bpf_verifier_env *env, |
9849 | struct bpf_func_state *st) | |
9850 | { | |
9851 | enum bpf_reg_liveness live; | |
9852 | int i, j; | |
9853 | ||
9854 | for (i = 0; i < BPF_REG_FP; i++) { | |
9855 | live = st->regs[i].live; | |
9856 | /* liveness must not touch this register anymore */ | |
9857 | st->regs[i].live |= REG_LIVE_DONE; | |
9858 | if (!(live & REG_LIVE_READ)) | |
9859 | /* since the register is unused, clear its state | |
9860 | * to make further comparison simpler | |
9861 | */ | |
f54c7898 | 9862 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
9863 | } |
9864 | ||
9865 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
9866 | live = st->stack[i].spilled_ptr.live; | |
9867 | /* liveness must not touch this stack slot anymore */ | |
9868 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
9869 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 9870 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
9871 | for (j = 0; j < BPF_REG_SIZE; j++) |
9872 | st->stack[i].slot_type[j] = STACK_INVALID; | |
9873 | } | |
9874 | } | |
9875 | } | |
9876 | ||
9877 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
9878 | struct bpf_verifier_state *st) | |
9879 | { | |
9880 | int i; | |
9881 | ||
9882 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
9883 | /* all regs in this state in all frames were already marked */ | |
9884 | return; | |
9885 | ||
9886 | for (i = 0; i <= st->curframe; i++) | |
9887 | clean_func_state(env, st->frame[i]); | |
9888 | } | |
9889 | ||
9890 | /* the parentage chains form a tree. | |
9891 | * the verifier states are added to state lists at given insn and | |
9892 | * pushed into state stack for future exploration. | |
9893 | * when the verifier reaches bpf_exit insn some of the verifer states | |
9894 | * stored in the state lists have their final liveness state already, | |
9895 | * but a lot of states will get revised from liveness point of view when | |
9896 | * the verifier explores other branches. | |
9897 | * Example: | |
9898 | * 1: r0 = 1 | |
9899 | * 2: if r1 == 100 goto pc+1 | |
9900 | * 3: r0 = 2 | |
9901 | * 4: exit | |
9902 | * when the verifier reaches exit insn the register r0 in the state list of | |
9903 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
9904 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
9905 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
9906 | * | |
9907 | * Since the verifier pushes the branch states as it sees them while exploring | |
9908 | * the program the condition of walking the branch instruction for the second | |
9909 | * time means that all states below this branch were already explored and | |
8fb33b60 | 9910 | * their final liveness marks are already propagated. |
9242b5f5 AS |
9911 | * Hence when the verifier completes the search of state list in is_state_visited() |
9912 | * we can call this clean_live_states() function to mark all liveness states | |
9913 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
9914 | * will not be used. | |
9915 | * This function also clears the registers and stack for states that !READ | |
9916 | * to simplify state merging. | |
9917 | * | |
9918 | * Important note here that walking the same branch instruction in the callee | |
9919 | * doesn't meant that the states are DONE. The verifier has to compare | |
9920 | * the callsites | |
9921 | */ | |
9922 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
9923 | struct bpf_verifier_state *cur) | |
9924 | { | |
9925 | struct bpf_verifier_state_list *sl; | |
9926 | int i; | |
9927 | ||
5d839021 | 9928 | sl = *explored_state(env, insn); |
a8f500af | 9929 | while (sl) { |
2589726d AS |
9930 | if (sl->state.branches) |
9931 | goto next; | |
dc2a4ebc AS |
9932 | if (sl->state.insn_idx != insn || |
9933 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
9934 | goto next; |
9935 | for (i = 0; i <= cur->curframe; i++) | |
9936 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
9937 | goto next; | |
9938 | clean_verifier_state(env, &sl->state); | |
9939 | next: | |
9940 | sl = sl->next; | |
9941 | } | |
9942 | } | |
9943 | ||
f1174f77 | 9944 | /* Returns true if (rold safe implies rcur safe) */ |
e042aa53 DB |
9945 | static bool regsafe(struct bpf_verifier_env *env, struct bpf_reg_state *rold, |
9946 | struct bpf_reg_state *rcur, struct bpf_id_pair *idmap) | |
f1174f77 | 9947 | { |
f4d7e40a AS |
9948 | bool equal; |
9949 | ||
dc503a8a EC |
9950 | if (!(rold->live & REG_LIVE_READ)) |
9951 | /* explored state didn't use this */ | |
9952 | return true; | |
9953 | ||
679c782d | 9954 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
9955 | |
9956 | if (rold->type == PTR_TO_STACK) | |
9957 | /* two stack pointers are equal only if they're pointing to | |
9958 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
9959 | */ | |
9960 | return equal && rold->frameno == rcur->frameno; | |
9961 | ||
9962 | if (equal) | |
969bf05e AS |
9963 | return true; |
9964 | ||
f1174f77 EC |
9965 | if (rold->type == NOT_INIT) |
9966 | /* explored state can't have used this */ | |
969bf05e | 9967 | return true; |
f1174f77 EC |
9968 | if (rcur->type == NOT_INIT) |
9969 | return false; | |
9970 | switch (rold->type) { | |
9971 | case SCALAR_VALUE: | |
e042aa53 DB |
9972 | if (env->explore_alu_limits) |
9973 | return false; | |
f1174f77 | 9974 | if (rcur->type == SCALAR_VALUE) { |
b5dc0163 AS |
9975 | if (!rold->precise && !rcur->precise) |
9976 | return true; | |
f1174f77 EC |
9977 | /* new val must satisfy old val knowledge */ |
9978 | return range_within(rold, rcur) && | |
9979 | tnum_in(rold->var_off, rcur->var_off); | |
9980 | } else { | |
179d1c56 JH |
9981 | /* We're trying to use a pointer in place of a scalar. |
9982 | * Even if the scalar was unbounded, this could lead to | |
9983 | * pointer leaks because scalars are allowed to leak | |
9984 | * while pointers are not. We could make this safe in | |
9985 | * special cases if root is calling us, but it's | |
9986 | * probably not worth the hassle. | |
f1174f77 | 9987 | */ |
179d1c56 | 9988 | return false; |
f1174f77 | 9989 | } |
69c087ba | 9990 | case PTR_TO_MAP_KEY: |
f1174f77 | 9991 | case PTR_TO_MAP_VALUE: |
1b688a19 EC |
9992 | /* If the new min/max/var_off satisfy the old ones and |
9993 | * everything else matches, we are OK. | |
d83525ca AS |
9994 | * 'id' is not compared, since it's only used for maps with |
9995 | * bpf_spin_lock inside map element and in such cases if | |
9996 | * the rest of the prog is valid for one map element then | |
9997 | * it's valid for all map elements regardless of the key | |
9998 | * used in bpf_map_lookup() | |
1b688a19 EC |
9999 | */ |
10000 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
10001 | range_within(rold, rcur) && | |
10002 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
10003 | case PTR_TO_MAP_VALUE_OR_NULL: |
10004 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
10005 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
10006 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
10007 | * checked, doing so could have affected others with the same | |
10008 | * id, and we can't check for that because we lost the id when | |
10009 | * we converted to a PTR_TO_MAP_VALUE. | |
10010 | */ | |
10011 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
10012 | return false; | |
10013 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
10014 | return false; | |
10015 | /* Check our ids match any regs they're supposed to */ | |
10016 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 10017 | case PTR_TO_PACKET_META: |
f1174f77 | 10018 | case PTR_TO_PACKET: |
de8f3a83 | 10019 | if (rcur->type != rold->type) |
f1174f77 EC |
10020 | return false; |
10021 | /* We must have at least as much range as the old ptr | |
10022 | * did, so that any accesses which were safe before are | |
10023 | * still safe. This is true even if old range < old off, | |
10024 | * since someone could have accessed through (ptr - k), or | |
10025 | * even done ptr -= k in a register, to get a safe access. | |
10026 | */ | |
10027 | if (rold->range > rcur->range) | |
10028 | return false; | |
10029 | /* If the offsets don't match, we can't trust our alignment; | |
10030 | * nor can we be sure that we won't fall out of range. | |
10031 | */ | |
10032 | if (rold->off != rcur->off) | |
10033 | return false; | |
10034 | /* id relations must be preserved */ | |
10035 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
10036 | return false; | |
10037 | /* new val must satisfy old val knowledge */ | |
10038 | return range_within(rold, rcur) && | |
10039 | tnum_in(rold->var_off, rcur->var_off); | |
10040 | case PTR_TO_CTX: | |
10041 | case CONST_PTR_TO_MAP: | |
f1174f77 | 10042 | case PTR_TO_PACKET_END: |
d58e468b | 10043 | case PTR_TO_FLOW_KEYS: |
c64b7983 JS |
10044 | case PTR_TO_SOCKET: |
10045 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
10046 | case PTR_TO_SOCK_COMMON: |
10047 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
10048 | case PTR_TO_TCP_SOCK: |
10049 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 10050 | case PTR_TO_XDP_SOCK: |
f1174f77 EC |
10051 | /* Only valid matches are exact, which memcmp() above |
10052 | * would have accepted | |
10053 | */ | |
10054 | default: | |
10055 | /* Don't know what's going on, just say it's not safe */ | |
10056 | return false; | |
10057 | } | |
969bf05e | 10058 | |
f1174f77 EC |
10059 | /* Shouldn't get here; if we do, say it's not safe */ |
10060 | WARN_ON_ONCE(1); | |
969bf05e AS |
10061 | return false; |
10062 | } | |
10063 | ||
e042aa53 DB |
10064 | static bool stacksafe(struct bpf_verifier_env *env, struct bpf_func_state *old, |
10065 | struct bpf_func_state *cur, struct bpf_id_pair *idmap) | |
638f5b90 AS |
10066 | { |
10067 | int i, spi; | |
10068 | ||
638f5b90 AS |
10069 | /* walk slots of the explored stack and ignore any additional |
10070 | * slots in the current stack, since explored(safe) state | |
10071 | * didn't use them | |
10072 | */ | |
10073 | for (i = 0; i < old->allocated_stack; i++) { | |
10074 | spi = i / BPF_REG_SIZE; | |
10075 | ||
b233920c AS |
10076 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
10077 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 10078 | /* explored state didn't use this */ |
fd05e57b | 10079 | continue; |
b233920c | 10080 | } |
cc2b14d5 | 10081 | |
638f5b90 AS |
10082 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
10083 | continue; | |
19e2dbb7 AS |
10084 | |
10085 | /* explored stack has more populated slots than current stack | |
10086 | * and these slots were used | |
10087 | */ | |
10088 | if (i >= cur->allocated_stack) | |
10089 | return false; | |
10090 | ||
cc2b14d5 AS |
10091 | /* if old state was safe with misc data in the stack |
10092 | * it will be safe with zero-initialized stack. | |
10093 | * The opposite is not true | |
10094 | */ | |
10095 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
10096 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
10097 | continue; | |
638f5b90 AS |
10098 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
10099 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
10100 | /* Ex: old explored (safe) state has STACK_SPILL in | |
b8c1a309 | 10101 | * this stack slot, but current has STACK_MISC -> |
638f5b90 AS |
10102 | * this verifier states are not equivalent, |
10103 | * return false to continue verification of this path | |
10104 | */ | |
10105 | return false; | |
10106 | if (i % BPF_REG_SIZE) | |
10107 | continue; | |
10108 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
10109 | continue; | |
e042aa53 DB |
10110 | if (!regsafe(env, &old->stack[spi].spilled_ptr, |
10111 | &cur->stack[spi].spilled_ptr, idmap)) | |
638f5b90 AS |
10112 | /* when explored and current stack slot are both storing |
10113 | * spilled registers, check that stored pointers types | |
10114 | * are the same as well. | |
10115 | * Ex: explored safe path could have stored | |
10116 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
10117 | * but current path has stored: | |
10118 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
10119 | * such verifier states are not equivalent. | |
10120 | * return false to continue verification of this path | |
10121 | */ | |
10122 | return false; | |
10123 | } | |
10124 | return true; | |
10125 | } | |
10126 | ||
fd978bf7 JS |
10127 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
10128 | { | |
10129 | if (old->acquired_refs != cur->acquired_refs) | |
10130 | return false; | |
10131 | return !memcmp(old->refs, cur->refs, | |
10132 | sizeof(*old->refs) * old->acquired_refs); | |
10133 | } | |
10134 | ||
f1bca824 AS |
10135 | /* compare two verifier states |
10136 | * | |
10137 | * all states stored in state_list are known to be valid, since | |
10138 | * verifier reached 'bpf_exit' instruction through them | |
10139 | * | |
10140 | * this function is called when verifier exploring different branches of | |
10141 | * execution popped from the state stack. If it sees an old state that has | |
10142 | * more strict register state and more strict stack state then this execution | |
10143 | * branch doesn't need to be explored further, since verifier already | |
10144 | * concluded that more strict state leads to valid finish. | |
10145 | * | |
10146 | * Therefore two states are equivalent if register state is more conservative | |
10147 | * and explored stack state is more conservative than the current one. | |
10148 | * Example: | |
10149 | * explored current | |
10150 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
10151 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
10152 | * | |
10153 | * In other words if current stack state (one being explored) has more | |
10154 | * valid slots than old one that already passed validation, it means | |
10155 | * the verifier can stop exploring and conclude that current state is valid too | |
10156 | * | |
10157 | * Similarly with registers. If explored state has register type as invalid | |
10158 | * whereas register type in current state is meaningful, it means that | |
10159 | * the current state will reach 'bpf_exit' instruction safely | |
10160 | */ | |
c9e73e3d | 10161 | static bool func_states_equal(struct bpf_verifier_env *env, struct bpf_func_state *old, |
f4d7e40a | 10162 | struct bpf_func_state *cur) |
f1bca824 AS |
10163 | { |
10164 | int i; | |
10165 | ||
c9e73e3d LB |
10166 | memset(env->idmap_scratch, 0, sizeof(env->idmap_scratch)); |
10167 | for (i = 0; i < MAX_BPF_REG; i++) | |
e042aa53 DB |
10168 | if (!regsafe(env, &old->regs[i], &cur->regs[i], |
10169 | env->idmap_scratch)) | |
c9e73e3d | 10170 | return false; |
f1bca824 | 10171 | |
e042aa53 | 10172 | if (!stacksafe(env, old, cur, env->idmap_scratch)) |
c9e73e3d | 10173 | return false; |
fd978bf7 JS |
10174 | |
10175 | if (!refsafe(old, cur)) | |
c9e73e3d LB |
10176 | return false; |
10177 | ||
10178 | return true; | |
f1bca824 AS |
10179 | } |
10180 | ||
f4d7e40a AS |
10181 | static bool states_equal(struct bpf_verifier_env *env, |
10182 | struct bpf_verifier_state *old, | |
10183 | struct bpf_verifier_state *cur) | |
10184 | { | |
10185 | int i; | |
10186 | ||
10187 | if (old->curframe != cur->curframe) | |
10188 | return false; | |
10189 | ||
979d63d5 DB |
10190 | /* Verification state from speculative execution simulation |
10191 | * must never prune a non-speculative execution one. | |
10192 | */ | |
10193 | if (old->speculative && !cur->speculative) | |
10194 | return false; | |
10195 | ||
d83525ca AS |
10196 | if (old->active_spin_lock != cur->active_spin_lock) |
10197 | return false; | |
10198 | ||
f4d7e40a AS |
10199 | /* for states to be equal callsites have to be the same |
10200 | * and all frame states need to be equivalent | |
10201 | */ | |
10202 | for (i = 0; i <= old->curframe; i++) { | |
10203 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
10204 | return false; | |
c9e73e3d | 10205 | if (!func_states_equal(env, old->frame[i], cur->frame[i])) |
f4d7e40a AS |
10206 | return false; |
10207 | } | |
10208 | return true; | |
10209 | } | |
10210 | ||
5327ed3d JW |
10211 | /* Return 0 if no propagation happened. Return negative error code if error |
10212 | * happened. Otherwise, return the propagated bit. | |
10213 | */ | |
55e7f3b5 JW |
10214 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
10215 | struct bpf_reg_state *reg, | |
10216 | struct bpf_reg_state *parent_reg) | |
10217 | { | |
5327ed3d JW |
10218 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
10219 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
10220 | int err; |
10221 | ||
5327ed3d JW |
10222 | /* When comes here, read flags of PARENT_REG or REG could be any of |
10223 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
10224 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
10225 | */ | |
10226 | if (parent_flag == REG_LIVE_READ64 || | |
10227 | /* Or if there is no read flag from REG. */ | |
10228 | !flag || | |
10229 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
10230 | parent_flag == flag) | |
55e7f3b5 JW |
10231 | return 0; |
10232 | ||
5327ed3d | 10233 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
10234 | if (err) |
10235 | return err; | |
10236 | ||
5327ed3d | 10237 | return flag; |
55e7f3b5 JW |
10238 | } |
10239 | ||
8e9cd9ce | 10240 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
10241 | * straight-line code between a state and its parent. When we arrive at an |
10242 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
10243 | * code, so read marks in the state must propagate to the parent regardless | |
10244 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 10245 | * in mark_reg_read() is for. |
8e9cd9ce | 10246 | */ |
f4d7e40a AS |
10247 | static int propagate_liveness(struct bpf_verifier_env *env, |
10248 | const struct bpf_verifier_state *vstate, | |
10249 | struct bpf_verifier_state *vparent) | |
dc503a8a | 10250 | { |
3f8cafa4 | 10251 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 10252 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 10253 | int i, frame, err = 0; |
dc503a8a | 10254 | |
f4d7e40a AS |
10255 | if (vparent->curframe != vstate->curframe) { |
10256 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
10257 | vparent->curframe, vstate->curframe); | |
10258 | return -EFAULT; | |
10259 | } | |
dc503a8a EC |
10260 | /* Propagate read liveness of registers... */ |
10261 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 10262 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
10263 | parent = vparent->frame[frame]; |
10264 | state = vstate->frame[frame]; | |
10265 | parent_reg = parent->regs; | |
10266 | state_reg = state->regs; | |
83d16312 JK |
10267 | /* We don't need to worry about FP liveness, it's read-only */ |
10268 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
10269 | err = propagate_liveness_reg(env, &state_reg[i], |
10270 | &parent_reg[i]); | |
5327ed3d | 10271 | if (err < 0) |
3f8cafa4 | 10272 | return err; |
5327ed3d JW |
10273 | if (err == REG_LIVE_READ64) |
10274 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 10275 | } |
f4d7e40a | 10276 | |
1b04aee7 | 10277 | /* Propagate stack slots. */ |
f4d7e40a AS |
10278 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
10279 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
10280 | parent_reg = &parent->stack[i].spilled_ptr; |
10281 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
10282 | err = propagate_liveness_reg(env, state_reg, |
10283 | parent_reg); | |
5327ed3d | 10284 | if (err < 0) |
3f8cafa4 | 10285 | return err; |
dc503a8a EC |
10286 | } |
10287 | } | |
5327ed3d | 10288 | return 0; |
dc503a8a EC |
10289 | } |
10290 | ||
a3ce685d AS |
10291 | /* find precise scalars in the previous equivalent state and |
10292 | * propagate them into the current state | |
10293 | */ | |
10294 | static int propagate_precision(struct bpf_verifier_env *env, | |
10295 | const struct bpf_verifier_state *old) | |
10296 | { | |
10297 | struct bpf_reg_state *state_reg; | |
10298 | struct bpf_func_state *state; | |
10299 | int i, err = 0; | |
10300 | ||
10301 | state = old->frame[old->curframe]; | |
10302 | state_reg = state->regs; | |
10303 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { | |
10304 | if (state_reg->type != SCALAR_VALUE || | |
10305 | !state_reg->precise) | |
10306 | continue; | |
10307 | if (env->log.level & BPF_LOG_LEVEL2) | |
10308 | verbose(env, "propagating r%d\n", i); | |
10309 | err = mark_chain_precision(env, i); | |
10310 | if (err < 0) | |
10311 | return err; | |
10312 | } | |
10313 | ||
10314 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
10315 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
10316 | continue; | |
10317 | state_reg = &state->stack[i].spilled_ptr; | |
10318 | if (state_reg->type != SCALAR_VALUE || | |
10319 | !state_reg->precise) | |
10320 | continue; | |
10321 | if (env->log.level & BPF_LOG_LEVEL2) | |
10322 | verbose(env, "propagating fp%d\n", | |
10323 | (-i - 1) * BPF_REG_SIZE); | |
10324 | err = mark_chain_precision_stack(env, i); | |
10325 | if (err < 0) | |
10326 | return err; | |
10327 | } | |
10328 | return 0; | |
10329 | } | |
10330 | ||
2589726d AS |
10331 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
10332 | struct bpf_verifier_state *cur) | |
10333 | { | |
10334 | struct bpf_func_state *fold, *fcur; | |
10335 | int i, fr = cur->curframe; | |
10336 | ||
10337 | if (old->curframe != fr) | |
10338 | return false; | |
10339 | ||
10340 | fold = old->frame[fr]; | |
10341 | fcur = cur->frame[fr]; | |
10342 | for (i = 0; i < MAX_BPF_REG; i++) | |
10343 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
10344 | offsetof(struct bpf_reg_state, parent))) | |
10345 | return false; | |
10346 | return true; | |
10347 | } | |
10348 | ||
10349 | ||
58e2af8b | 10350 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 10351 | { |
58e2af8b | 10352 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 10353 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 10354 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 10355 | int i, j, err, states_cnt = 0; |
10d274e8 | 10356 | bool add_new_state = env->test_state_freq ? true : false; |
f1bca824 | 10357 | |
b5dc0163 | 10358 | cur->last_insn_idx = env->prev_insn_idx; |
a8f500af | 10359 | if (!env->insn_aux_data[insn_idx].prune_point) |
f1bca824 AS |
10360 | /* this 'insn_idx' instruction wasn't marked, so we will not |
10361 | * be doing state search here | |
10362 | */ | |
10363 | return 0; | |
10364 | ||
2589726d AS |
10365 | /* bpf progs typically have pruning point every 4 instructions |
10366 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
10367 | * Do not add new state for future pruning if the verifier hasn't seen | |
10368 | * at least 2 jumps and at least 8 instructions. | |
10369 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
10370 | * In tests that amounts to up to 50% reduction into total verifier | |
10371 | * memory consumption and 20% verifier time speedup. | |
10372 | */ | |
10373 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
10374 | env->insn_processed - env->prev_insn_processed >= 8) | |
10375 | add_new_state = true; | |
10376 | ||
a8f500af AS |
10377 | pprev = explored_state(env, insn_idx); |
10378 | sl = *pprev; | |
10379 | ||
9242b5f5 AS |
10380 | clean_live_states(env, insn_idx, cur); |
10381 | ||
a8f500af | 10382 | while (sl) { |
dc2a4ebc AS |
10383 | states_cnt++; |
10384 | if (sl->state.insn_idx != insn_idx) | |
10385 | goto next; | |
2589726d AS |
10386 | if (sl->state.branches) { |
10387 | if (states_maybe_looping(&sl->state, cur) && | |
10388 | states_equal(env, &sl->state, cur)) { | |
10389 | verbose_linfo(env, insn_idx, "; "); | |
10390 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
10391 | return -EINVAL; | |
10392 | } | |
10393 | /* if the verifier is processing a loop, avoid adding new state | |
10394 | * too often, since different loop iterations have distinct | |
10395 | * states and may not help future pruning. | |
10396 | * This threshold shouldn't be too low to make sure that | |
10397 | * a loop with large bound will be rejected quickly. | |
10398 | * The most abusive loop will be: | |
10399 | * r1 += 1 | |
10400 | * if r1 < 1000000 goto pc-2 | |
10401 | * 1M insn_procssed limit / 100 == 10k peak states. | |
10402 | * This threshold shouldn't be too high either, since states | |
10403 | * at the end of the loop are likely to be useful in pruning. | |
10404 | */ | |
10405 | if (env->jmps_processed - env->prev_jmps_processed < 20 && | |
10406 | env->insn_processed - env->prev_insn_processed < 100) | |
10407 | add_new_state = false; | |
10408 | goto miss; | |
10409 | } | |
638f5b90 | 10410 | if (states_equal(env, &sl->state, cur)) { |
9f4686c4 | 10411 | sl->hit_cnt++; |
f1bca824 | 10412 | /* reached equivalent register/stack state, |
dc503a8a EC |
10413 | * prune the search. |
10414 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
10415 | * If we have any write marks in env->cur_state, they |
10416 | * will prevent corresponding reads in the continuation | |
10417 | * from reaching our parent (an explored_state). Our | |
10418 | * own state will get the read marks recorded, but | |
10419 | * they'll be immediately forgotten as we're pruning | |
10420 | * this state and will pop a new one. | |
f1bca824 | 10421 | */ |
f4d7e40a | 10422 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
10423 | |
10424 | /* if previous state reached the exit with precision and | |
10425 | * current state is equivalent to it (except precsion marks) | |
10426 | * the precision needs to be propagated back in | |
10427 | * the current state. | |
10428 | */ | |
10429 | err = err ? : push_jmp_history(env, cur); | |
10430 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
10431 | if (err) |
10432 | return err; | |
f1bca824 | 10433 | return 1; |
dc503a8a | 10434 | } |
2589726d AS |
10435 | miss: |
10436 | /* when new state is not going to be added do not increase miss count. | |
10437 | * Otherwise several loop iterations will remove the state | |
10438 | * recorded earlier. The goal of these heuristics is to have | |
10439 | * states from some iterations of the loop (some in the beginning | |
10440 | * and some at the end) to help pruning. | |
10441 | */ | |
10442 | if (add_new_state) | |
10443 | sl->miss_cnt++; | |
9f4686c4 AS |
10444 | /* heuristic to determine whether this state is beneficial |
10445 | * to keep checking from state equivalence point of view. | |
10446 | * Higher numbers increase max_states_per_insn and verification time, | |
10447 | * but do not meaningfully decrease insn_processed. | |
10448 | */ | |
10449 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
10450 | /* the state is unlikely to be useful. Remove it to | |
10451 | * speed up verification | |
10452 | */ | |
10453 | *pprev = sl->next; | |
10454 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
2589726d AS |
10455 | u32 br = sl->state.branches; |
10456 | ||
10457 | WARN_ONCE(br, | |
10458 | "BUG live_done but branches_to_explore %d\n", | |
10459 | br); | |
9f4686c4 AS |
10460 | free_verifier_state(&sl->state, false); |
10461 | kfree(sl); | |
10462 | env->peak_states--; | |
10463 | } else { | |
10464 | /* cannot free this state, since parentage chain may | |
10465 | * walk it later. Add it for free_list instead to | |
10466 | * be freed at the end of verification | |
10467 | */ | |
10468 | sl->next = env->free_list; | |
10469 | env->free_list = sl; | |
10470 | } | |
10471 | sl = *pprev; | |
10472 | continue; | |
10473 | } | |
dc2a4ebc | 10474 | next: |
9f4686c4 AS |
10475 | pprev = &sl->next; |
10476 | sl = *pprev; | |
f1bca824 AS |
10477 | } |
10478 | ||
06ee7115 AS |
10479 | if (env->max_states_per_insn < states_cnt) |
10480 | env->max_states_per_insn = states_cnt; | |
10481 | ||
2c78ee89 | 10482 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
b5dc0163 | 10483 | return push_jmp_history(env, cur); |
ceefbc96 | 10484 | |
2589726d | 10485 | if (!add_new_state) |
b5dc0163 | 10486 | return push_jmp_history(env, cur); |
ceefbc96 | 10487 | |
2589726d AS |
10488 | /* There were no equivalent states, remember the current one. |
10489 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 10490 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 10491 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 10492 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
10493 | * again on the way to bpf_exit. |
10494 | * When looping the sl->state.branches will be > 0 and this state | |
10495 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 10496 | */ |
638f5b90 | 10497 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
10498 | if (!new_sl) |
10499 | return -ENOMEM; | |
06ee7115 AS |
10500 | env->total_states++; |
10501 | env->peak_states++; | |
2589726d AS |
10502 | env->prev_jmps_processed = env->jmps_processed; |
10503 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 AS |
10504 | |
10505 | /* add new state to the head of linked list */ | |
679c782d EC |
10506 | new = &new_sl->state; |
10507 | err = copy_verifier_state(new, cur); | |
1969db47 | 10508 | if (err) { |
679c782d | 10509 | free_verifier_state(new, false); |
1969db47 AS |
10510 | kfree(new_sl); |
10511 | return err; | |
10512 | } | |
dc2a4ebc | 10513 | new->insn_idx = insn_idx; |
2589726d AS |
10514 | WARN_ONCE(new->branches != 1, |
10515 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 10516 | |
2589726d | 10517 | cur->parent = new; |
b5dc0163 AS |
10518 | cur->first_insn_idx = insn_idx; |
10519 | clear_jmp_history(cur); | |
5d839021 AS |
10520 | new_sl->next = *explored_state(env, insn_idx); |
10521 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
10522 | /* connect new state to parentage chain. Current frame needs all |
10523 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
10524 | * to the stack implicitly by JITs) so in callers' frames connect just | |
10525 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
10526 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
10527 | * from callee with its full parentage chain, anyway. | |
10528 | */ | |
8e9cd9ce EC |
10529 | /* clear write marks in current state: the writes we did are not writes |
10530 | * our child did, so they don't screen off its reads from us. | |
10531 | * (There are no read marks in current state, because reads always mark | |
10532 | * their parent and current state never has children yet. Only | |
10533 | * explored_states can get read marks.) | |
10534 | */ | |
eea1c227 AS |
10535 | for (j = 0; j <= cur->curframe; j++) { |
10536 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
10537 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
10538 | for (i = 0; i < BPF_REG_FP; i++) | |
10539 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
10540 | } | |
f4d7e40a AS |
10541 | |
10542 | /* all stack frames are accessible from callee, clear them all */ | |
10543 | for (j = 0; j <= cur->curframe; j++) { | |
10544 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 10545 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 10546 | |
679c782d | 10547 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 10548 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
10549 | frame->stack[i].spilled_ptr.parent = |
10550 | &newframe->stack[i].spilled_ptr; | |
10551 | } | |
f4d7e40a | 10552 | } |
f1bca824 AS |
10553 | return 0; |
10554 | } | |
10555 | ||
c64b7983 JS |
10556 | /* Return true if it's OK to have the same insn return a different type. */ |
10557 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
10558 | { | |
10559 | switch (type) { | |
10560 | case PTR_TO_CTX: | |
10561 | case PTR_TO_SOCKET: | |
10562 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
10563 | case PTR_TO_SOCK_COMMON: |
10564 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
10565 | case PTR_TO_TCP_SOCK: |
10566 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 10567 | case PTR_TO_XDP_SOCK: |
2a02759e | 10568 | case PTR_TO_BTF_ID: |
b121b341 | 10569 | case PTR_TO_BTF_ID_OR_NULL: |
c64b7983 JS |
10570 | return false; |
10571 | default: | |
10572 | return true; | |
10573 | } | |
10574 | } | |
10575 | ||
10576 | /* If an instruction was previously used with particular pointer types, then we | |
10577 | * need to be careful to avoid cases such as the below, where it may be ok | |
10578 | * for one branch accessing the pointer, but not ok for the other branch: | |
10579 | * | |
10580 | * R1 = sock_ptr | |
10581 | * goto X; | |
10582 | * ... | |
10583 | * R1 = some_other_valid_ptr; | |
10584 | * goto X; | |
10585 | * ... | |
10586 | * R2 = *(u32 *)(R1 + 0); | |
10587 | */ | |
10588 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
10589 | { | |
10590 | return src != prev && (!reg_type_mismatch_ok(src) || | |
10591 | !reg_type_mismatch_ok(prev)); | |
10592 | } | |
10593 | ||
58e2af8b | 10594 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 10595 | { |
6f8a57cc | 10596 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 10597 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 10598 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 10599 | struct bpf_reg_state *regs; |
06ee7115 | 10600 | int insn_cnt = env->prog->len; |
17a52670 | 10601 | bool do_print_state = false; |
b5dc0163 | 10602 | int prev_insn_idx = -1; |
17a52670 | 10603 | |
17a52670 AS |
10604 | for (;;) { |
10605 | struct bpf_insn *insn; | |
10606 | u8 class; | |
10607 | int err; | |
10608 | ||
b5dc0163 | 10609 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 10610 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 10611 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 10612 | env->insn_idx, insn_cnt); |
17a52670 AS |
10613 | return -EFAULT; |
10614 | } | |
10615 | ||
c08435ec | 10616 | insn = &insns[env->insn_idx]; |
17a52670 AS |
10617 | class = BPF_CLASS(insn->code); |
10618 | ||
06ee7115 | 10619 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
10620 | verbose(env, |
10621 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 10622 | env->insn_processed); |
17a52670 AS |
10623 | return -E2BIG; |
10624 | } | |
10625 | ||
c08435ec | 10626 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
10627 | if (err < 0) |
10628 | return err; | |
10629 | if (err == 1) { | |
10630 | /* found equivalent state, can prune the search */ | |
06ee7115 | 10631 | if (env->log.level & BPF_LOG_LEVEL) { |
f1bca824 | 10632 | if (do_print_state) |
979d63d5 DB |
10633 | verbose(env, "\nfrom %d to %d%s: safe\n", |
10634 | env->prev_insn_idx, env->insn_idx, | |
10635 | env->cur_state->speculative ? | |
10636 | " (speculative execution)" : ""); | |
f1bca824 | 10637 | else |
c08435ec | 10638 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
10639 | } |
10640 | goto process_bpf_exit; | |
10641 | } | |
10642 | ||
c3494801 AS |
10643 | if (signal_pending(current)) |
10644 | return -EAGAIN; | |
10645 | ||
3c2ce60b DB |
10646 | if (need_resched()) |
10647 | cond_resched(); | |
10648 | ||
06ee7115 AS |
10649 | if (env->log.level & BPF_LOG_LEVEL2 || |
10650 | (env->log.level & BPF_LOG_LEVEL && do_print_state)) { | |
10651 | if (env->log.level & BPF_LOG_LEVEL2) | |
c08435ec | 10652 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 10653 | else |
979d63d5 DB |
10654 | verbose(env, "\nfrom %d to %d%s:", |
10655 | env->prev_insn_idx, env->insn_idx, | |
10656 | env->cur_state->speculative ? | |
10657 | " (speculative execution)" : ""); | |
f4d7e40a | 10658 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
10659 | do_print_state = false; |
10660 | } | |
10661 | ||
06ee7115 | 10662 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 | 10663 | const struct bpf_insn_cbs cbs = { |
e6ac2450 | 10664 | .cb_call = disasm_kfunc_name, |
7105e828 | 10665 | .cb_print = verbose, |
abe08840 | 10666 | .private_data = env, |
7105e828 DB |
10667 | }; |
10668 | ||
c08435ec DB |
10669 | verbose_linfo(env, env->insn_idx, "; "); |
10670 | verbose(env, "%d: ", env->insn_idx); | |
abe08840 | 10671 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
10672 | } |
10673 | ||
cae1927c | 10674 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
10675 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
10676 | env->prev_insn_idx); | |
cae1927c JK |
10677 | if (err) |
10678 | return err; | |
10679 | } | |
13a27dfc | 10680 | |
638f5b90 | 10681 | regs = cur_regs(env); |
fe9a5ca7 | 10682 | sanitize_mark_insn_seen(env); |
b5dc0163 | 10683 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 10684 | |
17a52670 | 10685 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 10686 | err = check_alu_op(env, insn); |
17a52670 AS |
10687 | if (err) |
10688 | return err; | |
10689 | ||
10690 | } else if (class == BPF_LDX) { | |
3df126f3 | 10691 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
10692 | |
10693 | /* check for reserved fields is already done */ | |
10694 | ||
17a52670 | 10695 | /* check src operand */ |
dc503a8a | 10696 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10697 | if (err) |
10698 | return err; | |
10699 | ||
dc503a8a | 10700 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
10701 | if (err) |
10702 | return err; | |
10703 | ||
725f9dcd AS |
10704 | src_reg_type = regs[insn->src_reg].type; |
10705 | ||
17a52670 AS |
10706 | /* check that memory (src_reg + off) is readable, |
10707 | * the state of dst_reg will be updated by this func | |
10708 | */ | |
c08435ec DB |
10709 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
10710 | insn->off, BPF_SIZE(insn->code), | |
10711 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
10712 | if (err) |
10713 | return err; | |
10714 | ||
c08435ec | 10715 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
10716 | |
10717 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
10718 | /* saw a valid insn |
10719 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 10720 | * save type to validate intersecting paths |
9bac3d6d | 10721 | */ |
3df126f3 | 10722 | *prev_src_type = src_reg_type; |
9bac3d6d | 10723 | |
c64b7983 | 10724 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
10725 | /* ABuser program is trying to use the same insn |
10726 | * dst_reg = *(u32*) (src_reg + off) | |
10727 | * with different pointer types: | |
10728 | * src_reg == ctx in one branch and | |
10729 | * src_reg == stack|map in some other branch. | |
10730 | * Reject it. | |
10731 | */ | |
61bd5218 | 10732 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
10733 | return -EINVAL; |
10734 | } | |
10735 | ||
17a52670 | 10736 | } else if (class == BPF_STX) { |
3df126f3 | 10737 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 10738 | |
91c960b0 BJ |
10739 | if (BPF_MODE(insn->code) == BPF_ATOMIC) { |
10740 | err = check_atomic(env, env->insn_idx, insn); | |
17a52670 AS |
10741 | if (err) |
10742 | return err; | |
c08435ec | 10743 | env->insn_idx++; |
17a52670 AS |
10744 | continue; |
10745 | } | |
10746 | ||
5ca419f2 BJ |
10747 | if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) { |
10748 | verbose(env, "BPF_STX uses reserved fields\n"); | |
10749 | return -EINVAL; | |
10750 | } | |
10751 | ||
17a52670 | 10752 | /* check src1 operand */ |
dc503a8a | 10753 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10754 | if (err) |
10755 | return err; | |
10756 | /* check src2 operand */ | |
dc503a8a | 10757 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10758 | if (err) |
10759 | return err; | |
10760 | ||
d691f9e8 AS |
10761 | dst_reg_type = regs[insn->dst_reg].type; |
10762 | ||
17a52670 | 10763 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
10764 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
10765 | insn->off, BPF_SIZE(insn->code), | |
10766 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
10767 | if (err) |
10768 | return err; | |
10769 | ||
c08435ec | 10770 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
10771 | |
10772 | if (*prev_dst_type == NOT_INIT) { | |
10773 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 10774 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 10775 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
10776 | return -EINVAL; |
10777 | } | |
10778 | ||
17a52670 AS |
10779 | } else if (class == BPF_ST) { |
10780 | if (BPF_MODE(insn->code) != BPF_MEM || | |
10781 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 10782 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
10783 | return -EINVAL; |
10784 | } | |
10785 | /* check src operand */ | |
dc503a8a | 10786 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10787 | if (err) |
10788 | return err; | |
10789 | ||
f37a8cb8 | 10790 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 10791 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f DB |
10792 | insn->dst_reg, |
10793 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
10794 | return -EACCES; |
10795 | } | |
10796 | ||
17a52670 | 10797 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
10798 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
10799 | insn->off, BPF_SIZE(insn->code), | |
10800 | BPF_WRITE, -1, false); | |
17a52670 AS |
10801 | if (err) |
10802 | return err; | |
10803 | ||
092ed096 | 10804 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
10805 | u8 opcode = BPF_OP(insn->code); |
10806 | ||
2589726d | 10807 | env->jmps_processed++; |
17a52670 AS |
10808 | if (opcode == BPF_CALL) { |
10809 | if (BPF_SRC(insn->code) != BPF_K || | |
10810 | insn->off != 0 || | |
f4d7e40a | 10811 | (insn->src_reg != BPF_REG_0 && |
e6ac2450 MKL |
10812 | insn->src_reg != BPF_PSEUDO_CALL && |
10813 | insn->src_reg != BPF_PSEUDO_KFUNC_CALL) || | |
092ed096 JW |
10814 | insn->dst_reg != BPF_REG_0 || |
10815 | class == BPF_JMP32) { | |
61bd5218 | 10816 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
10817 | return -EINVAL; |
10818 | } | |
10819 | ||
d83525ca AS |
10820 | if (env->cur_state->active_spin_lock && |
10821 | (insn->src_reg == BPF_PSEUDO_CALL || | |
10822 | insn->imm != BPF_FUNC_spin_unlock)) { | |
10823 | verbose(env, "function calls are not allowed while holding a lock\n"); | |
10824 | return -EINVAL; | |
10825 | } | |
f4d7e40a | 10826 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 10827 | err = check_func_call(env, insn, &env->insn_idx); |
e6ac2450 MKL |
10828 | else if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) |
10829 | err = check_kfunc_call(env, insn); | |
f4d7e40a | 10830 | else |
69c087ba | 10831 | err = check_helper_call(env, insn, &env->insn_idx); |
17a52670 AS |
10832 | if (err) |
10833 | return err; | |
17a52670 AS |
10834 | } else if (opcode == BPF_JA) { |
10835 | if (BPF_SRC(insn->code) != BPF_K || | |
10836 | insn->imm != 0 || | |
10837 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
10838 | insn->dst_reg != BPF_REG_0 || |
10839 | class == BPF_JMP32) { | |
61bd5218 | 10840 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
10841 | return -EINVAL; |
10842 | } | |
10843 | ||
c08435ec | 10844 | env->insn_idx += insn->off + 1; |
17a52670 AS |
10845 | continue; |
10846 | ||
10847 | } else if (opcode == BPF_EXIT) { | |
10848 | if (BPF_SRC(insn->code) != BPF_K || | |
10849 | insn->imm != 0 || | |
10850 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
10851 | insn->dst_reg != BPF_REG_0 || |
10852 | class == BPF_JMP32) { | |
61bd5218 | 10853 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
10854 | return -EINVAL; |
10855 | } | |
10856 | ||
d83525ca AS |
10857 | if (env->cur_state->active_spin_lock) { |
10858 | verbose(env, "bpf_spin_unlock is missing\n"); | |
10859 | return -EINVAL; | |
10860 | } | |
10861 | ||
f4d7e40a AS |
10862 | if (state->curframe) { |
10863 | /* exit from nested function */ | |
c08435ec | 10864 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
10865 | if (err) |
10866 | return err; | |
10867 | do_print_state = true; | |
10868 | continue; | |
10869 | } | |
10870 | ||
fd978bf7 JS |
10871 | err = check_reference_leak(env); |
10872 | if (err) | |
10873 | return err; | |
10874 | ||
390ee7e2 AS |
10875 | err = check_return_code(env); |
10876 | if (err) | |
10877 | return err; | |
f1bca824 | 10878 | process_bpf_exit: |
2589726d | 10879 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 10880 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 10881 | &env->insn_idx, pop_log); |
638f5b90 AS |
10882 | if (err < 0) { |
10883 | if (err != -ENOENT) | |
10884 | return err; | |
17a52670 AS |
10885 | break; |
10886 | } else { | |
10887 | do_print_state = true; | |
10888 | continue; | |
10889 | } | |
10890 | } else { | |
c08435ec | 10891 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
10892 | if (err) |
10893 | return err; | |
10894 | } | |
10895 | } else if (class == BPF_LD) { | |
10896 | u8 mode = BPF_MODE(insn->code); | |
10897 | ||
10898 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
10899 | err = check_ld_abs(env, insn); |
10900 | if (err) | |
10901 | return err; | |
10902 | ||
17a52670 AS |
10903 | } else if (mode == BPF_IMM) { |
10904 | err = check_ld_imm(env, insn); | |
10905 | if (err) | |
10906 | return err; | |
10907 | ||
c08435ec | 10908 | env->insn_idx++; |
fe9a5ca7 | 10909 | sanitize_mark_insn_seen(env); |
17a52670 | 10910 | } else { |
61bd5218 | 10911 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
10912 | return -EINVAL; |
10913 | } | |
10914 | } else { | |
61bd5218 | 10915 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
10916 | return -EINVAL; |
10917 | } | |
10918 | ||
c08435ec | 10919 | env->insn_idx++; |
17a52670 AS |
10920 | } |
10921 | ||
10922 | return 0; | |
10923 | } | |
10924 | ||
541c3bad AN |
10925 | static int find_btf_percpu_datasec(struct btf *btf) |
10926 | { | |
10927 | const struct btf_type *t; | |
10928 | const char *tname; | |
10929 | int i, n; | |
10930 | ||
10931 | /* | |
10932 | * Both vmlinux and module each have their own ".data..percpu" | |
10933 | * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF | |
10934 | * types to look at only module's own BTF types. | |
10935 | */ | |
10936 | n = btf_nr_types(btf); | |
10937 | if (btf_is_module(btf)) | |
10938 | i = btf_nr_types(btf_vmlinux); | |
10939 | else | |
10940 | i = 1; | |
10941 | ||
10942 | for(; i < n; i++) { | |
10943 | t = btf_type_by_id(btf, i); | |
10944 | if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC) | |
10945 | continue; | |
10946 | ||
10947 | tname = btf_name_by_offset(btf, t->name_off); | |
10948 | if (!strcmp(tname, ".data..percpu")) | |
10949 | return i; | |
10950 | } | |
10951 | ||
10952 | return -ENOENT; | |
10953 | } | |
10954 | ||
4976b718 HL |
10955 | /* replace pseudo btf_id with kernel symbol address */ |
10956 | static int check_pseudo_btf_id(struct bpf_verifier_env *env, | |
10957 | struct bpf_insn *insn, | |
10958 | struct bpf_insn_aux_data *aux) | |
10959 | { | |
eaa6bcb7 HL |
10960 | const struct btf_var_secinfo *vsi; |
10961 | const struct btf_type *datasec; | |
541c3bad | 10962 | struct btf_mod_pair *btf_mod; |
4976b718 HL |
10963 | const struct btf_type *t; |
10964 | const char *sym_name; | |
eaa6bcb7 | 10965 | bool percpu = false; |
f16e6313 | 10966 | u32 type, id = insn->imm; |
541c3bad | 10967 | struct btf *btf; |
f16e6313 | 10968 | s32 datasec_id; |
4976b718 | 10969 | u64 addr; |
541c3bad | 10970 | int i, btf_fd, err; |
4976b718 | 10971 | |
541c3bad AN |
10972 | btf_fd = insn[1].imm; |
10973 | if (btf_fd) { | |
10974 | btf = btf_get_by_fd(btf_fd); | |
10975 | if (IS_ERR(btf)) { | |
10976 | verbose(env, "invalid module BTF object FD specified.\n"); | |
10977 | return -EINVAL; | |
10978 | } | |
10979 | } else { | |
10980 | if (!btf_vmlinux) { | |
10981 | verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n"); | |
10982 | return -EINVAL; | |
10983 | } | |
10984 | btf = btf_vmlinux; | |
10985 | btf_get(btf); | |
4976b718 HL |
10986 | } |
10987 | ||
541c3bad | 10988 | t = btf_type_by_id(btf, id); |
4976b718 HL |
10989 | if (!t) { |
10990 | verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id); | |
541c3bad AN |
10991 | err = -ENOENT; |
10992 | goto err_put; | |
4976b718 HL |
10993 | } |
10994 | ||
10995 | if (!btf_type_is_var(t)) { | |
541c3bad AN |
10996 | verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id); |
10997 | err = -EINVAL; | |
10998 | goto err_put; | |
4976b718 HL |
10999 | } |
11000 | ||
541c3bad | 11001 | sym_name = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
11002 | addr = kallsyms_lookup_name(sym_name); |
11003 | if (!addr) { | |
11004 | verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n", | |
11005 | sym_name); | |
541c3bad AN |
11006 | err = -ENOENT; |
11007 | goto err_put; | |
4976b718 HL |
11008 | } |
11009 | ||
541c3bad | 11010 | datasec_id = find_btf_percpu_datasec(btf); |
eaa6bcb7 | 11011 | if (datasec_id > 0) { |
541c3bad | 11012 | datasec = btf_type_by_id(btf, datasec_id); |
eaa6bcb7 HL |
11013 | for_each_vsi(i, datasec, vsi) { |
11014 | if (vsi->type == id) { | |
11015 | percpu = true; | |
11016 | break; | |
11017 | } | |
11018 | } | |
11019 | } | |
11020 | ||
4976b718 HL |
11021 | insn[0].imm = (u32)addr; |
11022 | insn[1].imm = addr >> 32; | |
11023 | ||
11024 | type = t->type; | |
541c3bad | 11025 | t = btf_type_skip_modifiers(btf, type, NULL); |
eaa6bcb7 HL |
11026 | if (percpu) { |
11027 | aux->btf_var.reg_type = PTR_TO_PERCPU_BTF_ID; | |
541c3bad | 11028 | aux->btf_var.btf = btf; |
eaa6bcb7 HL |
11029 | aux->btf_var.btf_id = type; |
11030 | } else if (!btf_type_is_struct(t)) { | |
4976b718 HL |
11031 | const struct btf_type *ret; |
11032 | const char *tname; | |
11033 | u32 tsize; | |
11034 | ||
11035 | /* resolve the type size of ksym. */ | |
541c3bad | 11036 | ret = btf_resolve_size(btf, t, &tsize); |
4976b718 | 11037 | if (IS_ERR(ret)) { |
541c3bad | 11038 | tname = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
11039 | verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n", |
11040 | tname, PTR_ERR(ret)); | |
541c3bad AN |
11041 | err = -EINVAL; |
11042 | goto err_put; | |
4976b718 HL |
11043 | } |
11044 | aux->btf_var.reg_type = PTR_TO_MEM; | |
11045 | aux->btf_var.mem_size = tsize; | |
11046 | } else { | |
11047 | aux->btf_var.reg_type = PTR_TO_BTF_ID; | |
541c3bad | 11048 | aux->btf_var.btf = btf; |
4976b718 HL |
11049 | aux->btf_var.btf_id = type; |
11050 | } | |
541c3bad AN |
11051 | |
11052 | /* check whether we recorded this BTF (and maybe module) already */ | |
11053 | for (i = 0; i < env->used_btf_cnt; i++) { | |
11054 | if (env->used_btfs[i].btf == btf) { | |
11055 | btf_put(btf); | |
11056 | return 0; | |
11057 | } | |
11058 | } | |
11059 | ||
11060 | if (env->used_btf_cnt >= MAX_USED_BTFS) { | |
11061 | err = -E2BIG; | |
11062 | goto err_put; | |
11063 | } | |
11064 | ||
11065 | btf_mod = &env->used_btfs[env->used_btf_cnt]; | |
11066 | btf_mod->btf = btf; | |
11067 | btf_mod->module = NULL; | |
11068 | ||
11069 | /* if we reference variables from kernel module, bump its refcount */ | |
11070 | if (btf_is_module(btf)) { | |
11071 | btf_mod->module = btf_try_get_module(btf); | |
11072 | if (!btf_mod->module) { | |
11073 | err = -ENXIO; | |
11074 | goto err_put; | |
11075 | } | |
11076 | } | |
11077 | ||
11078 | env->used_btf_cnt++; | |
11079 | ||
4976b718 | 11080 | return 0; |
541c3bad AN |
11081 | err_put: |
11082 | btf_put(btf); | |
11083 | return err; | |
4976b718 HL |
11084 | } |
11085 | ||
56f668df MKL |
11086 | static int check_map_prealloc(struct bpf_map *map) |
11087 | { | |
11088 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
11089 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
11090 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
11091 | !(map->map_flags & BPF_F_NO_PREALLOC); |
11092 | } | |
11093 | ||
d83525ca AS |
11094 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
11095 | { | |
11096 | switch (type) { | |
11097 | case BPF_PROG_TYPE_KPROBE: | |
11098 | case BPF_PROG_TYPE_TRACEPOINT: | |
11099 | case BPF_PROG_TYPE_PERF_EVENT: | |
11100 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
11101 | return true; | |
11102 | default: | |
11103 | return false; | |
11104 | } | |
11105 | } | |
11106 | ||
94dacdbd TG |
11107 | static bool is_preallocated_map(struct bpf_map *map) |
11108 | { | |
11109 | if (!check_map_prealloc(map)) | |
11110 | return false; | |
11111 | if (map->inner_map_meta && !check_map_prealloc(map->inner_map_meta)) | |
11112 | return false; | |
11113 | return true; | |
11114 | } | |
11115 | ||
61bd5218 JK |
11116 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
11117 | struct bpf_map *map, | |
fdc15d38 AS |
11118 | struct bpf_prog *prog) |
11119 | ||
11120 | { | |
7e40781c | 11121 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
94dacdbd TG |
11122 | /* |
11123 | * Validate that trace type programs use preallocated hash maps. | |
11124 | * | |
11125 | * For programs attached to PERF events this is mandatory as the | |
11126 | * perf NMI can hit any arbitrary code sequence. | |
11127 | * | |
11128 | * All other trace types using preallocated hash maps are unsafe as | |
11129 | * well because tracepoint or kprobes can be inside locked regions | |
11130 | * of the memory allocator or at a place where a recursion into the | |
11131 | * memory allocator would see inconsistent state. | |
11132 | * | |
2ed905c5 TG |
11133 | * On RT enabled kernels run-time allocation of all trace type |
11134 | * programs is strictly prohibited due to lock type constraints. On | |
11135 | * !RT kernels it is allowed for backwards compatibility reasons for | |
11136 | * now, but warnings are emitted so developers are made aware of | |
11137 | * the unsafety and can fix their programs before this is enforced. | |
56f668df | 11138 | */ |
7e40781c UP |
11139 | if (is_tracing_prog_type(prog_type) && !is_preallocated_map(map)) { |
11140 | if (prog_type == BPF_PROG_TYPE_PERF_EVENT) { | |
61bd5218 | 11141 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
11142 | return -EINVAL; |
11143 | } | |
2ed905c5 TG |
11144 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
11145 | verbose(env, "trace type programs can only use preallocated hash map\n"); | |
11146 | return -EINVAL; | |
11147 | } | |
94dacdbd TG |
11148 | WARN_ONCE(1, "trace type BPF program uses run-time allocation\n"); |
11149 | verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n"); | |
fdc15d38 | 11150 | } |
a3884572 | 11151 | |
9e7a4d98 KS |
11152 | if (map_value_has_spin_lock(map)) { |
11153 | if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) { | |
11154 | verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n"); | |
11155 | return -EINVAL; | |
11156 | } | |
11157 | ||
11158 | if (is_tracing_prog_type(prog_type)) { | |
11159 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
11160 | return -EINVAL; | |
11161 | } | |
11162 | ||
11163 | if (prog->aux->sleepable) { | |
11164 | verbose(env, "sleepable progs cannot use bpf_spin_lock yet\n"); | |
11165 | return -EINVAL; | |
11166 | } | |
d83525ca AS |
11167 | } |
11168 | ||
a3884572 | 11169 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && |
09728266 | 11170 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
11171 | verbose(env, "offload device mismatch between prog and map\n"); |
11172 | return -EINVAL; | |
11173 | } | |
11174 | ||
85d33df3 MKL |
11175 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
11176 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
11177 | return -EINVAL; | |
11178 | } | |
11179 | ||
1e6c62a8 AS |
11180 | if (prog->aux->sleepable) |
11181 | switch (map->map_type) { | |
11182 | case BPF_MAP_TYPE_HASH: | |
11183 | case BPF_MAP_TYPE_LRU_HASH: | |
11184 | case BPF_MAP_TYPE_ARRAY: | |
638e4b82 AS |
11185 | case BPF_MAP_TYPE_PERCPU_HASH: |
11186 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
11187 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
11188 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
11189 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
1e6c62a8 AS |
11190 | if (!is_preallocated_map(map)) { |
11191 | verbose(env, | |
638e4b82 | 11192 | "Sleepable programs can only use preallocated maps\n"); |
1e6c62a8 AS |
11193 | return -EINVAL; |
11194 | } | |
11195 | break; | |
ba90c2cc KS |
11196 | case BPF_MAP_TYPE_RINGBUF: |
11197 | break; | |
1e6c62a8 AS |
11198 | default: |
11199 | verbose(env, | |
ba90c2cc | 11200 | "Sleepable programs can only use array, hash, and ringbuf maps\n"); |
1e6c62a8 AS |
11201 | return -EINVAL; |
11202 | } | |
11203 | ||
fdc15d38 AS |
11204 | return 0; |
11205 | } | |
11206 | ||
b741f163 RG |
11207 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
11208 | { | |
11209 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
11210 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
11211 | } | |
11212 | ||
4976b718 HL |
11213 | /* find and rewrite pseudo imm in ld_imm64 instructions: |
11214 | * | |
11215 | * 1. if it accesses map FD, replace it with actual map pointer. | |
11216 | * 2. if it accesses btf_id of a VAR, replace it with pointer to the var. | |
11217 | * | |
11218 | * NOTE: btf_vmlinux is required for converting pseudo btf_id. | |
0246e64d | 11219 | */ |
4976b718 | 11220 | static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env) |
0246e64d AS |
11221 | { |
11222 | struct bpf_insn *insn = env->prog->insnsi; | |
11223 | int insn_cnt = env->prog->len; | |
fdc15d38 | 11224 | int i, j, err; |
0246e64d | 11225 | |
f1f7714e | 11226 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
11227 | if (err) |
11228 | return err; | |
11229 | ||
0246e64d | 11230 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 11231 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 11232 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 11233 | verbose(env, "BPF_LDX uses reserved fields\n"); |
d691f9e8 AS |
11234 | return -EINVAL; |
11235 | } | |
11236 | ||
0246e64d | 11237 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 11238 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
11239 | struct bpf_map *map; |
11240 | struct fd f; | |
d8eca5bb | 11241 | u64 addr; |
387544bf | 11242 | u32 fd; |
0246e64d AS |
11243 | |
11244 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
11245 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
11246 | insn[1].off != 0) { | |
61bd5218 | 11247 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
11248 | return -EINVAL; |
11249 | } | |
11250 | ||
d8eca5bb | 11251 | if (insn[0].src_reg == 0) |
0246e64d AS |
11252 | /* valid generic load 64-bit imm */ |
11253 | goto next_insn; | |
11254 | ||
4976b718 HL |
11255 | if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) { |
11256 | aux = &env->insn_aux_data[i]; | |
11257 | err = check_pseudo_btf_id(env, insn, aux); | |
11258 | if (err) | |
11259 | return err; | |
11260 | goto next_insn; | |
11261 | } | |
11262 | ||
69c087ba YS |
11263 | if (insn[0].src_reg == BPF_PSEUDO_FUNC) { |
11264 | aux = &env->insn_aux_data[i]; | |
11265 | aux->ptr_type = PTR_TO_FUNC; | |
11266 | goto next_insn; | |
11267 | } | |
11268 | ||
d8eca5bb DB |
11269 | /* In final convert_pseudo_ld_imm64() step, this is |
11270 | * converted into regular 64-bit imm load insn. | |
11271 | */ | |
387544bf AS |
11272 | switch (insn[0].src_reg) { |
11273 | case BPF_PSEUDO_MAP_VALUE: | |
11274 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
11275 | break; | |
11276 | case BPF_PSEUDO_MAP_FD: | |
11277 | case BPF_PSEUDO_MAP_IDX: | |
11278 | if (insn[1].imm == 0) | |
11279 | break; | |
11280 | fallthrough; | |
11281 | default: | |
11282 | verbose(env, "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
11283 | return -EINVAL; |
11284 | } | |
11285 | ||
387544bf AS |
11286 | switch (insn[0].src_reg) { |
11287 | case BPF_PSEUDO_MAP_IDX_VALUE: | |
11288 | case BPF_PSEUDO_MAP_IDX: | |
11289 | if (bpfptr_is_null(env->fd_array)) { | |
11290 | verbose(env, "fd_idx without fd_array is invalid\n"); | |
11291 | return -EPROTO; | |
11292 | } | |
11293 | if (copy_from_bpfptr_offset(&fd, env->fd_array, | |
11294 | insn[0].imm * sizeof(fd), | |
11295 | sizeof(fd))) | |
11296 | return -EFAULT; | |
11297 | break; | |
11298 | default: | |
11299 | fd = insn[0].imm; | |
11300 | break; | |
11301 | } | |
11302 | ||
11303 | f = fdget(fd); | |
c2101297 | 11304 | map = __bpf_map_get(f); |
0246e64d | 11305 | if (IS_ERR(map)) { |
61bd5218 | 11306 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 11307 | insn[0].imm); |
0246e64d AS |
11308 | return PTR_ERR(map); |
11309 | } | |
11310 | ||
61bd5218 | 11311 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
11312 | if (err) { |
11313 | fdput(f); | |
11314 | return err; | |
11315 | } | |
11316 | ||
d8eca5bb | 11317 | aux = &env->insn_aux_data[i]; |
387544bf AS |
11318 | if (insn[0].src_reg == BPF_PSEUDO_MAP_FD || |
11319 | insn[0].src_reg == BPF_PSEUDO_MAP_IDX) { | |
d8eca5bb DB |
11320 | addr = (unsigned long)map; |
11321 | } else { | |
11322 | u32 off = insn[1].imm; | |
11323 | ||
11324 | if (off >= BPF_MAX_VAR_OFF) { | |
11325 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
11326 | fdput(f); | |
11327 | return -EINVAL; | |
11328 | } | |
11329 | ||
11330 | if (!map->ops->map_direct_value_addr) { | |
11331 | verbose(env, "no direct value access support for this map type\n"); | |
11332 | fdput(f); | |
11333 | return -EINVAL; | |
11334 | } | |
11335 | ||
11336 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
11337 | if (err) { | |
11338 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
11339 | map->value_size, off); | |
11340 | fdput(f); | |
11341 | return err; | |
11342 | } | |
11343 | ||
11344 | aux->map_off = off; | |
11345 | addr += off; | |
11346 | } | |
11347 | ||
11348 | insn[0].imm = (u32)addr; | |
11349 | insn[1].imm = addr >> 32; | |
0246e64d AS |
11350 | |
11351 | /* check whether we recorded this map already */ | |
d8eca5bb | 11352 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 11353 | if (env->used_maps[j] == map) { |
d8eca5bb | 11354 | aux->map_index = j; |
0246e64d AS |
11355 | fdput(f); |
11356 | goto next_insn; | |
11357 | } | |
d8eca5bb | 11358 | } |
0246e64d AS |
11359 | |
11360 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
11361 | fdput(f); | |
11362 | return -E2BIG; | |
11363 | } | |
11364 | ||
0246e64d AS |
11365 | /* hold the map. If the program is rejected by verifier, |
11366 | * the map will be released by release_maps() or it | |
11367 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 11368 | * and all maps are released in free_used_maps() |
0246e64d | 11369 | */ |
1e0bd5a0 | 11370 | bpf_map_inc(map); |
d8eca5bb DB |
11371 | |
11372 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
11373 | env->used_maps[env->used_map_cnt++] = map; |
11374 | ||
b741f163 | 11375 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 11376 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 11377 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
11378 | fdput(f); |
11379 | return -EBUSY; | |
11380 | } | |
11381 | ||
0246e64d AS |
11382 | fdput(f); |
11383 | next_insn: | |
11384 | insn++; | |
11385 | i++; | |
5e581dad DB |
11386 | continue; |
11387 | } | |
11388 | ||
11389 | /* Basic sanity check before we invest more work here. */ | |
11390 | if (!bpf_opcode_in_insntable(insn->code)) { | |
11391 | verbose(env, "unknown opcode %02x\n", insn->code); | |
11392 | return -EINVAL; | |
0246e64d AS |
11393 | } |
11394 | } | |
11395 | ||
11396 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
11397 | * 'struct bpf_map *' into a register instead of user map_fd. | |
11398 | * These pointers will be used later by verifier to validate map access. | |
11399 | */ | |
11400 | return 0; | |
11401 | } | |
11402 | ||
11403 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 11404 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 11405 | { |
a2ea0746 DB |
11406 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
11407 | env->used_map_cnt); | |
0246e64d AS |
11408 | } |
11409 | ||
541c3bad AN |
11410 | /* drop refcnt of maps used by the rejected program */ |
11411 | static void release_btfs(struct bpf_verifier_env *env) | |
11412 | { | |
11413 | __bpf_free_used_btfs(env->prog->aux, env->used_btfs, | |
11414 | env->used_btf_cnt); | |
11415 | } | |
11416 | ||
0246e64d | 11417 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ |
58e2af8b | 11418 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
11419 | { |
11420 | struct bpf_insn *insn = env->prog->insnsi; | |
11421 | int insn_cnt = env->prog->len; | |
11422 | int i; | |
11423 | ||
69c087ba YS |
11424 | for (i = 0; i < insn_cnt; i++, insn++) { |
11425 | if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) | |
11426 | continue; | |
11427 | if (insn->src_reg == BPF_PSEUDO_FUNC) | |
11428 | continue; | |
11429 | insn->src_reg = 0; | |
11430 | } | |
0246e64d AS |
11431 | } |
11432 | ||
8041902d AS |
11433 | /* single env->prog->insni[off] instruction was replaced with the range |
11434 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
11435 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
11436 | */ | |
b325fbca JW |
11437 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, |
11438 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d AS |
11439 | { |
11440 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
b325fbca | 11441 | struct bpf_insn *insn = new_prog->insnsi; |
d203b0fd | 11442 | u32 old_seen = old_data[off].seen; |
b325fbca | 11443 | u32 prog_len; |
c131187d | 11444 | int i; |
8041902d | 11445 | |
b325fbca JW |
11446 | /* aux info at OFF always needs adjustment, no matter fast path |
11447 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
11448 | * original insn at old prog. | |
11449 | */ | |
11450 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
11451 | ||
8041902d AS |
11452 | if (cnt == 1) |
11453 | return 0; | |
b325fbca | 11454 | prog_len = new_prog->len; |
fad953ce KC |
11455 | new_data = vzalloc(array_size(prog_len, |
11456 | sizeof(struct bpf_insn_aux_data))); | |
8041902d AS |
11457 | if (!new_data) |
11458 | return -ENOMEM; | |
11459 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
11460 | memcpy(new_data + off + cnt - 1, old_data + off, | |
11461 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 11462 | for (i = off; i < off + cnt - 1; i++) { |
d203b0fd DB |
11463 | /* Expand insni[off]'s seen count to the patched range. */ |
11464 | new_data[i].seen = old_seen; | |
b325fbca JW |
11465 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
11466 | } | |
8041902d AS |
11467 | env->insn_aux_data = new_data; |
11468 | vfree(old_data); | |
11469 | return 0; | |
11470 | } | |
11471 | ||
cc8b0b92 AS |
11472 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
11473 | { | |
11474 | int i; | |
11475 | ||
11476 | if (len == 1) | |
11477 | return; | |
4cb3d99c JW |
11478 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
11479 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 11480 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 11481 | continue; |
9c8105bd | 11482 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
11483 | } |
11484 | } | |
11485 | ||
7506d211 | 11486 | static void adjust_poke_descs(struct bpf_prog *prog, u32 off, u32 len) |
a748c697 MF |
11487 | { |
11488 | struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab; | |
11489 | int i, sz = prog->aux->size_poke_tab; | |
11490 | struct bpf_jit_poke_descriptor *desc; | |
11491 | ||
11492 | for (i = 0; i < sz; i++) { | |
11493 | desc = &tab[i]; | |
7506d211 JF |
11494 | if (desc->insn_idx <= off) |
11495 | continue; | |
a748c697 MF |
11496 | desc->insn_idx += len - 1; |
11497 | } | |
11498 | } | |
11499 | ||
8041902d AS |
11500 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
11501 | const struct bpf_insn *patch, u32 len) | |
11502 | { | |
11503 | struct bpf_prog *new_prog; | |
11504 | ||
11505 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
11506 | if (IS_ERR(new_prog)) { |
11507 | if (PTR_ERR(new_prog) == -ERANGE) | |
11508 | verbose(env, | |
11509 | "insn %d cannot be patched due to 16-bit range\n", | |
11510 | env->insn_aux_data[off].orig_idx); | |
8041902d | 11511 | return NULL; |
4f73379e | 11512 | } |
b325fbca | 11513 | if (adjust_insn_aux_data(env, new_prog, off, len)) |
8041902d | 11514 | return NULL; |
cc8b0b92 | 11515 | adjust_subprog_starts(env, off, len); |
7506d211 | 11516 | adjust_poke_descs(new_prog, off, len); |
8041902d AS |
11517 | return new_prog; |
11518 | } | |
11519 | ||
52875a04 JK |
11520 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
11521 | u32 off, u32 cnt) | |
11522 | { | |
11523 | int i, j; | |
11524 | ||
11525 | /* find first prog starting at or after off (first to remove) */ | |
11526 | for (i = 0; i < env->subprog_cnt; i++) | |
11527 | if (env->subprog_info[i].start >= off) | |
11528 | break; | |
11529 | /* find first prog starting at or after off + cnt (first to stay) */ | |
11530 | for (j = i; j < env->subprog_cnt; j++) | |
11531 | if (env->subprog_info[j].start >= off + cnt) | |
11532 | break; | |
11533 | /* if j doesn't start exactly at off + cnt, we are just removing | |
11534 | * the front of previous prog | |
11535 | */ | |
11536 | if (env->subprog_info[j].start != off + cnt) | |
11537 | j--; | |
11538 | ||
11539 | if (j > i) { | |
11540 | struct bpf_prog_aux *aux = env->prog->aux; | |
11541 | int move; | |
11542 | ||
11543 | /* move fake 'exit' subprog as well */ | |
11544 | move = env->subprog_cnt + 1 - j; | |
11545 | ||
11546 | memmove(env->subprog_info + i, | |
11547 | env->subprog_info + j, | |
11548 | sizeof(*env->subprog_info) * move); | |
11549 | env->subprog_cnt -= j - i; | |
11550 | ||
11551 | /* remove func_info */ | |
11552 | if (aux->func_info) { | |
11553 | move = aux->func_info_cnt - j; | |
11554 | ||
11555 | memmove(aux->func_info + i, | |
11556 | aux->func_info + j, | |
11557 | sizeof(*aux->func_info) * move); | |
11558 | aux->func_info_cnt -= j - i; | |
11559 | /* func_info->insn_off is set after all code rewrites, | |
11560 | * in adjust_btf_func() - no need to adjust | |
11561 | */ | |
11562 | } | |
11563 | } else { | |
11564 | /* convert i from "first prog to remove" to "first to adjust" */ | |
11565 | if (env->subprog_info[i].start == off) | |
11566 | i++; | |
11567 | } | |
11568 | ||
11569 | /* update fake 'exit' subprog as well */ | |
11570 | for (; i <= env->subprog_cnt; i++) | |
11571 | env->subprog_info[i].start -= cnt; | |
11572 | ||
11573 | return 0; | |
11574 | } | |
11575 | ||
11576 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
11577 | u32 cnt) | |
11578 | { | |
11579 | struct bpf_prog *prog = env->prog; | |
11580 | u32 i, l_off, l_cnt, nr_linfo; | |
11581 | struct bpf_line_info *linfo; | |
11582 | ||
11583 | nr_linfo = prog->aux->nr_linfo; | |
11584 | if (!nr_linfo) | |
11585 | return 0; | |
11586 | ||
11587 | linfo = prog->aux->linfo; | |
11588 | ||
11589 | /* find first line info to remove, count lines to be removed */ | |
11590 | for (i = 0; i < nr_linfo; i++) | |
11591 | if (linfo[i].insn_off >= off) | |
11592 | break; | |
11593 | ||
11594 | l_off = i; | |
11595 | l_cnt = 0; | |
11596 | for (; i < nr_linfo; i++) | |
11597 | if (linfo[i].insn_off < off + cnt) | |
11598 | l_cnt++; | |
11599 | else | |
11600 | break; | |
11601 | ||
11602 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
11603 | * last removed linfo. prog is already modified, so prog->len == off | |
11604 | * means no live instructions after (tail of the program was removed). | |
11605 | */ | |
11606 | if (prog->len != off && l_cnt && | |
11607 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
11608 | l_cnt--; | |
11609 | linfo[--i].insn_off = off + cnt; | |
11610 | } | |
11611 | ||
11612 | /* remove the line info which refer to the removed instructions */ | |
11613 | if (l_cnt) { | |
11614 | memmove(linfo + l_off, linfo + i, | |
11615 | sizeof(*linfo) * (nr_linfo - i)); | |
11616 | ||
11617 | prog->aux->nr_linfo -= l_cnt; | |
11618 | nr_linfo = prog->aux->nr_linfo; | |
11619 | } | |
11620 | ||
11621 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
11622 | for (i = l_off; i < nr_linfo; i++) | |
11623 | linfo[i].insn_off -= cnt; | |
11624 | ||
11625 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
11626 | for (i = 0; i <= env->subprog_cnt; i++) | |
11627 | if (env->subprog_info[i].linfo_idx > l_off) { | |
11628 | /* program may have started in the removed region but | |
11629 | * may not be fully removed | |
11630 | */ | |
11631 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
11632 | env->subprog_info[i].linfo_idx -= l_cnt; | |
11633 | else | |
11634 | env->subprog_info[i].linfo_idx = l_off; | |
11635 | } | |
11636 | ||
11637 | return 0; | |
11638 | } | |
11639 | ||
11640 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
11641 | { | |
11642 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11643 | unsigned int orig_prog_len = env->prog->len; | |
11644 | int err; | |
11645 | ||
08ca90af JK |
11646 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
11647 | bpf_prog_offload_remove_insns(env, off, cnt); | |
11648 | ||
52875a04 JK |
11649 | err = bpf_remove_insns(env->prog, off, cnt); |
11650 | if (err) | |
11651 | return err; | |
11652 | ||
11653 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
11654 | if (err) | |
11655 | return err; | |
11656 | ||
11657 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
11658 | if (err) | |
11659 | return err; | |
11660 | ||
11661 | memmove(aux_data + off, aux_data + off + cnt, | |
11662 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
11663 | ||
11664 | return 0; | |
11665 | } | |
11666 | ||
2a5418a1 DB |
11667 | /* The verifier does more data flow analysis than llvm and will not |
11668 | * explore branches that are dead at run time. Malicious programs can | |
11669 | * have dead code too. Therefore replace all dead at-run-time code | |
11670 | * with 'ja -1'. | |
11671 | * | |
11672 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
11673 | * program and through another bug we would manage to jump there, then | |
11674 | * we'd execute beyond program memory otherwise. Returning exception | |
11675 | * code also wouldn't work since we can have subprogs where the dead | |
11676 | * code could be located. | |
c131187d AS |
11677 | */ |
11678 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
11679 | { | |
11680 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 11681 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
11682 | struct bpf_insn *insn = env->prog->insnsi; |
11683 | const int insn_cnt = env->prog->len; | |
11684 | int i; | |
11685 | ||
11686 | for (i = 0; i < insn_cnt; i++) { | |
11687 | if (aux_data[i].seen) | |
11688 | continue; | |
2a5418a1 | 11689 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
11690 | } |
11691 | } | |
11692 | ||
e2ae4ca2 JK |
11693 | static bool insn_is_cond_jump(u8 code) |
11694 | { | |
11695 | u8 op; | |
11696 | ||
092ed096 JW |
11697 | if (BPF_CLASS(code) == BPF_JMP32) |
11698 | return true; | |
11699 | ||
e2ae4ca2 JK |
11700 | if (BPF_CLASS(code) != BPF_JMP) |
11701 | return false; | |
11702 | ||
11703 | op = BPF_OP(code); | |
11704 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
11705 | } | |
11706 | ||
11707 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
11708 | { | |
11709 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11710 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
11711 | struct bpf_insn *insn = env->prog->insnsi; | |
11712 | const int insn_cnt = env->prog->len; | |
11713 | int i; | |
11714 | ||
11715 | for (i = 0; i < insn_cnt; i++, insn++) { | |
11716 | if (!insn_is_cond_jump(insn->code)) | |
11717 | continue; | |
11718 | ||
11719 | if (!aux_data[i + 1].seen) | |
11720 | ja.off = insn->off; | |
11721 | else if (!aux_data[i + 1 + insn->off].seen) | |
11722 | ja.off = 0; | |
11723 | else | |
11724 | continue; | |
11725 | ||
08ca90af JK |
11726 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
11727 | bpf_prog_offload_replace_insn(env, i, &ja); | |
11728 | ||
e2ae4ca2 JK |
11729 | memcpy(insn, &ja, sizeof(ja)); |
11730 | } | |
11731 | } | |
11732 | ||
52875a04 JK |
11733 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
11734 | { | |
11735 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11736 | int insn_cnt = env->prog->len; | |
11737 | int i, err; | |
11738 | ||
11739 | for (i = 0; i < insn_cnt; i++) { | |
11740 | int j; | |
11741 | ||
11742 | j = 0; | |
11743 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
11744 | j++; | |
11745 | if (!j) | |
11746 | continue; | |
11747 | ||
11748 | err = verifier_remove_insns(env, i, j); | |
11749 | if (err) | |
11750 | return err; | |
11751 | insn_cnt = env->prog->len; | |
11752 | } | |
11753 | ||
11754 | return 0; | |
11755 | } | |
11756 | ||
a1b14abc JK |
11757 | static int opt_remove_nops(struct bpf_verifier_env *env) |
11758 | { | |
11759 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
11760 | struct bpf_insn *insn = env->prog->insnsi; | |
11761 | int insn_cnt = env->prog->len; | |
11762 | int i, err; | |
11763 | ||
11764 | for (i = 0; i < insn_cnt; i++) { | |
11765 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
11766 | continue; | |
11767 | ||
11768 | err = verifier_remove_insns(env, i, 1); | |
11769 | if (err) | |
11770 | return err; | |
11771 | insn_cnt--; | |
11772 | i--; | |
11773 | } | |
11774 | ||
11775 | return 0; | |
11776 | } | |
11777 | ||
d6c2308c JW |
11778 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
11779 | const union bpf_attr *attr) | |
a4b1d3c1 | 11780 | { |
d6c2308c | 11781 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 11782 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 11783 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 11784 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 11785 | struct bpf_prog *new_prog; |
d6c2308c | 11786 | bool rnd_hi32; |
a4b1d3c1 | 11787 | |
d6c2308c | 11788 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 11789 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
11790 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
11791 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
11792 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
11793 | for (i = 0; i < len; i++) { |
11794 | int adj_idx = i + delta; | |
11795 | struct bpf_insn insn; | |
83a28819 | 11796 | int load_reg; |
a4b1d3c1 | 11797 | |
d6c2308c | 11798 | insn = insns[adj_idx]; |
83a28819 | 11799 | load_reg = insn_def_regno(&insn); |
d6c2308c JW |
11800 | if (!aux[adj_idx].zext_dst) { |
11801 | u8 code, class; | |
11802 | u32 imm_rnd; | |
11803 | ||
11804 | if (!rnd_hi32) | |
11805 | continue; | |
11806 | ||
11807 | code = insn.code; | |
11808 | class = BPF_CLASS(code); | |
83a28819 | 11809 | if (load_reg == -1) |
d6c2308c JW |
11810 | continue; |
11811 | ||
11812 | /* NOTE: arg "reg" (the fourth one) is only used for | |
83a28819 IL |
11813 | * BPF_STX + SRC_OP, so it is safe to pass NULL |
11814 | * here. | |
d6c2308c | 11815 | */ |
83a28819 | 11816 | if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) { |
d6c2308c JW |
11817 | if (class == BPF_LD && |
11818 | BPF_MODE(code) == BPF_IMM) | |
11819 | i++; | |
11820 | continue; | |
11821 | } | |
11822 | ||
11823 | /* ctx load could be transformed into wider load. */ | |
11824 | if (class == BPF_LDX && | |
11825 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
11826 | continue; | |
11827 | ||
11828 | imm_rnd = get_random_int(); | |
11829 | rnd_hi32_patch[0] = insn; | |
11830 | rnd_hi32_patch[1].imm = imm_rnd; | |
83a28819 | 11831 | rnd_hi32_patch[3].dst_reg = load_reg; |
d6c2308c JW |
11832 | patch = rnd_hi32_patch; |
11833 | patch_len = 4; | |
11834 | goto apply_patch_buffer; | |
11835 | } | |
11836 | ||
39491867 BJ |
11837 | /* Add in an zero-extend instruction if a) the JIT has requested |
11838 | * it or b) it's a CMPXCHG. | |
11839 | * | |
11840 | * The latter is because: BPF_CMPXCHG always loads a value into | |
11841 | * R0, therefore always zero-extends. However some archs' | |
11842 | * equivalent instruction only does this load when the | |
11843 | * comparison is successful. This detail of CMPXCHG is | |
11844 | * orthogonal to the general zero-extension behaviour of the | |
11845 | * CPU, so it's treated independently of bpf_jit_needs_zext. | |
11846 | */ | |
11847 | if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn)) | |
a4b1d3c1 JW |
11848 | continue; |
11849 | ||
83a28819 IL |
11850 | if (WARN_ON(load_reg == -1)) { |
11851 | verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n"); | |
11852 | return -EFAULT; | |
b2e37a71 IL |
11853 | } |
11854 | ||
a4b1d3c1 | 11855 | zext_patch[0] = insn; |
b2e37a71 IL |
11856 | zext_patch[1].dst_reg = load_reg; |
11857 | zext_patch[1].src_reg = load_reg; | |
d6c2308c JW |
11858 | patch = zext_patch; |
11859 | patch_len = 2; | |
11860 | apply_patch_buffer: | |
11861 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
11862 | if (!new_prog) |
11863 | return -ENOMEM; | |
11864 | env->prog = new_prog; | |
11865 | insns = new_prog->insnsi; | |
11866 | aux = env->insn_aux_data; | |
d6c2308c | 11867 | delta += patch_len - 1; |
a4b1d3c1 JW |
11868 | } |
11869 | ||
11870 | return 0; | |
11871 | } | |
11872 | ||
c64b7983 JS |
11873 | /* convert load instructions that access fields of a context type into a |
11874 | * sequence of instructions that access fields of the underlying structure: | |
11875 | * struct __sk_buff -> struct sk_buff | |
11876 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 11877 | */ |
58e2af8b | 11878 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 11879 | { |
00176a34 | 11880 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 11881 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 11882 | const int insn_cnt = env->prog->len; |
36bbef52 | 11883 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 11884 | u32 target_size, size_default, off; |
9bac3d6d | 11885 | struct bpf_prog *new_prog; |
d691f9e8 | 11886 | enum bpf_access_type type; |
f96da094 | 11887 | bool is_narrower_load; |
9bac3d6d | 11888 | |
b09928b9 DB |
11889 | if (ops->gen_prologue || env->seen_direct_write) { |
11890 | if (!ops->gen_prologue) { | |
11891 | verbose(env, "bpf verifier is misconfigured\n"); | |
11892 | return -EINVAL; | |
11893 | } | |
36bbef52 DB |
11894 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
11895 | env->prog); | |
11896 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 11897 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
11898 | return -EINVAL; |
11899 | } else if (cnt) { | |
8041902d | 11900 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
11901 | if (!new_prog) |
11902 | return -ENOMEM; | |
8041902d | 11903 | |
36bbef52 | 11904 | env->prog = new_prog; |
3df126f3 | 11905 | delta += cnt - 1; |
36bbef52 DB |
11906 | } |
11907 | } | |
11908 | ||
c64b7983 | 11909 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
11910 | return 0; |
11911 | ||
3df126f3 | 11912 | insn = env->prog->insnsi + delta; |
36bbef52 | 11913 | |
9bac3d6d | 11914 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 JS |
11915 | bpf_convert_ctx_access_t convert_ctx_access; |
11916 | ||
62c7989b DB |
11917 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
11918 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
11919 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 11920 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 11921 | type = BPF_READ; |
62c7989b DB |
11922 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
11923 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
11924 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 11925 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
11926 | type = BPF_WRITE; |
11927 | else | |
9bac3d6d AS |
11928 | continue; |
11929 | ||
af86ca4e AS |
11930 | if (type == BPF_WRITE && |
11931 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
11932 | struct bpf_insn patch[] = { | |
11933 | /* Sanitize suspicious stack slot with zero. | |
11934 | * There are no memory dependencies for this store, | |
11935 | * since it's only using frame pointer and immediate | |
11936 | * constant of zero | |
11937 | */ | |
11938 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
11939 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
11940 | 0), | |
11941 | /* the original STX instruction will immediately | |
11942 | * overwrite the same stack slot with appropriate value | |
11943 | */ | |
11944 | *insn, | |
11945 | }; | |
11946 | ||
11947 | cnt = ARRAY_SIZE(patch); | |
11948 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
11949 | if (!new_prog) | |
11950 | return -ENOMEM; | |
11951 | ||
11952 | delta += cnt - 1; | |
11953 | env->prog = new_prog; | |
11954 | insn = new_prog->insnsi + i + delta; | |
11955 | continue; | |
11956 | } | |
11957 | ||
c64b7983 JS |
11958 | switch (env->insn_aux_data[i + delta].ptr_type) { |
11959 | case PTR_TO_CTX: | |
11960 | if (!ops->convert_ctx_access) | |
11961 | continue; | |
11962 | convert_ctx_access = ops->convert_ctx_access; | |
11963 | break; | |
11964 | case PTR_TO_SOCKET: | |
46f8bc92 | 11965 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
11966 | convert_ctx_access = bpf_sock_convert_ctx_access; |
11967 | break; | |
655a51e5 MKL |
11968 | case PTR_TO_TCP_SOCK: |
11969 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
11970 | break; | |
fada7fdc JL |
11971 | case PTR_TO_XDP_SOCK: |
11972 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
11973 | break; | |
2a02759e | 11974 | case PTR_TO_BTF_ID: |
27ae7997 MKL |
11975 | if (type == BPF_READ) { |
11976 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
11977 | BPF_SIZE((insn)->code); | |
11978 | env->prog->aux->num_exentries++; | |
7e40781c | 11979 | } else if (resolve_prog_type(env->prog) != BPF_PROG_TYPE_STRUCT_OPS) { |
2a02759e AS |
11980 | verbose(env, "Writes through BTF pointers are not allowed\n"); |
11981 | return -EINVAL; | |
11982 | } | |
2a02759e | 11983 | continue; |
c64b7983 | 11984 | default: |
9bac3d6d | 11985 | continue; |
c64b7983 | 11986 | } |
9bac3d6d | 11987 | |
31fd8581 | 11988 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 11989 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
11990 | |
11991 | /* If the read access is a narrower load of the field, | |
11992 | * convert to a 4/8-byte load, to minimum program type specific | |
11993 | * convert_ctx_access changes. If conversion is successful, | |
11994 | * we will apply proper mask to the result. | |
11995 | */ | |
f96da094 | 11996 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
11997 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
11998 | off = insn->off; | |
31fd8581 | 11999 | if (is_narrower_load) { |
f96da094 DB |
12000 | u8 size_code; |
12001 | ||
12002 | if (type == BPF_WRITE) { | |
61bd5218 | 12003 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
12004 | return -EINVAL; |
12005 | } | |
31fd8581 | 12006 | |
f96da094 | 12007 | size_code = BPF_H; |
31fd8581 YS |
12008 | if (ctx_field_size == 4) |
12009 | size_code = BPF_W; | |
12010 | else if (ctx_field_size == 8) | |
12011 | size_code = BPF_DW; | |
f96da094 | 12012 | |
bc23105c | 12013 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
12014 | insn->code = BPF_LDX | BPF_MEM | size_code; |
12015 | } | |
f96da094 DB |
12016 | |
12017 | target_size = 0; | |
c64b7983 JS |
12018 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
12019 | &target_size); | |
f96da094 DB |
12020 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
12021 | (ctx_field_size && !target_size)) { | |
61bd5218 | 12022 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
12023 | return -EINVAL; |
12024 | } | |
f96da094 DB |
12025 | |
12026 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
12027 | u8 shift = bpf_ctx_narrow_access_offset( |
12028 | off, size, size_default) * 8; | |
46f53a65 AI |
12029 | if (ctx_field_size <= 4) { |
12030 | if (shift) | |
12031 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
12032 | insn->dst_reg, | |
12033 | shift); | |
31fd8581 | 12034 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 12035 | (1 << size * 8) - 1); |
46f53a65 AI |
12036 | } else { |
12037 | if (shift) | |
12038 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
12039 | insn->dst_reg, | |
12040 | shift); | |
31fd8581 | 12041 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 12042 | (1ULL << size * 8) - 1); |
46f53a65 | 12043 | } |
31fd8581 | 12044 | } |
9bac3d6d | 12045 | |
8041902d | 12046 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
12047 | if (!new_prog) |
12048 | return -ENOMEM; | |
12049 | ||
3df126f3 | 12050 | delta += cnt - 1; |
9bac3d6d AS |
12051 | |
12052 | /* keep walking new program and skip insns we just inserted */ | |
12053 | env->prog = new_prog; | |
3df126f3 | 12054 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
12055 | } |
12056 | ||
12057 | return 0; | |
12058 | } | |
12059 | ||
1c2a088a AS |
12060 | static int jit_subprogs(struct bpf_verifier_env *env) |
12061 | { | |
12062 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
12063 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
a748c697 | 12064 | struct bpf_map *map_ptr; |
7105e828 | 12065 | struct bpf_insn *insn; |
1c2a088a | 12066 | void *old_bpf_func; |
c4c0bdc0 | 12067 | int err, num_exentries; |
1c2a088a | 12068 | |
f910cefa | 12069 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
12070 | return 0; |
12071 | ||
7105e828 | 12072 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
69c087ba YS |
12073 | if (bpf_pseudo_func(insn)) { |
12074 | env->insn_aux_data[i].call_imm = insn->imm; | |
12075 | /* subprog is encoded in insn[1].imm */ | |
12076 | continue; | |
12077 | } | |
12078 | ||
23a2d70c | 12079 | if (!bpf_pseudo_call(insn)) |
1c2a088a | 12080 | continue; |
c7a89784 DB |
12081 | /* Upon error here we cannot fall back to interpreter but |
12082 | * need a hard reject of the program. Thus -EFAULT is | |
12083 | * propagated in any case. | |
12084 | */ | |
1c2a088a AS |
12085 | subprog = find_subprog(env, i + insn->imm + 1); |
12086 | if (subprog < 0) { | |
12087 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
12088 | i + insn->imm + 1); | |
12089 | return -EFAULT; | |
12090 | } | |
12091 | /* temporarily remember subprog id inside insn instead of | |
12092 | * aux_data, since next loop will split up all insns into funcs | |
12093 | */ | |
f910cefa | 12094 | insn->off = subprog; |
1c2a088a AS |
12095 | /* remember original imm in case JIT fails and fallback |
12096 | * to interpreter will be needed | |
12097 | */ | |
12098 | env->insn_aux_data[i].call_imm = insn->imm; | |
12099 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
12100 | insn->imm = 1; | |
12101 | } | |
12102 | ||
c454a46b MKL |
12103 | err = bpf_prog_alloc_jited_linfo(prog); |
12104 | if (err) | |
12105 | goto out_undo_insn; | |
12106 | ||
12107 | err = -ENOMEM; | |
6396bb22 | 12108 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 12109 | if (!func) |
c7a89784 | 12110 | goto out_undo_insn; |
1c2a088a | 12111 | |
f910cefa | 12112 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 12113 | subprog_start = subprog_end; |
4cb3d99c | 12114 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
12115 | |
12116 | len = subprog_end - subprog_start; | |
492ecee8 AS |
12117 | /* BPF_PROG_RUN doesn't call subprogs directly, |
12118 | * hence main prog stats include the runtime of subprogs. | |
12119 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
700d4796 | 12120 | * func[i]->stats will never be accessed and stays NULL |
492ecee8 AS |
12121 | */ |
12122 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
12123 | if (!func[i]) |
12124 | goto out_free; | |
12125 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
12126 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 12127 | func[i]->type = prog->type; |
1c2a088a | 12128 | func[i]->len = len; |
4f74d809 DB |
12129 | if (bpf_prog_calc_tag(func[i])) |
12130 | goto out_free; | |
1c2a088a | 12131 | func[i]->is_func = 1; |
ba64e7d8 | 12132 | func[i]->aux->func_idx = i; |
f263a814 | 12133 | /* Below members will be freed only at prog->aux */ |
ba64e7d8 YS |
12134 | func[i]->aux->btf = prog->aux->btf; |
12135 | func[i]->aux->func_info = prog->aux->func_info; | |
f263a814 JF |
12136 | func[i]->aux->poke_tab = prog->aux->poke_tab; |
12137 | func[i]->aux->size_poke_tab = prog->aux->size_poke_tab; | |
ba64e7d8 | 12138 | |
a748c697 | 12139 | for (j = 0; j < prog->aux->size_poke_tab; j++) { |
f263a814 | 12140 | struct bpf_jit_poke_descriptor *poke; |
a748c697 | 12141 | |
f263a814 JF |
12142 | poke = &prog->aux->poke_tab[j]; |
12143 | if (poke->insn_idx < subprog_end && | |
12144 | poke->insn_idx >= subprog_start) | |
12145 | poke->aux = func[i]->aux; | |
a748c697 MF |
12146 | } |
12147 | ||
1c2a088a AS |
12148 | /* Use bpf_prog_F_tag to indicate functions in stack traces. |
12149 | * Long term would need debug info to populate names | |
12150 | */ | |
12151 | func[i]->aux->name[0] = 'F'; | |
9c8105bd | 12152 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 12153 | func[i]->jit_requested = 1; |
e6ac2450 | 12154 | func[i]->aux->kfunc_tab = prog->aux->kfunc_tab; |
c454a46b MKL |
12155 | func[i]->aux->linfo = prog->aux->linfo; |
12156 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
12157 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
12158 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
c4c0bdc0 YS |
12159 | num_exentries = 0; |
12160 | insn = func[i]->insnsi; | |
12161 | for (j = 0; j < func[i]->len; j++, insn++) { | |
12162 | if (BPF_CLASS(insn->code) == BPF_LDX && | |
12163 | BPF_MODE(insn->code) == BPF_PROBE_MEM) | |
12164 | num_exentries++; | |
12165 | } | |
12166 | func[i]->aux->num_exentries = num_exentries; | |
ebf7d1f5 | 12167 | func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable; |
1c2a088a AS |
12168 | func[i] = bpf_int_jit_compile(func[i]); |
12169 | if (!func[i]->jited) { | |
12170 | err = -ENOTSUPP; | |
12171 | goto out_free; | |
12172 | } | |
12173 | cond_resched(); | |
12174 | } | |
a748c697 | 12175 | |
1c2a088a AS |
12176 | /* at this point all bpf functions were successfully JITed |
12177 | * now populate all bpf_calls with correct addresses and | |
12178 | * run last pass of JIT | |
12179 | */ | |
f910cefa | 12180 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
12181 | insn = func[i]->insnsi; |
12182 | for (j = 0; j < func[i]->len; j++, insn++) { | |
69c087ba YS |
12183 | if (bpf_pseudo_func(insn)) { |
12184 | subprog = insn[1].imm; | |
12185 | insn[0].imm = (u32)(long)func[subprog]->bpf_func; | |
12186 | insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32; | |
12187 | continue; | |
12188 | } | |
23a2d70c | 12189 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
12190 | continue; |
12191 | subprog = insn->off; | |
0d306c31 PB |
12192 | insn->imm = BPF_CAST_CALL(func[subprog]->bpf_func) - |
12193 | __bpf_call_base; | |
1c2a088a | 12194 | } |
2162fed4 SD |
12195 | |
12196 | /* we use the aux data to keep a list of the start addresses | |
12197 | * of the JITed images for each function in the program | |
12198 | * | |
12199 | * for some architectures, such as powerpc64, the imm field | |
12200 | * might not be large enough to hold the offset of the start | |
12201 | * address of the callee's JITed image from __bpf_call_base | |
12202 | * | |
12203 | * in such cases, we can lookup the start address of a callee | |
12204 | * by using its subprog id, available from the off field of | |
12205 | * the call instruction, as an index for this list | |
12206 | */ | |
12207 | func[i]->aux->func = func; | |
12208 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 12209 | } |
f910cefa | 12210 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
12211 | old_bpf_func = func[i]->bpf_func; |
12212 | tmp = bpf_int_jit_compile(func[i]); | |
12213 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
12214 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 12215 | err = -ENOTSUPP; |
1c2a088a AS |
12216 | goto out_free; |
12217 | } | |
12218 | cond_resched(); | |
12219 | } | |
12220 | ||
12221 | /* finally lock prog and jit images for all functions and | |
12222 | * populate kallsysm | |
12223 | */ | |
f910cefa | 12224 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
12225 | bpf_prog_lock_ro(func[i]); |
12226 | bpf_prog_kallsyms_add(func[i]); | |
12227 | } | |
7105e828 DB |
12228 | |
12229 | /* Last step: make now unused interpreter insns from main | |
12230 | * prog consistent for later dump requests, so they can | |
12231 | * later look the same as if they were interpreted only. | |
12232 | */ | |
12233 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
69c087ba YS |
12234 | if (bpf_pseudo_func(insn)) { |
12235 | insn[0].imm = env->insn_aux_data[i].call_imm; | |
12236 | insn[1].imm = find_subprog(env, i + insn[0].imm + 1); | |
12237 | continue; | |
12238 | } | |
23a2d70c | 12239 | if (!bpf_pseudo_call(insn)) |
7105e828 DB |
12240 | continue; |
12241 | insn->off = env->insn_aux_data[i].call_imm; | |
12242 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 12243 | insn->imm = subprog; |
7105e828 DB |
12244 | } |
12245 | ||
1c2a088a AS |
12246 | prog->jited = 1; |
12247 | prog->bpf_func = func[0]->bpf_func; | |
12248 | prog->aux->func = func; | |
f910cefa | 12249 | prog->aux->func_cnt = env->subprog_cnt; |
e16301fb | 12250 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
12251 | return 0; |
12252 | out_free: | |
f263a814 JF |
12253 | /* We failed JIT'ing, so at this point we need to unregister poke |
12254 | * descriptors from subprogs, so that kernel is not attempting to | |
12255 | * patch it anymore as we're freeing the subprog JIT memory. | |
12256 | */ | |
12257 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
12258 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
12259 | map_ptr->ops->map_poke_untrack(map_ptr, prog->aux); | |
12260 | } | |
12261 | /* At this point we're guaranteed that poke descriptors are not | |
12262 | * live anymore. We can just unlink its descriptor table as it's | |
12263 | * released with the main prog. | |
12264 | */ | |
a748c697 MF |
12265 | for (i = 0; i < env->subprog_cnt; i++) { |
12266 | if (!func[i]) | |
12267 | continue; | |
f263a814 | 12268 | func[i]->aux->poke_tab = NULL; |
a748c697 MF |
12269 | bpf_jit_free(func[i]); |
12270 | } | |
1c2a088a | 12271 | kfree(func); |
c7a89784 | 12272 | out_undo_insn: |
1c2a088a AS |
12273 | /* cleanup main prog to be interpreted */ |
12274 | prog->jit_requested = 0; | |
12275 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
23a2d70c | 12276 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
12277 | continue; |
12278 | insn->off = 0; | |
12279 | insn->imm = env->insn_aux_data[i].call_imm; | |
12280 | } | |
e16301fb | 12281 | bpf_prog_jit_attempt_done(prog); |
1c2a088a AS |
12282 | return err; |
12283 | } | |
12284 | ||
1ea47e01 AS |
12285 | static int fixup_call_args(struct bpf_verifier_env *env) |
12286 | { | |
19d28fbd | 12287 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
12288 | struct bpf_prog *prog = env->prog; |
12289 | struct bpf_insn *insn = prog->insnsi; | |
e6ac2450 | 12290 | bool has_kfunc_call = bpf_prog_has_kfunc_call(prog); |
1ea47e01 | 12291 | int i, depth; |
19d28fbd | 12292 | #endif |
e4052d06 | 12293 | int err = 0; |
1ea47e01 | 12294 | |
e4052d06 QM |
12295 | if (env->prog->jit_requested && |
12296 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
12297 | err = jit_subprogs(env); |
12298 | if (err == 0) | |
1c2a088a | 12299 | return 0; |
c7a89784 DB |
12300 | if (err == -EFAULT) |
12301 | return err; | |
19d28fbd DM |
12302 | } |
12303 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
e6ac2450 MKL |
12304 | if (has_kfunc_call) { |
12305 | verbose(env, "calling kernel functions are not allowed in non-JITed programs\n"); | |
12306 | return -EINVAL; | |
12307 | } | |
e411901c MF |
12308 | if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) { |
12309 | /* When JIT fails the progs with bpf2bpf calls and tail_calls | |
12310 | * have to be rejected, since interpreter doesn't support them yet. | |
12311 | */ | |
12312 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
12313 | return -EINVAL; | |
12314 | } | |
1ea47e01 | 12315 | for (i = 0; i < prog->len; i++, insn++) { |
69c087ba YS |
12316 | if (bpf_pseudo_func(insn)) { |
12317 | /* When JIT fails the progs with callback calls | |
12318 | * have to be rejected, since interpreter doesn't support them yet. | |
12319 | */ | |
12320 | verbose(env, "callbacks are not allowed in non-JITed programs\n"); | |
12321 | return -EINVAL; | |
12322 | } | |
12323 | ||
23a2d70c | 12324 | if (!bpf_pseudo_call(insn)) |
1ea47e01 AS |
12325 | continue; |
12326 | depth = get_callee_stack_depth(env, insn, i); | |
12327 | if (depth < 0) | |
12328 | return depth; | |
12329 | bpf_patch_call_args(insn, depth); | |
12330 | } | |
19d28fbd DM |
12331 | err = 0; |
12332 | #endif | |
12333 | return err; | |
1ea47e01 AS |
12334 | } |
12335 | ||
e6ac2450 MKL |
12336 | static int fixup_kfunc_call(struct bpf_verifier_env *env, |
12337 | struct bpf_insn *insn) | |
12338 | { | |
12339 | const struct bpf_kfunc_desc *desc; | |
12340 | ||
12341 | /* insn->imm has the btf func_id. Replace it with | |
12342 | * an address (relative to __bpf_base_call). | |
12343 | */ | |
12344 | desc = find_kfunc_desc(env->prog, insn->imm); | |
12345 | if (!desc) { | |
12346 | verbose(env, "verifier internal error: kernel function descriptor not found for func_id %u\n", | |
12347 | insn->imm); | |
12348 | return -EFAULT; | |
12349 | } | |
12350 | ||
12351 | insn->imm = desc->imm; | |
12352 | ||
12353 | return 0; | |
12354 | } | |
12355 | ||
e6ac5933 BJ |
12356 | /* Do various post-verification rewrites in a single program pass. |
12357 | * These rewrites simplify JIT and interpreter implementations. | |
e245c5c6 | 12358 | */ |
e6ac5933 | 12359 | static int do_misc_fixups(struct bpf_verifier_env *env) |
e245c5c6 | 12360 | { |
79741b3b | 12361 | struct bpf_prog *prog = env->prog; |
d2e4c1e6 | 12362 | bool expect_blinding = bpf_jit_blinding_enabled(prog); |
79741b3b | 12363 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 12364 | const struct bpf_func_proto *fn; |
79741b3b | 12365 | const int insn_cnt = prog->len; |
09772d92 | 12366 | const struct bpf_map_ops *ops; |
c93552c4 | 12367 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
12368 | struct bpf_insn insn_buf[16]; |
12369 | struct bpf_prog *new_prog; | |
12370 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 12371 | int i, ret, cnt, delta = 0; |
e245c5c6 | 12372 | |
79741b3b | 12373 | for (i = 0; i < insn_cnt; i++, insn++) { |
e6ac5933 | 12374 | /* Make divide-by-zero exceptions impossible. */ |
f6b1b3bf DB |
12375 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
12376 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
12377 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 12378 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf | 12379 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
e88b2c6e DB |
12380 | bool isdiv = BPF_OP(insn->code) == BPF_DIV; |
12381 | struct bpf_insn *patchlet; | |
12382 | struct bpf_insn chk_and_div[] = { | |
9b00f1b7 | 12383 | /* [R,W]x div 0 -> 0 */ |
e88b2c6e DB |
12384 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
12385 | BPF_JNE | BPF_K, insn->src_reg, | |
12386 | 0, 2, 0), | |
f6b1b3bf DB |
12387 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), |
12388 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
12389 | *insn, | |
12390 | }; | |
e88b2c6e | 12391 | struct bpf_insn chk_and_mod[] = { |
9b00f1b7 | 12392 | /* [R,W]x mod 0 -> [R,W]x */ |
e88b2c6e DB |
12393 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
12394 | BPF_JEQ | BPF_K, insn->src_reg, | |
9b00f1b7 | 12395 | 0, 1 + (is64 ? 0 : 1), 0), |
f6b1b3bf | 12396 | *insn, |
9b00f1b7 DB |
12397 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), |
12398 | BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), | |
f6b1b3bf | 12399 | }; |
f6b1b3bf | 12400 | |
e88b2c6e DB |
12401 | patchlet = isdiv ? chk_and_div : chk_and_mod; |
12402 | cnt = isdiv ? ARRAY_SIZE(chk_and_div) : | |
9b00f1b7 | 12403 | ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0); |
f6b1b3bf DB |
12404 | |
12405 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
12406 | if (!new_prog) |
12407 | return -ENOMEM; | |
12408 | ||
12409 | delta += cnt - 1; | |
12410 | env->prog = prog = new_prog; | |
12411 | insn = new_prog->insnsi + i + delta; | |
12412 | continue; | |
12413 | } | |
12414 | ||
e6ac5933 | 12415 | /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */ |
e0cea7ce DB |
12416 | if (BPF_CLASS(insn->code) == BPF_LD && |
12417 | (BPF_MODE(insn->code) == BPF_ABS || | |
12418 | BPF_MODE(insn->code) == BPF_IND)) { | |
12419 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
12420 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
12421 | verbose(env, "bpf verifier is misconfigured\n"); | |
12422 | return -EINVAL; | |
12423 | } | |
12424 | ||
12425 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
12426 | if (!new_prog) | |
12427 | return -ENOMEM; | |
12428 | ||
12429 | delta += cnt - 1; | |
12430 | env->prog = prog = new_prog; | |
12431 | insn = new_prog->insnsi + i + delta; | |
12432 | continue; | |
12433 | } | |
12434 | ||
e6ac5933 | 12435 | /* Rewrite pointer arithmetic to mitigate speculation attacks. */ |
979d63d5 DB |
12436 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
12437 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
12438 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
12439 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
979d63d5 | 12440 | struct bpf_insn *patch = &insn_buf[0]; |
801c6058 | 12441 | bool issrc, isneg, isimm; |
979d63d5 DB |
12442 | u32 off_reg; |
12443 | ||
12444 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
12445 | if (!aux->alu_state || |
12446 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
12447 | continue; |
12448 | ||
12449 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
12450 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
12451 | BPF_ALU_SANITIZE_SRC; | |
801c6058 | 12452 | isimm = aux->alu_state & BPF_ALU_IMMEDIATE; |
979d63d5 DB |
12453 | |
12454 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
801c6058 DB |
12455 | if (isimm) { |
12456 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
12457 | } else { | |
12458 | if (isneg) | |
12459 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
12460 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); | |
12461 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
12462 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
12463 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
12464 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
12465 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, off_reg); | |
12466 | } | |
b9b34ddb DB |
12467 | if (!issrc) |
12468 | *patch++ = BPF_MOV64_REG(insn->dst_reg, insn->src_reg); | |
12469 | insn->src_reg = BPF_REG_AX; | |
979d63d5 DB |
12470 | if (isneg) |
12471 | insn->code = insn->code == code_add ? | |
12472 | code_sub : code_add; | |
12473 | *patch++ = *insn; | |
801c6058 | 12474 | if (issrc && isneg && !isimm) |
979d63d5 DB |
12475 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); |
12476 | cnt = patch - insn_buf; | |
12477 | ||
12478 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
12479 | if (!new_prog) | |
12480 | return -ENOMEM; | |
12481 | ||
12482 | delta += cnt - 1; | |
12483 | env->prog = prog = new_prog; | |
12484 | insn = new_prog->insnsi + i + delta; | |
12485 | continue; | |
12486 | } | |
12487 | ||
79741b3b AS |
12488 | if (insn->code != (BPF_JMP | BPF_CALL)) |
12489 | continue; | |
cc8b0b92 AS |
12490 | if (insn->src_reg == BPF_PSEUDO_CALL) |
12491 | continue; | |
e6ac2450 MKL |
12492 | if (insn->src_reg == BPF_PSEUDO_KFUNC_CALL) { |
12493 | ret = fixup_kfunc_call(env, insn); | |
12494 | if (ret) | |
12495 | return ret; | |
12496 | continue; | |
12497 | } | |
e245c5c6 | 12498 | |
79741b3b AS |
12499 | if (insn->imm == BPF_FUNC_get_route_realm) |
12500 | prog->dst_needed = 1; | |
12501 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
12502 | bpf_user_rnd_init_once(); | |
9802d865 JB |
12503 | if (insn->imm == BPF_FUNC_override_return) |
12504 | prog->kprobe_override = 1; | |
79741b3b | 12505 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
12506 | /* If we tail call into other programs, we |
12507 | * cannot make any assumptions since they can | |
12508 | * be replaced dynamically during runtime in | |
12509 | * the program array. | |
12510 | */ | |
12511 | prog->cb_access = 1; | |
e411901c MF |
12512 | if (!allow_tail_call_in_subprogs(env)) |
12513 | prog->aux->stack_depth = MAX_BPF_STACK; | |
12514 | prog->aux->max_pkt_offset = MAX_PACKET_OFF; | |
7b9f6da1 | 12515 | |
79741b3b | 12516 | /* mark bpf_tail_call as different opcode to avoid |
8fb33b60 | 12517 | * conditional branch in the interpreter for every normal |
79741b3b AS |
12518 | * call and to prevent accidental JITing by JIT compiler |
12519 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 12520 | */ |
79741b3b | 12521 | insn->imm = 0; |
71189fa9 | 12522 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 12523 | |
c93552c4 | 12524 | aux = &env->insn_aux_data[i + delta]; |
2c78ee89 | 12525 | if (env->bpf_capable && !expect_blinding && |
cc52d914 | 12526 | prog->jit_requested && |
d2e4c1e6 DB |
12527 | !bpf_map_key_poisoned(aux) && |
12528 | !bpf_map_ptr_poisoned(aux) && | |
12529 | !bpf_map_ptr_unpriv(aux)) { | |
12530 | struct bpf_jit_poke_descriptor desc = { | |
12531 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
12532 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
12533 | .tail_call.key = bpf_map_key_immediate(aux), | |
a748c697 | 12534 | .insn_idx = i + delta, |
d2e4c1e6 DB |
12535 | }; |
12536 | ||
12537 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
12538 | if (ret < 0) { | |
12539 | verbose(env, "adding tail call poke descriptor failed\n"); | |
12540 | return ret; | |
12541 | } | |
12542 | ||
12543 | insn->imm = ret + 1; | |
12544 | continue; | |
12545 | } | |
12546 | ||
c93552c4 DB |
12547 | if (!bpf_map_ptr_unpriv(aux)) |
12548 | continue; | |
12549 | ||
b2157399 AS |
12550 | /* instead of changing every JIT dealing with tail_call |
12551 | * emit two extra insns: | |
12552 | * if (index >= max_entries) goto out; | |
12553 | * index &= array->index_mask; | |
12554 | * to avoid out-of-bounds cpu speculation | |
12555 | */ | |
c93552c4 | 12556 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 12557 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
12558 | return -EINVAL; |
12559 | } | |
c93552c4 | 12560 | |
d2e4c1e6 | 12561 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
12562 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
12563 | map_ptr->max_entries, 2); | |
12564 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
12565 | container_of(map_ptr, | |
12566 | struct bpf_array, | |
12567 | map)->index_mask); | |
12568 | insn_buf[2] = *insn; | |
12569 | cnt = 3; | |
12570 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
12571 | if (!new_prog) | |
12572 | return -ENOMEM; | |
12573 | ||
12574 | delta += cnt - 1; | |
12575 | env->prog = prog = new_prog; | |
12576 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
12577 | continue; |
12578 | } | |
e245c5c6 | 12579 | |
89c63074 | 12580 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
12581 | * and other inlining handlers are currently limited to 64 bit |
12582 | * only. | |
89c63074 | 12583 | */ |
60b58afc | 12584 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
12585 | (insn->imm == BPF_FUNC_map_lookup_elem || |
12586 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
12587 | insn->imm == BPF_FUNC_map_delete_elem || |
12588 | insn->imm == BPF_FUNC_map_push_elem || | |
12589 | insn->imm == BPF_FUNC_map_pop_elem || | |
e6a4750f BT |
12590 | insn->imm == BPF_FUNC_map_peek_elem || |
12591 | insn->imm == BPF_FUNC_redirect_map)) { | |
c93552c4 DB |
12592 | aux = &env->insn_aux_data[i + delta]; |
12593 | if (bpf_map_ptr_poisoned(aux)) | |
12594 | goto patch_call_imm; | |
12595 | ||
d2e4c1e6 | 12596 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
12597 | ops = map_ptr->ops; |
12598 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
12599 | ops->map_gen_lookup) { | |
12600 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
4a8f87e6 DB |
12601 | if (cnt == -EOPNOTSUPP) |
12602 | goto patch_map_ops_generic; | |
12603 | if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
09772d92 DB |
12604 | verbose(env, "bpf verifier is misconfigured\n"); |
12605 | return -EINVAL; | |
12606 | } | |
81ed18ab | 12607 | |
09772d92 DB |
12608 | new_prog = bpf_patch_insn_data(env, i + delta, |
12609 | insn_buf, cnt); | |
12610 | if (!new_prog) | |
12611 | return -ENOMEM; | |
81ed18ab | 12612 | |
09772d92 DB |
12613 | delta += cnt - 1; |
12614 | env->prog = prog = new_prog; | |
12615 | insn = new_prog->insnsi + i + delta; | |
12616 | continue; | |
12617 | } | |
81ed18ab | 12618 | |
09772d92 DB |
12619 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
12620 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
12621 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
12622 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
12623 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
12624 | (int (*)(struct bpf_map *map, void *key, void *value, | |
12625 | u64 flags))NULL)); | |
84430d42 DB |
12626 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
12627 | (int (*)(struct bpf_map *map, void *value, | |
12628 | u64 flags))NULL)); | |
12629 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
12630 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
12631 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
12632 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
e6a4750f BT |
12633 | BUILD_BUG_ON(!__same_type(ops->map_redirect, |
12634 | (int (*)(struct bpf_map *map, u32 ifindex, u64 flags))NULL)); | |
12635 | ||
4a8f87e6 | 12636 | patch_map_ops_generic: |
09772d92 DB |
12637 | switch (insn->imm) { |
12638 | case BPF_FUNC_map_lookup_elem: | |
12639 | insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) - | |
12640 | __bpf_call_base; | |
12641 | continue; | |
12642 | case BPF_FUNC_map_update_elem: | |
12643 | insn->imm = BPF_CAST_CALL(ops->map_update_elem) - | |
12644 | __bpf_call_base; | |
12645 | continue; | |
12646 | case BPF_FUNC_map_delete_elem: | |
12647 | insn->imm = BPF_CAST_CALL(ops->map_delete_elem) - | |
12648 | __bpf_call_base; | |
12649 | continue; | |
84430d42 DB |
12650 | case BPF_FUNC_map_push_elem: |
12651 | insn->imm = BPF_CAST_CALL(ops->map_push_elem) - | |
12652 | __bpf_call_base; | |
12653 | continue; | |
12654 | case BPF_FUNC_map_pop_elem: | |
12655 | insn->imm = BPF_CAST_CALL(ops->map_pop_elem) - | |
12656 | __bpf_call_base; | |
12657 | continue; | |
12658 | case BPF_FUNC_map_peek_elem: | |
12659 | insn->imm = BPF_CAST_CALL(ops->map_peek_elem) - | |
12660 | __bpf_call_base; | |
12661 | continue; | |
e6a4750f BT |
12662 | case BPF_FUNC_redirect_map: |
12663 | insn->imm = BPF_CAST_CALL(ops->map_redirect) - | |
12664 | __bpf_call_base; | |
12665 | continue; | |
09772d92 | 12666 | } |
81ed18ab | 12667 | |
09772d92 | 12668 | goto patch_call_imm; |
81ed18ab AS |
12669 | } |
12670 | ||
e6ac5933 | 12671 | /* Implement bpf_jiffies64 inline. */ |
5576b991 MKL |
12672 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
12673 | insn->imm == BPF_FUNC_jiffies64) { | |
12674 | struct bpf_insn ld_jiffies_addr[2] = { | |
12675 | BPF_LD_IMM64(BPF_REG_0, | |
12676 | (unsigned long)&jiffies), | |
12677 | }; | |
12678 | ||
12679 | insn_buf[0] = ld_jiffies_addr[0]; | |
12680 | insn_buf[1] = ld_jiffies_addr[1]; | |
12681 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
12682 | BPF_REG_0, 0); | |
12683 | cnt = 3; | |
12684 | ||
12685 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
12686 | cnt); | |
12687 | if (!new_prog) | |
12688 | return -ENOMEM; | |
12689 | ||
12690 | delta += cnt - 1; | |
12691 | env->prog = prog = new_prog; | |
12692 | insn = new_prog->insnsi + i + delta; | |
12693 | continue; | |
12694 | } | |
12695 | ||
81ed18ab | 12696 | patch_call_imm: |
5e43f899 | 12697 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
12698 | /* all functions that have prototype and verifier allowed |
12699 | * programs to call them, must be real in-kernel functions | |
12700 | */ | |
12701 | if (!fn->func) { | |
61bd5218 JK |
12702 | verbose(env, |
12703 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
12704 | func_id_name(insn->imm), insn->imm); |
12705 | return -EFAULT; | |
e245c5c6 | 12706 | } |
79741b3b | 12707 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 12708 | } |
e245c5c6 | 12709 | |
d2e4c1e6 DB |
12710 | /* Since poke tab is now finalized, publish aux to tracker. */ |
12711 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
12712 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
12713 | if (!map_ptr->ops->map_poke_track || | |
12714 | !map_ptr->ops->map_poke_untrack || | |
12715 | !map_ptr->ops->map_poke_run) { | |
12716 | verbose(env, "bpf verifier is misconfigured\n"); | |
12717 | return -EINVAL; | |
12718 | } | |
12719 | ||
12720 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
12721 | if (ret < 0) { | |
12722 | verbose(env, "tracking tail call prog failed\n"); | |
12723 | return ret; | |
12724 | } | |
12725 | } | |
12726 | ||
e6ac2450 MKL |
12727 | sort_kfunc_descs_by_imm(env->prog); |
12728 | ||
79741b3b AS |
12729 | return 0; |
12730 | } | |
e245c5c6 | 12731 | |
58e2af8b | 12732 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 12733 | { |
58e2af8b | 12734 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
12735 | int i; |
12736 | ||
9f4686c4 AS |
12737 | sl = env->free_list; |
12738 | while (sl) { | |
12739 | sln = sl->next; | |
12740 | free_verifier_state(&sl->state, false); | |
12741 | kfree(sl); | |
12742 | sl = sln; | |
12743 | } | |
51c39bb1 | 12744 | env->free_list = NULL; |
9f4686c4 | 12745 | |
f1bca824 AS |
12746 | if (!env->explored_states) |
12747 | return; | |
12748 | ||
dc2a4ebc | 12749 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
12750 | sl = env->explored_states[i]; |
12751 | ||
a8f500af AS |
12752 | while (sl) { |
12753 | sln = sl->next; | |
12754 | free_verifier_state(&sl->state, false); | |
12755 | kfree(sl); | |
12756 | sl = sln; | |
12757 | } | |
51c39bb1 | 12758 | env->explored_states[i] = NULL; |
f1bca824 | 12759 | } |
51c39bb1 | 12760 | } |
f1bca824 | 12761 | |
51c39bb1 AS |
12762 | static int do_check_common(struct bpf_verifier_env *env, int subprog) |
12763 | { | |
6f8a57cc | 12764 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
12765 | struct bpf_verifier_state *state; |
12766 | struct bpf_reg_state *regs; | |
12767 | int ret, i; | |
12768 | ||
12769 | env->prev_linfo = NULL; | |
12770 | env->pass_cnt++; | |
12771 | ||
12772 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
12773 | if (!state) | |
12774 | return -ENOMEM; | |
12775 | state->curframe = 0; | |
12776 | state->speculative = false; | |
12777 | state->branches = 1; | |
12778 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
12779 | if (!state->frame[0]) { | |
12780 | kfree(state); | |
12781 | return -ENOMEM; | |
12782 | } | |
12783 | env->cur_state = state; | |
12784 | init_func_state(env, state->frame[0], | |
12785 | BPF_MAIN_FUNC /* callsite */, | |
12786 | 0 /* frameno */, | |
12787 | subprog); | |
12788 | ||
12789 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 12790 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
51c39bb1 AS |
12791 | ret = btf_prepare_func_args(env, subprog, regs); |
12792 | if (ret) | |
12793 | goto out; | |
12794 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
12795 | if (regs[i].type == PTR_TO_CTX) | |
12796 | mark_reg_known_zero(env, regs, i); | |
12797 | else if (regs[i].type == SCALAR_VALUE) | |
12798 | mark_reg_unknown(env, regs, i); | |
e5069b9c DB |
12799 | else if (regs[i].type == PTR_TO_MEM_OR_NULL) { |
12800 | const u32 mem_size = regs[i].mem_size; | |
12801 | ||
12802 | mark_reg_known_zero(env, regs, i); | |
12803 | regs[i].mem_size = mem_size; | |
12804 | regs[i].id = ++env->id_gen; | |
12805 | } | |
51c39bb1 AS |
12806 | } |
12807 | } else { | |
12808 | /* 1st arg to a function */ | |
12809 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
12810 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
34747c41 | 12811 | ret = btf_check_subprog_arg_match(env, subprog, regs); |
51c39bb1 AS |
12812 | if (ret == -EFAULT) |
12813 | /* unlikely verifier bug. abort. | |
12814 | * ret == 0 and ret < 0 are sadly acceptable for | |
12815 | * main() function due to backward compatibility. | |
12816 | * Like socket filter program may be written as: | |
12817 | * int bpf_prog(struct pt_regs *ctx) | |
12818 | * and never dereference that ctx in the program. | |
12819 | * 'struct pt_regs' is a type mismatch for socket | |
12820 | * filter that should be using 'struct __sk_buff'. | |
12821 | */ | |
12822 | goto out; | |
12823 | } | |
12824 | ||
12825 | ret = do_check(env); | |
12826 | out: | |
f59bbfc2 AS |
12827 | /* check for NULL is necessary, since cur_state can be freed inside |
12828 | * do_check() under memory pressure. | |
12829 | */ | |
12830 | if (env->cur_state) { | |
12831 | free_verifier_state(env->cur_state, true); | |
12832 | env->cur_state = NULL; | |
12833 | } | |
6f8a57cc AN |
12834 | while (!pop_stack(env, NULL, NULL, false)); |
12835 | if (!ret && pop_log) | |
12836 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 | 12837 | free_states(env); |
51c39bb1 AS |
12838 | return ret; |
12839 | } | |
12840 | ||
12841 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
12842 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
12843 | * Consider: | |
12844 | * int bar(int); | |
12845 | * int foo(int f) | |
12846 | * { | |
12847 | * return bar(f); | |
12848 | * } | |
12849 | * int bar(int b) | |
12850 | * { | |
12851 | * ... | |
12852 | * } | |
12853 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
12854 | * will be assumed that bar() already verified successfully and call to bar() | |
12855 | * from foo() will be checked for type match only. Later bar() will be verified | |
12856 | * independently to check that it's safe for R1=any_scalar_value. | |
12857 | */ | |
12858 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
12859 | { | |
12860 | struct bpf_prog_aux *aux = env->prog->aux; | |
12861 | int i, ret; | |
12862 | ||
12863 | if (!aux->func_info) | |
12864 | return 0; | |
12865 | ||
12866 | for (i = 1; i < env->subprog_cnt; i++) { | |
12867 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
12868 | continue; | |
12869 | env->insn_idx = env->subprog_info[i].start; | |
12870 | WARN_ON_ONCE(env->insn_idx == 0); | |
12871 | ret = do_check_common(env, i); | |
12872 | if (ret) { | |
12873 | return ret; | |
12874 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
12875 | verbose(env, | |
12876 | "Func#%d is safe for any args that match its prototype\n", | |
12877 | i); | |
12878 | } | |
12879 | } | |
12880 | return 0; | |
12881 | } | |
12882 | ||
12883 | static int do_check_main(struct bpf_verifier_env *env) | |
12884 | { | |
12885 | int ret; | |
12886 | ||
12887 | env->insn_idx = 0; | |
12888 | ret = do_check_common(env, 0); | |
12889 | if (!ret) | |
12890 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
12891 | return ret; | |
12892 | } | |
12893 | ||
12894 | ||
06ee7115 AS |
12895 | static void print_verification_stats(struct bpf_verifier_env *env) |
12896 | { | |
12897 | int i; | |
12898 | ||
12899 | if (env->log.level & BPF_LOG_STATS) { | |
12900 | verbose(env, "verification time %lld usec\n", | |
12901 | div_u64(env->verification_time, 1000)); | |
12902 | verbose(env, "stack depth "); | |
12903 | for (i = 0; i < env->subprog_cnt; i++) { | |
12904 | u32 depth = env->subprog_info[i].stack_depth; | |
12905 | ||
12906 | verbose(env, "%d", depth); | |
12907 | if (i + 1 < env->subprog_cnt) | |
12908 | verbose(env, "+"); | |
12909 | } | |
12910 | verbose(env, "\n"); | |
12911 | } | |
12912 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
12913 | "total_states %d peak_states %d mark_read %d\n", | |
12914 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
12915 | env->max_states_per_insn, env->total_states, | |
12916 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
12917 | } |
12918 | ||
27ae7997 MKL |
12919 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
12920 | { | |
12921 | const struct btf_type *t, *func_proto; | |
12922 | const struct bpf_struct_ops *st_ops; | |
12923 | const struct btf_member *member; | |
12924 | struct bpf_prog *prog = env->prog; | |
12925 | u32 btf_id, member_idx; | |
12926 | const char *mname; | |
12927 | ||
12aa8a94 THJ |
12928 | if (!prog->gpl_compatible) { |
12929 | verbose(env, "struct ops programs must have a GPL compatible license\n"); | |
12930 | return -EINVAL; | |
12931 | } | |
12932 | ||
27ae7997 MKL |
12933 | btf_id = prog->aux->attach_btf_id; |
12934 | st_ops = bpf_struct_ops_find(btf_id); | |
12935 | if (!st_ops) { | |
12936 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
12937 | btf_id); | |
12938 | return -ENOTSUPP; | |
12939 | } | |
12940 | ||
12941 | t = st_ops->type; | |
12942 | member_idx = prog->expected_attach_type; | |
12943 | if (member_idx >= btf_type_vlen(t)) { | |
12944 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
12945 | member_idx, st_ops->name); | |
12946 | return -EINVAL; | |
12947 | } | |
12948 | ||
12949 | member = &btf_type_member(t)[member_idx]; | |
12950 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
12951 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
12952 | NULL); | |
12953 | if (!func_proto) { | |
12954 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
12955 | mname, member_idx, st_ops->name); | |
12956 | return -EINVAL; | |
12957 | } | |
12958 | ||
12959 | if (st_ops->check_member) { | |
12960 | int err = st_ops->check_member(t, member); | |
12961 | ||
12962 | if (err) { | |
12963 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
12964 | mname, st_ops->name); | |
12965 | return err; | |
12966 | } | |
12967 | } | |
12968 | ||
12969 | prog->aux->attach_func_proto = func_proto; | |
12970 | prog->aux->attach_func_name = mname; | |
12971 | env->ops = st_ops->verifier_ops; | |
12972 | ||
12973 | return 0; | |
12974 | } | |
6ba43b76 KS |
12975 | #define SECURITY_PREFIX "security_" |
12976 | ||
f7b12b6f | 12977 | static int check_attach_modify_return(unsigned long addr, const char *func_name) |
6ba43b76 | 12978 | { |
69191754 | 12979 | if (within_error_injection_list(addr) || |
f7b12b6f | 12980 | !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1)) |
6ba43b76 | 12981 | return 0; |
6ba43b76 | 12982 | |
6ba43b76 KS |
12983 | return -EINVAL; |
12984 | } | |
27ae7997 | 12985 | |
1e6c62a8 AS |
12986 | /* list of non-sleepable functions that are otherwise on |
12987 | * ALLOW_ERROR_INJECTION list | |
12988 | */ | |
12989 | BTF_SET_START(btf_non_sleepable_error_inject) | |
12990 | /* Three functions below can be called from sleepable and non-sleepable context. | |
12991 | * Assume non-sleepable from bpf safety point of view. | |
12992 | */ | |
12993 | BTF_ID(func, __add_to_page_cache_locked) | |
12994 | BTF_ID(func, should_fail_alloc_page) | |
12995 | BTF_ID(func, should_failslab) | |
12996 | BTF_SET_END(btf_non_sleepable_error_inject) | |
12997 | ||
12998 | static int check_non_sleepable_error_inject(u32 btf_id) | |
12999 | { | |
13000 | return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id); | |
13001 | } | |
13002 | ||
f7b12b6f THJ |
13003 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
13004 | const struct bpf_prog *prog, | |
13005 | const struct bpf_prog *tgt_prog, | |
13006 | u32 btf_id, | |
13007 | struct bpf_attach_target_info *tgt_info) | |
38207291 | 13008 | { |
be8704ff | 13009 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
f1b9509c | 13010 | const char prefix[] = "btf_trace_"; |
5b92a28a | 13011 | int ret = 0, subprog = -1, i; |
38207291 | 13012 | const struct btf_type *t; |
5b92a28a | 13013 | bool conservative = true; |
38207291 | 13014 | const char *tname; |
5b92a28a | 13015 | struct btf *btf; |
f7b12b6f | 13016 | long addr = 0; |
38207291 | 13017 | |
f1b9509c | 13018 | if (!btf_id) { |
efc68158 | 13019 | bpf_log(log, "Tracing programs must provide btf_id\n"); |
f1b9509c AS |
13020 | return -EINVAL; |
13021 | } | |
22dc4a0f | 13022 | btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf; |
5b92a28a | 13023 | if (!btf) { |
efc68158 | 13024 | bpf_log(log, |
5b92a28a AS |
13025 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); |
13026 | return -EINVAL; | |
13027 | } | |
13028 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c | 13029 | if (!t) { |
efc68158 | 13030 | bpf_log(log, "attach_btf_id %u is invalid\n", btf_id); |
f1b9509c AS |
13031 | return -EINVAL; |
13032 | } | |
5b92a28a | 13033 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c | 13034 | if (!tname) { |
efc68158 | 13035 | bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id); |
f1b9509c AS |
13036 | return -EINVAL; |
13037 | } | |
5b92a28a AS |
13038 | if (tgt_prog) { |
13039 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
13040 | ||
13041 | for (i = 0; i < aux->func_info_cnt; i++) | |
13042 | if (aux->func_info[i].type_id == btf_id) { | |
13043 | subprog = i; | |
13044 | break; | |
13045 | } | |
13046 | if (subprog == -1) { | |
efc68158 | 13047 | bpf_log(log, "Subprog %s doesn't exist\n", tname); |
5b92a28a AS |
13048 | return -EINVAL; |
13049 | } | |
13050 | conservative = aux->func_info_aux[subprog].unreliable; | |
be8704ff AS |
13051 | if (prog_extension) { |
13052 | if (conservative) { | |
efc68158 | 13053 | bpf_log(log, |
be8704ff AS |
13054 | "Cannot replace static functions\n"); |
13055 | return -EINVAL; | |
13056 | } | |
13057 | if (!prog->jit_requested) { | |
efc68158 | 13058 | bpf_log(log, |
be8704ff AS |
13059 | "Extension programs should be JITed\n"); |
13060 | return -EINVAL; | |
13061 | } | |
be8704ff AS |
13062 | } |
13063 | if (!tgt_prog->jited) { | |
efc68158 | 13064 | bpf_log(log, "Can attach to only JITed progs\n"); |
be8704ff AS |
13065 | return -EINVAL; |
13066 | } | |
13067 | if (tgt_prog->type == prog->type) { | |
13068 | /* Cannot fentry/fexit another fentry/fexit program. | |
13069 | * Cannot attach program extension to another extension. | |
13070 | * It's ok to attach fentry/fexit to extension program. | |
13071 | */ | |
efc68158 | 13072 | bpf_log(log, "Cannot recursively attach\n"); |
be8704ff AS |
13073 | return -EINVAL; |
13074 | } | |
13075 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
13076 | prog_extension && | |
13077 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
13078 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
13079 | /* Program extensions can extend all program types | |
13080 | * except fentry/fexit. The reason is the following. | |
13081 | * The fentry/fexit programs are used for performance | |
13082 | * analysis, stats and can be attached to any program | |
13083 | * type except themselves. When extension program is | |
13084 | * replacing XDP function it is necessary to allow | |
13085 | * performance analysis of all functions. Both original | |
13086 | * XDP program and its program extension. Hence | |
13087 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
13088 | * allowed. If extending of fentry/fexit was allowed it | |
13089 | * would be possible to create long call chain | |
13090 | * fentry->extension->fentry->extension beyond | |
13091 | * reasonable stack size. Hence extending fentry is not | |
13092 | * allowed. | |
13093 | */ | |
efc68158 | 13094 | bpf_log(log, "Cannot extend fentry/fexit\n"); |
be8704ff AS |
13095 | return -EINVAL; |
13096 | } | |
5b92a28a | 13097 | } else { |
be8704ff | 13098 | if (prog_extension) { |
efc68158 | 13099 | bpf_log(log, "Cannot replace kernel functions\n"); |
be8704ff AS |
13100 | return -EINVAL; |
13101 | } | |
5b92a28a | 13102 | } |
f1b9509c AS |
13103 | |
13104 | switch (prog->expected_attach_type) { | |
13105 | case BPF_TRACE_RAW_TP: | |
5b92a28a | 13106 | if (tgt_prog) { |
efc68158 | 13107 | bpf_log(log, |
5b92a28a AS |
13108 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); |
13109 | return -EINVAL; | |
13110 | } | |
38207291 | 13111 | if (!btf_type_is_typedef(t)) { |
efc68158 | 13112 | bpf_log(log, "attach_btf_id %u is not a typedef\n", |
38207291 MKL |
13113 | btf_id); |
13114 | return -EINVAL; | |
13115 | } | |
f1b9509c | 13116 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
efc68158 | 13117 | bpf_log(log, "attach_btf_id %u points to wrong type name %s\n", |
38207291 MKL |
13118 | btf_id, tname); |
13119 | return -EINVAL; | |
13120 | } | |
13121 | tname += sizeof(prefix) - 1; | |
5b92a28a | 13122 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
13123 | if (!btf_type_is_ptr(t)) |
13124 | /* should never happen in valid vmlinux build */ | |
13125 | return -EINVAL; | |
5b92a28a | 13126 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
13127 | if (!btf_type_is_func_proto(t)) |
13128 | /* should never happen in valid vmlinux build */ | |
13129 | return -EINVAL; | |
13130 | ||
f7b12b6f | 13131 | break; |
15d83c4d YS |
13132 | case BPF_TRACE_ITER: |
13133 | if (!btf_type_is_func(t)) { | |
efc68158 | 13134 | bpf_log(log, "attach_btf_id %u is not a function\n", |
15d83c4d YS |
13135 | btf_id); |
13136 | return -EINVAL; | |
13137 | } | |
13138 | t = btf_type_by_id(btf, t->type); | |
13139 | if (!btf_type_is_func_proto(t)) | |
13140 | return -EINVAL; | |
f7b12b6f THJ |
13141 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); |
13142 | if (ret) | |
13143 | return ret; | |
13144 | break; | |
be8704ff AS |
13145 | default: |
13146 | if (!prog_extension) | |
13147 | return -EINVAL; | |
df561f66 | 13148 | fallthrough; |
ae240823 | 13149 | case BPF_MODIFY_RETURN: |
9e4e01df | 13150 | case BPF_LSM_MAC: |
fec56f58 AS |
13151 | case BPF_TRACE_FENTRY: |
13152 | case BPF_TRACE_FEXIT: | |
13153 | if (!btf_type_is_func(t)) { | |
efc68158 | 13154 | bpf_log(log, "attach_btf_id %u is not a function\n", |
fec56f58 AS |
13155 | btf_id); |
13156 | return -EINVAL; | |
13157 | } | |
be8704ff | 13158 | if (prog_extension && |
efc68158 | 13159 | btf_check_type_match(log, prog, btf, t)) |
be8704ff | 13160 | return -EINVAL; |
5b92a28a | 13161 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
13162 | if (!btf_type_is_func_proto(t)) |
13163 | return -EINVAL; | |
f7b12b6f | 13164 | |
4a1e7c0c THJ |
13165 | if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) && |
13166 | (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type || | |
13167 | prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type)) | |
13168 | return -EINVAL; | |
13169 | ||
f7b12b6f | 13170 | if (tgt_prog && conservative) |
5b92a28a | 13171 | t = NULL; |
f7b12b6f THJ |
13172 | |
13173 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); | |
fec56f58 | 13174 | if (ret < 0) |
f7b12b6f THJ |
13175 | return ret; |
13176 | ||
5b92a28a | 13177 | if (tgt_prog) { |
e9eeec58 YS |
13178 | if (subprog == 0) |
13179 | addr = (long) tgt_prog->bpf_func; | |
13180 | else | |
13181 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a AS |
13182 | } else { |
13183 | addr = kallsyms_lookup_name(tname); | |
13184 | if (!addr) { | |
efc68158 | 13185 | bpf_log(log, |
5b92a28a AS |
13186 | "The address of function %s cannot be found\n", |
13187 | tname); | |
f7b12b6f | 13188 | return -ENOENT; |
5b92a28a | 13189 | } |
fec56f58 | 13190 | } |
18644cec | 13191 | |
1e6c62a8 AS |
13192 | if (prog->aux->sleepable) { |
13193 | ret = -EINVAL; | |
13194 | switch (prog->type) { | |
13195 | case BPF_PROG_TYPE_TRACING: | |
13196 | /* fentry/fexit/fmod_ret progs can be sleepable only if they are | |
13197 | * attached to ALLOW_ERROR_INJECTION and are not in denylist. | |
13198 | */ | |
13199 | if (!check_non_sleepable_error_inject(btf_id) && | |
13200 | within_error_injection_list(addr)) | |
13201 | ret = 0; | |
13202 | break; | |
13203 | case BPF_PROG_TYPE_LSM: | |
13204 | /* LSM progs check that they are attached to bpf_lsm_*() funcs. | |
13205 | * Only some of them are sleepable. | |
13206 | */ | |
423f1610 | 13207 | if (bpf_lsm_is_sleepable_hook(btf_id)) |
1e6c62a8 AS |
13208 | ret = 0; |
13209 | break; | |
13210 | default: | |
13211 | break; | |
13212 | } | |
f7b12b6f THJ |
13213 | if (ret) { |
13214 | bpf_log(log, "%s is not sleepable\n", tname); | |
13215 | return ret; | |
13216 | } | |
1e6c62a8 | 13217 | } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) { |
1af9270e | 13218 | if (tgt_prog) { |
efc68158 | 13219 | bpf_log(log, "can't modify return codes of BPF programs\n"); |
f7b12b6f THJ |
13220 | return -EINVAL; |
13221 | } | |
13222 | ret = check_attach_modify_return(addr, tname); | |
13223 | if (ret) { | |
13224 | bpf_log(log, "%s() is not modifiable\n", tname); | |
13225 | return ret; | |
1af9270e | 13226 | } |
18644cec | 13227 | } |
f7b12b6f THJ |
13228 | |
13229 | break; | |
13230 | } | |
13231 | tgt_info->tgt_addr = addr; | |
13232 | tgt_info->tgt_name = tname; | |
13233 | tgt_info->tgt_type = t; | |
13234 | return 0; | |
13235 | } | |
13236 | ||
35e3815f JO |
13237 | BTF_SET_START(btf_id_deny) |
13238 | BTF_ID_UNUSED | |
13239 | #ifdef CONFIG_SMP | |
13240 | BTF_ID(func, migrate_disable) | |
13241 | BTF_ID(func, migrate_enable) | |
13242 | #endif | |
13243 | #if !defined CONFIG_PREEMPT_RCU && !defined CONFIG_TINY_RCU | |
13244 | BTF_ID(func, rcu_read_unlock_strict) | |
13245 | #endif | |
13246 | BTF_SET_END(btf_id_deny) | |
13247 | ||
f7b12b6f THJ |
13248 | static int check_attach_btf_id(struct bpf_verifier_env *env) |
13249 | { | |
13250 | struct bpf_prog *prog = env->prog; | |
3aac1ead | 13251 | struct bpf_prog *tgt_prog = prog->aux->dst_prog; |
f7b12b6f THJ |
13252 | struct bpf_attach_target_info tgt_info = {}; |
13253 | u32 btf_id = prog->aux->attach_btf_id; | |
13254 | struct bpf_trampoline *tr; | |
13255 | int ret; | |
13256 | u64 key; | |
13257 | ||
79a7f8bd AS |
13258 | if (prog->type == BPF_PROG_TYPE_SYSCALL) { |
13259 | if (prog->aux->sleepable) | |
13260 | /* attach_btf_id checked to be zero already */ | |
13261 | return 0; | |
13262 | verbose(env, "Syscall programs can only be sleepable\n"); | |
13263 | return -EINVAL; | |
13264 | } | |
13265 | ||
f7b12b6f THJ |
13266 | if (prog->aux->sleepable && prog->type != BPF_PROG_TYPE_TRACING && |
13267 | prog->type != BPF_PROG_TYPE_LSM) { | |
13268 | verbose(env, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n"); | |
13269 | return -EINVAL; | |
13270 | } | |
13271 | ||
13272 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) | |
13273 | return check_struct_ops_btf_id(env); | |
13274 | ||
13275 | if (prog->type != BPF_PROG_TYPE_TRACING && | |
13276 | prog->type != BPF_PROG_TYPE_LSM && | |
13277 | prog->type != BPF_PROG_TYPE_EXT) | |
13278 | return 0; | |
13279 | ||
13280 | ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info); | |
13281 | if (ret) | |
fec56f58 | 13282 | return ret; |
f7b12b6f THJ |
13283 | |
13284 | if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) { | |
3aac1ead THJ |
13285 | /* to make freplace equivalent to their targets, they need to |
13286 | * inherit env->ops and expected_attach_type for the rest of the | |
13287 | * verification | |
13288 | */ | |
f7b12b6f THJ |
13289 | env->ops = bpf_verifier_ops[tgt_prog->type]; |
13290 | prog->expected_attach_type = tgt_prog->expected_attach_type; | |
13291 | } | |
13292 | ||
13293 | /* store info about the attachment target that will be used later */ | |
13294 | prog->aux->attach_func_proto = tgt_info.tgt_type; | |
13295 | prog->aux->attach_func_name = tgt_info.tgt_name; | |
13296 | ||
4a1e7c0c THJ |
13297 | if (tgt_prog) { |
13298 | prog->aux->saved_dst_prog_type = tgt_prog->type; | |
13299 | prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type; | |
13300 | } | |
13301 | ||
f7b12b6f THJ |
13302 | if (prog->expected_attach_type == BPF_TRACE_RAW_TP) { |
13303 | prog->aux->attach_btf_trace = true; | |
13304 | return 0; | |
13305 | } else if (prog->expected_attach_type == BPF_TRACE_ITER) { | |
13306 | if (!bpf_iter_prog_supported(prog)) | |
13307 | return -EINVAL; | |
13308 | return 0; | |
13309 | } | |
13310 | ||
13311 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
13312 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
13313 | if (ret < 0) | |
13314 | return ret; | |
35e3815f JO |
13315 | } else if (prog->type == BPF_PROG_TYPE_TRACING && |
13316 | btf_id_set_contains(&btf_id_deny, btf_id)) { | |
13317 | return -EINVAL; | |
38207291 | 13318 | } |
f7b12b6f | 13319 | |
22dc4a0f | 13320 | key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id); |
f7b12b6f THJ |
13321 | tr = bpf_trampoline_get(key, &tgt_info); |
13322 | if (!tr) | |
13323 | return -ENOMEM; | |
13324 | ||
3aac1ead | 13325 | prog->aux->dst_trampoline = tr; |
f7b12b6f | 13326 | return 0; |
38207291 MKL |
13327 | } |
13328 | ||
76654e67 AM |
13329 | struct btf *bpf_get_btf_vmlinux(void) |
13330 | { | |
13331 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { | |
13332 | mutex_lock(&bpf_verifier_lock); | |
13333 | if (!btf_vmlinux) | |
13334 | btf_vmlinux = btf_parse_vmlinux(); | |
13335 | mutex_unlock(&bpf_verifier_lock); | |
13336 | } | |
13337 | return btf_vmlinux; | |
13338 | } | |
13339 | ||
af2ac3e1 | 13340 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, bpfptr_t uattr) |
51580e79 | 13341 | { |
06ee7115 | 13342 | u64 start_time = ktime_get_ns(); |
58e2af8b | 13343 | struct bpf_verifier_env *env; |
b9193c1b | 13344 | struct bpf_verifier_log *log; |
9e4c24e7 | 13345 | int i, len, ret = -EINVAL; |
e2ae4ca2 | 13346 | bool is_priv; |
51580e79 | 13347 | |
eba0c929 AB |
13348 | /* no program is valid */ |
13349 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
13350 | return -EINVAL; | |
13351 | ||
58e2af8b | 13352 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
13353 | * allocate/free it every time bpf_check() is called |
13354 | */ | |
58e2af8b | 13355 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
13356 | if (!env) |
13357 | return -ENOMEM; | |
61bd5218 | 13358 | log = &env->log; |
cbd35700 | 13359 | |
9e4c24e7 | 13360 | len = (*prog)->len; |
fad953ce | 13361 | env->insn_aux_data = |
9e4c24e7 | 13362 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
13363 | ret = -ENOMEM; |
13364 | if (!env->insn_aux_data) | |
13365 | goto err_free_env; | |
9e4c24e7 JK |
13366 | for (i = 0; i < len; i++) |
13367 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 13368 | env->prog = *prog; |
00176a34 | 13369 | env->ops = bpf_verifier_ops[env->prog->type]; |
387544bf | 13370 | env->fd_array = make_bpfptr(attr->fd_array, uattr.is_kernel); |
2c78ee89 | 13371 | is_priv = bpf_capable(); |
0246e64d | 13372 | |
76654e67 | 13373 | bpf_get_btf_vmlinux(); |
8580ac94 | 13374 | |
cbd35700 | 13375 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
13376 | if (!is_priv) |
13377 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 AS |
13378 | |
13379 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
13380 | /* user requested verbose verifier output | |
13381 | * and supplied buffer to store the verification trace | |
13382 | */ | |
e7bf8249 JK |
13383 | log->level = attr->log_level; |
13384 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
13385 | log->len_total = attr->log_size; | |
cbd35700 AS |
13386 | |
13387 | ret = -EINVAL; | |
e7bf8249 | 13388 | /* log attributes have to be sane */ |
7a9f5c65 | 13389 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 || |
06ee7115 | 13390 | !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK) |
3df126f3 | 13391 | goto err_unlock; |
cbd35700 | 13392 | } |
1ad2f583 | 13393 | |
8580ac94 AS |
13394 | if (IS_ERR(btf_vmlinux)) { |
13395 | /* Either gcc or pahole or kernel are broken. */ | |
13396 | verbose(env, "in-kernel BTF is malformed\n"); | |
13397 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 13398 | goto skip_full_check; |
8580ac94 AS |
13399 | } |
13400 | ||
1ad2f583 DB |
13401 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
13402 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 13403 | env->strict_alignment = true; |
e9ee9efc DM |
13404 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
13405 | env->strict_alignment = false; | |
cbd35700 | 13406 | |
2c78ee89 | 13407 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
01f810ac | 13408 | env->allow_uninit_stack = bpf_allow_uninit_stack(); |
41c48f3a | 13409 | env->allow_ptr_to_map_access = bpf_allow_ptr_to_map_access(); |
2c78ee89 AS |
13410 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); |
13411 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
13412 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 13413 | |
10d274e8 AS |
13414 | if (is_priv) |
13415 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
13416 | ||
dc2a4ebc | 13417 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 13418 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
13419 | GFP_USER); |
13420 | ret = -ENOMEM; | |
13421 | if (!env->explored_states) | |
13422 | goto skip_full_check; | |
13423 | ||
e6ac2450 MKL |
13424 | ret = add_subprog_and_kfunc(env); |
13425 | if (ret < 0) | |
13426 | goto skip_full_check; | |
13427 | ||
d9762e84 | 13428 | ret = check_subprogs(env); |
475fb78f AS |
13429 | if (ret < 0) |
13430 | goto skip_full_check; | |
13431 | ||
c454a46b | 13432 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
13433 | if (ret < 0) |
13434 | goto skip_full_check; | |
13435 | ||
be8704ff AS |
13436 | ret = check_attach_btf_id(env); |
13437 | if (ret) | |
13438 | goto skip_full_check; | |
13439 | ||
4976b718 HL |
13440 | ret = resolve_pseudo_ldimm64(env); |
13441 | if (ret < 0) | |
13442 | goto skip_full_check; | |
13443 | ||
ceb11679 YZ |
13444 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
13445 | ret = bpf_prog_offload_verifier_prep(env->prog); | |
13446 | if (ret) | |
13447 | goto skip_full_check; | |
13448 | } | |
13449 | ||
d9762e84 MKL |
13450 | ret = check_cfg(env); |
13451 | if (ret < 0) | |
13452 | goto skip_full_check; | |
13453 | ||
51c39bb1 AS |
13454 | ret = do_check_subprogs(env); |
13455 | ret = ret ?: do_check_main(env); | |
cbd35700 | 13456 | |
c941ce9c QM |
13457 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
13458 | ret = bpf_prog_offload_finalize(env); | |
13459 | ||
0246e64d | 13460 | skip_full_check: |
51c39bb1 | 13461 | kvfree(env->explored_states); |
0246e64d | 13462 | |
c131187d | 13463 | if (ret == 0) |
9b38c405 | 13464 | ret = check_max_stack_depth(env); |
c131187d | 13465 | |
9b38c405 | 13466 | /* instruction rewrites happen after this point */ |
e2ae4ca2 JK |
13467 | if (is_priv) { |
13468 | if (ret == 0) | |
13469 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
13470 | if (ret == 0) |
13471 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
13472 | if (ret == 0) |
13473 | ret = opt_remove_nops(env); | |
52875a04 JK |
13474 | } else { |
13475 | if (ret == 0) | |
13476 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
13477 | } |
13478 | ||
9bac3d6d AS |
13479 | if (ret == 0) |
13480 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
13481 | ret = convert_ctx_accesses(env); | |
13482 | ||
e245c5c6 | 13483 | if (ret == 0) |
e6ac5933 | 13484 | ret = do_misc_fixups(env); |
e245c5c6 | 13485 | |
a4b1d3c1 JW |
13486 | /* do 32-bit optimization after insn patching has done so those patched |
13487 | * insns could be handled correctly. | |
13488 | */ | |
d6c2308c JW |
13489 | if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) { |
13490 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); | |
13491 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
13492 | : false; | |
a4b1d3c1 JW |
13493 | } |
13494 | ||
1ea47e01 AS |
13495 | if (ret == 0) |
13496 | ret = fixup_call_args(env); | |
13497 | ||
06ee7115 AS |
13498 | env->verification_time = ktime_get_ns() - start_time; |
13499 | print_verification_stats(env); | |
13500 | ||
a2a7d570 | 13501 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 13502 | ret = -ENOSPC; |
a2a7d570 | 13503 | if (log->level && !log->ubuf) { |
cbd35700 | 13504 | ret = -EFAULT; |
a2a7d570 | 13505 | goto err_release_maps; |
cbd35700 AS |
13506 | } |
13507 | ||
541c3bad AN |
13508 | if (ret) |
13509 | goto err_release_maps; | |
13510 | ||
13511 | if (env->used_map_cnt) { | |
0246e64d | 13512 | /* if program passed verifier, update used_maps in bpf_prog_info */ |
9bac3d6d AS |
13513 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
13514 | sizeof(env->used_maps[0]), | |
13515 | GFP_KERNEL); | |
0246e64d | 13516 | |
9bac3d6d | 13517 | if (!env->prog->aux->used_maps) { |
0246e64d | 13518 | ret = -ENOMEM; |
a2a7d570 | 13519 | goto err_release_maps; |
0246e64d AS |
13520 | } |
13521 | ||
9bac3d6d | 13522 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 13523 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 13524 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
541c3bad AN |
13525 | } |
13526 | if (env->used_btf_cnt) { | |
13527 | /* if program passed verifier, update used_btfs in bpf_prog_aux */ | |
13528 | env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt, | |
13529 | sizeof(env->used_btfs[0]), | |
13530 | GFP_KERNEL); | |
13531 | if (!env->prog->aux->used_btfs) { | |
13532 | ret = -ENOMEM; | |
13533 | goto err_release_maps; | |
13534 | } | |
0246e64d | 13535 | |
541c3bad AN |
13536 | memcpy(env->prog->aux->used_btfs, env->used_btfs, |
13537 | sizeof(env->used_btfs[0]) * env->used_btf_cnt); | |
13538 | env->prog->aux->used_btf_cnt = env->used_btf_cnt; | |
13539 | } | |
13540 | if (env->used_map_cnt || env->used_btf_cnt) { | |
0246e64d AS |
13541 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic |
13542 | * bpf_ld_imm64 instructions | |
13543 | */ | |
13544 | convert_pseudo_ld_imm64(env); | |
13545 | } | |
cbd35700 | 13546 | |
541c3bad | 13547 | adjust_btf_func(env); |
ba64e7d8 | 13548 | |
a2a7d570 | 13549 | err_release_maps: |
9bac3d6d | 13550 | if (!env->prog->aux->used_maps) |
0246e64d | 13551 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 13552 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
13553 | */ |
13554 | release_maps(env); | |
541c3bad AN |
13555 | if (!env->prog->aux->used_btfs) |
13556 | release_btfs(env); | |
03f87c0b THJ |
13557 | |
13558 | /* extension progs temporarily inherit the attach_type of their targets | |
13559 | for verification purposes, so set it back to zero before returning | |
13560 | */ | |
13561 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
13562 | env->prog->expected_attach_type = 0; | |
13563 | ||
9bac3d6d | 13564 | *prog = env->prog; |
3df126f3 | 13565 | err_unlock: |
45a73c17 AS |
13566 | if (!is_priv) |
13567 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
13568 | vfree(env->insn_aux_data); |
13569 | err_free_env: | |
13570 | kfree(env); | |
51580e79 AS |
13571 | return ret; |
13572 | } |