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. | |
50 | * Since it's analyzing all pathes 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 | |
135 | * returns ether pointer to map value or NULL. | |
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 | ||
69c087ba YS |
237 | static bool bpf_pseudo_func(const struct bpf_insn *insn) |
238 | { | |
239 | return insn->code == (BPF_LD | BPF_IMM | BPF_DW) && | |
240 | insn->src_reg == BPF_PSEUDO_FUNC; | |
241 | } | |
242 | ||
33ff9823 DB |
243 | struct bpf_call_arg_meta { |
244 | struct bpf_map *map_ptr; | |
435faee1 | 245 | bool raw_mode; |
36bbef52 | 246 | bool pkt_access; |
435faee1 DB |
247 | int regno; |
248 | int access_size; | |
457f4436 | 249 | int mem_size; |
10060503 | 250 | u64 msize_max_value; |
1b986589 | 251 | int ref_obj_id; |
d83525ca | 252 | int func_id; |
22dc4a0f | 253 | struct btf *btf; |
eaa6bcb7 | 254 | u32 btf_id; |
22dc4a0f | 255 | struct btf *ret_btf; |
eaa6bcb7 | 256 | u32 ret_btf_id; |
69c087ba | 257 | u32 subprogno; |
33ff9823 DB |
258 | }; |
259 | ||
8580ac94 AS |
260 | struct btf *btf_vmlinux; |
261 | ||
cbd35700 AS |
262 | static DEFINE_MUTEX(bpf_verifier_lock); |
263 | ||
d9762e84 MKL |
264 | static const struct bpf_line_info * |
265 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
266 | { | |
267 | const struct bpf_line_info *linfo; | |
268 | const struct bpf_prog *prog; | |
269 | u32 i, nr_linfo; | |
270 | ||
271 | prog = env->prog; | |
272 | nr_linfo = prog->aux->nr_linfo; | |
273 | ||
274 | if (!nr_linfo || insn_off >= prog->len) | |
275 | return NULL; | |
276 | ||
277 | linfo = prog->aux->linfo; | |
278 | for (i = 1; i < nr_linfo; i++) | |
279 | if (insn_off < linfo[i].insn_off) | |
280 | break; | |
281 | ||
282 | return &linfo[i - 1]; | |
283 | } | |
284 | ||
77d2e05a MKL |
285 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
286 | va_list args) | |
cbd35700 | 287 | { |
a2a7d570 | 288 | unsigned int n; |
cbd35700 | 289 | |
a2a7d570 | 290 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
291 | |
292 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
293 | "verifier log line truncated - local buffer too short\n"); | |
294 | ||
295 | n = min(log->len_total - log->len_used - 1, n); | |
296 | log->kbuf[n] = '\0'; | |
297 | ||
8580ac94 AS |
298 | if (log->level == BPF_LOG_KERNEL) { |
299 | pr_err("BPF:%s\n", log->kbuf); | |
300 | return; | |
301 | } | |
a2a7d570 JK |
302 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) |
303 | log->len_used += n; | |
304 | else | |
305 | log->ubuf = NULL; | |
cbd35700 | 306 | } |
abe08840 | 307 | |
6f8a57cc AN |
308 | static void bpf_vlog_reset(struct bpf_verifier_log *log, u32 new_pos) |
309 | { | |
310 | char zero = 0; | |
311 | ||
312 | if (!bpf_verifier_log_needed(log)) | |
313 | return; | |
314 | ||
315 | log->len_used = new_pos; | |
316 | if (put_user(zero, log->ubuf + new_pos)) | |
317 | log->ubuf = NULL; | |
318 | } | |
319 | ||
abe08840 JO |
320 | /* log_level controls verbosity level of eBPF verifier. |
321 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
322 | * so the user can figure out what's wrong with the program | |
430e68d1 | 323 | */ |
abe08840 JO |
324 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
325 | const char *fmt, ...) | |
326 | { | |
327 | va_list args; | |
328 | ||
77d2e05a MKL |
329 | if (!bpf_verifier_log_needed(&env->log)) |
330 | return; | |
331 | ||
abe08840 | 332 | va_start(args, fmt); |
77d2e05a | 333 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
334 | va_end(args); |
335 | } | |
336 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
337 | ||
338 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
339 | { | |
77d2e05a | 340 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
341 | va_list args; |
342 | ||
77d2e05a MKL |
343 | if (!bpf_verifier_log_needed(&env->log)) |
344 | return; | |
345 | ||
abe08840 | 346 | va_start(args, fmt); |
77d2e05a | 347 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
348 | va_end(args); |
349 | } | |
cbd35700 | 350 | |
9e15db66 AS |
351 | __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, |
352 | const char *fmt, ...) | |
353 | { | |
354 | va_list args; | |
355 | ||
356 | if (!bpf_verifier_log_needed(log)) | |
357 | return; | |
358 | ||
359 | va_start(args, fmt); | |
360 | bpf_verifier_vlog(log, fmt, args); | |
361 | va_end(args); | |
362 | } | |
363 | ||
d9762e84 MKL |
364 | static const char *ltrim(const char *s) |
365 | { | |
366 | while (isspace(*s)) | |
367 | s++; | |
368 | ||
369 | return s; | |
370 | } | |
371 | ||
372 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
373 | u32 insn_off, | |
374 | const char *prefix_fmt, ...) | |
375 | { | |
376 | const struct bpf_line_info *linfo; | |
377 | ||
378 | if (!bpf_verifier_log_needed(&env->log)) | |
379 | return; | |
380 | ||
381 | linfo = find_linfo(env, insn_off); | |
382 | if (!linfo || linfo == env->prev_linfo) | |
383 | return; | |
384 | ||
385 | if (prefix_fmt) { | |
386 | va_list args; | |
387 | ||
388 | va_start(args, prefix_fmt); | |
389 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
390 | va_end(args); | |
391 | } | |
392 | ||
393 | verbose(env, "%s\n", | |
394 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
395 | linfo->line_off))); | |
396 | ||
397 | env->prev_linfo = linfo; | |
398 | } | |
399 | ||
bc2591d6 YS |
400 | static void verbose_invalid_scalar(struct bpf_verifier_env *env, |
401 | struct bpf_reg_state *reg, | |
402 | struct tnum *range, const char *ctx, | |
403 | const char *reg_name) | |
404 | { | |
405 | char tn_buf[48]; | |
406 | ||
407 | verbose(env, "At %s the register %s ", ctx, reg_name); | |
408 | if (!tnum_is_unknown(reg->var_off)) { | |
409 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
410 | verbose(env, "has value %s", tn_buf); | |
411 | } else { | |
412 | verbose(env, "has unknown scalar value"); | |
413 | } | |
414 | tnum_strn(tn_buf, sizeof(tn_buf), *range); | |
415 | verbose(env, " should have been in %s\n", tn_buf); | |
416 | } | |
417 | ||
de8f3a83 DB |
418 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
419 | { | |
420 | return type == PTR_TO_PACKET || | |
421 | type == PTR_TO_PACKET_META; | |
422 | } | |
423 | ||
46f8bc92 MKL |
424 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
425 | { | |
426 | return type == PTR_TO_SOCKET || | |
655a51e5 | 427 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
428 | type == PTR_TO_TCP_SOCK || |
429 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
430 | } |
431 | ||
cac616db JF |
432 | static bool reg_type_not_null(enum bpf_reg_type type) |
433 | { | |
434 | return type == PTR_TO_SOCKET || | |
435 | type == PTR_TO_TCP_SOCK || | |
436 | type == PTR_TO_MAP_VALUE || | |
69c087ba | 437 | type == PTR_TO_MAP_KEY || |
01c66c48 | 438 | type == PTR_TO_SOCK_COMMON; |
cac616db JF |
439 | } |
440 | ||
840b9615 JS |
441 | static bool reg_type_may_be_null(enum bpf_reg_type type) |
442 | { | |
fd978bf7 | 443 | return type == PTR_TO_MAP_VALUE_OR_NULL || |
46f8bc92 | 444 | type == PTR_TO_SOCKET_OR_NULL || |
655a51e5 | 445 | type == PTR_TO_SOCK_COMMON_OR_NULL || |
b121b341 | 446 | type == PTR_TO_TCP_SOCK_OR_NULL || |
457f4436 | 447 | type == PTR_TO_BTF_ID_OR_NULL || |
afbf21dc YS |
448 | type == PTR_TO_MEM_OR_NULL || |
449 | type == PTR_TO_RDONLY_BUF_OR_NULL || | |
450 | type == PTR_TO_RDWR_BUF_OR_NULL; | |
fd978bf7 JS |
451 | } |
452 | ||
d83525ca AS |
453 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
454 | { | |
455 | return reg->type == PTR_TO_MAP_VALUE && | |
456 | map_value_has_spin_lock(reg->map_ptr); | |
457 | } | |
458 | ||
cba368c1 MKL |
459 | static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type) |
460 | { | |
461 | return type == PTR_TO_SOCKET || | |
462 | type == PTR_TO_SOCKET_OR_NULL || | |
463 | type == PTR_TO_TCP_SOCK || | |
457f4436 AN |
464 | type == PTR_TO_TCP_SOCK_OR_NULL || |
465 | type == PTR_TO_MEM || | |
466 | type == PTR_TO_MEM_OR_NULL; | |
cba368c1 MKL |
467 | } |
468 | ||
1b986589 | 469 | static bool arg_type_may_be_refcounted(enum bpf_arg_type type) |
fd978bf7 | 470 | { |
1b986589 | 471 | return type == ARG_PTR_TO_SOCK_COMMON; |
fd978bf7 JS |
472 | } |
473 | ||
fd1b0d60 LB |
474 | static bool arg_type_may_be_null(enum bpf_arg_type type) |
475 | { | |
476 | return type == ARG_PTR_TO_MAP_VALUE_OR_NULL || | |
477 | type == ARG_PTR_TO_MEM_OR_NULL || | |
478 | type == ARG_PTR_TO_CTX_OR_NULL || | |
479 | type == ARG_PTR_TO_SOCKET_OR_NULL || | |
69c087ba YS |
480 | type == ARG_PTR_TO_ALLOC_MEM_OR_NULL || |
481 | type == ARG_PTR_TO_STACK_OR_NULL; | |
fd1b0d60 LB |
482 | } |
483 | ||
fd978bf7 JS |
484 | /* Determine whether the function releases some resources allocated by another |
485 | * function call. The first reference type argument will be assumed to be | |
486 | * released by release_reference(). | |
487 | */ | |
488 | static bool is_release_function(enum bpf_func_id func_id) | |
489 | { | |
457f4436 AN |
490 | return func_id == BPF_FUNC_sk_release || |
491 | func_id == BPF_FUNC_ringbuf_submit || | |
492 | func_id == BPF_FUNC_ringbuf_discard; | |
840b9615 JS |
493 | } |
494 | ||
64d85290 | 495 | static bool may_be_acquire_function(enum bpf_func_id func_id) |
46f8bc92 MKL |
496 | { |
497 | return func_id == BPF_FUNC_sk_lookup_tcp || | |
edbf8c01 | 498 | func_id == BPF_FUNC_sk_lookup_udp || |
64d85290 | 499 | func_id == BPF_FUNC_skc_lookup_tcp || |
457f4436 AN |
500 | func_id == BPF_FUNC_map_lookup_elem || |
501 | func_id == BPF_FUNC_ringbuf_reserve; | |
64d85290 JS |
502 | } |
503 | ||
504 | static bool is_acquire_function(enum bpf_func_id func_id, | |
505 | const struct bpf_map *map) | |
506 | { | |
507 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
508 | ||
509 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
510 | func_id == BPF_FUNC_sk_lookup_udp || | |
457f4436 AN |
511 | func_id == BPF_FUNC_skc_lookup_tcp || |
512 | func_id == BPF_FUNC_ringbuf_reserve) | |
64d85290 JS |
513 | return true; |
514 | ||
515 | if (func_id == BPF_FUNC_map_lookup_elem && | |
516 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
517 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
518 | return true; | |
519 | ||
520 | return false; | |
46f8bc92 MKL |
521 | } |
522 | ||
1b986589 MKL |
523 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
524 | { | |
525 | return func_id == BPF_FUNC_tcp_sock || | |
1df8f55a MKL |
526 | func_id == BPF_FUNC_sk_fullsock || |
527 | func_id == BPF_FUNC_skc_to_tcp_sock || | |
528 | func_id == BPF_FUNC_skc_to_tcp6_sock || | |
529 | func_id == BPF_FUNC_skc_to_udp6_sock || | |
530 | func_id == BPF_FUNC_skc_to_tcp_timewait_sock || | |
531 | func_id == BPF_FUNC_skc_to_tcp_request_sock; | |
1b986589 MKL |
532 | } |
533 | ||
39491867 BJ |
534 | static bool is_cmpxchg_insn(const struct bpf_insn *insn) |
535 | { | |
536 | return BPF_CLASS(insn->code) == BPF_STX && | |
537 | BPF_MODE(insn->code) == BPF_ATOMIC && | |
538 | insn->imm == BPF_CMPXCHG; | |
539 | } | |
540 | ||
17a52670 AS |
541 | /* string representation of 'enum bpf_reg_type' */ |
542 | static const char * const reg_type_str[] = { | |
543 | [NOT_INIT] = "?", | |
f1174f77 | 544 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
545 | [PTR_TO_CTX] = "ctx", |
546 | [CONST_PTR_TO_MAP] = "map_ptr", | |
547 | [PTR_TO_MAP_VALUE] = "map_value", | |
548 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 549 | [PTR_TO_STACK] = "fp", |
969bf05e | 550 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 551 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 552 | [PTR_TO_PACKET_END] = "pkt_end", |
d58e468b | 553 | [PTR_TO_FLOW_KEYS] = "flow_keys", |
c64b7983 JS |
554 | [PTR_TO_SOCKET] = "sock", |
555 | [PTR_TO_SOCKET_OR_NULL] = "sock_or_null", | |
46f8bc92 MKL |
556 | [PTR_TO_SOCK_COMMON] = "sock_common", |
557 | [PTR_TO_SOCK_COMMON_OR_NULL] = "sock_common_or_null", | |
655a51e5 MKL |
558 | [PTR_TO_TCP_SOCK] = "tcp_sock", |
559 | [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", | |
9df1c28b | 560 | [PTR_TO_TP_BUFFER] = "tp_buffer", |
fada7fdc | 561 | [PTR_TO_XDP_SOCK] = "xdp_sock", |
9e15db66 | 562 | [PTR_TO_BTF_ID] = "ptr_", |
b121b341 | 563 | [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", |
eaa6bcb7 | 564 | [PTR_TO_PERCPU_BTF_ID] = "percpu_ptr_", |
457f4436 AN |
565 | [PTR_TO_MEM] = "mem", |
566 | [PTR_TO_MEM_OR_NULL] = "mem_or_null", | |
afbf21dc YS |
567 | [PTR_TO_RDONLY_BUF] = "rdonly_buf", |
568 | [PTR_TO_RDONLY_BUF_OR_NULL] = "rdonly_buf_or_null", | |
569 | [PTR_TO_RDWR_BUF] = "rdwr_buf", | |
570 | [PTR_TO_RDWR_BUF_OR_NULL] = "rdwr_buf_or_null", | |
69c087ba YS |
571 | [PTR_TO_FUNC] = "func", |
572 | [PTR_TO_MAP_KEY] = "map_key", | |
17a52670 AS |
573 | }; |
574 | ||
8efea21d EC |
575 | static char slot_type_char[] = { |
576 | [STACK_INVALID] = '?', | |
577 | [STACK_SPILL] = 'r', | |
578 | [STACK_MISC] = 'm', | |
579 | [STACK_ZERO] = '0', | |
580 | }; | |
581 | ||
4e92024a AS |
582 | static void print_liveness(struct bpf_verifier_env *env, |
583 | enum bpf_reg_liveness live) | |
584 | { | |
9242b5f5 | 585 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
586 | verbose(env, "_"); |
587 | if (live & REG_LIVE_READ) | |
588 | verbose(env, "r"); | |
589 | if (live & REG_LIVE_WRITTEN) | |
590 | verbose(env, "w"); | |
9242b5f5 AS |
591 | if (live & REG_LIVE_DONE) |
592 | verbose(env, "D"); | |
4e92024a AS |
593 | } |
594 | ||
f4d7e40a AS |
595 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
596 | const struct bpf_reg_state *reg) | |
597 | { | |
598 | struct bpf_verifier_state *cur = env->cur_state; | |
599 | ||
600 | return cur->frame[reg->frameno]; | |
601 | } | |
602 | ||
22dc4a0f | 603 | static const char *kernel_type_name(const struct btf* btf, u32 id) |
9e15db66 | 604 | { |
22dc4a0f | 605 | return btf_name_by_offset(btf, btf_type_by_id(btf, id)->name_off); |
9e15db66 AS |
606 | } |
607 | ||
61bd5218 | 608 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 609 | const struct bpf_func_state *state) |
17a52670 | 610 | { |
f4d7e40a | 611 | const struct bpf_reg_state *reg; |
17a52670 AS |
612 | enum bpf_reg_type t; |
613 | int i; | |
614 | ||
f4d7e40a AS |
615 | if (state->frameno) |
616 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 617 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
618 | reg = &state->regs[i]; |
619 | t = reg->type; | |
17a52670 AS |
620 | if (t == NOT_INIT) |
621 | continue; | |
4e92024a AS |
622 | verbose(env, " R%d", i); |
623 | print_liveness(env, reg->live); | |
624 | verbose(env, "=%s", reg_type_str[t]); | |
b5dc0163 AS |
625 | if (t == SCALAR_VALUE && reg->precise) |
626 | verbose(env, "P"); | |
f1174f77 EC |
627 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
628 | tnum_is_const(reg->var_off)) { | |
629 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 630 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 631 | } else { |
eaa6bcb7 HL |
632 | if (t == PTR_TO_BTF_ID || |
633 | t == PTR_TO_BTF_ID_OR_NULL || | |
634 | t == PTR_TO_PERCPU_BTF_ID) | |
22dc4a0f | 635 | verbose(env, "%s", kernel_type_name(reg->btf, reg->btf_id)); |
cba368c1 MKL |
636 | verbose(env, "(id=%d", reg->id); |
637 | if (reg_type_may_be_refcounted_or_null(t)) | |
638 | verbose(env, ",ref_obj_id=%d", reg->ref_obj_id); | |
f1174f77 | 639 | if (t != SCALAR_VALUE) |
61bd5218 | 640 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 641 | if (type_is_pkt_pointer(t)) |
61bd5218 | 642 | verbose(env, ",r=%d", reg->range); |
f1174f77 | 643 | else if (t == CONST_PTR_TO_MAP || |
69c087ba | 644 | t == PTR_TO_MAP_KEY || |
f1174f77 EC |
645 | t == PTR_TO_MAP_VALUE || |
646 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 647 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
648 | reg->map_ptr->key_size, |
649 | reg->map_ptr->value_size); | |
7d1238f2 EC |
650 | if (tnum_is_const(reg->var_off)) { |
651 | /* Typically an immediate SCALAR_VALUE, but | |
652 | * could be a pointer whose offset is too big | |
653 | * for reg->off | |
654 | */ | |
61bd5218 | 655 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
656 | } else { |
657 | if (reg->smin_value != reg->umin_value && | |
658 | reg->smin_value != S64_MIN) | |
61bd5218 | 659 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
660 | (long long)reg->smin_value); |
661 | if (reg->smax_value != reg->umax_value && | |
662 | reg->smax_value != S64_MAX) | |
61bd5218 | 663 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
664 | (long long)reg->smax_value); |
665 | if (reg->umin_value != 0) | |
61bd5218 | 666 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
667 | (unsigned long long)reg->umin_value); |
668 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 669 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
670 | (unsigned long long)reg->umax_value); |
671 | if (!tnum_is_unknown(reg->var_off)) { | |
672 | char tn_buf[48]; | |
f1174f77 | 673 | |
7d1238f2 | 674 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 675 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 676 | } |
3f50f132 JF |
677 | if (reg->s32_min_value != reg->smin_value && |
678 | reg->s32_min_value != S32_MIN) | |
679 | verbose(env, ",s32_min_value=%d", | |
680 | (int)(reg->s32_min_value)); | |
681 | if (reg->s32_max_value != reg->smax_value && | |
682 | reg->s32_max_value != S32_MAX) | |
683 | verbose(env, ",s32_max_value=%d", | |
684 | (int)(reg->s32_max_value)); | |
685 | if (reg->u32_min_value != reg->umin_value && | |
686 | reg->u32_min_value != U32_MIN) | |
687 | verbose(env, ",u32_min_value=%d", | |
688 | (int)(reg->u32_min_value)); | |
689 | if (reg->u32_max_value != reg->umax_value && | |
690 | reg->u32_max_value != U32_MAX) | |
691 | verbose(env, ",u32_max_value=%d", | |
692 | (int)(reg->u32_max_value)); | |
f1174f77 | 693 | } |
61bd5218 | 694 | verbose(env, ")"); |
f1174f77 | 695 | } |
17a52670 | 696 | } |
638f5b90 | 697 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
698 | char types_buf[BPF_REG_SIZE + 1]; |
699 | bool valid = false; | |
700 | int j; | |
701 | ||
702 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
703 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
704 | valid = true; | |
705 | types_buf[j] = slot_type_char[ | |
706 | state->stack[i].slot_type[j]]; | |
707 | } | |
708 | types_buf[BPF_REG_SIZE] = 0; | |
709 | if (!valid) | |
710 | continue; | |
711 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
712 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
b5dc0163 AS |
713 | if (state->stack[i].slot_type[0] == STACK_SPILL) { |
714 | reg = &state->stack[i].spilled_ptr; | |
715 | t = reg->type; | |
716 | verbose(env, "=%s", reg_type_str[t]); | |
717 | if (t == SCALAR_VALUE && reg->precise) | |
718 | verbose(env, "P"); | |
719 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
720 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
721 | } else { | |
8efea21d | 722 | verbose(env, "=%s", types_buf); |
b5dc0163 | 723 | } |
17a52670 | 724 | } |
fd978bf7 JS |
725 | if (state->acquired_refs && state->refs[0].id) { |
726 | verbose(env, " refs=%d", state->refs[0].id); | |
727 | for (i = 1; i < state->acquired_refs; i++) | |
728 | if (state->refs[i].id) | |
729 | verbose(env, ",%d", state->refs[i].id); | |
730 | } | |
61bd5218 | 731 | verbose(env, "\n"); |
17a52670 AS |
732 | } |
733 | ||
84dbf350 JS |
734 | #define COPY_STATE_FN(NAME, COUNT, FIELD, SIZE) \ |
735 | static int copy_##NAME##_state(struct bpf_func_state *dst, \ | |
736 | const struct bpf_func_state *src) \ | |
737 | { \ | |
738 | if (!src->FIELD) \ | |
739 | return 0; \ | |
740 | if (WARN_ON_ONCE(dst->COUNT < src->COUNT)) { \ | |
741 | /* internal bug, make state invalid to reject the program */ \ | |
742 | memset(dst, 0, sizeof(*dst)); \ | |
743 | return -EFAULT; \ | |
744 | } \ | |
745 | memcpy(dst->FIELD, src->FIELD, \ | |
746 | sizeof(*src->FIELD) * (src->COUNT / SIZE)); \ | |
747 | return 0; \ | |
638f5b90 | 748 | } |
fd978bf7 JS |
749 | /* copy_reference_state() */ |
750 | COPY_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
751 | /* copy_stack_state() */ |
752 | COPY_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
753 | #undef COPY_STATE_FN | |
754 | ||
755 | #define REALLOC_STATE_FN(NAME, COUNT, FIELD, SIZE) \ | |
756 | static int realloc_##NAME##_state(struct bpf_func_state *state, int size, \ | |
757 | bool copy_old) \ | |
758 | { \ | |
759 | u32 old_size = state->COUNT; \ | |
760 | struct bpf_##NAME##_state *new_##FIELD; \ | |
761 | int slot = size / SIZE; \ | |
762 | \ | |
763 | if (size <= old_size || !size) { \ | |
764 | if (copy_old) \ | |
765 | return 0; \ | |
766 | state->COUNT = slot * SIZE; \ | |
767 | if (!size && old_size) { \ | |
768 | kfree(state->FIELD); \ | |
769 | state->FIELD = NULL; \ | |
770 | } \ | |
771 | return 0; \ | |
772 | } \ | |
773 | new_##FIELD = kmalloc_array(slot, sizeof(struct bpf_##NAME##_state), \ | |
774 | GFP_KERNEL); \ | |
775 | if (!new_##FIELD) \ | |
776 | return -ENOMEM; \ | |
777 | if (copy_old) { \ | |
778 | if (state->FIELD) \ | |
779 | memcpy(new_##FIELD, state->FIELD, \ | |
780 | sizeof(*new_##FIELD) * (old_size / SIZE)); \ | |
781 | memset(new_##FIELD + old_size / SIZE, 0, \ | |
782 | sizeof(*new_##FIELD) * (size - old_size) / SIZE); \ | |
783 | } \ | |
784 | state->COUNT = slot * SIZE; \ | |
785 | kfree(state->FIELD); \ | |
786 | state->FIELD = new_##FIELD; \ | |
787 | return 0; \ | |
788 | } | |
fd978bf7 JS |
789 | /* realloc_reference_state() */ |
790 | REALLOC_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
791 | /* realloc_stack_state() */ |
792 | REALLOC_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
793 | #undef REALLOC_STATE_FN | |
638f5b90 AS |
794 | |
795 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
796 | * make it consume minimal amount of memory. check_stack_write() access from | |
f4d7e40a | 797 | * the program calls into realloc_func_state() to grow the stack size. |
84dbf350 JS |
798 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state |
799 | * which realloc_stack_state() copies over. It points to previous | |
800 | * bpf_verifier_state which is never reallocated. | |
638f5b90 | 801 | */ |
fd978bf7 JS |
802 | static int realloc_func_state(struct bpf_func_state *state, int stack_size, |
803 | int refs_size, bool copy_old) | |
638f5b90 | 804 | { |
fd978bf7 JS |
805 | int err = realloc_reference_state(state, refs_size, copy_old); |
806 | if (err) | |
807 | return err; | |
808 | return realloc_stack_state(state, stack_size, copy_old); | |
809 | } | |
810 | ||
811 | /* Acquire a pointer id from the env and update the state->refs to include | |
812 | * this new pointer reference. | |
813 | * On success, returns a valid pointer id to associate with the register | |
814 | * On failure, returns a negative errno. | |
638f5b90 | 815 | */ |
fd978bf7 | 816 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 817 | { |
fd978bf7 JS |
818 | struct bpf_func_state *state = cur_func(env); |
819 | int new_ofs = state->acquired_refs; | |
820 | int id, err; | |
821 | ||
822 | err = realloc_reference_state(state, state->acquired_refs + 1, true); | |
823 | if (err) | |
824 | return err; | |
825 | id = ++env->id_gen; | |
826 | state->refs[new_ofs].id = id; | |
827 | state->refs[new_ofs].insn_idx = insn_idx; | |
638f5b90 | 828 | |
fd978bf7 JS |
829 | return id; |
830 | } | |
831 | ||
832 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 833 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
834 | { |
835 | int i, last_idx; | |
836 | ||
fd978bf7 JS |
837 | last_idx = state->acquired_refs - 1; |
838 | for (i = 0; i < state->acquired_refs; i++) { | |
839 | if (state->refs[i].id == ptr_id) { | |
840 | if (last_idx && i != last_idx) | |
841 | memcpy(&state->refs[i], &state->refs[last_idx], | |
842 | sizeof(*state->refs)); | |
843 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
844 | state->acquired_refs--; | |
638f5b90 | 845 | return 0; |
638f5b90 | 846 | } |
638f5b90 | 847 | } |
46f8bc92 | 848 | return -EINVAL; |
fd978bf7 JS |
849 | } |
850 | ||
851 | static int transfer_reference_state(struct bpf_func_state *dst, | |
852 | struct bpf_func_state *src) | |
853 | { | |
854 | int err = realloc_reference_state(dst, src->acquired_refs, false); | |
855 | if (err) | |
856 | return err; | |
857 | err = copy_reference_state(dst, src); | |
858 | if (err) | |
859 | return err; | |
638f5b90 AS |
860 | return 0; |
861 | } | |
862 | ||
f4d7e40a AS |
863 | static void free_func_state(struct bpf_func_state *state) |
864 | { | |
5896351e AS |
865 | if (!state) |
866 | return; | |
fd978bf7 | 867 | kfree(state->refs); |
f4d7e40a AS |
868 | kfree(state->stack); |
869 | kfree(state); | |
870 | } | |
871 | ||
b5dc0163 AS |
872 | static void clear_jmp_history(struct bpf_verifier_state *state) |
873 | { | |
874 | kfree(state->jmp_history); | |
875 | state->jmp_history = NULL; | |
876 | state->jmp_history_cnt = 0; | |
877 | } | |
878 | ||
1969db47 AS |
879 | static void free_verifier_state(struct bpf_verifier_state *state, |
880 | bool free_self) | |
638f5b90 | 881 | { |
f4d7e40a AS |
882 | int i; |
883 | ||
884 | for (i = 0; i <= state->curframe; i++) { | |
885 | free_func_state(state->frame[i]); | |
886 | state->frame[i] = NULL; | |
887 | } | |
b5dc0163 | 888 | clear_jmp_history(state); |
1969db47 AS |
889 | if (free_self) |
890 | kfree(state); | |
638f5b90 AS |
891 | } |
892 | ||
893 | /* copy verifier state from src to dst growing dst stack space | |
894 | * when necessary to accommodate larger src stack | |
895 | */ | |
f4d7e40a AS |
896 | static int copy_func_state(struct bpf_func_state *dst, |
897 | const struct bpf_func_state *src) | |
638f5b90 AS |
898 | { |
899 | int err; | |
900 | ||
fd978bf7 JS |
901 | err = realloc_func_state(dst, src->allocated_stack, src->acquired_refs, |
902 | false); | |
903 | if (err) | |
904 | return err; | |
905 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); | |
906 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
907 | if (err) |
908 | return err; | |
638f5b90 AS |
909 | return copy_stack_state(dst, src); |
910 | } | |
911 | ||
f4d7e40a AS |
912 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
913 | const struct bpf_verifier_state *src) | |
914 | { | |
915 | struct bpf_func_state *dst; | |
b5dc0163 | 916 | u32 jmp_sz = sizeof(struct bpf_idx_pair) * src->jmp_history_cnt; |
f4d7e40a AS |
917 | int i, err; |
918 | ||
b5dc0163 AS |
919 | if (dst_state->jmp_history_cnt < src->jmp_history_cnt) { |
920 | kfree(dst_state->jmp_history); | |
921 | dst_state->jmp_history = kmalloc(jmp_sz, GFP_USER); | |
922 | if (!dst_state->jmp_history) | |
923 | return -ENOMEM; | |
924 | } | |
925 | memcpy(dst_state->jmp_history, src->jmp_history, jmp_sz); | |
926 | dst_state->jmp_history_cnt = src->jmp_history_cnt; | |
927 | ||
f4d7e40a AS |
928 | /* if dst has more stack frames then src frame, free them */ |
929 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
930 | free_func_state(dst_state->frame[i]); | |
931 | dst_state->frame[i] = NULL; | |
932 | } | |
979d63d5 | 933 | dst_state->speculative = src->speculative; |
f4d7e40a | 934 | dst_state->curframe = src->curframe; |
d83525ca | 935 | dst_state->active_spin_lock = src->active_spin_lock; |
2589726d AS |
936 | dst_state->branches = src->branches; |
937 | dst_state->parent = src->parent; | |
b5dc0163 AS |
938 | dst_state->first_insn_idx = src->first_insn_idx; |
939 | dst_state->last_insn_idx = src->last_insn_idx; | |
f4d7e40a AS |
940 | for (i = 0; i <= src->curframe; i++) { |
941 | dst = dst_state->frame[i]; | |
942 | if (!dst) { | |
943 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
944 | if (!dst) | |
945 | return -ENOMEM; | |
946 | dst_state->frame[i] = dst; | |
947 | } | |
948 | err = copy_func_state(dst, src->frame[i]); | |
949 | if (err) | |
950 | return err; | |
951 | } | |
952 | return 0; | |
953 | } | |
954 | ||
2589726d AS |
955 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
956 | { | |
957 | while (st) { | |
958 | u32 br = --st->branches; | |
959 | ||
960 | /* WARN_ON(br > 1) technically makes sense here, | |
961 | * but see comment in push_stack(), hence: | |
962 | */ | |
963 | WARN_ONCE((int)br < 0, | |
964 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
965 | br); | |
966 | if (br) | |
967 | break; | |
968 | st = st->parent; | |
969 | } | |
970 | } | |
971 | ||
638f5b90 | 972 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 973 | int *insn_idx, bool pop_log) |
638f5b90 AS |
974 | { |
975 | struct bpf_verifier_state *cur = env->cur_state; | |
976 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
977 | int err; | |
17a52670 AS |
978 | |
979 | if (env->head == NULL) | |
638f5b90 | 980 | return -ENOENT; |
17a52670 | 981 | |
638f5b90 AS |
982 | if (cur) { |
983 | err = copy_verifier_state(cur, &head->st); | |
984 | if (err) | |
985 | return err; | |
986 | } | |
6f8a57cc AN |
987 | if (pop_log) |
988 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
989 | if (insn_idx) |
990 | *insn_idx = head->insn_idx; | |
17a52670 | 991 | if (prev_insn_idx) |
638f5b90 AS |
992 | *prev_insn_idx = head->prev_insn_idx; |
993 | elem = head->next; | |
1969db47 | 994 | free_verifier_state(&head->st, false); |
638f5b90 | 995 | kfree(head); |
17a52670 AS |
996 | env->head = elem; |
997 | env->stack_size--; | |
638f5b90 | 998 | return 0; |
17a52670 AS |
999 | } |
1000 | ||
58e2af8b | 1001 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
1002 | int insn_idx, int prev_insn_idx, |
1003 | bool speculative) | |
17a52670 | 1004 | { |
638f5b90 | 1005 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 1006 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 1007 | int err; |
17a52670 | 1008 | |
638f5b90 | 1009 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
1010 | if (!elem) |
1011 | goto err; | |
1012 | ||
17a52670 AS |
1013 | elem->insn_idx = insn_idx; |
1014 | elem->prev_insn_idx = prev_insn_idx; | |
1015 | elem->next = env->head; | |
6f8a57cc | 1016 | elem->log_pos = env->log.len_used; |
17a52670 AS |
1017 | env->head = elem; |
1018 | env->stack_size++; | |
1969db47 AS |
1019 | err = copy_verifier_state(&elem->st, cur); |
1020 | if (err) | |
1021 | goto err; | |
979d63d5 | 1022 | elem->st.speculative |= speculative; |
b285fcb7 AS |
1023 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
1024 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
1025 | env->stack_size); | |
17a52670 AS |
1026 | goto err; |
1027 | } | |
2589726d AS |
1028 | if (elem->st.parent) { |
1029 | ++elem->st.parent->branches; | |
1030 | /* WARN_ON(branches > 2) technically makes sense here, | |
1031 | * but | |
1032 | * 1. speculative states will bump 'branches' for non-branch | |
1033 | * instructions | |
1034 | * 2. is_state_visited() heuristics may decide not to create | |
1035 | * a new state for a sequence of branches and all such current | |
1036 | * and cloned states will be pointing to a single parent state | |
1037 | * which might have large 'branches' count. | |
1038 | */ | |
1039 | } | |
17a52670 AS |
1040 | return &elem->st; |
1041 | err: | |
5896351e AS |
1042 | free_verifier_state(env->cur_state, true); |
1043 | env->cur_state = NULL; | |
17a52670 | 1044 | /* pop all elements and return */ |
6f8a57cc | 1045 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
1046 | return NULL; |
1047 | } | |
1048 | ||
1049 | #define CALLER_SAVED_REGS 6 | |
1050 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
1051 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
1052 | }; | |
1053 | ||
f54c7898 DB |
1054 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1055 | struct bpf_reg_state *reg); | |
f1174f77 | 1056 | |
e688c3db AS |
1057 | /* This helper doesn't clear reg->id */ |
1058 | static void ___mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
b03c9f9f | 1059 | { |
b03c9f9f EC |
1060 | reg->var_off = tnum_const(imm); |
1061 | reg->smin_value = (s64)imm; | |
1062 | reg->smax_value = (s64)imm; | |
1063 | reg->umin_value = imm; | |
1064 | reg->umax_value = imm; | |
3f50f132 JF |
1065 | |
1066 | reg->s32_min_value = (s32)imm; | |
1067 | reg->s32_max_value = (s32)imm; | |
1068 | reg->u32_min_value = (u32)imm; | |
1069 | reg->u32_max_value = (u32)imm; | |
1070 | } | |
1071 | ||
e688c3db AS |
1072 | /* Mark the unknown part of a register (variable offset or scalar value) as |
1073 | * known to have the value @imm. | |
1074 | */ | |
1075 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
1076 | { | |
1077 | /* Clear id, off, and union(map_ptr, range) */ | |
1078 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
1079 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
1080 | ___mark_reg_known(reg, imm); | |
1081 | } | |
1082 | ||
3f50f132 JF |
1083 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) |
1084 | { | |
1085 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
1086 | reg->s32_min_value = (s32)imm; | |
1087 | reg->s32_max_value = (s32)imm; | |
1088 | reg->u32_min_value = (u32)imm; | |
1089 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
1090 | } |
1091 | ||
f1174f77 EC |
1092 | /* Mark the 'variable offset' part of a register as zero. This should be |
1093 | * used only on registers holding a pointer type. | |
1094 | */ | |
1095 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 1096 | { |
b03c9f9f | 1097 | __mark_reg_known(reg, 0); |
f1174f77 | 1098 | } |
a9789ef9 | 1099 | |
cc2b14d5 AS |
1100 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
1101 | { | |
1102 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
1103 | reg->type = SCALAR_VALUE; |
1104 | } | |
1105 | ||
61bd5218 JK |
1106 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
1107 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1108 | { |
1109 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1110 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
1111 | /* Something bad happened, let's kill all regs */ |
1112 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 1113 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1114 | return; |
1115 | } | |
1116 | __mark_reg_known_zero(regs + regno); | |
1117 | } | |
1118 | ||
4ddb7416 DB |
1119 | static void mark_ptr_not_null_reg(struct bpf_reg_state *reg) |
1120 | { | |
1121 | switch (reg->type) { | |
1122 | case PTR_TO_MAP_VALUE_OR_NULL: { | |
1123 | const struct bpf_map *map = reg->map_ptr; | |
1124 | ||
1125 | if (map->inner_map_meta) { | |
1126 | reg->type = CONST_PTR_TO_MAP; | |
1127 | reg->map_ptr = map->inner_map_meta; | |
1128 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { | |
1129 | reg->type = PTR_TO_XDP_SOCK; | |
1130 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || | |
1131 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
1132 | reg->type = PTR_TO_SOCKET; | |
1133 | } else { | |
1134 | reg->type = PTR_TO_MAP_VALUE; | |
1135 | } | |
1136 | break; | |
1137 | } | |
1138 | case PTR_TO_SOCKET_OR_NULL: | |
1139 | reg->type = PTR_TO_SOCKET; | |
1140 | break; | |
1141 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
1142 | reg->type = PTR_TO_SOCK_COMMON; | |
1143 | break; | |
1144 | case PTR_TO_TCP_SOCK_OR_NULL: | |
1145 | reg->type = PTR_TO_TCP_SOCK; | |
1146 | break; | |
1147 | case PTR_TO_BTF_ID_OR_NULL: | |
1148 | reg->type = PTR_TO_BTF_ID; | |
1149 | break; | |
1150 | case PTR_TO_MEM_OR_NULL: | |
1151 | reg->type = PTR_TO_MEM; | |
1152 | break; | |
1153 | case PTR_TO_RDONLY_BUF_OR_NULL: | |
1154 | reg->type = PTR_TO_RDONLY_BUF; | |
1155 | break; | |
1156 | case PTR_TO_RDWR_BUF_OR_NULL: | |
1157 | reg->type = PTR_TO_RDWR_BUF; | |
1158 | break; | |
1159 | default: | |
33ccec5f | 1160 | WARN_ONCE(1, "unknown nullable register type"); |
4ddb7416 DB |
1161 | } |
1162 | } | |
1163 | ||
de8f3a83 DB |
1164 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
1165 | { | |
1166 | return type_is_pkt_pointer(reg->type); | |
1167 | } | |
1168 | ||
1169 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
1170 | { | |
1171 | return reg_is_pkt_pointer(reg) || | |
1172 | reg->type == PTR_TO_PACKET_END; | |
1173 | } | |
1174 | ||
1175 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
1176 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
1177 | enum bpf_reg_type which) | |
1178 | { | |
1179 | /* The register can already have a range from prior markings. | |
1180 | * This is fine as long as it hasn't been advanced from its | |
1181 | * origin. | |
1182 | */ | |
1183 | return reg->type == which && | |
1184 | reg->id == 0 && | |
1185 | reg->off == 0 && | |
1186 | tnum_equals_const(reg->var_off, 0); | |
1187 | } | |
1188 | ||
3f50f132 JF |
1189 | /* Reset the min/max bounds of a register */ |
1190 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
1191 | { | |
1192 | reg->smin_value = S64_MIN; | |
1193 | reg->smax_value = S64_MAX; | |
1194 | reg->umin_value = 0; | |
1195 | reg->umax_value = U64_MAX; | |
1196 | ||
1197 | reg->s32_min_value = S32_MIN; | |
1198 | reg->s32_max_value = S32_MAX; | |
1199 | reg->u32_min_value = 0; | |
1200 | reg->u32_max_value = U32_MAX; | |
1201 | } | |
1202 | ||
1203 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
1204 | { | |
1205 | reg->smin_value = S64_MIN; | |
1206 | reg->smax_value = S64_MAX; | |
1207 | reg->umin_value = 0; | |
1208 | reg->umax_value = U64_MAX; | |
1209 | } | |
1210 | ||
1211 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
1212 | { | |
1213 | reg->s32_min_value = S32_MIN; | |
1214 | reg->s32_max_value = S32_MAX; | |
1215 | reg->u32_min_value = 0; | |
1216 | reg->u32_max_value = U32_MAX; | |
1217 | } | |
1218 | ||
1219 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
1220 | { | |
1221 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
1222 | ||
1223 | /* min signed is max(sign bit) | min(other bits) */ | |
1224 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
1225 | var32_off.value | (var32_off.mask & S32_MIN)); | |
1226 | /* max signed is min(sign bit) | max(other bits) */ | |
1227 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
1228 | var32_off.value | (var32_off.mask & S32_MAX)); | |
1229 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
1230 | reg->u32_max_value = min(reg->u32_max_value, | |
1231 | (u32)(var32_off.value | var32_off.mask)); | |
1232 | } | |
1233 | ||
1234 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1235 | { |
1236 | /* min signed is max(sign bit) | min(other bits) */ | |
1237 | reg->smin_value = max_t(s64, reg->smin_value, | |
1238 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
1239 | /* max signed is min(sign bit) | max(other bits) */ | |
1240 | reg->smax_value = min_t(s64, reg->smax_value, | |
1241 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
1242 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
1243 | reg->umax_value = min(reg->umax_value, | |
1244 | reg->var_off.value | reg->var_off.mask); | |
1245 | } | |
1246 | ||
3f50f132 JF |
1247 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
1248 | { | |
1249 | __update_reg32_bounds(reg); | |
1250 | __update_reg64_bounds(reg); | |
1251 | } | |
1252 | ||
b03c9f9f | 1253 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
1254 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
1255 | { | |
1256 | /* Learn sign from signed bounds. | |
1257 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1258 | * are the same, so combine. This works even in the negative case, e.g. | |
1259 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1260 | */ | |
1261 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
1262 | reg->s32_min_value = reg->u32_min_value = | |
1263 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1264 | reg->s32_max_value = reg->u32_max_value = | |
1265 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1266 | return; | |
1267 | } | |
1268 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1269 | * boundary, so we must be careful. | |
1270 | */ | |
1271 | if ((s32)reg->u32_max_value >= 0) { | |
1272 | /* Positive. We can't learn anything from the smin, but smax | |
1273 | * is positive, hence safe. | |
1274 | */ | |
1275 | reg->s32_min_value = reg->u32_min_value; | |
1276 | reg->s32_max_value = reg->u32_max_value = | |
1277 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1278 | } else if ((s32)reg->u32_min_value < 0) { | |
1279 | /* Negative. We can't learn anything from the smax, but smin | |
1280 | * is negative, hence safe. | |
1281 | */ | |
1282 | reg->s32_min_value = reg->u32_min_value = | |
1283 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1284 | reg->s32_max_value = reg->u32_max_value; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1289 | { |
1290 | /* Learn sign from signed bounds. | |
1291 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1292 | * are the same, so combine. This works even in the negative case, e.g. | |
1293 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1294 | */ | |
1295 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
1296 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1297 | reg->umin_value); | |
1298 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1299 | reg->umax_value); | |
1300 | return; | |
1301 | } | |
1302 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1303 | * boundary, so we must be careful. | |
1304 | */ | |
1305 | if ((s64)reg->umax_value >= 0) { | |
1306 | /* Positive. We can't learn anything from the smin, but smax | |
1307 | * is positive, hence safe. | |
1308 | */ | |
1309 | reg->smin_value = reg->umin_value; | |
1310 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1311 | reg->umax_value); | |
1312 | } else if ((s64)reg->umin_value < 0) { | |
1313 | /* Negative. We can't learn anything from the smax, but smin | |
1314 | * is negative, hence safe. | |
1315 | */ | |
1316 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1317 | reg->umin_value); | |
1318 | reg->smax_value = reg->umax_value; | |
1319 | } | |
1320 | } | |
1321 | ||
3f50f132 JF |
1322 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
1323 | { | |
1324 | __reg32_deduce_bounds(reg); | |
1325 | __reg64_deduce_bounds(reg); | |
1326 | } | |
1327 | ||
b03c9f9f EC |
1328 | /* Attempts to improve var_off based on unsigned min/max information */ |
1329 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
1330 | { | |
3f50f132 JF |
1331 | struct tnum var64_off = tnum_intersect(reg->var_off, |
1332 | tnum_range(reg->umin_value, | |
1333 | reg->umax_value)); | |
1334 | struct tnum var32_off = tnum_intersect(tnum_subreg(reg->var_off), | |
1335 | tnum_range(reg->u32_min_value, | |
1336 | reg->u32_max_value)); | |
1337 | ||
1338 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
1339 | } |
1340 | ||
3f50f132 | 1341 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 1342 | { |
3f50f132 JF |
1343 | reg->umin_value = reg->u32_min_value; |
1344 | reg->umax_value = reg->u32_max_value; | |
1345 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds | |
1346 | * but must be positive otherwise set to worse case bounds | |
1347 | * and refine later from tnum. | |
1348 | */ | |
3a71dc36 | 1349 | if (reg->s32_min_value >= 0 && reg->s32_max_value >= 0) |
3f50f132 JF |
1350 | reg->smax_value = reg->s32_max_value; |
1351 | else | |
1352 | reg->smax_value = U32_MAX; | |
3a71dc36 JF |
1353 | if (reg->s32_min_value >= 0) |
1354 | reg->smin_value = reg->s32_min_value; | |
1355 | else | |
1356 | reg->smin_value = 0; | |
3f50f132 JF |
1357 | } |
1358 | ||
1359 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
1360 | { | |
1361 | /* special case when 64-bit register has upper 32-bit register | |
1362 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
1363 | * allowing us to use 32-bit bounds directly, | |
1364 | */ | |
1365 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
1366 | __reg_assign_32_into_64(reg); | |
1367 | } else { | |
1368 | /* Otherwise the best we can do is push lower 32bit known and | |
1369 | * unknown bits into register (var_off set from jmp logic) | |
1370 | * then learn as much as possible from the 64-bit tnum | |
1371 | * known and unknown bits. The previous smin/smax bounds are | |
1372 | * invalid here because of jmp32 compare so mark them unknown | |
1373 | * so they do not impact tnum bounds calculation. | |
1374 | */ | |
1375 | __mark_reg64_unbounded(reg); | |
1376 | __update_reg_bounds(reg); | |
1377 | } | |
1378 | ||
1379 | /* Intersecting with the old var_off might have improved our bounds | |
1380 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1381 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1382 | */ | |
1383 | __reg_deduce_bounds(reg); | |
1384 | __reg_bound_offset(reg); | |
1385 | __update_reg_bounds(reg); | |
1386 | } | |
1387 | ||
1388 | static bool __reg64_bound_s32(s64 a) | |
1389 | { | |
b0270958 | 1390 | return a > S32_MIN && a < S32_MAX; |
3f50f132 JF |
1391 | } |
1392 | ||
1393 | static bool __reg64_bound_u32(u64 a) | |
1394 | { | |
1395 | if (a > U32_MIN && a < U32_MAX) | |
1396 | return true; | |
1397 | return false; | |
1398 | } | |
1399 | ||
1400 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
1401 | { | |
1402 | __mark_reg32_unbounded(reg); | |
1403 | ||
b0270958 | 1404 | if (__reg64_bound_s32(reg->smin_value) && __reg64_bound_s32(reg->smax_value)) { |
3f50f132 | 1405 | reg->s32_min_value = (s32)reg->smin_value; |
3f50f132 | 1406 | reg->s32_max_value = (s32)reg->smax_value; |
b0270958 | 1407 | } |
3f50f132 JF |
1408 | if (__reg64_bound_u32(reg->umin_value)) |
1409 | reg->u32_min_value = (u32)reg->umin_value; | |
1410 | if (__reg64_bound_u32(reg->umax_value)) | |
1411 | reg->u32_max_value = (u32)reg->umax_value; | |
1412 | ||
1413 | /* Intersecting with the old var_off might have improved our bounds | |
1414 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1415 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1416 | */ | |
1417 | __reg_deduce_bounds(reg); | |
1418 | __reg_bound_offset(reg); | |
1419 | __update_reg_bounds(reg); | |
b03c9f9f EC |
1420 | } |
1421 | ||
f1174f77 | 1422 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
1423 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1424 | struct bpf_reg_state *reg) | |
f1174f77 | 1425 | { |
a9c676bc AS |
1426 | /* |
1427 | * Clear type, id, off, and union(map_ptr, range) and | |
1428 | * padding between 'type' and union | |
1429 | */ | |
1430 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 1431 | reg->type = SCALAR_VALUE; |
f1174f77 | 1432 | reg->var_off = tnum_unknown; |
f4d7e40a | 1433 | reg->frameno = 0; |
2c78ee89 | 1434 | reg->precise = env->subprog_cnt > 1 || !env->bpf_capable; |
b03c9f9f | 1435 | __mark_reg_unbounded(reg); |
f1174f77 EC |
1436 | } |
1437 | ||
61bd5218 JK |
1438 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
1439 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1440 | { |
1441 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1442 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
1443 | /* Something bad happened, let's kill all regs except FP */ |
1444 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1445 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1446 | return; |
1447 | } | |
f54c7898 | 1448 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
1449 | } |
1450 | ||
f54c7898 DB |
1451 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1452 | struct bpf_reg_state *reg) | |
f1174f77 | 1453 | { |
f54c7898 | 1454 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
1455 | reg->type = NOT_INIT; |
1456 | } | |
1457 | ||
61bd5218 JK |
1458 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
1459 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1460 | { |
1461 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1462 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
1463 | /* Something bad happened, let's kill all regs except FP */ |
1464 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1465 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1466 | return; |
1467 | } | |
f54c7898 | 1468 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
1469 | } |
1470 | ||
41c48f3a AI |
1471 | static void mark_btf_ld_reg(struct bpf_verifier_env *env, |
1472 | struct bpf_reg_state *regs, u32 regno, | |
22dc4a0f AN |
1473 | enum bpf_reg_type reg_type, |
1474 | struct btf *btf, u32 btf_id) | |
41c48f3a AI |
1475 | { |
1476 | if (reg_type == SCALAR_VALUE) { | |
1477 | mark_reg_unknown(env, regs, regno); | |
1478 | return; | |
1479 | } | |
1480 | mark_reg_known_zero(env, regs, regno); | |
1481 | regs[regno].type = PTR_TO_BTF_ID; | |
22dc4a0f | 1482 | regs[regno].btf = btf; |
41c48f3a AI |
1483 | regs[regno].btf_id = btf_id; |
1484 | } | |
1485 | ||
5327ed3d | 1486 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 1487 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 1488 | struct bpf_func_state *state) |
17a52670 | 1489 | { |
f4d7e40a | 1490 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
1491 | int i; |
1492 | ||
dc503a8a | 1493 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 1494 | mark_reg_not_init(env, regs, i); |
dc503a8a | 1495 | regs[i].live = REG_LIVE_NONE; |
679c782d | 1496 | regs[i].parent = NULL; |
5327ed3d | 1497 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 1498 | } |
17a52670 AS |
1499 | |
1500 | /* frame pointer */ | |
f1174f77 | 1501 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 1502 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 1503 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
1504 | } |
1505 | ||
f4d7e40a AS |
1506 | #define BPF_MAIN_FUNC (-1) |
1507 | static void init_func_state(struct bpf_verifier_env *env, | |
1508 | struct bpf_func_state *state, | |
1509 | int callsite, int frameno, int subprogno) | |
1510 | { | |
1511 | state->callsite = callsite; | |
1512 | state->frameno = frameno; | |
1513 | state->subprogno = subprogno; | |
1514 | init_reg_state(env, state); | |
1515 | } | |
1516 | ||
17a52670 AS |
1517 | enum reg_arg_type { |
1518 | SRC_OP, /* register is used as source operand */ | |
1519 | DST_OP, /* register is used as destination operand */ | |
1520 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1521 | }; | |
1522 | ||
cc8b0b92 AS |
1523 | static int cmp_subprogs(const void *a, const void *b) |
1524 | { | |
9c8105bd JW |
1525 | return ((struct bpf_subprog_info *)a)->start - |
1526 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1527 | } |
1528 | ||
1529 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1530 | { | |
9c8105bd | 1531 | struct bpf_subprog_info *p; |
cc8b0b92 | 1532 | |
9c8105bd JW |
1533 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1534 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1535 | if (!p) |
1536 | return -ENOENT; | |
9c8105bd | 1537 | return p - env->subprog_info; |
cc8b0b92 AS |
1538 | |
1539 | } | |
1540 | ||
1541 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1542 | { | |
1543 | int insn_cnt = env->prog->len; | |
1544 | int ret; | |
1545 | ||
1546 | if (off >= insn_cnt || off < 0) { | |
1547 | verbose(env, "call to invalid destination\n"); | |
1548 | return -EINVAL; | |
1549 | } | |
1550 | ret = find_subprog(env, off); | |
1551 | if (ret >= 0) | |
282a0f46 | 1552 | return ret; |
4cb3d99c | 1553 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1554 | verbose(env, "too many subprograms\n"); |
1555 | return -E2BIG; | |
1556 | } | |
9c8105bd JW |
1557 | env->subprog_info[env->subprog_cnt++].start = off; |
1558 | sort(env->subprog_info, env->subprog_cnt, | |
1559 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
282a0f46 | 1560 | return env->subprog_cnt - 1; |
cc8b0b92 AS |
1561 | } |
1562 | ||
1563 | static int check_subprogs(struct bpf_verifier_env *env) | |
1564 | { | |
1565 | int i, ret, subprog_start, subprog_end, off, cur_subprog = 0; | |
9c8105bd | 1566 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 AS |
1567 | struct bpf_insn *insn = env->prog->insnsi; |
1568 | int insn_cnt = env->prog->len; | |
1569 | ||
f910cefa JW |
1570 | /* Add entry function. */ |
1571 | ret = add_subprog(env, 0); | |
1572 | if (ret < 0) | |
1573 | return ret; | |
1574 | ||
cc8b0b92 AS |
1575 | /* determine subprog starts. The end is one before the next starts */ |
1576 | for (i = 0; i < insn_cnt; i++) { | |
69c087ba YS |
1577 | if (bpf_pseudo_func(insn + i)) { |
1578 | if (!env->bpf_capable) { | |
1579 | verbose(env, | |
1580 | "function pointers are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); | |
1581 | return -EPERM; | |
1582 | } | |
1583 | ret = add_subprog(env, i + insn[i].imm + 1); | |
1584 | if (ret < 0) | |
1585 | return ret; | |
1586 | /* remember subprog */ | |
1587 | insn[i + 1].imm = ret; | |
1588 | continue; | |
1589 | } | |
23a2d70c | 1590 | if (!bpf_pseudo_call(insn + i)) |
cc8b0b92 | 1591 | continue; |
2c78ee89 AS |
1592 | if (!env->bpf_capable) { |
1593 | verbose(env, | |
1594 | "function calls to other bpf functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); | |
cc8b0b92 AS |
1595 | return -EPERM; |
1596 | } | |
cc8b0b92 AS |
1597 | ret = add_subprog(env, i + insn[i].imm + 1); |
1598 | if (ret < 0) | |
1599 | return ret; | |
1600 | } | |
1601 | ||
4cb3d99c JW |
1602 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
1603 | * logic. 'subprog_cnt' should not be increased. | |
1604 | */ | |
1605 | subprog[env->subprog_cnt].start = insn_cnt; | |
1606 | ||
06ee7115 | 1607 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 1608 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 1609 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 AS |
1610 | |
1611 | /* now check that all jumps are within the same subprog */ | |
4cb3d99c JW |
1612 | subprog_start = subprog[cur_subprog].start; |
1613 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
1614 | for (i = 0; i < insn_cnt; i++) { |
1615 | u8 code = insn[i].code; | |
1616 | ||
7f6e4312 MF |
1617 | if (code == (BPF_JMP | BPF_CALL) && |
1618 | insn[i].imm == BPF_FUNC_tail_call && | |
1619 | insn[i].src_reg != BPF_PSEUDO_CALL) | |
1620 | subprog[cur_subprog].has_tail_call = true; | |
09b28d76 AS |
1621 | if (BPF_CLASS(code) == BPF_LD && |
1622 | (BPF_MODE(code) == BPF_ABS || BPF_MODE(code) == BPF_IND)) | |
1623 | subprog[cur_subprog].has_ld_abs = true; | |
092ed096 | 1624 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
1625 | goto next; |
1626 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
1627 | goto next; | |
1628 | off = i + insn[i].off + 1; | |
1629 | if (off < subprog_start || off >= subprog_end) { | |
1630 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
1631 | return -EINVAL; | |
1632 | } | |
1633 | next: | |
1634 | if (i == subprog_end - 1) { | |
1635 | /* to avoid fall-through from one subprog into another | |
1636 | * the last insn of the subprog should be either exit | |
1637 | * or unconditional jump back | |
1638 | */ | |
1639 | if (code != (BPF_JMP | BPF_EXIT) && | |
1640 | code != (BPF_JMP | BPF_JA)) { | |
1641 | verbose(env, "last insn is not an exit or jmp\n"); | |
1642 | return -EINVAL; | |
1643 | } | |
1644 | subprog_start = subprog_end; | |
4cb3d99c JW |
1645 | cur_subprog++; |
1646 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 1647 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
1648 | } |
1649 | } | |
1650 | return 0; | |
1651 | } | |
1652 | ||
679c782d EC |
1653 | /* Parentage chain of this register (or stack slot) should take care of all |
1654 | * issues like callee-saved registers, stack slot allocation time, etc. | |
1655 | */ | |
f4d7e40a | 1656 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 1657 | const struct bpf_reg_state *state, |
5327ed3d | 1658 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
1659 | { |
1660 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 1661 | int cnt = 0; |
dc503a8a EC |
1662 | |
1663 | while (parent) { | |
1664 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 1665 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 1666 | break; |
9242b5f5 AS |
1667 | if (parent->live & REG_LIVE_DONE) { |
1668 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
1669 | reg_type_str[parent->type], | |
1670 | parent->var_off.value, parent->off); | |
1671 | return -EFAULT; | |
1672 | } | |
5327ed3d JW |
1673 | /* The first condition is more likely to be true than the |
1674 | * second, checked it first. | |
1675 | */ | |
1676 | if ((parent->live & REG_LIVE_READ) == flag || | |
1677 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
1678 | /* The parentage chain never changes and |
1679 | * this parent was already marked as LIVE_READ. | |
1680 | * There is no need to keep walking the chain again and | |
1681 | * keep re-marking all parents as LIVE_READ. | |
1682 | * This case happens when the same register is read | |
1683 | * multiple times without writes into it in-between. | |
5327ed3d JW |
1684 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
1685 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
1686 | */ |
1687 | break; | |
dc503a8a | 1688 | /* ... then we depend on parent's value */ |
5327ed3d JW |
1689 | parent->live |= flag; |
1690 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
1691 | if (flag == REG_LIVE_READ64) | |
1692 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
1693 | state = parent; |
1694 | parent = state->parent; | |
f4d7e40a | 1695 | writes = true; |
06ee7115 | 1696 | cnt++; |
dc503a8a | 1697 | } |
06ee7115 AS |
1698 | |
1699 | if (env->longest_mark_read_walk < cnt) | |
1700 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 1701 | return 0; |
dc503a8a EC |
1702 | } |
1703 | ||
5327ed3d JW |
1704 | /* This function is supposed to be used by the following 32-bit optimization |
1705 | * code only. It returns TRUE if the source or destination register operates | |
1706 | * on 64-bit, otherwise return FALSE. | |
1707 | */ | |
1708 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
1709 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
1710 | { | |
1711 | u8 code, class, op; | |
1712 | ||
1713 | code = insn->code; | |
1714 | class = BPF_CLASS(code); | |
1715 | op = BPF_OP(code); | |
1716 | if (class == BPF_JMP) { | |
1717 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
1718 | * conservatively. | |
1719 | */ | |
1720 | if (op == BPF_EXIT) | |
1721 | return true; | |
1722 | if (op == BPF_CALL) { | |
1723 | /* BPF to BPF call will reach here because of marking | |
1724 | * caller saved clobber with DST_OP_NO_MARK for which we | |
1725 | * don't care the register def because they are anyway | |
1726 | * marked as NOT_INIT already. | |
1727 | */ | |
1728 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
1729 | return false; | |
1730 | /* Helper call will reach here because of arg type | |
1731 | * check, conservatively return TRUE. | |
1732 | */ | |
1733 | if (t == SRC_OP) | |
1734 | return true; | |
1735 | ||
1736 | return false; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | if (class == BPF_ALU64 || class == BPF_JMP || | |
1741 | /* BPF_END always use BPF_ALU class. */ | |
1742 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) | |
1743 | return true; | |
1744 | ||
1745 | if (class == BPF_ALU || class == BPF_JMP32) | |
1746 | return false; | |
1747 | ||
1748 | if (class == BPF_LDX) { | |
1749 | if (t != SRC_OP) | |
1750 | return BPF_SIZE(code) == BPF_DW; | |
1751 | /* LDX source must be ptr. */ | |
1752 | return true; | |
1753 | } | |
1754 | ||
1755 | if (class == BPF_STX) { | |
83a28819 IL |
1756 | /* BPF_STX (including atomic variants) has multiple source |
1757 | * operands, one of which is a ptr. Check whether the caller is | |
1758 | * asking about it. | |
1759 | */ | |
1760 | if (t == SRC_OP && reg->type != SCALAR_VALUE) | |
5327ed3d JW |
1761 | return true; |
1762 | return BPF_SIZE(code) == BPF_DW; | |
1763 | } | |
1764 | ||
1765 | if (class == BPF_LD) { | |
1766 | u8 mode = BPF_MODE(code); | |
1767 | ||
1768 | /* LD_IMM64 */ | |
1769 | if (mode == BPF_IMM) | |
1770 | return true; | |
1771 | ||
1772 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
1773 | if (t != SRC_OP) | |
1774 | return false; | |
1775 | ||
1776 | /* Implicit ctx ptr. */ | |
1777 | if (regno == BPF_REG_6) | |
1778 | return true; | |
1779 | ||
1780 | /* Explicit source could be any width. */ | |
1781 | return true; | |
1782 | } | |
1783 | ||
1784 | if (class == BPF_ST) | |
1785 | /* The only source register for BPF_ST is a ptr. */ | |
1786 | return true; | |
1787 | ||
1788 | /* Conservatively return true at default. */ | |
1789 | return true; | |
1790 | } | |
1791 | ||
83a28819 IL |
1792 | /* Return the regno defined by the insn, or -1. */ |
1793 | static int insn_def_regno(const struct bpf_insn *insn) | |
b325fbca | 1794 | { |
83a28819 IL |
1795 | switch (BPF_CLASS(insn->code)) { |
1796 | case BPF_JMP: | |
1797 | case BPF_JMP32: | |
1798 | case BPF_ST: | |
1799 | return -1; | |
1800 | case BPF_STX: | |
1801 | if (BPF_MODE(insn->code) == BPF_ATOMIC && | |
1802 | (insn->imm & BPF_FETCH)) { | |
1803 | if (insn->imm == BPF_CMPXCHG) | |
1804 | return BPF_REG_0; | |
1805 | else | |
1806 | return insn->src_reg; | |
1807 | } else { | |
1808 | return -1; | |
1809 | } | |
1810 | default: | |
1811 | return insn->dst_reg; | |
1812 | } | |
b325fbca JW |
1813 | } |
1814 | ||
1815 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
1816 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
1817 | { | |
83a28819 IL |
1818 | int dst_reg = insn_def_regno(insn); |
1819 | ||
1820 | if (dst_reg == -1) | |
b325fbca JW |
1821 | return false; |
1822 | ||
83a28819 | 1823 | return !is_reg64(env, insn, dst_reg, NULL, DST_OP); |
b325fbca JW |
1824 | } |
1825 | ||
5327ed3d JW |
1826 | static void mark_insn_zext(struct bpf_verifier_env *env, |
1827 | struct bpf_reg_state *reg) | |
1828 | { | |
1829 | s32 def_idx = reg->subreg_def; | |
1830 | ||
1831 | if (def_idx == DEF_NOT_SUBREG) | |
1832 | return; | |
1833 | ||
1834 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
1835 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
1836 | reg->subreg_def = DEF_NOT_SUBREG; | |
1837 | } | |
1838 | ||
dc503a8a | 1839 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
1840 | enum reg_arg_type t) |
1841 | { | |
f4d7e40a AS |
1842 | struct bpf_verifier_state *vstate = env->cur_state; |
1843 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 1844 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 1845 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 1846 | bool rw64; |
dc503a8a | 1847 | |
17a52670 | 1848 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 1849 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
1850 | return -EINVAL; |
1851 | } | |
1852 | ||
c342dc10 | 1853 | reg = ®s[regno]; |
5327ed3d | 1854 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
1855 | if (t == SRC_OP) { |
1856 | /* check whether register used as source operand can be read */ | |
c342dc10 | 1857 | if (reg->type == NOT_INIT) { |
61bd5218 | 1858 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
1859 | return -EACCES; |
1860 | } | |
679c782d | 1861 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
1862 | if (regno == BPF_REG_FP) |
1863 | return 0; | |
1864 | ||
5327ed3d JW |
1865 | if (rw64) |
1866 | mark_insn_zext(env, reg); | |
1867 | ||
1868 | return mark_reg_read(env, reg, reg->parent, | |
1869 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
1870 | } else { |
1871 | /* check whether register used as dest operand can be written to */ | |
1872 | if (regno == BPF_REG_FP) { | |
61bd5218 | 1873 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
1874 | return -EACCES; |
1875 | } | |
c342dc10 | 1876 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 1877 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 1878 | if (t == DST_OP) |
61bd5218 | 1879 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
1880 | } |
1881 | return 0; | |
1882 | } | |
1883 | ||
b5dc0163 AS |
1884 | /* for any branch, call, exit record the history of jmps in the given state */ |
1885 | static int push_jmp_history(struct bpf_verifier_env *env, | |
1886 | struct bpf_verifier_state *cur) | |
1887 | { | |
1888 | u32 cnt = cur->jmp_history_cnt; | |
1889 | struct bpf_idx_pair *p; | |
1890 | ||
1891 | cnt++; | |
1892 | p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER); | |
1893 | if (!p) | |
1894 | return -ENOMEM; | |
1895 | p[cnt - 1].idx = env->insn_idx; | |
1896 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
1897 | cur->jmp_history = p; | |
1898 | cur->jmp_history_cnt = cnt; | |
1899 | return 0; | |
1900 | } | |
1901 | ||
1902 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
1903 | * history then previous instruction came from straight line execution. | |
1904 | */ | |
1905 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
1906 | u32 *history) | |
1907 | { | |
1908 | u32 cnt = *history; | |
1909 | ||
1910 | if (cnt && st->jmp_history[cnt - 1].idx == i) { | |
1911 | i = st->jmp_history[cnt - 1].prev_idx; | |
1912 | (*history)--; | |
1913 | } else { | |
1914 | i--; | |
1915 | } | |
1916 | return i; | |
1917 | } | |
1918 | ||
1919 | /* For given verifier state backtrack_insn() is called from the last insn to | |
1920 | * the first insn. Its purpose is to compute a bitmask of registers and | |
1921 | * stack slots that needs precision in the parent verifier state. | |
1922 | */ | |
1923 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, | |
1924 | u32 *reg_mask, u64 *stack_mask) | |
1925 | { | |
1926 | const struct bpf_insn_cbs cbs = { | |
1927 | .cb_print = verbose, | |
1928 | .private_data = env, | |
1929 | }; | |
1930 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
1931 | u8 class = BPF_CLASS(insn->code); | |
1932 | u8 opcode = BPF_OP(insn->code); | |
1933 | u8 mode = BPF_MODE(insn->code); | |
1934 | u32 dreg = 1u << insn->dst_reg; | |
1935 | u32 sreg = 1u << insn->src_reg; | |
1936 | u32 spi; | |
1937 | ||
1938 | if (insn->code == 0) | |
1939 | return 0; | |
1940 | if (env->log.level & BPF_LOG_LEVEL) { | |
1941 | verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask); | |
1942 | verbose(env, "%d: ", idx); | |
1943 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
1944 | } | |
1945 | ||
1946 | if (class == BPF_ALU || class == BPF_ALU64) { | |
1947 | if (!(*reg_mask & dreg)) | |
1948 | return 0; | |
1949 | if (opcode == BPF_MOV) { | |
1950 | if (BPF_SRC(insn->code) == BPF_X) { | |
1951 | /* dreg = sreg | |
1952 | * dreg needs precision after this insn | |
1953 | * sreg needs precision before this insn | |
1954 | */ | |
1955 | *reg_mask &= ~dreg; | |
1956 | *reg_mask |= sreg; | |
1957 | } else { | |
1958 | /* dreg = K | |
1959 | * dreg needs precision after this insn. | |
1960 | * Corresponding register is already marked | |
1961 | * as precise=true in this verifier state. | |
1962 | * No further markings in parent are necessary | |
1963 | */ | |
1964 | *reg_mask &= ~dreg; | |
1965 | } | |
1966 | } else { | |
1967 | if (BPF_SRC(insn->code) == BPF_X) { | |
1968 | /* dreg += sreg | |
1969 | * both dreg and sreg need precision | |
1970 | * before this insn | |
1971 | */ | |
1972 | *reg_mask |= sreg; | |
1973 | } /* else dreg += K | |
1974 | * dreg still needs precision before this insn | |
1975 | */ | |
1976 | } | |
1977 | } else if (class == BPF_LDX) { | |
1978 | if (!(*reg_mask & dreg)) | |
1979 | return 0; | |
1980 | *reg_mask &= ~dreg; | |
1981 | ||
1982 | /* scalars can only be spilled into stack w/o losing precision. | |
1983 | * Load from any other memory can be zero extended. | |
1984 | * The desire to keep that precision is already indicated | |
1985 | * by 'precise' mark in corresponding register of this state. | |
1986 | * No further tracking necessary. | |
1987 | */ | |
1988 | if (insn->src_reg != BPF_REG_FP) | |
1989 | return 0; | |
1990 | if (BPF_SIZE(insn->code) != BPF_DW) | |
1991 | return 0; | |
1992 | ||
1993 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
1994 | * that [fp - off] slot contains scalar that needs to be | |
1995 | * tracked with precision | |
1996 | */ | |
1997 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
1998 | if (spi >= 64) { | |
1999 | verbose(env, "BUG spi %d\n", spi); | |
2000 | WARN_ONCE(1, "verifier backtracking bug"); | |
2001 | return -EFAULT; | |
2002 | } | |
2003 | *stack_mask |= 1ull << spi; | |
b3b50f05 | 2004 | } else if (class == BPF_STX || class == BPF_ST) { |
b5dc0163 | 2005 | if (*reg_mask & dreg) |
b3b50f05 | 2006 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
2007 | * to access memory. It means backtracking |
2008 | * encountered a case of pointer subtraction. | |
2009 | */ | |
2010 | return -ENOTSUPP; | |
2011 | /* scalars can only be spilled into stack */ | |
2012 | if (insn->dst_reg != BPF_REG_FP) | |
2013 | return 0; | |
2014 | if (BPF_SIZE(insn->code) != BPF_DW) | |
2015 | return 0; | |
2016 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
2017 | if (spi >= 64) { | |
2018 | verbose(env, "BUG spi %d\n", spi); | |
2019 | WARN_ONCE(1, "verifier backtracking bug"); | |
2020 | return -EFAULT; | |
2021 | } | |
2022 | if (!(*stack_mask & (1ull << spi))) | |
2023 | return 0; | |
2024 | *stack_mask &= ~(1ull << spi); | |
b3b50f05 AN |
2025 | if (class == BPF_STX) |
2026 | *reg_mask |= sreg; | |
b5dc0163 AS |
2027 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
2028 | if (opcode == BPF_CALL) { | |
2029 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
2030 | return -ENOTSUPP; | |
2031 | /* regular helper call sets R0 */ | |
2032 | *reg_mask &= ~1; | |
2033 | if (*reg_mask & 0x3f) { | |
2034 | /* if backtracing was looking for registers R1-R5 | |
2035 | * they should have been found already. | |
2036 | */ | |
2037 | verbose(env, "BUG regs %x\n", *reg_mask); | |
2038 | WARN_ONCE(1, "verifier backtracking bug"); | |
2039 | return -EFAULT; | |
2040 | } | |
2041 | } else if (opcode == BPF_EXIT) { | |
2042 | return -ENOTSUPP; | |
2043 | } | |
2044 | } else if (class == BPF_LD) { | |
2045 | if (!(*reg_mask & dreg)) | |
2046 | return 0; | |
2047 | *reg_mask &= ~dreg; | |
2048 | /* It's ld_imm64 or ld_abs or ld_ind. | |
2049 | * For ld_imm64 no further tracking of precision | |
2050 | * into parent is necessary | |
2051 | */ | |
2052 | if (mode == BPF_IND || mode == BPF_ABS) | |
2053 | /* to be analyzed */ | |
2054 | return -ENOTSUPP; | |
b5dc0163 AS |
2055 | } |
2056 | return 0; | |
2057 | } | |
2058 | ||
2059 | /* the scalar precision tracking algorithm: | |
2060 | * . at the start all registers have precise=false. | |
2061 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
2062 | * . once precise value of the scalar register is used in: | |
2063 | * . ptr + scalar alu | |
2064 | * . if (scalar cond K|scalar) | |
2065 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
2066 | * backtrack through the verifier states and mark all registers and | |
2067 | * stack slots with spilled constants that these scalar regisers | |
2068 | * should be precise. | |
2069 | * . during state pruning two registers (or spilled stack slots) | |
2070 | * are equivalent if both are not precise. | |
2071 | * | |
2072 | * Note the verifier cannot simply walk register parentage chain, | |
2073 | * since many different registers and stack slots could have been | |
2074 | * used to compute single precise scalar. | |
2075 | * | |
2076 | * The approach of starting with precise=true for all registers and then | |
2077 | * backtrack to mark a register as not precise when the verifier detects | |
2078 | * that program doesn't care about specific value (e.g., when helper | |
2079 | * takes register as ARG_ANYTHING parameter) is not safe. | |
2080 | * | |
2081 | * It's ok to walk single parentage chain of the verifier states. | |
2082 | * It's possible that this backtracking will go all the way till 1st insn. | |
2083 | * All other branches will be explored for needing precision later. | |
2084 | * | |
2085 | * The backtracking needs to deal with cases like: | |
2086 | * 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) | |
2087 | * r9 -= r8 | |
2088 | * r5 = r9 | |
2089 | * if r5 > 0x79f goto pc+7 | |
2090 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
2091 | * r5 += 1 | |
2092 | * ... | |
2093 | * call bpf_perf_event_output#25 | |
2094 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
2095 | * | |
2096 | * and this case: | |
2097 | * r6 = 1 | |
2098 | * call foo // uses callee's r6 inside to compute r0 | |
2099 | * r0 += r6 | |
2100 | * if r0 == 0 goto | |
2101 | * | |
2102 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
2103 | * | |
2104 | * Also if parent's curframe > frame where backtracking started, | |
2105 | * the verifier need to mark registers in both frames, otherwise callees | |
2106 | * may incorrectly prune callers. This is similar to | |
2107 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
2108 | * | |
2109 | * For now backtracking falls back into conservative marking. | |
2110 | */ | |
2111 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
2112 | struct bpf_verifier_state *st) | |
2113 | { | |
2114 | struct bpf_func_state *func; | |
2115 | struct bpf_reg_state *reg; | |
2116 | int i, j; | |
2117 | ||
2118 | /* big hammer: mark all scalars precise in this path. | |
2119 | * pop_stack may still get !precise scalars. | |
2120 | */ | |
2121 | for (; st; st = st->parent) | |
2122 | for (i = 0; i <= st->curframe; i++) { | |
2123 | func = st->frame[i]; | |
2124 | for (j = 0; j < BPF_REG_FP; j++) { | |
2125 | reg = &func->regs[j]; | |
2126 | if (reg->type != SCALAR_VALUE) | |
2127 | continue; | |
2128 | reg->precise = true; | |
2129 | } | |
2130 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
2131 | if (func->stack[j].slot_type[0] != STACK_SPILL) | |
2132 | continue; | |
2133 | reg = &func->stack[j].spilled_ptr; | |
2134 | if (reg->type != SCALAR_VALUE) | |
2135 | continue; | |
2136 | reg->precise = true; | |
2137 | } | |
2138 | } | |
2139 | } | |
2140 | ||
a3ce685d AS |
2141 | static int __mark_chain_precision(struct bpf_verifier_env *env, int regno, |
2142 | int spi) | |
b5dc0163 AS |
2143 | { |
2144 | struct bpf_verifier_state *st = env->cur_state; | |
2145 | int first_idx = st->first_insn_idx; | |
2146 | int last_idx = env->insn_idx; | |
2147 | struct bpf_func_state *func; | |
2148 | struct bpf_reg_state *reg; | |
a3ce685d AS |
2149 | u32 reg_mask = regno >= 0 ? 1u << regno : 0; |
2150 | u64 stack_mask = spi >= 0 ? 1ull << spi : 0; | |
b5dc0163 | 2151 | bool skip_first = true; |
a3ce685d | 2152 | bool new_marks = false; |
b5dc0163 AS |
2153 | int i, err; |
2154 | ||
2c78ee89 | 2155 | if (!env->bpf_capable) |
b5dc0163 AS |
2156 | return 0; |
2157 | ||
2158 | func = st->frame[st->curframe]; | |
a3ce685d AS |
2159 | if (regno >= 0) { |
2160 | reg = &func->regs[regno]; | |
2161 | if (reg->type != SCALAR_VALUE) { | |
2162 | WARN_ONCE(1, "backtracing misuse"); | |
2163 | return -EFAULT; | |
2164 | } | |
2165 | if (!reg->precise) | |
2166 | new_marks = true; | |
2167 | else | |
2168 | reg_mask = 0; | |
2169 | reg->precise = true; | |
b5dc0163 | 2170 | } |
b5dc0163 | 2171 | |
a3ce685d AS |
2172 | while (spi >= 0) { |
2173 | if (func->stack[spi].slot_type[0] != STACK_SPILL) { | |
2174 | stack_mask = 0; | |
2175 | break; | |
2176 | } | |
2177 | reg = &func->stack[spi].spilled_ptr; | |
2178 | if (reg->type != SCALAR_VALUE) { | |
2179 | stack_mask = 0; | |
2180 | break; | |
2181 | } | |
2182 | if (!reg->precise) | |
2183 | new_marks = true; | |
2184 | else | |
2185 | stack_mask = 0; | |
2186 | reg->precise = true; | |
2187 | break; | |
2188 | } | |
2189 | ||
2190 | if (!new_marks) | |
2191 | return 0; | |
2192 | if (!reg_mask && !stack_mask) | |
2193 | return 0; | |
b5dc0163 AS |
2194 | for (;;) { |
2195 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
2196 | u32 history = st->jmp_history_cnt; |
2197 | ||
2198 | if (env->log.level & BPF_LOG_LEVEL) | |
2199 | verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx); | |
2200 | for (i = last_idx;;) { | |
2201 | if (skip_first) { | |
2202 | err = 0; | |
2203 | skip_first = false; | |
2204 | } else { | |
2205 | err = backtrack_insn(env, i, ®_mask, &stack_mask); | |
2206 | } | |
2207 | if (err == -ENOTSUPP) { | |
2208 | mark_all_scalars_precise(env, st); | |
2209 | return 0; | |
2210 | } else if (err) { | |
2211 | return err; | |
2212 | } | |
2213 | if (!reg_mask && !stack_mask) | |
2214 | /* Found assignment(s) into tracked register in this state. | |
2215 | * Since this state is already marked, just return. | |
2216 | * Nothing to be tracked further in the parent state. | |
2217 | */ | |
2218 | return 0; | |
2219 | if (i == first_idx) | |
2220 | break; | |
2221 | i = get_prev_insn_idx(st, i, &history); | |
2222 | if (i >= env->prog->len) { | |
2223 | /* This can happen if backtracking reached insn 0 | |
2224 | * and there are still reg_mask or stack_mask | |
2225 | * to backtrack. | |
2226 | * It means the backtracking missed the spot where | |
2227 | * particular register was initialized with a constant. | |
2228 | */ | |
2229 | verbose(env, "BUG backtracking idx %d\n", i); | |
2230 | WARN_ONCE(1, "verifier backtracking bug"); | |
2231 | return -EFAULT; | |
2232 | } | |
2233 | } | |
2234 | st = st->parent; | |
2235 | if (!st) | |
2236 | break; | |
2237 | ||
a3ce685d | 2238 | new_marks = false; |
b5dc0163 AS |
2239 | func = st->frame[st->curframe]; |
2240 | bitmap_from_u64(mask, reg_mask); | |
2241 | for_each_set_bit(i, mask, 32) { | |
2242 | reg = &func->regs[i]; | |
a3ce685d AS |
2243 | if (reg->type != SCALAR_VALUE) { |
2244 | reg_mask &= ~(1u << i); | |
b5dc0163 | 2245 | continue; |
a3ce685d | 2246 | } |
b5dc0163 AS |
2247 | if (!reg->precise) |
2248 | new_marks = true; | |
2249 | reg->precise = true; | |
2250 | } | |
2251 | ||
2252 | bitmap_from_u64(mask, stack_mask); | |
2253 | for_each_set_bit(i, mask, 64) { | |
2254 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
2339cd6c AS |
2255 | /* the sequence of instructions: |
2256 | * 2: (bf) r3 = r10 | |
2257 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
2258 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
2259 | * doesn't contain jmps. It's backtracked | |
2260 | * as a single block. | |
2261 | * During backtracking insn 3 is not recognized as | |
2262 | * stack access, so at the end of backtracking | |
2263 | * stack slot fp-8 is still marked in stack_mask. | |
2264 | * However the parent state may not have accessed | |
2265 | * fp-8 and it's "unallocated" stack space. | |
2266 | * In such case fallback to conservative. | |
b5dc0163 | 2267 | */ |
2339cd6c AS |
2268 | mark_all_scalars_precise(env, st); |
2269 | return 0; | |
b5dc0163 AS |
2270 | } |
2271 | ||
a3ce685d AS |
2272 | if (func->stack[i].slot_type[0] != STACK_SPILL) { |
2273 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2274 | continue; |
a3ce685d | 2275 | } |
b5dc0163 | 2276 | reg = &func->stack[i].spilled_ptr; |
a3ce685d AS |
2277 | if (reg->type != SCALAR_VALUE) { |
2278 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2279 | continue; |
a3ce685d | 2280 | } |
b5dc0163 AS |
2281 | if (!reg->precise) |
2282 | new_marks = true; | |
2283 | reg->precise = true; | |
2284 | } | |
2285 | if (env->log.level & BPF_LOG_LEVEL) { | |
2286 | print_verifier_state(env, func); | |
2287 | verbose(env, "parent %s regs=%x stack=%llx marks\n", | |
2288 | new_marks ? "didn't have" : "already had", | |
2289 | reg_mask, stack_mask); | |
2290 | } | |
2291 | ||
a3ce685d AS |
2292 | if (!reg_mask && !stack_mask) |
2293 | break; | |
b5dc0163 AS |
2294 | if (!new_marks) |
2295 | break; | |
2296 | ||
2297 | last_idx = st->last_insn_idx; | |
2298 | first_idx = st->first_insn_idx; | |
2299 | } | |
2300 | return 0; | |
2301 | } | |
2302 | ||
a3ce685d AS |
2303 | static int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
2304 | { | |
2305 | return __mark_chain_precision(env, regno, -1); | |
2306 | } | |
2307 | ||
2308 | static int mark_chain_precision_stack(struct bpf_verifier_env *env, int spi) | |
2309 | { | |
2310 | return __mark_chain_precision(env, -1, spi); | |
2311 | } | |
b5dc0163 | 2312 | |
1be7f75d AS |
2313 | static bool is_spillable_regtype(enum bpf_reg_type type) |
2314 | { | |
2315 | switch (type) { | |
2316 | case PTR_TO_MAP_VALUE: | |
2317 | case PTR_TO_MAP_VALUE_OR_NULL: | |
2318 | case PTR_TO_STACK: | |
2319 | case PTR_TO_CTX: | |
969bf05e | 2320 | case PTR_TO_PACKET: |
de8f3a83 | 2321 | case PTR_TO_PACKET_META: |
969bf05e | 2322 | case PTR_TO_PACKET_END: |
d58e468b | 2323 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 2324 | case CONST_PTR_TO_MAP: |
c64b7983 JS |
2325 | case PTR_TO_SOCKET: |
2326 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
2327 | case PTR_TO_SOCK_COMMON: |
2328 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
2329 | case PTR_TO_TCP_SOCK: |
2330 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 2331 | case PTR_TO_XDP_SOCK: |
65726b5b | 2332 | case PTR_TO_BTF_ID: |
b121b341 | 2333 | case PTR_TO_BTF_ID_OR_NULL: |
afbf21dc YS |
2334 | case PTR_TO_RDONLY_BUF: |
2335 | case PTR_TO_RDONLY_BUF_OR_NULL: | |
2336 | case PTR_TO_RDWR_BUF: | |
2337 | case PTR_TO_RDWR_BUF_OR_NULL: | |
eaa6bcb7 | 2338 | case PTR_TO_PERCPU_BTF_ID: |
744ea4e3 GR |
2339 | case PTR_TO_MEM: |
2340 | case PTR_TO_MEM_OR_NULL: | |
69c087ba YS |
2341 | case PTR_TO_FUNC: |
2342 | case PTR_TO_MAP_KEY: | |
1be7f75d AS |
2343 | return true; |
2344 | default: | |
2345 | return false; | |
2346 | } | |
2347 | } | |
2348 | ||
cc2b14d5 AS |
2349 | /* Does this register contain a constant zero? */ |
2350 | static bool register_is_null(struct bpf_reg_state *reg) | |
2351 | { | |
2352 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
2353 | } | |
2354 | ||
f7cf25b2 AS |
2355 | static bool register_is_const(struct bpf_reg_state *reg) |
2356 | { | |
2357 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
2358 | } | |
2359 | ||
5689d49b YS |
2360 | static bool __is_scalar_unbounded(struct bpf_reg_state *reg) |
2361 | { | |
2362 | return tnum_is_unknown(reg->var_off) && | |
2363 | reg->smin_value == S64_MIN && reg->smax_value == S64_MAX && | |
2364 | reg->umin_value == 0 && reg->umax_value == U64_MAX && | |
2365 | reg->s32_min_value == S32_MIN && reg->s32_max_value == S32_MAX && | |
2366 | reg->u32_min_value == 0 && reg->u32_max_value == U32_MAX; | |
2367 | } | |
2368 | ||
2369 | static bool register_is_bounded(struct bpf_reg_state *reg) | |
2370 | { | |
2371 | return reg->type == SCALAR_VALUE && !__is_scalar_unbounded(reg); | |
2372 | } | |
2373 | ||
6e7e63cb JH |
2374 | static bool __is_pointer_value(bool allow_ptr_leaks, |
2375 | const struct bpf_reg_state *reg) | |
2376 | { | |
2377 | if (allow_ptr_leaks) | |
2378 | return false; | |
2379 | ||
2380 | return reg->type != SCALAR_VALUE; | |
2381 | } | |
2382 | ||
f7cf25b2 AS |
2383 | static void save_register_state(struct bpf_func_state *state, |
2384 | int spi, struct bpf_reg_state *reg) | |
2385 | { | |
2386 | int i; | |
2387 | ||
2388 | state->stack[spi].spilled_ptr = *reg; | |
2389 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2390 | ||
2391 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2392 | state->stack[spi].slot_type[i] = STACK_SPILL; | |
2393 | } | |
2394 | ||
01f810ac | 2395 | /* check_stack_{read,write}_fixed_off functions track spill/fill of registers, |
17a52670 AS |
2396 | * stack boundary and alignment are checked in check_mem_access() |
2397 | */ | |
01f810ac AM |
2398 | static int check_stack_write_fixed_off(struct bpf_verifier_env *env, |
2399 | /* stack frame we're writing to */ | |
2400 | struct bpf_func_state *state, | |
2401 | int off, int size, int value_regno, | |
2402 | int insn_idx) | |
17a52670 | 2403 | { |
f4d7e40a | 2404 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 2405 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
b5dc0163 | 2406 | u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg; |
f7cf25b2 | 2407 | struct bpf_reg_state *reg = NULL; |
638f5b90 | 2408 | |
f4d7e40a | 2409 | err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE), |
fd978bf7 | 2410 | state->acquired_refs, true); |
638f5b90 AS |
2411 | if (err) |
2412 | return err; | |
9c399760 AS |
2413 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
2414 | * so it's aligned access and [off, off + size) are within stack limits | |
2415 | */ | |
638f5b90 AS |
2416 | if (!env->allow_ptr_leaks && |
2417 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
2418 | size != BPF_REG_SIZE) { | |
2419 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
2420 | return -EACCES; | |
2421 | } | |
17a52670 | 2422 | |
f4d7e40a | 2423 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
2424 | if (value_regno >= 0) |
2425 | reg = &cur->regs[value_regno]; | |
17a52670 | 2426 | |
5689d49b | 2427 | if (reg && size == BPF_REG_SIZE && register_is_bounded(reg) && |
2c78ee89 | 2428 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
2429 | if (dst_reg != BPF_REG_FP) { |
2430 | /* The backtracking logic can only recognize explicit | |
2431 | * stack slot address like [fp - 8]. Other spill of | |
2432 | * scalar via different register has to be conervative. | |
2433 | * Backtrack from here and mark all registers as precise | |
2434 | * that contributed into 'reg' being a constant. | |
2435 | */ | |
2436 | err = mark_chain_precision(env, value_regno); | |
2437 | if (err) | |
2438 | return err; | |
2439 | } | |
f7cf25b2 AS |
2440 | save_register_state(state, spi, reg); |
2441 | } else if (reg && is_spillable_regtype(reg->type)) { | |
17a52670 | 2442 | /* register containing pointer is being spilled into stack */ |
9c399760 | 2443 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 2444 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 2445 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
2446 | return -EACCES; |
2447 | } | |
2448 | ||
f7cf25b2 | 2449 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
2450 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
2451 | return -EINVAL; | |
2452 | } | |
2453 | ||
2c78ee89 | 2454 | if (!env->bypass_spec_v4) { |
f7cf25b2 | 2455 | bool sanitize = false; |
17a52670 | 2456 | |
f7cf25b2 AS |
2457 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
2458 | register_is_const(&state->stack[spi].spilled_ptr)) | |
2459 | sanitize = true; | |
2460 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2461 | if (state->stack[spi].slot_type[i] == STACK_MISC) { | |
2462 | sanitize = true; | |
2463 | break; | |
2464 | } | |
2465 | if (sanitize) { | |
af86ca4e AS |
2466 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; |
2467 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
2468 | ||
2469 | /* detected reuse of integer stack slot with a pointer | |
2470 | * which means either llvm is reusing stack slot or | |
2471 | * an attacker is trying to exploit CVE-2018-3639 | |
2472 | * (speculative store bypass) | |
2473 | * Have to sanitize that slot with preemptive | |
2474 | * store of zero. | |
2475 | */ | |
2476 | if (*poff && *poff != soff) { | |
2477 | /* disallow programs where single insn stores | |
2478 | * into two different stack slots, since verifier | |
2479 | * cannot sanitize them | |
2480 | */ | |
2481 | verbose(env, | |
2482 | "insn %d cannot access two stack slots fp%d and fp%d", | |
2483 | insn_idx, *poff, soff); | |
2484 | return -EINVAL; | |
2485 | } | |
2486 | *poff = soff; | |
2487 | } | |
af86ca4e | 2488 | } |
f7cf25b2 | 2489 | save_register_state(state, spi, reg); |
9c399760 | 2490 | } else { |
cc2b14d5 AS |
2491 | u8 type = STACK_MISC; |
2492 | ||
679c782d EC |
2493 | /* regular write of data into stack destroys any spilled ptr */ |
2494 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d JW |
2495 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
2496 | if (state->stack[spi].slot_type[0] == STACK_SPILL) | |
2497 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2498 | state->stack[spi].slot_type[i] = STACK_MISC; | |
9c399760 | 2499 | |
cc2b14d5 AS |
2500 | /* only mark the slot as written if all 8 bytes were written |
2501 | * otherwise read propagation may incorrectly stop too soon | |
2502 | * when stack slots are partially written. | |
2503 | * This heuristic means that read propagation will be | |
2504 | * conservative, since it will add reg_live_read marks | |
2505 | * to stack slots all the way to first state when programs | |
2506 | * writes+reads less than 8 bytes | |
2507 | */ | |
2508 | if (size == BPF_REG_SIZE) | |
2509 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2510 | ||
2511 | /* when we zero initialize stack slots mark them as such */ | |
b5dc0163 AS |
2512 | if (reg && register_is_null(reg)) { |
2513 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
2514 | err = mark_chain_precision(env, value_regno); | |
2515 | if (err) | |
2516 | return err; | |
cc2b14d5 | 2517 | type = STACK_ZERO; |
b5dc0163 | 2518 | } |
cc2b14d5 | 2519 | |
0bae2d4d | 2520 | /* Mark slots affected by this stack write. */ |
9c399760 | 2521 | for (i = 0; i < size; i++) |
638f5b90 | 2522 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 2523 | type; |
17a52670 AS |
2524 | } |
2525 | return 0; | |
2526 | } | |
2527 | ||
01f810ac AM |
2528 | /* Write the stack: 'stack[ptr_regno + off] = value_regno'. 'ptr_regno' is |
2529 | * known to contain a variable offset. | |
2530 | * This function checks whether the write is permitted and conservatively | |
2531 | * tracks the effects of the write, considering that each stack slot in the | |
2532 | * dynamic range is potentially written to. | |
2533 | * | |
2534 | * 'off' includes 'regno->off'. | |
2535 | * 'value_regno' can be -1, meaning that an unknown value is being written to | |
2536 | * the stack. | |
2537 | * | |
2538 | * Spilled pointers in range are not marked as written because we don't know | |
2539 | * what's going to be actually written. This means that read propagation for | |
2540 | * future reads cannot be terminated by this write. | |
2541 | * | |
2542 | * For privileged programs, uninitialized stack slots are considered | |
2543 | * initialized by this write (even though we don't know exactly what offsets | |
2544 | * are going to be written to). The idea is that we don't want the verifier to | |
2545 | * reject future reads that access slots written to through variable offsets. | |
2546 | */ | |
2547 | static int check_stack_write_var_off(struct bpf_verifier_env *env, | |
2548 | /* func where register points to */ | |
2549 | struct bpf_func_state *state, | |
2550 | int ptr_regno, int off, int size, | |
2551 | int value_regno, int insn_idx) | |
2552 | { | |
2553 | struct bpf_func_state *cur; /* state of the current function */ | |
2554 | int min_off, max_off; | |
2555 | int i, err; | |
2556 | struct bpf_reg_state *ptr_reg = NULL, *value_reg = NULL; | |
2557 | bool writing_zero = false; | |
2558 | /* set if the fact that we're writing a zero is used to let any | |
2559 | * stack slots remain STACK_ZERO | |
2560 | */ | |
2561 | bool zero_used = false; | |
2562 | ||
2563 | cur = env->cur_state->frame[env->cur_state->curframe]; | |
2564 | ptr_reg = &cur->regs[ptr_regno]; | |
2565 | min_off = ptr_reg->smin_value + off; | |
2566 | max_off = ptr_reg->smax_value + off + size; | |
2567 | if (value_regno >= 0) | |
2568 | value_reg = &cur->regs[value_regno]; | |
2569 | if (value_reg && register_is_null(value_reg)) | |
2570 | writing_zero = true; | |
2571 | ||
2572 | err = realloc_func_state(state, round_up(-min_off, BPF_REG_SIZE), | |
2573 | state->acquired_refs, true); | |
2574 | if (err) | |
2575 | return err; | |
2576 | ||
2577 | ||
2578 | /* Variable offset writes destroy any spilled pointers in range. */ | |
2579 | for (i = min_off; i < max_off; i++) { | |
2580 | u8 new_type, *stype; | |
2581 | int slot, spi; | |
2582 | ||
2583 | slot = -i - 1; | |
2584 | spi = slot / BPF_REG_SIZE; | |
2585 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
2586 | ||
2587 | if (!env->allow_ptr_leaks | |
2588 | && *stype != NOT_INIT | |
2589 | && *stype != SCALAR_VALUE) { | |
2590 | /* Reject the write if there's are spilled pointers in | |
2591 | * range. If we didn't reject here, the ptr status | |
2592 | * would be erased below (even though not all slots are | |
2593 | * actually overwritten), possibly opening the door to | |
2594 | * leaks. | |
2595 | */ | |
2596 | verbose(env, "spilled ptr in range of var-offset stack write; insn %d, ptr off: %d", | |
2597 | insn_idx, i); | |
2598 | return -EINVAL; | |
2599 | } | |
2600 | ||
2601 | /* Erase all spilled pointers. */ | |
2602 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
2603 | ||
2604 | /* Update the slot type. */ | |
2605 | new_type = STACK_MISC; | |
2606 | if (writing_zero && *stype == STACK_ZERO) { | |
2607 | new_type = STACK_ZERO; | |
2608 | zero_used = true; | |
2609 | } | |
2610 | /* If the slot is STACK_INVALID, we check whether it's OK to | |
2611 | * pretend that it will be initialized by this write. The slot | |
2612 | * might not actually be written to, and so if we mark it as | |
2613 | * initialized future reads might leak uninitialized memory. | |
2614 | * For privileged programs, we will accept such reads to slots | |
2615 | * that may or may not be written because, if we're reject | |
2616 | * them, the error would be too confusing. | |
2617 | */ | |
2618 | if (*stype == STACK_INVALID && !env->allow_uninit_stack) { | |
2619 | verbose(env, "uninit stack in range of var-offset write prohibited for !root; insn %d, off: %d", | |
2620 | insn_idx, i); | |
2621 | return -EINVAL; | |
2622 | } | |
2623 | *stype = new_type; | |
2624 | } | |
2625 | if (zero_used) { | |
2626 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
2627 | err = mark_chain_precision(env, value_regno); | |
2628 | if (err) | |
2629 | return err; | |
2630 | } | |
2631 | return 0; | |
2632 | } | |
2633 | ||
2634 | /* When register 'dst_regno' is assigned some values from stack[min_off, | |
2635 | * max_off), we set the register's type according to the types of the | |
2636 | * respective stack slots. If all the stack values are known to be zeros, then | |
2637 | * so is the destination reg. Otherwise, the register is considered to be | |
2638 | * SCALAR. This function does not deal with register filling; the caller must | |
2639 | * ensure that all spilled registers in the stack range have been marked as | |
2640 | * read. | |
2641 | */ | |
2642 | static void mark_reg_stack_read(struct bpf_verifier_env *env, | |
2643 | /* func where src register points to */ | |
2644 | struct bpf_func_state *ptr_state, | |
2645 | int min_off, int max_off, int dst_regno) | |
2646 | { | |
2647 | struct bpf_verifier_state *vstate = env->cur_state; | |
2648 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
2649 | int i, slot, spi; | |
2650 | u8 *stype; | |
2651 | int zeros = 0; | |
2652 | ||
2653 | for (i = min_off; i < max_off; i++) { | |
2654 | slot = -i - 1; | |
2655 | spi = slot / BPF_REG_SIZE; | |
2656 | stype = ptr_state->stack[spi].slot_type; | |
2657 | if (stype[slot % BPF_REG_SIZE] != STACK_ZERO) | |
2658 | break; | |
2659 | zeros++; | |
2660 | } | |
2661 | if (zeros == max_off - min_off) { | |
2662 | /* any access_size read into register is zero extended, | |
2663 | * so the whole register == const_zero | |
2664 | */ | |
2665 | __mark_reg_const_zero(&state->regs[dst_regno]); | |
2666 | /* backtracking doesn't support STACK_ZERO yet, | |
2667 | * so mark it precise here, so that later | |
2668 | * backtracking can stop here. | |
2669 | * Backtracking may not need this if this register | |
2670 | * doesn't participate in pointer adjustment. | |
2671 | * Forward propagation of precise flag is not | |
2672 | * necessary either. This mark is only to stop | |
2673 | * backtracking. Any register that contributed | |
2674 | * to const 0 was marked precise before spill. | |
2675 | */ | |
2676 | state->regs[dst_regno].precise = true; | |
2677 | } else { | |
2678 | /* have read misc data from the stack */ | |
2679 | mark_reg_unknown(env, state->regs, dst_regno); | |
2680 | } | |
2681 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
2682 | } | |
2683 | ||
2684 | /* Read the stack at 'off' and put the results into the register indicated by | |
2685 | * 'dst_regno'. It handles reg filling if the addressed stack slot is a | |
2686 | * spilled reg. | |
2687 | * | |
2688 | * 'dst_regno' can be -1, meaning that the read value is not going to a | |
2689 | * register. | |
2690 | * | |
2691 | * The access is assumed to be within the current stack bounds. | |
2692 | */ | |
2693 | static int check_stack_read_fixed_off(struct bpf_verifier_env *env, | |
2694 | /* func where src register points to */ | |
2695 | struct bpf_func_state *reg_state, | |
2696 | int off, int size, int dst_regno) | |
17a52670 | 2697 | { |
f4d7e40a AS |
2698 | struct bpf_verifier_state *vstate = env->cur_state; |
2699 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 2700 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 2701 | struct bpf_reg_state *reg; |
638f5b90 | 2702 | u8 *stype; |
17a52670 | 2703 | |
f4d7e40a | 2704 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 2705 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 2706 | |
638f5b90 | 2707 | if (stype[0] == STACK_SPILL) { |
9c399760 | 2708 | if (size != BPF_REG_SIZE) { |
f7cf25b2 AS |
2709 | if (reg->type != SCALAR_VALUE) { |
2710 | verbose_linfo(env, env->insn_idx, "; "); | |
2711 | verbose(env, "invalid size of register fill\n"); | |
2712 | return -EACCES; | |
2713 | } | |
01f810ac AM |
2714 | if (dst_regno >= 0) { |
2715 | mark_reg_unknown(env, state->regs, dst_regno); | |
2716 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; | |
f7cf25b2 AS |
2717 | } |
2718 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
2719 | return 0; | |
17a52670 | 2720 | } |
9c399760 | 2721 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 2722 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 2723 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
2724 | return -EACCES; |
2725 | } | |
2726 | } | |
2727 | ||
01f810ac | 2728 | if (dst_regno >= 0) { |
17a52670 | 2729 | /* restore register state from stack */ |
01f810ac | 2730 | state->regs[dst_regno] = *reg; |
2f18f62e AS |
2731 | /* mark reg as written since spilled pointer state likely |
2732 | * has its liveness marks cleared by is_state_visited() | |
2733 | * which resets stack/reg liveness for state transitions | |
2734 | */ | |
01f810ac | 2735 | state->regs[dst_regno].live |= REG_LIVE_WRITTEN; |
6e7e63cb | 2736 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
01f810ac | 2737 | /* If dst_regno==-1, the caller is asking us whether |
6e7e63cb JH |
2738 | * it is acceptable to use this value as a SCALAR_VALUE |
2739 | * (e.g. for XADD). | |
2740 | * We must not allow unprivileged callers to do that | |
2741 | * with spilled pointers. | |
2742 | */ | |
2743 | verbose(env, "leaking pointer from stack off %d\n", | |
2744 | off); | |
2745 | return -EACCES; | |
dc503a8a | 2746 | } |
f7cf25b2 | 2747 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 | 2748 | } else { |
01f810ac | 2749 | u8 type; |
cc2b14d5 | 2750 | |
17a52670 | 2751 | for (i = 0; i < size; i++) { |
01f810ac AM |
2752 | type = stype[(slot - i) % BPF_REG_SIZE]; |
2753 | if (type == STACK_MISC) | |
cc2b14d5 | 2754 | continue; |
01f810ac | 2755 | if (type == STACK_ZERO) |
cc2b14d5 | 2756 | continue; |
cc2b14d5 AS |
2757 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
2758 | off, i, size); | |
2759 | return -EACCES; | |
2760 | } | |
f7cf25b2 | 2761 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
01f810ac AM |
2762 | if (dst_regno >= 0) |
2763 | mark_reg_stack_read(env, reg_state, off, off + size, dst_regno); | |
17a52670 | 2764 | } |
f7cf25b2 | 2765 | return 0; |
17a52670 AS |
2766 | } |
2767 | ||
01f810ac AM |
2768 | enum stack_access_src { |
2769 | ACCESS_DIRECT = 1, /* the access is performed by an instruction */ | |
2770 | ACCESS_HELPER = 2, /* the access is performed by a helper */ | |
2771 | }; | |
2772 | ||
2773 | static int check_stack_range_initialized(struct bpf_verifier_env *env, | |
2774 | int regno, int off, int access_size, | |
2775 | bool zero_size_allowed, | |
2776 | enum stack_access_src type, | |
2777 | struct bpf_call_arg_meta *meta); | |
2778 | ||
2779 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) | |
2780 | { | |
2781 | return cur_regs(env) + regno; | |
2782 | } | |
2783 | ||
2784 | /* Read the stack at 'ptr_regno + off' and put the result into the register | |
2785 | * 'dst_regno'. | |
2786 | * 'off' includes the pointer register's fixed offset(i.e. 'ptr_regno.off'), | |
2787 | * but not its variable offset. | |
2788 | * 'size' is assumed to be <= reg size and the access is assumed to be aligned. | |
2789 | * | |
2790 | * As opposed to check_stack_read_fixed_off, this function doesn't deal with | |
2791 | * filling registers (i.e. reads of spilled register cannot be detected when | |
2792 | * the offset is not fixed). We conservatively mark 'dst_regno' as containing | |
2793 | * SCALAR_VALUE. That's why we assert that the 'ptr_regno' has a variable | |
2794 | * offset; for a fixed offset check_stack_read_fixed_off should be used | |
2795 | * instead. | |
2796 | */ | |
2797 | static int check_stack_read_var_off(struct bpf_verifier_env *env, | |
2798 | int ptr_regno, int off, int size, int dst_regno) | |
e4298d25 | 2799 | { |
01f810ac AM |
2800 | /* The state of the source register. */ |
2801 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
2802 | struct bpf_func_state *ptr_state = func(env, reg); | |
2803 | int err; | |
2804 | int min_off, max_off; | |
2805 | ||
2806 | /* Note that we pass a NULL meta, so raw access will not be permitted. | |
e4298d25 | 2807 | */ |
01f810ac AM |
2808 | err = check_stack_range_initialized(env, ptr_regno, off, size, |
2809 | false, ACCESS_DIRECT, NULL); | |
2810 | if (err) | |
2811 | return err; | |
2812 | ||
2813 | min_off = reg->smin_value + off; | |
2814 | max_off = reg->smax_value + off; | |
2815 | mark_reg_stack_read(env, ptr_state, min_off, max_off + size, dst_regno); | |
2816 | return 0; | |
2817 | } | |
2818 | ||
2819 | /* check_stack_read dispatches to check_stack_read_fixed_off or | |
2820 | * check_stack_read_var_off. | |
2821 | * | |
2822 | * The caller must ensure that the offset falls within the allocated stack | |
2823 | * bounds. | |
2824 | * | |
2825 | * 'dst_regno' is a register which will receive the value from the stack. It | |
2826 | * can be -1, meaning that the read value is not going to a register. | |
2827 | */ | |
2828 | static int check_stack_read(struct bpf_verifier_env *env, | |
2829 | int ptr_regno, int off, int size, | |
2830 | int dst_regno) | |
2831 | { | |
2832 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
2833 | struct bpf_func_state *state = func(env, reg); | |
2834 | int err; | |
2835 | /* Some accesses are only permitted with a static offset. */ | |
2836 | bool var_off = !tnum_is_const(reg->var_off); | |
2837 | ||
2838 | /* The offset is required to be static when reads don't go to a | |
2839 | * register, in order to not leak pointers (see | |
2840 | * check_stack_read_fixed_off). | |
2841 | */ | |
2842 | if (dst_regno < 0 && var_off) { | |
e4298d25 DB |
2843 | char tn_buf[48]; |
2844 | ||
2845 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac | 2846 | verbose(env, "variable offset stack pointer cannot be passed into helper function; var_off=%s off=%d size=%d\n", |
e4298d25 DB |
2847 | tn_buf, off, size); |
2848 | return -EACCES; | |
2849 | } | |
01f810ac AM |
2850 | /* Variable offset is prohibited for unprivileged mode for simplicity |
2851 | * since it requires corresponding support in Spectre masking for stack | |
2852 | * ALU. See also retrieve_ptr_limit(). | |
2853 | */ | |
2854 | if (!env->bypass_spec_v1 && var_off) { | |
2855 | char tn_buf[48]; | |
e4298d25 | 2856 | |
01f810ac AM |
2857 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
2858 | verbose(env, "R%d variable offset stack access prohibited for !root, var_off=%s\n", | |
2859 | ptr_regno, tn_buf); | |
e4298d25 DB |
2860 | return -EACCES; |
2861 | } | |
2862 | ||
01f810ac AM |
2863 | if (!var_off) { |
2864 | off += reg->var_off.value; | |
2865 | err = check_stack_read_fixed_off(env, state, off, size, | |
2866 | dst_regno); | |
2867 | } else { | |
2868 | /* Variable offset stack reads need more conservative handling | |
2869 | * than fixed offset ones. Note that dst_regno >= 0 on this | |
2870 | * branch. | |
2871 | */ | |
2872 | err = check_stack_read_var_off(env, ptr_regno, off, size, | |
2873 | dst_regno); | |
2874 | } | |
2875 | return err; | |
2876 | } | |
2877 | ||
2878 | ||
2879 | /* check_stack_write dispatches to check_stack_write_fixed_off or | |
2880 | * check_stack_write_var_off. | |
2881 | * | |
2882 | * 'ptr_regno' is the register used as a pointer into the stack. | |
2883 | * 'off' includes 'ptr_regno->off', but not its variable offset (if any). | |
2884 | * 'value_regno' is the register whose value we're writing to the stack. It can | |
2885 | * be -1, meaning that we're not writing from a register. | |
2886 | * | |
2887 | * The caller must ensure that the offset falls within the maximum stack size. | |
2888 | */ | |
2889 | static int check_stack_write(struct bpf_verifier_env *env, | |
2890 | int ptr_regno, int off, int size, | |
2891 | int value_regno, int insn_idx) | |
2892 | { | |
2893 | struct bpf_reg_state *reg = reg_state(env, ptr_regno); | |
2894 | struct bpf_func_state *state = func(env, reg); | |
2895 | int err; | |
2896 | ||
2897 | if (tnum_is_const(reg->var_off)) { | |
2898 | off += reg->var_off.value; | |
2899 | err = check_stack_write_fixed_off(env, state, off, size, | |
2900 | value_regno, insn_idx); | |
2901 | } else { | |
2902 | /* Variable offset stack reads need more conservative handling | |
2903 | * than fixed offset ones. | |
2904 | */ | |
2905 | err = check_stack_write_var_off(env, state, | |
2906 | ptr_regno, off, size, | |
2907 | value_regno, insn_idx); | |
2908 | } | |
2909 | return err; | |
e4298d25 DB |
2910 | } |
2911 | ||
591fe988 DB |
2912 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
2913 | int off, int size, enum bpf_access_type type) | |
2914 | { | |
2915 | struct bpf_reg_state *regs = cur_regs(env); | |
2916 | struct bpf_map *map = regs[regno].map_ptr; | |
2917 | u32 cap = bpf_map_flags_to_cap(map); | |
2918 | ||
2919 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
2920 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
2921 | map->value_size, off, size); | |
2922 | return -EACCES; | |
2923 | } | |
2924 | ||
2925 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
2926 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
2927 | map->value_size, off, size); | |
2928 | return -EACCES; | |
2929 | } | |
2930 | ||
2931 | return 0; | |
2932 | } | |
2933 | ||
457f4436 AN |
2934 | /* check read/write into memory region (e.g., map value, ringbuf sample, etc) */ |
2935 | static int __check_mem_access(struct bpf_verifier_env *env, int regno, | |
2936 | int off, int size, u32 mem_size, | |
2937 | bool zero_size_allowed) | |
17a52670 | 2938 | { |
457f4436 AN |
2939 | bool size_ok = size > 0 || (size == 0 && zero_size_allowed); |
2940 | struct bpf_reg_state *reg; | |
2941 | ||
2942 | if (off >= 0 && size_ok && (u64)off + size <= mem_size) | |
2943 | return 0; | |
17a52670 | 2944 | |
457f4436 AN |
2945 | reg = &cur_regs(env)[regno]; |
2946 | switch (reg->type) { | |
69c087ba YS |
2947 | case PTR_TO_MAP_KEY: |
2948 | verbose(env, "invalid access to map key, key_size=%d off=%d size=%d\n", | |
2949 | mem_size, off, size); | |
2950 | break; | |
457f4436 | 2951 | case PTR_TO_MAP_VALUE: |
61bd5218 | 2952 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
457f4436 AN |
2953 | mem_size, off, size); |
2954 | break; | |
2955 | case PTR_TO_PACKET: | |
2956 | case PTR_TO_PACKET_META: | |
2957 | case PTR_TO_PACKET_END: | |
2958 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", | |
2959 | off, size, regno, reg->id, off, mem_size); | |
2960 | break; | |
2961 | case PTR_TO_MEM: | |
2962 | default: | |
2963 | verbose(env, "invalid access to memory, mem_size=%u off=%d size=%d\n", | |
2964 | mem_size, off, size); | |
17a52670 | 2965 | } |
457f4436 AN |
2966 | |
2967 | return -EACCES; | |
17a52670 AS |
2968 | } |
2969 | ||
457f4436 AN |
2970 | /* check read/write into a memory region with possible variable offset */ |
2971 | static int check_mem_region_access(struct bpf_verifier_env *env, u32 regno, | |
2972 | int off, int size, u32 mem_size, | |
2973 | bool zero_size_allowed) | |
dbcfe5f7 | 2974 | { |
f4d7e40a AS |
2975 | struct bpf_verifier_state *vstate = env->cur_state; |
2976 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
2977 | struct bpf_reg_state *reg = &state->regs[regno]; |
2978 | int err; | |
2979 | ||
457f4436 | 2980 | /* We may have adjusted the register pointing to memory region, so we |
f1174f77 EC |
2981 | * need to try adding each of min_value and max_value to off |
2982 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 2983 | */ |
06ee7115 | 2984 | if (env->log.level & BPF_LOG_LEVEL) |
61bd5218 | 2985 | print_verifier_state(env, state); |
b7137c4e | 2986 | |
dbcfe5f7 GB |
2987 | /* The minimum value is only important with signed |
2988 | * comparisons where we can't assume the floor of a | |
2989 | * value is 0. If we are using signed variables for our | |
2990 | * index'es we need to make sure that whatever we use | |
2991 | * will have a set floor within our range. | |
2992 | */ | |
b7137c4e DB |
2993 | if (reg->smin_value < 0 && |
2994 | (reg->smin_value == S64_MIN || | |
2995 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
2996 | reg->smin_value + off < 0)) { | |
61bd5218 | 2997 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
2998 | regno); |
2999 | return -EACCES; | |
3000 | } | |
457f4436 AN |
3001 | err = __check_mem_access(env, regno, reg->smin_value + off, size, |
3002 | mem_size, zero_size_allowed); | |
dbcfe5f7 | 3003 | if (err) { |
457f4436 | 3004 | verbose(env, "R%d min value is outside of the allowed memory range\n", |
61bd5218 | 3005 | regno); |
dbcfe5f7 GB |
3006 | return err; |
3007 | } | |
3008 | ||
b03c9f9f EC |
3009 | /* If we haven't set a max value then we need to bail since we can't be |
3010 | * sure we won't do bad things. | |
3011 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 3012 | */ |
b03c9f9f | 3013 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
457f4436 | 3014 | verbose(env, "R%d unbounded memory access, make sure to bounds check any such access\n", |
dbcfe5f7 GB |
3015 | regno); |
3016 | return -EACCES; | |
3017 | } | |
457f4436 AN |
3018 | err = __check_mem_access(env, regno, reg->umax_value + off, size, |
3019 | mem_size, zero_size_allowed); | |
3020 | if (err) { | |
3021 | verbose(env, "R%d max value is outside of the allowed memory range\n", | |
61bd5218 | 3022 | regno); |
457f4436 AN |
3023 | return err; |
3024 | } | |
3025 | ||
3026 | return 0; | |
3027 | } | |
d83525ca | 3028 | |
457f4436 AN |
3029 | /* check read/write into a map element with possible variable offset */ |
3030 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
3031 | int off, int size, bool zero_size_allowed) | |
3032 | { | |
3033 | struct bpf_verifier_state *vstate = env->cur_state; | |
3034 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3035 | struct bpf_reg_state *reg = &state->regs[regno]; | |
3036 | struct bpf_map *map = reg->map_ptr; | |
3037 | int err; | |
3038 | ||
3039 | err = check_mem_region_access(env, regno, off, size, map->value_size, | |
3040 | zero_size_allowed); | |
3041 | if (err) | |
3042 | return err; | |
3043 | ||
3044 | if (map_value_has_spin_lock(map)) { | |
3045 | u32 lock = map->spin_lock_off; | |
d83525ca AS |
3046 | |
3047 | /* if any part of struct bpf_spin_lock can be touched by | |
3048 | * load/store reject this program. | |
3049 | * To check that [x1, x2) overlaps with [y1, y2) | |
3050 | * it is sufficient to check x1 < y2 && y1 < x2. | |
3051 | */ | |
3052 | if (reg->smin_value + off < lock + sizeof(struct bpf_spin_lock) && | |
3053 | lock < reg->umax_value + off + size) { | |
3054 | verbose(env, "bpf_spin_lock cannot be accessed directly by load/store\n"); | |
3055 | return -EACCES; | |
3056 | } | |
3057 | } | |
f1174f77 | 3058 | return err; |
dbcfe5f7 GB |
3059 | } |
3060 | ||
969bf05e AS |
3061 | #define MAX_PACKET_OFF 0xffff |
3062 | ||
7e40781c UP |
3063 | static enum bpf_prog_type resolve_prog_type(struct bpf_prog *prog) |
3064 | { | |
3aac1ead | 3065 | return prog->aux->dst_prog ? prog->aux->dst_prog->type : prog->type; |
7e40781c UP |
3066 | } |
3067 | ||
58e2af8b | 3068 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
3069 | const struct bpf_call_arg_meta *meta, |
3070 | enum bpf_access_type t) | |
4acf6c0b | 3071 | { |
7e40781c UP |
3072 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
3073 | ||
3074 | switch (prog_type) { | |
5d66fa7d | 3075 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
3076 | case BPF_PROG_TYPE_LWT_IN: |
3077 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 3078 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 3079 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 3080 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 3081 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
3082 | if (t == BPF_WRITE) |
3083 | return false; | |
8731745e | 3084 | fallthrough; |
5d66fa7d DB |
3085 | |
3086 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
3087 | case BPF_PROG_TYPE_SCHED_CLS: |
3088 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 3089 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 3090 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 3091 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 3092 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
3093 | if (meta) |
3094 | return meta->pkt_access; | |
3095 | ||
3096 | env->seen_direct_write = true; | |
4acf6c0b | 3097 | return true; |
0d01da6a SF |
3098 | |
3099 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
3100 | if (t == BPF_WRITE) | |
3101 | env->seen_direct_write = true; | |
3102 | ||
3103 | return true; | |
3104 | ||
4acf6c0b BB |
3105 | default: |
3106 | return false; | |
3107 | } | |
3108 | } | |
3109 | ||
f1174f77 | 3110 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 3111 | int size, bool zero_size_allowed) |
f1174f77 | 3112 | { |
638f5b90 | 3113 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
3114 | struct bpf_reg_state *reg = ®s[regno]; |
3115 | int err; | |
3116 | ||
3117 | /* We may have added a variable offset to the packet pointer; but any | |
3118 | * reg->range we have comes after that. We are only checking the fixed | |
3119 | * offset. | |
3120 | */ | |
3121 | ||
3122 | /* We don't allow negative numbers, because we aren't tracking enough | |
3123 | * detail to prove they're safe. | |
3124 | */ | |
b03c9f9f | 3125 | if (reg->smin_value < 0) { |
61bd5218 | 3126 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
3127 | regno); |
3128 | return -EACCES; | |
3129 | } | |
6d94e741 AS |
3130 | |
3131 | err = reg->range < 0 ? -EINVAL : | |
3132 | __check_mem_access(env, regno, off, size, reg->range, | |
457f4436 | 3133 | zero_size_allowed); |
f1174f77 | 3134 | if (err) { |
61bd5218 | 3135 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
3136 | return err; |
3137 | } | |
e647815a | 3138 | |
457f4436 | 3139 | /* __check_mem_access has made sure "off + size - 1" is within u16. |
e647815a JW |
3140 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, |
3141 | * otherwise find_good_pkt_pointers would have refused to set range info | |
457f4436 | 3142 | * that __check_mem_access would have rejected this pkt access. |
e647815a JW |
3143 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. |
3144 | */ | |
3145 | env->prog->aux->max_pkt_offset = | |
3146 | max_t(u32, env->prog->aux->max_pkt_offset, | |
3147 | off + reg->umax_value + size - 1); | |
3148 | ||
f1174f77 EC |
3149 | return err; |
3150 | } | |
3151 | ||
3152 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 3153 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 | 3154 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
22dc4a0f | 3155 | struct btf **btf, u32 *btf_id) |
17a52670 | 3156 | { |
f96da094 DB |
3157 | struct bpf_insn_access_aux info = { |
3158 | .reg_type = *reg_type, | |
9e15db66 | 3159 | .log = &env->log, |
f96da094 | 3160 | }; |
31fd8581 | 3161 | |
4f9218aa | 3162 | if (env->ops->is_valid_access && |
5e43f899 | 3163 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
3164 | /* A non zero info.ctx_field_size indicates that this field is a |
3165 | * candidate for later verifier transformation to load the whole | |
3166 | * field and then apply a mask when accessed with a narrower | |
3167 | * access than actual ctx access size. A zero info.ctx_field_size | |
3168 | * will only allow for whole field access and rejects any other | |
3169 | * type of narrower access. | |
31fd8581 | 3170 | */ |
23994631 | 3171 | *reg_type = info.reg_type; |
31fd8581 | 3172 | |
22dc4a0f AN |
3173 | if (*reg_type == PTR_TO_BTF_ID || *reg_type == PTR_TO_BTF_ID_OR_NULL) { |
3174 | *btf = info.btf; | |
9e15db66 | 3175 | *btf_id = info.btf_id; |
22dc4a0f | 3176 | } else { |
9e15db66 | 3177 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
22dc4a0f | 3178 | } |
32bbe007 AS |
3179 | /* remember the offset of last byte accessed in ctx */ |
3180 | if (env->prog->aux->max_ctx_offset < off + size) | |
3181 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 3182 | return 0; |
32bbe007 | 3183 | } |
17a52670 | 3184 | |
61bd5218 | 3185 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
3186 | return -EACCES; |
3187 | } | |
3188 | ||
d58e468b PP |
3189 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
3190 | int size) | |
3191 | { | |
3192 | if (size < 0 || off < 0 || | |
3193 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
3194 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
3195 | off, size); | |
3196 | return -EACCES; | |
3197 | } | |
3198 | return 0; | |
3199 | } | |
3200 | ||
5f456649 MKL |
3201 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
3202 | u32 regno, int off, int size, | |
3203 | enum bpf_access_type t) | |
c64b7983 JS |
3204 | { |
3205 | struct bpf_reg_state *regs = cur_regs(env); | |
3206 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 3207 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 3208 | bool valid; |
c64b7983 JS |
3209 | |
3210 | if (reg->smin_value < 0) { | |
3211 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
3212 | regno); | |
3213 | return -EACCES; | |
3214 | } | |
3215 | ||
46f8bc92 MKL |
3216 | switch (reg->type) { |
3217 | case PTR_TO_SOCK_COMMON: | |
3218 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
3219 | break; | |
3220 | case PTR_TO_SOCKET: | |
3221 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
3222 | break; | |
655a51e5 MKL |
3223 | case PTR_TO_TCP_SOCK: |
3224 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
3225 | break; | |
fada7fdc JL |
3226 | case PTR_TO_XDP_SOCK: |
3227 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
3228 | break; | |
46f8bc92 MKL |
3229 | default: |
3230 | valid = false; | |
c64b7983 JS |
3231 | } |
3232 | ||
5f456649 | 3233 | |
46f8bc92 MKL |
3234 | if (valid) { |
3235 | env->insn_aux_data[insn_idx].ctx_field_size = | |
3236 | info.ctx_field_size; | |
3237 | return 0; | |
3238 | } | |
3239 | ||
3240 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
3241 | regno, reg_type_str[reg->type], off, size); | |
3242 | ||
3243 | return -EACCES; | |
c64b7983 JS |
3244 | } |
3245 | ||
4cabc5b1 DB |
3246 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
3247 | { | |
2a159c6f | 3248 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
3249 | } |
3250 | ||
f37a8cb8 DB |
3251 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
3252 | { | |
2a159c6f | 3253 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 3254 | |
46f8bc92 MKL |
3255 | return reg->type == PTR_TO_CTX; |
3256 | } | |
3257 | ||
3258 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
3259 | { | |
3260 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
3261 | ||
3262 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
3263 | } |
3264 | ||
ca369602 DB |
3265 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
3266 | { | |
2a159c6f | 3267 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
3268 | |
3269 | return type_is_pkt_pointer(reg->type); | |
3270 | } | |
3271 | ||
4b5defde DB |
3272 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
3273 | { | |
3274 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
3275 | ||
3276 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
3277 | return reg->type == PTR_TO_FLOW_KEYS; | |
3278 | } | |
3279 | ||
61bd5218 JK |
3280 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
3281 | const struct bpf_reg_state *reg, | |
d1174416 | 3282 | int off, int size, bool strict) |
969bf05e | 3283 | { |
f1174f77 | 3284 | struct tnum reg_off; |
e07b98d9 | 3285 | int ip_align; |
d1174416 DM |
3286 | |
3287 | /* Byte size accesses are always allowed. */ | |
3288 | if (!strict || size == 1) | |
3289 | return 0; | |
3290 | ||
e4eda884 DM |
3291 | /* For platforms that do not have a Kconfig enabling |
3292 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
3293 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
3294 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
3295 | * to this code only in strict mode where we want to emulate | |
3296 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
3297 | * unconditional IP align value of '2'. | |
e07b98d9 | 3298 | */ |
e4eda884 | 3299 | ip_align = 2; |
f1174f77 EC |
3300 | |
3301 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
3302 | if (!tnum_is_aligned(reg_off, size)) { | |
3303 | char tn_buf[48]; | |
3304 | ||
3305 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
3306 | verbose(env, |
3307 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 3308 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
3309 | return -EACCES; |
3310 | } | |
79adffcd | 3311 | |
969bf05e AS |
3312 | return 0; |
3313 | } | |
3314 | ||
61bd5218 JK |
3315 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
3316 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
3317 | const char *pointer_desc, |
3318 | int off, int size, bool strict) | |
79adffcd | 3319 | { |
f1174f77 EC |
3320 | struct tnum reg_off; |
3321 | ||
3322 | /* Byte size accesses are always allowed. */ | |
3323 | if (!strict || size == 1) | |
3324 | return 0; | |
3325 | ||
3326 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
3327 | if (!tnum_is_aligned(reg_off, size)) { | |
3328 | char tn_buf[48]; | |
3329 | ||
3330 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3331 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 3332 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
3333 | return -EACCES; |
3334 | } | |
3335 | ||
969bf05e AS |
3336 | return 0; |
3337 | } | |
3338 | ||
e07b98d9 | 3339 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
3340 | const struct bpf_reg_state *reg, int off, |
3341 | int size, bool strict_alignment_once) | |
79adffcd | 3342 | { |
ca369602 | 3343 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 3344 | const char *pointer_desc = ""; |
d1174416 | 3345 | |
79adffcd DB |
3346 | switch (reg->type) { |
3347 | case PTR_TO_PACKET: | |
de8f3a83 DB |
3348 | case PTR_TO_PACKET_META: |
3349 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
3350 | * right in front, treat it the very same way. | |
3351 | */ | |
61bd5218 | 3352 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
3353 | case PTR_TO_FLOW_KEYS: |
3354 | pointer_desc = "flow keys "; | |
3355 | break; | |
69c087ba YS |
3356 | case PTR_TO_MAP_KEY: |
3357 | pointer_desc = "key "; | |
3358 | break; | |
f1174f77 EC |
3359 | case PTR_TO_MAP_VALUE: |
3360 | pointer_desc = "value "; | |
3361 | break; | |
3362 | case PTR_TO_CTX: | |
3363 | pointer_desc = "context "; | |
3364 | break; | |
3365 | case PTR_TO_STACK: | |
3366 | pointer_desc = "stack "; | |
01f810ac AM |
3367 | /* The stack spill tracking logic in check_stack_write_fixed_off() |
3368 | * and check_stack_read_fixed_off() relies on stack accesses being | |
a5ec6ae1 JH |
3369 | * aligned. |
3370 | */ | |
3371 | strict = true; | |
f1174f77 | 3372 | break; |
c64b7983 JS |
3373 | case PTR_TO_SOCKET: |
3374 | pointer_desc = "sock "; | |
3375 | break; | |
46f8bc92 MKL |
3376 | case PTR_TO_SOCK_COMMON: |
3377 | pointer_desc = "sock_common "; | |
3378 | break; | |
655a51e5 MKL |
3379 | case PTR_TO_TCP_SOCK: |
3380 | pointer_desc = "tcp_sock "; | |
3381 | break; | |
fada7fdc JL |
3382 | case PTR_TO_XDP_SOCK: |
3383 | pointer_desc = "xdp_sock "; | |
3384 | break; | |
79adffcd | 3385 | default: |
f1174f77 | 3386 | break; |
79adffcd | 3387 | } |
61bd5218 JK |
3388 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
3389 | strict); | |
79adffcd DB |
3390 | } |
3391 | ||
f4d7e40a AS |
3392 | static int update_stack_depth(struct bpf_verifier_env *env, |
3393 | const struct bpf_func_state *func, | |
3394 | int off) | |
3395 | { | |
9c8105bd | 3396 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
3397 | |
3398 | if (stack >= -off) | |
3399 | return 0; | |
3400 | ||
3401 | /* update known max for given subprogram */ | |
9c8105bd | 3402 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
3403 | return 0; |
3404 | } | |
f4d7e40a | 3405 | |
70a87ffe AS |
3406 | /* starting from main bpf function walk all instructions of the function |
3407 | * and recursively walk all callees that given function can call. | |
3408 | * Ignore jump and exit insns. | |
3409 | * Since recursion is prevented by check_cfg() this algorithm | |
3410 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
3411 | */ | |
3412 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
3413 | { | |
9c8105bd JW |
3414 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
3415 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 3416 | struct bpf_insn *insn = env->prog->insnsi; |
ebf7d1f5 | 3417 | bool tail_call_reachable = false; |
70a87ffe AS |
3418 | int ret_insn[MAX_CALL_FRAMES]; |
3419 | int ret_prog[MAX_CALL_FRAMES]; | |
ebf7d1f5 | 3420 | int j; |
f4d7e40a | 3421 | |
70a87ffe | 3422 | process_func: |
7f6e4312 MF |
3423 | /* protect against potential stack overflow that might happen when |
3424 | * bpf2bpf calls get combined with tailcalls. Limit the caller's stack | |
3425 | * depth for such case down to 256 so that the worst case scenario | |
3426 | * would result in 8k stack size (32 which is tailcall limit * 256 = | |
3427 | * 8k). | |
3428 | * | |
3429 | * To get the idea what might happen, see an example: | |
3430 | * func1 -> sub rsp, 128 | |
3431 | * subfunc1 -> sub rsp, 256 | |
3432 | * tailcall1 -> add rsp, 256 | |
3433 | * func2 -> sub rsp, 192 (total stack size = 128 + 192 = 320) | |
3434 | * subfunc2 -> sub rsp, 64 | |
3435 | * subfunc22 -> sub rsp, 128 | |
3436 | * tailcall2 -> add rsp, 128 | |
3437 | * func3 -> sub rsp, 32 (total stack size 128 + 192 + 64 + 32 = 416) | |
3438 | * | |
3439 | * tailcall will unwind the current stack frame but it will not get rid | |
3440 | * of caller's stack as shown on the example above. | |
3441 | */ | |
3442 | if (idx && subprog[idx].has_tail_call && depth >= 256) { | |
3443 | verbose(env, | |
3444 | "tail_calls are not allowed when call stack of previous frames is %d bytes. Too large\n", | |
3445 | depth); | |
3446 | return -EACCES; | |
3447 | } | |
70a87ffe AS |
3448 | /* round up to 32-bytes, since this is granularity |
3449 | * of interpreter stack size | |
3450 | */ | |
9c8105bd | 3451 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 3452 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 3453 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 3454 | frame + 1, depth); |
f4d7e40a AS |
3455 | return -EACCES; |
3456 | } | |
70a87ffe | 3457 | continue_func: |
4cb3d99c | 3458 | subprog_end = subprog[idx + 1].start; |
70a87ffe | 3459 | for (; i < subprog_end; i++) { |
69c087ba | 3460 | if (!bpf_pseudo_call(insn + i) && !bpf_pseudo_func(insn + i)) |
70a87ffe AS |
3461 | continue; |
3462 | /* remember insn and function to return to */ | |
3463 | ret_insn[frame] = i + 1; | |
9c8105bd | 3464 | ret_prog[frame] = idx; |
70a87ffe AS |
3465 | |
3466 | /* find the callee */ | |
3467 | i = i + insn[i].imm + 1; | |
9c8105bd JW |
3468 | idx = find_subprog(env, i); |
3469 | if (idx < 0) { | |
70a87ffe AS |
3470 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
3471 | i); | |
3472 | return -EFAULT; | |
3473 | } | |
ebf7d1f5 MF |
3474 | |
3475 | if (subprog[idx].has_tail_call) | |
3476 | tail_call_reachable = true; | |
3477 | ||
70a87ffe AS |
3478 | frame++; |
3479 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
3480 | verbose(env, "the call stack of %d frames is too deep !\n", |
3481 | frame); | |
3482 | return -E2BIG; | |
70a87ffe AS |
3483 | } |
3484 | goto process_func; | |
3485 | } | |
ebf7d1f5 MF |
3486 | /* if tail call got detected across bpf2bpf calls then mark each of the |
3487 | * currently present subprog frames as tail call reachable subprogs; | |
3488 | * this info will be utilized by JIT so that we will be preserving the | |
3489 | * tail call counter throughout bpf2bpf calls combined with tailcalls | |
3490 | */ | |
3491 | if (tail_call_reachable) | |
3492 | for (j = 0; j < frame; j++) | |
3493 | subprog[ret_prog[j]].tail_call_reachable = true; | |
3494 | ||
70a87ffe AS |
3495 | /* end of for() loop means the last insn of the 'subprog' |
3496 | * was reached. Doesn't matter whether it was JA or EXIT | |
3497 | */ | |
3498 | if (frame == 0) | |
3499 | return 0; | |
9c8105bd | 3500 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
3501 | frame--; |
3502 | i = ret_insn[frame]; | |
9c8105bd | 3503 | idx = ret_prog[frame]; |
70a87ffe | 3504 | goto continue_func; |
f4d7e40a AS |
3505 | } |
3506 | ||
19d28fbd | 3507 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
3508 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
3509 | const struct bpf_insn *insn, int idx) | |
3510 | { | |
3511 | int start = idx + insn->imm + 1, subprog; | |
3512 | ||
3513 | subprog = find_subprog(env, start); | |
3514 | if (subprog < 0) { | |
3515 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
3516 | start); | |
3517 | return -EFAULT; | |
3518 | } | |
9c8105bd | 3519 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 3520 | } |
19d28fbd | 3521 | #endif |
1ea47e01 | 3522 | |
51c39bb1 AS |
3523 | int check_ctx_reg(struct bpf_verifier_env *env, |
3524 | const struct bpf_reg_state *reg, int regno) | |
58990d1f DB |
3525 | { |
3526 | /* Access to ctx or passing it to a helper is only allowed in | |
3527 | * its original, unmodified form. | |
3528 | */ | |
3529 | ||
3530 | if (reg->off) { | |
3531 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
3532 | regno, reg->off); | |
3533 | return -EACCES; | |
3534 | } | |
3535 | ||
3536 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3537 | char tn_buf[48]; | |
3538 | ||
3539 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3540 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
3541 | return -EACCES; | |
3542 | } | |
3543 | ||
3544 | return 0; | |
3545 | } | |
3546 | ||
afbf21dc YS |
3547 | static int __check_buffer_access(struct bpf_verifier_env *env, |
3548 | const char *buf_info, | |
3549 | const struct bpf_reg_state *reg, | |
3550 | int regno, int off, int size) | |
9df1c28b MM |
3551 | { |
3552 | if (off < 0) { | |
3553 | verbose(env, | |
4fc00b79 | 3554 | "R%d invalid %s buffer access: off=%d, size=%d\n", |
afbf21dc | 3555 | regno, buf_info, off, size); |
9df1c28b MM |
3556 | return -EACCES; |
3557 | } | |
3558 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3559 | char tn_buf[48]; | |
3560 | ||
3561 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3562 | verbose(env, | |
4fc00b79 | 3563 | "R%d invalid variable buffer offset: off=%d, var_off=%s\n", |
9df1c28b MM |
3564 | regno, off, tn_buf); |
3565 | return -EACCES; | |
3566 | } | |
afbf21dc YS |
3567 | |
3568 | return 0; | |
3569 | } | |
3570 | ||
3571 | static int check_tp_buffer_access(struct bpf_verifier_env *env, | |
3572 | const struct bpf_reg_state *reg, | |
3573 | int regno, int off, int size) | |
3574 | { | |
3575 | int err; | |
3576 | ||
3577 | err = __check_buffer_access(env, "tracepoint", reg, regno, off, size); | |
3578 | if (err) | |
3579 | return err; | |
3580 | ||
9df1c28b MM |
3581 | if (off + size > env->prog->aux->max_tp_access) |
3582 | env->prog->aux->max_tp_access = off + size; | |
3583 | ||
3584 | return 0; | |
3585 | } | |
3586 | ||
afbf21dc YS |
3587 | static int check_buffer_access(struct bpf_verifier_env *env, |
3588 | const struct bpf_reg_state *reg, | |
3589 | int regno, int off, int size, | |
3590 | bool zero_size_allowed, | |
3591 | const char *buf_info, | |
3592 | u32 *max_access) | |
3593 | { | |
3594 | int err; | |
3595 | ||
3596 | err = __check_buffer_access(env, buf_info, reg, regno, off, size); | |
3597 | if (err) | |
3598 | return err; | |
3599 | ||
3600 | if (off + size > *max_access) | |
3601 | *max_access = off + size; | |
3602 | ||
3603 | return 0; | |
3604 | } | |
3605 | ||
3f50f132 JF |
3606 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
3607 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
3608 | { | |
3609 | reg->var_off = tnum_subreg(reg->var_off); | |
3610 | __reg_assign_32_into_64(reg); | |
3611 | } | |
9df1c28b | 3612 | |
0c17d1d2 JH |
3613 | /* truncate register to smaller size (in bytes) |
3614 | * must be called with size < BPF_REG_SIZE | |
3615 | */ | |
3616 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
3617 | { | |
3618 | u64 mask; | |
3619 | ||
3620 | /* clear high bits in bit representation */ | |
3621 | reg->var_off = tnum_cast(reg->var_off, size); | |
3622 | ||
3623 | /* fix arithmetic bounds */ | |
3624 | mask = ((u64)1 << (size * 8)) - 1; | |
3625 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
3626 | reg->umin_value &= mask; | |
3627 | reg->umax_value &= mask; | |
3628 | } else { | |
3629 | reg->umin_value = 0; | |
3630 | reg->umax_value = mask; | |
3631 | } | |
3632 | reg->smin_value = reg->umin_value; | |
3633 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
3634 | |
3635 | /* If size is smaller than 32bit register the 32bit register | |
3636 | * values are also truncated so we push 64-bit bounds into | |
3637 | * 32-bit bounds. Above were truncated < 32-bits already. | |
3638 | */ | |
3639 | if (size >= 4) | |
3640 | return; | |
3641 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
3642 | } |
3643 | ||
a23740ec AN |
3644 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
3645 | { | |
3646 | return (map->map_flags & BPF_F_RDONLY_PROG) && map->frozen; | |
3647 | } | |
3648 | ||
3649 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val) | |
3650 | { | |
3651 | void *ptr; | |
3652 | u64 addr; | |
3653 | int err; | |
3654 | ||
3655 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
3656 | if (err) | |
3657 | return err; | |
2dedd7d2 | 3658 | ptr = (void *)(long)addr + off; |
a23740ec AN |
3659 | |
3660 | switch (size) { | |
3661 | case sizeof(u8): | |
3662 | *val = (u64)*(u8 *)ptr; | |
3663 | break; | |
3664 | case sizeof(u16): | |
3665 | *val = (u64)*(u16 *)ptr; | |
3666 | break; | |
3667 | case sizeof(u32): | |
3668 | *val = (u64)*(u32 *)ptr; | |
3669 | break; | |
3670 | case sizeof(u64): | |
3671 | *val = *(u64 *)ptr; | |
3672 | break; | |
3673 | default: | |
3674 | return -EINVAL; | |
3675 | } | |
3676 | return 0; | |
3677 | } | |
3678 | ||
9e15db66 AS |
3679 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
3680 | struct bpf_reg_state *regs, | |
3681 | int regno, int off, int size, | |
3682 | enum bpf_access_type atype, | |
3683 | int value_regno) | |
3684 | { | |
3685 | struct bpf_reg_state *reg = regs + regno; | |
22dc4a0f AN |
3686 | const struct btf_type *t = btf_type_by_id(reg->btf, reg->btf_id); |
3687 | const char *tname = btf_name_by_offset(reg->btf, t->name_off); | |
9e15db66 AS |
3688 | u32 btf_id; |
3689 | int ret; | |
3690 | ||
9e15db66 AS |
3691 | if (off < 0) { |
3692 | verbose(env, | |
3693 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
3694 | regno, tname, off); | |
3695 | return -EACCES; | |
3696 | } | |
3697 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3698 | char tn_buf[48]; | |
3699 | ||
3700 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3701 | verbose(env, | |
3702 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
3703 | regno, tname, off, tn_buf); | |
3704 | return -EACCES; | |
3705 | } | |
3706 | ||
27ae7997 | 3707 | if (env->ops->btf_struct_access) { |
22dc4a0f AN |
3708 | ret = env->ops->btf_struct_access(&env->log, reg->btf, t, |
3709 | off, size, atype, &btf_id); | |
27ae7997 MKL |
3710 | } else { |
3711 | if (atype != BPF_READ) { | |
3712 | verbose(env, "only read is supported\n"); | |
3713 | return -EACCES; | |
3714 | } | |
3715 | ||
22dc4a0f AN |
3716 | ret = btf_struct_access(&env->log, reg->btf, t, off, size, |
3717 | atype, &btf_id); | |
27ae7997 MKL |
3718 | } |
3719 | ||
9e15db66 AS |
3720 | if (ret < 0) |
3721 | return ret; | |
3722 | ||
41c48f3a | 3723 | if (atype == BPF_READ && value_regno >= 0) |
22dc4a0f | 3724 | mark_btf_ld_reg(env, regs, value_regno, ret, reg->btf, btf_id); |
41c48f3a AI |
3725 | |
3726 | return 0; | |
3727 | } | |
3728 | ||
3729 | static int check_ptr_to_map_access(struct bpf_verifier_env *env, | |
3730 | struct bpf_reg_state *regs, | |
3731 | int regno, int off, int size, | |
3732 | enum bpf_access_type atype, | |
3733 | int value_regno) | |
3734 | { | |
3735 | struct bpf_reg_state *reg = regs + regno; | |
3736 | struct bpf_map *map = reg->map_ptr; | |
3737 | const struct btf_type *t; | |
3738 | const char *tname; | |
3739 | u32 btf_id; | |
3740 | int ret; | |
3741 | ||
3742 | if (!btf_vmlinux) { | |
3743 | verbose(env, "map_ptr access not supported without CONFIG_DEBUG_INFO_BTF\n"); | |
3744 | return -ENOTSUPP; | |
3745 | } | |
3746 | ||
3747 | if (!map->ops->map_btf_id || !*map->ops->map_btf_id) { | |
3748 | verbose(env, "map_ptr access not supported for map type %d\n", | |
3749 | map->map_type); | |
3750 | return -ENOTSUPP; | |
3751 | } | |
3752 | ||
3753 | t = btf_type_by_id(btf_vmlinux, *map->ops->map_btf_id); | |
3754 | tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
3755 | ||
3756 | if (!env->allow_ptr_to_map_access) { | |
3757 | verbose(env, | |
3758 | "%s access is allowed only to CAP_PERFMON and CAP_SYS_ADMIN\n", | |
3759 | tname); | |
3760 | return -EPERM; | |
9e15db66 | 3761 | } |
27ae7997 | 3762 | |
41c48f3a AI |
3763 | if (off < 0) { |
3764 | verbose(env, "R%d is %s invalid negative access: off=%d\n", | |
3765 | regno, tname, off); | |
3766 | return -EACCES; | |
3767 | } | |
3768 | ||
3769 | if (atype != BPF_READ) { | |
3770 | verbose(env, "only read from %s is supported\n", tname); | |
3771 | return -EACCES; | |
3772 | } | |
3773 | ||
22dc4a0f | 3774 | ret = btf_struct_access(&env->log, btf_vmlinux, t, off, size, atype, &btf_id); |
41c48f3a AI |
3775 | if (ret < 0) |
3776 | return ret; | |
3777 | ||
3778 | if (value_regno >= 0) | |
22dc4a0f | 3779 | mark_btf_ld_reg(env, regs, value_regno, ret, btf_vmlinux, btf_id); |
41c48f3a | 3780 | |
9e15db66 AS |
3781 | return 0; |
3782 | } | |
3783 | ||
01f810ac AM |
3784 | /* Check that the stack access at the given offset is within bounds. The |
3785 | * maximum valid offset is -1. | |
3786 | * | |
3787 | * The minimum valid offset is -MAX_BPF_STACK for writes, and | |
3788 | * -state->allocated_stack for reads. | |
3789 | */ | |
3790 | static int check_stack_slot_within_bounds(int off, | |
3791 | struct bpf_func_state *state, | |
3792 | enum bpf_access_type t) | |
3793 | { | |
3794 | int min_valid_off; | |
3795 | ||
3796 | if (t == BPF_WRITE) | |
3797 | min_valid_off = -MAX_BPF_STACK; | |
3798 | else | |
3799 | min_valid_off = -state->allocated_stack; | |
3800 | ||
3801 | if (off < min_valid_off || off > -1) | |
3802 | return -EACCES; | |
3803 | return 0; | |
3804 | } | |
3805 | ||
3806 | /* Check that the stack access at 'regno + off' falls within the maximum stack | |
3807 | * bounds. | |
3808 | * | |
3809 | * 'off' includes `regno->offset`, but not its dynamic part (if any). | |
3810 | */ | |
3811 | static int check_stack_access_within_bounds( | |
3812 | struct bpf_verifier_env *env, | |
3813 | int regno, int off, int access_size, | |
3814 | enum stack_access_src src, enum bpf_access_type type) | |
3815 | { | |
3816 | struct bpf_reg_state *regs = cur_regs(env); | |
3817 | struct bpf_reg_state *reg = regs + regno; | |
3818 | struct bpf_func_state *state = func(env, reg); | |
3819 | int min_off, max_off; | |
3820 | int err; | |
3821 | char *err_extra; | |
3822 | ||
3823 | if (src == ACCESS_HELPER) | |
3824 | /* We don't know if helpers are reading or writing (or both). */ | |
3825 | err_extra = " indirect access to"; | |
3826 | else if (type == BPF_READ) | |
3827 | err_extra = " read from"; | |
3828 | else | |
3829 | err_extra = " write to"; | |
3830 | ||
3831 | if (tnum_is_const(reg->var_off)) { | |
3832 | min_off = reg->var_off.value + off; | |
3833 | if (access_size > 0) | |
3834 | max_off = min_off + access_size - 1; | |
3835 | else | |
3836 | max_off = min_off; | |
3837 | } else { | |
3838 | if (reg->smax_value >= BPF_MAX_VAR_OFF || | |
3839 | reg->smin_value <= -BPF_MAX_VAR_OFF) { | |
3840 | verbose(env, "invalid unbounded variable-offset%s stack R%d\n", | |
3841 | err_extra, regno); | |
3842 | return -EACCES; | |
3843 | } | |
3844 | min_off = reg->smin_value + off; | |
3845 | if (access_size > 0) | |
3846 | max_off = reg->smax_value + off + access_size - 1; | |
3847 | else | |
3848 | max_off = min_off; | |
3849 | } | |
3850 | ||
3851 | err = check_stack_slot_within_bounds(min_off, state, type); | |
3852 | if (!err) | |
3853 | err = check_stack_slot_within_bounds(max_off, state, type); | |
3854 | ||
3855 | if (err) { | |
3856 | if (tnum_is_const(reg->var_off)) { | |
3857 | verbose(env, "invalid%s stack R%d off=%d size=%d\n", | |
3858 | err_extra, regno, off, access_size); | |
3859 | } else { | |
3860 | char tn_buf[48]; | |
3861 | ||
3862 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3863 | verbose(env, "invalid variable-offset%s stack R%d var_off=%s size=%d\n", | |
3864 | err_extra, regno, tn_buf, access_size); | |
3865 | } | |
3866 | } | |
3867 | return err; | |
3868 | } | |
41c48f3a | 3869 | |
17a52670 AS |
3870 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
3871 | * if t==write, value_regno is a register which value is stored into memory | |
3872 | * if t==read, value_regno is a register which will receive the value from memory | |
3873 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
3874 | * if t==read && value_regno==-1, don't care what we read from memory | |
3875 | */ | |
ca369602 DB |
3876 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
3877 | int off, int bpf_size, enum bpf_access_type t, | |
3878 | int value_regno, bool strict_alignment_once) | |
17a52670 | 3879 | { |
638f5b90 AS |
3880 | struct bpf_reg_state *regs = cur_regs(env); |
3881 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 3882 | struct bpf_func_state *state; |
17a52670 AS |
3883 | int size, err = 0; |
3884 | ||
3885 | size = bpf_size_to_bytes(bpf_size); | |
3886 | if (size < 0) | |
3887 | return size; | |
3888 | ||
f1174f77 | 3889 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 3890 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
3891 | if (err) |
3892 | return err; | |
17a52670 | 3893 | |
f1174f77 EC |
3894 | /* for access checks, reg->off is just part of off */ |
3895 | off += reg->off; | |
3896 | ||
69c087ba YS |
3897 | if (reg->type == PTR_TO_MAP_KEY) { |
3898 | if (t == BPF_WRITE) { | |
3899 | verbose(env, "write to change key R%d not allowed\n", regno); | |
3900 | return -EACCES; | |
3901 | } | |
3902 | ||
3903 | err = check_mem_region_access(env, regno, off, size, | |
3904 | reg->map_ptr->key_size, false); | |
3905 | if (err) | |
3906 | return err; | |
3907 | if (value_regno >= 0) | |
3908 | mark_reg_unknown(env, regs, value_regno); | |
3909 | } else if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
3910 | if (t == BPF_WRITE && value_regno >= 0 && |
3911 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 3912 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
3913 | return -EACCES; |
3914 | } | |
591fe988 DB |
3915 | err = check_map_access_type(env, regno, off, size, t); |
3916 | if (err) | |
3917 | return err; | |
9fd29c08 | 3918 | err = check_map_access(env, regno, off, size, false); |
a23740ec AN |
3919 | if (!err && t == BPF_READ && value_regno >= 0) { |
3920 | struct bpf_map *map = reg->map_ptr; | |
3921 | ||
3922 | /* if map is read-only, track its contents as scalars */ | |
3923 | if (tnum_is_const(reg->var_off) && | |
3924 | bpf_map_is_rdonly(map) && | |
3925 | map->ops->map_direct_value_addr) { | |
3926 | int map_off = off + reg->var_off.value; | |
3927 | u64 val = 0; | |
3928 | ||
3929 | err = bpf_map_direct_read(map, map_off, size, | |
3930 | &val); | |
3931 | if (err) | |
3932 | return err; | |
3933 | ||
3934 | regs[value_regno].type = SCALAR_VALUE; | |
3935 | __mark_reg_known(®s[value_regno], val); | |
3936 | } else { | |
3937 | mark_reg_unknown(env, regs, value_regno); | |
3938 | } | |
3939 | } | |
457f4436 AN |
3940 | } else if (reg->type == PTR_TO_MEM) { |
3941 | if (t == BPF_WRITE && value_regno >= 0 && | |
3942 | is_pointer_value(env, value_regno)) { | |
3943 | verbose(env, "R%d leaks addr into mem\n", value_regno); | |
3944 | return -EACCES; | |
3945 | } | |
3946 | err = check_mem_region_access(env, regno, off, size, | |
3947 | reg->mem_size, false); | |
3948 | if (!err && t == BPF_READ && value_regno >= 0) | |
3949 | mark_reg_unknown(env, regs, value_regno); | |
1a0dc1ac | 3950 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 3951 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
22dc4a0f | 3952 | struct btf *btf = NULL; |
9e15db66 | 3953 | u32 btf_id = 0; |
19de99f7 | 3954 | |
1be7f75d AS |
3955 | if (t == BPF_WRITE && value_regno >= 0 && |
3956 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 3957 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
3958 | return -EACCES; |
3959 | } | |
f1174f77 | 3960 | |
58990d1f DB |
3961 | err = check_ctx_reg(env, reg, regno); |
3962 | if (err < 0) | |
3963 | return err; | |
3964 | ||
22dc4a0f | 3965 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf, &btf_id); |
9e15db66 AS |
3966 | if (err) |
3967 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 3968 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 3969 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
3970 | * PTR_TO_PACKET[_META,_END]. In the latter |
3971 | * case, we know the offset is zero. | |
f1174f77 | 3972 | */ |
46f8bc92 | 3973 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 3974 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 3975 | } else { |
638f5b90 | 3976 | mark_reg_known_zero(env, regs, |
61bd5218 | 3977 | value_regno); |
46f8bc92 MKL |
3978 | if (reg_type_may_be_null(reg_type)) |
3979 | regs[value_regno].id = ++env->id_gen; | |
5327ed3d JW |
3980 | /* A load of ctx field could have different |
3981 | * actual load size with the one encoded in the | |
3982 | * insn. When the dst is PTR, it is for sure not | |
3983 | * a sub-register. | |
3984 | */ | |
3985 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
b121b341 | 3986 | if (reg_type == PTR_TO_BTF_ID || |
22dc4a0f AN |
3987 | reg_type == PTR_TO_BTF_ID_OR_NULL) { |
3988 | regs[value_regno].btf = btf; | |
9e15db66 | 3989 | regs[value_regno].btf_id = btf_id; |
22dc4a0f | 3990 | } |
46f8bc92 | 3991 | } |
638f5b90 | 3992 | regs[value_regno].type = reg_type; |
969bf05e | 3993 | } |
17a52670 | 3994 | |
f1174f77 | 3995 | } else if (reg->type == PTR_TO_STACK) { |
01f810ac AM |
3996 | /* Basic bounds checks. */ |
3997 | err = check_stack_access_within_bounds(env, regno, off, size, ACCESS_DIRECT, t); | |
e4298d25 DB |
3998 | if (err) |
3999 | return err; | |
8726679a | 4000 | |
f4d7e40a AS |
4001 | state = func(env, reg); |
4002 | err = update_stack_depth(env, state, off); | |
4003 | if (err) | |
4004 | return err; | |
8726679a | 4005 | |
01f810ac AM |
4006 | if (t == BPF_READ) |
4007 | err = check_stack_read(env, regno, off, size, | |
61bd5218 | 4008 | value_regno); |
01f810ac AM |
4009 | else |
4010 | err = check_stack_write(env, regno, off, size, | |
4011 | value_regno, insn_idx); | |
de8f3a83 | 4012 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 4013 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 4014 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
4015 | return -EACCES; |
4016 | } | |
4acf6c0b BB |
4017 | if (t == BPF_WRITE && value_regno >= 0 && |
4018 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
4019 | verbose(env, "R%d leaks addr into packet\n", |
4020 | value_regno); | |
4acf6c0b BB |
4021 | return -EACCES; |
4022 | } | |
9fd29c08 | 4023 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 4024 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 4025 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
4026 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
4027 | if (t == BPF_WRITE && value_regno >= 0 && | |
4028 | is_pointer_value(env, value_regno)) { | |
4029 | verbose(env, "R%d leaks addr into flow keys\n", | |
4030 | value_regno); | |
4031 | return -EACCES; | |
4032 | } | |
4033 | ||
4034 | err = check_flow_keys_access(env, off, size); | |
4035 | if (!err && t == BPF_READ && value_regno >= 0) | |
4036 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 4037 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 4038 | if (t == BPF_WRITE) { |
46f8bc92 MKL |
4039 | verbose(env, "R%d cannot write into %s\n", |
4040 | regno, reg_type_str[reg->type]); | |
c64b7983 JS |
4041 | return -EACCES; |
4042 | } | |
5f456649 | 4043 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
4044 | if (!err && value_regno >= 0) |
4045 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
4046 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
4047 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
4048 | if (!err && t == BPF_READ && value_regno >= 0) | |
4049 | mark_reg_unknown(env, regs, value_regno); | |
9e15db66 AS |
4050 | } else if (reg->type == PTR_TO_BTF_ID) { |
4051 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, | |
4052 | value_regno); | |
41c48f3a AI |
4053 | } else if (reg->type == CONST_PTR_TO_MAP) { |
4054 | err = check_ptr_to_map_access(env, regs, regno, off, size, t, | |
4055 | value_regno); | |
afbf21dc YS |
4056 | } else if (reg->type == PTR_TO_RDONLY_BUF) { |
4057 | if (t == BPF_WRITE) { | |
4058 | verbose(env, "R%d cannot write into %s\n", | |
4059 | regno, reg_type_str[reg->type]); | |
4060 | return -EACCES; | |
4061 | } | |
f6dfbe31 CIK |
4062 | err = check_buffer_access(env, reg, regno, off, size, false, |
4063 | "rdonly", | |
afbf21dc YS |
4064 | &env->prog->aux->max_rdonly_access); |
4065 | if (!err && value_regno >= 0) | |
4066 | mark_reg_unknown(env, regs, value_regno); | |
4067 | } else if (reg->type == PTR_TO_RDWR_BUF) { | |
f6dfbe31 CIK |
4068 | err = check_buffer_access(env, reg, regno, off, size, false, |
4069 | "rdwr", | |
afbf21dc YS |
4070 | &env->prog->aux->max_rdwr_access); |
4071 | if (!err && t == BPF_READ && value_regno >= 0) | |
4072 | mark_reg_unknown(env, regs, value_regno); | |
17a52670 | 4073 | } else { |
61bd5218 JK |
4074 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
4075 | reg_type_str[reg->type]); | |
17a52670 AS |
4076 | return -EACCES; |
4077 | } | |
969bf05e | 4078 | |
f1174f77 | 4079 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 4080 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 4081 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 4082 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 4083 | } |
17a52670 AS |
4084 | return err; |
4085 | } | |
4086 | ||
91c960b0 | 4087 | static int check_atomic(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 4088 | { |
5ffa2550 | 4089 | int load_reg; |
17a52670 AS |
4090 | int err; |
4091 | ||
5ca419f2 BJ |
4092 | switch (insn->imm) { |
4093 | case BPF_ADD: | |
4094 | case BPF_ADD | BPF_FETCH: | |
981f94c3 BJ |
4095 | case BPF_AND: |
4096 | case BPF_AND | BPF_FETCH: | |
4097 | case BPF_OR: | |
4098 | case BPF_OR | BPF_FETCH: | |
4099 | case BPF_XOR: | |
4100 | case BPF_XOR | BPF_FETCH: | |
5ffa2550 BJ |
4101 | case BPF_XCHG: |
4102 | case BPF_CMPXCHG: | |
5ca419f2 BJ |
4103 | break; |
4104 | default: | |
91c960b0 BJ |
4105 | verbose(env, "BPF_ATOMIC uses invalid atomic opcode %02x\n", insn->imm); |
4106 | return -EINVAL; | |
4107 | } | |
4108 | ||
4109 | if (BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) { | |
4110 | verbose(env, "invalid atomic operand size\n"); | |
17a52670 AS |
4111 | return -EINVAL; |
4112 | } | |
4113 | ||
4114 | /* check src1 operand */ | |
dc503a8a | 4115 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4116 | if (err) |
4117 | return err; | |
4118 | ||
4119 | /* check src2 operand */ | |
dc503a8a | 4120 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4121 | if (err) |
4122 | return err; | |
4123 | ||
5ffa2550 BJ |
4124 | if (insn->imm == BPF_CMPXCHG) { |
4125 | /* Check comparison of R0 with memory location */ | |
4126 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
4127 | if (err) | |
4128 | return err; | |
4129 | } | |
4130 | ||
6bdf6abc | 4131 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 4132 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
4133 | return -EACCES; |
4134 | } | |
4135 | ||
ca369602 | 4136 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 4137 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
4138 | is_flow_key_reg(env, insn->dst_reg) || |
4139 | is_sk_reg(env, insn->dst_reg)) { | |
91c960b0 | 4140 | verbose(env, "BPF_ATOMIC stores into R%d %s is not allowed\n", |
2a159c6f DB |
4141 | insn->dst_reg, |
4142 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
4143 | return -EACCES; |
4144 | } | |
4145 | ||
37086bfd BJ |
4146 | if (insn->imm & BPF_FETCH) { |
4147 | if (insn->imm == BPF_CMPXCHG) | |
4148 | load_reg = BPF_REG_0; | |
4149 | else | |
4150 | load_reg = insn->src_reg; | |
4151 | ||
4152 | /* check and record load of old value */ | |
4153 | err = check_reg_arg(env, load_reg, DST_OP); | |
4154 | if (err) | |
4155 | return err; | |
4156 | } else { | |
4157 | /* This instruction accesses a memory location but doesn't | |
4158 | * actually load it into a register. | |
4159 | */ | |
4160 | load_reg = -1; | |
4161 | } | |
4162 | ||
91c960b0 | 4163 | /* check whether we can read the memory */ |
31fd8581 | 4164 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
37086bfd | 4165 | BPF_SIZE(insn->code), BPF_READ, load_reg, true); |
17a52670 AS |
4166 | if (err) |
4167 | return err; | |
4168 | ||
91c960b0 | 4169 | /* check whether we can write into the same memory */ |
5ca419f2 BJ |
4170 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
4171 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); | |
4172 | if (err) | |
4173 | return err; | |
4174 | ||
5ca419f2 | 4175 | return 0; |
17a52670 AS |
4176 | } |
4177 | ||
01f810ac AM |
4178 | /* When register 'regno' is used to read the stack (either directly or through |
4179 | * a helper function) make sure that it's within stack boundary and, depending | |
4180 | * on the access type, that all elements of the stack are initialized. | |
4181 | * | |
4182 | * 'off' includes 'regno->off', but not its dynamic part (if any). | |
4183 | * | |
4184 | * All registers that have been spilled on the stack in the slots within the | |
4185 | * read offsets are marked as read. | |
4186 | */ | |
4187 | static int check_stack_range_initialized( | |
4188 | struct bpf_verifier_env *env, int regno, int off, | |
4189 | int access_size, bool zero_size_allowed, | |
4190 | enum stack_access_src type, struct bpf_call_arg_meta *meta) | |
2011fccf AI |
4191 | { |
4192 | struct bpf_reg_state *reg = reg_state(env, regno); | |
01f810ac AM |
4193 | struct bpf_func_state *state = func(env, reg); |
4194 | int err, min_off, max_off, i, j, slot, spi; | |
4195 | char *err_extra = type == ACCESS_HELPER ? " indirect" : ""; | |
4196 | enum bpf_access_type bounds_check_type; | |
4197 | /* Some accesses can write anything into the stack, others are | |
4198 | * read-only. | |
4199 | */ | |
4200 | bool clobber = false; | |
2011fccf | 4201 | |
01f810ac AM |
4202 | if (access_size == 0 && !zero_size_allowed) { |
4203 | verbose(env, "invalid zero-sized read\n"); | |
2011fccf AI |
4204 | return -EACCES; |
4205 | } | |
2011fccf | 4206 | |
01f810ac AM |
4207 | if (type == ACCESS_HELPER) { |
4208 | /* The bounds checks for writes are more permissive than for | |
4209 | * reads. However, if raw_mode is not set, we'll do extra | |
4210 | * checks below. | |
4211 | */ | |
4212 | bounds_check_type = BPF_WRITE; | |
4213 | clobber = true; | |
4214 | } else { | |
4215 | bounds_check_type = BPF_READ; | |
4216 | } | |
4217 | err = check_stack_access_within_bounds(env, regno, off, access_size, | |
4218 | type, bounds_check_type); | |
4219 | if (err) | |
4220 | return err; | |
4221 | ||
17a52670 | 4222 | |
2011fccf | 4223 | if (tnum_is_const(reg->var_off)) { |
01f810ac | 4224 | min_off = max_off = reg->var_off.value + off; |
2011fccf | 4225 | } else { |
088ec26d AI |
4226 | /* Variable offset is prohibited for unprivileged mode for |
4227 | * simplicity since it requires corresponding support in | |
4228 | * Spectre masking for stack ALU. | |
4229 | * See also retrieve_ptr_limit(). | |
4230 | */ | |
2c78ee89 | 4231 | if (!env->bypass_spec_v1) { |
088ec26d | 4232 | char tn_buf[48]; |
f1174f77 | 4233 | |
088ec26d | 4234 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
01f810ac AM |
4235 | verbose(env, "R%d%s variable offset stack access prohibited for !root, var_off=%s\n", |
4236 | regno, err_extra, tn_buf); | |
088ec26d AI |
4237 | return -EACCES; |
4238 | } | |
f2bcd05e AI |
4239 | /* Only initialized buffer on stack is allowed to be accessed |
4240 | * with variable offset. With uninitialized buffer it's hard to | |
4241 | * guarantee that whole memory is marked as initialized on | |
4242 | * helper return since specific bounds are unknown what may | |
4243 | * cause uninitialized stack leaking. | |
4244 | */ | |
4245 | if (meta && meta->raw_mode) | |
4246 | meta = NULL; | |
4247 | ||
01f810ac AM |
4248 | min_off = reg->smin_value + off; |
4249 | max_off = reg->smax_value + off; | |
17a52670 AS |
4250 | } |
4251 | ||
435faee1 DB |
4252 | if (meta && meta->raw_mode) { |
4253 | meta->access_size = access_size; | |
4254 | meta->regno = regno; | |
4255 | return 0; | |
4256 | } | |
4257 | ||
2011fccf | 4258 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
4259 | u8 *stype; |
4260 | ||
2011fccf | 4261 | slot = -i - 1; |
638f5b90 | 4262 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
4263 | if (state->allocated_stack <= slot) |
4264 | goto err; | |
4265 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
4266 | if (*stype == STACK_MISC) | |
4267 | goto mark; | |
4268 | if (*stype == STACK_ZERO) { | |
01f810ac AM |
4269 | if (clobber) { |
4270 | /* helper can write anything into the stack */ | |
4271 | *stype = STACK_MISC; | |
4272 | } | |
cc2b14d5 | 4273 | goto mark; |
17a52670 | 4274 | } |
1d68f22b YS |
4275 | |
4276 | if (state->stack[spi].slot_type[0] == STACK_SPILL && | |
4277 | state->stack[spi].spilled_ptr.type == PTR_TO_BTF_ID) | |
4278 | goto mark; | |
4279 | ||
f7cf25b2 | 4280 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
cd17d38f YS |
4281 | (state->stack[spi].spilled_ptr.type == SCALAR_VALUE || |
4282 | env->allow_ptr_leaks)) { | |
01f810ac AM |
4283 | if (clobber) { |
4284 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); | |
4285 | for (j = 0; j < BPF_REG_SIZE; j++) | |
4286 | state->stack[spi].slot_type[j] = STACK_MISC; | |
4287 | } | |
f7cf25b2 AS |
4288 | goto mark; |
4289 | } | |
4290 | ||
cc2b14d5 | 4291 | err: |
2011fccf | 4292 | if (tnum_is_const(reg->var_off)) { |
01f810ac AM |
4293 | verbose(env, "invalid%s read from stack R%d off %d+%d size %d\n", |
4294 | err_extra, regno, min_off, i - min_off, access_size); | |
2011fccf AI |
4295 | } else { |
4296 | char tn_buf[48]; | |
4297 | ||
4298 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
01f810ac AM |
4299 | verbose(env, "invalid%s read from stack R%d var_off %s+%d size %d\n", |
4300 | err_extra, regno, tn_buf, i - min_off, access_size); | |
2011fccf | 4301 | } |
cc2b14d5 AS |
4302 | return -EACCES; |
4303 | mark: | |
4304 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
4305 | * the whole slot to be marked as 'read' | |
4306 | */ | |
679c782d | 4307 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
4308 | state->stack[spi].spilled_ptr.parent, |
4309 | REG_LIVE_READ64); | |
17a52670 | 4310 | } |
2011fccf | 4311 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
4312 | } |
4313 | ||
06c1c049 GB |
4314 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
4315 | int access_size, bool zero_size_allowed, | |
4316 | struct bpf_call_arg_meta *meta) | |
4317 | { | |
638f5b90 | 4318 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 4319 | |
f1174f77 | 4320 | switch (reg->type) { |
06c1c049 | 4321 | case PTR_TO_PACKET: |
de8f3a83 | 4322 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
4323 | return check_packet_access(env, regno, reg->off, access_size, |
4324 | zero_size_allowed); | |
69c087ba YS |
4325 | case PTR_TO_MAP_KEY: |
4326 | return check_mem_region_access(env, regno, reg->off, access_size, | |
4327 | reg->map_ptr->key_size, false); | |
06c1c049 | 4328 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
4329 | if (check_map_access_type(env, regno, reg->off, access_size, |
4330 | meta && meta->raw_mode ? BPF_WRITE : | |
4331 | BPF_READ)) | |
4332 | return -EACCES; | |
9fd29c08 YS |
4333 | return check_map_access(env, regno, reg->off, access_size, |
4334 | zero_size_allowed); | |
457f4436 AN |
4335 | case PTR_TO_MEM: |
4336 | return check_mem_region_access(env, regno, reg->off, | |
4337 | access_size, reg->mem_size, | |
4338 | zero_size_allowed); | |
afbf21dc YS |
4339 | case PTR_TO_RDONLY_BUF: |
4340 | if (meta && meta->raw_mode) | |
4341 | return -EACCES; | |
4342 | return check_buffer_access(env, reg, regno, reg->off, | |
4343 | access_size, zero_size_allowed, | |
4344 | "rdonly", | |
4345 | &env->prog->aux->max_rdonly_access); | |
4346 | case PTR_TO_RDWR_BUF: | |
4347 | return check_buffer_access(env, reg, regno, reg->off, | |
4348 | access_size, zero_size_allowed, | |
4349 | "rdwr", | |
4350 | &env->prog->aux->max_rdwr_access); | |
0d004c02 | 4351 | case PTR_TO_STACK: |
01f810ac AM |
4352 | return check_stack_range_initialized( |
4353 | env, | |
4354 | regno, reg->off, access_size, | |
4355 | zero_size_allowed, ACCESS_HELPER, meta); | |
0d004c02 LB |
4356 | default: /* scalar_value or invalid ptr */ |
4357 | /* Allow zero-byte read from NULL, regardless of pointer type */ | |
4358 | if (zero_size_allowed && access_size == 0 && | |
4359 | register_is_null(reg)) | |
4360 | return 0; | |
4361 | ||
4362 | verbose(env, "R%d type=%s expected=%s\n", regno, | |
4363 | reg_type_str[reg->type], | |
4364 | reg_type_str[PTR_TO_STACK]); | |
4365 | return -EACCES; | |
06c1c049 GB |
4366 | } |
4367 | } | |
4368 | ||
e5069b9c DB |
4369 | int check_mem_reg(struct bpf_verifier_env *env, struct bpf_reg_state *reg, |
4370 | u32 regno, u32 mem_size) | |
4371 | { | |
4372 | if (register_is_null(reg)) | |
4373 | return 0; | |
4374 | ||
4375 | if (reg_type_may_be_null(reg->type)) { | |
4376 | /* Assuming that the register contains a value check if the memory | |
4377 | * access is safe. Temporarily save and restore the register's state as | |
4378 | * the conversion shouldn't be visible to a caller. | |
4379 | */ | |
4380 | const struct bpf_reg_state saved_reg = *reg; | |
4381 | int rv; | |
4382 | ||
4383 | mark_ptr_not_null_reg(reg); | |
4384 | rv = check_helper_mem_access(env, regno, mem_size, true, NULL); | |
4385 | *reg = saved_reg; | |
4386 | return rv; | |
4387 | } | |
4388 | ||
4389 | return check_helper_mem_access(env, regno, mem_size, true, NULL); | |
4390 | } | |
4391 | ||
d83525ca AS |
4392 | /* Implementation details: |
4393 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL | |
4394 | * Two bpf_map_lookups (even with the same key) will have different reg->id. | |
4395 | * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after | |
4396 | * value_or_null->value transition, since the verifier only cares about | |
4397 | * the range of access to valid map value pointer and doesn't care about actual | |
4398 | * address of the map element. | |
4399 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps | |
4400 | * reg->id > 0 after value_or_null->value transition. By doing so | |
4401 | * two bpf_map_lookups will be considered two different pointers that | |
4402 | * point to different bpf_spin_locks. | |
4403 | * The verifier allows taking only one bpf_spin_lock at a time to avoid | |
4404 | * dead-locks. | |
4405 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
4406 | * reg_is_refcounted() logic. The verifier needs to remember only | |
4407 | * one spin_lock instead of array of acquired_refs. | |
4408 | * cur_state->active_spin_lock remembers which map value element got locked | |
4409 | * and clears it after bpf_spin_unlock. | |
4410 | */ | |
4411 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
4412 | bool is_lock) | |
4413 | { | |
4414 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
4415 | struct bpf_verifier_state *cur = env->cur_state; | |
4416 | bool is_const = tnum_is_const(reg->var_off); | |
4417 | struct bpf_map *map = reg->map_ptr; | |
4418 | u64 val = reg->var_off.value; | |
4419 | ||
d83525ca AS |
4420 | if (!is_const) { |
4421 | verbose(env, | |
4422 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
4423 | regno); | |
4424 | return -EINVAL; | |
4425 | } | |
4426 | if (!map->btf) { | |
4427 | verbose(env, | |
4428 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
4429 | map->name); | |
4430 | return -EINVAL; | |
4431 | } | |
4432 | if (!map_value_has_spin_lock(map)) { | |
4433 | if (map->spin_lock_off == -E2BIG) | |
4434 | verbose(env, | |
4435 | "map '%s' has more than one 'struct bpf_spin_lock'\n", | |
4436 | map->name); | |
4437 | else if (map->spin_lock_off == -ENOENT) | |
4438 | verbose(env, | |
4439 | "map '%s' doesn't have 'struct bpf_spin_lock'\n", | |
4440 | map->name); | |
4441 | else | |
4442 | verbose(env, | |
4443 | "map '%s' is not a struct type or bpf_spin_lock is mangled\n", | |
4444 | map->name); | |
4445 | return -EINVAL; | |
4446 | } | |
4447 | if (map->spin_lock_off != val + reg->off) { | |
4448 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock'\n", | |
4449 | val + reg->off); | |
4450 | return -EINVAL; | |
4451 | } | |
4452 | if (is_lock) { | |
4453 | if (cur->active_spin_lock) { | |
4454 | verbose(env, | |
4455 | "Locking two bpf_spin_locks are not allowed\n"); | |
4456 | return -EINVAL; | |
4457 | } | |
4458 | cur->active_spin_lock = reg->id; | |
4459 | } else { | |
4460 | if (!cur->active_spin_lock) { | |
4461 | verbose(env, "bpf_spin_unlock without taking a lock\n"); | |
4462 | return -EINVAL; | |
4463 | } | |
4464 | if (cur->active_spin_lock != reg->id) { | |
4465 | verbose(env, "bpf_spin_unlock of different lock\n"); | |
4466 | return -EINVAL; | |
4467 | } | |
4468 | cur->active_spin_lock = 0; | |
4469 | } | |
4470 | return 0; | |
4471 | } | |
4472 | ||
90133415 DB |
4473 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
4474 | { | |
4475 | return type == ARG_PTR_TO_MEM || | |
4476 | type == ARG_PTR_TO_MEM_OR_NULL || | |
4477 | type == ARG_PTR_TO_UNINIT_MEM; | |
4478 | } | |
4479 | ||
4480 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
4481 | { | |
4482 | return type == ARG_CONST_SIZE || | |
4483 | type == ARG_CONST_SIZE_OR_ZERO; | |
4484 | } | |
4485 | ||
457f4436 AN |
4486 | static bool arg_type_is_alloc_size(enum bpf_arg_type type) |
4487 | { | |
4488 | return type == ARG_CONST_ALLOC_SIZE_OR_ZERO; | |
4489 | } | |
4490 | ||
57c3bb72 AI |
4491 | static bool arg_type_is_int_ptr(enum bpf_arg_type type) |
4492 | { | |
4493 | return type == ARG_PTR_TO_INT || | |
4494 | type == ARG_PTR_TO_LONG; | |
4495 | } | |
4496 | ||
4497 | static int int_ptr_type_to_size(enum bpf_arg_type type) | |
4498 | { | |
4499 | if (type == ARG_PTR_TO_INT) | |
4500 | return sizeof(u32); | |
4501 | else if (type == ARG_PTR_TO_LONG) | |
4502 | return sizeof(u64); | |
4503 | ||
4504 | return -EINVAL; | |
4505 | } | |
4506 | ||
912f442c LB |
4507 | static int resolve_map_arg_type(struct bpf_verifier_env *env, |
4508 | const struct bpf_call_arg_meta *meta, | |
4509 | enum bpf_arg_type *arg_type) | |
4510 | { | |
4511 | if (!meta->map_ptr) { | |
4512 | /* kernel subsystem misconfigured verifier */ | |
4513 | verbose(env, "invalid map_ptr to access map->type\n"); | |
4514 | return -EACCES; | |
4515 | } | |
4516 | ||
4517 | switch (meta->map_ptr->map_type) { | |
4518 | case BPF_MAP_TYPE_SOCKMAP: | |
4519 | case BPF_MAP_TYPE_SOCKHASH: | |
4520 | if (*arg_type == ARG_PTR_TO_MAP_VALUE) { | |
6550f2dd | 4521 | *arg_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON; |
912f442c LB |
4522 | } else { |
4523 | verbose(env, "invalid arg_type for sockmap/sockhash\n"); | |
4524 | return -EINVAL; | |
4525 | } | |
4526 | break; | |
4527 | ||
4528 | default: | |
4529 | break; | |
4530 | } | |
4531 | return 0; | |
4532 | } | |
4533 | ||
f79e7ea5 LB |
4534 | struct bpf_reg_types { |
4535 | const enum bpf_reg_type types[10]; | |
1df8f55a | 4536 | u32 *btf_id; |
f79e7ea5 LB |
4537 | }; |
4538 | ||
4539 | static const struct bpf_reg_types map_key_value_types = { | |
4540 | .types = { | |
4541 | PTR_TO_STACK, | |
4542 | PTR_TO_PACKET, | |
4543 | PTR_TO_PACKET_META, | |
69c087ba | 4544 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4545 | PTR_TO_MAP_VALUE, |
4546 | }, | |
4547 | }; | |
4548 | ||
4549 | static const struct bpf_reg_types sock_types = { | |
4550 | .types = { | |
4551 | PTR_TO_SOCK_COMMON, | |
4552 | PTR_TO_SOCKET, | |
4553 | PTR_TO_TCP_SOCK, | |
4554 | PTR_TO_XDP_SOCK, | |
4555 | }, | |
4556 | }; | |
4557 | ||
49a2a4d4 | 4558 | #ifdef CONFIG_NET |
1df8f55a MKL |
4559 | static const struct bpf_reg_types btf_id_sock_common_types = { |
4560 | .types = { | |
4561 | PTR_TO_SOCK_COMMON, | |
4562 | PTR_TO_SOCKET, | |
4563 | PTR_TO_TCP_SOCK, | |
4564 | PTR_TO_XDP_SOCK, | |
4565 | PTR_TO_BTF_ID, | |
4566 | }, | |
4567 | .btf_id = &btf_sock_ids[BTF_SOCK_TYPE_SOCK_COMMON], | |
4568 | }; | |
49a2a4d4 | 4569 | #endif |
1df8f55a | 4570 | |
f79e7ea5 LB |
4571 | static const struct bpf_reg_types mem_types = { |
4572 | .types = { | |
4573 | PTR_TO_STACK, | |
4574 | PTR_TO_PACKET, | |
4575 | PTR_TO_PACKET_META, | |
69c087ba | 4576 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4577 | PTR_TO_MAP_VALUE, |
4578 | PTR_TO_MEM, | |
4579 | PTR_TO_RDONLY_BUF, | |
4580 | PTR_TO_RDWR_BUF, | |
4581 | }, | |
4582 | }; | |
4583 | ||
4584 | static const struct bpf_reg_types int_ptr_types = { | |
4585 | .types = { | |
4586 | PTR_TO_STACK, | |
4587 | PTR_TO_PACKET, | |
4588 | PTR_TO_PACKET_META, | |
69c087ba | 4589 | PTR_TO_MAP_KEY, |
f79e7ea5 LB |
4590 | PTR_TO_MAP_VALUE, |
4591 | }, | |
4592 | }; | |
4593 | ||
4594 | static const struct bpf_reg_types fullsock_types = { .types = { PTR_TO_SOCKET } }; | |
4595 | static const struct bpf_reg_types scalar_types = { .types = { SCALAR_VALUE } }; | |
4596 | static const struct bpf_reg_types context_types = { .types = { PTR_TO_CTX } }; | |
4597 | static const struct bpf_reg_types alloc_mem_types = { .types = { PTR_TO_MEM } }; | |
4598 | static const struct bpf_reg_types const_map_ptr_types = { .types = { CONST_PTR_TO_MAP } }; | |
4599 | static const struct bpf_reg_types btf_ptr_types = { .types = { PTR_TO_BTF_ID } }; | |
4600 | static const struct bpf_reg_types spin_lock_types = { .types = { PTR_TO_MAP_VALUE } }; | |
eaa6bcb7 | 4601 | static const struct bpf_reg_types percpu_btf_ptr_types = { .types = { PTR_TO_PERCPU_BTF_ID } }; |
69c087ba YS |
4602 | static const struct bpf_reg_types func_ptr_types = { .types = { PTR_TO_FUNC } }; |
4603 | static const struct bpf_reg_types stack_ptr_types = { .types = { PTR_TO_STACK } }; | |
f79e7ea5 | 4604 | |
0789e13b | 4605 | static const struct bpf_reg_types *compatible_reg_types[__BPF_ARG_TYPE_MAX] = { |
f79e7ea5 LB |
4606 | [ARG_PTR_TO_MAP_KEY] = &map_key_value_types, |
4607 | [ARG_PTR_TO_MAP_VALUE] = &map_key_value_types, | |
4608 | [ARG_PTR_TO_UNINIT_MAP_VALUE] = &map_key_value_types, | |
4609 | [ARG_PTR_TO_MAP_VALUE_OR_NULL] = &map_key_value_types, | |
4610 | [ARG_CONST_SIZE] = &scalar_types, | |
4611 | [ARG_CONST_SIZE_OR_ZERO] = &scalar_types, | |
4612 | [ARG_CONST_ALLOC_SIZE_OR_ZERO] = &scalar_types, | |
4613 | [ARG_CONST_MAP_PTR] = &const_map_ptr_types, | |
4614 | [ARG_PTR_TO_CTX] = &context_types, | |
4615 | [ARG_PTR_TO_CTX_OR_NULL] = &context_types, | |
4616 | [ARG_PTR_TO_SOCK_COMMON] = &sock_types, | |
49a2a4d4 | 4617 | #ifdef CONFIG_NET |
1df8f55a | 4618 | [ARG_PTR_TO_BTF_ID_SOCK_COMMON] = &btf_id_sock_common_types, |
49a2a4d4 | 4619 | #endif |
f79e7ea5 LB |
4620 | [ARG_PTR_TO_SOCKET] = &fullsock_types, |
4621 | [ARG_PTR_TO_SOCKET_OR_NULL] = &fullsock_types, | |
4622 | [ARG_PTR_TO_BTF_ID] = &btf_ptr_types, | |
4623 | [ARG_PTR_TO_SPIN_LOCK] = &spin_lock_types, | |
4624 | [ARG_PTR_TO_MEM] = &mem_types, | |
4625 | [ARG_PTR_TO_MEM_OR_NULL] = &mem_types, | |
4626 | [ARG_PTR_TO_UNINIT_MEM] = &mem_types, | |
4627 | [ARG_PTR_TO_ALLOC_MEM] = &alloc_mem_types, | |
4628 | [ARG_PTR_TO_ALLOC_MEM_OR_NULL] = &alloc_mem_types, | |
4629 | [ARG_PTR_TO_INT] = &int_ptr_types, | |
4630 | [ARG_PTR_TO_LONG] = &int_ptr_types, | |
eaa6bcb7 | 4631 | [ARG_PTR_TO_PERCPU_BTF_ID] = &percpu_btf_ptr_types, |
69c087ba YS |
4632 | [ARG_PTR_TO_FUNC] = &func_ptr_types, |
4633 | [ARG_PTR_TO_STACK_OR_NULL] = &stack_ptr_types, | |
f79e7ea5 LB |
4634 | }; |
4635 | ||
4636 | static int check_reg_type(struct bpf_verifier_env *env, u32 regno, | |
a968d5e2 MKL |
4637 | enum bpf_arg_type arg_type, |
4638 | const u32 *arg_btf_id) | |
f79e7ea5 LB |
4639 | { |
4640 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
4641 | enum bpf_reg_type expected, type = reg->type; | |
a968d5e2 | 4642 | const struct bpf_reg_types *compatible; |
f79e7ea5 LB |
4643 | int i, j; |
4644 | ||
a968d5e2 MKL |
4645 | compatible = compatible_reg_types[arg_type]; |
4646 | if (!compatible) { | |
4647 | verbose(env, "verifier internal error: unsupported arg type %d\n", arg_type); | |
4648 | return -EFAULT; | |
4649 | } | |
4650 | ||
f79e7ea5 LB |
4651 | for (i = 0; i < ARRAY_SIZE(compatible->types); i++) { |
4652 | expected = compatible->types[i]; | |
4653 | if (expected == NOT_INIT) | |
4654 | break; | |
4655 | ||
4656 | if (type == expected) | |
a968d5e2 | 4657 | goto found; |
f79e7ea5 LB |
4658 | } |
4659 | ||
4660 | verbose(env, "R%d type=%s expected=", regno, reg_type_str[type]); | |
4661 | for (j = 0; j + 1 < i; j++) | |
4662 | verbose(env, "%s, ", reg_type_str[compatible->types[j]]); | |
4663 | verbose(env, "%s\n", reg_type_str[compatible->types[j]]); | |
4664 | return -EACCES; | |
a968d5e2 MKL |
4665 | |
4666 | found: | |
4667 | if (type == PTR_TO_BTF_ID) { | |
1df8f55a MKL |
4668 | if (!arg_btf_id) { |
4669 | if (!compatible->btf_id) { | |
4670 | verbose(env, "verifier internal error: missing arg compatible BTF ID\n"); | |
4671 | return -EFAULT; | |
4672 | } | |
4673 | arg_btf_id = compatible->btf_id; | |
4674 | } | |
4675 | ||
22dc4a0f AN |
4676 | if (!btf_struct_ids_match(&env->log, reg->btf, reg->btf_id, reg->off, |
4677 | btf_vmlinux, *arg_btf_id)) { | |
a968d5e2 | 4678 | verbose(env, "R%d is of type %s but %s is expected\n", |
22dc4a0f AN |
4679 | regno, kernel_type_name(reg->btf, reg->btf_id), |
4680 | kernel_type_name(btf_vmlinux, *arg_btf_id)); | |
a968d5e2 MKL |
4681 | return -EACCES; |
4682 | } | |
4683 | ||
4684 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
4685 | verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n", | |
4686 | regno); | |
4687 | return -EACCES; | |
4688 | } | |
4689 | } | |
4690 | ||
4691 | return 0; | |
f79e7ea5 LB |
4692 | } |
4693 | ||
af7ec138 YS |
4694 | static int check_func_arg(struct bpf_verifier_env *env, u32 arg, |
4695 | struct bpf_call_arg_meta *meta, | |
4696 | const struct bpf_func_proto *fn) | |
17a52670 | 4697 | { |
af7ec138 | 4698 | u32 regno = BPF_REG_1 + arg; |
638f5b90 | 4699 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
af7ec138 | 4700 | enum bpf_arg_type arg_type = fn->arg_type[arg]; |
f79e7ea5 | 4701 | enum bpf_reg_type type = reg->type; |
17a52670 AS |
4702 | int err = 0; |
4703 | ||
80f1d68c | 4704 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
4705 | return 0; |
4706 | ||
dc503a8a EC |
4707 | err = check_reg_arg(env, regno, SRC_OP); |
4708 | if (err) | |
4709 | return err; | |
17a52670 | 4710 | |
1be7f75d AS |
4711 | if (arg_type == ARG_ANYTHING) { |
4712 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
4713 | verbose(env, "R%d leaks addr into helper function\n", |
4714 | regno); | |
1be7f75d AS |
4715 | return -EACCES; |
4716 | } | |
80f1d68c | 4717 | return 0; |
1be7f75d | 4718 | } |
80f1d68c | 4719 | |
de8f3a83 | 4720 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 4721 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 4722 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
4723 | return -EACCES; |
4724 | } | |
4725 | ||
912f442c LB |
4726 | if (arg_type == ARG_PTR_TO_MAP_VALUE || |
4727 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE || | |
4728 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) { | |
4729 | err = resolve_map_arg_type(env, meta, &arg_type); | |
4730 | if (err) | |
4731 | return err; | |
4732 | } | |
4733 | ||
fd1b0d60 LB |
4734 | if (register_is_null(reg) && arg_type_may_be_null(arg_type)) |
4735 | /* A NULL register has a SCALAR_VALUE type, so skip | |
4736 | * type checking. | |
4737 | */ | |
4738 | goto skip_type_check; | |
4739 | ||
a968d5e2 | 4740 | err = check_reg_type(env, regno, arg_type, fn->arg_btf_id[arg]); |
f79e7ea5 LB |
4741 | if (err) |
4742 | return err; | |
4743 | ||
a968d5e2 | 4744 | if (type == PTR_TO_CTX) { |
feec7040 LB |
4745 | err = check_ctx_reg(env, reg, regno); |
4746 | if (err < 0) | |
4747 | return err; | |
d7b9454a LB |
4748 | } |
4749 | ||
fd1b0d60 | 4750 | skip_type_check: |
02f7c958 | 4751 | if (reg->ref_obj_id) { |
457f4436 AN |
4752 | if (meta->ref_obj_id) { |
4753 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
4754 | regno, reg->ref_obj_id, | |
4755 | meta->ref_obj_id); | |
4756 | return -EFAULT; | |
4757 | } | |
4758 | meta->ref_obj_id = reg->ref_obj_id; | |
17a52670 AS |
4759 | } |
4760 | ||
17a52670 AS |
4761 | if (arg_type == ARG_CONST_MAP_PTR) { |
4762 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 4763 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
4764 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
4765 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
4766 | * check that [key, key + map->key_size) are within | |
4767 | * stack limits and initialized | |
4768 | */ | |
33ff9823 | 4769 | if (!meta->map_ptr) { |
17a52670 AS |
4770 | /* in function declaration map_ptr must come before |
4771 | * map_key, so that it's verified and known before | |
4772 | * we have to check map_key here. Otherwise it means | |
4773 | * that kernel subsystem misconfigured verifier | |
4774 | */ | |
61bd5218 | 4775 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
4776 | return -EACCES; |
4777 | } | |
d71962f3 PC |
4778 | err = check_helper_mem_access(env, regno, |
4779 | meta->map_ptr->key_size, false, | |
4780 | NULL); | |
2ea864c5 | 4781 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
4782 | (arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL && |
4783 | !register_is_null(reg)) || | |
2ea864c5 | 4784 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { |
17a52670 AS |
4785 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
4786 | * check [value, value + map->value_size) validity | |
4787 | */ | |
33ff9823 | 4788 | if (!meta->map_ptr) { |
17a52670 | 4789 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 4790 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
4791 | return -EACCES; |
4792 | } | |
2ea864c5 | 4793 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE); |
d71962f3 PC |
4794 | err = check_helper_mem_access(env, regno, |
4795 | meta->map_ptr->value_size, false, | |
2ea864c5 | 4796 | meta); |
eaa6bcb7 HL |
4797 | } else if (arg_type == ARG_PTR_TO_PERCPU_BTF_ID) { |
4798 | if (!reg->btf_id) { | |
4799 | verbose(env, "Helper has invalid btf_id in R%d\n", regno); | |
4800 | return -EACCES; | |
4801 | } | |
22dc4a0f | 4802 | meta->ret_btf = reg->btf; |
eaa6bcb7 | 4803 | meta->ret_btf_id = reg->btf_id; |
c18f0b6a LB |
4804 | } else if (arg_type == ARG_PTR_TO_SPIN_LOCK) { |
4805 | if (meta->func_id == BPF_FUNC_spin_lock) { | |
4806 | if (process_spin_lock(env, regno, true)) | |
4807 | return -EACCES; | |
4808 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { | |
4809 | if (process_spin_lock(env, regno, false)) | |
4810 | return -EACCES; | |
4811 | } else { | |
4812 | verbose(env, "verifier internal error\n"); | |
4813 | return -EFAULT; | |
4814 | } | |
69c087ba YS |
4815 | } else if (arg_type == ARG_PTR_TO_FUNC) { |
4816 | meta->subprogno = reg->subprogno; | |
a2bbe7cc LB |
4817 | } else if (arg_type_is_mem_ptr(arg_type)) { |
4818 | /* The access to this pointer is only checked when we hit the | |
4819 | * next is_mem_size argument below. | |
4820 | */ | |
4821 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MEM); | |
90133415 | 4822 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 4823 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 4824 | |
10060503 JF |
4825 | /* This is used to refine r0 return value bounds for helpers |
4826 | * that enforce this value as an upper bound on return values. | |
4827 | * See do_refine_retval_range() for helpers that can refine | |
4828 | * the return value. C type of helper is u32 so we pull register | |
4829 | * bound from umax_value however, if negative verifier errors | |
4830 | * out. Only upper bounds can be learned because retval is an | |
4831 | * int type and negative retvals are allowed. | |
849fa506 | 4832 | */ |
10060503 | 4833 | meta->msize_max_value = reg->umax_value; |
849fa506 | 4834 | |
f1174f77 EC |
4835 | /* The register is SCALAR_VALUE; the access check |
4836 | * happens using its boundaries. | |
06c1c049 | 4837 | */ |
f1174f77 | 4838 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
4839 | /* For unprivileged variable accesses, disable raw |
4840 | * mode so that the program is required to | |
4841 | * initialize all the memory that the helper could | |
4842 | * just partially fill up. | |
4843 | */ | |
4844 | meta = NULL; | |
4845 | ||
b03c9f9f | 4846 | if (reg->smin_value < 0) { |
61bd5218 | 4847 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
4848 | regno); |
4849 | return -EACCES; | |
4850 | } | |
06c1c049 | 4851 | |
b03c9f9f | 4852 | if (reg->umin_value == 0) { |
f1174f77 EC |
4853 | err = check_helper_mem_access(env, regno - 1, 0, |
4854 | zero_size_allowed, | |
4855 | meta); | |
06c1c049 GB |
4856 | if (err) |
4857 | return err; | |
06c1c049 | 4858 | } |
f1174f77 | 4859 | |
b03c9f9f | 4860 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 4861 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
4862 | regno); |
4863 | return -EACCES; | |
4864 | } | |
4865 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 4866 | reg->umax_value, |
f1174f77 | 4867 | zero_size_allowed, meta); |
b5dc0163 AS |
4868 | if (!err) |
4869 | err = mark_chain_precision(env, regno); | |
457f4436 AN |
4870 | } else if (arg_type_is_alloc_size(arg_type)) { |
4871 | if (!tnum_is_const(reg->var_off)) { | |
28a8add6 | 4872 | verbose(env, "R%d is not a known constant'\n", |
457f4436 AN |
4873 | regno); |
4874 | return -EACCES; | |
4875 | } | |
4876 | meta->mem_size = reg->var_off.value; | |
57c3bb72 AI |
4877 | } else if (arg_type_is_int_ptr(arg_type)) { |
4878 | int size = int_ptr_type_to_size(arg_type); | |
4879 | ||
4880 | err = check_helper_mem_access(env, regno, size, false, meta); | |
4881 | if (err) | |
4882 | return err; | |
4883 | err = check_ptr_alignment(env, reg, 0, size, true); | |
17a52670 AS |
4884 | } |
4885 | ||
4886 | return err; | |
4887 | } | |
4888 | ||
0126240f LB |
4889 | static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) |
4890 | { | |
4891 | enum bpf_attach_type eatype = env->prog->expected_attach_type; | |
7e40781c | 4892 | enum bpf_prog_type type = resolve_prog_type(env->prog); |
0126240f LB |
4893 | |
4894 | if (func_id != BPF_FUNC_map_update_elem) | |
4895 | return false; | |
4896 | ||
4897 | /* It's not possible to get access to a locked struct sock in these | |
4898 | * contexts, so updating is safe. | |
4899 | */ | |
4900 | switch (type) { | |
4901 | case BPF_PROG_TYPE_TRACING: | |
4902 | if (eatype == BPF_TRACE_ITER) | |
4903 | return true; | |
4904 | break; | |
4905 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
4906 | case BPF_PROG_TYPE_SCHED_CLS: | |
4907 | case BPF_PROG_TYPE_SCHED_ACT: | |
4908 | case BPF_PROG_TYPE_XDP: | |
4909 | case BPF_PROG_TYPE_SK_REUSEPORT: | |
4910 | case BPF_PROG_TYPE_FLOW_DISSECTOR: | |
4911 | case BPF_PROG_TYPE_SK_LOOKUP: | |
4912 | return true; | |
4913 | default: | |
4914 | break; | |
4915 | } | |
4916 | ||
4917 | verbose(env, "cannot update sockmap in this context\n"); | |
4918 | return false; | |
4919 | } | |
4920 | ||
e411901c MF |
4921 | static bool allow_tail_call_in_subprogs(struct bpf_verifier_env *env) |
4922 | { | |
4923 | return env->prog->jit_requested && IS_ENABLED(CONFIG_X86_64); | |
4924 | } | |
4925 | ||
61bd5218 JK |
4926 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
4927 | struct bpf_map *map, int func_id) | |
35578d79 | 4928 | { |
35578d79 KX |
4929 | if (!map) |
4930 | return 0; | |
4931 | ||
6aff67c8 AS |
4932 | /* We need a two way check, first is from map perspective ... */ |
4933 | switch (map->map_type) { | |
4934 | case BPF_MAP_TYPE_PROG_ARRAY: | |
4935 | if (func_id != BPF_FUNC_tail_call) | |
4936 | goto error; | |
4937 | break; | |
4938 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
4939 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 4940 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 4941 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
4942 | func_id != BPF_FUNC_perf_event_read_value && |
4943 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
4944 | goto error; |
4945 | break; | |
457f4436 AN |
4946 | case BPF_MAP_TYPE_RINGBUF: |
4947 | if (func_id != BPF_FUNC_ringbuf_output && | |
4948 | func_id != BPF_FUNC_ringbuf_reserve && | |
4949 | func_id != BPF_FUNC_ringbuf_submit && | |
4950 | func_id != BPF_FUNC_ringbuf_discard && | |
4951 | func_id != BPF_FUNC_ringbuf_query) | |
4952 | goto error; | |
4953 | break; | |
6aff67c8 AS |
4954 | case BPF_MAP_TYPE_STACK_TRACE: |
4955 | if (func_id != BPF_FUNC_get_stackid) | |
4956 | goto error; | |
4957 | break; | |
4ed8ec52 | 4958 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 4959 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 4960 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
4961 | goto error; |
4962 | break; | |
cd339431 | 4963 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 4964 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
4965 | if (func_id != BPF_FUNC_get_local_storage) |
4966 | goto error; | |
4967 | break; | |
546ac1ff | 4968 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 4969 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
4970 | if (func_id != BPF_FUNC_redirect_map && |
4971 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
4972 | goto error; |
4973 | break; | |
fbfc504a BT |
4974 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
4975 | * appear. | |
4976 | */ | |
6710e112 JDB |
4977 | case BPF_MAP_TYPE_CPUMAP: |
4978 | if (func_id != BPF_FUNC_redirect_map) | |
4979 | goto error; | |
4980 | break; | |
fada7fdc JL |
4981 | case BPF_MAP_TYPE_XSKMAP: |
4982 | if (func_id != BPF_FUNC_redirect_map && | |
4983 | func_id != BPF_FUNC_map_lookup_elem) | |
4984 | goto error; | |
4985 | break; | |
56f668df | 4986 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 4987 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
4988 | if (func_id != BPF_FUNC_map_lookup_elem) |
4989 | goto error; | |
16a43625 | 4990 | break; |
174a79ff JF |
4991 | case BPF_MAP_TYPE_SOCKMAP: |
4992 | if (func_id != BPF_FUNC_sk_redirect_map && | |
4993 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 4994 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 4995 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 | 4996 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
4997 | func_id != BPF_FUNC_map_lookup_elem && |
4998 | !may_update_sockmap(env, func_id)) | |
174a79ff JF |
4999 | goto error; |
5000 | break; | |
81110384 JF |
5001 | case BPF_MAP_TYPE_SOCKHASH: |
5002 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
5003 | func_id != BPF_FUNC_sock_hash_update && | |
5004 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 5005 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 | 5006 | func_id != BPF_FUNC_sk_select_reuseport && |
0126240f LB |
5007 | func_id != BPF_FUNC_map_lookup_elem && |
5008 | !may_update_sockmap(env, func_id)) | |
81110384 JF |
5009 | goto error; |
5010 | break; | |
2dbb9b9e MKL |
5011 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
5012 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
5013 | goto error; | |
5014 | break; | |
f1a2e44a MV |
5015 | case BPF_MAP_TYPE_QUEUE: |
5016 | case BPF_MAP_TYPE_STACK: | |
5017 | if (func_id != BPF_FUNC_map_peek_elem && | |
5018 | func_id != BPF_FUNC_map_pop_elem && | |
5019 | func_id != BPF_FUNC_map_push_elem) | |
5020 | goto error; | |
5021 | break; | |
6ac99e8f MKL |
5022 | case BPF_MAP_TYPE_SK_STORAGE: |
5023 | if (func_id != BPF_FUNC_sk_storage_get && | |
5024 | func_id != BPF_FUNC_sk_storage_delete) | |
5025 | goto error; | |
5026 | break; | |
8ea63684 KS |
5027 | case BPF_MAP_TYPE_INODE_STORAGE: |
5028 | if (func_id != BPF_FUNC_inode_storage_get && | |
5029 | func_id != BPF_FUNC_inode_storage_delete) | |
5030 | goto error; | |
5031 | break; | |
4cf1bc1f KS |
5032 | case BPF_MAP_TYPE_TASK_STORAGE: |
5033 | if (func_id != BPF_FUNC_task_storage_get && | |
5034 | func_id != BPF_FUNC_task_storage_delete) | |
5035 | goto error; | |
5036 | break; | |
6aff67c8 AS |
5037 | default: |
5038 | break; | |
5039 | } | |
5040 | ||
5041 | /* ... and second from the function itself. */ | |
5042 | switch (func_id) { | |
5043 | case BPF_FUNC_tail_call: | |
5044 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
5045 | goto error; | |
e411901c MF |
5046 | if (env->subprog_cnt > 1 && !allow_tail_call_in_subprogs(env)) { |
5047 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
f4d7e40a AS |
5048 | return -EINVAL; |
5049 | } | |
6aff67c8 AS |
5050 | break; |
5051 | case BPF_FUNC_perf_event_read: | |
5052 | case BPF_FUNC_perf_event_output: | |
908432ca | 5053 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 5054 | case BPF_FUNC_skb_output: |
d831ee84 | 5055 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
5056 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
5057 | goto error; | |
5058 | break; | |
5059 | case BPF_FUNC_get_stackid: | |
5060 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
5061 | goto error; | |
5062 | break; | |
60d20f91 | 5063 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 5064 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
5065 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
5066 | goto error; | |
5067 | break; | |
97f91a7c | 5068 | case BPF_FUNC_redirect_map: |
9c270af3 | 5069 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 5070 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
5071 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
5072 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
5073 | goto error; |
5074 | break; | |
174a79ff | 5075 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 5076 | case BPF_FUNC_msg_redirect_map: |
81110384 | 5077 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
5078 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
5079 | goto error; | |
5080 | break; | |
81110384 JF |
5081 | case BPF_FUNC_sk_redirect_hash: |
5082 | case BPF_FUNC_msg_redirect_hash: | |
5083 | case BPF_FUNC_sock_hash_update: | |
5084 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
5085 | goto error; |
5086 | break; | |
cd339431 | 5087 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
5088 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
5089 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
5090 | goto error; |
5091 | break; | |
2dbb9b9e | 5092 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
5093 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
5094 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
5095 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
5096 | goto error; |
5097 | break; | |
f1a2e44a MV |
5098 | case BPF_FUNC_map_peek_elem: |
5099 | case BPF_FUNC_map_pop_elem: | |
5100 | case BPF_FUNC_map_push_elem: | |
5101 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
5102 | map->map_type != BPF_MAP_TYPE_STACK) | |
5103 | goto error; | |
5104 | break; | |
6ac99e8f MKL |
5105 | case BPF_FUNC_sk_storage_get: |
5106 | case BPF_FUNC_sk_storage_delete: | |
5107 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
5108 | goto error; | |
5109 | break; | |
8ea63684 KS |
5110 | case BPF_FUNC_inode_storage_get: |
5111 | case BPF_FUNC_inode_storage_delete: | |
5112 | if (map->map_type != BPF_MAP_TYPE_INODE_STORAGE) | |
5113 | goto error; | |
5114 | break; | |
4cf1bc1f KS |
5115 | case BPF_FUNC_task_storage_get: |
5116 | case BPF_FUNC_task_storage_delete: | |
5117 | if (map->map_type != BPF_MAP_TYPE_TASK_STORAGE) | |
5118 | goto error; | |
5119 | break; | |
6aff67c8 AS |
5120 | default: |
5121 | break; | |
35578d79 KX |
5122 | } |
5123 | ||
5124 | return 0; | |
6aff67c8 | 5125 | error: |
61bd5218 | 5126 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 5127 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 5128 | return -EINVAL; |
35578d79 KX |
5129 | } |
5130 | ||
90133415 | 5131 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
5132 | { |
5133 | int count = 0; | |
5134 | ||
39f19ebb | 5135 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5136 | count++; |
39f19ebb | 5137 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5138 | count++; |
39f19ebb | 5139 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5140 | count++; |
39f19ebb | 5141 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 5142 | count++; |
39f19ebb | 5143 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
5144 | count++; |
5145 | ||
90133415 DB |
5146 | /* We only support one arg being in raw mode at the moment, |
5147 | * which is sufficient for the helper functions we have | |
5148 | * right now. | |
5149 | */ | |
5150 | return count <= 1; | |
5151 | } | |
5152 | ||
5153 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
5154 | enum bpf_arg_type arg_next) | |
5155 | { | |
5156 | return (arg_type_is_mem_ptr(arg_curr) && | |
5157 | !arg_type_is_mem_size(arg_next)) || | |
5158 | (!arg_type_is_mem_ptr(arg_curr) && | |
5159 | arg_type_is_mem_size(arg_next)); | |
5160 | } | |
5161 | ||
5162 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
5163 | { | |
5164 | /* bpf_xxx(..., buf, len) call will access 'len' | |
5165 | * bytes from memory 'buf'. Both arg types need | |
5166 | * to be paired, so make sure there's no buggy | |
5167 | * helper function specification. | |
5168 | */ | |
5169 | if (arg_type_is_mem_size(fn->arg1_type) || | |
5170 | arg_type_is_mem_ptr(fn->arg5_type) || | |
5171 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
5172 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
5173 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
5174 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
5175 | return false; | |
5176 | ||
5177 | return true; | |
5178 | } | |
5179 | ||
1b986589 | 5180 | static bool check_refcount_ok(const struct bpf_func_proto *fn, int func_id) |
fd978bf7 JS |
5181 | { |
5182 | int count = 0; | |
5183 | ||
1b986589 | 5184 | if (arg_type_may_be_refcounted(fn->arg1_type)) |
fd978bf7 | 5185 | count++; |
1b986589 | 5186 | if (arg_type_may_be_refcounted(fn->arg2_type)) |
fd978bf7 | 5187 | count++; |
1b986589 | 5188 | if (arg_type_may_be_refcounted(fn->arg3_type)) |
fd978bf7 | 5189 | count++; |
1b986589 | 5190 | if (arg_type_may_be_refcounted(fn->arg4_type)) |
fd978bf7 | 5191 | count++; |
1b986589 | 5192 | if (arg_type_may_be_refcounted(fn->arg5_type)) |
fd978bf7 JS |
5193 | count++; |
5194 | ||
1b986589 MKL |
5195 | /* A reference acquiring function cannot acquire |
5196 | * another refcounted ptr. | |
5197 | */ | |
64d85290 | 5198 | if (may_be_acquire_function(func_id) && count) |
1b986589 MKL |
5199 | return false; |
5200 | ||
fd978bf7 JS |
5201 | /* We only support one arg being unreferenced at the moment, |
5202 | * which is sufficient for the helper functions we have right now. | |
5203 | */ | |
5204 | return count <= 1; | |
5205 | } | |
5206 | ||
9436ef6e LB |
5207 | static bool check_btf_id_ok(const struct bpf_func_proto *fn) |
5208 | { | |
5209 | int i; | |
5210 | ||
1df8f55a | 5211 | for (i = 0; i < ARRAY_SIZE(fn->arg_type); i++) { |
9436ef6e LB |
5212 | if (fn->arg_type[i] == ARG_PTR_TO_BTF_ID && !fn->arg_btf_id[i]) |
5213 | return false; | |
5214 | ||
1df8f55a MKL |
5215 | if (fn->arg_type[i] != ARG_PTR_TO_BTF_ID && fn->arg_btf_id[i]) |
5216 | return false; | |
5217 | } | |
5218 | ||
9436ef6e LB |
5219 | return true; |
5220 | } | |
5221 | ||
1b986589 | 5222 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
5223 | { |
5224 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 5225 | check_arg_pair_ok(fn) && |
9436ef6e | 5226 | check_btf_id_ok(fn) && |
1b986589 | 5227 | check_refcount_ok(fn, func_id) ? 0 : -EINVAL; |
435faee1 DB |
5228 | } |
5229 | ||
de8f3a83 DB |
5230 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
5231 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 5232 | */ |
f4d7e40a AS |
5233 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
5234 | struct bpf_func_state *state) | |
969bf05e | 5235 | { |
58e2af8b | 5236 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
5237 | int i; |
5238 | ||
5239 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 5240 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 5241 | mark_reg_unknown(env, regs, i); |
969bf05e | 5242 | |
f3709f69 JS |
5243 | bpf_for_each_spilled_reg(i, state, reg) { |
5244 | if (!reg) | |
969bf05e | 5245 | continue; |
de8f3a83 | 5246 | if (reg_is_pkt_pointer_any(reg)) |
f54c7898 | 5247 | __mark_reg_unknown(env, reg); |
969bf05e AS |
5248 | } |
5249 | } | |
5250 | ||
f4d7e40a AS |
5251 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
5252 | { | |
5253 | struct bpf_verifier_state *vstate = env->cur_state; | |
5254 | int i; | |
5255 | ||
5256 | for (i = 0; i <= vstate->curframe; i++) | |
5257 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
5258 | } | |
5259 | ||
6d94e741 AS |
5260 | enum { |
5261 | AT_PKT_END = -1, | |
5262 | BEYOND_PKT_END = -2, | |
5263 | }; | |
5264 | ||
5265 | static void mark_pkt_end(struct bpf_verifier_state *vstate, int regn, bool range_open) | |
5266 | { | |
5267 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5268 | struct bpf_reg_state *reg = &state->regs[regn]; | |
5269 | ||
5270 | if (reg->type != PTR_TO_PACKET) | |
5271 | /* PTR_TO_PACKET_META is not supported yet */ | |
5272 | return; | |
5273 | ||
5274 | /* The 'reg' is pkt > pkt_end or pkt >= pkt_end. | |
5275 | * How far beyond pkt_end it goes is unknown. | |
5276 | * if (!range_open) it's the case of pkt >= pkt_end | |
5277 | * if (range_open) it's the case of pkt > pkt_end | |
5278 | * hence this pointer is at least 1 byte bigger than pkt_end | |
5279 | */ | |
5280 | if (range_open) | |
5281 | reg->range = BEYOND_PKT_END; | |
5282 | else | |
5283 | reg->range = AT_PKT_END; | |
5284 | } | |
5285 | ||
fd978bf7 | 5286 | static void release_reg_references(struct bpf_verifier_env *env, |
1b986589 MKL |
5287 | struct bpf_func_state *state, |
5288 | int ref_obj_id) | |
fd978bf7 JS |
5289 | { |
5290 | struct bpf_reg_state *regs = state->regs, *reg; | |
5291 | int i; | |
5292 | ||
5293 | for (i = 0; i < MAX_BPF_REG; i++) | |
1b986589 | 5294 | if (regs[i].ref_obj_id == ref_obj_id) |
fd978bf7 JS |
5295 | mark_reg_unknown(env, regs, i); |
5296 | ||
5297 | bpf_for_each_spilled_reg(i, state, reg) { | |
5298 | if (!reg) | |
5299 | continue; | |
1b986589 | 5300 | if (reg->ref_obj_id == ref_obj_id) |
f54c7898 | 5301 | __mark_reg_unknown(env, reg); |
fd978bf7 JS |
5302 | } |
5303 | } | |
5304 | ||
5305 | /* The pointer with the specified id has released its reference to kernel | |
5306 | * resources. Identify all copies of the same pointer and clear the reference. | |
5307 | */ | |
5308 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 5309 | int ref_obj_id) |
fd978bf7 JS |
5310 | { |
5311 | struct bpf_verifier_state *vstate = env->cur_state; | |
1b986589 | 5312 | int err; |
fd978bf7 JS |
5313 | int i; |
5314 | ||
1b986589 MKL |
5315 | err = release_reference_state(cur_func(env), ref_obj_id); |
5316 | if (err) | |
5317 | return err; | |
5318 | ||
fd978bf7 | 5319 | for (i = 0; i <= vstate->curframe; i++) |
1b986589 | 5320 | release_reg_references(env, vstate->frame[i], ref_obj_id); |
fd978bf7 | 5321 | |
1b986589 | 5322 | return 0; |
fd978bf7 JS |
5323 | } |
5324 | ||
51c39bb1 AS |
5325 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
5326 | struct bpf_reg_state *regs) | |
5327 | { | |
5328 | int i; | |
5329 | ||
5330 | /* after the call registers r0 - r5 were scratched */ | |
5331 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
5332 | mark_reg_not_init(env, regs, caller_saved[i]); | |
5333 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
5334 | } | |
5335 | } | |
5336 | ||
14351375 YS |
5337 | typedef int (*set_callee_state_fn)(struct bpf_verifier_env *env, |
5338 | struct bpf_func_state *caller, | |
5339 | struct bpf_func_state *callee, | |
5340 | int insn_idx); | |
5341 | ||
5342 | static int __check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
5343 | int *insn_idx, int subprog, | |
5344 | set_callee_state_fn set_callee_state_cb) | |
f4d7e40a AS |
5345 | { |
5346 | struct bpf_verifier_state *state = env->cur_state; | |
51c39bb1 | 5347 | struct bpf_func_info_aux *func_info_aux; |
f4d7e40a | 5348 | struct bpf_func_state *caller, *callee; |
14351375 | 5349 | int err; |
51c39bb1 | 5350 | bool is_global = false; |
f4d7e40a | 5351 | |
aada9ce6 | 5352 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 5353 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 5354 | state->curframe + 2); |
f4d7e40a AS |
5355 | return -E2BIG; |
5356 | } | |
5357 | ||
f4d7e40a AS |
5358 | caller = state->frame[state->curframe]; |
5359 | if (state->frame[state->curframe + 1]) { | |
5360 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
5361 | state->curframe + 1); | |
5362 | return -EFAULT; | |
5363 | } | |
5364 | ||
51c39bb1 AS |
5365 | func_info_aux = env->prog->aux->func_info_aux; |
5366 | if (func_info_aux) | |
5367 | is_global = func_info_aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
5368 | err = btf_check_func_arg_match(env, subprog, caller->regs); | |
5369 | if (err == -EFAULT) | |
5370 | return err; | |
5371 | if (is_global) { | |
5372 | if (err) { | |
5373 | verbose(env, "Caller passes invalid args into func#%d\n", | |
5374 | subprog); | |
5375 | return err; | |
5376 | } else { | |
5377 | if (env->log.level & BPF_LOG_LEVEL) | |
5378 | verbose(env, | |
5379 | "Func#%d is global and valid. Skipping.\n", | |
5380 | subprog); | |
5381 | clear_caller_saved_regs(env, caller->regs); | |
5382 | ||
45159b27 | 5383 | /* All global functions return a 64-bit SCALAR_VALUE */ |
51c39bb1 | 5384 | mark_reg_unknown(env, caller->regs, BPF_REG_0); |
45159b27 | 5385 | caller->regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; |
51c39bb1 AS |
5386 | |
5387 | /* continue with next insn after call */ | |
5388 | return 0; | |
5389 | } | |
5390 | } | |
5391 | ||
f4d7e40a AS |
5392 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); |
5393 | if (!callee) | |
5394 | return -ENOMEM; | |
5395 | state->frame[state->curframe + 1] = callee; | |
5396 | ||
5397 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
5398 | * into its own stack before reading from it. | |
5399 | * callee can read/write into caller's stack | |
5400 | */ | |
5401 | init_func_state(env, callee, | |
5402 | /* remember the callsite, it will be used by bpf_exit */ | |
5403 | *insn_idx /* callsite */, | |
5404 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 5405 | subprog /* subprog number within this prog */); |
f4d7e40a | 5406 | |
fd978bf7 JS |
5407 | /* Transfer references to the callee */ |
5408 | err = transfer_reference_state(callee, caller); | |
5409 | if (err) | |
5410 | return err; | |
5411 | ||
14351375 YS |
5412 | err = set_callee_state_cb(env, caller, callee, *insn_idx); |
5413 | if (err) | |
5414 | return err; | |
f4d7e40a | 5415 | |
51c39bb1 | 5416 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a AS |
5417 | |
5418 | /* only increment it after check_reg_arg() finished */ | |
5419 | state->curframe++; | |
5420 | ||
5421 | /* and go analyze first insn of the callee */ | |
14351375 | 5422 | *insn_idx = env->subprog_info[subprog].start - 1; |
f4d7e40a | 5423 | |
06ee7115 | 5424 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
5425 | verbose(env, "caller:\n"); |
5426 | print_verifier_state(env, caller); | |
5427 | verbose(env, "callee:\n"); | |
5428 | print_verifier_state(env, callee); | |
5429 | } | |
5430 | return 0; | |
5431 | } | |
5432 | ||
314ee05e YS |
5433 | int map_set_for_each_callback_args(struct bpf_verifier_env *env, |
5434 | struct bpf_func_state *caller, | |
5435 | struct bpf_func_state *callee) | |
5436 | { | |
5437 | /* bpf_for_each_map_elem(struct bpf_map *map, void *callback_fn, | |
5438 | * void *callback_ctx, u64 flags); | |
5439 | * callback_fn(struct bpf_map *map, void *key, void *value, | |
5440 | * void *callback_ctx); | |
5441 | */ | |
5442 | callee->regs[BPF_REG_1] = caller->regs[BPF_REG_1]; | |
5443 | ||
5444 | callee->regs[BPF_REG_2].type = PTR_TO_MAP_KEY; | |
5445 | __mark_reg_known_zero(&callee->regs[BPF_REG_2]); | |
5446 | callee->regs[BPF_REG_2].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
5447 | ||
5448 | callee->regs[BPF_REG_3].type = PTR_TO_MAP_VALUE; | |
5449 | __mark_reg_known_zero(&callee->regs[BPF_REG_3]); | |
5450 | callee->regs[BPF_REG_3].map_ptr = caller->regs[BPF_REG_1].map_ptr; | |
5451 | ||
5452 | /* pointer to stack or null */ | |
5453 | callee->regs[BPF_REG_4] = caller->regs[BPF_REG_3]; | |
5454 | ||
5455 | /* unused */ | |
5456 | __mark_reg_not_init(env, &callee->regs[BPF_REG_5]); | |
5457 | return 0; | |
5458 | } | |
5459 | ||
14351375 YS |
5460 | static int set_callee_state(struct bpf_verifier_env *env, |
5461 | struct bpf_func_state *caller, | |
5462 | struct bpf_func_state *callee, int insn_idx) | |
5463 | { | |
5464 | int i; | |
5465 | ||
5466 | /* copy r1 - r5 args that callee can access. The copy includes parent | |
5467 | * pointers, which connects us up to the liveness chain | |
5468 | */ | |
5469 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
5470 | callee->regs[i] = caller->regs[i]; | |
5471 | return 0; | |
5472 | } | |
5473 | ||
5474 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
5475 | int *insn_idx) | |
5476 | { | |
5477 | int subprog, target_insn; | |
5478 | ||
5479 | target_insn = *insn_idx + insn->imm + 1; | |
5480 | subprog = find_subprog(env, target_insn); | |
5481 | if (subprog < 0) { | |
5482 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
5483 | target_insn); | |
5484 | return -EFAULT; | |
5485 | } | |
5486 | ||
5487 | return __check_func_call(env, insn, insn_idx, subprog, set_callee_state); | |
5488 | } | |
5489 | ||
69c087ba YS |
5490 | static int set_map_elem_callback_state(struct bpf_verifier_env *env, |
5491 | struct bpf_func_state *caller, | |
5492 | struct bpf_func_state *callee, | |
5493 | int insn_idx) | |
5494 | { | |
5495 | struct bpf_insn_aux_data *insn_aux = &env->insn_aux_data[insn_idx]; | |
5496 | struct bpf_map *map; | |
5497 | int err; | |
5498 | ||
5499 | if (bpf_map_ptr_poisoned(insn_aux)) { | |
5500 | verbose(env, "tail_call abusing map_ptr\n"); | |
5501 | return -EINVAL; | |
5502 | } | |
5503 | ||
5504 | map = BPF_MAP_PTR(insn_aux->map_ptr_state); | |
5505 | if (!map->ops->map_set_for_each_callback_args || | |
5506 | !map->ops->map_for_each_callback) { | |
5507 | verbose(env, "callback function not allowed for map\n"); | |
5508 | return -ENOTSUPP; | |
5509 | } | |
5510 | ||
5511 | err = map->ops->map_set_for_each_callback_args(env, caller, callee); | |
5512 | if (err) | |
5513 | return err; | |
5514 | ||
5515 | callee->in_callback_fn = true; | |
5516 | return 0; | |
5517 | } | |
5518 | ||
f4d7e40a AS |
5519 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) |
5520 | { | |
5521 | struct bpf_verifier_state *state = env->cur_state; | |
5522 | struct bpf_func_state *caller, *callee; | |
5523 | struct bpf_reg_state *r0; | |
fd978bf7 | 5524 | int err; |
f4d7e40a AS |
5525 | |
5526 | callee = state->frame[state->curframe]; | |
5527 | r0 = &callee->regs[BPF_REG_0]; | |
5528 | if (r0->type == PTR_TO_STACK) { | |
5529 | /* technically it's ok to return caller's stack pointer | |
5530 | * (or caller's caller's pointer) back to the caller, | |
5531 | * since these pointers are valid. Only current stack | |
5532 | * pointer will be invalid as soon as function exits, | |
5533 | * but let's be conservative | |
5534 | */ | |
5535 | verbose(env, "cannot return stack pointer to the caller\n"); | |
5536 | return -EINVAL; | |
5537 | } | |
5538 | ||
5539 | state->curframe--; | |
5540 | caller = state->frame[state->curframe]; | |
69c087ba YS |
5541 | if (callee->in_callback_fn) { |
5542 | /* enforce R0 return value range [0, 1]. */ | |
5543 | struct tnum range = tnum_range(0, 1); | |
5544 | ||
5545 | if (r0->type != SCALAR_VALUE) { | |
5546 | verbose(env, "R0 not a scalar value\n"); | |
5547 | return -EACCES; | |
5548 | } | |
5549 | if (!tnum_in(range, r0->var_off)) { | |
5550 | verbose_invalid_scalar(env, r0, &range, "callback return", "R0"); | |
5551 | return -EINVAL; | |
5552 | } | |
5553 | } else { | |
5554 | /* return to the caller whatever r0 had in the callee */ | |
5555 | caller->regs[BPF_REG_0] = *r0; | |
5556 | } | |
f4d7e40a | 5557 | |
fd978bf7 JS |
5558 | /* Transfer references to the caller */ |
5559 | err = transfer_reference_state(caller, callee); | |
5560 | if (err) | |
5561 | return err; | |
5562 | ||
f4d7e40a | 5563 | *insn_idx = callee->callsite + 1; |
06ee7115 | 5564 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
5565 | verbose(env, "returning from callee:\n"); |
5566 | print_verifier_state(env, callee); | |
5567 | verbose(env, "to caller at %d:\n", *insn_idx); | |
5568 | print_verifier_state(env, caller); | |
5569 | } | |
5570 | /* clear everything in the callee */ | |
5571 | free_func_state(callee); | |
5572 | state->frame[state->curframe + 1] = NULL; | |
5573 | return 0; | |
5574 | } | |
5575 | ||
849fa506 YS |
5576 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
5577 | int func_id, | |
5578 | struct bpf_call_arg_meta *meta) | |
5579 | { | |
5580 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
5581 | ||
5582 | if (ret_type != RET_INTEGER || | |
5583 | (func_id != BPF_FUNC_get_stack && | |
47cc0ed5 DB |
5584 | func_id != BPF_FUNC_probe_read_str && |
5585 | func_id != BPF_FUNC_probe_read_kernel_str && | |
5586 | func_id != BPF_FUNC_probe_read_user_str)) | |
849fa506 YS |
5587 | return; |
5588 | ||
10060503 | 5589 | ret_reg->smax_value = meta->msize_max_value; |
fa123ac0 | 5590 | ret_reg->s32_max_value = meta->msize_max_value; |
b0270958 AS |
5591 | ret_reg->smin_value = -MAX_ERRNO; |
5592 | ret_reg->s32_min_value = -MAX_ERRNO; | |
849fa506 YS |
5593 | __reg_deduce_bounds(ret_reg); |
5594 | __reg_bound_offset(ret_reg); | |
10060503 | 5595 | __update_reg_bounds(ret_reg); |
849fa506 YS |
5596 | } |
5597 | ||
c93552c4 DB |
5598 | static int |
5599 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
5600 | int func_id, int insn_idx) | |
5601 | { | |
5602 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 5603 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
5604 | |
5605 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
5606 | func_id != BPF_FUNC_map_lookup_elem && |
5607 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
5608 | func_id != BPF_FUNC_map_delete_elem && |
5609 | func_id != BPF_FUNC_map_push_elem && | |
5610 | func_id != BPF_FUNC_map_pop_elem && | |
69c087ba | 5611 | func_id != BPF_FUNC_map_peek_elem && |
e6a4750f BT |
5612 | func_id != BPF_FUNC_for_each_map_elem && |
5613 | func_id != BPF_FUNC_redirect_map) | |
c93552c4 | 5614 | return 0; |
09772d92 | 5615 | |
591fe988 | 5616 | if (map == NULL) { |
c93552c4 DB |
5617 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
5618 | return -EINVAL; | |
5619 | } | |
5620 | ||
591fe988 DB |
5621 | /* In case of read-only, some additional restrictions |
5622 | * need to be applied in order to prevent altering the | |
5623 | * state of the map from program side. | |
5624 | */ | |
5625 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
5626 | (func_id == BPF_FUNC_map_delete_elem || | |
5627 | func_id == BPF_FUNC_map_update_elem || | |
5628 | func_id == BPF_FUNC_map_push_elem || | |
5629 | func_id == BPF_FUNC_map_pop_elem)) { | |
5630 | verbose(env, "write into map forbidden\n"); | |
5631 | return -EACCES; | |
5632 | } | |
5633 | ||
d2e4c1e6 | 5634 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 5635 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 5636 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 5637 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 5638 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 5639 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
5640 | return 0; |
5641 | } | |
5642 | ||
d2e4c1e6 DB |
5643 | static int |
5644 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
5645 | int func_id, int insn_idx) | |
5646 | { | |
5647 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
5648 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
5649 | struct bpf_map *map = meta->map_ptr; | |
5650 | struct tnum range; | |
5651 | u64 val; | |
cc52d914 | 5652 | int err; |
d2e4c1e6 DB |
5653 | |
5654 | if (func_id != BPF_FUNC_tail_call) | |
5655 | return 0; | |
5656 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
5657 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
5658 | return -EINVAL; | |
5659 | } | |
5660 | ||
5661 | range = tnum_range(0, map->max_entries - 1); | |
5662 | reg = ®s[BPF_REG_3]; | |
5663 | ||
5664 | if (!register_is_const(reg) || !tnum_in(range, reg->var_off)) { | |
5665 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
5666 | return 0; | |
5667 | } | |
5668 | ||
cc52d914 DB |
5669 | err = mark_chain_precision(env, BPF_REG_3); |
5670 | if (err) | |
5671 | return err; | |
5672 | ||
d2e4c1e6 DB |
5673 | val = reg->var_off.value; |
5674 | if (bpf_map_key_unseen(aux)) | |
5675 | bpf_map_key_store(aux, val); | |
5676 | else if (!bpf_map_key_poisoned(aux) && | |
5677 | bpf_map_key_immediate(aux) != val) | |
5678 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
5679 | return 0; | |
5680 | } | |
5681 | ||
fd978bf7 JS |
5682 | static int check_reference_leak(struct bpf_verifier_env *env) |
5683 | { | |
5684 | struct bpf_func_state *state = cur_func(env); | |
5685 | int i; | |
5686 | ||
5687 | for (i = 0; i < state->acquired_refs; i++) { | |
5688 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", | |
5689 | state->refs[i].id, state->refs[i].insn_idx); | |
5690 | } | |
5691 | return state->acquired_refs ? -EINVAL : 0; | |
5692 | } | |
5693 | ||
69c087ba YS |
5694 | static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
5695 | int *insn_idx_p) | |
17a52670 | 5696 | { |
17a52670 | 5697 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 5698 | struct bpf_reg_state *regs; |
33ff9823 | 5699 | struct bpf_call_arg_meta meta; |
69c087ba | 5700 | int insn_idx = *insn_idx_p; |
969bf05e | 5701 | bool changes_data; |
69c087ba | 5702 | int i, err, func_id; |
17a52670 AS |
5703 | |
5704 | /* find function prototype */ | |
69c087ba | 5705 | func_id = insn->imm; |
17a52670 | 5706 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { |
61bd5218 JK |
5707 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
5708 | func_id); | |
17a52670 AS |
5709 | return -EINVAL; |
5710 | } | |
5711 | ||
00176a34 | 5712 | if (env->ops->get_func_proto) |
5e43f899 | 5713 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 5714 | if (!fn) { |
61bd5218 JK |
5715 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
5716 | func_id); | |
17a52670 AS |
5717 | return -EINVAL; |
5718 | } | |
5719 | ||
5720 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 5721 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 5722 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
5723 | return -EINVAL; |
5724 | } | |
5725 | ||
eae2e83e JO |
5726 | if (fn->allowed && !fn->allowed(env->prog)) { |
5727 | verbose(env, "helper call is not allowed in probe\n"); | |
5728 | return -EINVAL; | |
5729 | } | |
5730 | ||
04514d13 | 5731 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 5732 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
5733 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
5734 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
5735 | func_id_name(func_id), func_id); | |
5736 | return -EINVAL; | |
5737 | } | |
969bf05e | 5738 | |
33ff9823 | 5739 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 5740 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 5741 | |
1b986589 | 5742 | err = check_func_proto(fn, func_id); |
435faee1 | 5743 | if (err) { |
61bd5218 | 5744 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 5745 | func_id_name(func_id), func_id); |
435faee1 DB |
5746 | return err; |
5747 | } | |
5748 | ||
d83525ca | 5749 | meta.func_id = func_id; |
17a52670 | 5750 | /* check args */ |
523a4cf4 | 5751 | for (i = 0; i < MAX_BPF_FUNC_REG_ARGS; i++) { |
af7ec138 | 5752 | err = check_func_arg(env, i, &meta, fn); |
a7658e1a AS |
5753 | if (err) |
5754 | return err; | |
5755 | } | |
17a52670 | 5756 | |
c93552c4 DB |
5757 | err = record_func_map(env, &meta, func_id, insn_idx); |
5758 | if (err) | |
5759 | return err; | |
5760 | ||
d2e4c1e6 DB |
5761 | err = record_func_key(env, &meta, func_id, insn_idx); |
5762 | if (err) | |
5763 | return err; | |
5764 | ||
435faee1 DB |
5765 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
5766 | * is inferred from register state. | |
5767 | */ | |
5768 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
5769 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
5770 | BPF_WRITE, -1, false); | |
435faee1 DB |
5771 | if (err) |
5772 | return err; | |
5773 | } | |
5774 | ||
fd978bf7 JS |
5775 | if (func_id == BPF_FUNC_tail_call) { |
5776 | err = check_reference_leak(env); | |
5777 | if (err) { | |
5778 | verbose(env, "tail_call would lead to reference leak\n"); | |
5779 | return err; | |
5780 | } | |
5781 | } else if (is_release_function(func_id)) { | |
1b986589 | 5782 | err = release_reference(env, meta.ref_obj_id); |
46f8bc92 MKL |
5783 | if (err) { |
5784 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
5785 | func_id_name(func_id), func_id); | |
fd978bf7 | 5786 | return err; |
46f8bc92 | 5787 | } |
fd978bf7 JS |
5788 | } |
5789 | ||
638f5b90 | 5790 | regs = cur_regs(env); |
cd339431 RG |
5791 | |
5792 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
5793 | * this is required because get_local_storage() can't return an error. | |
5794 | */ | |
5795 | if (func_id == BPF_FUNC_get_local_storage && | |
5796 | !register_is_null(®s[BPF_REG_2])) { | |
5797 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
5798 | return -EINVAL; | |
5799 | } | |
5800 | ||
69c087ba YS |
5801 | if (func_id == BPF_FUNC_for_each_map_elem) { |
5802 | err = __check_func_call(env, insn, insn_idx_p, meta.subprogno, | |
5803 | set_map_elem_callback_state); | |
5804 | if (err < 0) | |
5805 | return -EINVAL; | |
5806 | } | |
5807 | ||
17a52670 | 5808 | /* reset caller saved regs */ |
dc503a8a | 5809 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 5810 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
5811 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
5812 | } | |
17a52670 | 5813 | |
5327ed3d JW |
5814 | /* helper call returns 64-bit value. */ |
5815 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
5816 | ||
dc503a8a | 5817 | /* update return register (already marked as written above) */ |
17a52670 | 5818 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 5819 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 5820 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
5821 | } else if (fn->ret_type == RET_VOID) { |
5822 | regs[BPF_REG_0].type = NOT_INIT; | |
3e6a4b3e RG |
5823 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL || |
5824 | fn->ret_type == RET_PTR_TO_MAP_VALUE) { | |
f1174f77 | 5825 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 5826 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
5827 | /* remember map_ptr, so that check_map_access() |
5828 | * can check 'value_size' boundary of memory access | |
5829 | * to map element returned from bpf_map_lookup_elem() | |
5830 | */ | |
33ff9823 | 5831 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
5832 | verbose(env, |
5833 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
5834 | return -EINVAL; |
5835 | } | |
33ff9823 | 5836 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
4d31f301 DB |
5837 | if (fn->ret_type == RET_PTR_TO_MAP_VALUE) { |
5838 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE; | |
e16d2f1a AS |
5839 | if (map_value_has_spin_lock(meta.map_ptr)) |
5840 | regs[BPF_REG_0].id = ++env->id_gen; | |
4d31f301 DB |
5841 | } else { |
5842 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
4d31f301 | 5843 | } |
c64b7983 JS |
5844 | } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) { |
5845 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
5846 | regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL; | |
85a51f8c LB |
5847 | } else if (fn->ret_type == RET_PTR_TO_SOCK_COMMON_OR_NULL) { |
5848 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
5849 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON_OR_NULL; | |
655a51e5 MKL |
5850 | } else if (fn->ret_type == RET_PTR_TO_TCP_SOCK_OR_NULL) { |
5851 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
5852 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK_OR_NULL; | |
457f4436 AN |
5853 | } else if (fn->ret_type == RET_PTR_TO_ALLOC_MEM_OR_NULL) { |
5854 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
5855 | regs[BPF_REG_0].type = PTR_TO_MEM_OR_NULL; | |
457f4436 | 5856 | regs[BPF_REG_0].mem_size = meta.mem_size; |
63d9b80d HL |
5857 | } else if (fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID_OR_NULL || |
5858 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID) { | |
eaa6bcb7 HL |
5859 | const struct btf_type *t; |
5860 | ||
5861 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
22dc4a0f | 5862 | t = btf_type_skip_modifiers(meta.ret_btf, meta.ret_btf_id, NULL); |
eaa6bcb7 HL |
5863 | if (!btf_type_is_struct(t)) { |
5864 | u32 tsize; | |
5865 | const struct btf_type *ret; | |
5866 | const char *tname; | |
5867 | ||
5868 | /* resolve the type size of ksym. */ | |
22dc4a0f | 5869 | ret = btf_resolve_size(meta.ret_btf, t, &tsize); |
eaa6bcb7 | 5870 | if (IS_ERR(ret)) { |
22dc4a0f | 5871 | tname = btf_name_by_offset(meta.ret_btf, t->name_off); |
eaa6bcb7 HL |
5872 | verbose(env, "unable to resolve the size of type '%s': %ld\n", |
5873 | tname, PTR_ERR(ret)); | |
5874 | return -EINVAL; | |
5875 | } | |
63d9b80d HL |
5876 | regs[BPF_REG_0].type = |
5877 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ? | |
5878 | PTR_TO_MEM : PTR_TO_MEM_OR_NULL; | |
eaa6bcb7 HL |
5879 | regs[BPF_REG_0].mem_size = tsize; |
5880 | } else { | |
63d9b80d HL |
5881 | regs[BPF_REG_0].type = |
5882 | fn->ret_type == RET_PTR_TO_MEM_OR_BTF_ID ? | |
5883 | PTR_TO_BTF_ID : PTR_TO_BTF_ID_OR_NULL; | |
22dc4a0f | 5884 | regs[BPF_REG_0].btf = meta.ret_btf; |
eaa6bcb7 HL |
5885 | regs[BPF_REG_0].btf_id = meta.ret_btf_id; |
5886 | } | |
3ca1032a KS |
5887 | } else if (fn->ret_type == RET_PTR_TO_BTF_ID_OR_NULL || |
5888 | fn->ret_type == RET_PTR_TO_BTF_ID) { | |
af7ec138 YS |
5889 | int ret_btf_id; |
5890 | ||
5891 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
3ca1032a KS |
5892 | regs[BPF_REG_0].type = fn->ret_type == RET_PTR_TO_BTF_ID ? |
5893 | PTR_TO_BTF_ID : | |
5894 | PTR_TO_BTF_ID_OR_NULL; | |
af7ec138 YS |
5895 | ret_btf_id = *fn->ret_btf_id; |
5896 | if (ret_btf_id == 0) { | |
5897 | verbose(env, "invalid return type %d of func %s#%d\n", | |
5898 | fn->ret_type, func_id_name(func_id), func_id); | |
5899 | return -EINVAL; | |
5900 | } | |
22dc4a0f AN |
5901 | /* current BPF helper definitions are only coming from |
5902 | * built-in code with type IDs from vmlinux BTF | |
5903 | */ | |
5904 | regs[BPF_REG_0].btf = btf_vmlinux; | |
af7ec138 | 5905 | regs[BPF_REG_0].btf_id = ret_btf_id; |
17a52670 | 5906 | } else { |
61bd5218 | 5907 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 5908 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
5909 | return -EINVAL; |
5910 | } | |
04fd61ab | 5911 | |
93c230e3 MKL |
5912 | if (reg_type_may_be_null(regs[BPF_REG_0].type)) |
5913 | regs[BPF_REG_0].id = ++env->id_gen; | |
5914 | ||
0f3adc28 | 5915 | if (is_ptr_cast_function(func_id)) { |
1b986589 MKL |
5916 | /* For release_reference() */ |
5917 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 5918 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
5919 | int id = acquire_reference_state(env, insn_idx); |
5920 | ||
5921 | if (id < 0) | |
5922 | return id; | |
5923 | /* For mark_ptr_or_null_reg() */ | |
5924 | regs[BPF_REG_0].id = id; | |
5925 | /* For release_reference() */ | |
5926 | regs[BPF_REG_0].ref_obj_id = id; | |
5927 | } | |
1b986589 | 5928 | |
849fa506 YS |
5929 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
5930 | ||
61bd5218 | 5931 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
5932 | if (err) |
5933 | return err; | |
04fd61ab | 5934 | |
fa28dcb8 SL |
5935 | if ((func_id == BPF_FUNC_get_stack || |
5936 | func_id == BPF_FUNC_get_task_stack) && | |
5937 | !env->prog->has_callchain_buf) { | |
c195651e YS |
5938 | const char *err_str; |
5939 | ||
5940 | #ifdef CONFIG_PERF_EVENTS | |
5941 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
5942 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
5943 | #else | |
5944 | err = -ENOTSUPP; | |
5945 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
5946 | #endif | |
5947 | if (err) { | |
5948 | verbose(env, err_str, func_id_name(func_id), func_id); | |
5949 | return err; | |
5950 | } | |
5951 | ||
5952 | env->prog->has_callchain_buf = true; | |
5953 | } | |
5954 | ||
5d99cb2c SL |
5955 | if (func_id == BPF_FUNC_get_stackid || func_id == BPF_FUNC_get_stack) |
5956 | env->prog->call_get_stack = true; | |
5957 | ||
969bf05e AS |
5958 | if (changes_data) |
5959 | clear_all_pkt_pointers(env); | |
5960 | return 0; | |
5961 | } | |
5962 | ||
b03c9f9f EC |
5963 | static bool signed_add_overflows(s64 a, s64 b) |
5964 | { | |
5965 | /* Do the add in u64, where overflow is well-defined */ | |
5966 | s64 res = (s64)((u64)a + (u64)b); | |
5967 | ||
5968 | if (b < 0) | |
5969 | return res > a; | |
5970 | return res < a; | |
5971 | } | |
5972 | ||
bc895e8b | 5973 | static bool signed_add32_overflows(s32 a, s32 b) |
3f50f132 JF |
5974 | { |
5975 | /* Do the add in u32, where overflow is well-defined */ | |
5976 | s32 res = (s32)((u32)a + (u32)b); | |
5977 | ||
5978 | if (b < 0) | |
5979 | return res > a; | |
5980 | return res < a; | |
5981 | } | |
5982 | ||
bc895e8b | 5983 | static bool signed_sub_overflows(s64 a, s64 b) |
b03c9f9f EC |
5984 | { |
5985 | /* Do the sub in u64, where overflow is well-defined */ | |
5986 | s64 res = (s64)((u64)a - (u64)b); | |
5987 | ||
5988 | if (b < 0) | |
5989 | return res < a; | |
5990 | return res > a; | |
969bf05e AS |
5991 | } |
5992 | ||
3f50f132 JF |
5993 | static bool signed_sub32_overflows(s32 a, s32 b) |
5994 | { | |
bc895e8b | 5995 | /* Do the sub in u32, where overflow is well-defined */ |
3f50f132 JF |
5996 | s32 res = (s32)((u32)a - (u32)b); |
5997 | ||
5998 | if (b < 0) | |
5999 | return res < a; | |
6000 | return res > a; | |
6001 | } | |
6002 | ||
bb7f0f98 AS |
6003 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
6004 | const struct bpf_reg_state *reg, | |
6005 | enum bpf_reg_type type) | |
6006 | { | |
6007 | bool known = tnum_is_const(reg->var_off); | |
6008 | s64 val = reg->var_off.value; | |
6009 | s64 smin = reg->smin_value; | |
6010 | ||
6011 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
6012 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
6013 | reg_type_str[type], val); | |
6014 | return false; | |
6015 | } | |
6016 | ||
6017 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
6018 | verbose(env, "%s pointer offset %d is not allowed\n", | |
6019 | reg_type_str[type], reg->off); | |
6020 | return false; | |
6021 | } | |
6022 | ||
6023 | if (smin == S64_MIN) { | |
6024 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
6025 | reg_type_str[type]); | |
6026 | return false; | |
6027 | } | |
6028 | ||
6029 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
6030 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
6031 | smin, reg_type_str[type]); | |
6032 | return false; | |
6033 | } | |
6034 | ||
6035 | return true; | |
6036 | } | |
6037 | ||
979d63d5 DB |
6038 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
6039 | { | |
6040 | return &env->insn_aux_data[env->insn_idx]; | |
6041 | } | |
6042 | ||
6043 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, | |
6044 | u32 *ptr_limit, u8 opcode, bool off_is_neg) | |
6045 | { | |
6046 | bool mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
6047 | (opcode == BPF_SUB && !off_is_neg); | |
1b1597e6 | 6048 | u32 off, max; |
979d63d5 DB |
6049 | |
6050 | switch (ptr_reg->type) { | |
6051 | case PTR_TO_STACK: | |
1b1597e6 PK |
6052 | /* Offset 0 is out-of-bounds, but acceptable start for the |
6053 | * left direction, see BPF_REG_FP. | |
6054 | */ | |
6055 | max = MAX_BPF_STACK + mask_to_left; | |
088ec26d AI |
6056 | /* Indirect variable offset stack access is prohibited in |
6057 | * unprivileged mode so it's not handled here. | |
6058 | */ | |
979d63d5 DB |
6059 | off = ptr_reg->off + ptr_reg->var_off.value; |
6060 | if (mask_to_left) | |
6061 | *ptr_limit = MAX_BPF_STACK + off; | |
6062 | else | |
b5871dca | 6063 | *ptr_limit = -off - 1; |
1b1597e6 | 6064 | return *ptr_limit >= max ? -ERANGE : 0; |
979d63d5 | 6065 | case PTR_TO_MAP_VALUE: |
1b1597e6 | 6066 | max = ptr_reg->map_ptr->value_size; |
979d63d5 DB |
6067 | if (mask_to_left) { |
6068 | *ptr_limit = ptr_reg->umax_value + ptr_reg->off; | |
6069 | } else { | |
6070 | off = ptr_reg->smin_value + ptr_reg->off; | |
b5871dca | 6071 | *ptr_limit = ptr_reg->map_ptr->value_size - off - 1; |
979d63d5 | 6072 | } |
1b1597e6 | 6073 | return *ptr_limit >= max ? -ERANGE : 0; |
979d63d5 DB |
6074 | default: |
6075 | return -EINVAL; | |
6076 | } | |
6077 | } | |
6078 | ||
d3bd7413 DB |
6079 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
6080 | const struct bpf_insn *insn) | |
6081 | { | |
2c78ee89 | 6082 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
6083 | } |
6084 | ||
6085 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
6086 | u32 alu_state, u32 alu_limit) | |
6087 | { | |
6088 | /* If we arrived here from different branches with different | |
6089 | * state or limits to sanitize, then this won't work. | |
6090 | */ | |
6091 | if (aux->alu_state && | |
6092 | (aux->alu_state != alu_state || | |
6093 | aux->alu_limit != alu_limit)) | |
6094 | return -EACCES; | |
6095 | ||
e6ac5933 | 6096 | /* Corresponding fixup done in do_misc_fixups(). */ |
d3bd7413 DB |
6097 | aux->alu_state = alu_state; |
6098 | aux->alu_limit = alu_limit; | |
6099 | return 0; | |
6100 | } | |
6101 | ||
6102 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
6103 | struct bpf_insn *insn) | |
6104 | { | |
6105 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
6106 | ||
6107 | if (can_skip_alu_sanitation(env, insn)) | |
6108 | return 0; | |
6109 | ||
6110 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
6111 | } | |
6112 | ||
979d63d5 DB |
6113 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
6114 | struct bpf_insn *insn, | |
6115 | const struct bpf_reg_state *ptr_reg, | |
6116 | struct bpf_reg_state *dst_reg, | |
6117 | bool off_is_neg) | |
6118 | { | |
6119 | struct bpf_verifier_state *vstate = env->cur_state; | |
6120 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
6121 | bool ptr_is_dst_reg = ptr_reg == dst_reg; | |
6122 | u8 opcode = BPF_OP(insn->code); | |
6123 | u32 alu_state, alu_limit; | |
6124 | struct bpf_reg_state tmp; | |
6125 | bool ret; | |
f232326f | 6126 | int err; |
979d63d5 | 6127 | |
d3bd7413 | 6128 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
6129 | return 0; |
6130 | ||
6131 | /* We already marked aux for masking from non-speculative | |
6132 | * paths, thus we got here in the first place. We only care | |
6133 | * to explore bad access from here. | |
6134 | */ | |
6135 | if (vstate->speculative) | |
6136 | goto do_sim; | |
6137 | ||
6138 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
6139 | alu_state |= ptr_is_dst_reg ? | |
6140 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
6141 | ||
f232326f PK |
6142 | err = retrieve_ptr_limit(ptr_reg, &alu_limit, opcode, off_is_neg); |
6143 | if (err < 0) | |
6144 | return err; | |
6145 | ||
6146 | err = update_alu_sanitation_state(aux, alu_state, alu_limit); | |
6147 | if (err < 0) | |
6148 | return err; | |
979d63d5 DB |
6149 | do_sim: |
6150 | /* Simulate and find potential out-of-bounds access under | |
6151 | * speculative execution from truncation as a result of | |
6152 | * masking when off was not within expected range. If off | |
6153 | * sits in dst, then we temporarily need to move ptr there | |
6154 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
6155 | * for cases where we use K-based arithmetic in one direction | |
6156 | * and truncated reg-based in the other in order to explore | |
6157 | * bad access. | |
6158 | */ | |
6159 | if (!ptr_is_dst_reg) { | |
6160 | tmp = *dst_reg; | |
6161 | *dst_reg = *ptr_reg; | |
6162 | } | |
6163 | ret = push_stack(env, env->insn_idx + 1, env->insn_idx, true); | |
0803278b | 6164 | if (!ptr_is_dst_reg && ret) |
979d63d5 DB |
6165 | *dst_reg = tmp; |
6166 | return !ret ? -EFAULT : 0; | |
6167 | } | |
6168 | ||
01f810ac AM |
6169 | /* check that stack access falls within stack limits and that 'reg' doesn't |
6170 | * have a variable offset. | |
6171 | * | |
6172 | * Variable offset is prohibited for unprivileged mode for simplicity since it | |
6173 | * requires corresponding support in Spectre masking for stack ALU. See also | |
6174 | * retrieve_ptr_limit(). | |
6175 | * | |
6176 | * | |
6177 | * 'off' includes 'reg->off'. | |
6178 | */ | |
6179 | static int check_stack_access_for_ptr_arithmetic( | |
6180 | struct bpf_verifier_env *env, | |
6181 | int regno, | |
6182 | const struct bpf_reg_state *reg, | |
6183 | int off) | |
6184 | { | |
6185 | if (!tnum_is_const(reg->var_off)) { | |
6186 | char tn_buf[48]; | |
6187 | ||
6188 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
6189 | verbose(env, "R%d variable stack access prohibited for !root, var_off=%s off=%d\n", | |
6190 | regno, tn_buf, off); | |
6191 | return -EACCES; | |
6192 | } | |
6193 | ||
6194 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
6195 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
6196 | "prohibited for !root; off=%d\n", regno, off); | |
6197 | return -EACCES; | |
6198 | } | |
6199 | ||
6200 | return 0; | |
6201 | } | |
6202 | ||
6203 | ||
f1174f77 | 6204 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
6205 | * Caller should also handle BPF_MOV case separately. |
6206 | * If we return -EACCES, caller may want to try again treating pointer as a | |
6207 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
6208 | */ | |
6209 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
6210 | struct bpf_insn *insn, | |
6211 | const struct bpf_reg_state *ptr_reg, | |
6212 | const struct bpf_reg_state *off_reg) | |
969bf05e | 6213 | { |
f4d7e40a AS |
6214 | struct bpf_verifier_state *vstate = env->cur_state; |
6215 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
6216 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 6217 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
6218 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
6219 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
6220 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
6221 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
9d7eceed | 6222 | u32 dst = insn->dst_reg, src = insn->src_reg; |
969bf05e | 6223 | u8 opcode = BPF_OP(insn->code); |
979d63d5 | 6224 | int ret; |
969bf05e | 6225 | |
f1174f77 | 6226 | dst_reg = ®s[dst]; |
969bf05e | 6227 | |
6f16101e DB |
6228 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
6229 | smin_val > smax_val || umin_val > umax_val) { | |
6230 | /* Taint dst register if offset had invalid bounds derived from | |
6231 | * e.g. dead branches. | |
6232 | */ | |
f54c7898 | 6233 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 6234 | return 0; |
f1174f77 EC |
6235 | } |
6236 | ||
6237 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
6238 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
6c693541 YS |
6239 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
6240 | __mark_reg_unknown(env, dst_reg); | |
6241 | return 0; | |
6242 | } | |
6243 | ||
82abbf8d AS |
6244 | verbose(env, |
6245 | "R%d 32-bit pointer arithmetic prohibited\n", | |
6246 | dst); | |
f1174f77 | 6247 | return -EACCES; |
969bf05e AS |
6248 | } |
6249 | ||
aad2eeaf JS |
6250 | switch (ptr_reg->type) { |
6251 | case PTR_TO_MAP_VALUE_OR_NULL: | |
6252 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", | |
6253 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 6254 | return -EACCES; |
aad2eeaf | 6255 | case CONST_PTR_TO_MAP: |
7c696732 YS |
6256 | /* smin_val represents the known value */ |
6257 | if (known && smin_val == 0 && opcode == BPF_ADD) | |
6258 | break; | |
8731745e | 6259 | fallthrough; |
aad2eeaf | 6260 | case PTR_TO_PACKET_END: |
c64b7983 JS |
6261 | case PTR_TO_SOCKET: |
6262 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
6263 | case PTR_TO_SOCK_COMMON: |
6264 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
6265 | case PTR_TO_TCP_SOCK: |
6266 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 6267 | case PTR_TO_XDP_SOCK: |
aad2eeaf JS |
6268 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
6269 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 6270 | return -EACCES; |
9d7eceed DB |
6271 | case PTR_TO_MAP_VALUE: |
6272 | if (!env->allow_ptr_leaks && !known && (smin_val < 0) != (smax_val < 0)) { | |
6273 | verbose(env, "R%d has unknown scalar with mixed signed bounds, pointer arithmetic with it prohibited for !root\n", | |
6274 | off_reg == dst_reg ? dst : src); | |
6275 | return -EACCES; | |
6276 | } | |
df561f66 | 6277 | fallthrough; |
aad2eeaf JS |
6278 | default: |
6279 | break; | |
f1174f77 EC |
6280 | } |
6281 | ||
6282 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
6283 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 6284 | */ |
f1174f77 EC |
6285 | dst_reg->type = ptr_reg->type; |
6286 | dst_reg->id = ptr_reg->id; | |
969bf05e | 6287 | |
bb7f0f98 AS |
6288 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
6289 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
6290 | return -EINVAL; | |
6291 | ||
3f50f132 JF |
6292 | /* pointer types do not carry 32-bit bounds at the moment. */ |
6293 | __mark_reg32_unbounded(dst_reg); | |
6294 | ||
f1174f77 EC |
6295 | switch (opcode) { |
6296 | case BPF_ADD: | |
979d63d5 DB |
6297 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
6298 | if (ret < 0) { | |
f232326f | 6299 | verbose(env, "R%d tried to add from different maps, paths, or prohibited types\n", dst); |
979d63d5 DB |
6300 | return ret; |
6301 | } | |
f1174f77 EC |
6302 | /* We can take a fixed offset as long as it doesn't overflow |
6303 | * the s32 'off' field | |
969bf05e | 6304 | */ |
b03c9f9f EC |
6305 | if (known && (ptr_reg->off + smin_val == |
6306 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 6307 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
6308 | dst_reg->smin_value = smin_ptr; |
6309 | dst_reg->smax_value = smax_ptr; | |
6310 | dst_reg->umin_value = umin_ptr; | |
6311 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 6312 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 6313 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 6314 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
6315 | break; |
6316 | } | |
f1174f77 EC |
6317 | /* A new variable offset is created. Note that off_reg->off |
6318 | * == 0, since it's a scalar. | |
6319 | * dst_reg gets the pointer type and since some positive | |
6320 | * integer value was added to the pointer, give it a new 'id' | |
6321 | * if it's a PTR_TO_PACKET. | |
6322 | * this creates a new 'base' pointer, off_reg (variable) gets | |
6323 | * added into the variable offset, and we copy the fixed offset | |
6324 | * from ptr_reg. | |
969bf05e | 6325 | */ |
b03c9f9f EC |
6326 | if (signed_add_overflows(smin_ptr, smin_val) || |
6327 | signed_add_overflows(smax_ptr, smax_val)) { | |
6328 | dst_reg->smin_value = S64_MIN; | |
6329 | dst_reg->smax_value = S64_MAX; | |
6330 | } else { | |
6331 | dst_reg->smin_value = smin_ptr + smin_val; | |
6332 | dst_reg->smax_value = smax_ptr + smax_val; | |
6333 | } | |
6334 | if (umin_ptr + umin_val < umin_ptr || | |
6335 | umax_ptr + umax_val < umax_ptr) { | |
6336 | dst_reg->umin_value = 0; | |
6337 | dst_reg->umax_value = U64_MAX; | |
6338 | } else { | |
6339 | dst_reg->umin_value = umin_ptr + umin_val; | |
6340 | dst_reg->umax_value = umax_ptr + umax_val; | |
6341 | } | |
f1174f77 EC |
6342 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
6343 | dst_reg->off = ptr_reg->off; | |
0962590e | 6344 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 6345 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
6346 | dst_reg->id = ++env->id_gen; |
6347 | /* something was added to pkt_ptr, set range to zero */ | |
22dc4a0f | 6348 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
f1174f77 EC |
6349 | } |
6350 | break; | |
6351 | case BPF_SUB: | |
979d63d5 DB |
6352 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
6353 | if (ret < 0) { | |
f232326f | 6354 | verbose(env, "R%d tried to sub from different maps, paths, or prohibited types\n", dst); |
979d63d5 DB |
6355 | return ret; |
6356 | } | |
f1174f77 EC |
6357 | if (dst_reg == off_reg) { |
6358 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
6359 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
6360 | dst); | |
f1174f77 EC |
6361 | return -EACCES; |
6362 | } | |
6363 | /* We don't allow subtraction from FP, because (according to | |
6364 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
6365 | * be able to deal with it. | |
969bf05e | 6366 | */ |
f1174f77 | 6367 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
6368 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
6369 | dst); | |
f1174f77 EC |
6370 | return -EACCES; |
6371 | } | |
b03c9f9f EC |
6372 | if (known && (ptr_reg->off - smin_val == |
6373 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 6374 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
6375 | dst_reg->smin_value = smin_ptr; |
6376 | dst_reg->smax_value = smax_ptr; | |
6377 | dst_reg->umin_value = umin_ptr; | |
6378 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
6379 | dst_reg->var_off = ptr_reg->var_off; |
6380 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 6381 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 6382 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
6383 | break; |
6384 | } | |
f1174f77 EC |
6385 | /* A new variable offset is created. If the subtrahend is known |
6386 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 6387 | */ |
b03c9f9f EC |
6388 | if (signed_sub_overflows(smin_ptr, smax_val) || |
6389 | signed_sub_overflows(smax_ptr, smin_val)) { | |
6390 | /* Overflow possible, we know nothing */ | |
6391 | dst_reg->smin_value = S64_MIN; | |
6392 | dst_reg->smax_value = S64_MAX; | |
6393 | } else { | |
6394 | dst_reg->smin_value = smin_ptr - smax_val; | |
6395 | dst_reg->smax_value = smax_ptr - smin_val; | |
6396 | } | |
6397 | if (umin_ptr < umax_val) { | |
6398 | /* Overflow possible, we know nothing */ | |
6399 | dst_reg->umin_value = 0; | |
6400 | dst_reg->umax_value = U64_MAX; | |
6401 | } else { | |
6402 | /* Cannot overflow (as long as bounds are consistent) */ | |
6403 | dst_reg->umin_value = umin_ptr - umax_val; | |
6404 | dst_reg->umax_value = umax_ptr - umin_val; | |
6405 | } | |
f1174f77 EC |
6406 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
6407 | dst_reg->off = ptr_reg->off; | |
0962590e | 6408 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 6409 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
6410 | dst_reg->id = ++env->id_gen; |
6411 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 6412 | if (smin_val < 0) |
22dc4a0f | 6413 | memset(&dst_reg->raw, 0, sizeof(dst_reg->raw)); |
43188702 | 6414 | } |
f1174f77 EC |
6415 | break; |
6416 | case BPF_AND: | |
6417 | case BPF_OR: | |
6418 | case BPF_XOR: | |
82abbf8d AS |
6419 | /* bitwise ops on pointers are troublesome, prohibit. */ |
6420 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
6421 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
6422 | return -EACCES; |
6423 | default: | |
6424 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
6425 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
6426 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 6427 | return -EACCES; |
43188702 JF |
6428 | } |
6429 | ||
bb7f0f98 AS |
6430 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
6431 | return -EINVAL; | |
6432 | ||
b03c9f9f EC |
6433 | __update_reg_bounds(dst_reg); |
6434 | __reg_deduce_bounds(dst_reg); | |
6435 | __reg_bound_offset(dst_reg); | |
0d6303db DB |
6436 | |
6437 | /* For unprivileged we require that resulting offset must be in bounds | |
6438 | * in order to be able to sanitize access later on. | |
6439 | */ | |
2c78ee89 | 6440 | if (!env->bypass_spec_v1) { |
e4298d25 DB |
6441 | if (dst_reg->type == PTR_TO_MAP_VALUE && |
6442 | check_map_access(env, dst, dst_reg->off, 1, false)) { | |
6443 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
6444 | "prohibited for !root\n", dst); | |
6445 | return -EACCES; | |
6446 | } else if (dst_reg->type == PTR_TO_STACK && | |
01f810ac AM |
6447 | check_stack_access_for_ptr_arithmetic( |
6448 | env, dst, dst_reg, dst_reg->off + | |
6449 | dst_reg->var_off.value)) { | |
e4298d25 DB |
6450 | return -EACCES; |
6451 | } | |
0d6303db DB |
6452 | } |
6453 | ||
43188702 JF |
6454 | return 0; |
6455 | } | |
6456 | ||
3f50f132 JF |
6457 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
6458 | struct bpf_reg_state *src_reg) | |
6459 | { | |
6460 | s32 smin_val = src_reg->s32_min_value; | |
6461 | s32 smax_val = src_reg->s32_max_value; | |
6462 | u32 umin_val = src_reg->u32_min_value; | |
6463 | u32 umax_val = src_reg->u32_max_value; | |
6464 | ||
6465 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
6466 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
6467 | dst_reg->s32_min_value = S32_MIN; | |
6468 | dst_reg->s32_max_value = S32_MAX; | |
6469 | } else { | |
6470 | dst_reg->s32_min_value += smin_val; | |
6471 | dst_reg->s32_max_value += smax_val; | |
6472 | } | |
6473 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
6474 | dst_reg->u32_max_value + umax_val < umax_val) { | |
6475 | dst_reg->u32_min_value = 0; | |
6476 | dst_reg->u32_max_value = U32_MAX; | |
6477 | } else { | |
6478 | dst_reg->u32_min_value += umin_val; | |
6479 | dst_reg->u32_max_value += umax_val; | |
6480 | } | |
6481 | } | |
6482 | ||
07cd2631 JF |
6483 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
6484 | struct bpf_reg_state *src_reg) | |
6485 | { | |
6486 | s64 smin_val = src_reg->smin_value; | |
6487 | s64 smax_val = src_reg->smax_value; | |
6488 | u64 umin_val = src_reg->umin_value; | |
6489 | u64 umax_val = src_reg->umax_value; | |
6490 | ||
6491 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
6492 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
6493 | dst_reg->smin_value = S64_MIN; | |
6494 | dst_reg->smax_value = S64_MAX; | |
6495 | } else { | |
6496 | dst_reg->smin_value += smin_val; | |
6497 | dst_reg->smax_value += smax_val; | |
6498 | } | |
6499 | if (dst_reg->umin_value + umin_val < umin_val || | |
6500 | dst_reg->umax_value + umax_val < umax_val) { | |
6501 | dst_reg->umin_value = 0; | |
6502 | dst_reg->umax_value = U64_MAX; | |
6503 | } else { | |
6504 | dst_reg->umin_value += umin_val; | |
6505 | dst_reg->umax_value += umax_val; | |
6506 | } | |
3f50f132 JF |
6507 | } |
6508 | ||
6509 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
6510 | struct bpf_reg_state *src_reg) | |
6511 | { | |
6512 | s32 smin_val = src_reg->s32_min_value; | |
6513 | s32 smax_val = src_reg->s32_max_value; | |
6514 | u32 umin_val = src_reg->u32_min_value; | |
6515 | u32 umax_val = src_reg->u32_max_value; | |
6516 | ||
6517 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
6518 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
6519 | /* Overflow possible, we know nothing */ | |
6520 | dst_reg->s32_min_value = S32_MIN; | |
6521 | dst_reg->s32_max_value = S32_MAX; | |
6522 | } else { | |
6523 | dst_reg->s32_min_value -= smax_val; | |
6524 | dst_reg->s32_max_value -= smin_val; | |
6525 | } | |
6526 | if (dst_reg->u32_min_value < umax_val) { | |
6527 | /* Overflow possible, we know nothing */ | |
6528 | dst_reg->u32_min_value = 0; | |
6529 | dst_reg->u32_max_value = U32_MAX; | |
6530 | } else { | |
6531 | /* Cannot overflow (as long as bounds are consistent) */ | |
6532 | dst_reg->u32_min_value -= umax_val; | |
6533 | dst_reg->u32_max_value -= umin_val; | |
6534 | } | |
07cd2631 JF |
6535 | } |
6536 | ||
6537 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
6538 | struct bpf_reg_state *src_reg) | |
6539 | { | |
6540 | s64 smin_val = src_reg->smin_value; | |
6541 | s64 smax_val = src_reg->smax_value; | |
6542 | u64 umin_val = src_reg->umin_value; | |
6543 | u64 umax_val = src_reg->umax_value; | |
6544 | ||
6545 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
6546 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
6547 | /* Overflow possible, we know nothing */ | |
6548 | dst_reg->smin_value = S64_MIN; | |
6549 | dst_reg->smax_value = S64_MAX; | |
6550 | } else { | |
6551 | dst_reg->smin_value -= smax_val; | |
6552 | dst_reg->smax_value -= smin_val; | |
6553 | } | |
6554 | if (dst_reg->umin_value < umax_val) { | |
6555 | /* Overflow possible, we know nothing */ | |
6556 | dst_reg->umin_value = 0; | |
6557 | dst_reg->umax_value = U64_MAX; | |
6558 | } else { | |
6559 | /* Cannot overflow (as long as bounds are consistent) */ | |
6560 | dst_reg->umin_value -= umax_val; | |
6561 | dst_reg->umax_value -= umin_val; | |
6562 | } | |
3f50f132 JF |
6563 | } |
6564 | ||
6565 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
6566 | struct bpf_reg_state *src_reg) | |
6567 | { | |
6568 | s32 smin_val = src_reg->s32_min_value; | |
6569 | u32 umin_val = src_reg->u32_min_value; | |
6570 | u32 umax_val = src_reg->u32_max_value; | |
6571 | ||
6572 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
6573 | /* Ain't nobody got time to multiply that sign */ | |
6574 | __mark_reg32_unbounded(dst_reg); | |
6575 | return; | |
6576 | } | |
6577 | /* Both values are positive, so we can work with unsigned and | |
6578 | * copy the result to signed (unless it exceeds S32_MAX). | |
6579 | */ | |
6580 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
6581 | /* Potential overflow, we know nothing */ | |
6582 | __mark_reg32_unbounded(dst_reg); | |
6583 | return; | |
6584 | } | |
6585 | dst_reg->u32_min_value *= umin_val; | |
6586 | dst_reg->u32_max_value *= umax_val; | |
6587 | if (dst_reg->u32_max_value > S32_MAX) { | |
6588 | /* Overflow possible, we know nothing */ | |
6589 | dst_reg->s32_min_value = S32_MIN; | |
6590 | dst_reg->s32_max_value = S32_MAX; | |
6591 | } else { | |
6592 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
6593 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
6594 | } | |
07cd2631 JF |
6595 | } |
6596 | ||
6597 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
6598 | struct bpf_reg_state *src_reg) | |
6599 | { | |
6600 | s64 smin_val = src_reg->smin_value; | |
6601 | u64 umin_val = src_reg->umin_value; | |
6602 | u64 umax_val = src_reg->umax_value; | |
6603 | ||
07cd2631 JF |
6604 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
6605 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 6606 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
6607 | return; |
6608 | } | |
6609 | /* Both values are positive, so we can work with unsigned and | |
6610 | * copy the result to signed (unless it exceeds S64_MAX). | |
6611 | */ | |
6612 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
6613 | /* Potential overflow, we know nothing */ | |
3f50f132 | 6614 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
6615 | return; |
6616 | } | |
6617 | dst_reg->umin_value *= umin_val; | |
6618 | dst_reg->umax_value *= umax_val; | |
6619 | if (dst_reg->umax_value > S64_MAX) { | |
6620 | /* Overflow possible, we know nothing */ | |
6621 | dst_reg->smin_value = S64_MIN; | |
6622 | dst_reg->smax_value = S64_MAX; | |
6623 | } else { | |
6624 | dst_reg->smin_value = dst_reg->umin_value; | |
6625 | dst_reg->smax_value = dst_reg->umax_value; | |
6626 | } | |
6627 | } | |
6628 | ||
3f50f132 JF |
6629 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
6630 | struct bpf_reg_state *src_reg) | |
6631 | { | |
6632 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
6633 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
6634 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
6635 | s32 smin_val = src_reg->s32_min_value; | |
6636 | u32 umax_val = src_reg->u32_max_value; | |
6637 | ||
6638 | /* Assuming scalar64_min_max_and will be called so its safe | |
6639 | * to skip updating register for known 32-bit case. | |
6640 | */ | |
6641 | if (src_known && dst_known) | |
6642 | return; | |
6643 | ||
6644 | /* We get our minimum from the var_off, since that's inherently | |
6645 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
6646 | */ | |
6647 | dst_reg->u32_min_value = var32_off.value; | |
6648 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
6649 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
6650 | /* Lose signed bounds when ANDing negative numbers, | |
6651 | * ain't nobody got time for that. | |
6652 | */ | |
6653 | dst_reg->s32_min_value = S32_MIN; | |
6654 | dst_reg->s32_max_value = S32_MAX; | |
6655 | } else { | |
6656 | /* ANDing two positives gives a positive, so safe to | |
6657 | * cast result into s64. | |
6658 | */ | |
6659 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
6660 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
6661 | } | |
6662 | ||
6663 | } | |
6664 | ||
07cd2631 JF |
6665 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
6666 | struct bpf_reg_state *src_reg) | |
6667 | { | |
3f50f132 JF |
6668 | bool src_known = tnum_is_const(src_reg->var_off); |
6669 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
6670 | s64 smin_val = src_reg->smin_value; |
6671 | u64 umax_val = src_reg->umax_value; | |
6672 | ||
3f50f132 | 6673 | if (src_known && dst_known) { |
4fbb38a3 | 6674 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
6675 | return; |
6676 | } | |
6677 | ||
07cd2631 JF |
6678 | /* We get our minimum from the var_off, since that's inherently |
6679 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
6680 | */ | |
07cd2631 JF |
6681 | dst_reg->umin_value = dst_reg->var_off.value; |
6682 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
6683 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
6684 | /* Lose signed bounds when ANDing negative numbers, | |
6685 | * ain't nobody got time for that. | |
6686 | */ | |
6687 | dst_reg->smin_value = S64_MIN; | |
6688 | dst_reg->smax_value = S64_MAX; | |
6689 | } else { | |
6690 | /* ANDing two positives gives a positive, so safe to | |
6691 | * cast result into s64. | |
6692 | */ | |
6693 | dst_reg->smin_value = dst_reg->umin_value; | |
6694 | dst_reg->smax_value = dst_reg->umax_value; | |
6695 | } | |
6696 | /* We may learn something more from the var_off */ | |
6697 | __update_reg_bounds(dst_reg); | |
6698 | } | |
6699 | ||
3f50f132 JF |
6700 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
6701 | struct bpf_reg_state *src_reg) | |
6702 | { | |
6703 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
6704 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
6705 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5b9fbeb7 DB |
6706 | s32 smin_val = src_reg->s32_min_value; |
6707 | u32 umin_val = src_reg->u32_min_value; | |
3f50f132 JF |
6708 | |
6709 | /* Assuming scalar64_min_max_or will be called so it is safe | |
6710 | * to skip updating register for known case. | |
6711 | */ | |
6712 | if (src_known && dst_known) | |
6713 | return; | |
6714 | ||
6715 | /* We get our maximum from the var_off, and our minimum is the | |
6716 | * maximum of the operands' minima | |
6717 | */ | |
6718 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
6719 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
6720 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
6721 | /* Lose signed bounds when ORing negative numbers, | |
6722 | * ain't nobody got time for that. | |
6723 | */ | |
6724 | dst_reg->s32_min_value = S32_MIN; | |
6725 | dst_reg->s32_max_value = S32_MAX; | |
6726 | } else { | |
6727 | /* ORing two positives gives a positive, so safe to | |
6728 | * cast result into s64. | |
6729 | */ | |
5b9fbeb7 DB |
6730 | dst_reg->s32_min_value = dst_reg->u32_min_value; |
6731 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
3f50f132 JF |
6732 | } |
6733 | } | |
6734 | ||
07cd2631 JF |
6735 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
6736 | struct bpf_reg_state *src_reg) | |
6737 | { | |
3f50f132 JF |
6738 | bool src_known = tnum_is_const(src_reg->var_off); |
6739 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
6740 | s64 smin_val = src_reg->smin_value; |
6741 | u64 umin_val = src_reg->umin_value; | |
6742 | ||
3f50f132 | 6743 | if (src_known && dst_known) { |
4fbb38a3 | 6744 | __mark_reg_known(dst_reg, dst_reg->var_off.value); |
3f50f132 JF |
6745 | return; |
6746 | } | |
6747 | ||
07cd2631 JF |
6748 | /* We get our maximum from the var_off, and our minimum is the |
6749 | * maximum of the operands' minima | |
6750 | */ | |
07cd2631 JF |
6751 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
6752 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
6753 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
6754 | /* Lose signed bounds when ORing negative numbers, | |
6755 | * ain't nobody got time for that. | |
6756 | */ | |
6757 | dst_reg->smin_value = S64_MIN; | |
6758 | dst_reg->smax_value = S64_MAX; | |
6759 | } else { | |
6760 | /* ORing two positives gives a positive, so safe to | |
6761 | * cast result into s64. | |
6762 | */ | |
6763 | dst_reg->smin_value = dst_reg->umin_value; | |
6764 | dst_reg->smax_value = dst_reg->umax_value; | |
6765 | } | |
6766 | /* We may learn something more from the var_off */ | |
6767 | __update_reg_bounds(dst_reg); | |
6768 | } | |
6769 | ||
2921c90d YS |
6770 | static void scalar32_min_max_xor(struct bpf_reg_state *dst_reg, |
6771 | struct bpf_reg_state *src_reg) | |
6772 | { | |
6773 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
6774 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
6775 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
6776 | s32 smin_val = src_reg->s32_min_value; | |
6777 | ||
6778 | /* Assuming scalar64_min_max_xor will be called so it is safe | |
6779 | * to skip updating register for known case. | |
6780 | */ | |
6781 | if (src_known && dst_known) | |
6782 | return; | |
6783 | ||
6784 | /* We get both minimum and maximum from the var32_off. */ | |
6785 | dst_reg->u32_min_value = var32_off.value; | |
6786 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
6787 | ||
6788 | if (dst_reg->s32_min_value >= 0 && smin_val >= 0) { | |
6789 | /* XORing two positive sign numbers gives a positive, | |
6790 | * so safe to cast u32 result into s32. | |
6791 | */ | |
6792 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
6793 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
6794 | } else { | |
6795 | dst_reg->s32_min_value = S32_MIN; | |
6796 | dst_reg->s32_max_value = S32_MAX; | |
6797 | } | |
6798 | } | |
6799 | ||
6800 | static void scalar_min_max_xor(struct bpf_reg_state *dst_reg, | |
6801 | struct bpf_reg_state *src_reg) | |
6802 | { | |
6803 | bool src_known = tnum_is_const(src_reg->var_off); | |
6804 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
6805 | s64 smin_val = src_reg->smin_value; | |
6806 | ||
6807 | if (src_known && dst_known) { | |
6808 | /* dst_reg->var_off.value has been updated earlier */ | |
6809 | __mark_reg_known(dst_reg, dst_reg->var_off.value); | |
6810 | return; | |
6811 | } | |
6812 | ||
6813 | /* We get both minimum and maximum from the var_off. */ | |
6814 | dst_reg->umin_value = dst_reg->var_off.value; | |
6815 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
6816 | ||
6817 | if (dst_reg->smin_value >= 0 && smin_val >= 0) { | |
6818 | /* XORing two positive sign numbers gives a positive, | |
6819 | * so safe to cast u64 result into s64. | |
6820 | */ | |
6821 | dst_reg->smin_value = dst_reg->umin_value; | |
6822 | dst_reg->smax_value = dst_reg->umax_value; | |
6823 | } else { | |
6824 | dst_reg->smin_value = S64_MIN; | |
6825 | dst_reg->smax_value = S64_MAX; | |
6826 | } | |
6827 | ||
6828 | __update_reg_bounds(dst_reg); | |
6829 | } | |
6830 | ||
3f50f132 JF |
6831 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
6832 | u64 umin_val, u64 umax_val) | |
07cd2631 | 6833 | { |
07cd2631 JF |
6834 | /* We lose all sign bit information (except what we can pick |
6835 | * up from var_off) | |
6836 | */ | |
3f50f132 JF |
6837 | dst_reg->s32_min_value = S32_MIN; |
6838 | dst_reg->s32_max_value = S32_MAX; | |
6839 | /* If we might shift our top bit out, then we know nothing */ | |
6840 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
6841 | dst_reg->u32_min_value = 0; | |
6842 | dst_reg->u32_max_value = U32_MAX; | |
6843 | } else { | |
6844 | dst_reg->u32_min_value <<= umin_val; | |
6845 | dst_reg->u32_max_value <<= umax_val; | |
6846 | } | |
6847 | } | |
6848 | ||
6849 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
6850 | struct bpf_reg_state *src_reg) | |
6851 | { | |
6852 | u32 umax_val = src_reg->u32_max_value; | |
6853 | u32 umin_val = src_reg->u32_min_value; | |
6854 | /* u32 alu operation will zext upper bits */ | |
6855 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
6856 | ||
6857 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
6858 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
6859 | /* Not required but being careful mark reg64 bounds as unknown so | |
6860 | * that we are forced to pick them up from tnum and zext later and | |
6861 | * if some path skips this step we are still safe. | |
6862 | */ | |
6863 | __mark_reg64_unbounded(dst_reg); | |
6864 | __update_reg32_bounds(dst_reg); | |
6865 | } | |
6866 | ||
6867 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
6868 | u64 umin_val, u64 umax_val) | |
6869 | { | |
6870 | /* Special case <<32 because it is a common compiler pattern to sign | |
6871 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
6872 | * positive we know this shift will also be positive so we can track | |
6873 | * bounds correctly. Otherwise we lose all sign bit information except | |
6874 | * what we can pick up from var_off. Perhaps we can generalize this | |
6875 | * later to shifts of any length. | |
6876 | */ | |
6877 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
6878 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
6879 | else | |
6880 | dst_reg->smax_value = S64_MAX; | |
6881 | ||
6882 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
6883 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
6884 | else | |
6885 | dst_reg->smin_value = S64_MIN; | |
6886 | ||
07cd2631 JF |
6887 | /* If we might shift our top bit out, then we know nothing */ |
6888 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
6889 | dst_reg->umin_value = 0; | |
6890 | dst_reg->umax_value = U64_MAX; | |
6891 | } else { | |
6892 | dst_reg->umin_value <<= umin_val; | |
6893 | dst_reg->umax_value <<= umax_val; | |
6894 | } | |
3f50f132 JF |
6895 | } |
6896 | ||
6897 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
6898 | struct bpf_reg_state *src_reg) | |
6899 | { | |
6900 | u64 umax_val = src_reg->umax_value; | |
6901 | u64 umin_val = src_reg->umin_value; | |
6902 | ||
6903 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
6904 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
6905 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
6906 | ||
07cd2631 JF |
6907 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
6908 | /* We may learn something more from the var_off */ | |
6909 | __update_reg_bounds(dst_reg); | |
6910 | } | |
6911 | ||
3f50f132 JF |
6912 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
6913 | struct bpf_reg_state *src_reg) | |
6914 | { | |
6915 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
6916 | u32 umax_val = src_reg->u32_max_value; | |
6917 | u32 umin_val = src_reg->u32_min_value; | |
6918 | ||
6919 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
6920 | * be negative, then either: | |
6921 | * 1) src_reg might be zero, so the sign bit of the result is | |
6922 | * unknown, so we lose our signed bounds | |
6923 | * 2) it's known negative, thus the unsigned bounds capture the | |
6924 | * signed bounds | |
6925 | * 3) the signed bounds cross zero, so they tell us nothing | |
6926 | * about the result | |
6927 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 6928 | * unsigned bounds capture the signed bounds. |
3f50f132 JF |
6929 | * Thus, in all cases it suffices to blow away our signed bounds |
6930 | * and rely on inferring new ones from the unsigned bounds and | |
6931 | * var_off of the result. | |
6932 | */ | |
6933 | dst_reg->s32_min_value = S32_MIN; | |
6934 | dst_reg->s32_max_value = S32_MAX; | |
6935 | ||
6936 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
6937 | dst_reg->u32_min_value >>= umax_val; | |
6938 | dst_reg->u32_max_value >>= umin_val; | |
6939 | ||
6940 | __mark_reg64_unbounded(dst_reg); | |
6941 | __update_reg32_bounds(dst_reg); | |
6942 | } | |
6943 | ||
07cd2631 JF |
6944 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
6945 | struct bpf_reg_state *src_reg) | |
6946 | { | |
6947 | u64 umax_val = src_reg->umax_value; | |
6948 | u64 umin_val = src_reg->umin_value; | |
6949 | ||
6950 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
6951 | * be negative, then either: | |
6952 | * 1) src_reg might be zero, so the sign bit of the result is | |
6953 | * unknown, so we lose our signed bounds | |
6954 | * 2) it's known negative, thus the unsigned bounds capture the | |
6955 | * signed bounds | |
6956 | * 3) the signed bounds cross zero, so they tell us nothing | |
6957 | * about the result | |
6958 | * If the value in dst_reg is known nonnegative, then again the | |
18b24d78 | 6959 | * unsigned bounds capture the signed bounds. |
07cd2631 JF |
6960 | * Thus, in all cases it suffices to blow away our signed bounds |
6961 | * and rely on inferring new ones from the unsigned bounds and | |
6962 | * var_off of the result. | |
6963 | */ | |
6964 | dst_reg->smin_value = S64_MIN; | |
6965 | dst_reg->smax_value = S64_MAX; | |
6966 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
6967 | dst_reg->umin_value >>= umax_val; | |
6968 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
6969 | |
6970 | /* Its not easy to operate on alu32 bounds here because it depends | |
6971 | * on bits being shifted in. Take easy way out and mark unbounded | |
6972 | * so we can recalculate later from tnum. | |
6973 | */ | |
6974 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
6975 | __update_reg_bounds(dst_reg); |
6976 | } | |
6977 | ||
3f50f132 JF |
6978 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
6979 | struct bpf_reg_state *src_reg) | |
07cd2631 | 6980 | { |
3f50f132 | 6981 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
6982 | |
6983 | /* Upon reaching here, src_known is true and | |
6984 | * umax_val is equal to umin_val. | |
6985 | */ | |
3f50f132 JF |
6986 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
6987 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 6988 | |
3f50f132 JF |
6989 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
6990 | ||
6991 | /* blow away the dst_reg umin_value/umax_value and rely on | |
6992 | * dst_reg var_off to refine the result. | |
6993 | */ | |
6994 | dst_reg->u32_min_value = 0; | |
6995 | dst_reg->u32_max_value = U32_MAX; | |
6996 | ||
6997 | __mark_reg64_unbounded(dst_reg); | |
6998 | __update_reg32_bounds(dst_reg); | |
6999 | } | |
7000 | ||
7001 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
7002 | struct bpf_reg_state *src_reg) | |
7003 | { | |
7004 | u64 umin_val = src_reg->umin_value; | |
7005 | ||
7006 | /* Upon reaching here, src_known is true and umax_val is equal | |
7007 | * to umin_val. | |
7008 | */ | |
7009 | dst_reg->smin_value >>= umin_val; | |
7010 | dst_reg->smax_value >>= umin_val; | |
7011 | ||
7012 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
7013 | |
7014 | /* blow away the dst_reg umin_value/umax_value and rely on | |
7015 | * dst_reg var_off to refine the result. | |
7016 | */ | |
7017 | dst_reg->umin_value = 0; | |
7018 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
7019 | |
7020 | /* Its not easy to operate on alu32 bounds here because it depends | |
7021 | * on bits being shifted in from upper 32-bits. Take easy way out | |
7022 | * and mark unbounded so we can recalculate later from tnum. | |
7023 | */ | |
7024 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
7025 | __update_reg_bounds(dst_reg); |
7026 | } | |
7027 | ||
468f6eaf JH |
7028 | /* WARNING: This function does calculations on 64-bit values, but the actual |
7029 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
7030 | * need extra checks in the 32-bit case. | |
7031 | */ | |
f1174f77 EC |
7032 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
7033 | struct bpf_insn *insn, | |
7034 | struct bpf_reg_state *dst_reg, | |
7035 | struct bpf_reg_state src_reg) | |
969bf05e | 7036 | { |
638f5b90 | 7037 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 7038 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 7039 | bool src_known; |
b03c9f9f EC |
7040 | s64 smin_val, smax_val; |
7041 | u64 umin_val, umax_val; | |
3f50f132 JF |
7042 | s32 s32_min_val, s32_max_val; |
7043 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 7044 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
d3bd7413 DB |
7045 | u32 dst = insn->dst_reg; |
7046 | int ret; | |
3f50f132 | 7047 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
b799207e | 7048 | |
b03c9f9f EC |
7049 | smin_val = src_reg.smin_value; |
7050 | smax_val = src_reg.smax_value; | |
7051 | umin_val = src_reg.umin_value; | |
7052 | umax_val = src_reg.umax_value; | |
f23cc643 | 7053 | |
3f50f132 JF |
7054 | s32_min_val = src_reg.s32_min_value; |
7055 | s32_max_val = src_reg.s32_max_value; | |
7056 | u32_min_val = src_reg.u32_min_value; | |
7057 | u32_max_val = src_reg.u32_max_value; | |
7058 | ||
7059 | if (alu32) { | |
7060 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
7061 | if ((src_known && |
7062 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
7063 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
7064 | /* Taint dst register if offset had invalid bounds | |
7065 | * derived from e.g. dead branches. | |
7066 | */ | |
7067 | __mark_reg_unknown(env, dst_reg); | |
7068 | return 0; | |
7069 | } | |
7070 | } else { | |
7071 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
7072 | if ((src_known && |
7073 | (smin_val != smax_val || umin_val != umax_val)) || | |
7074 | smin_val > smax_val || umin_val > umax_val) { | |
7075 | /* Taint dst register if offset had invalid bounds | |
7076 | * derived from e.g. dead branches. | |
7077 | */ | |
7078 | __mark_reg_unknown(env, dst_reg); | |
7079 | return 0; | |
7080 | } | |
6f16101e DB |
7081 | } |
7082 | ||
bb7f0f98 AS |
7083 | if (!src_known && |
7084 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 7085 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
7086 | return 0; |
7087 | } | |
7088 | ||
3f50f132 JF |
7089 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
7090 | * There are two classes of instructions: The first class we track both | |
7091 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
7092 | * greatest amount of precision when alu operations are mixed with jmp32 | |
7093 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
7094 | * and BPF_OR. This is possible because these ops have fairly easy to | |
7095 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
7096 | * See alu32 verifier tests for examples. The second class of | |
7097 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
7098 | * with regards to tracking sign/unsigned bounds because the bits may | |
7099 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
7100 | * the reg unbounded in the subreg bound space and use the resulting | |
7101 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
7102 | */ | |
48461135 JB |
7103 | switch (opcode) { |
7104 | case BPF_ADD: | |
d3bd7413 DB |
7105 | ret = sanitize_val_alu(env, insn); |
7106 | if (ret < 0) { | |
7107 | verbose(env, "R%d tried to add from different pointers or scalars\n", dst); | |
7108 | return ret; | |
7109 | } | |
3f50f132 | 7110 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 7111 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 7112 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
7113 | break; |
7114 | case BPF_SUB: | |
d3bd7413 DB |
7115 | ret = sanitize_val_alu(env, insn); |
7116 | if (ret < 0) { | |
7117 | verbose(env, "R%d tried to sub from different pointers or scalars\n", dst); | |
7118 | return ret; | |
7119 | } | |
3f50f132 | 7120 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 7121 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 7122 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
7123 | break; |
7124 | case BPF_MUL: | |
3f50f132 JF |
7125 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
7126 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 7127 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
7128 | break; |
7129 | case BPF_AND: | |
3f50f132 JF |
7130 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
7131 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 7132 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
7133 | break; |
7134 | case BPF_OR: | |
3f50f132 JF |
7135 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
7136 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 7137 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 | 7138 | break; |
2921c90d YS |
7139 | case BPF_XOR: |
7140 | dst_reg->var_off = tnum_xor(dst_reg->var_off, src_reg.var_off); | |
7141 | scalar32_min_max_xor(dst_reg, &src_reg); | |
7142 | scalar_min_max_xor(dst_reg, &src_reg); | |
7143 | break; | |
48461135 | 7144 | case BPF_LSH: |
468f6eaf JH |
7145 | if (umax_val >= insn_bitness) { |
7146 | /* Shifts greater than 31 or 63 are undefined. | |
7147 | * This includes shifts by a negative number. | |
b03c9f9f | 7148 | */ |
61bd5218 | 7149 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
7150 | break; |
7151 | } | |
3f50f132 JF |
7152 | if (alu32) |
7153 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
7154 | else | |
7155 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
7156 | break; |
7157 | case BPF_RSH: | |
468f6eaf JH |
7158 | if (umax_val >= insn_bitness) { |
7159 | /* Shifts greater than 31 or 63 are undefined. | |
7160 | * This includes shifts by a negative number. | |
b03c9f9f | 7161 | */ |
61bd5218 | 7162 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
7163 | break; |
7164 | } | |
3f50f132 JF |
7165 | if (alu32) |
7166 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
7167 | else | |
7168 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 7169 | break; |
9cbe1f5a YS |
7170 | case BPF_ARSH: |
7171 | if (umax_val >= insn_bitness) { | |
7172 | /* Shifts greater than 31 or 63 are undefined. | |
7173 | * This includes shifts by a negative number. | |
7174 | */ | |
7175 | mark_reg_unknown(env, regs, insn->dst_reg); | |
7176 | break; | |
7177 | } | |
3f50f132 JF |
7178 | if (alu32) |
7179 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
7180 | else | |
7181 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 7182 | break; |
48461135 | 7183 | default: |
61bd5218 | 7184 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
7185 | break; |
7186 | } | |
7187 | ||
3f50f132 JF |
7188 | /* ALU32 ops are zero extended into 64bit register */ |
7189 | if (alu32) | |
7190 | zext_32_to_64(dst_reg); | |
468f6eaf | 7191 | |
294f2fc6 | 7192 | __update_reg_bounds(dst_reg); |
b03c9f9f EC |
7193 | __reg_deduce_bounds(dst_reg); |
7194 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
7195 | return 0; |
7196 | } | |
7197 | ||
7198 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
7199 | * and var_off. | |
7200 | */ | |
7201 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
7202 | struct bpf_insn *insn) | |
7203 | { | |
f4d7e40a AS |
7204 | struct bpf_verifier_state *vstate = env->cur_state; |
7205 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
7206 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
7207 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
7208 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 7209 | int err; |
f1174f77 EC |
7210 | |
7211 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
7212 | src_reg = NULL; |
7213 | if (dst_reg->type != SCALAR_VALUE) | |
7214 | ptr_reg = dst_reg; | |
75748837 AS |
7215 | else |
7216 | /* Make sure ID is cleared otherwise dst_reg min/max could be | |
7217 | * incorrectly propagated into other registers by find_equal_scalars() | |
7218 | */ | |
7219 | dst_reg->id = 0; | |
f1174f77 EC |
7220 | if (BPF_SRC(insn->code) == BPF_X) { |
7221 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
7222 | if (src_reg->type != SCALAR_VALUE) { |
7223 | if (dst_reg->type != SCALAR_VALUE) { | |
7224 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
7225 | * an arbitrary scalar. Disallow all math except |
7226 | * pointer subtraction | |
f1174f77 | 7227 | */ |
dd066823 | 7228 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
7229 | mark_reg_unknown(env, regs, insn->dst_reg); |
7230 | return 0; | |
f1174f77 | 7231 | } |
82abbf8d AS |
7232 | verbose(env, "R%d pointer %s pointer prohibited\n", |
7233 | insn->dst_reg, | |
7234 | bpf_alu_string[opcode >> 4]); | |
7235 | return -EACCES; | |
f1174f77 EC |
7236 | } else { |
7237 | /* scalar += pointer | |
7238 | * This is legal, but we have to reverse our | |
7239 | * src/dest handling in computing the range | |
7240 | */ | |
b5dc0163 AS |
7241 | err = mark_chain_precision(env, insn->dst_reg); |
7242 | if (err) | |
7243 | return err; | |
82abbf8d AS |
7244 | return adjust_ptr_min_max_vals(env, insn, |
7245 | src_reg, dst_reg); | |
f1174f77 EC |
7246 | } |
7247 | } else if (ptr_reg) { | |
7248 | /* pointer += scalar */ | |
b5dc0163 AS |
7249 | err = mark_chain_precision(env, insn->src_reg); |
7250 | if (err) | |
7251 | return err; | |
82abbf8d AS |
7252 | return adjust_ptr_min_max_vals(env, insn, |
7253 | dst_reg, src_reg); | |
f1174f77 EC |
7254 | } |
7255 | } else { | |
7256 | /* Pretend the src is a reg with a known value, since we only | |
7257 | * need to be able to read from this state. | |
7258 | */ | |
7259 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 7260 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 7261 | src_reg = &off_reg; |
82abbf8d AS |
7262 | if (ptr_reg) /* pointer += K */ |
7263 | return adjust_ptr_min_max_vals(env, insn, | |
7264 | ptr_reg, src_reg); | |
f1174f77 EC |
7265 | } |
7266 | ||
7267 | /* Got here implies adding two SCALAR_VALUEs */ | |
7268 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 7269 | print_verifier_state(env, state); |
61bd5218 | 7270 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
7271 | return -EINVAL; |
7272 | } | |
7273 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 7274 | print_verifier_state(env, state); |
61bd5218 | 7275 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
7276 | return -EINVAL; |
7277 | } | |
7278 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
7279 | } |
7280 | ||
17a52670 | 7281 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 7282 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 7283 | { |
638f5b90 | 7284 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
7285 | u8 opcode = BPF_OP(insn->code); |
7286 | int err; | |
7287 | ||
7288 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
7289 | if (opcode == BPF_NEG) { | |
7290 | if (BPF_SRC(insn->code) != 0 || | |
7291 | insn->src_reg != BPF_REG_0 || | |
7292 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 7293 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
7294 | return -EINVAL; |
7295 | } | |
7296 | } else { | |
7297 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
7298 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
7299 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 7300 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
7301 | return -EINVAL; |
7302 | } | |
7303 | } | |
7304 | ||
7305 | /* check src operand */ | |
dc503a8a | 7306 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
7307 | if (err) |
7308 | return err; | |
7309 | ||
1be7f75d | 7310 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 7311 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
7312 | insn->dst_reg); |
7313 | return -EACCES; | |
7314 | } | |
7315 | ||
17a52670 | 7316 | /* check dest operand */ |
dc503a8a | 7317 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
7318 | if (err) |
7319 | return err; | |
7320 | ||
7321 | } else if (opcode == BPF_MOV) { | |
7322 | ||
7323 | if (BPF_SRC(insn->code) == BPF_X) { | |
7324 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 7325 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
7326 | return -EINVAL; |
7327 | } | |
7328 | ||
7329 | /* check src operand */ | |
dc503a8a | 7330 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
7331 | if (err) |
7332 | return err; | |
7333 | } else { | |
7334 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 7335 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
7336 | return -EINVAL; |
7337 | } | |
7338 | } | |
7339 | ||
fbeb1603 AF |
7340 | /* check dest operand, mark as required later */ |
7341 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
7342 | if (err) |
7343 | return err; | |
7344 | ||
7345 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
7346 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
7347 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
7348 | ||
17a52670 AS |
7349 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
7350 | /* case: R1 = R2 | |
7351 | * copy register state to dest reg | |
7352 | */ | |
75748837 AS |
7353 | if (src_reg->type == SCALAR_VALUE && !src_reg->id) |
7354 | /* Assign src and dst registers the same ID | |
7355 | * that will be used by find_equal_scalars() | |
7356 | * to propagate min/max range. | |
7357 | */ | |
7358 | src_reg->id = ++env->id_gen; | |
e434b8cd JW |
7359 | *dst_reg = *src_reg; |
7360 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 7361 | dst_reg->subreg_def = DEF_NOT_SUBREG; |
17a52670 | 7362 | } else { |
f1174f77 | 7363 | /* R1 = (u32) R2 */ |
1be7f75d | 7364 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
7365 | verbose(env, |
7366 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
7367 | insn->src_reg); |
7368 | return -EACCES; | |
e434b8cd JW |
7369 | } else if (src_reg->type == SCALAR_VALUE) { |
7370 | *dst_reg = *src_reg; | |
75748837 AS |
7371 | /* Make sure ID is cleared otherwise |
7372 | * dst_reg min/max could be incorrectly | |
7373 | * propagated into src_reg by find_equal_scalars() | |
7374 | */ | |
7375 | dst_reg->id = 0; | |
e434b8cd | 7376 | dst_reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 7377 | dst_reg->subreg_def = env->insn_idx + 1; |
e434b8cd JW |
7378 | } else { |
7379 | mark_reg_unknown(env, regs, | |
7380 | insn->dst_reg); | |
1be7f75d | 7381 | } |
3f50f132 | 7382 | zext_32_to_64(dst_reg); |
17a52670 AS |
7383 | } |
7384 | } else { | |
7385 | /* case: R = imm | |
7386 | * remember the value we stored into this reg | |
7387 | */ | |
fbeb1603 AF |
7388 | /* clear any state __mark_reg_known doesn't set */ |
7389 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 7390 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
7391 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
7392 | __mark_reg_known(regs + insn->dst_reg, | |
7393 | insn->imm); | |
7394 | } else { | |
7395 | __mark_reg_known(regs + insn->dst_reg, | |
7396 | (u32)insn->imm); | |
7397 | } | |
17a52670 AS |
7398 | } |
7399 | ||
7400 | } else if (opcode > BPF_END) { | |
61bd5218 | 7401 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
7402 | return -EINVAL; |
7403 | ||
7404 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
7405 | ||
17a52670 AS |
7406 | if (BPF_SRC(insn->code) == BPF_X) { |
7407 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 7408 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
7409 | return -EINVAL; |
7410 | } | |
7411 | /* check src1 operand */ | |
dc503a8a | 7412 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
7413 | if (err) |
7414 | return err; | |
7415 | } else { | |
7416 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 7417 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
7418 | return -EINVAL; |
7419 | } | |
7420 | } | |
7421 | ||
7422 | /* check src2 operand */ | |
dc503a8a | 7423 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
7424 | if (err) |
7425 | return err; | |
7426 | ||
7427 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
7428 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 7429 | verbose(env, "div by zero\n"); |
17a52670 AS |
7430 | return -EINVAL; |
7431 | } | |
7432 | ||
229394e8 RV |
7433 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
7434 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
7435 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
7436 | ||
7437 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 7438 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
7439 | return -EINVAL; |
7440 | } | |
7441 | } | |
7442 | ||
1a0dc1ac | 7443 | /* check dest operand */ |
dc503a8a | 7444 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
7445 | if (err) |
7446 | return err; | |
7447 | ||
f1174f77 | 7448 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
7449 | } |
7450 | ||
7451 | return 0; | |
7452 | } | |
7453 | ||
c6a9efa1 PC |
7454 | static void __find_good_pkt_pointers(struct bpf_func_state *state, |
7455 | struct bpf_reg_state *dst_reg, | |
6d94e741 | 7456 | enum bpf_reg_type type, int new_range) |
c6a9efa1 PC |
7457 | { |
7458 | struct bpf_reg_state *reg; | |
7459 | int i; | |
7460 | ||
7461 | for (i = 0; i < MAX_BPF_REG; i++) { | |
7462 | reg = &state->regs[i]; | |
7463 | if (reg->type == type && reg->id == dst_reg->id) | |
7464 | /* keep the maximum range already checked */ | |
7465 | reg->range = max(reg->range, new_range); | |
7466 | } | |
7467 | ||
7468 | bpf_for_each_spilled_reg(i, state, reg) { | |
7469 | if (!reg) | |
7470 | continue; | |
7471 | if (reg->type == type && reg->id == dst_reg->id) | |
7472 | reg->range = max(reg->range, new_range); | |
7473 | } | |
7474 | } | |
7475 | ||
f4d7e40a | 7476 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 7477 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 7478 | enum bpf_reg_type type, |
fb2a311a | 7479 | bool range_right_open) |
969bf05e | 7480 | { |
6d94e741 | 7481 | int new_range, i; |
2d2be8ca | 7482 | |
fb2a311a DB |
7483 | if (dst_reg->off < 0 || |
7484 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
7485 | /* This doesn't give us any range */ |
7486 | return; | |
7487 | ||
b03c9f9f EC |
7488 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
7489 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
7490 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
7491 | * than pkt_end, but that's because it's also less than pkt. | |
7492 | */ | |
7493 | return; | |
7494 | ||
fb2a311a DB |
7495 | new_range = dst_reg->off; |
7496 | if (range_right_open) | |
7497 | new_range--; | |
7498 | ||
7499 | /* Examples for register markings: | |
2d2be8ca | 7500 | * |
fb2a311a | 7501 | * pkt_data in dst register: |
2d2be8ca DB |
7502 | * |
7503 | * r2 = r3; | |
7504 | * r2 += 8; | |
7505 | * if (r2 > pkt_end) goto <handle exception> | |
7506 | * <access okay> | |
7507 | * | |
b4e432f1 DB |
7508 | * r2 = r3; |
7509 | * r2 += 8; | |
7510 | * if (r2 < pkt_end) goto <access okay> | |
7511 | * <handle exception> | |
7512 | * | |
2d2be8ca DB |
7513 | * Where: |
7514 | * r2 == dst_reg, pkt_end == src_reg | |
7515 | * r2=pkt(id=n,off=8,r=0) | |
7516 | * r3=pkt(id=n,off=0,r=0) | |
7517 | * | |
fb2a311a | 7518 | * pkt_data in src register: |
2d2be8ca DB |
7519 | * |
7520 | * r2 = r3; | |
7521 | * r2 += 8; | |
7522 | * if (pkt_end >= r2) goto <access okay> | |
7523 | * <handle exception> | |
7524 | * | |
b4e432f1 DB |
7525 | * r2 = r3; |
7526 | * r2 += 8; | |
7527 | * if (pkt_end <= r2) goto <handle exception> | |
7528 | * <access okay> | |
7529 | * | |
2d2be8ca DB |
7530 | * Where: |
7531 | * pkt_end == dst_reg, r2 == src_reg | |
7532 | * r2=pkt(id=n,off=8,r=0) | |
7533 | * r3=pkt(id=n,off=0,r=0) | |
7534 | * | |
7535 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
7536 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
7537 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
7538 | * the check. | |
969bf05e | 7539 | */ |
2d2be8ca | 7540 | |
f1174f77 EC |
7541 | /* If our ids match, then we must have the same max_value. And we |
7542 | * don't care about the other reg's fixed offset, since if it's too big | |
7543 | * the range won't allow anything. | |
7544 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
7545 | */ | |
c6a9efa1 PC |
7546 | for (i = 0; i <= vstate->curframe; i++) |
7547 | __find_good_pkt_pointers(vstate->frame[i], dst_reg, type, | |
7548 | new_range); | |
969bf05e AS |
7549 | } |
7550 | ||
3f50f132 | 7551 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 7552 | { |
3f50f132 JF |
7553 | struct tnum subreg = tnum_subreg(reg->var_off); |
7554 | s32 sval = (s32)val; | |
a72dafaf | 7555 | |
3f50f132 JF |
7556 | switch (opcode) { |
7557 | case BPF_JEQ: | |
7558 | if (tnum_is_const(subreg)) | |
7559 | return !!tnum_equals_const(subreg, val); | |
7560 | break; | |
7561 | case BPF_JNE: | |
7562 | if (tnum_is_const(subreg)) | |
7563 | return !tnum_equals_const(subreg, val); | |
7564 | break; | |
7565 | case BPF_JSET: | |
7566 | if ((~subreg.mask & subreg.value) & val) | |
7567 | return 1; | |
7568 | if (!((subreg.mask | subreg.value) & val)) | |
7569 | return 0; | |
7570 | break; | |
7571 | case BPF_JGT: | |
7572 | if (reg->u32_min_value > val) | |
7573 | return 1; | |
7574 | else if (reg->u32_max_value <= val) | |
7575 | return 0; | |
7576 | break; | |
7577 | case BPF_JSGT: | |
7578 | if (reg->s32_min_value > sval) | |
7579 | return 1; | |
ee114dd6 | 7580 | else if (reg->s32_max_value <= sval) |
3f50f132 JF |
7581 | return 0; |
7582 | break; | |
7583 | case BPF_JLT: | |
7584 | if (reg->u32_max_value < val) | |
7585 | return 1; | |
7586 | else if (reg->u32_min_value >= val) | |
7587 | return 0; | |
7588 | break; | |
7589 | case BPF_JSLT: | |
7590 | if (reg->s32_max_value < sval) | |
7591 | return 1; | |
7592 | else if (reg->s32_min_value >= sval) | |
7593 | return 0; | |
7594 | break; | |
7595 | case BPF_JGE: | |
7596 | if (reg->u32_min_value >= val) | |
7597 | return 1; | |
7598 | else if (reg->u32_max_value < val) | |
7599 | return 0; | |
7600 | break; | |
7601 | case BPF_JSGE: | |
7602 | if (reg->s32_min_value >= sval) | |
7603 | return 1; | |
7604 | else if (reg->s32_max_value < sval) | |
7605 | return 0; | |
7606 | break; | |
7607 | case BPF_JLE: | |
7608 | if (reg->u32_max_value <= val) | |
7609 | return 1; | |
7610 | else if (reg->u32_min_value > val) | |
7611 | return 0; | |
7612 | break; | |
7613 | case BPF_JSLE: | |
7614 | if (reg->s32_max_value <= sval) | |
7615 | return 1; | |
7616 | else if (reg->s32_min_value > sval) | |
7617 | return 0; | |
7618 | break; | |
7619 | } | |
4f7b3e82 | 7620 | |
3f50f132 JF |
7621 | return -1; |
7622 | } | |
092ed096 | 7623 | |
3f50f132 JF |
7624 | |
7625 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
7626 | { | |
7627 | s64 sval = (s64)val; | |
a72dafaf | 7628 | |
4f7b3e82 AS |
7629 | switch (opcode) { |
7630 | case BPF_JEQ: | |
7631 | if (tnum_is_const(reg->var_off)) | |
7632 | return !!tnum_equals_const(reg->var_off, val); | |
7633 | break; | |
7634 | case BPF_JNE: | |
7635 | if (tnum_is_const(reg->var_off)) | |
7636 | return !tnum_equals_const(reg->var_off, val); | |
7637 | break; | |
960ea056 JK |
7638 | case BPF_JSET: |
7639 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
7640 | return 1; | |
7641 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
7642 | return 0; | |
7643 | break; | |
4f7b3e82 AS |
7644 | case BPF_JGT: |
7645 | if (reg->umin_value > val) | |
7646 | return 1; | |
7647 | else if (reg->umax_value <= val) | |
7648 | return 0; | |
7649 | break; | |
7650 | case BPF_JSGT: | |
a72dafaf | 7651 | if (reg->smin_value > sval) |
4f7b3e82 | 7652 | return 1; |
ee114dd6 | 7653 | else if (reg->smax_value <= sval) |
4f7b3e82 AS |
7654 | return 0; |
7655 | break; | |
7656 | case BPF_JLT: | |
7657 | if (reg->umax_value < val) | |
7658 | return 1; | |
7659 | else if (reg->umin_value >= val) | |
7660 | return 0; | |
7661 | break; | |
7662 | case BPF_JSLT: | |
a72dafaf | 7663 | if (reg->smax_value < sval) |
4f7b3e82 | 7664 | return 1; |
a72dafaf | 7665 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
7666 | return 0; |
7667 | break; | |
7668 | case BPF_JGE: | |
7669 | if (reg->umin_value >= val) | |
7670 | return 1; | |
7671 | else if (reg->umax_value < val) | |
7672 | return 0; | |
7673 | break; | |
7674 | case BPF_JSGE: | |
a72dafaf | 7675 | if (reg->smin_value >= sval) |
4f7b3e82 | 7676 | return 1; |
a72dafaf | 7677 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
7678 | return 0; |
7679 | break; | |
7680 | case BPF_JLE: | |
7681 | if (reg->umax_value <= val) | |
7682 | return 1; | |
7683 | else if (reg->umin_value > val) | |
7684 | return 0; | |
7685 | break; | |
7686 | case BPF_JSLE: | |
a72dafaf | 7687 | if (reg->smax_value <= sval) |
4f7b3e82 | 7688 | return 1; |
a72dafaf | 7689 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
7690 | return 0; |
7691 | break; | |
7692 | } | |
7693 | ||
7694 | return -1; | |
7695 | } | |
7696 | ||
3f50f132 JF |
7697 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
7698 | * and return: | |
7699 | * 1 - branch will be taken and "goto target" will be executed | |
7700 | * 0 - branch will not be taken and fall-through to next insn | |
7701 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
7702 | * range [0,10] | |
604dca5e | 7703 | */ |
3f50f132 JF |
7704 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
7705 | bool is_jmp32) | |
604dca5e | 7706 | { |
cac616db JF |
7707 | if (__is_pointer_value(false, reg)) { |
7708 | if (!reg_type_not_null(reg->type)) | |
7709 | return -1; | |
7710 | ||
7711 | /* If pointer is valid tests against zero will fail so we can | |
7712 | * use this to direct branch taken. | |
7713 | */ | |
7714 | if (val != 0) | |
7715 | return -1; | |
7716 | ||
7717 | switch (opcode) { | |
7718 | case BPF_JEQ: | |
7719 | return 0; | |
7720 | case BPF_JNE: | |
7721 | return 1; | |
7722 | default: | |
7723 | return -1; | |
7724 | } | |
7725 | } | |
604dca5e | 7726 | |
3f50f132 JF |
7727 | if (is_jmp32) |
7728 | return is_branch32_taken(reg, val, opcode); | |
7729 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
7730 | } |
7731 | ||
6d94e741 AS |
7732 | static int flip_opcode(u32 opcode) |
7733 | { | |
7734 | /* How can we transform "a <op> b" into "b <op> a"? */ | |
7735 | static const u8 opcode_flip[16] = { | |
7736 | /* these stay the same */ | |
7737 | [BPF_JEQ >> 4] = BPF_JEQ, | |
7738 | [BPF_JNE >> 4] = BPF_JNE, | |
7739 | [BPF_JSET >> 4] = BPF_JSET, | |
7740 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
7741 | [BPF_JGE >> 4] = BPF_JLE, | |
7742 | [BPF_JGT >> 4] = BPF_JLT, | |
7743 | [BPF_JLE >> 4] = BPF_JGE, | |
7744 | [BPF_JLT >> 4] = BPF_JGT, | |
7745 | [BPF_JSGE >> 4] = BPF_JSLE, | |
7746 | [BPF_JSGT >> 4] = BPF_JSLT, | |
7747 | [BPF_JSLE >> 4] = BPF_JSGE, | |
7748 | [BPF_JSLT >> 4] = BPF_JSGT | |
7749 | }; | |
7750 | return opcode_flip[opcode >> 4]; | |
7751 | } | |
7752 | ||
7753 | static int is_pkt_ptr_branch_taken(struct bpf_reg_state *dst_reg, | |
7754 | struct bpf_reg_state *src_reg, | |
7755 | u8 opcode) | |
7756 | { | |
7757 | struct bpf_reg_state *pkt; | |
7758 | ||
7759 | if (src_reg->type == PTR_TO_PACKET_END) { | |
7760 | pkt = dst_reg; | |
7761 | } else if (dst_reg->type == PTR_TO_PACKET_END) { | |
7762 | pkt = src_reg; | |
7763 | opcode = flip_opcode(opcode); | |
7764 | } else { | |
7765 | return -1; | |
7766 | } | |
7767 | ||
7768 | if (pkt->range >= 0) | |
7769 | return -1; | |
7770 | ||
7771 | switch (opcode) { | |
7772 | case BPF_JLE: | |
7773 | /* pkt <= pkt_end */ | |
7774 | fallthrough; | |
7775 | case BPF_JGT: | |
7776 | /* pkt > pkt_end */ | |
7777 | if (pkt->range == BEYOND_PKT_END) | |
7778 | /* pkt has at last one extra byte beyond pkt_end */ | |
7779 | return opcode == BPF_JGT; | |
7780 | break; | |
7781 | case BPF_JLT: | |
7782 | /* pkt < pkt_end */ | |
7783 | fallthrough; | |
7784 | case BPF_JGE: | |
7785 | /* pkt >= pkt_end */ | |
7786 | if (pkt->range == BEYOND_PKT_END || pkt->range == AT_PKT_END) | |
7787 | return opcode == BPF_JGE; | |
7788 | break; | |
7789 | } | |
7790 | return -1; | |
7791 | } | |
7792 | ||
48461135 JB |
7793 | /* Adjusts the register min/max values in the case that the dst_reg is the |
7794 | * variable register that we are working on, and src_reg is a constant or we're | |
7795 | * simply doing a BPF_K check. | |
f1174f77 | 7796 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
7797 | */ |
7798 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
7799 | struct bpf_reg_state *false_reg, |
7800 | u64 val, u32 val32, | |
092ed096 | 7801 | u8 opcode, bool is_jmp32) |
48461135 | 7802 | { |
3f50f132 JF |
7803 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
7804 | struct tnum false_64off = false_reg->var_off; | |
7805 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
7806 | struct tnum true_64off = true_reg->var_off; | |
7807 | s64 sval = (s64)val; | |
7808 | s32 sval32 = (s32)val32; | |
a72dafaf | 7809 | |
f1174f77 EC |
7810 | /* If the dst_reg is a pointer, we can't learn anything about its |
7811 | * variable offset from the compare (unless src_reg were a pointer into | |
7812 | * the same object, but we don't bother with that. | |
7813 | * Since false_reg and true_reg have the same type by construction, we | |
7814 | * only need to check one of them for pointerness. | |
7815 | */ | |
7816 | if (__is_pointer_value(false, false_reg)) | |
7817 | return; | |
4cabc5b1 | 7818 | |
48461135 JB |
7819 | switch (opcode) { |
7820 | case BPF_JEQ: | |
48461135 | 7821 | case BPF_JNE: |
a72dafaf JW |
7822 | { |
7823 | struct bpf_reg_state *reg = | |
7824 | opcode == BPF_JEQ ? true_reg : false_reg; | |
7825 | ||
e688c3db AS |
7826 | /* JEQ/JNE comparison doesn't change the register equivalence. |
7827 | * r1 = r2; | |
7828 | * if (r1 == 42) goto label; | |
7829 | * ... | |
7830 | * label: // here both r1 and r2 are known to be 42. | |
7831 | * | |
7832 | * Hence when marking register as known preserve it's ID. | |
48461135 | 7833 | */ |
3f50f132 JF |
7834 | if (is_jmp32) |
7835 | __mark_reg32_known(reg, val32); | |
7836 | else | |
e688c3db | 7837 | ___mark_reg_known(reg, val); |
48461135 | 7838 | break; |
a72dafaf | 7839 | } |
960ea056 | 7840 | case BPF_JSET: |
3f50f132 JF |
7841 | if (is_jmp32) { |
7842 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
7843 | if (is_power_of_2(val32)) | |
7844 | true_32off = tnum_or(true_32off, | |
7845 | tnum_const(val32)); | |
7846 | } else { | |
7847 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
7848 | if (is_power_of_2(val)) | |
7849 | true_64off = tnum_or(true_64off, | |
7850 | tnum_const(val)); | |
7851 | } | |
960ea056 | 7852 | break; |
48461135 | 7853 | case BPF_JGE: |
a72dafaf JW |
7854 | case BPF_JGT: |
7855 | { | |
3f50f132 JF |
7856 | if (is_jmp32) { |
7857 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
7858 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
7859 | ||
7860 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
7861 | false_umax); | |
7862 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
7863 | true_umin); | |
7864 | } else { | |
7865 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
7866 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
7867 | ||
7868 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
7869 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
7870 | } | |
b03c9f9f | 7871 | break; |
a72dafaf | 7872 | } |
48461135 | 7873 | case BPF_JSGE: |
a72dafaf JW |
7874 | case BPF_JSGT: |
7875 | { | |
3f50f132 JF |
7876 | if (is_jmp32) { |
7877 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
7878 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 7879 | |
3f50f132 JF |
7880 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
7881 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
7882 | } else { | |
7883 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
7884 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
7885 | ||
7886 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
7887 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
7888 | } | |
48461135 | 7889 | break; |
a72dafaf | 7890 | } |
b4e432f1 | 7891 | case BPF_JLE: |
a72dafaf JW |
7892 | case BPF_JLT: |
7893 | { | |
3f50f132 JF |
7894 | if (is_jmp32) { |
7895 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
7896 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
7897 | ||
7898 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
7899 | false_umin); | |
7900 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
7901 | true_umax); | |
7902 | } else { | |
7903 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
7904 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
7905 | ||
7906 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
7907 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
7908 | } | |
b4e432f1 | 7909 | break; |
a72dafaf | 7910 | } |
b4e432f1 | 7911 | case BPF_JSLE: |
a72dafaf JW |
7912 | case BPF_JSLT: |
7913 | { | |
3f50f132 JF |
7914 | if (is_jmp32) { |
7915 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
7916 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 7917 | |
3f50f132 JF |
7918 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
7919 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
7920 | } else { | |
7921 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
7922 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
7923 | ||
7924 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
7925 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
7926 | } | |
b4e432f1 | 7927 | break; |
a72dafaf | 7928 | } |
48461135 | 7929 | default: |
0fc31b10 | 7930 | return; |
48461135 JB |
7931 | } |
7932 | ||
3f50f132 JF |
7933 | if (is_jmp32) { |
7934 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
7935 | tnum_subreg(false_32off)); | |
7936 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
7937 | tnum_subreg(true_32off)); | |
7938 | __reg_combine_32_into_64(false_reg); | |
7939 | __reg_combine_32_into_64(true_reg); | |
7940 | } else { | |
7941 | false_reg->var_off = false_64off; | |
7942 | true_reg->var_off = true_64off; | |
7943 | __reg_combine_64_into_32(false_reg); | |
7944 | __reg_combine_64_into_32(true_reg); | |
7945 | } | |
48461135 JB |
7946 | } |
7947 | ||
f1174f77 EC |
7948 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
7949 | * the variable reg. | |
48461135 JB |
7950 | */ |
7951 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
7952 | struct bpf_reg_state *false_reg, |
7953 | u64 val, u32 val32, | |
092ed096 | 7954 | u8 opcode, bool is_jmp32) |
48461135 | 7955 | { |
6d94e741 | 7956 | opcode = flip_opcode(opcode); |
0fc31b10 JH |
7957 | /* This uses zero as "not present in table"; luckily the zero opcode, |
7958 | * BPF_JA, can't get here. | |
b03c9f9f | 7959 | */ |
0fc31b10 | 7960 | if (opcode) |
3f50f132 | 7961 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
7962 | } |
7963 | ||
7964 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
7965 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
7966 | struct bpf_reg_state *dst_reg) | |
7967 | { | |
b03c9f9f EC |
7968 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
7969 | dst_reg->umin_value); | |
7970 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
7971 | dst_reg->umax_value); | |
7972 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
7973 | dst_reg->smin_value); | |
7974 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
7975 | dst_reg->smax_value); | |
f1174f77 EC |
7976 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
7977 | dst_reg->var_off); | |
b03c9f9f EC |
7978 | /* We might have learned new bounds from the var_off. */ |
7979 | __update_reg_bounds(src_reg); | |
7980 | __update_reg_bounds(dst_reg); | |
7981 | /* We might have learned something about the sign bit. */ | |
7982 | __reg_deduce_bounds(src_reg); | |
7983 | __reg_deduce_bounds(dst_reg); | |
7984 | /* We might have learned some bits from the bounds. */ | |
7985 | __reg_bound_offset(src_reg); | |
7986 | __reg_bound_offset(dst_reg); | |
7987 | /* Intersecting with the old var_off might have improved our bounds | |
7988 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
7989 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
7990 | */ | |
7991 | __update_reg_bounds(src_reg); | |
7992 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
7993 | } |
7994 | ||
7995 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
7996 | struct bpf_reg_state *true_dst, | |
7997 | struct bpf_reg_state *false_src, | |
7998 | struct bpf_reg_state *false_dst, | |
7999 | u8 opcode) | |
8000 | { | |
8001 | switch (opcode) { | |
8002 | case BPF_JEQ: | |
8003 | __reg_combine_min_max(true_src, true_dst); | |
8004 | break; | |
8005 | case BPF_JNE: | |
8006 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 8007 | break; |
4cabc5b1 | 8008 | } |
48461135 JB |
8009 | } |
8010 | ||
fd978bf7 JS |
8011 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
8012 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 8013 | bool is_null) |
57a09bf0 | 8014 | { |
93c230e3 MKL |
8015 | if (reg_type_may_be_null(reg->type) && reg->id == id && |
8016 | !WARN_ON_ONCE(!reg->id)) { | |
f1174f77 EC |
8017 | /* Old offset (both fixed and variable parts) should |
8018 | * have been known-zero, because we don't allow pointer | |
8019 | * arithmetic on pointers that might be NULL. | |
8020 | */ | |
b03c9f9f EC |
8021 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
8022 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 8023 | reg->off)) { |
b03c9f9f EC |
8024 | __mark_reg_known_zero(reg); |
8025 | reg->off = 0; | |
f1174f77 EC |
8026 | } |
8027 | if (is_null) { | |
8028 | reg->type = SCALAR_VALUE; | |
1b986589 MKL |
8029 | /* We don't need id and ref_obj_id from this point |
8030 | * onwards anymore, thus we should better reset it, | |
8031 | * so that state pruning has chances to take effect. | |
8032 | */ | |
8033 | reg->id = 0; | |
8034 | reg->ref_obj_id = 0; | |
4ddb7416 DB |
8035 | |
8036 | return; | |
8037 | } | |
8038 | ||
8039 | mark_ptr_not_null_reg(reg); | |
8040 | ||
8041 | if (!reg_may_point_to_spin_lock(reg)) { | |
1b986589 MKL |
8042 | /* For not-NULL ptr, reg->ref_obj_id will be reset |
8043 | * in release_reg_references(). | |
8044 | * | |
8045 | * reg->id is still used by spin_lock ptr. Other | |
8046 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
8047 | */ |
8048 | reg->id = 0; | |
56f668df | 8049 | } |
57a09bf0 TG |
8050 | } |
8051 | } | |
8052 | ||
c6a9efa1 PC |
8053 | static void __mark_ptr_or_null_regs(struct bpf_func_state *state, u32 id, |
8054 | bool is_null) | |
8055 | { | |
8056 | struct bpf_reg_state *reg; | |
8057 | int i; | |
8058 | ||
8059 | for (i = 0; i < MAX_BPF_REG; i++) | |
8060 | mark_ptr_or_null_reg(state, &state->regs[i], id, is_null); | |
8061 | ||
8062 | bpf_for_each_spilled_reg(i, state, reg) { | |
8063 | if (!reg) | |
8064 | continue; | |
8065 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
8066 | } | |
8067 | } | |
8068 | ||
57a09bf0 TG |
8069 | /* The logic is similar to find_good_pkt_pointers(), both could eventually |
8070 | * be folded together at some point. | |
8071 | */ | |
840b9615 JS |
8072 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
8073 | bool is_null) | |
57a09bf0 | 8074 | { |
f4d7e40a | 8075 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
c6a9efa1 | 8076 | struct bpf_reg_state *regs = state->regs; |
1b986589 | 8077 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 8078 | u32 id = regs[regno].id; |
c6a9efa1 | 8079 | int i; |
57a09bf0 | 8080 | |
1b986589 MKL |
8081 | if (ref_obj_id && ref_obj_id == id && is_null) |
8082 | /* regs[regno] is in the " == NULL" branch. | |
8083 | * No one could have freed the reference state before | |
8084 | * doing the NULL check. | |
8085 | */ | |
8086 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 8087 | |
c6a9efa1 PC |
8088 | for (i = 0; i <= vstate->curframe; i++) |
8089 | __mark_ptr_or_null_regs(vstate->frame[i], id, is_null); | |
57a09bf0 TG |
8090 | } |
8091 | ||
5beca081 DB |
8092 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
8093 | struct bpf_reg_state *dst_reg, | |
8094 | struct bpf_reg_state *src_reg, | |
8095 | struct bpf_verifier_state *this_branch, | |
8096 | struct bpf_verifier_state *other_branch) | |
8097 | { | |
8098 | if (BPF_SRC(insn->code) != BPF_X) | |
8099 | return false; | |
8100 | ||
092ed096 JW |
8101 | /* Pointers are always 64-bit. */ |
8102 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
8103 | return false; | |
8104 | ||
5beca081 DB |
8105 | switch (BPF_OP(insn->code)) { |
8106 | case BPF_JGT: | |
8107 | if ((dst_reg->type == PTR_TO_PACKET && | |
8108 | src_reg->type == PTR_TO_PACKET_END) || | |
8109 | (dst_reg->type == PTR_TO_PACKET_META && | |
8110 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8111 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
8112 | find_good_pkt_pointers(this_branch, dst_reg, | |
8113 | dst_reg->type, false); | |
6d94e741 | 8114 | mark_pkt_end(other_branch, insn->dst_reg, true); |
5beca081 DB |
8115 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8116 | src_reg->type == PTR_TO_PACKET) || | |
8117 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8118 | src_reg->type == PTR_TO_PACKET_META)) { | |
8119 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
8120 | find_good_pkt_pointers(other_branch, src_reg, | |
8121 | src_reg->type, true); | |
6d94e741 | 8122 | mark_pkt_end(this_branch, insn->src_reg, false); |
5beca081 DB |
8123 | } else { |
8124 | return false; | |
8125 | } | |
8126 | break; | |
8127 | case BPF_JLT: | |
8128 | if ((dst_reg->type == PTR_TO_PACKET && | |
8129 | src_reg->type == PTR_TO_PACKET_END) || | |
8130 | (dst_reg->type == PTR_TO_PACKET_META && | |
8131 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8132 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
8133 | find_good_pkt_pointers(other_branch, dst_reg, | |
8134 | dst_reg->type, true); | |
6d94e741 | 8135 | mark_pkt_end(this_branch, insn->dst_reg, false); |
5beca081 DB |
8136 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8137 | src_reg->type == PTR_TO_PACKET) || | |
8138 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8139 | src_reg->type == PTR_TO_PACKET_META)) { | |
8140 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
8141 | find_good_pkt_pointers(this_branch, src_reg, | |
8142 | src_reg->type, false); | |
6d94e741 | 8143 | mark_pkt_end(other_branch, insn->src_reg, true); |
5beca081 DB |
8144 | } else { |
8145 | return false; | |
8146 | } | |
8147 | break; | |
8148 | case BPF_JGE: | |
8149 | if ((dst_reg->type == PTR_TO_PACKET && | |
8150 | src_reg->type == PTR_TO_PACKET_END) || | |
8151 | (dst_reg->type == PTR_TO_PACKET_META && | |
8152 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8153 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
8154 | find_good_pkt_pointers(this_branch, dst_reg, | |
8155 | dst_reg->type, true); | |
6d94e741 | 8156 | mark_pkt_end(other_branch, insn->dst_reg, false); |
5beca081 DB |
8157 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8158 | src_reg->type == PTR_TO_PACKET) || | |
8159 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8160 | src_reg->type == PTR_TO_PACKET_META)) { | |
8161 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
8162 | find_good_pkt_pointers(other_branch, src_reg, | |
8163 | src_reg->type, false); | |
6d94e741 | 8164 | mark_pkt_end(this_branch, insn->src_reg, true); |
5beca081 DB |
8165 | } else { |
8166 | return false; | |
8167 | } | |
8168 | break; | |
8169 | case BPF_JLE: | |
8170 | if ((dst_reg->type == PTR_TO_PACKET && | |
8171 | src_reg->type == PTR_TO_PACKET_END) || | |
8172 | (dst_reg->type == PTR_TO_PACKET_META && | |
8173 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
8174 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
8175 | find_good_pkt_pointers(other_branch, dst_reg, | |
8176 | dst_reg->type, false); | |
6d94e741 | 8177 | mark_pkt_end(this_branch, insn->dst_reg, true); |
5beca081 DB |
8178 | } else if ((dst_reg->type == PTR_TO_PACKET_END && |
8179 | src_reg->type == PTR_TO_PACKET) || | |
8180 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
8181 | src_reg->type == PTR_TO_PACKET_META)) { | |
8182 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
8183 | find_good_pkt_pointers(this_branch, src_reg, | |
8184 | src_reg->type, true); | |
6d94e741 | 8185 | mark_pkt_end(other_branch, insn->src_reg, false); |
5beca081 DB |
8186 | } else { |
8187 | return false; | |
8188 | } | |
8189 | break; | |
8190 | default: | |
8191 | return false; | |
8192 | } | |
8193 | ||
8194 | return true; | |
8195 | } | |
8196 | ||
75748837 AS |
8197 | static void find_equal_scalars(struct bpf_verifier_state *vstate, |
8198 | struct bpf_reg_state *known_reg) | |
8199 | { | |
8200 | struct bpf_func_state *state; | |
8201 | struct bpf_reg_state *reg; | |
8202 | int i, j; | |
8203 | ||
8204 | for (i = 0; i <= vstate->curframe; i++) { | |
8205 | state = vstate->frame[i]; | |
8206 | for (j = 0; j < MAX_BPF_REG; j++) { | |
8207 | reg = &state->regs[j]; | |
8208 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
8209 | *reg = *known_reg; | |
8210 | } | |
8211 | ||
8212 | bpf_for_each_spilled_reg(j, state, reg) { | |
8213 | if (!reg) | |
8214 | continue; | |
8215 | if (reg->type == SCALAR_VALUE && reg->id == known_reg->id) | |
8216 | *reg = *known_reg; | |
8217 | } | |
8218 | } | |
8219 | } | |
8220 | ||
58e2af8b | 8221 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
8222 | struct bpf_insn *insn, int *insn_idx) |
8223 | { | |
f4d7e40a AS |
8224 | struct bpf_verifier_state *this_branch = env->cur_state; |
8225 | struct bpf_verifier_state *other_branch; | |
8226 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 8227 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
17a52670 | 8228 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 8229 | bool is_jmp32; |
fb8d251e | 8230 | int pred = -1; |
17a52670 AS |
8231 | int err; |
8232 | ||
092ed096 JW |
8233 | /* Only conditional jumps are expected to reach here. */ |
8234 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
8235 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
8236 | return -EINVAL; |
8237 | } | |
8238 | ||
8239 | if (BPF_SRC(insn->code) == BPF_X) { | |
8240 | if (insn->imm != 0) { | |
092ed096 | 8241 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
8242 | return -EINVAL; |
8243 | } | |
8244 | ||
8245 | /* check src1 operand */ | |
dc503a8a | 8246 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
8247 | if (err) |
8248 | return err; | |
1be7f75d AS |
8249 | |
8250 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 8251 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
8252 | insn->src_reg); |
8253 | return -EACCES; | |
8254 | } | |
fb8d251e | 8255 | src_reg = ®s[insn->src_reg]; |
17a52670 AS |
8256 | } else { |
8257 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 8258 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
8259 | return -EINVAL; |
8260 | } | |
8261 | } | |
8262 | ||
8263 | /* check src2 operand */ | |
dc503a8a | 8264 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
8265 | if (err) |
8266 | return err; | |
8267 | ||
1a0dc1ac | 8268 | dst_reg = ®s[insn->dst_reg]; |
092ed096 | 8269 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 8270 | |
3f50f132 JF |
8271 | if (BPF_SRC(insn->code) == BPF_K) { |
8272 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
8273 | } else if (src_reg->type == SCALAR_VALUE && | |
8274 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
8275 | pred = is_branch_taken(dst_reg, | |
8276 | tnum_subreg(src_reg->var_off).value, | |
8277 | opcode, | |
8278 | is_jmp32); | |
8279 | } else if (src_reg->type == SCALAR_VALUE && | |
8280 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
8281 | pred = is_branch_taken(dst_reg, | |
8282 | src_reg->var_off.value, | |
8283 | opcode, | |
8284 | is_jmp32); | |
6d94e741 AS |
8285 | } else if (reg_is_pkt_pointer_any(dst_reg) && |
8286 | reg_is_pkt_pointer_any(src_reg) && | |
8287 | !is_jmp32) { | |
8288 | pred = is_pkt_ptr_branch_taken(dst_reg, src_reg, opcode); | |
3f50f132 JF |
8289 | } |
8290 | ||
b5dc0163 | 8291 | if (pred >= 0) { |
cac616db JF |
8292 | /* If we get here with a dst_reg pointer type it is because |
8293 | * above is_branch_taken() special cased the 0 comparison. | |
8294 | */ | |
8295 | if (!__is_pointer_value(false, dst_reg)) | |
8296 | err = mark_chain_precision(env, insn->dst_reg); | |
6d94e741 AS |
8297 | if (BPF_SRC(insn->code) == BPF_X && !err && |
8298 | !__is_pointer_value(false, src_reg)) | |
b5dc0163 AS |
8299 | err = mark_chain_precision(env, insn->src_reg); |
8300 | if (err) | |
8301 | return err; | |
8302 | } | |
fb8d251e AS |
8303 | if (pred == 1) { |
8304 | /* only follow the goto, ignore fall-through */ | |
8305 | *insn_idx += insn->off; | |
8306 | return 0; | |
8307 | } else if (pred == 0) { | |
8308 | /* only follow fall-through branch, since | |
8309 | * that's where the program will go | |
8310 | */ | |
8311 | return 0; | |
17a52670 AS |
8312 | } |
8313 | ||
979d63d5 DB |
8314 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
8315 | false); | |
17a52670 AS |
8316 | if (!other_branch) |
8317 | return -EFAULT; | |
f4d7e40a | 8318 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 8319 | |
48461135 JB |
8320 | /* detect if we are comparing against a constant value so we can adjust |
8321 | * our min/max values for our dst register. | |
f1174f77 EC |
8322 | * this is only legit if both are scalars (or pointers to the same |
8323 | * object, I suppose, but we don't support that right now), because | |
8324 | * otherwise the different base pointers mean the offsets aren't | |
8325 | * comparable. | |
48461135 JB |
8326 | */ |
8327 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 8328 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 8329 | |
f1174f77 | 8330 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
8331 | src_reg->type == SCALAR_VALUE) { |
8332 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
8333 | (is_jmp32 && |
8334 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 8335 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 8336 | dst_reg, |
3f50f132 JF |
8337 | src_reg->var_off.value, |
8338 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
8339 | opcode, is_jmp32); |
8340 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
8341 | (is_jmp32 && |
8342 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 8343 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 8344 | src_reg, |
3f50f132 JF |
8345 | dst_reg->var_off.value, |
8346 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
8347 | opcode, is_jmp32); |
8348 | else if (!is_jmp32 && | |
8349 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 8350 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
8351 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
8352 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 8353 | src_reg, dst_reg, opcode); |
e688c3db AS |
8354 | if (src_reg->id && |
8355 | !WARN_ON_ONCE(src_reg->id != other_branch_regs[insn->src_reg].id)) { | |
75748837 AS |
8356 | find_equal_scalars(this_branch, src_reg); |
8357 | find_equal_scalars(other_branch, &other_branch_regs[insn->src_reg]); | |
8358 | } | |
8359 | ||
f1174f77 EC |
8360 | } |
8361 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 8362 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
8363 | dst_reg, insn->imm, (u32)insn->imm, |
8364 | opcode, is_jmp32); | |
48461135 JB |
8365 | } |
8366 | ||
e688c3db AS |
8367 | if (dst_reg->type == SCALAR_VALUE && dst_reg->id && |
8368 | !WARN_ON_ONCE(dst_reg->id != other_branch_regs[insn->dst_reg].id)) { | |
75748837 AS |
8369 | find_equal_scalars(this_branch, dst_reg); |
8370 | find_equal_scalars(other_branch, &other_branch_regs[insn->dst_reg]); | |
8371 | } | |
8372 | ||
092ed096 JW |
8373 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
8374 | * NOTE: these optimizations below are related with pointer comparison | |
8375 | * which will never be JMP32. | |
8376 | */ | |
8377 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 8378 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
840b9615 JS |
8379 | reg_type_may_be_null(dst_reg->type)) { |
8380 | /* Mark all identical registers in each branch as either | |
57a09bf0 TG |
8381 | * safe or unknown depending R == 0 or R != 0 conditional. |
8382 | */ | |
840b9615 JS |
8383 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
8384 | opcode == BPF_JNE); | |
8385 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
8386 | opcode == BPF_JEQ); | |
5beca081 DB |
8387 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
8388 | this_branch, other_branch) && | |
8389 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
8390 | verbose(env, "R%d pointer comparison prohibited\n", |
8391 | insn->dst_reg); | |
1be7f75d | 8392 | return -EACCES; |
17a52670 | 8393 | } |
06ee7115 | 8394 | if (env->log.level & BPF_LOG_LEVEL) |
f4d7e40a | 8395 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
8396 | return 0; |
8397 | } | |
8398 | ||
17a52670 | 8399 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 8400 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 8401 | { |
d8eca5bb | 8402 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 8403 | struct bpf_reg_state *regs = cur_regs(env); |
4976b718 | 8404 | struct bpf_reg_state *dst_reg; |
d8eca5bb | 8405 | struct bpf_map *map; |
17a52670 AS |
8406 | int err; |
8407 | ||
8408 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 8409 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
8410 | return -EINVAL; |
8411 | } | |
8412 | if (insn->off != 0) { | |
61bd5218 | 8413 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
8414 | return -EINVAL; |
8415 | } | |
8416 | ||
dc503a8a | 8417 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
8418 | if (err) |
8419 | return err; | |
8420 | ||
4976b718 | 8421 | dst_reg = ®s[insn->dst_reg]; |
6b173873 | 8422 | if (insn->src_reg == 0) { |
6b173873 JK |
8423 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
8424 | ||
4976b718 | 8425 | dst_reg->type = SCALAR_VALUE; |
b03c9f9f | 8426 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 8427 | return 0; |
6b173873 | 8428 | } |
17a52670 | 8429 | |
4976b718 HL |
8430 | if (insn->src_reg == BPF_PSEUDO_BTF_ID) { |
8431 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
8432 | ||
8433 | dst_reg->type = aux->btf_var.reg_type; | |
8434 | switch (dst_reg->type) { | |
8435 | case PTR_TO_MEM: | |
8436 | dst_reg->mem_size = aux->btf_var.mem_size; | |
8437 | break; | |
8438 | case PTR_TO_BTF_ID: | |
eaa6bcb7 | 8439 | case PTR_TO_PERCPU_BTF_ID: |
22dc4a0f | 8440 | dst_reg->btf = aux->btf_var.btf; |
4976b718 HL |
8441 | dst_reg->btf_id = aux->btf_var.btf_id; |
8442 | break; | |
8443 | default: | |
8444 | verbose(env, "bpf verifier is misconfigured\n"); | |
8445 | return -EFAULT; | |
8446 | } | |
8447 | return 0; | |
8448 | } | |
8449 | ||
69c087ba YS |
8450 | if (insn->src_reg == BPF_PSEUDO_FUNC) { |
8451 | struct bpf_prog_aux *aux = env->prog->aux; | |
8452 | u32 subprogno = insn[1].imm; | |
8453 | ||
8454 | if (!aux->func_info) { | |
8455 | verbose(env, "missing btf func_info\n"); | |
8456 | return -EINVAL; | |
8457 | } | |
8458 | if (aux->func_info_aux[subprogno].linkage != BTF_FUNC_STATIC) { | |
8459 | verbose(env, "callback function not static\n"); | |
8460 | return -EINVAL; | |
8461 | } | |
8462 | ||
8463 | dst_reg->type = PTR_TO_FUNC; | |
8464 | dst_reg->subprogno = subprogno; | |
8465 | return 0; | |
8466 | } | |
8467 | ||
d8eca5bb DB |
8468 | map = env->used_maps[aux->map_index]; |
8469 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
4976b718 | 8470 | dst_reg->map_ptr = map; |
d8eca5bb DB |
8471 | |
8472 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE) { | |
4976b718 HL |
8473 | dst_reg->type = PTR_TO_MAP_VALUE; |
8474 | dst_reg->off = aux->map_off; | |
d8eca5bb | 8475 | if (map_value_has_spin_lock(map)) |
4976b718 | 8476 | dst_reg->id = ++env->id_gen; |
d8eca5bb | 8477 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD) { |
4976b718 | 8478 | dst_reg->type = CONST_PTR_TO_MAP; |
d8eca5bb DB |
8479 | } else { |
8480 | verbose(env, "bpf verifier is misconfigured\n"); | |
8481 | return -EINVAL; | |
8482 | } | |
17a52670 | 8483 | |
17a52670 AS |
8484 | return 0; |
8485 | } | |
8486 | ||
96be4325 DB |
8487 | static bool may_access_skb(enum bpf_prog_type type) |
8488 | { | |
8489 | switch (type) { | |
8490 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
8491 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 8492 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
8493 | return true; |
8494 | default: | |
8495 | return false; | |
8496 | } | |
8497 | } | |
8498 | ||
ddd872bc AS |
8499 | /* verify safety of LD_ABS|LD_IND instructions: |
8500 | * - they can only appear in the programs where ctx == skb | |
8501 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
8502 | * preserve R6-R9, and store return value into R0 | |
8503 | * | |
8504 | * Implicit input: | |
8505 | * ctx == skb == R6 == CTX | |
8506 | * | |
8507 | * Explicit input: | |
8508 | * SRC == any register | |
8509 | * IMM == 32-bit immediate | |
8510 | * | |
8511 | * Output: | |
8512 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
8513 | */ | |
58e2af8b | 8514 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 8515 | { |
638f5b90 | 8516 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 8517 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 8518 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
8519 | int i, err; |
8520 | ||
7e40781c | 8521 | if (!may_access_skb(resolve_prog_type(env->prog))) { |
61bd5218 | 8522 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
8523 | return -EINVAL; |
8524 | } | |
8525 | ||
e0cea7ce DB |
8526 | if (!env->ops->gen_ld_abs) { |
8527 | verbose(env, "bpf verifier is misconfigured\n"); | |
8528 | return -EINVAL; | |
8529 | } | |
8530 | ||
ddd872bc | 8531 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 8532 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 8533 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 8534 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
8535 | return -EINVAL; |
8536 | } | |
8537 | ||
8538 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 8539 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
8540 | if (err) |
8541 | return err; | |
8542 | ||
fd978bf7 JS |
8543 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
8544 | * gen_ld_abs() may terminate the program at runtime, leading to | |
8545 | * reference leak. | |
8546 | */ | |
8547 | err = check_reference_leak(env); | |
8548 | if (err) { | |
8549 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
8550 | return err; | |
8551 | } | |
8552 | ||
d83525ca AS |
8553 | if (env->cur_state->active_spin_lock) { |
8554 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); | |
8555 | return -EINVAL; | |
8556 | } | |
8557 | ||
6d4f151a | 8558 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
8559 | verbose(env, |
8560 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
8561 | return -EINVAL; |
8562 | } | |
8563 | ||
8564 | if (mode == BPF_IND) { | |
8565 | /* check explicit source operand */ | |
dc503a8a | 8566 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
8567 | if (err) |
8568 | return err; | |
8569 | } | |
8570 | ||
6d4f151a DB |
8571 | err = check_ctx_reg(env, ®s[ctx_reg], ctx_reg); |
8572 | if (err < 0) | |
8573 | return err; | |
8574 | ||
ddd872bc | 8575 | /* reset caller saved regs to unreadable */ |
dc503a8a | 8576 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 8577 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
8578 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
8579 | } | |
ddd872bc AS |
8580 | |
8581 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
8582 | * the value fetched from the packet. |
8583 | * Already marked as written above. | |
ddd872bc | 8584 | */ |
61bd5218 | 8585 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
8586 | /* ld_abs load up to 32-bit skb data. */ |
8587 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
8588 | return 0; |
8589 | } | |
8590 | ||
390ee7e2 AS |
8591 | static int check_return_code(struct bpf_verifier_env *env) |
8592 | { | |
5cf1e914 | 8593 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 8594 | const struct bpf_prog *prog = env->prog; |
390ee7e2 AS |
8595 | struct bpf_reg_state *reg; |
8596 | struct tnum range = tnum_range(0, 1); | |
7e40781c | 8597 | enum bpf_prog_type prog_type = resolve_prog_type(env->prog); |
27ae7997 | 8598 | int err; |
f782e2c3 | 8599 | const bool is_subprog = env->cur_state->frame[0]->subprogno; |
27ae7997 | 8600 | |
9e4e01df | 8601 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
f782e2c3 DB |
8602 | if (!is_subprog && |
8603 | (prog_type == BPF_PROG_TYPE_STRUCT_OPS || | |
7e40781c | 8604 | prog_type == BPF_PROG_TYPE_LSM) && |
27ae7997 MKL |
8605 | !prog->aux->attach_func_proto->type) |
8606 | return 0; | |
8607 | ||
8608 | /* eBPF calling convetion is such that R0 is used | |
8609 | * to return the value from eBPF program. | |
8610 | * Make sure that it's readable at this time | |
8611 | * of bpf_exit, which means that program wrote | |
8612 | * something into it earlier | |
8613 | */ | |
8614 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
8615 | if (err) | |
8616 | return err; | |
8617 | ||
8618 | if (is_pointer_value(env, BPF_REG_0)) { | |
8619 | verbose(env, "R0 leaks addr as return value\n"); | |
8620 | return -EACCES; | |
8621 | } | |
390ee7e2 | 8622 | |
f782e2c3 DB |
8623 | reg = cur_regs(env) + BPF_REG_0; |
8624 | if (is_subprog) { | |
8625 | if (reg->type != SCALAR_VALUE) { | |
8626 | verbose(env, "At subprogram exit the register R0 is not a scalar value (%s)\n", | |
8627 | reg_type_str[reg->type]); | |
8628 | return -EINVAL; | |
8629 | } | |
8630 | return 0; | |
8631 | } | |
8632 | ||
7e40781c | 8633 | switch (prog_type) { |
983695fa DB |
8634 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
8635 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 DB |
8636 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
8637 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || | |
8638 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
8639 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || | |
8640 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME) | |
983695fa | 8641 | range = tnum_range(1, 1); |
77241217 SF |
8642 | if (env->prog->expected_attach_type == BPF_CGROUP_INET4_BIND || |
8643 | env->prog->expected_attach_type == BPF_CGROUP_INET6_BIND) | |
8644 | range = tnum_range(0, 3); | |
ed4ed404 | 8645 | break; |
390ee7e2 | 8646 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 8647 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
8648 | range = tnum_range(0, 3); | |
8649 | enforce_attach_type_range = tnum_range(2, 3); | |
8650 | } | |
ed4ed404 | 8651 | break; |
390ee7e2 AS |
8652 | case BPF_PROG_TYPE_CGROUP_SOCK: |
8653 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 8654 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 8655 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 8656 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 8657 | break; |
15ab09bd AS |
8658 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
8659 | if (!env->prog->aux->attach_btf_id) | |
8660 | return 0; | |
8661 | range = tnum_const(0); | |
8662 | break; | |
15d83c4d | 8663 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
8664 | switch (env->prog->expected_attach_type) { |
8665 | case BPF_TRACE_FENTRY: | |
8666 | case BPF_TRACE_FEXIT: | |
8667 | range = tnum_const(0); | |
8668 | break; | |
8669 | case BPF_TRACE_RAW_TP: | |
8670 | case BPF_MODIFY_RETURN: | |
15d83c4d | 8671 | return 0; |
2ec0616e DB |
8672 | case BPF_TRACE_ITER: |
8673 | break; | |
e92888c7 YS |
8674 | default: |
8675 | return -ENOTSUPP; | |
8676 | } | |
15d83c4d | 8677 | break; |
e9ddbb77 JS |
8678 | case BPF_PROG_TYPE_SK_LOOKUP: |
8679 | range = tnum_range(SK_DROP, SK_PASS); | |
8680 | break; | |
e92888c7 YS |
8681 | case BPF_PROG_TYPE_EXT: |
8682 | /* freplace program can return anything as its return value | |
8683 | * depends on the to-be-replaced kernel func or bpf program. | |
8684 | */ | |
390ee7e2 AS |
8685 | default: |
8686 | return 0; | |
8687 | } | |
8688 | ||
390ee7e2 | 8689 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 8690 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
8691 | reg_type_str[reg->type]); |
8692 | return -EINVAL; | |
8693 | } | |
8694 | ||
8695 | if (!tnum_in(range, reg->var_off)) { | |
bc2591d6 | 8696 | verbose_invalid_scalar(env, reg, &range, "program exit", "R0"); |
390ee7e2 AS |
8697 | return -EINVAL; |
8698 | } | |
5cf1e914 | 8699 | |
8700 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
8701 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
8702 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
8703 | return 0; |
8704 | } | |
8705 | ||
475fb78f AS |
8706 | /* non-recursive DFS pseudo code |
8707 | * 1 procedure DFS-iterative(G,v): | |
8708 | * 2 label v as discovered | |
8709 | * 3 let S be a stack | |
8710 | * 4 S.push(v) | |
8711 | * 5 while S is not empty | |
8712 | * 6 t <- S.pop() | |
8713 | * 7 if t is what we're looking for: | |
8714 | * 8 return t | |
8715 | * 9 for all edges e in G.adjacentEdges(t) do | |
8716 | * 10 if edge e is already labelled | |
8717 | * 11 continue with the next edge | |
8718 | * 12 w <- G.adjacentVertex(t,e) | |
8719 | * 13 if vertex w is not discovered and not explored | |
8720 | * 14 label e as tree-edge | |
8721 | * 15 label w as discovered | |
8722 | * 16 S.push(w) | |
8723 | * 17 continue at 5 | |
8724 | * 18 else if vertex w is discovered | |
8725 | * 19 label e as back-edge | |
8726 | * 20 else | |
8727 | * 21 // vertex w is explored | |
8728 | * 22 label e as forward- or cross-edge | |
8729 | * 23 label t as explored | |
8730 | * 24 S.pop() | |
8731 | * | |
8732 | * convention: | |
8733 | * 0x10 - discovered | |
8734 | * 0x11 - discovered and fall-through edge labelled | |
8735 | * 0x12 - discovered and fall-through and branch edges labelled | |
8736 | * 0x20 - explored | |
8737 | */ | |
8738 | ||
8739 | enum { | |
8740 | DISCOVERED = 0x10, | |
8741 | EXPLORED = 0x20, | |
8742 | FALLTHROUGH = 1, | |
8743 | BRANCH = 2, | |
8744 | }; | |
8745 | ||
dc2a4ebc AS |
8746 | static u32 state_htab_size(struct bpf_verifier_env *env) |
8747 | { | |
8748 | return env->prog->len; | |
8749 | } | |
8750 | ||
5d839021 AS |
8751 | static struct bpf_verifier_state_list **explored_state( |
8752 | struct bpf_verifier_env *env, | |
8753 | int idx) | |
8754 | { | |
dc2a4ebc AS |
8755 | struct bpf_verifier_state *cur = env->cur_state; |
8756 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
8757 | ||
8758 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
8759 | } |
8760 | ||
8761 | static void init_explored_state(struct bpf_verifier_env *env, int idx) | |
8762 | { | |
a8f500af | 8763 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 8764 | } |
f1bca824 | 8765 | |
59e2e27d WAF |
8766 | enum { |
8767 | DONE_EXPLORING = 0, | |
8768 | KEEP_EXPLORING = 1, | |
8769 | }; | |
8770 | ||
475fb78f AS |
8771 | /* t, w, e - match pseudo-code above: |
8772 | * t - index of current instruction | |
8773 | * w - next instruction | |
8774 | * e - edge | |
8775 | */ | |
2589726d AS |
8776 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env, |
8777 | bool loop_ok) | |
475fb78f | 8778 | { |
7df737e9 AS |
8779 | int *insn_stack = env->cfg.insn_stack; |
8780 | int *insn_state = env->cfg.insn_state; | |
8781 | ||
475fb78f | 8782 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
59e2e27d | 8783 | return DONE_EXPLORING; |
475fb78f AS |
8784 | |
8785 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
59e2e27d | 8786 | return DONE_EXPLORING; |
475fb78f AS |
8787 | |
8788 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 8789 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 8790 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
8791 | return -EINVAL; |
8792 | } | |
8793 | ||
f1bca824 AS |
8794 | if (e == BRANCH) |
8795 | /* mark branch target for state pruning */ | |
5d839021 | 8796 | init_explored_state(env, w); |
f1bca824 | 8797 | |
475fb78f AS |
8798 | if (insn_state[w] == 0) { |
8799 | /* tree-edge */ | |
8800 | insn_state[t] = DISCOVERED | e; | |
8801 | insn_state[w] = DISCOVERED; | |
7df737e9 | 8802 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 8803 | return -E2BIG; |
7df737e9 | 8804 | insn_stack[env->cfg.cur_stack++] = w; |
59e2e27d | 8805 | return KEEP_EXPLORING; |
475fb78f | 8806 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { |
2c78ee89 | 8807 | if (loop_ok && env->bpf_capable) |
59e2e27d | 8808 | return DONE_EXPLORING; |
d9762e84 MKL |
8809 | verbose_linfo(env, t, "%d: ", t); |
8810 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 8811 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
8812 | return -EINVAL; |
8813 | } else if (insn_state[w] == EXPLORED) { | |
8814 | /* forward- or cross-edge */ | |
8815 | insn_state[t] = DISCOVERED | e; | |
8816 | } else { | |
61bd5218 | 8817 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
8818 | return -EFAULT; |
8819 | } | |
59e2e27d WAF |
8820 | return DONE_EXPLORING; |
8821 | } | |
8822 | ||
efdb22de YS |
8823 | static int visit_func_call_insn(int t, int insn_cnt, |
8824 | struct bpf_insn *insns, | |
8825 | struct bpf_verifier_env *env, | |
8826 | bool visit_callee) | |
8827 | { | |
8828 | int ret; | |
8829 | ||
8830 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); | |
8831 | if (ret) | |
8832 | return ret; | |
8833 | ||
8834 | if (t + 1 < insn_cnt) | |
8835 | init_explored_state(env, t + 1); | |
8836 | if (visit_callee) { | |
8837 | init_explored_state(env, t); | |
8838 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, | |
8839 | env, false); | |
8840 | } | |
8841 | return ret; | |
8842 | } | |
8843 | ||
59e2e27d WAF |
8844 | /* Visits the instruction at index t and returns one of the following: |
8845 | * < 0 - an error occurred | |
8846 | * DONE_EXPLORING - the instruction was fully explored | |
8847 | * KEEP_EXPLORING - there is still work to be done before it is fully explored | |
8848 | */ | |
8849 | static int visit_insn(int t, int insn_cnt, struct bpf_verifier_env *env) | |
8850 | { | |
8851 | struct bpf_insn *insns = env->prog->insnsi; | |
8852 | int ret; | |
8853 | ||
69c087ba YS |
8854 | if (bpf_pseudo_func(insns + t)) |
8855 | return visit_func_call_insn(t, insn_cnt, insns, env, true); | |
8856 | ||
59e2e27d WAF |
8857 | /* All non-branch instructions have a single fall-through edge. */ |
8858 | if (BPF_CLASS(insns[t].code) != BPF_JMP && | |
8859 | BPF_CLASS(insns[t].code) != BPF_JMP32) | |
8860 | return push_insn(t, t + 1, FALLTHROUGH, env, false); | |
8861 | ||
8862 | switch (BPF_OP(insns[t].code)) { | |
8863 | case BPF_EXIT: | |
8864 | return DONE_EXPLORING; | |
8865 | ||
8866 | case BPF_CALL: | |
efdb22de YS |
8867 | return visit_func_call_insn(t, insn_cnt, insns, env, |
8868 | insns[t].src_reg == BPF_PSEUDO_CALL); | |
59e2e27d WAF |
8869 | |
8870 | case BPF_JA: | |
8871 | if (BPF_SRC(insns[t].code) != BPF_K) | |
8872 | return -EINVAL; | |
8873 | ||
8874 | /* unconditional jump with single edge */ | |
8875 | ret = push_insn(t, t + insns[t].off + 1, FALLTHROUGH, env, | |
8876 | true); | |
8877 | if (ret) | |
8878 | return ret; | |
8879 | ||
8880 | /* unconditional jmp is not a good pruning point, | |
8881 | * but it's marked, since backtracking needs | |
8882 | * to record jmp history in is_state_visited(). | |
8883 | */ | |
8884 | init_explored_state(env, t + insns[t].off + 1); | |
8885 | /* tell verifier to check for equivalent states | |
8886 | * after every call and jump | |
8887 | */ | |
8888 | if (t + 1 < insn_cnt) | |
8889 | init_explored_state(env, t + 1); | |
8890 | ||
8891 | return ret; | |
8892 | ||
8893 | default: | |
8894 | /* conditional jump with two edges */ | |
8895 | init_explored_state(env, t); | |
8896 | ret = push_insn(t, t + 1, FALLTHROUGH, env, true); | |
8897 | if (ret) | |
8898 | return ret; | |
8899 | ||
8900 | return push_insn(t, t + insns[t].off + 1, BRANCH, env, true); | |
8901 | } | |
475fb78f AS |
8902 | } |
8903 | ||
8904 | /* non-recursive depth-first-search to detect loops in BPF program | |
8905 | * loop == back-edge in directed graph | |
8906 | */ | |
58e2af8b | 8907 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f | 8908 | { |
475fb78f | 8909 | int insn_cnt = env->prog->len; |
7df737e9 | 8910 | int *insn_stack, *insn_state; |
475fb78f | 8911 | int ret = 0; |
59e2e27d | 8912 | int i; |
475fb78f | 8913 | |
7df737e9 | 8914 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
8915 | if (!insn_state) |
8916 | return -ENOMEM; | |
8917 | ||
7df737e9 | 8918 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 8919 | if (!insn_stack) { |
71dde681 | 8920 | kvfree(insn_state); |
475fb78f AS |
8921 | return -ENOMEM; |
8922 | } | |
8923 | ||
8924 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
8925 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 8926 | env->cfg.cur_stack = 1; |
475fb78f | 8927 | |
59e2e27d WAF |
8928 | while (env->cfg.cur_stack > 0) { |
8929 | int t = insn_stack[env->cfg.cur_stack - 1]; | |
475fb78f | 8930 | |
59e2e27d WAF |
8931 | ret = visit_insn(t, insn_cnt, env); |
8932 | switch (ret) { | |
8933 | case DONE_EXPLORING: | |
8934 | insn_state[t] = EXPLORED; | |
8935 | env->cfg.cur_stack--; | |
8936 | break; | |
8937 | case KEEP_EXPLORING: | |
8938 | break; | |
8939 | default: | |
8940 | if (ret > 0) { | |
8941 | verbose(env, "visit_insn internal bug\n"); | |
8942 | ret = -EFAULT; | |
475fb78f | 8943 | } |
475fb78f | 8944 | goto err_free; |
59e2e27d | 8945 | } |
475fb78f AS |
8946 | } |
8947 | ||
59e2e27d | 8948 | if (env->cfg.cur_stack < 0) { |
61bd5218 | 8949 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
8950 | ret = -EFAULT; |
8951 | goto err_free; | |
8952 | } | |
475fb78f | 8953 | |
475fb78f AS |
8954 | for (i = 0; i < insn_cnt; i++) { |
8955 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 8956 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
8957 | ret = -EINVAL; |
8958 | goto err_free; | |
8959 | } | |
8960 | } | |
8961 | ret = 0; /* cfg looks good */ | |
8962 | ||
8963 | err_free: | |
71dde681 AS |
8964 | kvfree(insn_state); |
8965 | kvfree(insn_stack); | |
7df737e9 | 8966 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
8967 | return ret; |
8968 | } | |
8969 | ||
09b28d76 AS |
8970 | static int check_abnormal_return(struct bpf_verifier_env *env) |
8971 | { | |
8972 | int i; | |
8973 | ||
8974 | for (i = 1; i < env->subprog_cnt; i++) { | |
8975 | if (env->subprog_info[i].has_ld_abs) { | |
8976 | verbose(env, "LD_ABS is not allowed in subprogs without BTF\n"); | |
8977 | return -EINVAL; | |
8978 | } | |
8979 | if (env->subprog_info[i].has_tail_call) { | |
8980 | verbose(env, "tail_call is not allowed in subprogs without BTF\n"); | |
8981 | return -EINVAL; | |
8982 | } | |
8983 | } | |
8984 | return 0; | |
8985 | } | |
8986 | ||
838e9690 YS |
8987 | /* The minimum supported BTF func info size */ |
8988 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
8989 | #define MAX_FUNCINFO_REC_SIZE 252 | |
8990 | ||
c454a46b MKL |
8991 | static int check_btf_func(struct bpf_verifier_env *env, |
8992 | const union bpf_attr *attr, | |
8993 | union bpf_attr __user *uattr) | |
838e9690 | 8994 | { |
09b28d76 | 8995 | const struct btf_type *type, *func_proto, *ret_type; |
d0b2818e | 8996 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 8997 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 8998 | struct bpf_func_info *krecord; |
8c1b6e69 | 8999 | struct bpf_func_info_aux *info_aux = NULL; |
c454a46b MKL |
9000 | struct bpf_prog *prog; |
9001 | const struct btf *btf; | |
838e9690 | 9002 | void __user *urecord; |
d0b2818e | 9003 | u32 prev_offset = 0; |
09b28d76 | 9004 | bool scalar_return; |
e7ed83d6 | 9005 | int ret = -ENOMEM; |
838e9690 YS |
9006 | |
9007 | nfuncs = attr->func_info_cnt; | |
09b28d76 AS |
9008 | if (!nfuncs) { |
9009 | if (check_abnormal_return(env)) | |
9010 | return -EINVAL; | |
838e9690 | 9011 | return 0; |
09b28d76 | 9012 | } |
838e9690 YS |
9013 | |
9014 | if (nfuncs != env->subprog_cnt) { | |
9015 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
9016 | return -EINVAL; | |
9017 | } | |
9018 | ||
9019 | urec_size = attr->func_info_rec_size; | |
9020 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
9021 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
9022 | urec_size % sizeof(u32)) { | |
9023 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
9024 | return -EINVAL; | |
9025 | } | |
9026 | ||
c454a46b MKL |
9027 | prog = env->prog; |
9028 | btf = prog->aux->btf; | |
838e9690 YS |
9029 | |
9030 | urecord = u64_to_user_ptr(attr->func_info); | |
9031 | min_size = min_t(u32, krec_size, urec_size); | |
9032 | ||
ba64e7d8 | 9033 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
9034 | if (!krecord) |
9035 | return -ENOMEM; | |
8c1b6e69 AS |
9036 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); |
9037 | if (!info_aux) | |
9038 | goto err_free; | |
ba64e7d8 | 9039 | |
838e9690 YS |
9040 | for (i = 0; i < nfuncs; i++) { |
9041 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
9042 | if (ret) { | |
9043 | if (ret == -E2BIG) { | |
9044 | verbose(env, "nonzero tailing record in func info"); | |
9045 | /* set the size kernel expects so loader can zero | |
9046 | * out the rest of the record. | |
9047 | */ | |
9048 | if (put_user(min_size, &uattr->func_info_rec_size)) | |
9049 | ret = -EFAULT; | |
9050 | } | |
c454a46b | 9051 | goto err_free; |
838e9690 YS |
9052 | } |
9053 | ||
ba64e7d8 | 9054 | if (copy_from_user(&krecord[i], urecord, min_size)) { |
838e9690 | 9055 | ret = -EFAULT; |
c454a46b | 9056 | goto err_free; |
838e9690 YS |
9057 | } |
9058 | ||
d30d42e0 | 9059 | /* check insn_off */ |
09b28d76 | 9060 | ret = -EINVAL; |
838e9690 | 9061 | if (i == 0) { |
d30d42e0 | 9062 | if (krecord[i].insn_off) { |
838e9690 | 9063 | verbose(env, |
d30d42e0 MKL |
9064 | "nonzero insn_off %u for the first func info record", |
9065 | krecord[i].insn_off); | |
c454a46b | 9066 | goto err_free; |
838e9690 | 9067 | } |
d30d42e0 | 9068 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
9069 | verbose(env, |
9070 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 9071 | krecord[i].insn_off, prev_offset); |
c454a46b | 9072 | goto err_free; |
838e9690 YS |
9073 | } |
9074 | ||
d30d42e0 | 9075 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 | 9076 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
c454a46b | 9077 | goto err_free; |
838e9690 YS |
9078 | } |
9079 | ||
9080 | /* check type_id */ | |
ba64e7d8 | 9081 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 9082 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 9083 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 9084 | krecord[i].type_id); |
c454a46b | 9085 | goto err_free; |
838e9690 | 9086 | } |
51c39bb1 | 9087 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); |
09b28d76 AS |
9088 | |
9089 | func_proto = btf_type_by_id(btf, type->type); | |
9090 | if (unlikely(!func_proto || !btf_type_is_func_proto(func_proto))) | |
9091 | /* btf_func_check() already verified it during BTF load */ | |
9092 | goto err_free; | |
9093 | ret_type = btf_type_skip_modifiers(btf, func_proto->type, NULL); | |
9094 | scalar_return = | |
9095 | btf_type_is_small_int(ret_type) || btf_type_is_enum(ret_type); | |
9096 | if (i && !scalar_return && env->subprog_info[i].has_ld_abs) { | |
9097 | verbose(env, "LD_ABS is only allowed in functions that return 'int'.\n"); | |
9098 | goto err_free; | |
9099 | } | |
9100 | if (i && !scalar_return && env->subprog_info[i].has_tail_call) { | |
9101 | verbose(env, "tail_call is only allowed in functions that return 'int'.\n"); | |
9102 | goto err_free; | |
9103 | } | |
9104 | ||
d30d42e0 | 9105 | prev_offset = krecord[i].insn_off; |
838e9690 YS |
9106 | urecord += urec_size; |
9107 | } | |
9108 | ||
ba64e7d8 YS |
9109 | prog->aux->func_info = krecord; |
9110 | prog->aux->func_info_cnt = nfuncs; | |
8c1b6e69 | 9111 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
9112 | return 0; |
9113 | ||
c454a46b | 9114 | err_free: |
ba64e7d8 | 9115 | kvfree(krecord); |
8c1b6e69 | 9116 | kfree(info_aux); |
838e9690 YS |
9117 | return ret; |
9118 | } | |
9119 | ||
ba64e7d8 YS |
9120 | static void adjust_btf_func(struct bpf_verifier_env *env) |
9121 | { | |
8c1b6e69 | 9122 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
9123 | int i; |
9124 | ||
8c1b6e69 | 9125 | if (!aux->func_info) |
ba64e7d8 YS |
9126 | return; |
9127 | ||
9128 | for (i = 0; i < env->subprog_cnt; i++) | |
8c1b6e69 | 9129 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
9130 | } |
9131 | ||
c454a46b MKL |
9132 | #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \ |
9133 | sizeof(((struct bpf_line_info *)(0))->line_col)) | |
9134 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE | |
9135 | ||
9136 | static int check_btf_line(struct bpf_verifier_env *env, | |
9137 | const union bpf_attr *attr, | |
9138 | union bpf_attr __user *uattr) | |
9139 | { | |
9140 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
9141 | struct bpf_subprog_info *sub; | |
9142 | struct bpf_line_info *linfo; | |
9143 | struct bpf_prog *prog; | |
9144 | const struct btf *btf; | |
9145 | void __user *ulinfo; | |
9146 | int err; | |
9147 | ||
9148 | nr_linfo = attr->line_info_cnt; | |
9149 | if (!nr_linfo) | |
9150 | return 0; | |
9151 | ||
9152 | rec_size = attr->line_info_rec_size; | |
9153 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
9154 | rec_size > MAX_LINEINFO_REC_SIZE || | |
9155 | rec_size & (sizeof(u32) - 1)) | |
9156 | return -EINVAL; | |
9157 | ||
9158 | /* Need to zero it in case the userspace may | |
9159 | * pass in a smaller bpf_line_info object. | |
9160 | */ | |
9161 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
9162 | GFP_KERNEL | __GFP_NOWARN); | |
9163 | if (!linfo) | |
9164 | return -ENOMEM; | |
9165 | ||
9166 | prog = env->prog; | |
9167 | btf = prog->aux->btf; | |
9168 | ||
9169 | s = 0; | |
9170 | sub = env->subprog_info; | |
9171 | ulinfo = u64_to_user_ptr(attr->line_info); | |
9172 | expected_size = sizeof(struct bpf_line_info); | |
9173 | ncopy = min_t(u32, expected_size, rec_size); | |
9174 | for (i = 0; i < nr_linfo; i++) { | |
9175 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
9176 | if (err) { | |
9177 | if (err == -E2BIG) { | |
9178 | verbose(env, "nonzero tailing record in line_info"); | |
9179 | if (put_user(expected_size, | |
9180 | &uattr->line_info_rec_size)) | |
9181 | err = -EFAULT; | |
9182 | } | |
9183 | goto err_free; | |
9184 | } | |
9185 | ||
9186 | if (copy_from_user(&linfo[i], ulinfo, ncopy)) { | |
9187 | err = -EFAULT; | |
9188 | goto err_free; | |
9189 | } | |
9190 | ||
9191 | /* | |
9192 | * Check insn_off to ensure | |
9193 | * 1) strictly increasing AND | |
9194 | * 2) bounded by prog->len | |
9195 | * | |
9196 | * The linfo[0].insn_off == 0 check logically falls into | |
9197 | * the later "missing bpf_line_info for func..." case | |
9198 | * because the first linfo[0].insn_off must be the | |
9199 | * first sub also and the first sub must have | |
9200 | * subprog_info[0].start == 0. | |
9201 | */ | |
9202 | if ((i && linfo[i].insn_off <= prev_offset) || | |
9203 | linfo[i].insn_off >= prog->len) { | |
9204 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
9205 | i, linfo[i].insn_off, prev_offset, | |
9206 | prog->len); | |
9207 | err = -EINVAL; | |
9208 | goto err_free; | |
9209 | } | |
9210 | ||
fdbaa0be MKL |
9211 | if (!prog->insnsi[linfo[i].insn_off].code) { |
9212 | verbose(env, | |
9213 | "Invalid insn code at line_info[%u].insn_off\n", | |
9214 | i); | |
9215 | err = -EINVAL; | |
9216 | goto err_free; | |
9217 | } | |
9218 | ||
23127b33 MKL |
9219 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
9220 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
9221 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
9222 | err = -EINVAL; | |
9223 | goto err_free; | |
9224 | } | |
9225 | ||
9226 | if (s != env->subprog_cnt) { | |
9227 | if (linfo[i].insn_off == sub[s].start) { | |
9228 | sub[s].linfo_idx = i; | |
9229 | s++; | |
9230 | } else if (sub[s].start < linfo[i].insn_off) { | |
9231 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
9232 | err = -EINVAL; | |
9233 | goto err_free; | |
9234 | } | |
9235 | } | |
9236 | ||
9237 | prev_offset = linfo[i].insn_off; | |
9238 | ulinfo += rec_size; | |
9239 | } | |
9240 | ||
9241 | if (s != env->subprog_cnt) { | |
9242 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
9243 | env->subprog_cnt - s, s); | |
9244 | err = -EINVAL; | |
9245 | goto err_free; | |
9246 | } | |
9247 | ||
9248 | prog->aux->linfo = linfo; | |
9249 | prog->aux->nr_linfo = nr_linfo; | |
9250 | ||
9251 | return 0; | |
9252 | ||
9253 | err_free: | |
9254 | kvfree(linfo); | |
9255 | return err; | |
9256 | } | |
9257 | ||
9258 | static int check_btf_info(struct bpf_verifier_env *env, | |
9259 | const union bpf_attr *attr, | |
9260 | union bpf_attr __user *uattr) | |
9261 | { | |
9262 | struct btf *btf; | |
9263 | int err; | |
9264 | ||
09b28d76 AS |
9265 | if (!attr->func_info_cnt && !attr->line_info_cnt) { |
9266 | if (check_abnormal_return(env)) | |
9267 | return -EINVAL; | |
c454a46b | 9268 | return 0; |
09b28d76 | 9269 | } |
c454a46b MKL |
9270 | |
9271 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
9272 | if (IS_ERR(btf)) | |
9273 | return PTR_ERR(btf); | |
350a5c4d AS |
9274 | if (btf_is_kernel(btf)) { |
9275 | btf_put(btf); | |
9276 | return -EACCES; | |
9277 | } | |
c454a46b MKL |
9278 | env->prog->aux->btf = btf; |
9279 | ||
9280 | err = check_btf_func(env, attr, uattr); | |
9281 | if (err) | |
9282 | return err; | |
9283 | ||
9284 | err = check_btf_line(env, attr, uattr); | |
9285 | if (err) | |
9286 | return err; | |
9287 | ||
9288 | return 0; | |
ba64e7d8 YS |
9289 | } |
9290 | ||
f1174f77 EC |
9291 | /* check %cur's range satisfies %old's */ |
9292 | static bool range_within(struct bpf_reg_state *old, | |
9293 | struct bpf_reg_state *cur) | |
9294 | { | |
b03c9f9f EC |
9295 | return old->umin_value <= cur->umin_value && |
9296 | old->umax_value >= cur->umax_value && | |
9297 | old->smin_value <= cur->smin_value && | |
fd675184 DB |
9298 | old->smax_value >= cur->smax_value && |
9299 | old->u32_min_value <= cur->u32_min_value && | |
9300 | old->u32_max_value >= cur->u32_max_value && | |
9301 | old->s32_min_value <= cur->s32_min_value && | |
9302 | old->s32_max_value >= cur->s32_max_value; | |
f1174f77 EC |
9303 | } |
9304 | ||
9305 | /* Maximum number of register states that can exist at once */ | |
9306 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
9307 | struct idpair { | |
9308 | u32 old; | |
9309 | u32 cur; | |
9310 | }; | |
9311 | ||
9312 | /* If in the old state two registers had the same id, then they need to have | |
9313 | * the same id in the new state as well. But that id could be different from | |
9314 | * the old state, so we need to track the mapping from old to new ids. | |
9315 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
9316 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
9317 | * regs with a different old id could still have new id 9, we don't care about | |
9318 | * that. | |
9319 | * So we look through our idmap to see if this old id has been seen before. If | |
9320 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 9321 | */ |
f1174f77 | 9322 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 9323 | { |
f1174f77 | 9324 | unsigned int i; |
969bf05e | 9325 | |
f1174f77 EC |
9326 | for (i = 0; i < ID_MAP_SIZE; i++) { |
9327 | if (!idmap[i].old) { | |
9328 | /* Reached an empty slot; haven't seen this id before */ | |
9329 | idmap[i].old = old_id; | |
9330 | idmap[i].cur = cur_id; | |
9331 | return true; | |
9332 | } | |
9333 | if (idmap[i].old == old_id) | |
9334 | return idmap[i].cur == cur_id; | |
9335 | } | |
9336 | /* We ran out of idmap slots, which should be impossible */ | |
9337 | WARN_ON_ONCE(1); | |
9338 | return false; | |
9339 | } | |
9340 | ||
9242b5f5 AS |
9341 | static void clean_func_state(struct bpf_verifier_env *env, |
9342 | struct bpf_func_state *st) | |
9343 | { | |
9344 | enum bpf_reg_liveness live; | |
9345 | int i, j; | |
9346 | ||
9347 | for (i = 0; i < BPF_REG_FP; i++) { | |
9348 | live = st->regs[i].live; | |
9349 | /* liveness must not touch this register anymore */ | |
9350 | st->regs[i].live |= REG_LIVE_DONE; | |
9351 | if (!(live & REG_LIVE_READ)) | |
9352 | /* since the register is unused, clear its state | |
9353 | * to make further comparison simpler | |
9354 | */ | |
f54c7898 | 9355 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
9356 | } |
9357 | ||
9358 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
9359 | live = st->stack[i].spilled_ptr.live; | |
9360 | /* liveness must not touch this stack slot anymore */ | |
9361 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
9362 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 9363 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
9364 | for (j = 0; j < BPF_REG_SIZE; j++) |
9365 | st->stack[i].slot_type[j] = STACK_INVALID; | |
9366 | } | |
9367 | } | |
9368 | } | |
9369 | ||
9370 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
9371 | struct bpf_verifier_state *st) | |
9372 | { | |
9373 | int i; | |
9374 | ||
9375 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
9376 | /* all regs in this state in all frames were already marked */ | |
9377 | return; | |
9378 | ||
9379 | for (i = 0; i <= st->curframe; i++) | |
9380 | clean_func_state(env, st->frame[i]); | |
9381 | } | |
9382 | ||
9383 | /* the parentage chains form a tree. | |
9384 | * the verifier states are added to state lists at given insn and | |
9385 | * pushed into state stack for future exploration. | |
9386 | * when the verifier reaches bpf_exit insn some of the verifer states | |
9387 | * stored in the state lists have their final liveness state already, | |
9388 | * but a lot of states will get revised from liveness point of view when | |
9389 | * the verifier explores other branches. | |
9390 | * Example: | |
9391 | * 1: r0 = 1 | |
9392 | * 2: if r1 == 100 goto pc+1 | |
9393 | * 3: r0 = 2 | |
9394 | * 4: exit | |
9395 | * when the verifier reaches exit insn the register r0 in the state list of | |
9396 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
9397 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
9398 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
9399 | * | |
9400 | * Since the verifier pushes the branch states as it sees them while exploring | |
9401 | * the program the condition of walking the branch instruction for the second | |
9402 | * time means that all states below this branch were already explored and | |
9403 | * their final liveness markes are already propagated. | |
9404 | * Hence when the verifier completes the search of state list in is_state_visited() | |
9405 | * we can call this clean_live_states() function to mark all liveness states | |
9406 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
9407 | * will not be used. | |
9408 | * This function also clears the registers and stack for states that !READ | |
9409 | * to simplify state merging. | |
9410 | * | |
9411 | * Important note here that walking the same branch instruction in the callee | |
9412 | * doesn't meant that the states are DONE. The verifier has to compare | |
9413 | * the callsites | |
9414 | */ | |
9415 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
9416 | struct bpf_verifier_state *cur) | |
9417 | { | |
9418 | struct bpf_verifier_state_list *sl; | |
9419 | int i; | |
9420 | ||
5d839021 | 9421 | sl = *explored_state(env, insn); |
a8f500af | 9422 | while (sl) { |
2589726d AS |
9423 | if (sl->state.branches) |
9424 | goto next; | |
dc2a4ebc AS |
9425 | if (sl->state.insn_idx != insn || |
9426 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
9427 | goto next; |
9428 | for (i = 0; i <= cur->curframe; i++) | |
9429 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
9430 | goto next; | |
9431 | clean_verifier_state(env, &sl->state); | |
9432 | next: | |
9433 | sl = sl->next; | |
9434 | } | |
9435 | } | |
9436 | ||
f1174f77 | 9437 | /* Returns true if (rold safe implies rcur safe) */ |
1b688a19 EC |
9438 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
9439 | struct idpair *idmap) | |
f1174f77 | 9440 | { |
f4d7e40a AS |
9441 | bool equal; |
9442 | ||
dc503a8a EC |
9443 | if (!(rold->live & REG_LIVE_READ)) |
9444 | /* explored state didn't use this */ | |
9445 | return true; | |
9446 | ||
679c782d | 9447 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
9448 | |
9449 | if (rold->type == PTR_TO_STACK) | |
9450 | /* two stack pointers are equal only if they're pointing to | |
9451 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
9452 | */ | |
9453 | return equal && rold->frameno == rcur->frameno; | |
9454 | ||
9455 | if (equal) | |
969bf05e AS |
9456 | return true; |
9457 | ||
f1174f77 EC |
9458 | if (rold->type == NOT_INIT) |
9459 | /* explored state can't have used this */ | |
969bf05e | 9460 | return true; |
f1174f77 EC |
9461 | if (rcur->type == NOT_INIT) |
9462 | return false; | |
9463 | switch (rold->type) { | |
9464 | case SCALAR_VALUE: | |
9465 | if (rcur->type == SCALAR_VALUE) { | |
b5dc0163 AS |
9466 | if (!rold->precise && !rcur->precise) |
9467 | return true; | |
f1174f77 EC |
9468 | /* new val must satisfy old val knowledge */ |
9469 | return range_within(rold, rcur) && | |
9470 | tnum_in(rold->var_off, rcur->var_off); | |
9471 | } else { | |
179d1c56 JH |
9472 | /* We're trying to use a pointer in place of a scalar. |
9473 | * Even if the scalar was unbounded, this could lead to | |
9474 | * pointer leaks because scalars are allowed to leak | |
9475 | * while pointers are not. We could make this safe in | |
9476 | * special cases if root is calling us, but it's | |
9477 | * probably not worth the hassle. | |
f1174f77 | 9478 | */ |
179d1c56 | 9479 | return false; |
f1174f77 | 9480 | } |
69c087ba | 9481 | case PTR_TO_MAP_KEY: |
f1174f77 | 9482 | case PTR_TO_MAP_VALUE: |
1b688a19 EC |
9483 | /* If the new min/max/var_off satisfy the old ones and |
9484 | * everything else matches, we are OK. | |
d83525ca AS |
9485 | * 'id' is not compared, since it's only used for maps with |
9486 | * bpf_spin_lock inside map element and in such cases if | |
9487 | * the rest of the prog is valid for one map element then | |
9488 | * it's valid for all map elements regardless of the key | |
9489 | * used in bpf_map_lookup() | |
1b688a19 EC |
9490 | */ |
9491 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
9492 | range_within(rold, rcur) && | |
9493 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
9494 | case PTR_TO_MAP_VALUE_OR_NULL: |
9495 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
9496 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
9497 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
9498 | * checked, doing so could have affected others with the same | |
9499 | * id, and we can't check for that because we lost the id when | |
9500 | * we converted to a PTR_TO_MAP_VALUE. | |
9501 | */ | |
9502 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
9503 | return false; | |
9504 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
9505 | return false; | |
9506 | /* Check our ids match any regs they're supposed to */ | |
9507 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 9508 | case PTR_TO_PACKET_META: |
f1174f77 | 9509 | case PTR_TO_PACKET: |
de8f3a83 | 9510 | if (rcur->type != rold->type) |
f1174f77 EC |
9511 | return false; |
9512 | /* We must have at least as much range as the old ptr | |
9513 | * did, so that any accesses which were safe before are | |
9514 | * still safe. This is true even if old range < old off, | |
9515 | * since someone could have accessed through (ptr - k), or | |
9516 | * even done ptr -= k in a register, to get a safe access. | |
9517 | */ | |
9518 | if (rold->range > rcur->range) | |
9519 | return false; | |
9520 | /* If the offsets don't match, we can't trust our alignment; | |
9521 | * nor can we be sure that we won't fall out of range. | |
9522 | */ | |
9523 | if (rold->off != rcur->off) | |
9524 | return false; | |
9525 | /* id relations must be preserved */ | |
9526 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
9527 | return false; | |
9528 | /* new val must satisfy old val knowledge */ | |
9529 | return range_within(rold, rcur) && | |
9530 | tnum_in(rold->var_off, rcur->var_off); | |
9531 | case PTR_TO_CTX: | |
9532 | case CONST_PTR_TO_MAP: | |
f1174f77 | 9533 | case PTR_TO_PACKET_END: |
d58e468b | 9534 | case PTR_TO_FLOW_KEYS: |
c64b7983 JS |
9535 | case PTR_TO_SOCKET: |
9536 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
9537 | case PTR_TO_SOCK_COMMON: |
9538 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
9539 | case PTR_TO_TCP_SOCK: |
9540 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 9541 | case PTR_TO_XDP_SOCK: |
f1174f77 EC |
9542 | /* Only valid matches are exact, which memcmp() above |
9543 | * would have accepted | |
9544 | */ | |
9545 | default: | |
9546 | /* Don't know what's going on, just say it's not safe */ | |
9547 | return false; | |
9548 | } | |
969bf05e | 9549 | |
f1174f77 EC |
9550 | /* Shouldn't get here; if we do, say it's not safe */ |
9551 | WARN_ON_ONCE(1); | |
969bf05e AS |
9552 | return false; |
9553 | } | |
9554 | ||
f4d7e40a AS |
9555 | static bool stacksafe(struct bpf_func_state *old, |
9556 | struct bpf_func_state *cur, | |
638f5b90 AS |
9557 | struct idpair *idmap) |
9558 | { | |
9559 | int i, spi; | |
9560 | ||
638f5b90 AS |
9561 | /* walk slots of the explored stack and ignore any additional |
9562 | * slots in the current stack, since explored(safe) state | |
9563 | * didn't use them | |
9564 | */ | |
9565 | for (i = 0; i < old->allocated_stack; i++) { | |
9566 | spi = i / BPF_REG_SIZE; | |
9567 | ||
b233920c AS |
9568 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
9569 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 9570 | /* explored state didn't use this */ |
fd05e57b | 9571 | continue; |
b233920c | 9572 | } |
cc2b14d5 | 9573 | |
638f5b90 AS |
9574 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
9575 | continue; | |
19e2dbb7 AS |
9576 | |
9577 | /* explored stack has more populated slots than current stack | |
9578 | * and these slots were used | |
9579 | */ | |
9580 | if (i >= cur->allocated_stack) | |
9581 | return false; | |
9582 | ||
cc2b14d5 AS |
9583 | /* if old state was safe with misc data in the stack |
9584 | * it will be safe with zero-initialized stack. | |
9585 | * The opposite is not true | |
9586 | */ | |
9587 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
9588 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
9589 | continue; | |
638f5b90 AS |
9590 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
9591 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
9592 | /* Ex: old explored (safe) state has STACK_SPILL in | |
b8c1a309 | 9593 | * this stack slot, but current has STACK_MISC -> |
638f5b90 AS |
9594 | * this verifier states are not equivalent, |
9595 | * return false to continue verification of this path | |
9596 | */ | |
9597 | return false; | |
9598 | if (i % BPF_REG_SIZE) | |
9599 | continue; | |
9600 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
9601 | continue; | |
9602 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
9603 | &cur->stack[spi].spilled_ptr, | |
9604 | idmap)) | |
9605 | /* when explored and current stack slot are both storing | |
9606 | * spilled registers, check that stored pointers types | |
9607 | * are the same as well. | |
9608 | * Ex: explored safe path could have stored | |
9609 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
9610 | * but current path has stored: | |
9611 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
9612 | * such verifier states are not equivalent. | |
9613 | * return false to continue verification of this path | |
9614 | */ | |
9615 | return false; | |
9616 | } | |
9617 | return true; | |
9618 | } | |
9619 | ||
fd978bf7 JS |
9620 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
9621 | { | |
9622 | if (old->acquired_refs != cur->acquired_refs) | |
9623 | return false; | |
9624 | return !memcmp(old->refs, cur->refs, | |
9625 | sizeof(*old->refs) * old->acquired_refs); | |
9626 | } | |
9627 | ||
f1bca824 AS |
9628 | /* compare two verifier states |
9629 | * | |
9630 | * all states stored in state_list are known to be valid, since | |
9631 | * verifier reached 'bpf_exit' instruction through them | |
9632 | * | |
9633 | * this function is called when verifier exploring different branches of | |
9634 | * execution popped from the state stack. If it sees an old state that has | |
9635 | * more strict register state and more strict stack state then this execution | |
9636 | * branch doesn't need to be explored further, since verifier already | |
9637 | * concluded that more strict state leads to valid finish. | |
9638 | * | |
9639 | * Therefore two states are equivalent if register state is more conservative | |
9640 | * and explored stack state is more conservative than the current one. | |
9641 | * Example: | |
9642 | * explored current | |
9643 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
9644 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
9645 | * | |
9646 | * In other words if current stack state (one being explored) has more | |
9647 | * valid slots than old one that already passed validation, it means | |
9648 | * the verifier can stop exploring and conclude that current state is valid too | |
9649 | * | |
9650 | * Similarly with registers. If explored state has register type as invalid | |
9651 | * whereas register type in current state is meaningful, it means that | |
9652 | * the current state will reach 'bpf_exit' instruction safely | |
9653 | */ | |
f4d7e40a AS |
9654 | static bool func_states_equal(struct bpf_func_state *old, |
9655 | struct bpf_func_state *cur) | |
f1bca824 | 9656 | { |
f1174f77 EC |
9657 | struct idpair *idmap; |
9658 | bool ret = false; | |
f1bca824 AS |
9659 | int i; |
9660 | ||
f1174f77 EC |
9661 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
9662 | /* If we failed to allocate the idmap, just say it's not safe */ | |
9663 | if (!idmap) | |
1a0dc1ac | 9664 | return false; |
f1174f77 EC |
9665 | |
9666 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 9667 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 9668 | goto out_free; |
f1bca824 AS |
9669 | } |
9670 | ||
638f5b90 AS |
9671 | if (!stacksafe(old, cur, idmap)) |
9672 | goto out_free; | |
fd978bf7 JS |
9673 | |
9674 | if (!refsafe(old, cur)) | |
9675 | goto out_free; | |
f1174f77 EC |
9676 | ret = true; |
9677 | out_free: | |
9678 | kfree(idmap); | |
9679 | return ret; | |
f1bca824 AS |
9680 | } |
9681 | ||
f4d7e40a AS |
9682 | static bool states_equal(struct bpf_verifier_env *env, |
9683 | struct bpf_verifier_state *old, | |
9684 | struct bpf_verifier_state *cur) | |
9685 | { | |
9686 | int i; | |
9687 | ||
9688 | if (old->curframe != cur->curframe) | |
9689 | return false; | |
9690 | ||
979d63d5 DB |
9691 | /* Verification state from speculative execution simulation |
9692 | * must never prune a non-speculative execution one. | |
9693 | */ | |
9694 | if (old->speculative && !cur->speculative) | |
9695 | return false; | |
9696 | ||
d83525ca AS |
9697 | if (old->active_spin_lock != cur->active_spin_lock) |
9698 | return false; | |
9699 | ||
f4d7e40a AS |
9700 | /* for states to be equal callsites have to be the same |
9701 | * and all frame states need to be equivalent | |
9702 | */ | |
9703 | for (i = 0; i <= old->curframe; i++) { | |
9704 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
9705 | return false; | |
9706 | if (!func_states_equal(old->frame[i], cur->frame[i])) | |
9707 | return false; | |
9708 | } | |
9709 | return true; | |
9710 | } | |
9711 | ||
5327ed3d JW |
9712 | /* Return 0 if no propagation happened. Return negative error code if error |
9713 | * happened. Otherwise, return the propagated bit. | |
9714 | */ | |
55e7f3b5 JW |
9715 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
9716 | struct bpf_reg_state *reg, | |
9717 | struct bpf_reg_state *parent_reg) | |
9718 | { | |
5327ed3d JW |
9719 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
9720 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
9721 | int err; |
9722 | ||
5327ed3d JW |
9723 | /* When comes here, read flags of PARENT_REG or REG could be any of |
9724 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
9725 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
9726 | */ | |
9727 | if (parent_flag == REG_LIVE_READ64 || | |
9728 | /* Or if there is no read flag from REG. */ | |
9729 | !flag || | |
9730 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
9731 | parent_flag == flag) | |
55e7f3b5 JW |
9732 | return 0; |
9733 | ||
5327ed3d | 9734 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
9735 | if (err) |
9736 | return err; | |
9737 | ||
5327ed3d | 9738 | return flag; |
55e7f3b5 JW |
9739 | } |
9740 | ||
8e9cd9ce | 9741 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
9742 | * straight-line code between a state and its parent. When we arrive at an |
9743 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
9744 | * code, so read marks in the state must propagate to the parent regardless | |
9745 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 9746 | * in mark_reg_read() is for. |
8e9cd9ce | 9747 | */ |
f4d7e40a AS |
9748 | static int propagate_liveness(struct bpf_verifier_env *env, |
9749 | const struct bpf_verifier_state *vstate, | |
9750 | struct bpf_verifier_state *vparent) | |
dc503a8a | 9751 | { |
3f8cafa4 | 9752 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 9753 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 9754 | int i, frame, err = 0; |
dc503a8a | 9755 | |
f4d7e40a AS |
9756 | if (vparent->curframe != vstate->curframe) { |
9757 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
9758 | vparent->curframe, vstate->curframe); | |
9759 | return -EFAULT; | |
9760 | } | |
dc503a8a EC |
9761 | /* Propagate read liveness of registers... */ |
9762 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 9763 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
9764 | parent = vparent->frame[frame]; |
9765 | state = vstate->frame[frame]; | |
9766 | parent_reg = parent->regs; | |
9767 | state_reg = state->regs; | |
83d16312 JK |
9768 | /* We don't need to worry about FP liveness, it's read-only */ |
9769 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
9770 | err = propagate_liveness_reg(env, &state_reg[i], |
9771 | &parent_reg[i]); | |
5327ed3d | 9772 | if (err < 0) |
3f8cafa4 | 9773 | return err; |
5327ed3d JW |
9774 | if (err == REG_LIVE_READ64) |
9775 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 9776 | } |
f4d7e40a | 9777 | |
1b04aee7 | 9778 | /* Propagate stack slots. */ |
f4d7e40a AS |
9779 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
9780 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
9781 | parent_reg = &parent->stack[i].spilled_ptr; |
9782 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
9783 | err = propagate_liveness_reg(env, state_reg, |
9784 | parent_reg); | |
5327ed3d | 9785 | if (err < 0) |
3f8cafa4 | 9786 | return err; |
dc503a8a EC |
9787 | } |
9788 | } | |
5327ed3d | 9789 | return 0; |
dc503a8a EC |
9790 | } |
9791 | ||
a3ce685d AS |
9792 | /* find precise scalars in the previous equivalent state and |
9793 | * propagate them into the current state | |
9794 | */ | |
9795 | static int propagate_precision(struct bpf_verifier_env *env, | |
9796 | const struct bpf_verifier_state *old) | |
9797 | { | |
9798 | struct bpf_reg_state *state_reg; | |
9799 | struct bpf_func_state *state; | |
9800 | int i, err = 0; | |
9801 | ||
9802 | state = old->frame[old->curframe]; | |
9803 | state_reg = state->regs; | |
9804 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { | |
9805 | if (state_reg->type != SCALAR_VALUE || | |
9806 | !state_reg->precise) | |
9807 | continue; | |
9808 | if (env->log.level & BPF_LOG_LEVEL2) | |
9809 | verbose(env, "propagating r%d\n", i); | |
9810 | err = mark_chain_precision(env, i); | |
9811 | if (err < 0) | |
9812 | return err; | |
9813 | } | |
9814 | ||
9815 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
9816 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
9817 | continue; | |
9818 | state_reg = &state->stack[i].spilled_ptr; | |
9819 | if (state_reg->type != SCALAR_VALUE || | |
9820 | !state_reg->precise) | |
9821 | continue; | |
9822 | if (env->log.level & BPF_LOG_LEVEL2) | |
9823 | verbose(env, "propagating fp%d\n", | |
9824 | (-i - 1) * BPF_REG_SIZE); | |
9825 | err = mark_chain_precision_stack(env, i); | |
9826 | if (err < 0) | |
9827 | return err; | |
9828 | } | |
9829 | return 0; | |
9830 | } | |
9831 | ||
2589726d AS |
9832 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
9833 | struct bpf_verifier_state *cur) | |
9834 | { | |
9835 | struct bpf_func_state *fold, *fcur; | |
9836 | int i, fr = cur->curframe; | |
9837 | ||
9838 | if (old->curframe != fr) | |
9839 | return false; | |
9840 | ||
9841 | fold = old->frame[fr]; | |
9842 | fcur = cur->frame[fr]; | |
9843 | for (i = 0; i < MAX_BPF_REG; i++) | |
9844 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
9845 | offsetof(struct bpf_reg_state, parent))) | |
9846 | return false; | |
9847 | return true; | |
9848 | } | |
9849 | ||
9850 | ||
58e2af8b | 9851 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 9852 | { |
58e2af8b | 9853 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 9854 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 9855 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 9856 | int i, j, err, states_cnt = 0; |
10d274e8 | 9857 | bool add_new_state = env->test_state_freq ? true : false; |
f1bca824 | 9858 | |
b5dc0163 | 9859 | cur->last_insn_idx = env->prev_insn_idx; |
a8f500af | 9860 | if (!env->insn_aux_data[insn_idx].prune_point) |
f1bca824 AS |
9861 | /* this 'insn_idx' instruction wasn't marked, so we will not |
9862 | * be doing state search here | |
9863 | */ | |
9864 | return 0; | |
9865 | ||
2589726d AS |
9866 | /* bpf progs typically have pruning point every 4 instructions |
9867 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
9868 | * Do not add new state for future pruning if the verifier hasn't seen | |
9869 | * at least 2 jumps and at least 8 instructions. | |
9870 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
9871 | * In tests that amounts to up to 50% reduction into total verifier | |
9872 | * memory consumption and 20% verifier time speedup. | |
9873 | */ | |
9874 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
9875 | env->insn_processed - env->prev_insn_processed >= 8) | |
9876 | add_new_state = true; | |
9877 | ||
a8f500af AS |
9878 | pprev = explored_state(env, insn_idx); |
9879 | sl = *pprev; | |
9880 | ||
9242b5f5 AS |
9881 | clean_live_states(env, insn_idx, cur); |
9882 | ||
a8f500af | 9883 | while (sl) { |
dc2a4ebc AS |
9884 | states_cnt++; |
9885 | if (sl->state.insn_idx != insn_idx) | |
9886 | goto next; | |
2589726d AS |
9887 | if (sl->state.branches) { |
9888 | if (states_maybe_looping(&sl->state, cur) && | |
9889 | states_equal(env, &sl->state, cur)) { | |
9890 | verbose_linfo(env, insn_idx, "; "); | |
9891 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
9892 | return -EINVAL; | |
9893 | } | |
9894 | /* if the verifier is processing a loop, avoid adding new state | |
9895 | * too often, since different loop iterations have distinct | |
9896 | * states and may not help future pruning. | |
9897 | * This threshold shouldn't be too low to make sure that | |
9898 | * a loop with large bound will be rejected quickly. | |
9899 | * The most abusive loop will be: | |
9900 | * r1 += 1 | |
9901 | * if r1 < 1000000 goto pc-2 | |
9902 | * 1M insn_procssed limit / 100 == 10k peak states. | |
9903 | * This threshold shouldn't be too high either, since states | |
9904 | * at the end of the loop are likely to be useful in pruning. | |
9905 | */ | |
9906 | if (env->jmps_processed - env->prev_jmps_processed < 20 && | |
9907 | env->insn_processed - env->prev_insn_processed < 100) | |
9908 | add_new_state = false; | |
9909 | goto miss; | |
9910 | } | |
638f5b90 | 9911 | if (states_equal(env, &sl->state, cur)) { |
9f4686c4 | 9912 | sl->hit_cnt++; |
f1bca824 | 9913 | /* reached equivalent register/stack state, |
dc503a8a EC |
9914 | * prune the search. |
9915 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
9916 | * If we have any write marks in env->cur_state, they |
9917 | * will prevent corresponding reads in the continuation | |
9918 | * from reaching our parent (an explored_state). Our | |
9919 | * own state will get the read marks recorded, but | |
9920 | * they'll be immediately forgotten as we're pruning | |
9921 | * this state and will pop a new one. | |
f1bca824 | 9922 | */ |
f4d7e40a | 9923 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
9924 | |
9925 | /* if previous state reached the exit with precision and | |
9926 | * current state is equivalent to it (except precsion marks) | |
9927 | * the precision needs to be propagated back in | |
9928 | * the current state. | |
9929 | */ | |
9930 | err = err ? : push_jmp_history(env, cur); | |
9931 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
9932 | if (err) |
9933 | return err; | |
f1bca824 | 9934 | return 1; |
dc503a8a | 9935 | } |
2589726d AS |
9936 | miss: |
9937 | /* when new state is not going to be added do not increase miss count. | |
9938 | * Otherwise several loop iterations will remove the state | |
9939 | * recorded earlier. The goal of these heuristics is to have | |
9940 | * states from some iterations of the loop (some in the beginning | |
9941 | * and some at the end) to help pruning. | |
9942 | */ | |
9943 | if (add_new_state) | |
9944 | sl->miss_cnt++; | |
9f4686c4 AS |
9945 | /* heuristic to determine whether this state is beneficial |
9946 | * to keep checking from state equivalence point of view. | |
9947 | * Higher numbers increase max_states_per_insn and verification time, | |
9948 | * but do not meaningfully decrease insn_processed. | |
9949 | */ | |
9950 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
9951 | /* the state is unlikely to be useful. Remove it to | |
9952 | * speed up verification | |
9953 | */ | |
9954 | *pprev = sl->next; | |
9955 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
2589726d AS |
9956 | u32 br = sl->state.branches; |
9957 | ||
9958 | WARN_ONCE(br, | |
9959 | "BUG live_done but branches_to_explore %d\n", | |
9960 | br); | |
9f4686c4 AS |
9961 | free_verifier_state(&sl->state, false); |
9962 | kfree(sl); | |
9963 | env->peak_states--; | |
9964 | } else { | |
9965 | /* cannot free this state, since parentage chain may | |
9966 | * walk it later. Add it for free_list instead to | |
9967 | * be freed at the end of verification | |
9968 | */ | |
9969 | sl->next = env->free_list; | |
9970 | env->free_list = sl; | |
9971 | } | |
9972 | sl = *pprev; | |
9973 | continue; | |
9974 | } | |
dc2a4ebc | 9975 | next: |
9f4686c4 AS |
9976 | pprev = &sl->next; |
9977 | sl = *pprev; | |
f1bca824 AS |
9978 | } |
9979 | ||
06ee7115 AS |
9980 | if (env->max_states_per_insn < states_cnt) |
9981 | env->max_states_per_insn = states_cnt; | |
9982 | ||
2c78ee89 | 9983 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
b5dc0163 | 9984 | return push_jmp_history(env, cur); |
ceefbc96 | 9985 | |
2589726d | 9986 | if (!add_new_state) |
b5dc0163 | 9987 | return push_jmp_history(env, cur); |
ceefbc96 | 9988 | |
2589726d AS |
9989 | /* There were no equivalent states, remember the current one. |
9990 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 9991 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 9992 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 9993 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
9994 | * again on the way to bpf_exit. |
9995 | * When looping the sl->state.branches will be > 0 and this state | |
9996 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 9997 | */ |
638f5b90 | 9998 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
9999 | if (!new_sl) |
10000 | return -ENOMEM; | |
06ee7115 AS |
10001 | env->total_states++; |
10002 | env->peak_states++; | |
2589726d AS |
10003 | env->prev_jmps_processed = env->jmps_processed; |
10004 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 AS |
10005 | |
10006 | /* add new state to the head of linked list */ | |
679c782d EC |
10007 | new = &new_sl->state; |
10008 | err = copy_verifier_state(new, cur); | |
1969db47 | 10009 | if (err) { |
679c782d | 10010 | free_verifier_state(new, false); |
1969db47 AS |
10011 | kfree(new_sl); |
10012 | return err; | |
10013 | } | |
dc2a4ebc | 10014 | new->insn_idx = insn_idx; |
2589726d AS |
10015 | WARN_ONCE(new->branches != 1, |
10016 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 10017 | |
2589726d | 10018 | cur->parent = new; |
b5dc0163 AS |
10019 | cur->first_insn_idx = insn_idx; |
10020 | clear_jmp_history(cur); | |
5d839021 AS |
10021 | new_sl->next = *explored_state(env, insn_idx); |
10022 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
10023 | /* connect new state to parentage chain. Current frame needs all |
10024 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
10025 | * to the stack implicitly by JITs) so in callers' frames connect just | |
10026 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
10027 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
10028 | * from callee with its full parentage chain, anyway. | |
10029 | */ | |
8e9cd9ce EC |
10030 | /* clear write marks in current state: the writes we did are not writes |
10031 | * our child did, so they don't screen off its reads from us. | |
10032 | * (There are no read marks in current state, because reads always mark | |
10033 | * their parent and current state never has children yet. Only | |
10034 | * explored_states can get read marks.) | |
10035 | */ | |
eea1c227 AS |
10036 | for (j = 0; j <= cur->curframe; j++) { |
10037 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
10038 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
10039 | for (i = 0; i < BPF_REG_FP; i++) | |
10040 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
10041 | } | |
f4d7e40a AS |
10042 | |
10043 | /* all stack frames are accessible from callee, clear them all */ | |
10044 | for (j = 0; j <= cur->curframe; j++) { | |
10045 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 10046 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 10047 | |
679c782d | 10048 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 10049 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
10050 | frame->stack[i].spilled_ptr.parent = |
10051 | &newframe->stack[i].spilled_ptr; | |
10052 | } | |
f4d7e40a | 10053 | } |
f1bca824 AS |
10054 | return 0; |
10055 | } | |
10056 | ||
c64b7983 JS |
10057 | /* Return true if it's OK to have the same insn return a different type. */ |
10058 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
10059 | { | |
10060 | switch (type) { | |
10061 | case PTR_TO_CTX: | |
10062 | case PTR_TO_SOCKET: | |
10063 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
10064 | case PTR_TO_SOCK_COMMON: |
10065 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
10066 | case PTR_TO_TCP_SOCK: |
10067 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 10068 | case PTR_TO_XDP_SOCK: |
2a02759e | 10069 | case PTR_TO_BTF_ID: |
b121b341 | 10070 | case PTR_TO_BTF_ID_OR_NULL: |
c64b7983 JS |
10071 | return false; |
10072 | default: | |
10073 | return true; | |
10074 | } | |
10075 | } | |
10076 | ||
10077 | /* If an instruction was previously used with particular pointer types, then we | |
10078 | * need to be careful to avoid cases such as the below, where it may be ok | |
10079 | * for one branch accessing the pointer, but not ok for the other branch: | |
10080 | * | |
10081 | * R1 = sock_ptr | |
10082 | * goto X; | |
10083 | * ... | |
10084 | * R1 = some_other_valid_ptr; | |
10085 | * goto X; | |
10086 | * ... | |
10087 | * R2 = *(u32 *)(R1 + 0); | |
10088 | */ | |
10089 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
10090 | { | |
10091 | return src != prev && (!reg_type_mismatch_ok(src) || | |
10092 | !reg_type_mismatch_ok(prev)); | |
10093 | } | |
10094 | ||
58e2af8b | 10095 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 10096 | { |
6f8a57cc | 10097 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 10098 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 10099 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 10100 | struct bpf_reg_state *regs; |
06ee7115 | 10101 | int insn_cnt = env->prog->len; |
17a52670 | 10102 | bool do_print_state = false; |
b5dc0163 | 10103 | int prev_insn_idx = -1; |
17a52670 | 10104 | |
17a52670 AS |
10105 | for (;;) { |
10106 | struct bpf_insn *insn; | |
10107 | u8 class; | |
10108 | int err; | |
10109 | ||
b5dc0163 | 10110 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 10111 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 10112 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 10113 | env->insn_idx, insn_cnt); |
17a52670 AS |
10114 | return -EFAULT; |
10115 | } | |
10116 | ||
c08435ec | 10117 | insn = &insns[env->insn_idx]; |
17a52670 AS |
10118 | class = BPF_CLASS(insn->code); |
10119 | ||
06ee7115 | 10120 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
10121 | verbose(env, |
10122 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 10123 | env->insn_processed); |
17a52670 AS |
10124 | return -E2BIG; |
10125 | } | |
10126 | ||
c08435ec | 10127 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
10128 | if (err < 0) |
10129 | return err; | |
10130 | if (err == 1) { | |
10131 | /* found equivalent state, can prune the search */ | |
06ee7115 | 10132 | if (env->log.level & BPF_LOG_LEVEL) { |
f1bca824 | 10133 | if (do_print_state) |
979d63d5 DB |
10134 | verbose(env, "\nfrom %d to %d%s: safe\n", |
10135 | env->prev_insn_idx, env->insn_idx, | |
10136 | env->cur_state->speculative ? | |
10137 | " (speculative execution)" : ""); | |
f1bca824 | 10138 | else |
c08435ec | 10139 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
10140 | } |
10141 | goto process_bpf_exit; | |
10142 | } | |
10143 | ||
c3494801 AS |
10144 | if (signal_pending(current)) |
10145 | return -EAGAIN; | |
10146 | ||
3c2ce60b DB |
10147 | if (need_resched()) |
10148 | cond_resched(); | |
10149 | ||
06ee7115 AS |
10150 | if (env->log.level & BPF_LOG_LEVEL2 || |
10151 | (env->log.level & BPF_LOG_LEVEL && do_print_state)) { | |
10152 | if (env->log.level & BPF_LOG_LEVEL2) | |
c08435ec | 10153 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 10154 | else |
979d63d5 DB |
10155 | verbose(env, "\nfrom %d to %d%s:", |
10156 | env->prev_insn_idx, env->insn_idx, | |
10157 | env->cur_state->speculative ? | |
10158 | " (speculative execution)" : ""); | |
f4d7e40a | 10159 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
10160 | do_print_state = false; |
10161 | } | |
10162 | ||
06ee7115 | 10163 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 DB |
10164 | const struct bpf_insn_cbs cbs = { |
10165 | .cb_print = verbose, | |
abe08840 | 10166 | .private_data = env, |
7105e828 DB |
10167 | }; |
10168 | ||
c08435ec DB |
10169 | verbose_linfo(env, env->insn_idx, "; "); |
10170 | verbose(env, "%d: ", env->insn_idx); | |
abe08840 | 10171 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
10172 | } |
10173 | ||
cae1927c | 10174 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
10175 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
10176 | env->prev_insn_idx); | |
cae1927c JK |
10177 | if (err) |
10178 | return err; | |
10179 | } | |
13a27dfc | 10180 | |
638f5b90 | 10181 | regs = cur_regs(env); |
51c39bb1 | 10182 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; |
b5dc0163 | 10183 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 10184 | |
17a52670 | 10185 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 10186 | err = check_alu_op(env, insn); |
17a52670 AS |
10187 | if (err) |
10188 | return err; | |
10189 | ||
10190 | } else if (class == BPF_LDX) { | |
3df126f3 | 10191 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
10192 | |
10193 | /* check for reserved fields is already done */ | |
10194 | ||
17a52670 | 10195 | /* check src operand */ |
dc503a8a | 10196 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10197 | if (err) |
10198 | return err; | |
10199 | ||
dc503a8a | 10200 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
10201 | if (err) |
10202 | return err; | |
10203 | ||
725f9dcd AS |
10204 | src_reg_type = regs[insn->src_reg].type; |
10205 | ||
17a52670 AS |
10206 | /* check that memory (src_reg + off) is readable, |
10207 | * the state of dst_reg will be updated by this func | |
10208 | */ | |
c08435ec DB |
10209 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
10210 | insn->off, BPF_SIZE(insn->code), | |
10211 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
10212 | if (err) |
10213 | return err; | |
10214 | ||
c08435ec | 10215 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
10216 | |
10217 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
10218 | /* saw a valid insn |
10219 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 10220 | * save type to validate intersecting paths |
9bac3d6d | 10221 | */ |
3df126f3 | 10222 | *prev_src_type = src_reg_type; |
9bac3d6d | 10223 | |
c64b7983 | 10224 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
10225 | /* ABuser program is trying to use the same insn |
10226 | * dst_reg = *(u32*) (src_reg + off) | |
10227 | * with different pointer types: | |
10228 | * src_reg == ctx in one branch and | |
10229 | * src_reg == stack|map in some other branch. | |
10230 | * Reject it. | |
10231 | */ | |
61bd5218 | 10232 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
10233 | return -EINVAL; |
10234 | } | |
10235 | ||
17a52670 | 10236 | } else if (class == BPF_STX) { |
3df126f3 | 10237 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 10238 | |
91c960b0 BJ |
10239 | if (BPF_MODE(insn->code) == BPF_ATOMIC) { |
10240 | err = check_atomic(env, env->insn_idx, insn); | |
17a52670 AS |
10241 | if (err) |
10242 | return err; | |
c08435ec | 10243 | env->insn_idx++; |
17a52670 AS |
10244 | continue; |
10245 | } | |
10246 | ||
5ca419f2 BJ |
10247 | if (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0) { |
10248 | verbose(env, "BPF_STX uses reserved fields\n"); | |
10249 | return -EINVAL; | |
10250 | } | |
10251 | ||
17a52670 | 10252 | /* check src1 operand */ |
dc503a8a | 10253 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
10254 | if (err) |
10255 | return err; | |
10256 | /* check src2 operand */ | |
dc503a8a | 10257 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10258 | if (err) |
10259 | return err; | |
10260 | ||
d691f9e8 AS |
10261 | dst_reg_type = regs[insn->dst_reg].type; |
10262 | ||
17a52670 | 10263 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
10264 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
10265 | insn->off, BPF_SIZE(insn->code), | |
10266 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
10267 | if (err) |
10268 | return err; | |
10269 | ||
c08435ec | 10270 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
10271 | |
10272 | if (*prev_dst_type == NOT_INIT) { | |
10273 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 10274 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 10275 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
10276 | return -EINVAL; |
10277 | } | |
10278 | ||
17a52670 AS |
10279 | } else if (class == BPF_ST) { |
10280 | if (BPF_MODE(insn->code) != BPF_MEM || | |
10281 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 10282 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
10283 | return -EINVAL; |
10284 | } | |
10285 | /* check src operand */ | |
dc503a8a | 10286 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
10287 | if (err) |
10288 | return err; | |
10289 | ||
f37a8cb8 | 10290 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 10291 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f DB |
10292 | insn->dst_reg, |
10293 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
10294 | return -EACCES; |
10295 | } | |
10296 | ||
17a52670 | 10297 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
10298 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
10299 | insn->off, BPF_SIZE(insn->code), | |
10300 | BPF_WRITE, -1, false); | |
17a52670 AS |
10301 | if (err) |
10302 | return err; | |
10303 | ||
092ed096 | 10304 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
10305 | u8 opcode = BPF_OP(insn->code); |
10306 | ||
2589726d | 10307 | env->jmps_processed++; |
17a52670 AS |
10308 | if (opcode == BPF_CALL) { |
10309 | if (BPF_SRC(insn->code) != BPF_K || | |
10310 | insn->off != 0 || | |
f4d7e40a AS |
10311 | (insn->src_reg != BPF_REG_0 && |
10312 | insn->src_reg != BPF_PSEUDO_CALL) || | |
092ed096 JW |
10313 | insn->dst_reg != BPF_REG_0 || |
10314 | class == BPF_JMP32) { | |
61bd5218 | 10315 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
10316 | return -EINVAL; |
10317 | } | |
10318 | ||
d83525ca AS |
10319 | if (env->cur_state->active_spin_lock && |
10320 | (insn->src_reg == BPF_PSEUDO_CALL || | |
10321 | insn->imm != BPF_FUNC_spin_unlock)) { | |
10322 | verbose(env, "function calls are not allowed while holding a lock\n"); | |
10323 | return -EINVAL; | |
10324 | } | |
f4d7e40a | 10325 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 10326 | err = check_func_call(env, insn, &env->insn_idx); |
f4d7e40a | 10327 | else |
69c087ba | 10328 | err = check_helper_call(env, insn, &env->insn_idx); |
17a52670 AS |
10329 | if (err) |
10330 | return err; | |
17a52670 AS |
10331 | } else if (opcode == BPF_JA) { |
10332 | if (BPF_SRC(insn->code) != BPF_K || | |
10333 | insn->imm != 0 || | |
10334 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
10335 | insn->dst_reg != BPF_REG_0 || |
10336 | class == BPF_JMP32) { | |
61bd5218 | 10337 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
10338 | return -EINVAL; |
10339 | } | |
10340 | ||
c08435ec | 10341 | env->insn_idx += insn->off + 1; |
17a52670 AS |
10342 | continue; |
10343 | ||
10344 | } else if (opcode == BPF_EXIT) { | |
10345 | if (BPF_SRC(insn->code) != BPF_K || | |
10346 | insn->imm != 0 || | |
10347 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
10348 | insn->dst_reg != BPF_REG_0 || |
10349 | class == BPF_JMP32) { | |
61bd5218 | 10350 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
10351 | return -EINVAL; |
10352 | } | |
10353 | ||
d83525ca AS |
10354 | if (env->cur_state->active_spin_lock) { |
10355 | verbose(env, "bpf_spin_unlock is missing\n"); | |
10356 | return -EINVAL; | |
10357 | } | |
10358 | ||
f4d7e40a AS |
10359 | if (state->curframe) { |
10360 | /* exit from nested function */ | |
c08435ec | 10361 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
10362 | if (err) |
10363 | return err; | |
10364 | do_print_state = true; | |
10365 | continue; | |
10366 | } | |
10367 | ||
fd978bf7 JS |
10368 | err = check_reference_leak(env); |
10369 | if (err) | |
10370 | return err; | |
10371 | ||
390ee7e2 AS |
10372 | err = check_return_code(env); |
10373 | if (err) | |
10374 | return err; | |
f1bca824 | 10375 | process_bpf_exit: |
2589726d | 10376 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 10377 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 10378 | &env->insn_idx, pop_log); |
638f5b90 AS |
10379 | if (err < 0) { |
10380 | if (err != -ENOENT) | |
10381 | return err; | |
17a52670 AS |
10382 | break; |
10383 | } else { | |
10384 | do_print_state = true; | |
10385 | continue; | |
10386 | } | |
10387 | } else { | |
c08435ec | 10388 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
10389 | if (err) |
10390 | return err; | |
10391 | } | |
10392 | } else if (class == BPF_LD) { | |
10393 | u8 mode = BPF_MODE(insn->code); | |
10394 | ||
10395 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
10396 | err = check_ld_abs(env, insn); |
10397 | if (err) | |
10398 | return err; | |
10399 | ||
17a52670 AS |
10400 | } else if (mode == BPF_IMM) { |
10401 | err = check_ld_imm(env, insn); | |
10402 | if (err) | |
10403 | return err; | |
10404 | ||
c08435ec | 10405 | env->insn_idx++; |
51c39bb1 | 10406 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; |
17a52670 | 10407 | } else { |
61bd5218 | 10408 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
10409 | return -EINVAL; |
10410 | } | |
10411 | } else { | |
61bd5218 | 10412 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
10413 | return -EINVAL; |
10414 | } | |
10415 | ||
c08435ec | 10416 | env->insn_idx++; |
17a52670 AS |
10417 | } |
10418 | ||
10419 | return 0; | |
10420 | } | |
10421 | ||
541c3bad AN |
10422 | static int find_btf_percpu_datasec(struct btf *btf) |
10423 | { | |
10424 | const struct btf_type *t; | |
10425 | const char *tname; | |
10426 | int i, n; | |
10427 | ||
10428 | /* | |
10429 | * Both vmlinux and module each have their own ".data..percpu" | |
10430 | * DATASECs in BTF. So for module's case, we need to skip vmlinux BTF | |
10431 | * types to look at only module's own BTF types. | |
10432 | */ | |
10433 | n = btf_nr_types(btf); | |
10434 | if (btf_is_module(btf)) | |
10435 | i = btf_nr_types(btf_vmlinux); | |
10436 | else | |
10437 | i = 1; | |
10438 | ||
10439 | for(; i < n; i++) { | |
10440 | t = btf_type_by_id(btf, i); | |
10441 | if (BTF_INFO_KIND(t->info) != BTF_KIND_DATASEC) | |
10442 | continue; | |
10443 | ||
10444 | tname = btf_name_by_offset(btf, t->name_off); | |
10445 | if (!strcmp(tname, ".data..percpu")) | |
10446 | return i; | |
10447 | } | |
10448 | ||
10449 | return -ENOENT; | |
10450 | } | |
10451 | ||
4976b718 HL |
10452 | /* replace pseudo btf_id with kernel symbol address */ |
10453 | static int check_pseudo_btf_id(struct bpf_verifier_env *env, | |
10454 | struct bpf_insn *insn, | |
10455 | struct bpf_insn_aux_data *aux) | |
10456 | { | |
eaa6bcb7 HL |
10457 | const struct btf_var_secinfo *vsi; |
10458 | const struct btf_type *datasec; | |
541c3bad | 10459 | struct btf_mod_pair *btf_mod; |
4976b718 HL |
10460 | const struct btf_type *t; |
10461 | const char *sym_name; | |
eaa6bcb7 | 10462 | bool percpu = false; |
f16e6313 | 10463 | u32 type, id = insn->imm; |
541c3bad | 10464 | struct btf *btf; |
f16e6313 | 10465 | s32 datasec_id; |
4976b718 | 10466 | u64 addr; |
541c3bad | 10467 | int i, btf_fd, err; |
4976b718 | 10468 | |
541c3bad AN |
10469 | btf_fd = insn[1].imm; |
10470 | if (btf_fd) { | |
10471 | btf = btf_get_by_fd(btf_fd); | |
10472 | if (IS_ERR(btf)) { | |
10473 | verbose(env, "invalid module BTF object FD specified.\n"); | |
10474 | return -EINVAL; | |
10475 | } | |
10476 | } else { | |
10477 | if (!btf_vmlinux) { | |
10478 | verbose(env, "kernel is missing BTF, make sure CONFIG_DEBUG_INFO_BTF=y is specified in Kconfig.\n"); | |
10479 | return -EINVAL; | |
10480 | } | |
10481 | btf = btf_vmlinux; | |
10482 | btf_get(btf); | |
4976b718 HL |
10483 | } |
10484 | ||
541c3bad | 10485 | t = btf_type_by_id(btf, id); |
4976b718 HL |
10486 | if (!t) { |
10487 | verbose(env, "ldimm64 insn specifies invalid btf_id %d.\n", id); | |
541c3bad AN |
10488 | err = -ENOENT; |
10489 | goto err_put; | |
4976b718 HL |
10490 | } |
10491 | ||
10492 | if (!btf_type_is_var(t)) { | |
541c3bad AN |
10493 | verbose(env, "pseudo btf_id %d in ldimm64 isn't KIND_VAR.\n", id); |
10494 | err = -EINVAL; | |
10495 | goto err_put; | |
4976b718 HL |
10496 | } |
10497 | ||
541c3bad | 10498 | sym_name = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
10499 | addr = kallsyms_lookup_name(sym_name); |
10500 | if (!addr) { | |
10501 | verbose(env, "ldimm64 failed to find the address for kernel symbol '%s'.\n", | |
10502 | sym_name); | |
541c3bad AN |
10503 | err = -ENOENT; |
10504 | goto err_put; | |
4976b718 HL |
10505 | } |
10506 | ||
541c3bad | 10507 | datasec_id = find_btf_percpu_datasec(btf); |
eaa6bcb7 | 10508 | if (datasec_id > 0) { |
541c3bad | 10509 | datasec = btf_type_by_id(btf, datasec_id); |
eaa6bcb7 HL |
10510 | for_each_vsi(i, datasec, vsi) { |
10511 | if (vsi->type == id) { | |
10512 | percpu = true; | |
10513 | break; | |
10514 | } | |
10515 | } | |
10516 | } | |
10517 | ||
4976b718 HL |
10518 | insn[0].imm = (u32)addr; |
10519 | insn[1].imm = addr >> 32; | |
10520 | ||
10521 | type = t->type; | |
541c3bad | 10522 | t = btf_type_skip_modifiers(btf, type, NULL); |
eaa6bcb7 HL |
10523 | if (percpu) { |
10524 | aux->btf_var.reg_type = PTR_TO_PERCPU_BTF_ID; | |
541c3bad | 10525 | aux->btf_var.btf = btf; |
eaa6bcb7 HL |
10526 | aux->btf_var.btf_id = type; |
10527 | } else if (!btf_type_is_struct(t)) { | |
4976b718 HL |
10528 | const struct btf_type *ret; |
10529 | const char *tname; | |
10530 | u32 tsize; | |
10531 | ||
10532 | /* resolve the type size of ksym. */ | |
541c3bad | 10533 | ret = btf_resolve_size(btf, t, &tsize); |
4976b718 | 10534 | if (IS_ERR(ret)) { |
541c3bad | 10535 | tname = btf_name_by_offset(btf, t->name_off); |
4976b718 HL |
10536 | verbose(env, "ldimm64 unable to resolve the size of type '%s': %ld\n", |
10537 | tname, PTR_ERR(ret)); | |
541c3bad AN |
10538 | err = -EINVAL; |
10539 | goto err_put; | |
4976b718 HL |
10540 | } |
10541 | aux->btf_var.reg_type = PTR_TO_MEM; | |
10542 | aux->btf_var.mem_size = tsize; | |
10543 | } else { | |
10544 | aux->btf_var.reg_type = PTR_TO_BTF_ID; | |
541c3bad | 10545 | aux->btf_var.btf = btf; |
4976b718 HL |
10546 | aux->btf_var.btf_id = type; |
10547 | } | |
541c3bad AN |
10548 | |
10549 | /* check whether we recorded this BTF (and maybe module) already */ | |
10550 | for (i = 0; i < env->used_btf_cnt; i++) { | |
10551 | if (env->used_btfs[i].btf == btf) { | |
10552 | btf_put(btf); | |
10553 | return 0; | |
10554 | } | |
10555 | } | |
10556 | ||
10557 | if (env->used_btf_cnt >= MAX_USED_BTFS) { | |
10558 | err = -E2BIG; | |
10559 | goto err_put; | |
10560 | } | |
10561 | ||
10562 | btf_mod = &env->used_btfs[env->used_btf_cnt]; | |
10563 | btf_mod->btf = btf; | |
10564 | btf_mod->module = NULL; | |
10565 | ||
10566 | /* if we reference variables from kernel module, bump its refcount */ | |
10567 | if (btf_is_module(btf)) { | |
10568 | btf_mod->module = btf_try_get_module(btf); | |
10569 | if (!btf_mod->module) { | |
10570 | err = -ENXIO; | |
10571 | goto err_put; | |
10572 | } | |
10573 | } | |
10574 | ||
10575 | env->used_btf_cnt++; | |
10576 | ||
4976b718 | 10577 | return 0; |
541c3bad AN |
10578 | err_put: |
10579 | btf_put(btf); | |
10580 | return err; | |
4976b718 HL |
10581 | } |
10582 | ||
56f668df MKL |
10583 | static int check_map_prealloc(struct bpf_map *map) |
10584 | { | |
10585 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
10586 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
10587 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
10588 | !(map->map_flags & BPF_F_NO_PREALLOC); |
10589 | } | |
10590 | ||
d83525ca AS |
10591 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
10592 | { | |
10593 | switch (type) { | |
10594 | case BPF_PROG_TYPE_KPROBE: | |
10595 | case BPF_PROG_TYPE_TRACEPOINT: | |
10596 | case BPF_PROG_TYPE_PERF_EVENT: | |
10597 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
10598 | return true; | |
10599 | default: | |
10600 | return false; | |
10601 | } | |
10602 | } | |
10603 | ||
94dacdbd TG |
10604 | static bool is_preallocated_map(struct bpf_map *map) |
10605 | { | |
10606 | if (!check_map_prealloc(map)) | |
10607 | return false; | |
10608 | if (map->inner_map_meta && !check_map_prealloc(map->inner_map_meta)) | |
10609 | return false; | |
10610 | return true; | |
10611 | } | |
10612 | ||
61bd5218 JK |
10613 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
10614 | struct bpf_map *map, | |
fdc15d38 AS |
10615 | struct bpf_prog *prog) |
10616 | ||
10617 | { | |
7e40781c | 10618 | enum bpf_prog_type prog_type = resolve_prog_type(prog); |
94dacdbd TG |
10619 | /* |
10620 | * Validate that trace type programs use preallocated hash maps. | |
10621 | * | |
10622 | * For programs attached to PERF events this is mandatory as the | |
10623 | * perf NMI can hit any arbitrary code sequence. | |
10624 | * | |
10625 | * All other trace types using preallocated hash maps are unsafe as | |
10626 | * well because tracepoint or kprobes can be inside locked regions | |
10627 | * of the memory allocator or at a place where a recursion into the | |
10628 | * memory allocator would see inconsistent state. | |
10629 | * | |
2ed905c5 TG |
10630 | * On RT enabled kernels run-time allocation of all trace type |
10631 | * programs is strictly prohibited due to lock type constraints. On | |
10632 | * !RT kernels it is allowed for backwards compatibility reasons for | |
10633 | * now, but warnings are emitted so developers are made aware of | |
10634 | * the unsafety and can fix their programs before this is enforced. | |
56f668df | 10635 | */ |
7e40781c UP |
10636 | if (is_tracing_prog_type(prog_type) && !is_preallocated_map(map)) { |
10637 | if (prog_type == BPF_PROG_TYPE_PERF_EVENT) { | |
61bd5218 | 10638 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
10639 | return -EINVAL; |
10640 | } | |
2ed905c5 TG |
10641 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
10642 | verbose(env, "trace type programs can only use preallocated hash map\n"); | |
10643 | return -EINVAL; | |
10644 | } | |
94dacdbd TG |
10645 | WARN_ONCE(1, "trace type BPF program uses run-time allocation\n"); |
10646 | verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n"); | |
fdc15d38 | 10647 | } |
a3884572 | 10648 | |
9e7a4d98 KS |
10649 | if (map_value_has_spin_lock(map)) { |
10650 | if (prog_type == BPF_PROG_TYPE_SOCKET_FILTER) { | |
10651 | verbose(env, "socket filter progs cannot use bpf_spin_lock yet\n"); | |
10652 | return -EINVAL; | |
10653 | } | |
10654 | ||
10655 | if (is_tracing_prog_type(prog_type)) { | |
10656 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
10657 | return -EINVAL; | |
10658 | } | |
10659 | ||
10660 | if (prog->aux->sleepable) { | |
10661 | verbose(env, "sleepable progs cannot use bpf_spin_lock yet\n"); | |
10662 | return -EINVAL; | |
10663 | } | |
d83525ca AS |
10664 | } |
10665 | ||
a3884572 | 10666 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && |
09728266 | 10667 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
10668 | verbose(env, "offload device mismatch between prog and map\n"); |
10669 | return -EINVAL; | |
10670 | } | |
10671 | ||
85d33df3 MKL |
10672 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
10673 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
10674 | return -EINVAL; | |
10675 | } | |
10676 | ||
1e6c62a8 AS |
10677 | if (prog->aux->sleepable) |
10678 | switch (map->map_type) { | |
10679 | case BPF_MAP_TYPE_HASH: | |
10680 | case BPF_MAP_TYPE_LRU_HASH: | |
10681 | case BPF_MAP_TYPE_ARRAY: | |
638e4b82 AS |
10682 | case BPF_MAP_TYPE_PERCPU_HASH: |
10683 | case BPF_MAP_TYPE_PERCPU_ARRAY: | |
10684 | case BPF_MAP_TYPE_LRU_PERCPU_HASH: | |
10685 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: | |
10686 | case BPF_MAP_TYPE_HASH_OF_MAPS: | |
1e6c62a8 AS |
10687 | if (!is_preallocated_map(map)) { |
10688 | verbose(env, | |
638e4b82 | 10689 | "Sleepable programs can only use preallocated maps\n"); |
1e6c62a8 AS |
10690 | return -EINVAL; |
10691 | } | |
10692 | break; | |
ba90c2cc KS |
10693 | case BPF_MAP_TYPE_RINGBUF: |
10694 | break; | |
1e6c62a8 AS |
10695 | default: |
10696 | verbose(env, | |
ba90c2cc | 10697 | "Sleepable programs can only use array, hash, and ringbuf maps\n"); |
1e6c62a8 AS |
10698 | return -EINVAL; |
10699 | } | |
10700 | ||
fdc15d38 AS |
10701 | return 0; |
10702 | } | |
10703 | ||
b741f163 RG |
10704 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
10705 | { | |
10706 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
10707 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
10708 | } | |
10709 | ||
4976b718 HL |
10710 | /* find and rewrite pseudo imm in ld_imm64 instructions: |
10711 | * | |
10712 | * 1. if it accesses map FD, replace it with actual map pointer. | |
10713 | * 2. if it accesses btf_id of a VAR, replace it with pointer to the var. | |
10714 | * | |
10715 | * NOTE: btf_vmlinux is required for converting pseudo btf_id. | |
0246e64d | 10716 | */ |
4976b718 | 10717 | static int resolve_pseudo_ldimm64(struct bpf_verifier_env *env) |
0246e64d AS |
10718 | { |
10719 | struct bpf_insn *insn = env->prog->insnsi; | |
10720 | int insn_cnt = env->prog->len; | |
fdc15d38 | 10721 | int i, j, err; |
0246e64d | 10722 | |
f1f7714e | 10723 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
10724 | if (err) |
10725 | return err; | |
10726 | ||
0246e64d | 10727 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 10728 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 10729 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 10730 | verbose(env, "BPF_LDX uses reserved fields\n"); |
d691f9e8 AS |
10731 | return -EINVAL; |
10732 | } | |
10733 | ||
0246e64d | 10734 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 10735 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
10736 | struct bpf_map *map; |
10737 | struct fd f; | |
d8eca5bb | 10738 | u64 addr; |
0246e64d AS |
10739 | |
10740 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
10741 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
10742 | insn[1].off != 0) { | |
61bd5218 | 10743 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
10744 | return -EINVAL; |
10745 | } | |
10746 | ||
d8eca5bb | 10747 | if (insn[0].src_reg == 0) |
0246e64d AS |
10748 | /* valid generic load 64-bit imm */ |
10749 | goto next_insn; | |
10750 | ||
4976b718 HL |
10751 | if (insn[0].src_reg == BPF_PSEUDO_BTF_ID) { |
10752 | aux = &env->insn_aux_data[i]; | |
10753 | err = check_pseudo_btf_id(env, insn, aux); | |
10754 | if (err) | |
10755 | return err; | |
10756 | goto next_insn; | |
10757 | } | |
10758 | ||
69c087ba YS |
10759 | if (insn[0].src_reg == BPF_PSEUDO_FUNC) { |
10760 | aux = &env->insn_aux_data[i]; | |
10761 | aux->ptr_type = PTR_TO_FUNC; | |
10762 | goto next_insn; | |
10763 | } | |
10764 | ||
d8eca5bb DB |
10765 | /* In final convert_pseudo_ld_imm64() step, this is |
10766 | * converted into regular 64-bit imm load insn. | |
10767 | */ | |
10768 | if ((insn[0].src_reg != BPF_PSEUDO_MAP_FD && | |
10769 | insn[0].src_reg != BPF_PSEUDO_MAP_VALUE) || | |
10770 | (insn[0].src_reg == BPF_PSEUDO_MAP_FD && | |
10771 | insn[1].imm != 0)) { | |
10772 | verbose(env, | |
10773 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
10774 | return -EINVAL; |
10775 | } | |
10776 | ||
20182390 | 10777 | f = fdget(insn[0].imm); |
c2101297 | 10778 | map = __bpf_map_get(f); |
0246e64d | 10779 | if (IS_ERR(map)) { |
61bd5218 | 10780 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 10781 | insn[0].imm); |
0246e64d AS |
10782 | return PTR_ERR(map); |
10783 | } | |
10784 | ||
61bd5218 | 10785 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
10786 | if (err) { |
10787 | fdput(f); | |
10788 | return err; | |
10789 | } | |
10790 | ||
d8eca5bb DB |
10791 | aux = &env->insn_aux_data[i]; |
10792 | if (insn->src_reg == BPF_PSEUDO_MAP_FD) { | |
10793 | addr = (unsigned long)map; | |
10794 | } else { | |
10795 | u32 off = insn[1].imm; | |
10796 | ||
10797 | if (off >= BPF_MAX_VAR_OFF) { | |
10798 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
10799 | fdput(f); | |
10800 | return -EINVAL; | |
10801 | } | |
10802 | ||
10803 | if (!map->ops->map_direct_value_addr) { | |
10804 | verbose(env, "no direct value access support for this map type\n"); | |
10805 | fdput(f); | |
10806 | return -EINVAL; | |
10807 | } | |
10808 | ||
10809 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
10810 | if (err) { | |
10811 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
10812 | map->value_size, off); | |
10813 | fdput(f); | |
10814 | return err; | |
10815 | } | |
10816 | ||
10817 | aux->map_off = off; | |
10818 | addr += off; | |
10819 | } | |
10820 | ||
10821 | insn[0].imm = (u32)addr; | |
10822 | insn[1].imm = addr >> 32; | |
0246e64d AS |
10823 | |
10824 | /* check whether we recorded this map already */ | |
d8eca5bb | 10825 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 10826 | if (env->used_maps[j] == map) { |
d8eca5bb | 10827 | aux->map_index = j; |
0246e64d AS |
10828 | fdput(f); |
10829 | goto next_insn; | |
10830 | } | |
d8eca5bb | 10831 | } |
0246e64d AS |
10832 | |
10833 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
10834 | fdput(f); | |
10835 | return -E2BIG; | |
10836 | } | |
10837 | ||
0246e64d AS |
10838 | /* hold the map. If the program is rejected by verifier, |
10839 | * the map will be released by release_maps() or it | |
10840 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 10841 | * and all maps are released in free_used_maps() |
0246e64d | 10842 | */ |
1e0bd5a0 | 10843 | bpf_map_inc(map); |
d8eca5bb DB |
10844 | |
10845 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
10846 | env->used_maps[env->used_map_cnt++] = map; |
10847 | ||
b741f163 | 10848 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 10849 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 10850 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
10851 | fdput(f); |
10852 | return -EBUSY; | |
10853 | } | |
10854 | ||
0246e64d AS |
10855 | fdput(f); |
10856 | next_insn: | |
10857 | insn++; | |
10858 | i++; | |
5e581dad DB |
10859 | continue; |
10860 | } | |
10861 | ||
10862 | /* Basic sanity check before we invest more work here. */ | |
10863 | if (!bpf_opcode_in_insntable(insn->code)) { | |
10864 | verbose(env, "unknown opcode %02x\n", insn->code); | |
10865 | return -EINVAL; | |
0246e64d AS |
10866 | } |
10867 | } | |
10868 | ||
10869 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
10870 | * 'struct bpf_map *' into a register instead of user map_fd. | |
10871 | * These pointers will be used later by verifier to validate map access. | |
10872 | */ | |
10873 | return 0; | |
10874 | } | |
10875 | ||
10876 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 10877 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 10878 | { |
a2ea0746 DB |
10879 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
10880 | env->used_map_cnt); | |
0246e64d AS |
10881 | } |
10882 | ||
541c3bad AN |
10883 | /* drop refcnt of maps used by the rejected program */ |
10884 | static void release_btfs(struct bpf_verifier_env *env) | |
10885 | { | |
10886 | __bpf_free_used_btfs(env->prog->aux, env->used_btfs, | |
10887 | env->used_btf_cnt); | |
10888 | } | |
10889 | ||
0246e64d | 10890 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ |
58e2af8b | 10891 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
10892 | { |
10893 | struct bpf_insn *insn = env->prog->insnsi; | |
10894 | int insn_cnt = env->prog->len; | |
10895 | int i; | |
10896 | ||
69c087ba YS |
10897 | for (i = 0; i < insn_cnt; i++, insn++) { |
10898 | if (insn->code != (BPF_LD | BPF_IMM | BPF_DW)) | |
10899 | continue; | |
10900 | if (insn->src_reg == BPF_PSEUDO_FUNC) | |
10901 | continue; | |
10902 | insn->src_reg = 0; | |
10903 | } | |
0246e64d AS |
10904 | } |
10905 | ||
8041902d AS |
10906 | /* single env->prog->insni[off] instruction was replaced with the range |
10907 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
10908 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
10909 | */ | |
b325fbca JW |
10910 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, |
10911 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d AS |
10912 | { |
10913 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
b325fbca JW |
10914 | struct bpf_insn *insn = new_prog->insnsi; |
10915 | u32 prog_len; | |
c131187d | 10916 | int i; |
8041902d | 10917 | |
b325fbca JW |
10918 | /* aux info at OFF always needs adjustment, no matter fast path |
10919 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
10920 | * original insn at old prog. | |
10921 | */ | |
10922 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
10923 | ||
8041902d AS |
10924 | if (cnt == 1) |
10925 | return 0; | |
b325fbca | 10926 | prog_len = new_prog->len; |
fad953ce KC |
10927 | new_data = vzalloc(array_size(prog_len, |
10928 | sizeof(struct bpf_insn_aux_data))); | |
8041902d AS |
10929 | if (!new_data) |
10930 | return -ENOMEM; | |
10931 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
10932 | memcpy(new_data + off + cnt - 1, old_data + off, | |
10933 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 10934 | for (i = off; i < off + cnt - 1; i++) { |
51c39bb1 | 10935 | new_data[i].seen = env->pass_cnt; |
b325fbca JW |
10936 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
10937 | } | |
8041902d AS |
10938 | env->insn_aux_data = new_data; |
10939 | vfree(old_data); | |
10940 | return 0; | |
10941 | } | |
10942 | ||
cc8b0b92 AS |
10943 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
10944 | { | |
10945 | int i; | |
10946 | ||
10947 | if (len == 1) | |
10948 | return; | |
4cb3d99c JW |
10949 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
10950 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 10951 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 10952 | continue; |
9c8105bd | 10953 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
10954 | } |
10955 | } | |
10956 | ||
a748c697 MF |
10957 | static void adjust_poke_descs(struct bpf_prog *prog, u32 len) |
10958 | { | |
10959 | struct bpf_jit_poke_descriptor *tab = prog->aux->poke_tab; | |
10960 | int i, sz = prog->aux->size_poke_tab; | |
10961 | struct bpf_jit_poke_descriptor *desc; | |
10962 | ||
10963 | for (i = 0; i < sz; i++) { | |
10964 | desc = &tab[i]; | |
10965 | desc->insn_idx += len - 1; | |
10966 | } | |
10967 | } | |
10968 | ||
8041902d AS |
10969 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
10970 | const struct bpf_insn *patch, u32 len) | |
10971 | { | |
10972 | struct bpf_prog *new_prog; | |
10973 | ||
10974 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
10975 | if (IS_ERR(new_prog)) { |
10976 | if (PTR_ERR(new_prog) == -ERANGE) | |
10977 | verbose(env, | |
10978 | "insn %d cannot be patched due to 16-bit range\n", | |
10979 | env->insn_aux_data[off].orig_idx); | |
8041902d | 10980 | return NULL; |
4f73379e | 10981 | } |
b325fbca | 10982 | if (adjust_insn_aux_data(env, new_prog, off, len)) |
8041902d | 10983 | return NULL; |
cc8b0b92 | 10984 | adjust_subprog_starts(env, off, len); |
a748c697 | 10985 | adjust_poke_descs(new_prog, len); |
8041902d AS |
10986 | return new_prog; |
10987 | } | |
10988 | ||
52875a04 JK |
10989 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
10990 | u32 off, u32 cnt) | |
10991 | { | |
10992 | int i, j; | |
10993 | ||
10994 | /* find first prog starting at or after off (first to remove) */ | |
10995 | for (i = 0; i < env->subprog_cnt; i++) | |
10996 | if (env->subprog_info[i].start >= off) | |
10997 | break; | |
10998 | /* find first prog starting at or after off + cnt (first to stay) */ | |
10999 | for (j = i; j < env->subprog_cnt; j++) | |
11000 | if (env->subprog_info[j].start >= off + cnt) | |
11001 | break; | |
11002 | /* if j doesn't start exactly at off + cnt, we are just removing | |
11003 | * the front of previous prog | |
11004 | */ | |
11005 | if (env->subprog_info[j].start != off + cnt) | |
11006 | j--; | |
11007 | ||
11008 | if (j > i) { | |
11009 | struct bpf_prog_aux *aux = env->prog->aux; | |
11010 | int move; | |
11011 | ||
11012 | /* move fake 'exit' subprog as well */ | |
11013 | move = env->subprog_cnt + 1 - j; | |
11014 | ||
11015 | memmove(env->subprog_info + i, | |
11016 | env->subprog_info + j, | |
11017 | sizeof(*env->subprog_info) * move); | |
11018 | env->subprog_cnt -= j - i; | |
11019 | ||
11020 | /* remove func_info */ | |
11021 | if (aux->func_info) { | |
11022 | move = aux->func_info_cnt - j; | |
11023 | ||
11024 | memmove(aux->func_info + i, | |
11025 | aux->func_info + j, | |
11026 | sizeof(*aux->func_info) * move); | |
11027 | aux->func_info_cnt -= j - i; | |
11028 | /* func_info->insn_off is set after all code rewrites, | |
11029 | * in adjust_btf_func() - no need to adjust | |
11030 | */ | |
11031 | } | |
11032 | } else { | |
11033 | /* convert i from "first prog to remove" to "first to adjust" */ | |
11034 | if (env->subprog_info[i].start == off) | |
11035 | i++; | |
11036 | } | |
11037 | ||
11038 | /* update fake 'exit' subprog as well */ | |
11039 | for (; i <= env->subprog_cnt; i++) | |
11040 | env->subprog_info[i].start -= cnt; | |
11041 | ||
11042 | return 0; | |
11043 | } | |
11044 | ||
11045 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
11046 | u32 cnt) | |
11047 | { | |
11048 | struct bpf_prog *prog = env->prog; | |
11049 | u32 i, l_off, l_cnt, nr_linfo; | |
11050 | struct bpf_line_info *linfo; | |
11051 | ||
11052 | nr_linfo = prog->aux->nr_linfo; | |
11053 | if (!nr_linfo) | |
11054 | return 0; | |
11055 | ||
11056 | linfo = prog->aux->linfo; | |
11057 | ||
11058 | /* find first line info to remove, count lines to be removed */ | |
11059 | for (i = 0; i < nr_linfo; i++) | |
11060 | if (linfo[i].insn_off >= off) | |
11061 | break; | |
11062 | ||
11063 | l_off = i; | |
11064 | l_cnt = 0; | |
11065 | for (; i < nr_linfo; i++) | |
11066 | if (linfo[i].insn_off < off + cnt) | |
11067 | l_cnt++; | |
11068 | else | |
11069 | break; | |
11070 | ||
11071 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
11072 | * last removed linfo. prog is already modified, so prog->len == off | |
11073 | * means no live instructions after (tail of the program was removed). | |
11074 | */ | |
11075 | if (prog->len != off && l_cnt && | |
11076 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
11077 | l_cnt--; | |
11078 | linfo[--i].insn_off = off + cnt; | |
11079 | } | |
11080 | ||
11081 | /* remove the line info which refer to the removed instructions */ | |
11082 | if (l_cnt) { | |
11083 | memmove(linfo + l_off, linfo + i, | |
11084 | sizeof(*linfo) * (nr_linfo - i)); | |
11085 | ||
11086 | prog->aux->nr_linfo -= l_cnt; | |
11087 | nr_linfo = prog->aux->nr_linfo; | |
11088 | } | |
11089 | ||
11090 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
11091 | for (i = l_off; i < nr_linfo; i++) | |
11092 | linfo[i].insn_off -= cnt; | |
11093 | ||
11094 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
11095 | for (i = 0; i <= env->subprog_cnt; i++) | |
11096 | if (env->subprog_info[i].linfo_idx > l_off) { | |
11097 | /* program may have started in the removed region but | |
11098 | * may not be fully removed | |
11099 | */ | |
11100 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
11101 | env->subprog_info[i].linfo_idx -= l_cnt; | |
11102 | else | |
11103 | env->subprog_info[i].linfo_idx = l_off; | |
11104 | } | |
11105 | ||
11106 | return 0; | |
11107 | } | |
11108 | ||
11109 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
11110 | { | |
11111 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11112 | unsigned int orig_prog_len = env->prog->len; | |
11113 | int err; | |
11114 | ||
08ca90af JK |
11115 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
11116 | bpf_prog_offload_remove_insns(env, off, cnt); | |
11117 | ||
52875a04 JK |
11118 | err = bpf_remove_insns(env->prog, off, cnt); |
11119 | if (err) | |
11120 | return err; | |
11121 | ||
11122 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
11123 | if (err) | |
11124 | return err; | |
11125 | ||
11126 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
11127 | if (err) | |
11128 | return err; | |
11129 | ||
11130 | memmove(aux_data + off, aux_data + off + cnt, | |
11131 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
11132 | ||
11133 | return 0; | |
11134 | } | |
11135 | ||
2a5418a1 DB |
11136 | /* The verifier does more data flow analysis than llvm and will not |
11137 | * explore branches that are dead at run time. Malicious programs can | |
11138 | * have dead code too. Therefore replace all dead at-run-time code | |
11139 | * with 'ja -1'. | |
11140 | * | |
11141 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
11142 | * program and through another bug we would manage to jump there, then | |
11143 | * we'd execute beyond program memory otherwise. Returning exception | |
11144 | * code also wouldn't work since we can have subprogs where the dead | |
11145 | * code could be located. | |
c131187d AS |
11146 | */ |
11147 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
11148 | { | |
11149 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 11150 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
11151 | struct bpf_insn *insn = env->prog->insnsi; |
11152 | const int insn_cnt = env->prog->len; | |
11153 | int i; | |
11154 | ||
11155 | for (i = 0; i < insn_cnt; i++) { | |
11156 | if (aux_data[i].seen) | |
11157 | continue; | |
2a5418a1 | 11158 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
11159 | } |
11160 | } | |
11161 | ||
e2ae4ca2 JK |
11162 | static bool insn_is_cond_jump(u8 code) |
11163 | { | |
11164 | u8 op; | |
11165 | ||
092ed096 JW |
11166 | if (BPF_CLASS(code) == BPF_JMP32) |
11167 | return true; | |
11168 | ||
e2ae4ca2 JK |
11169 | if (BPF_CLASS(code) != BPF_JMP) |
11170 | return false; | |
11171 | ||
11172 | op = BPF_OP(code); | |
11173 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
11174 | } | |
11175 | ||
11176 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
11177 | { | |
11178 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11179 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
11180 | struct bpf_insn *insn = env->prog->insnsi; | |
11181 | const int insn_cnt = env->prog->len; | |
11182 | int i; | |
11183 | ||
11184 | for (i = 0; i < insn_cnt; i++, insn++) { | |
11185 | if (!insn_is_cond_jump(insn->code)) | |
11186 | continue; | |
11187 | ||
11188 | if (!aux_data[i + 1].seen) | |
11189 | ja.off = insn->off; | |
11190 | else if (!aux_data[i + 1 + insn->off].seen) | |
11191 | ja.off = 0; | |
11192 | else | |
11193 | continue; | |
11194 | ||
08ca90af JK |
11195 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
11196 | bpf_prog_offload_replace_insn(env, i, &ja); | |
11197 | ||
e2ae4ca2 JK |
11198 | memcpy(insn, &ja, sizeof(ja)); |
11199 | } | |
11200 | } | |
11201 | ||
52875a04 JK |
11202 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
11203 | { | |
11204 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
11205 | int insn_cnt = env->prog->len; | |
11206 | int i, err; | |
11207 | ||
11208 | for (i = 0; i < insn_cnt; i++) { | |
11209 | int j; | |
11210 | ||
11211 | j = 0; | |
11212 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
11213 | j++; | |
11214 | if (!j) | |
11215 | continue; | |
11216 | ||
11217 | err = verifier_remove_insns(env, i, j); | |
11218 | if (err) | |
11219 | return err; | |
11220 | insn_cnt = env->prog->len; | |
11221 | } | |
11222 | ||
11223 | return 0; | |
11224 | } | |
11225 | ||
a1b14abc JK |
11226 | static int opt_remove_nops(struct bpf_verifier_env *env) |
11227 | { | |
11228 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
11229 | struct bpf_insn *insn = env->prog->insnsi; | |
11230 | int insn_cnt = env->prog->len; | |
11231 | int i, err; | |
11232 | ||
11233 | for (i = 0; i < insn_cnt; i++) { | |
11234 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
11235 | continue; | |
11236 | ||
11237 | err = verifier_remove_insns(env, i, 1); | |
11238 | if (err) | |
11239 | return err; | |
11240 | insn_cnt--; | |
11241 | i--; | |
11242 | } | |
11243 | ||
11244 | return 0; | |
11245 | } | |
11246 | ||
d6c2308c JW |
11247 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
11248 | const union bpf_attr *attr) | |
a4b1d3c1 | 11249 | { |
d6c2308c | 11250 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 11251 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 11252 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 11253 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 11254 | struct bpf_prog *new_prog; |
d6c2308c | 11255 | bool rnd_hi32; |
a4b1d3c1 | 11256 | |
d6c2308c | 11257 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 11258 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
11259 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
11260 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
11261 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
11262 | for (i = 0; i < len; i++) { |
11263 | int adj_idx = i + delta; | |
11264 | struct bpf_insn insn; | |
83a28819 | 11265 | int load_reg; |
a4b1d3c1 | 11266 | |
d6c2308c | 11267 | insn = insns[adj_idx]; |
83a28819 | 11268 | load_reg = insn_def_regno(&insn); |
d6c2308c JW |
11269 | if (!aux[adj_idx].zext_dst) { |
11270 | u8 code, class; | |
11271 | u32 imm_rnd; | |
11272 | ||
11273 | if (!rnd_hi32) | |
11274 | continue; | |
11275 | ||
11276 | code = insn.code; | |
11277 | class = BPF_CLASS(code); | |
83a28819 | 11278 | if (load_reg == -1) |
d6c2308c JW |
11279 | continue; |
11280 | ||
11281 | /* NOTE: arg "reg" (the fourth one) is only used for | |
83a28819 IL |
11282 | * BPF_STX + SRC_OP, so it is safe to pass NULL |
11283 | * here. | |
d6c2308c | 11284 | */ |
83a28819 | 11285 | if (is_reg64(env, &insn, load_reg, NULL, DST_OP)) { |
d6c2308c JW |
11286 | if (class == BPF_LD && |
11287 | BPF_MODE(code) == BPF_IMM) | |
11288 | i++; | |
11289 | continue; | |
11290 | } | |
11291 | ||
11292 | /* ctx load could be transformed into wider load. */ | |
11293 | if (class == BPF_LDX && | |
11294 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
11295 | continue; | |
11296 | ||
11297 | imm_rnd = get_random_int(); | |
11298 | rnd_hi32_patch[0] = insn; | |
11299 | rnd_hi32_patch[1].imm = imm_rnd; | |
83a28819 | 11300 | rnd_hi32_patch[3].dst_reg = load_reg; |
d6c2308c JW |
11301 | patch = rnd_hi32_patch; |
11302 | patch_len = 4; | |
11303 | goto apply_patch_buffer; | |
11304 | } | |
11305 | ||
39491867 BJ |
11306 | /* Add in an zero-extend instruction if a) the JIT has requested |
11307 | * it or b) it's a CMPXCHG. | |
11308 | * | |
11309 | * The latter is because: BPF_CMPXCHG always loads a value into | |
11310 | * R0, therefore always zero-extends. However some archs' | |
11311 | * equivalent instruction only does this load when the | |
11312 | * comparison is successful. This detail of CMPXCHG is | |
11313 | * orthogonal to the general zero-extension behaviour of the | |
11314 | * CPU, so it's treated independently of bpf_jit_needs_zext. | |
11315 | */ | |
11316 | if (!bpf_jit_needs_zext() && !is_cmpxchg_insn(&insn)) | |
a4b1d3c1 JW |
11317 | continue; |
11318 | ||
83a28819 IL |
11319 | if (WARN_ON(load_reg == -1)) { |
11320 | verbose(env, "verifier bug. zext_dst is set, but no reg is defined\n"); | |
11321 | return -EFAULT; | |
b2e37a71 IL |
11322 | } |
11323 | ||
a4b1d3c1 | 11324 | zext_patch[0] = insn; |
b2e37a71 IL |
11325 | zext_patch[1].dst_reg = load_reg; |
11326 | zext_patch[1].src_reg = load_reg; | |
d6c2308c JW |
11327 | patch = zext_patch; |
11328 | patch_len = 2; | |
11329 | apply_patch_buffer: | |
11330 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
11331 | if (!new_prog) |
11332 | return -ENOMEM; | |
11333 | env->prog = new_prog; | |
11334 | insns = new_prog->insnsi; | |
11335 | aux = env->insn_aux_data; | |
d6c2308c | 11336 | delta += patch_len - 1; |
a4b1d3c1 JW |
11337 | } |
11338 | ||
11339 | return 0; | |
11340 | } | |
11341 | ||
c64b7983 JS |
11342 | /* convert load instructions that access fields of a context type into a |
11343 | * sequence of instructions that access fields of the underlying structure: | |
11344 | * struct __sk_buff -> struct sk_buff | |
11345 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 11346 | */ |
58e2af8b | 11347 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 11348 | { |
00176a34 | 11349 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 11350 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 11351 | const int insn_cnt = env->prog->len; |
36bbef52 | 11352 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 11353 | u32 target_size, size_default, off; |
9bac3d6d | 11354 | struct bpf_prog *new_prog; |
d691f9e8 | 11355 | enum bpf_access_type type; |
f96da094 | 11356 | bool is_narrower_load; |
9bac3d6d | 11357 | |
b09928b9 DB |
11358 | if (ops->gen_prologue || env->seen_direct_write) { |
11359 | if (!ops->gen_prologue) { | |
11360 | verbose(env, "bpf verifier is misconfigured\n"); | |
11361 | return -EINVAL; | |
11362 | } | |
36bbef52 DB |
11363 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
11364 | env->prog); | |
11365 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 11366 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
11367 | return -EINVAL; |
11368 | } else if (cnt) { | |
8041902d | 11369 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
11370 | if (!new_prog) |
11371 | return -ENOMEM; | |
8041902d | 11372 | |
36bbef52 | 11373 | env->prog = new_prog; |
3df126f3 | 11374 | delta += cnt - 1; |
36bbef52 DB |
11375 | } |
11376 | } | |
11377 | ||
c64b7983 | 11378 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
11379 | return 0; |
11380 | ||
3df126f3 | 11381 | insn = env->prog->insnsi + delta; |
36bbef52 | 11382 | |
9bac3d6d | 11383 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 JS |
11384 | bpf_convert_ctx_access_t convert_ctx_access; |
11385 | ||
62c7989b DB |
11386 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
11387 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
11388 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 11389 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 11390 | type = BPF_READ; |
62c7989b DB |
11391 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
11392 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
11393 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 11394 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
11395 | type = BPF_WRITE; |
11396 | else | |
9bac3d6d AS |
11397 | continue; |
11398 | ||
af86ca4e AS |
11399 | if (type == BPF_WRITE && |
11400 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
11401 | struct bpf_insn patch[] = { | |
11402 | /* Sanitize suspicious stack slot with zero. | |
11403 | * There are no memory dependencies for this store, | |
11404 | * since it's only using frame pointer and immediate | |
11405 | * constant of zero | |
11406 | */ | |
11407 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
11408 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
11409 | 0), | |
11410 | /* the original STX instruction will immediately | |
11411 | * overwrite the same stack slot with appropriate value | |
11412 | */ | |
11413 | *insn, | |
11414 | }; | |
11415 | ||
11416 | cnt = ARRAY_SIZE(patch); | |
11417 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
11418 | if (!new_prog) | |
11419 | return -ENOMEM; | |
11420 | ||
11421 | delta += cnt - 1; | |
11422 | env->prog = new_prog; | |
11423 | insn = new_prog->insnsi + i + delta; | |
11424 | continue; | |
11425 | } | |
11426 | ||
c64b7983 JS |
11427 | switch (env->insn_aux_data[i + delta].ptr_type) { |
11428 | case PTR_TO_CTX: | |
11429 | if (!ops->convert_ctx_access) | |
11430 | continue; | |
11431 | convert_ctx_access = ops->convert_ctx_access; | |
11432 | break; | |
11433 | case PTR_TO_SOCKET: | |
46f8bc92 | 11434 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
11435 | convert_ctx_access = bpf_sock_convert_ctx_access; |
11436 | break; | |
655a51e5 MKL |
11437 | case PTR_TO_TCP_SOCK: |
11438 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
11439 | break; | |
fada7fdc JL |
11440 | case PTR_TO_XDP_SOCK: |
11441 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
11442 | break; | |
2a02759e | 11443 | case PTR_TO_BTF_ID: |
27ae7997 MKL |
11444 | if (type == BPF_READ) { |
11445 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
11446 | BPF_SIZE((insn)->code); | |
11447 | env->prog->aux->num_exentries++; | |
7e40781c | 11448 | } else if (resolve_prog_type(env->prog) != BPF_PROG_TYPE_STRUCT_OPS) { |
2a02759e AS |
11449 | verbose(env, "Writes through BTF pointers are not allowed\n"); |
11450 | return -EINVAL; | |
11451 | } | |
2a02759e | 11452 | continue; |
c64b7983 | 11453 | default: |
9bac3d6d | 11454 | continue; |
c64b7983 | 11455 | } |
9bac3d6d | 11456 | |
31fd8581 | 11457 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 11458 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
11459 | |
11460 | /* If the read access is a narrower load of the field, | |
11461 | * convert to a 4/8-byte load, to minimum program type specific | |
11462 | * convert_ctx_access changes. If conversion is successful, | |
11463 | * we will apply proper mask to the result. | |
11464 | */ | |
f96da094 | 11465 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
11466 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
11467 | off = insn->off; | |
31fd8581 | 11468 | if (is_narrower_load) { |
f96da094 DB |
11469 | u8 size_code; |
11470 | ||
11471 | if (type == BPF_WRITE) { | |
61bd5218 | 11472 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
11473 | return -EINVAL; |
11474 | } | |
31fd8581 | 11475 | |
f96da094 | 11476 | size_code = BPF_H; |
31fd8581 YS |
11477 | if (ctx_field_size == 4) |
11478 | size_code = BPF_W; | |
11479 | else if (ctx_field_size == 8) | |
11480 | size_code = BPF_DW; | |
f96da094 | 11481 | |
bc23105c | 11482 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
11483 | insn->code = BPF_LDX | BPF_MEM | size_code; |
11484 | } | |
f96da094 DB |
11485 | |
11486 | target_size = 0; | |
c64b7983 JS |
11487 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
11488 | &target_size); | |
f96da094 DB |
11489 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
11490 | (ctx_field_size && !target_size)) { | |
61bd5218 | 11491 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
11492 | return -EINVAL; |
11493 | } | |
f96da094 DB |
11494 | |
11495 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
11496 | u8 shift = bpf_ctx_narrow_access_offset( |
11497 | off, size, size_default) * 8; | |
46f53a65 AI |
11498 | if (ctx_field_size <= 4) { |
11499 | if (shift) | |
11500 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
11501 | insn->dst_reg, | |
11502 | shift); | |
31fd8581 | 11503 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 11504 | (1 << size * 8) - 1); |
46f53a65 AI |
11505 | } else { |
11506 | if (shift) | |
11507 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
11508 | insn->dst_reg, | |
11509 | shift); | |
31fd8581 | 11510 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 11511 | (1ULL << size * 8) - 1); |
46f53a65 | 11512 | } |
31fd8581 | 11513 | } |
9bac3d6d | 11514 | |
8041902d | 11515 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
11516 | if (!new_prog) |
11517 | return -ENOMEM; | |
11518 | ||
3df126f3 | 11519 | delta += cnt - 1; |
9bac3d6d AS |
11520 | |
11521 | /* keep walking new program and skip insns we just inserted */ | |
11522 | env->prog = new_prog; | |
3df126f3 | 11523 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
11524 | } |
11525 | ||
11526 | return 0; | |
11527 | } | |
11528 | ||
1c2a088a AS |
11529 | static int jit_subprogs(struct bpf_verifier_env *env) |
11530 | { | |
11531 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
11532 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
a748c697 | 11533 | struct bpf_map *map_ptr; |
7105e828 | 11534 | struct bpf_insn *insn; |
1c2a088a | 11535 | void *old_bpf_func; |
c4c0bdc0 | 11536 | int err, num_exentries; |
1c2a088a | 11537 | |
f910cefa | 11538 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
11539 | return 0; |
11540 | ||
7105e828 | 11541 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
69c087ba YS |
11542 | if (bpf_pseudo_func(insn)) { |
11543 | env->insn_aux_data[i].call_imm = insn->imm; | |
11544 | /* subprog is encoded in insn[1].imm */ | |
11545 | continue; | |
11546 | } | |
11547 | ||
23a2d70c | 11548 | if (!bpf_pseudo_call(insn)) |
1c2a088a | 11549 | continue; |
c7a89784 DB |
11550 | /* Upon error here we cannot fall back to interpreter but |
11551 | * need a hard reject of the program. Thus -EFAULT is | |
11552 | * propagated in any case. | |
11553 | */ | |
1c2a088a AS |
11554 | subprog = find_subprog(env, i + insn->imm + 1); |
11555 | if (subprog < 0) { | |
11556 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
11557 | i + insn->imm + 1); | |
11558 | return -EFAULT; | |
11559 | } | |
11560 | /* temporarily remember subprog id inside insn instead of | |
11561 | * aux_data, since next loop will split up all insns into funcs | |
11562 | */ | |
f910cefa | 11563 | insn->off = subprog; |
1c2a088a AS |
11564 | /* remember original imm in case JIT fails and fallback |
11565 | * to interpreter will be needed | |
11566 | */ | |
11567 | env->insn_aux_data[i].call_imm = insn->imm; | |
11568 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
11569 | insn->imm = 1; | |
11570 | } | |
11571 | ||
c454a46b MKL |
11572 | err = bpf_prog_alloc_jited_linfo(prog); |
11573 | if (err) | |
11574 | goto out_undo_insn; | |
11575 | ||
11576 | err = -ENOMEM; | |
6396bb22 | 11577 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 11578 | if (!func) |
c7a89784 | 11579 | goto out_undo_insn; |
1c2a088a | 11580 | |
f910cefa | 11581 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 11582 | subprog_start = subprog_end; |
4cb3d99c | 11583 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
11584 | |
11585 | len = subprog_end - subprog_start; | |
492ecee8 AS |
11586 | /* BPF_PROG_RUN doesn't call subprogs directly, |
11587 | * hence main prog stats include the runtime of subprogs. | |
11588 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
700d4796 | 11589 | * func[i]->stats will never be accessed and stays NULL |
492ecee8 AS |
11590 | */ |
11591 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
11592 | if (!func[i]) |
11593 | goto out_free; | |
11594 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
11595 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 11596 | func[i]->type = prog->type; |
1c2a088a | 11597 | func[i]->len = len; |
4f74d809 DB |
11598 | if (bpf_prog_calc_tag(func[i])) |
11599 | goto out_free; | |
1c2a088a | 11600 | func[i]->is_func = 1; |
ba64e7d8 YS |
11601 | func[i]->aux->func_idx = i; |
11602 | /* the btf and func_info will be freed only at prog->aux */ | |
11603 | func[i]->aux->btf = prog->aux->btf; | |
11604 | func[i]->aux->func_info = prog->aux->func_info; | |
11605 | ||
a748c697 MF |
11606 | for (j = 0; j < prog->aux->size_poke_tab; j++) { |
11607 | u32 insn_idx = prog->aux->poke_tab[j].insn_idx; | |
11608 | int ret; | |
11609 | ||
11610 | if (!(insn_idx >= subprog_start && | |
11611 | insn_idx <= subprog_end)) | |
11612 | continue; | |
11613 | ||
11614 | ret = bpf_jit_add_poke_descriptor(func[i], | |
11615 | &prog->aux->poke_tab[j]); | |
11616 | if (ret < 0) { | |
11617 | verbose(env, "adding tail call poke descriptor failed\n"); | |
11618 | goto out_free; | |
11619 | } | |
11620 | ||
11621 | func[i]->insnsi[insn_idx - subprog_start].imm = ret + 1; | |
11622 | ||
11623 | map_ptr = func[i]->aux->poke_tab[ret].tail_call.map; | |
11624 | ret = map_ptr->ops->map_poke_track(map_ptr, func[i]->aux); | |
11625 | if (ret < 0) { | |
11626 | verbose(env, "tracking tail call prog failed\n"); | |
11627 | goto out_free; | |
11628 | } | |
11629 | } | |
11630 | ||
1c2a088a AS |
11631 | /* Use bpf_prog_F_tag to indicate functions in stack traces. |
11632 | * Long term would need debug info to populate names | |
11633 | */ | |
11634 | func[i]->aux->name[0] = 'F'; | |
9c8105bd | 11635 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 11636 | func[i]->jit_requested = 1; |
c454a46b MKL |
11637 | func[i]->aux->linfo = prog->aux->linfo; |
11638 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
11639 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
11640 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
c4c0bdc0 YS |
11641 | num_exentries = 0; |
11642 | insn = func[i]->insnsi; | |
11643 | for (j = 0; j < func[i]->len; j++, insn++) { | |
11644 | if (BPF_CLASS(insn->code) == BPF_LDX && | |
11645 | BPF_MODE(insn->code) == BPF_PROBE_MEM) | |
11646 | num_exentries++; | |
11647 | } | |
11648 | func[i]->aux->num_exentries = num_exentries; | |
ebf7d1f5 | 11649 | func[i]->aux->tail_call_reachable = env->subprog_info[i].tail_call_reachable; |
1c2a088a AS |
11650 | func[i] = bpf_int_jit_compile(func[i]); |
11651 | if (!func[i]->jited) { | |
11652 | err = -ENOTSUPP; | |
11653 | goto out_free; | |
11654 | } | |
11655 | cond_resched(); | |
11656 | } | |
a748c697 MF |
11657 | |
11658 | /* Untrack main program's aux structs so that during map_poke_run() | |
11659 | * we will not stumble upon the unfilled poke descriptors; each | |
11660 | * of the main program's poke descs got distributed across subprogs | |
11661 | * and got tracked onto map, so we are sure that none of them will | |
11662 | * be missed after the operation below | |
11663 | */ | |
11664 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
11665 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
11666 | ||
11667 | map_ptr->ops->map_poke_untrack(map_ptr, prog->aux); | |
11668 | } | |
11669 | ||
1c2a088a AS |
11670 | /* at this point all bpf functions were successfully JITed |
11671 | * now populate all bpf_calls with correct addresses and | |
11672 | * run last pass of JIT | |
11673 | */ | |
f910cefa | 11674 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
11675 | insn = func[i]->insnsi; |
11676 | for (j = 0; j < func[i]->len; j++, insn++) { | |
69c087ba YS |
11677 | if (bpf_pseudo_func(insn)) { |
11678 | subprog = insn[1].imm; | |
11679 | insn[0].imm = (u32)(long)func[subprog]->bpf_func; | |
11680 | insn[1].imm = ((u64)(long)func[subprog]->bpf_func) >> 32; | |
11681 | continue; | |
11682 | } | |
23a2d70c | 11683 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
11684 | continue; |
11685 | subprog = insn->off; | |
0d306c31 PB |
11686 | insn->imm = BPF_CAST_CALL(func[subprog]->bpf_func) - |
11687 | __bpf_call_base; | |
1c2a088a | 11688 | } |
2162fed4 SD |
11689 | |
11690 | /* we use the aux data to keep a list of the start addresses | |
11691 | * of the JITed images for each function in the program | |
11692 | * | |
11693 | * for some architectures, such as powerpc64, the imm field | |
11694 | * might not be large enough to hold the offset of the start | |
11695 | * address of the callee's JITed image from __bpf_call_base | |
11696 | * | |
11697 | * in such cases, we can lookup the start address of a callee | |
11698 | * by using its subprog id, available from the off field of | |
11699 | * the call instruction, as an index for this list | |
11700 | */ | |
11701 | func[i]->aux->func = func; | |
11702 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 11703 | } |
f910cefa | 11704 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
11705 | old_bpf_func = func[i]->bpf_func; |
11706 | tmp = bpf_int_jit_compile(func[i]); | |
11707 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
11708 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 11709 | err = -ENOTSUPP; |
1c2a088a AS |
11710 | goto out_free; |
11711 | } | |
11712 | cond_resched(); | |
11713 | } | |
11714 | ||
11715 | /* finally lock prog and jit images for all functions and | |
11716 | * populate kallsysm | |
11717 | */ | |
f910cefa | 11718 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
11719 | bpf_prog_lock_ro(func[i]); |
11720 | bpf_prog_kallsyms_add(func[i]); | |
11721 | } | |
7105e828 DB |
11722 | |
11723 | /* Last step: make now unused interpreter insns from main | |
11724 | * prog consistent for later dump requests, so they can | |
11725 | * later look the same as if they were interpreted only. | |
11726 | */ | |
11727 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
69c087ba YS |
11728 | if (bpf_pseudo_func(insn)) { |
11729 | insn[0].imm = env->insn_aux_data[i].call_imm; | |
11730 | insn[1].imm = find_subprog(env, i + insn[0].imm + 1); | |
11731 | continue; | |
11732 | } | |
23a2d70c | 11733 | if (!bpf_pseudo_call(insn)) |
7105e828 DB |
11734 | continue; |
11735 | insn->off = env->insn_aux_data[i].call_imm; | |
11736 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 11737 | insn->imm = subprog; |
7105e828 DB |
11738 | } |
11739 | ||
1c2a088a AS |
11740 | prog->jited = 1; |
11741 | prog->bpf_func = func[0]->bpf_func; | |
11742 | prog->aux->func = func; | |
f910cefa | 11743 | prog->aux->func_cnt = env->subprog_cnt; |
c454a46b | 11744 | bpf_prog_free_unused_jited_linfo(prog); |
1c2a088a AS |
11745 | return 0; |
11746 | out_free: | |
a748c697 MF |
11747 | for (i = 0; i < env->subprog_cnt; i++) { |
11748 | if (!func[i]) | |
11749 | continue; | |
11750 | ||
11751 | for (j = 0; j < func[i]->aux->size_poke_tab; j++) { | |
11752 | map_ptr = func[i]->aux->poke_tab[j].tail_call.map; | |
11753 | map_ptr->ops->map_poke_untrack(map_ptr, func[i]->aux); | |
11754 | } | |
11755 | bpf_jit_free(func[i]); | |
11756 | } | |
1c2a088a | 11757 | kfree(func); |
c7a89784 | 11758 | out_undo_insn: |
1c2a088a AS |
11759 | /* cleanup main prog to be interpreted */ |
11760 | prog->jit_requested = 0; | |
11761 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
23a2d70c | 11762 | if (!bpf_pseudo_call(insn)) |
1c2a088a AS |
11763 | continue; |
11764 | insn->off = 0; | |
11765 | insn->imm = env->insn_aux_data[i].call_imm; | |
11766 | } | |
c454a46b | 11767 | bpf_prog_free_jited_linfo(prog); |
1c2a088a AS |
11768 | return err; |
11769 | } | |
11770 | ||
1ea47e01 AS |
11771 | static int fixup_call_args(struct bpf_verifier_env *env) |
11772 | { | |
19d28fbd | 11773 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
11774 | struct bpf_prog *prog = env->prog; |
11775 | struct bpf_insn *insn = prog->insnsi; | |
11776 | int i, depth; | |
19d28fbd | 11777 | #endif |
e4052d06 | 11778 | int err = 0; |
1ea47e01 | 11779 | |
e4052d06 QM |
11780 | if (env->prog->jit_requested && |
11781 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
11782 | err = jit_subprogs(env); |
11783 | if (err == 0) | |
1c2a088a | 11784 | return 0; |
c7a89784 DB |
11785 | if (err == -EFAULT) |
11786 | return err; | |
19d28fbd DM |
11787 | } |
11788 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
e411901c MF |
11789 | if (env->subprog_cnt > 1 && env->prog->aux->tail_call_reachable) { |
11790 | /* When JIT fails the progs with bpf2bpf calls and tail_calls | |
11791 | * have to be rejected, since interpreter doesn't support them yet. | |
11792 | */ | |
11793 | verbose(env, "tail_calls are not allowed in non-JITed programs with bpf-to-bpf calls\n"); | |
11794 | return -EINVAL; | |
11795 | } | |
1ea47e01 | 11796 | for (i = 0; i < prog->len; i++, insn++) { |
69c087ba YS |
11797 | if (bpf_pseudo_func(insn)) { |
11798 | /* When JIT fails the progs with callback calls | |
11799 | * have to be rejected, since interpreter doesn't support them yet. | |
11800 | */ | |
11801 | verbose(env, "callbacks are not allowed in non-JITed programs\n"); | |
11802 | return -EINVAL; | |
11803 | } | |
11804 | ||
23a2d70c | 11805 | if (!bpf_pseudo_call(insn)) |
1ea47e01 AS |
11806 | continue; |
11807 | depth = get_callee_stack_depth(env, insn, i); | |
11808 | if (depth < 0) | |
11809 | return depth; | |
11810 | bpf_patch_call_args(insn, depth); | |
11811 | } | |
19d28fbd DM |
11812 | err = 0; |
11813 | #endif | |
11814 | return err; | |
1ea47e01 AS |
11815 | } |
11816 | ||
e6ac5933 BJ |
11817 | /* Do various post-verification rewrites in a single program pass. |
11818 | * These rewrites simplify JIT and interpreter implementations. | |
e245c5c6 | 11819 | */ |
e6ac5933 | 11820 | static int do_misc_fixups(struct bpf_verifier_env *env) |
e245c5c6 | 11821 | { |
79741b3b | 11822 | struct bpf_prog *prog = env->prog; |
d2e4c1e6 | 11823 | bool expect_blinding = bpf_jit_blinding_enabled(prog); |
79741b3b | 11824 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 11825 | const struct bpf_func_proto *fn; |
79741b3b | 11826 | const int insn_cnt = prog->len; |
09772d92 | 11827 | const struct bpf_map_ops *ops; |
c93552c4 | 11828 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
11829 | struct bpf_insn insn_buf[16]; |
11830 | struct bpf_prog *new_prog; | |
11831 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 11832 | int i, ret, cnt, delta = 0; |
e245c5c6 | 11833 | |
79741b3b | 11834 | for (i = 0; i < insn_cnt; i++, insn++) { |
e6ac5933 | 11835 | /* Make divide-by-zero exceptions impossible. */ |
f6b1b3bf DB |
11836 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
11837 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
11838 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 11839 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf | 11840 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
e88b2c6e DB |
11841 | bool isdiv = BPF_OP(insn->code) == BPF_DIV; |
11842 | struct bpf_insn *patchlet; | |
11843 | struct bpf_insn chk_and_div[] = { | |
9b00f1b7 | 11844 | /* [R,W]x div 0 -> 0 */ |
e88b2c6e DB |
11845 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
11846 | BPF_JNE | BPF_K, insn->src_reg, | |
11847 | 0, 2, 0), | |
f6b1b3bf DB |
11848 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), |
11849 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
11850 | *insn, | |
11851 | }; | |
e88b2c6e | 11852 | struct bpf_insn chk_and_mod[] = { |
9b00f1b7 | 11853 | /* [R,W]x mod 0 -> [R,W]x */ |
e88b2c6e DB |
11854 | BPF_RAW_INSN((is64 ? BPF_JMP : BPF_JMP32) | |
11855 | BPF_JEQ | BPF_K, insn->src_reg, | |
9b00f1b7 | 11856 | 0, 1 + (is64 ? 0 : 1), 0), |
f6b1b3bf | 11857 | *insn, |
9b00f1b7 DB |
11858 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), |
11859 | BPF_MOV32_REG(insn->dst_reg, insn->dst_reg), | |
f6b1b3bf | 11860 | }; |
f6b1b3bf | 11861 | |
e88b2c6e DB |
11862 | patchlet = isdiv ? chk_and_div : chk_and_mod; |
11863 | cnt = isdiv ? ARRAY_SIZE(chk_and_div) : | |
9b00f1b7 | 11864 | ARRAY_SIZE(chk_and_mod) - (is64 ? 2 : 0); |
f6b1b3bf DB |
11865 | |
11866 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
11867 | if (!new_prog) |
11868 | return -ENOMEM; | |
11869 | ||
11870 | delta += cnt - 1; | |
11871 | env->prog = prog = new_prog; | |
11872 | insn = new_prog->insnsi + i + delta; | |
11873 | continue; | |
11874 | } | |
11875 | ||
e6ac5933 | 11876 | /* Implement LD_ABS and LD_IND with a rewrite, if supported by the program type. */ |
e0cea7ce DB |
11877 | if (BPF_CLASS(insn->code) == BPF_LD && |
11878 | (BPF_MODE(insn->code) == BPF_ABS || | |
11879 | BPF_MODE(insn->code) == BPF_IND)) { | |
11880 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
11881 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
11882 | verbose(env, "bpf verifier is misconfigured\n"); | |
11883 | return -EINVAL; | |
11884 | } | |
11885 | ||
11886 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
11887 | if (!new_prog) | |
11888 | return -ENOMEM; | |
11889 | ||
11890 | delta += cnt - 1; | |
11891 | env->prog = prog = new_prog; | |
11892 | insn = new_prog->insnsi + i + delta; | |
11893 | continue; | |
11894 | } | |
11895 | ||
e6ac5933 | 11896 | /* Rewrite pointer arithmetic to mitigate speculation attacks. */ |
979d63d5 DB |
11897 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
11898 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
11899 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
11900 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
979d63d5 DB |
11901 | struct bpf_insn *patch = &insn_buf[0]; |
11902 | bool issrc, isneg; | |
11903 | u32 off_reg; | |
11904 | ||
11905 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
11906 | if (!aux->alu_state || |
11907 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
11908 | continue; |
11909 | ||
11910 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
11911 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
11912 | BPF_ALU_SANITIZE_SRC; | |
11913 | ||
11914 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
11915 | if (isneg) | |
11916 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
b5871dca | 11917 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit); |
979d63d5 DB |
11918 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); |
11919 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
11920 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
11921 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
11922 | if (issrc) { | |
11923 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, | |
11924 | off_reg); | |
11925 | insn->src_reg = BPF_REG_AX; | |
11926 | } else { | |
11927 | *patch++ = BPF_ALU64_REG(BPF_AND, off_reg, | |
11928 | BPF_REG_AX); | |
11929 | } | |
11930 | if (isneg) | |
11931 | insn->code = insn->code == code_add ? | |
11932 | code_sub : code_add; | |
11933 | *patch++ = *insn; | |
11934 | if (issrc && isneg) | |
11935 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
11936 | cnt = patch - insn_buf; | |
11937 | ||
11938 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
11939 | if (!new_prog) | |
11940 | return -ENOMEM; | |
11941 | ||
11942 | delta += cnt - 1; | |
11943 | env->prog = prog = new_prog; | |
11944 | insn = new_prog->insnsi + i + delta; | |
11945 | continue; | |
11946 | } | |
11947 | ||
79741b3b AS |
11948 | if (insn->code != (BPF_JMP | BPF_CALL)) |
11949 | continue; | |
cc8b0b92 AS |
11950 | if (insn->src_reg == BPF_PSEUDO_CALL) |
11951 | continue; | |
e245c5c6 | 11952 | |
79741b3b AS |
11953 | if (insn->imm == BPF_FUNC_get_route_realm) |
11954 | prog->dst_needed = 1; | |
11955 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
11956 | bpf_user_rnd_init_once(); | |
9802d865 JB |
11957 | if (insn->imm == BPF_FUNC_override_return) |
11958 | prog->kprobe_override = 1; | |
79741b3b | 11959 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
11960 | /* If we tail call into other programs, we |
11961 | * cannot make any assumptions since they can | |
11962 | * be replaced dynamically during runtime in | |
11963 | * the program array. | |
11964 | */ | |
11965 | prog->cb_access = 1; | |
e411901c MF |
11966 | if (!allow_tail_call_in_subprogs(env)) |
11967 | prog->aux->stack_depth = MAX_BPF_STACK; | |
11968 | prog->aux->max_pkt_offset = MAX_PACKET_OFF; | |
7b9f6da1 | 11969 | |
79741b3b AS |
11970 | /* mark bpf_tail_call as different opcode to avoid |
11971 | * conditional branch in the interpeter for every normal | |
11972 | * call and to prevent accidental JITing by JIT compiler | |
11973 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 11974 | */ |
79741b3b | 11975 | insn->imm = 0; |
71189fa9 | 11976 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 11977 | |
c93552c4 | 11978 | aux = &env->insn_aux_data[i + delta]; |
2c78ee89 | 11979 | if (env->bpf_capable && !expect_blinding && |
cc52d914 | 11980 | prog->jit_requested && |
d2e4c1e6 DB |
11981 | !bpf_map_key_poisoned(aux) && |
11982 | !bpf_map_ptr_poisoned(aux) && | |
11983 | !bpf_map_ptr_unpriv(aux)) { | |
11984 | struct bpf_jit_poke_descriptor desc = { | |
11985 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
11986 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
11987 | .tail_call.key = bpf_map_key_immediate(aux), | |
a748c697 | 11988 | .insn_idx = i + delta, |
d2e4c1e6 DB |
11989 | }; |
11990 | ||
11991 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
11992 | if (ret < 0) { | |
11993 | verbose(env, "adding tail call poke descriptor failed\n"); | |
11994 | return ret; | |
11995 | } | |
11996 | ||
11997 | insn->imm = ret + 1; | |
11998 | continue; | |
11999 | } | |
12000 | ||
c93552c4 DB |
12001 | if (!bpf_map_ptr_unpriv(aux)) |
12002 | continue; | |
12003 | ||
b2157399 AS |
12004 | /* instead of changing every JIT dealing with tail_call |
12005 | * emit two extra insns: | |
12006 | * if (index >= max_entries) goto out; | |
12007 | * index &= array->index_mask; | |
12008 | * to avoid out-of-bounds cpu speculation | |
12009 | */ | |
c93552c4 | 12010 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 12011 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
12012 | return -EINVAL; |
12013 | } | |
c93552c4 | 12014 | |
d2e4c1e6 | 12015 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
12016 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
12017 | map_ptr->max_entries, 2); | |
12018 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
12019 | container_of(map_ptr, | |
12020 | struct bpf_array, | |
12021 | map)->index_mask); | |
12022 | insn_buf[2] = *insn; | |
12023 | cnt = 3; | |
12024 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
12025 | if (!new_prog) | |
12026 | return -ENOMEM; | |
12027 | ||
12028 | delta += cnt - 1; | |
12029 | env->prog = prog = new_prog; | |
12030 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
12031 | continue; |
12032 | } | |
e245c5c6 | 12033 | |
89c63074 | 12034 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
12035 | * and other inlining handlers are currently limited to 64 bit |
12036 | * only. | |
89c63074 | 12037 | */ |
60b58afc | 12038 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
12039 | (insn->imm == BPF_FUNC_map_lookup_elem || |
12040 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
12041 | insn->imm == BPF_FUNC_map_delete_elem || |
12042 | insn->imm == BPF_FUNC_map_push_elem || | |
12043 | insn->imm == BPF_FUNC_map_pop_elem || | |
e6a4750f BT |
12044 | insn->imm == BPF_FUNC_map_peek_elem || |
12045 | insn->imm == BPF_FUNC_redirect_map)) { | |
c93552c4 DB |
12046 | aux = &env->insn_aux_data[i + delta]; |
12047 | if (bpf_map_ptr_poisoned(aux)) | |
12048 | goto patch_call_imm; | |
12049 | ||
d2e4c1e6 | 12050 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
12051 | ops = map_ptr->ops; |
12052 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
12053 | ops->map_gen_lookup) { | |
12054 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
4a8f87e6 DB |
12055 | if (cnt == -EOPNOTSUPP) |
12056 | goto patch_map_ops_generic; | |
12057 | if (cnt <= 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
09772d92 DB |
12058 | verbose(env, "bpf verifier is misconfigured\n"); |
12059 | return -EINVAL; | |
12060 | } | |
81ed18ab | 12061 | |
09772d92 DB |
12062 | new_prog = bpf_patch_insn_data(env, i + delta, |
12063 | insn_buf, cnt); | |
12064 | if (!new_prog) | |
12065 | return -ENOMEM; | |
81ed18ab | 12066 | |
09772d92 DB |
12067 | delta += cnt - 1; |
12068 | env->prog = prog = new_prog; | |
12069 | insn = new_prog->insnsi + i + delta; | |
12070 | continue; | |
12071 | } | |
81ed18ab | 12072 | |
09772d92 DB |
12073 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
12074 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
12075 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
12076 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
12077 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
12078 | (int (*)(struct bpf_map *map, void *key, void *value, | |
12079 | u64 flags))NULL)); | |
84430d42 DB |
12080 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
12081 | (int (*)(struct bpf_map *map, void *value, | |
12082 | u64 flags))NULL)); | |
12083 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
12084 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
12085 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
12086 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
e6a4750f BT |
12087 | BUILD_BUG_ON(!__same_type(ops->map_redirect, |
12088 | (int (*)(struct bpf_map *map, u32 ifindex, u64 flags))NULL)); | |
12089 | ||
4a8f87e6 | 12090 | patch_map_ops_generic: |
09772d92 DB |
12091 | switch (insn->imm) { |
12092 | case BPF_FUNC_map_lookup_elem: | |
12093 | insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) - | |
12094 | __bpf_call_base; | |
12095 | continue; | |
12096 | case BPF_FUNC_map_update_elem: | |
12097 | insn->imm = BPF_CAST_CALL(ops->map_update_elem) - | |
12098 | __bpf_call_base; | |
12099 | continue; | |
12100 | case BPF_FUNC_map_delete_elem: | |
12101 | insn->imm = BPF_CAST_CALL(ops->map_delete_elem) - | |
12102 | __bpf_call_base; | |
12103 | continue; | |
84430d42 DB |
12104 | case BPF_FUNC_map_push_elem: |
12105 | insn->imm = BPF_CAST_CALL(ops->map_push_elem) - | |
12106 | __bpf_call_base; | |
12107 | continue; | |
12108 | case BPF_FUNC_map_pop_elem: | |
12109 | insn->imm = BPF_CAST_CALL(ops->map_pop_elem) - | |
12110 | __bpf_call_base; | |
12111 | continue; | |
12112 | case BPF_FUNC_map_peek_elem: | |
12113 | insn->imm = BPF_CAST_CALL(ops->map_peek_elem) - | |
12114 | __bpf_call_base; | |
12115 | continue; | |
e6a4750f BT |
12116 | case BPF_FUNC_redirect_map: |
12117 | insn->imm = BPF_CAST_CALL(ops->map_redirect) - | |
12118 | __bpf_call_base; | |
12119 | continue; | |
09772d92 | 12120 | } |
81ed18ab | 12121 | |
09772d92 | 12122 | goto patch_call_imm; |
81ed18ab AS |
12123 | } |
12124 | ||
e6ac5933 | 12125 | /* Implement bpf_jiffies64 inline. */ |
5576b991 MKL |
12126 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
12127 | insn->imm == BPF_FUNC_jiffies64) { | |
12128 | struct bpf_insn ld_jiffies_addr[2] = { | |
12129 | BPF_LD_IMM64(BPF_REG_0, | |
12130 | (unsigned long)&jiffies), | |
12131 | }; | |
12132 | ||
12133 | insn_buf[0] = ld_jiffies_addr[0]; | |
12134 | insn_buf[1] = ld_jiffies_addr[1]; | |
12135 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
12136 | BPF_REG_0, 0); | |
12137 | cnt = 3; | |
12138 | ||
12139 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
12140 | cnt); | |
12141 | if (!new_prog) | |
12142 | return -ENOMEM; | |
12143 | ||
12144 | delta += cnt - 1; | |
12145 | env->prog = prog = new_prog; | |
12146 | insn = new_prog->insnsi + i + delta; | |
12147 | continue; | |
12148 | } | |
12149 | ||
81ed18ab | 12150 | patch_call_imm: |
5e43f899 | 12151 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
12152 | /* all functions that have prototype and verifier allowed |
12153 | * programs to call them, must be real in-kernel functions | |
12154 | */ | |
12155 | if (!fn->func) { | |
61bd5218 JK |
12156 | verbose(env, |
12157 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
12158 | func_id_name(insn->imm), insn->imm); |
12159 | return -EFAULT; | |
e245c5c6 | 12160 | } |
79741b3b | 12161 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 12162 | } |
e245c5c6 | 12163 | |
d2e4c1e6 DB |
12164 | /* Since poke tab is now finalized, publish aux to tracker. */ |
12165 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
12166 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
12167 | if (!map_ptr->ops->map_poke_track || | |
12168 | !map_ptr->ops->map_poke_untrack || | |
12169 | !map_ptr->ops->map_poke_run) { | |
12170 | verbose(env, "bpf verifier is misconfigured\n"); | |
12171 | return -EINVAL; | |
12172 | } | |
12173 | ||
12174 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
12175 | if (ret < 0) { | |
12176 | verbose(env, "tracking tail call prog failed\n"); | |
12177 | return ret; | |
12178 | } | |
12179 | } | |
12180 | ||
79741b3b AS |
12181 | return 0; |
12182 | } | |
e245c5c6 | 12183 | |
58e2af8b | 12184 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 12185 | { |
58e2af8b | 12186 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
12187 | int i; |
12188 | ||
9f4686c4 AS |
12189 | sl = env->free_list; |
12190 | while (sl) { | |
12191 | sln = sl->next; | |
12192 | free_verifier_state(&sl->state, false); | |
12193 | kfree(sl); | |
12194 | sl = sln; | |
12195 | } | |
51c39bb1 | 12196 | env->free_list = NULL; |
9f4686c4 | 12197 | |
f1bca824 AS |
12198 | if (!env->explored_states) |
12199 | return; | |
12200 | ||
dc2a4ebc | 12201 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
12202 | sl = env->explored_states[i]; |
12203 | ||
a8f500af AS |
12204 | while (sl) { |
12205 | sln = sl->next; | |
12206 | free_verifier_state(&sl->state, false); | |
12207 | kfree(sl); | |
12208 | sl = sln; | |
12209 | } | |
51c39bb1 | 12210 | env->explored_states[i] = NULL; |
f1bca824 | 12211 | } |
51c39bb1 | 12212 | } |
f1bca824 | 12213 | |
51c39bb1 AS |
12214 | /* The verifier is using insn_aux_data[] to store temporary data during |
12215 | * verification and to store information for passes that run after the | |
12216 | * verification like dead code sanitization. do_check_common() for subprogram N | |
12217 | * may analyze many other subprograms. sanitize_insn_aux_data() clears all | |
12218 | * temporary data after do_check_common() finds that subprogram N cannot be | |
12219 | * verified independently. pass_cnt counts the number of times | |
12220 | * do_check_common() was run and insn->aux->seen tells the pass number | |
12221 | * insn_aux_data was touched. These variables are compared to clear temporary | |
12222 | * data from failed pass. For testing and experiments do_check_common() can be | |
12223 | * run multiple times even when prior attempt to verify is unsuccessful. | |
12224 | */ | |
12225 | static void sanitize_insn_aux_data(struct bpf_verifier_env *env) | |
12226 | { | |
12227 | struct bpf_insn *insn = env->prog->insnsi; | |
12228 | struct bpf_insn_aux_data *aux; | |
12229 | int i, class; | |
12230 | ||
12231 | for (i = 0; i < env->prog->len; i++) { | |
12232 | class = BPF_CLASS(insn[i].code); | |
12233 | if (class != BPF_LDX && class != BPF_STX) | |
12234 | continue; | |
12235 | aux = &env->insn_aux_data[i]; | |
12236 | if (aux->seen != env->pass_cnt) | |
12237 | continue; | |
12238 | memset(aux, 0, offsetof(typeof(*aux), orig_idx)); | |
12239 | } | |
f1bca824 AS |
12240 | } |
12241 | ||
51c39bb1 AS |
12242 | static int do_check_common(struct bpf_verifier_env *env, int subprog) |
12243 | { | |
6f8a57cc | 12244 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
12245 | struct bpf_verifier_state *state; |
12246 | struct bpf_reg_state *regs; | |
12247 | int ret, i; | |
12248 | ||
12249 | env->prev_linfo = NULL; | |
12250 | env->pass_cnt++; | |
12251 | ||
12252 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
12253 | if (!state) | |
12254 | return -ENOMEM; | |
12255 | state->curframe = 0; | |
12256 | state->speculative = false; | |
12257 | state->branches = 1; | |
12258 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
12259 | if (!state->frame[0]) { | |
12260 | kfree(state); | |
12261 | return -ENOMEM; | |
12262 | } | |
12263 | env->cur_state = state; | |
12264 | init_func_state(env, state->frame[0], | |
12265 | BPF_MAIN_FUNC /* callsite */, | |
12266 | 0 /* frameno */, | |
12267 | subprog); | |
12268 | ||
12269 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 12270 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
51c39bb1 AS |
12271 | ret = btf_prepare_func_args(env, subprog, regs); |
12272 | if (ret) | |
12273 | goto out; | |
12274 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
12275 | if (regs[i].type == PTR_TO_CTX) | |
12276 | mark_reg_known_zero(env, regs, i); | |
12277 | else if (regs[i].type == SCALAR_VALUE) | |
12278 | mark_reg_unknown(env, regs, i); | |
e5069b9c DB |
12279 | else if (regs[i].type == PTR_TO_MEM_OR_NULL) { |
12280 | const u32 mem_size = regs[i].mem_size; | |
12281 | ||
12282 | mark_reg_known_zero(env, regs, i); | |
12283 | regs[i].mem_size = mem_size; | |
12284 | regs[i].id = ++env->id_gen; | |
12285 | } | |
51c39bb1 AS |
12286 | } |
12287 | } else { | |
12288 | /* 1st arg to a function */ | |
12289 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
12290 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
12291 | ret = btf_check_func_arg_match(env, subprog, regs); | |
12292 | if (ret == -EFAULT) | |
12293 | /* unlikely verifier bug. abort. | |
12294 | * ret == 0 and ret < 0 are sadly acceptable for | |
12295 | * main() function due to backward compatibility. | |
12296 | * Like socket filter program may be written as: | |
12297 | * int bpf_prog(struct pt_regs *ctx) | |
12298 | * and never dereference that ctx in the program. | |
12299 | * 'struct pt_regs' is a type mismatch for socket | |
12300 | * filter that should be using 'struct __sk_buff'. | |
12301 | */ | |
12302 | goto out; | |
12303 | } | |
12304 | ||
12305 | ret = do_check(env); | |
12306 | out: | |
f59bbfc2 AS |
12307 | /* check for NULL is necessary, since cur_state can be freed inside |
12308 | * do_check() under memory pressure. | |
12309 | */ | |
12310 | if (env->cur_state) { | |
12311 | free_verifier_state(env->cur_state, true); | |
12312 | env->cur_state = NULL; | |
12313 | } | |
6f8a57cc AN |
12314 | while (!pop_stack(env, NULL, NULL, false)); |
12315 | if (!ret && pop_log) | |
12316 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 AS |
12317 | free_states(env); |
12318 | if (ret) | |
12319 | /* clean aux data in case subprog was rejected */ | |
12320 | sanitize_insn_aux_data(env); | |
12321 | return ret; | |
12322 | } | |
12323 | ||
12324 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
12325 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
12326 | * Consider: | |
12327 | * int bar(int); | |
12328 | * int foo(int f) | |
12329 | * { | |
12330 | * return bar(f); | |
12331 | * } | |
12332 | * int bar(int b) | |
12333 | * { | |
12334 | * ... | |
12335 | * } | |
12336 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
12337 | * will be assumed that bar() already verified successfully and call to bar() | |
12338 | * from foo() will be checked for type match only. Later bar() will be verified | |
12339 | * independently to check that it's safe for R1=any_scalar_value. | |
12340 | */ | |
12341 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
12342 | { | |
12343 | struct bpf_prog_aux *aux = env->prog->aux; | |
12344 | int i, ret; | |
12345 | ||
12346 | if (!aux->func_info) | |
12347 | return 0; | |
12348 | ||
12349 | for (i = 1; i < env->subprog_cnt; i++) { | |
12350 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
12351 | continue; | |
12352 | env->insn_idx = env->subprog_info[i].start; | |
12353 | WARN_ON_ONCE(env->insn_idx == 0); | |
12354 | ret = do_check_common(env, i); | |
12355 | if (ret) { | |
12356 | return ret; | |
12357 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
12358 | verbose(env, | |
12359 | "Func#%d is safe for any args that match its prototype\n", | |
12360 | i); | |
12361 | } | |
12362 | } | |
12363 | return 0; | |
12364 | } | |
12365 | ||
12366 | static int do_check_main(struct bpf_verifier_env *env) | |
12367 | { | |
12368 | int ret; | |
12369 | ||
12370 | env->insn_idx = 0; | |
12371 | ret = do_check_common(env, 0); | |
12372 | if (!ret) | |
12373 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
12374 | return ret; | |
12375 | } | |
12376 | ||
12377 | ||
06ee7115 AS |
12378 | static void print_verification_stats(struct bpf_verifier_env *env) |
12379 | { | |
12380 | int i; | |
12381 | ||
12382 | if (env->log.level & BPF_LOG_STATS) { | |
12383 | verbose(env, "verification time %lld usec\n", | |
12384 | div_u64(env->verification_time, 1000)); | |
12385 | verbose(env, "stack depth "); | |
12386 | for (i = 0; i < env->subprog_cnt; i++) { | |
12387 | u32 depth = env->subprog_info[i].stack_depth; | |
12388 | ||
12389 | verbose(env, "%d", depth); | |
12390 | if (i + 1 < env->subprog_cnt) | |
12391 | verbose(env, "+"); | |
12392 | } | |
12393 | verbose(env, "\n"); | |
12394 | } | |
12395 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
12396 | "total_states %d peak_states %d mark_read %d\n", | |
12397 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
12398 | env->max_states_per_insn, env->total_states, | |
12399 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
12400 | } |
12401 | ||
27ae7997 MKL |
12402 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
12403 | { | |
12404 | const struct btf_type *t, *func_proto; | |
12405 | const struct bpf_struct_ops *st_ops; | |
12406 | const struct btf_member *member; | |
12407 | struct bpf_prog *prog = env->prog; | |
12408 | u32 btf_id, member_idx; | |
12409 | const char *mname; | |
12410 | ||
12411 | btf_id = prog->aux->attach_btf_id; | |
12412 | st_ops = bpf_struct_ops_find(btf_id); | |
12413 | if (!st_ops) { | |
12414 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
12415 | btf_id); | |
12416 | return -ENOTSUPP; | |
12417 | } | |
12418 | ||
12419 | t = st_ops->type; | |
12420 | member_idx = prog->expected_attach_type; | |
12421 | if (member_idx >= btf_type_vlen(t)) { | |
12422 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
12423 | member_idx, st_ops->name); | |
12424 | return -EINVAL; | |
12425 | } | |
12426 | ||
12427 | member = &btf_type_member(t)[member_idx]; | |
12428 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
12429 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
12430 | NULL); | |
12431 | if (!func_proto) { | |
12432 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
12433 | mname, member_idx, st_ops->name); | |
12434 | return -EINVAL; | |
12435 | } | |
12436 | ||
12437 | if (st_ops->check_member) { | |
12438 | int err = st_ops->check_member(t, member); | |
12439 | ||
12440 | if (err) { | |
12441 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
12442 | mname, st_ops->name); | |
12443 | return err; | |
12444 | } | |
12445 | } | |
12446 | ||
12447 | prog->aux->attach_func_proto = func_proto; | |
12448 | prog->aux->attach_func_name = mname; | |
12449 | env->ops = st_ops->verifier_ops; | |
12450 | ||
12451 | return 0; | |
12452 | } | |
6ba43b76 KS |
12453 | #define SECURITY_PREFIX "security_" |
12454 | ||
f7b12b6f | 12455 | static int check_attach_modify_return(unsigned long addr, const char *func_name) |
6ba43b76 | 12456 | { |
69191754 | 12457 | if (within_error_injection_list(addr) || |
f7b12b6f | 12458 | !strncmp(SECURITY_PREFIX, func_name, sizeof(SECURITY_PREFIX) - 1)) |
6ba43b76 | 12459 | return 0; |
6ba43b76 | 12460 | |
6ba43b76 KS |
12461 | return -EINVAL; |
12462 | } | |
27ae7997 | 12463 | |
1e6c62a8 AS |
12464 | /* list of non-sleepable functions that are otherwise on |
12465 | * ALLOW_ERROR_INJECTION list | |
12466 | */ | |
12467 | BTF_SET_START(btf_non_sleepable_error_inject) | |
12468 | /* Three functions below can be called from sleepable and non-sleepable context. | |
12469 | * Assume non-sleepable from bpf safety point of view. | |
12470 | */ | |
12471 | BTF_ID(func, __add_to_page_cache_locked) | |
12472 | BTF_ID(func, should_fail_alloc_page) | |
12473 | BTF_ID(func, should_failslab) | |
12474 | BTF_SET_END(btf_non_sleepable_error_inject) | |
12475 | ||
12476 | static int check_non_sleepable_error_inject(u32 btf_id) | |
12477 | { | |
12478 | return btf_id_set_contains(&btf_non_sleepable_error_inject, btf_id); | |
12479 | } | |
12480 | ||
f7b12b6f THJ |
12481 | int bpf_check_attach_target(struct bpf_verifier_log *log, |
12482 | const struct bpf_prog *prog, | |
12483 | const struct bpf_prog *tgt_prog, | |
12484 | u32 btf_id, | |
12485 | struct bpf_attach_target_info *tgt_info) | |
38207291 | 12486 | { |
be8704ff | 12487 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
f1b9509c | 12488 | const char prefix[] = "btf_trace_"; |
5b92a28a | 12489 | int ret = 0, subprog = -1, i; |
38207291 | 12490 | const struct btf_type *t; |
5b92a28a | 12491 | bool conservative = true; |
38207291 | 12492 | const char *tname; |
5b92a28a | 12493 | struct btf *btf; |
f7b12b6f | 12494 | long addr = 0; |
38207291 | 12495 | |
f1b9509c | 12496 | if (!btf_id) { |
efc68158 | 12497 | bpf_log(log, "Tracing programs must provide btf_id\n"); |
f1b9509c AS |
12498 | return -EINVAL; |
12499 | } | |
22dc4a0f | 12500 | btf = tgt_prog ? tgt_prog->aux->btf : prog->aux->attach_btf; |
5b92a28a | 12501 | if (!btf) { |
efc68158 | 12502 | bpf_log(log, |
5b92a28a AS |
12503 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); |
12504 | return -EINVAL; | |
12505 | } | |
12506 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c | 12507 | if (!t) { |
efc68158 | 12508 | bpf_log(log, "attach_btf_id %u is invalid\n", btf_id); |
f1b9509c AS |
12509 | return -EINVAL; |
12510 | } | |
5b92a28a | 12511 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c | 12512 | if (!tname) { |
efc68158 | 12513 | bpf_log(log, "attach_btf_id %u doesn't have a name\n", btf_id); |
f1b9509c AS |
12514 | return -EINVAL; |
12515 | } | |
5b92a28a AS |
12516 | if (tgt_prog) { |
12517 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
12518 | ||
12519 | for (i = 0; i < aux->func_info_cnt; i++) | |
12520 | if (aux->func_info[i].type_id == btf_id) { | |
12521 | subprog = i; | |
12522 | break; | |
12523 | } | |
12524 | if (subprog == -1) { | |
efc68158 | 12525 | bpf_log(log, "Subprog %s doesn't exist\n", tname); |
5b92a28a AS |
12526 | return -EINVAL; |
12527 | } | |
12528 | conservative = aux->func_info_aux[subprog].unreliable; | |
be8704ff AS |
12529 | if (prog_extension) { |
12530 | if (conservative) { | |
efc68158 | 12531 | bpf_log(log, |
be8704ff AS |
12532 | "Cannot replace static functions\n"); |
12533 | return -EINVAL; | |
12534 | } | |
12535 | if (!prog->jit_requested) { | |
efc68158 | 12536 | bpf_log(log, |
be8704ff AS |
12537 | "Extension programs should be JITed\n"); |
12538 | return -EINVAL; | |
12539 | } | |
be8704ff AS |
12540 | } |
12541 | if (!tgt_prog->jited) { | |
efc68158 | 12542 | bpf_log(log, "Can attach to only JITed progs\n"); |
be8704ff AS |
12543 | return -EINVAL; |
12544 | } | |
12545 | if (tgt_prog->type == prog->type) { | |
12546 | /* Cannot fentry/fexit another fentry/fexit program. | |
12547 | * Cannot attach program extension to another extension. | |
12548 | * It's ok to attach fentry/fexit to extension program. | |
12549 | */ | |
efc68158 | 12550 | bpf_log(log, "Cannot recursively attach\n"); |
be8704ff AS |
12551 | return -EINVAL; |
12552 | } | |
12553 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
12554 | prog_extension && | |
12555 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
12556 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
12557 | /* Program extensions can extend all program types | |
12558 | * except fentry/fexit. The reason is the following. | |
12559 | * The fentry/fexit programs are used for performance | |
12560 | * analysis, stats and can be attached to any program | |
12561 | * type except themselves. When extension program is | |
12562 | * replacing XDP function it is necessary to allow | |
12563 | * performance analysis of all functions. Both original | |
12564 | * XDP program and its program extension. Hence | |
12565 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
12566 | * allowed. If extending of fentry/fexit was allowed it | |
12567 | * would be possible to create long call chain | |
12568 | * fentry->extension->fentry->extension beyond | |
12569 | * reasonable stack size. Hence extending fentry is not | |
12570 | * allowed. | |
12571 | */ | |
efc68158 | 12572 | bpf_log(log, "Cannot extend fentry/fexit\n"); |
be8704ff AS |
12573 | return -EINVAL; |
12574 | } | |
5b92a28a | 12575 | } else { |
be8704ff | 12576 | if (prog_extension) { |
efc68158 | 12577 | bpf_log(log, "Cannot replace kernel functions\n"); |
be8704ff AS |
12578 | return -EINVAL; |
12579 | } | |
5b92a28a | 12580 | } |
f1b9509c AS |
12581 | |
12582 | switch (prog->expected_attach_type) { | |
12583 | case BPF_TRACE_RAW_TP: | |
5b92a28a | 12584 | if (tgt_prog) { |
efc68158 | 12585 | bpf_log(log, |
5b92a28a AS |
12586 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); |
12587 | return -EINVAL; | |
12588 | } | |
38207291 | 12589 | if (!btf_type_is_typedef(t)) { |
efc68158 | 12590 | bpf_log(log, "attach_btf_id %u is not a typedef\n", |
38207291 MKL |
12591 | btf_id); |
12592 | return -EINVAL; | |
12593 | } | |
f1b9509c | 12594 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
efc68158 | 12595 | bpf_log(log, "attach_btf_id %u points to wrong type name %s\n", |
38207291 MKL |
12596 | btf_id, tname); |
12597 | return -EINVAL; | |
12598 | } | |
12599 | tname += sizeof(prefix) - 1; | |
5b92a28a | 12600 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
12601 | if (!btf_type_is_ptr(t)) |
12602 | /* should never happen in valid vmlinux build */ | |
12603 | return -EINVAL; | |
5b92a28a | 12604 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
12605 | if (!btf_type_is_func_proto(t)) |
12606 | /* should never happen in valid vmlinux build */ | |
12607 | return -EINVAL; | |
12608 | ||
f7b12b6f | 12609 | break; |
15d83c4d YS |
12610 | case BPF_TRACE_ITER: |
12611 | if (!btf_type_is_func(t)) { | |
efc68158 | 12612 | bpf_log(log, "attach_btf_id %u is not a function\n", |
15d83c4d YS |
12613 | btf_id); |
12614 | return -EINVAL; | |
12615 | } | |
12616 | t = btf_type_by_id(btf, t->type); | |
12617 | if (!btf_type_is_func_proto(t)) | |
12618 | return -EINVAL; | |
f7b12b6f THJ |
12619 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); |
12620 | if (ret) | |
12621 | return ret; | |
12622 | break; | |
be8704ff AS |
12623 | default: |
12624 | if (!prog_extension) | |
12625 | return -EINVAL; | |
df561f66 | 12626 | fallthrough; |
ae240823 | 12627 | case BPF_MODIFY_RETURN: |
9e4e01df | 12628 | case BPF_LSM_MAC: |
fec56f58 AS |
12629 | case BPF_TRACE_FENTRY: |
12630 | case BPF_TRACE_FEXIT: | |
12631 | if (!btf_type_is_func(t)) { | |
efc68158 | 12632 | bpf_log(log, "attach_btf_id %u is not a function\n", |
fec56f58 AS |
12633 | btf_id); |
12634 | return -EINVAL; | |
12635 | } | |
be8704ff | 12636 | if (prog_extension && |
efc68158 | 12637 | btf_check_type_match(log, prog, btf, t)) |
be8704ff | 12638 | return -EINVAL; |
5b92a28a | 12639 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
12640 | if (!btf_type_is_func_proto(t)) |
12641 | return -EINVAL; | |
f7b12b6f | 12642 | |
4a1e7c0c THJ |
12643 | if ((prog->aux->saved_dst_prog_type || prog->aux->saved_dst_attach_type) && |
12644 | (!tgt_prog || prog->aux->saved_dst_prog_type != tgt_prog->type || | |
12645 | prog->aux->saved_dst_attach_type != tgt_prog->expected_attach_type)) | |
12646 | return -EINVAL; | |
12647 | ||
f7b12b6f | 12648 | if (tgt_prog && conservative) |
5b92a28a | 12649 | t = NULL; |
f7b12b6f THJ |
12650 | |
12651 | ret = btf_distill_func_proto(log, btf, t, tname, &tgt_info->fmodel); | |
fec56f58 | 12652 | if (ret < 0) |
f7b12b6f THJ |
12653 | return ret; |
12654 | ||
5b92a28a | 12655 | if (tgt_prog) { |
e9eeec58 YS |
12656 | if (subprog == 0) |
12657 | addr = (long) tgt_prog->bpf_func; | |
12658 | else | |
12659 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a AS |
12660 | } else { |
12661 | addr = kallsyms_lookup_name(tname); | |
12662 | if (!addr) { | |
efc68158 | 12663 | bpf_log(log, |
5b92a28a AS |
12664 | "The address of function %s cannot be found\n", |
12665 | tname); | |
f7b12b6f | 12666 | return -ENOENT; |
5b92a28a | 12667 | } |
fec56f58 | 12668 | } |
18644cec | 12669 | |
1e6c62a8 AS |
12670 | if (prog->aux->sleepable) { |
12671 | ret = -EINVAL; | |
12672 | switch (prog->type) { | |
12673 | case BPF_PROG_TYPE_TRACING: | |
12674 | /* fentry/fexit/fmod_ret progs can be sleepable only if they are | |
12675 | * attached to ALLOW_ERROR_INJECTION and are not in denylist. | |
12676 | */ | |
12677 | if (!check_non_sleepable_error_inject(btf_id) && | |
12678 | within_error_injection_list(addr)) | |
12679 | ret = 0; | |
12680 | break; | |
12681 | case BPF_PROG_TYPE_LSM: | |
12682 | /* LSM progs check that they are attached to bpf_lsm_*() funcs. | |
12683 | * Only some of them are sleepable. | |
12684 | */ | |
423f1610 | 12685 | if (bpf_lsm_is_sleepable_hook(btf_id)) |
1e6c62a8 AS |
12686 | ret = 0; |
12687 | break; | |
12688 | default: | |
12689 | break; | |
12690 | } | |
f7b12b6f THJ |
12691 | if (ret) { |
12692 | bpf_log(log, "%s is not sleepable\n", tname); | |
12693 | return ret; | |
12694 | } | |
1e6c62a8 | 12695 | } else if (prog->expected_attach_type == BPF_MODIFY_RETURN) { |
1af9270e | 12696 | if (tgt_prog) { |
efc68158 | 12697 | bpf_log(log, "can't modify return codes of BPF programs\n"); |
f7b12b6f THJ |
12698 | return -EINVAL; |
12699 | } | |
12700 | ret = check_attach_modify_return(addr, tname); | |
12701 | if (ret) { | |
12702 | bpf_log(log, "%s() is not modifiable\n", tname); | |
12703 | return ret; | |
1af9270e | 12704 | } |
18644cec | 12705 | } |
f7b12b6f THJ |
12706 | |
12707 | break; | |
12708 | } | |
12709 | tgt_info->tgt_addr = addr; | |
12710 | tgt_info->tgt_name = tname; | |
12711 | tgt_info->tgt_type = t; | |
12712 | return 0; | |
12713 | } | |
12714 | ||
12715 | static int check_attach_btf_id(struct bpf_verifier_env *env) | |
12716 | { | |
12717 | struct bpf_prog *prog = env->prog; | |
3aac1ead | 12718 | struct bpf_prog *tgt_prog = prog->aux->dst_prog; |
f7b12b6f THJ |
12719 | struct bpf_attach_target_info tgt_info = {}; |
12720 | u32 btf_id = prog->aux->attach_btf_id; | |
12721 | struct bpf_trampoline *tr; | |
12722 | int ret; | |
12723 | u64 key; | |
12724 | ||
12725 | if (prog->aux->sleepable && prog->type != BPF_PROG_TYPE_TRACING && | |
12726 | prog->type != BPF_PROG_TYPE_LSM) { | |
12727 | verbose(env, "Only fentry/fexit/fmod_ret and lsm programs can be sleepable\n"); | |
12728 | return -EINVAL; | |
12729 | } | |
12730 | ||
12731 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) | |
12732 | return check_struct_ops_btf_id(env); | |
12733 | ||
12734 | if (prog->type != BPF_PROG_TYPE_TRACING && | |
12735 | prog->type != BPF_PROG_TYPE_LSM && | |
12736 | prog->type != BPF_PROG_TYPE_EXT) | |
12737 | return 0; | |
12738 | ||
12739 | ret = bpf_check_attach_target(&env->log, prog, tgt_prog, btf_id, &tgt_info); | |
12740 | if (ret) | |
fec56f58 | 12741 | return ret; |
f7b12b6f THJ |
12742 | |
12743 | if (tgt_prog && prog->type == BPF_PROG_TYPE_EXT) { | |
3aac1ead THJ |
12744 | /* to make freplace equivalent to their targets, they need to |
12745 | * inherit env->ops and expected_attach_type for the rest of the | |
12746 | * verification | |
12747 | */ | |
f7b12b6f THJ |
12748 | env->ops = bpf_verifier_ops[tgt_prog->type]; |
12749 | prog->expected_attach_type = tgt_prog->expected_attach_type; | |
12750 | } | |
12751 | ||
12752 | /* store info about the attachment target that will be used later */ | |
12753 | prog->aux->attach_func_proto = tgt_info.tgt_type; | |
12754 | prog->aux->attach_func_name = tgt_info.tgt_name; | |
12755 | ||
4a1e7c0c THJ |
12756 | if (tgt_prog) { |
12757 | prog->aux->saved_dst_prog_type = tgt_prog->type; | |
12758 | prog->aux->saved_dst_attach_type = tgt_prog->expected_attach_type; | |
12759 | } | |
12760 | ||
f7b12b6f THJ |
12761 | if (prog->expected_attach_type == BPF_TRACE_RAW_TP) { |
12762 | prog->aux->attach_btf_trace = true; | |
12763 | return 0; | |
12764 | } else if (prog->expected_attach_type == BPF_TRACE_ITER) { | |
12765 | if (!bpf_iter_prog_supported(prog)) | |
12766 | return -EINVAL; | |
12767 | return 0; | |
12768 | } | |
12769 | ||
12770 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
12771 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
12772 | if (ret < 0) | |
12773 | return ret; | |
38207291 | 12774 | } |
f7b12b6f | 12775 | |
22dc4a0f | 12776 | key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id); |
f7b12b6f THJ |
12777 | tr = bpf_trampoline_get(key, &tgt_info); |
12778 | if (!tr) | |
12779 | return -ENOMEM; | |
12780 | ||
3aac1ead | 12781 | prog->aux->dst_trampoline = tr; |
f7b12b6f | 12782 | return 0; |
38207291 MKL |
12783 | } |
12784 | ||
76654e67 AM |
12785 | struct btf *bpf_get_btf_vmlinux(void) |
12786 | { | |
12787 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { | |
12788 | mutex_lock(&bpf_verifier_lock); | |
12789 | if (!btf_vmlinux) | |
12790 | btf_vmlinux = btf_parse_vmlinux(); | |
12791 | mutex_unlock(&bpf_verifier_lock); | |
12792 | } | |
12793 | return btf_vmlinux; | |
12794 | } | |
12795 | ||
838e9690 YS |
12796 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, |
12797 | union bpf_attr __user *uattr) | |
51580e79 | 12798 | { |
06ee7115 | 12799 | u64 start_time = ktime_get_ns(); |
58e2af8b | 12800 | struct bpf_verifier_env *env; |
b9193c1b | 12801 | struct bpf_verifier_log *log; |
9e4c24e7 | 12802 | int i, len, ret = -EINVAL; |
e2ae4ca2 | 12803 | bool is_priv; |
51580e79 | 12804 | |
eba0c929 AB |
12805 | /* no program is valid */ |
12806 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
12807 | return -EINVAL; | |
12808 | ||
58e2af8b | 12809 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
12810 | * allocate/free it every time bpf_check() is called |
12811 | */ | |
58e2af8b | 12812 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
12813 | if (!env) |
12814 | return -ENOMEM; | |
61bd5218 | 12815 | log = &env->log; |
cbd35700 | 12816 | |
9e4c24e7 | 12817 | len = (*prog)->len; |
fad953ce | 12818 | env->insn_aux_data = |
9e4c24e7 | 12819 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
12820 | ret = -ENOMEM; |
12821 | if (!env->insn_aux_data) | |
12822 | goto err_free_env; | |
9e4c24e7 JK |
12823 | for (i = 0; i < len; i++) |
12824 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 12825 | env->prog = *prog; |
00176a34 | 12826 | env->ops = bpf_verifier_ops[env->prog->type]; |
2c78ee89 | 12827 | is_priv = bpf_capable(); |
0246e64d | 12828 | |
76654e67 | 12829 | bpf_get_btf_vmlinux(); |
8580ac94 | 12830 | |
cbd35700 | 12831 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
12832 | if (!is_priv) |
12833 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 AS |
12834 | |
12835 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
12836 | /* user requested verbose verifier output | |
12837 | * and supplied buffer to store the verification trace | |
12838 | */ | |
e7bf8249 JK |
12839 | log->level = attr->log_level; |
12840 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
12841 | log->len_total = attr->log_size; | |
cbd35700 AS |
12842 | |
12843 | ret = -EINVAL; | |
e7bf8249 | 12844 | /* log attributes have to be sane */ |
7a9f5c65 | 12845 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 || |
06ee7115 | 12846 | !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK) |
3df126f3 | 12847 | goto err_unlock; |
cbd35700 | 12848 | } |
1ad2f583 | 12849 | |
8580ac94 AS |
12850 | if (IS_ERR(btf_vmlinux)) { |
12851 | /* Either gcc or pahole or kernel are broken. */ | |
12852 | verbose(env, "in-kernel BTF is malformed\n"); | |
12853 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 12854 | goto skip_full_check; |
8580ac94 AS |
12855 | } |
12856 | ||
1ad2f583 DB |
12857 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
12858 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 12859 | env->strict_alignment = true; |
e9ee9efc DM |
12860 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
12861 | env->strict_alignment = false; | |
cbd35700 | 12862 | |
2c78ee89 | 12863 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
01f810ac | 12864 | env->allow_uninit_stack = bpf_allow_uninit_stack(); |
41c48f3a | 12865 | env->allow_ptr_to_map_access = bpf_allow_ptr_to_map_access(); |
2c78ee89 AS |
12866 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); |
12867 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
12868 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 12869 | |
10d274e8 AS |
12870 | if (is_priv) |
12871 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
12872 | ||
cae1927c | 12873 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
a40a2632 | 12874 | ret = bpf_prog_offload_verifier_prep(env->prog); |
ab3f0063 | 12875 | if (ret) |
f4e3ec0d | 12876 | goto skip_full_check; |
ab3f0063 JK |
12877 | } |
12878 | ||
dc2a4ebc | 12879 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 12880 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
12881 | GFP_USER); |
12882 | ret = -ENOMEM; | |
12883 | if (!env->explored_states) | |
12884 | goto skip_full_check; | |
12885 | ||
d9762e84 | 12886 | ret = check_subprogs(env); |
475fb78f AS |
12887 | if (ret < 0) |
12888 | goto skip_full_check; | |
12889 | ||
c454a46b | 12890 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
12891 | if (ret < 0) |
12892 | goto skip_full_check; | |
12893 | ||
be8704ff AS |
12894 | ret = check_attach_btf_id(env); |
12895 | if (ret) | |
12896 | goto skip_full_check; | |
12897 | ||
4976b718 HL |
12898 | ret = resolve_pseudo_ldimm64(env); |
12899 | if (ret < 0) | |
12900 | goto skip_full_check; | |
12901 | ||
d9762e84 MKL |
12902 | ret = check_cfg(env); |
12903 | if (ret < 0) | |
12904 | goto skip_full_check; | |
12905 | ||
51c39bb1 AS |
12906 | ret = do_check_subprogs(env); |
12907 | ret = ret ?: do_check_main(env); | |
cbd35700 | 12908 | |
c941ce9c QM |
12909 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
12910 | ret = bpf_prog_offload_finalize(env); | |
12911 | ||
0246e64d | 12912 | skip_full_check: |
51c39bb1 | 12913 | kvfree(env->explored_states); |
0246e64d | 12914 | |
c131187d | 12915 | if (ret == 0) |
9b38c405 | 12916 | ret = check_max_stack_depth(env); |
c131187d | 12917 | |
9b38c405 | 12918 | /* instruction rewrites happen after this point */ |
e2ae4ca2 JK |
12919 | if (is_priv) { |
12920 | if (ret == 0) | |
12921 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
12922 | if (ret == 0) |
12923 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
12924 | if (ret == 0) |
12925 | ret = opt_remove_nops(env); | |
52875a04 JK |
12926 | } else { |
12927 | if (ret == 0) | |
12928 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
12929 | } |
12930 | ||
9bac3d6d AS |
12931 | if (ret == 0) |
12932 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
12933 | ret = convert_ctx_accesses(env); | |
12934 | ||
e245c5c6 | 12935 | if (ret == 0) |
e6ac5933 | 12936 | ret = do_misc_fixups(env); |
e245c5c6 | 12937 | |
a4b1d3c1 JW |
12938 | /* do 32-bit optimization after insn patching has done so those patched |
12939 | * insns could be handled correctly. | |
12940 | */ | |
d6c2308c JW |
12941 | if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) { |
12942 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); | |
12943 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
12944 | : false; | |
a4b1d3c1 JW |
12945 | } |
12946 | ||
1ea47e01 AS |
12947 | if (ret == 0) |
12948 | ret = fixup_call_args(env); | |
12949 | ||
06ee7115 AS |
12950 | env->verification_time = ktime_get_ns() - start_time; |
12951 | print_verification_stats(env); | |
12952 | ||
a2a7d570 | 12953 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 12954 | ret = -ENOSPC; |
a2a7d570 | 12955 | if (log->level && !log->ubuf) { |
cbd35700 | 12956 | ret = -EFAULT; |
a2a7d570 | 12957 | goto err_release_maps; |
cbd35700 AS |
12958 | } |
12959 | ||
541c3bad AN |
12960 | if (ret) |
12961 | goto err_release_maps; | |
12962 | ||
12963 | if (env->used_map_cnt) { | |
0246e64d | 12964 | /* if program passed verifier, update used_maps in bpf_prog_info */ |
9bac3d6d AS |
12965 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
12966 | sizeof(env->used_maps[0]), | |
12967 | GFP_KERNEL); | |
0246e64d | 12968 | |
9bac3d6d | 12969 | if (!env->prog->aux->used_maps) { |
0246e64d | 12970 | ret = -ENOMEM; |
a2a7d570 | 12971 | goto err_release_maps; |
0246e64d AS |
12972 | } |
12973 | ||
9bac3d6d | 12974 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 12975 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 12976 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
541c3bad AN |
12977 | } |
12978 | if (env->used_btf_cnt) { | |
12979 | /* if program passed verifier, update used_btfs in bpf_prog_aux */ | |
12980 | env->prog->aux->used_btfs = kmalloc_array(env->used_btf_cnt, | |
12981 | sizeof(env->used_btfs[0]), | |
12982 | GFP_KERNEL); | |
12983 | if (!env->prog->aux->used_btfs) { | |
12984 | ret = -ENOMEM; | |
12985 | goto err_release_maps; | |
12986 | } | |
0246e64d | 12987 | |
541c3bad AN |
12988 | memcpy(env->prog->aux->used_btfs, env->used_btfs, |
12989 | sizeof(env->used_btfs[0]) * env->used_btf_cnt); | |
12990 | env->prog->aux->used_btf_cnt = env->used_btf_cnt; | |
12991 | } | |
12992 | if (env->used_map_cnt || env->used_btf_cnt) { | |
0246e64d AS |
12993 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic |
12994 | * bpf_ld_imm64 instructions | |
12995 | */ | |
12996 | convert_pseudo_ld_imm64(env); | |
12997 | } | |
cbd35700 | 12998 | |
541c3bad | 12999 | adjust_btf_func(env); |
ba64e7d8 | 13000 | |
a2a7d570 | 13001 | err_release_maps: |
9bac3d6d | 13002 | if (!env->prog->aux->used_maps) |
0246e64d | 13003 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 13004 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
13005 | */ |
13006 | release_maps(env); | |
541c3bad AN |
13007 | if (!env->prog->aux->used_btfs) |
13008 | release_btfs(env); | |
03f87c0b THJ |
13009 | |
13010 | /* extension progs temporarily inherit the attach_type of their targets | |
13011 | for verification purposes, so set it back to zero before returning | |
13012 | */ | |
13013 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
13014 | env->prog->expected_attach_type = 0; | |
13015 | ||
9bac3d6d | 13016 | *prog = env->prog; |
3df126f3 | 13017 | err_unlock: |
45a73c17 AS |
13018 | if (!is_priv) |
13019 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
13020 | vfree(env->insn_aux_data); |
13021 | err_free_env: | |
13022 | kfree(env); | |
51580e79 AS |
13023 | return ret; |
13024 | } |