]>
Commit | Line | Data |
---|---|---|
51580e79 | 1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 2 | * Copyright (c) 2016 Facebook |
51580e79 AS |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | */ | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/types.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/bpf.h> | |
58e2af8b | 17 | #include <linux/bpf_verifier.h> |
51580e79 AS |
18 | #include <linux/filter.h> |
19 | #include <net/netlink.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/vmalloc.h> | |
ebb676da | 22 | #include <linux/stringify.h> |
1ec7c3bd | 23 | #include <linux/sched/signal.h> |
51580e79 | 24 | |
f4ac7e0b JK |
25 | #include "disasm.h" |
26 | ||
00176a34 JK |
27 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
28 | #define BPF_PROG_TYPE(_id, _name) \ | |
29 | [_id] = & _name ## _verifier_ops, | |
30 | #define BPF_MAP_TYPE(_id, _ops) | |
31 | #include <linux/bpf_types.h> | |
32 | #undef BPF_PROG_TYPE | |
33 | #undef BPF_MAP_TYPE | |
34 | }; | |
35 | ||
51580e79 AS |
36 | /* bpf_check() is a static code analyzer that walks eBPF program |
37 | * instruction by instruction and updates register/stack state. | |
38 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
39 | * | |
40 | * The first pass is depth-first-search to check that the program is a DAG. | |
41 | * It rejects the following programs: | |
42 | * - larger than BPF_MAXINSNS insns | |
43 | * - if loop is present (detected via back-edge) | |
44 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
45 | * - out of bounds or malformed jumps | |
46 | * The second pass is all possible path descent from the 1st insn. | |
47 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 48 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
49 | * insn is less then 4K, but there are too many branches that change stack/regs. |
50 | * Number of 'branches to be analyzed' is limited to 1k | |
51 | * | |
52 | * On entry to each instruction, each register has a type, and the instruction | |
53 | * changes the types of the registers depending on instruction semantics. | |
54 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
55 | * copied to R1. | |
56 | * | |
57 | * All registers are 64-bit. | |
58 | * R0 - return register | |
59 | * R1-R5 argument passing registers | |
60 | * R6-R9 callee saved registers | |
61 | * R10 - frame pointer read-only | |
62 | * | |
63 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
64 | * and has type PTR_TO_CTX. | |
65 | * | |
66 | * Verifier tracks arithmetic operations on pointers in case: | |
67 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
68 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
69 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
70 | * and 2nd arithmetic instruction is pattern matched to recognize | |
71 | * that it wants to construct a pointer to some element within stack. | |
72 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
73 | * (and -20 constant is saved for further stack bounds checking). | |
74 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
75 | * | |
f1174f77 | 76 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 77 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 78 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
79 | * |
80 | * When verifier sees load or store instructions the type of base register | |
f1174f77 | 81 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer |
51580e79 AS |
82 | * types recognized by check_mem_access() function. |
83 | * | |
84 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
85 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
86 | * | |
87 | * registers used to pass values to function calls are checked against | |
88 | * function argument constraints. | |
89 | * | |
90 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
91 | * It means that the register type passed to this function must be | |
92 | * PTR_TO_STACK and it will be used inside the function as | |
93 | * 'pointer to map element key' | |
94 | * | |
95 | * For example the argument constraints for bpf_map_lookup_elem(): | |
96 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
97 | * .arg1_type = ARG_CONST_MAP_PTR, | |
98 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
99 | * | |
100 | * ret_type says that this function returns 'pointer to map elem value or null' | |
101 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
102 | * 2nd argument should be a pointer to stack, which will be used inside | |
103 | * the helper function as a pointer to map element key. | |
104 | * | |
105 | * On the kernel side the helper function looks like: | |
106 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
107 | * { | |
108 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
109 | * void *key = (void *) (unsigned long) r2; | |
110 | * void *value; | |
111 | * | |
112 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
113 | * [key, key + map->key_size) bytes are valid and were initialized on | |
114 | * the stack of eBPF program. | |
115 | * } | |
116 | * | |
117 | * Corresponding eBPF program may look like: | |
118 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
119 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
120 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
121 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
122 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
123 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
124 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
125 | * | |
126 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
127 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
128 | * and were initialized prior to this call. | |
129 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
130 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
131 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
132 | * returns ether pointer to map value or NULL. | |
133 | * | |
134 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
135 | * insn, the register holding that pointer in the true branch changes state to | |
136 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
137 | * branch. See check_cond_jmp_op(). | |
138 | * | |
139 | * After the call R0 is set to return type of the function and registers R1-R5 | |
140 | * are set to NOT_INIT to indicate that they are no longer readable. | |
141 | */ | |
142 | ||
17a52670 | 143 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 144 | struct bpf_verifier_stack_elem { |
17a52670 AS |
145 | /* verifer state is 'st' |
146 | * before processing instruction 'insn_idx' | |
147 | * and after processing instruction 'prev_insn_idx' | |
148 | */ | |
58e2af8b | 149 | struct bpf_verifier_state st; |
17a52670 AS |
150 | int insn_idx; |
151 | int prev_insn_idx; | |
58e2af8b | 152 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
153 | }; |
154 | ||
8e17c1b1 | 155 | #define BPF_COMPLEXITY_LIMIT_INSNS 131072 |
07016151 | 156 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 |
adec32a5 | 157 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 158 | |
3f62bcc9 DB |
159 | #define BPF_MAP_PTR_UNPRIV 1UL |
160 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
161 | POISON_POINTER_DELTA)) | |
162 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
163 | ||
164 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) | |
165 | { | |
166 | return BPF_MAP_PTR(aux->map_state) == BPF_MAP_PTR_POISON; | |
167 | } | |
168 | ||
169 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
170 | { | |
171 | return aux->map_state & BPF_MAP_PTR_UNPRIV; | |
172 | } | |
173 | ||
174 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
175 | const struct bpf_map *map, bool unpriv) | |
176 | { | |
177 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
178 | unpriv |= bpf_map_ptr_unpriv(aux); | |
179 | aux->map_state = (unsigned long)map | | |
180 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
181 | } | |
fad73a1a | 182 | |
33ff9823 DB |
183 | struct bpf_call_arg_meta { |
184 | struct bpf_map *map_ptr; | |
435faee1 | 185 | bool raw_mode; |
36bbef52 | 186 | bool pkt_access; |
435faee1 DB |
187 | int regno; |
188 | int access_size; | |
33ff9823 DB |
189 | }; |
190 | ||
cbd35700 AS |
191 | static DEFINE_MUTEX(bpf_verifier_lock); |
192 | ||
193 | /* log_level controls verbosity level of eBPF verifier. | |
194 | * verbose() is used to dump the verification trace to the log, so the user | |
195 | * can figure out what's wrong with the program | |
196 | */ | |
61bd5218 JK |
197 | static __printf(2, 3) void verbose(struct bpf_verifier_env *env, |
198 | const char *fmt, ...) | |
cbd35700 | 199 | { |
61bd5218 | 200 | struct bpf_verifer_log *log = &env->log; |
a2a7d570 | 201 | unsigned int n; |
cbd35700 AS |
202 | va_list args; |
203 | ||
a2a7d570 | 204 | if (!log->level || !log->ubuf || bpf_verifier_log_full(log)) |
cbd35700 AS |
205 | return; |
206 | ||
207 | va_start(args, fmt); | |
a2a7d570 | 208 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
cbd35700 | 209 | va_end(args); |
a2a7d570 JK |
210 | |
211 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
212 | "verifier log line truncated - local buffer too short\n"); | |
213 | ||
214 | n = min(log->len_total - log->len_used - 1, n); | |
215 | log->kbuf[n] = '\0'; | |
216 | ||
217 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) | |
218 | log->len_used += n; | |
219 | else | |
220 | log->ubuf = NULL; | |
cbd35700 AS |
221 | } |
222 | ||
de8f3a83 DB |
223 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
224 | { | |
225 | return type == PTR_TO_PACKET || | |
226 | type == PTR_TO_PACKET_META; | |
227 | } | |
228 | ||
17a52670 AS |
229 | /* string representation of 'enum bpf_reg_type' */ |
230 | static const char * const reg_type_str[] = { | |
231 | [NOT_INIT] = "?", | |
f1174f77 | 232 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
233 | [PTR_TO_CTX] = "ctx", |
234 | [CONST_PTR_TO_MAP] = "map_ptr", | |
235 | [PTR_TO_MAP_VALUE] = "map_value", | |
236 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 237 | [PTR_TO_STACK] = "fp", |
969bf05e | 238 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 239 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 240 | [PTR_TO_PACKET_END] = "pkt_end", |
17a52670 AS |
241 | }; |
242 | ||
61bd5218 JK |
243 | static void print_verifier_state(struct bpf_verifier_env *env, |
244 | struct bpf_verifier_state *state) | |
17a52670 | 245 | { |
58e2af8b | 246 | struct bpf_reg_state *reg; |
17a52670 AS |
247 | enum bpf_reg_type t; |
248 | int i; | |
249 | ||
250 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
251 | reg = &state->regs[i]; |
252 | t = reg->type; | |
17a52670 AS |
253 | if (t == NOT_INIT) |
254 | continue; | |
61bd5218 | 255 | verbose(env, " R%d=%s", i, reg_type_str[t]); |
f1174f77 EC |
256 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
257 | tnum_is_const(reg->var_off)) { | |
258 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 259 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 260 | } else { |
61bd5218 | 261 | verbose(env, "(id=%d", reg->id); |
f1174f77 | 262 | if (t != SCALAR_VALUE) |
61bd5218 | 263 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 264 | if (type_is_pkt_pointer(t)) |
61bd5218 | 265 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
266 | else if (t == CONST_PTR_TO_MAP || |
267 | t == PTR_TO_MAP_VALUE || | |
268 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 269 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
270 | reg->map_ptr->key_size, |
271 | reg->map_ptr->value_size); | |
7d1238f2 EC |
272 | if (tnum_is_const(reg->var_off)) { |
273 | /* Typically an immediate SCALAR_VALUE, but | |
274 | * could be a pointer whose offset is too big | |
275 | * for reg->off | |
276 | */ | |
61bd5218 | 277 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
278 | } else { |
279 | if (reg->smin_value != reg->umin_value && | |
280 | reg->smin_value != S64_MIN) | |
61bd5218 | 281 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
282 | (long long)reg->smin_value); |
283 | if (reg->smax_value != reg->umax_value && | |
284 | reg->smax_value != S64_MAX) | |
61bd5218 | 285 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
286 | (long long)reg->smax_value); |
287 | if (reg->umin_value != 0) | |
61bd5218 | 288 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
289 | (unsigned long long)reg->umin_value); |
290 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 291 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
292 | (unsigned long long)reg->umax_value); |
293 | if (!tnum_is_unknown(reg->var_off)) { | |
294 | char tn_buf[48]; | |
f1174f77 | 295 | |
7d1238f2 | 296 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 297 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 298 | } |
f1174f77 | 299 | } |
61bd5218 | 300 | verbose(env, ")"); |
f1174f77 | 301 | } |
17a52670 | 302 | } |
638f5b90 AS |
303 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
304 | if (state->stack[i].slot_type[0] == STACK_SPILL) | |
305 | verbose(env, " fp%d=%s", | |
e8988995 | 306 | (-i - 1) * BPF_REG_SIZE, |
638f5b90 | 307 | reg_type_str[state->stack[i].spilled_ptr.type]); |
17a52670 | 308 | } |
61bd5218 | 309 | verbose(env, "\n"); |
17a52670 AS |
310 | } |
311 | ||
638f5b90 AS |
312 | static int copy_stack_state(struct bpf_verifier_state *dst, |
313 | const struct bpf_verifier_state *src) | |
17a52670 | 314 | { |
638f5b90 AS |
315 | if (!src->stack) |
316 | return 0; | |
317 | if (WARN_ON_ONCE(dst->allocated_stack < src->allocated_stack)) { | |
318 | /* internal bug, make state invalid to reject the program */ | |
319 | memset(dst, 0, sizeof(*dst)); | |
320 | return -EFAULT; | |
321 | } | |
322 | memcpy(dst->stack, src->stack, | |
323 | sizeof(*src->stack) * (src->allocated_stack / BPF_REG_SIZE)); | |
324 | return 0; | |
325 | } | |
326 | ||
327 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
328 | * make it consume minimal amount of memory. check_stack_write() access from | |
329 | * the program calls into realloc_verifier_state() to grow the stack size. | |
330 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state | |
331 | * which this function copies over. It points to previous bpf_verifier_state | |
332 | * which is never reallocated | |
333 | */ | |
334 | static int realloc_verifier_state(struct bpf_verifier_state *state, int size, | |
335 | bool copy_old) | |
336 | { | |
337 | u32 old_size = state->allocated_stack; | |
338 | struct bpf_stack_state *new_stack; | |
339 | int slot = size / BPF_REG_SIZE; | |
340 | ||
341 | if (size <= old_size || !size) { | |
342 | if (copy_old) | |
343 | return 0; | |
344 | state->allocated_stack = slot * BPF_REG_SIZE; | |
345 | if (!size && old_size) { | |
346 | kfree(state->stack); | |
347 | state->stack = NULL; | |
348 | } | |
349 | return 0; | |
350 | } | |
351 | new_stack = kmalloc_array(slot, sizeof(struct bpf_stack_state), | |
352 | GFP_KERNEL); | |
353 | if (!new_stack) | |
354 | return -ENOMEM; | |
355 | if (copy_old) { | |
356 | if (state->stack) | |
357 | memcpy(new_stack, state->stack, | |
358 | sizeof(*new_stack) * (old_size / BPF_REG_SIZE)); | |
359 | memset(new_stack + old_size / BPF_REG_SIZE, 0, | |
360 | sizeof(*new_stack) * (size - old_size) / BPF_REG_SIZE); | |
361 | } | |
362 | state->allocated_stack = slot * BPF_REG_SIZE; | |
363 | kfree(state->stack); | |
364 | state->stack = new_stack; | |
365 | return 0; | |
366 | } | |
367 | ||
1969db47 AS |
368 | static void free_verifier_state(struct bpf_verifier_state *state, |
369 | bool free_self) | |
638f5b90 AS |
370 | { |
371 | kfree(state->stack); | |
1969db47 AS |
372 | if (free_self) |
373 | kfree(state); | |
638f5b90 AS |
374 | } |
375 | ||
376 | /* copy verifier state from src to dst growing dst stack space | |
377 | * when necessary to accommodate larger src stack | |
378 | */ | |
379 | static int copy_verifier_state(struct bpf_verifier_state *dst, | |
380 | const struct bpf_verifier_state *src) | |
381 | { | |
382 | int err; | |
383 | ||
384 | err = realloc_verifier_state(dst, src->allocated_stack, false); | |
385 | if (err) | |
386 | return err; | |
387 | memcpy(dst, src, offsetof(struct bpf_verifier_state, allocated_stack)); | |
388 | return copy_stack_state(dst, src); | |
389 | } | |
390 | ||
391 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, | |
392 | int *insn_idx) | |
393 | { | |
394 | struct bpf_verifier_state *cur = env->cur_state; | |
395 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
396 | int err; | |
17a52670 AS |
397 | |
398 | if (env->head == NULL) | |
638f5b90 | 399 | return -ENOENT; |
17a52670 | 400 | |
638f5b90 AS |
401 | if (cur) { |
402 | err = copy_verifier_state(cur, &head->st); | |
403 | if (err) | |
404 | return err; | |
405 | } | |
406 | if (insn_idx) | |
407 | *insn_idx = head->insn_idx; | |
17a52670 | 408 | if (prev_insn_idx) |
638f5b90 AS |
409 | *prev_insn_idx = head->prev_insn_idx; |
410 | elem = head->next; | |
1969db47 | 411 | free_verifier_state(&head->st, false); |
638f5b90 | 412 | kfree(head); |
17a52670 AS |
413 | env->head = elem; |
414 | env->stack_size--; | |
638f5b90 | 415 | return 0; |
17a52670 AS |
416 | } |
417 | ||
58e2af8b | 418 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
010150ff DB |
419 | int insn_idx, int prev_insn_idx, |
420 | bool speculative) | |
17a52670 | 421 | { |
638f5b90 | 422 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 423 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 424 | int err; |
17a52670 | 425 | |
638f5b90 | 426 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
427 | if (!elem) |
428 | goto err; | |
429 | ||
17a52670 AS |
430 | elem->insn_idx = insn_idx; |
431 | elem->prev_insn_idx = prev_insn_idx; | |
432 | elem->next = env->head; | |
433 | env->head = elem; | |
434 | env->stack_size++; | |
1969db47 AS |
435 | err = copy_verifier_state(&elem->st, cur); |
436 | if (err) | |
437 | goto err; | |
010150ff | 438 | elem->st.speculative |= speculative; |
07016151 | 439 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
61bd5218 | 440 | verbose(env, "BPF program is too complex\n"); |
17a52670 AS |
441 | goto err; |
442 | } | |
443 | return &elem->st; | |
444 | err: | |
445 | /* pop all elements and return */ | |
638f5b90 | 446 | while (!pop_stack(env, NULL, NULL)); |
17a52670 AS |
447 | return NULL; |
448 | } | |
449 | ||
450 | #define CALLER_SAVED_REGS 6 | |
451 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
452 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
453 | }; | |
454 | ||
f1174f77 EC |
455 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
456 | ||
b03c9f9f EC |
457 | /* Mark the unknown part of a register (variable offset or scalar value) as |
458 | * known to have the value @imm. | |
459 | */ | |
460 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
461 | { | |
8f594301 AS |
462 | /* Clear id, off, and union(map_ptr, range) */ |
463 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
464 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
b03c9f9f EC |
465 | reg->var_off = tnum_const(imm); |
466 | reg->smin_value = (s64)imm; | |
467 | reg->smax_value = (s64)imm; | |
468 | reg->umin_value = imm; | |
469 | reg->umax_value = imm; | |
470 | } | |
471 | ||
f1174f77 EC |
472 | /* Mark the 'variable offset' part of a register as zero. This should be |
473 | * used only on registers holding a pointer type. | |
474 | */ | |
475 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 476 | { |
b03c9f9f | 477 | __mark_reg_known(reg, 0); |
f1174f77 | 478 | } |
a9789ef9 | 479 | |
61bd5218 JK |
480 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
481 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
482 | { |
483 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 484 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
485 | /* Something bad happened, let's kill all regs */ |
486 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
487 | __mark_reg_not_init(regs + regno); | |
488 | return; | |
489 | } | |
490 | __mark_reg_known_zero(regs + regno); | |
491 | } | |
492 | ||
de8f3a83 DB |
493 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
494 | { | |
495 | return type_is_pkt_pointer(reg->type); | |
496 | } | |
497 | ||
498 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
499 | { | |
500 | return reg_is_pkt_pointer(reg) || | |
501 | reg->type == PTR_TO_PACKET_END; | |
502 | } | |
503 | ||
504 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
505 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
506 | enum bpf_reg_type which) | |
507 | { | |
508 | /* The register can already have a range from prior markings. | |
509 | * This is fine as long as it hasn't been advanced from its | |
510 | * origin. | |
511 | */ | |
512 | return reg->type == which && | |
513 | reg->id == 0 && | |
514 | reg->off == 0 && | |
515 | tnum_equals_const(reg->var_off, 0); | |
516 | } | |
517 | ||
b03c9f9f EC |
518 | /* Attempts to improve min/max values based on var_off information */ |
519 | static void __update_reg_bounds(struct bpf_reg_state *reg) | |
520 | { | |
521 | /* min signed is max(sign bit) | min(other bits) */ | |
522 | reg->smin_value = max_t(s64, reg->smin_value, | |
523 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
524 | /* max signed is min(sign bit) | max(other bits) */ | |
525 | reg->smax_value = min_t(s64, reg->smax_value, | |
526 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
527 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
528 | reg->umax_value = min(reg->umax_value, | |
529 | reg->var_off.value | reg->var_off.mask); | |
530 | } | |
531 | ||
532 | /* Uses signed min/max values to inform unsigned, and vice-versa */ | |
533 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) | |
534 | { | |
535 | /* Learn sign from signed bounds. | |
536 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
537 | * are the same, so combine. This works even in the negative case, e.g. | |
538 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
539 | */ | |
540 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
541 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
542 | reg->umin_value); | |
543 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
544 | reg->umax_value); | |
545 | return; | |
546 | } | |
547 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
548 | * boundary, so we must be careful. | |
549 | */ | |
550 | if ((s64)reg->umax_value >= 0) { | |
551 | /* Positive. We can't learn anything from the smin, but smax | |
552 | * is positive, hence safe. | |
553 | */ | |
554 | reg->smin_value = reg->umin_value; | |
555 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
556 | reg->umax_value); | |
557 | } else if ((s64)reg->umin_value < 0) { | |
558 | /* Negative. We can't learn anything from the smax, but smin | |
559 | * is negative, hence safe. | |
560 | */ | |
561 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
562 | reg->umin_value); | |
563 | reg->smax_value = reg->umax_value; | |
564 | } | |
565 | } | |
566 | ||
567 | /* Attempts to improve var_off based on unsigned min/max information */ | |
568 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
569 | { | |
570 | reg->var_off = tnum_intersect(reg->var_off, | |
571 | tnum_range(reg->umin_value, | |
572 | reg->umax_value)); | |
573 | } | |
574 | ||
575 | /* Reset the min/max bounds of a register */ | |
576 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
577 | { | |
578 | reg->smin_value = S64_MIN; | |
579 | reg->smax_value = S64_MAX; | |
580 | reg->umin_value = 0; | |
581 | reg->umax_value = U64_MAX; | |
582 | } | |
583 | ||
f1174f77 EC |
584 | /* Mark a register as having a completely unknown (scalar) value. */ |
585 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
586 | { | |
8f594301 AS |
587 | /* |
588 | * Clear type, id, off, and union(map_ptr, range) and | |
589 | * padding between 'type' and union | |
590 | */ | |
591 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 592 | reg->type = SCALAR_VALUE; |
f1174f77 | 593 | reg->var_off = tnum_unknown; |
b03c9f9f | 594 | __mark_reg_unbounded(reg); |
f1174f77 EC |
595 | } |
596 | ||
61bd5218 JK |
597 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
598 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
599 | { |
600 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 601 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
f1174f77 EC |
602 | /* Something bad happened, let's kill all regs */ |
603 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
604 | __mark_reg_not_init(regs + regno); | |
605 | return; | |
606 | } | |
607 | __mark_reg_unknown(regs + regno); | |
608 | } | |
609 | ||
610 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
611 | { | |
612 | __mark_reg_unknown(reg); | |
613 | reg->type = NOT_INIT; | |
614 | } | |
615 | ||
61bd5218 JK |
616 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
617 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
618 | { |
619 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 620 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
f1174f77 EC |
621 | /* Something bad happened, let's kill all regs */ |
622 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
623 | __mark_reg_not_init(regs + regno); | |
624 | return; | |
625 | } | |
626 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
627 | } |
628 | ||
61bd5218 JK |
629 | static void init_reg_state(struct bpf_verifier_env *env, |
630 | struct bpf_reg_state *regs) | |
17a52670 AS |
631 | { |
632 | int i; | |
633 | ||
dc503a8a | 634 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 635 | mark_reg_not_init(env, regs, i); |
dc503a8a EC |
636 | regs[i].live = REG_LIVE_NONE; |
637 | } | |
17a52670 AS |
638 | |
639 | /* frame pointer */ | |
f1174f77 | 640 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 641 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
17a52670 AS |
642 | |
643 | /* 1st arg to a function */ | |
644 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
61bd5218 | 645 | mark_reg_known_zero(env, regs, BPF_REG_1); |
6760bf2d DB |
646 | } |
647 | ||
17a52670 AS |
648 | enum reg_arg_type { |
649 | SRC_OP, /* register is used as source operand */ | |
650 | DST_OP, /* register is used as destination operand */ | |
651 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
652 | }; | |
653 | ||
dc503a8a EC |
654 | static void mark_reg_read(const struct bpf_verifier_state *state, u32 regno) |
655 | { | |
656 | struct bpf_verifier_state *parent = state->parent; | |
657 | ||
8fe2d6cc AS |
658 | if (regno == BPF_REG_FP) |
659 | /* We don't need to worry about FP liveness because it's read-only */ | |
660 | return; | |
661 | ||
dc503a8a EC |
662 | while (parent) { |
663 | /* if read wasn't screened by an earlier write ... */ | |
664 | if (state->regs[regno].live & REG_LIVE_WRITTEN) | |
665 | break; | |
666 | /* ... then we depend on parent's value */ | |
667 | parent->regs[regno].live |= REG_LIVE_READ; | |
668 | state = parent; | |
669 | parent = state->parent; | |
670 | } | |
671 | } | |
672 | ||
673 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, | |
17a52670 AS |
674 | enum reg_arg_type t) |
675 | { | |
638f5b90 | 676 | struct bpf_reg_state *regs = env->cur_state->regs; |
dc503a8a | 677 | |
17a52670 | 678 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 679 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
680 | return -EINVAL; |
681 | } | |
682 | ||
683 | if (t == SRC_OP) { | |
684 | /* check whether register used as source operand can be read */ | |
685 | if (regs[regno].type == NOT_INIT) { | |
61bd5218 | 686 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
687 | return -EACCES; |
688 | } | |
638f5b90 | 689 | mark_reg_read(env->cur_state, regno); |
17a52670 AS |
690 | } else { |
691 | /* check whether register used as dest operand can be written to */ | |
692 | if (regno == BPF_REG_FP) { | |
61bd5218 | 693 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
694 | return -EACCES; |
695 | } | |
dc503a8a | 696 | regs[regno].live |= REG_LIVE_WRITTEN; |
17a52670 | 697 | if (t == DST_OP) |
61bd5218 | 698 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
699 | } |
700 | return 0; | |
701 | } | |
702 | ||
1be7f75d AS |
703 | static bool is_spillable_regtype(enum bpf_reg_type type) |
704 | { | |
705 | switch (type) { | |
706 | case PTR_TO_MAP_VALUE: | |
707 | case PTR_TO_MAP_VALUE_OR_NULL: | |
708 | case PTR_TO_STACK: | |
709 | case PTR_TO_CTX: | |
969bf05e | 710 | case PTR_TO_PACKET: |
de8f3a83 | 711 | case PTR_TO_PACKET_META: |
969bf05e | 712 | case PTR_TO_PACKET_END: |
1be7f75d AS |
713 | case CONST_PTR_TO_MAP: |
714 | return true; | |
715 | default: | |
716 | return false; | |
717 | } | |
718 | } | |
719 | ||
17a52670 AS |
720 | /* check_stack_read/write functions track spill/fill of registers, |
721 | * stack boundary and alignment are checked in check_mem_access() | |
722 | */ | |
61bd5218 JK |
723 | static int check_stack_write(struct bpf_verifier_env *env, |
724 | struct bpf_verifier_state *state, int off, | |
abd098e0 | 725 | int size, int value_regno, int insn_idx) |
17a52670 | 726 | { |
638f5b90 AS |
727 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
728 | ||
729 | err = realloc_verifier_state(state, round_up(slot + 1, BPF_REG_SIZE), | |
730 | true); | |
731 | if (err) | |
732 | return err; | |
9c399760 AS |
733 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
734 | * so it's aligned access and [off, off + size) are within stack limits | |
735 | */ | |
638f5b90 AS |
736 | if (!env->allow_ptr_leaks && |
737 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
738 | size != BPF_REG_SIZE) { | |
739 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
740 | return -EACCES; | |
741 | } | |
17a52670 AS |
742 | |
743 | if (value_regno >= 0 && | |
1be7f75d | 744 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
745 | |
746 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 747 | if (size != BPF_REG_SIZE) { |
61bd5218 | 748 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
749 | return -EACCES; |
750 | } | |
751 | ||
17a52670 | 752 | /* save register state */ |
638f5b90 AS |
753 | state->stack[spi].spilled_ptr = state->regs[value_regno]; |
754 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
17a52670 | 755 | |
abd098e0 AS |
756 | for (i = 0; i < BPF_REG_SIZE; i++) { |
757 | if (state->stack[spi].slot_type[i] == STACK_MISC && | |
758 | !env->allow_ptr_leaks) { | |
759 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; | |
760 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
761 | ||
762 | /* detected reuse of integer stack slot with a pointer | |
763 | * which means either llvm is reusing stack slot or | |
764 | * an attacker is trying to exploit CVE-2018-3639 | |
765 | * (speculative store bypass) | |
766 | * Have to sanitize that slot with preemptive | |
767 | * store of zero. | |
768 | */ | |
769 | if (*poff && *poff != soff) { | |
770 | /* disallow programs where single insn stores | |
771 | * into two different stack slots, since verifier | |
772 | * cannot sanitize them | |
773 | */ | |
774 | verbose(env, | |
775 | "insn %d cannot access two stack slots fp%d and fp%d", | |
776 | insn_idx, *poff, soff); | |
777 | return -EINVAL; | |
778 | } | |
779 | *poff = soff; | |
780 | } | |
638f5b90 | 781 | state->stack[spi].slot_type[i] = STACK_SPILL; |
abd098e0 | 782 | } |
9c399760 | 783 | } else { |
17a52670 | 784 | /* regular write of data into stack */ |
638f5b90 | 785 | state->stack[spi].spilled_ptr = (struct bpf_reg_state) {}; |
9c399760 AS |
786 | |
787 | for (i = 0; i < size; i++) | |
638f5b90 AS |
788 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
789 | STACK_MISC; | |
17a52670 AS |
790 | } |
791 | return 0; | |
792 | } | |
793 | ||
dc503a8a EC |
794 | static void mark_stack_slot_read(const struct bpf_verifier_state *state, int slot) |
795 | { | |
796 | struct bpf_verifier_state *parent = state->parent; | |
797 | ||
798 | while (parent) { | |
799 | /* if read wasn't screened by an earlier write ... */ | |
638f5b90 | 800 | if (state->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN) |
dc503a8a EC |
801 | break; |
802 | /* ... then we depend on parent's value */ | |
638f5b90 | 803 | parent->stack[slot].spilled_ptr.live |= REG_LIVE_READ; |
dc503a8a EC |
804 | state = parent; |
805 | parent = state->parent; | |
806 | } | |
807 | } | |
808 | ||
61bd5218 JK |
809 | static int check_stack_read(struct bpf_verifier_env *env, |
810 | struct bpf_verifier_state *state, int off, int size, | |
17a52670 AS |
811 | int value_regno) |
812 | { | |
638f5b90 AS |
813 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
814 | u8 *stype; | |
17a52670 | 815 | |
638f5b90 AS |
816 | if (state->allocated_stack <= slot) { |
817 | verbose(env, "invalid read from stack off %d+0 size %d\n", | |
818 | off, size); | |
819 | return -EACCES; | |
820 | } | |
821 | stype = state->stack[spi].slot_type; | |
17a52670 | 822 | |
638f5b90 | 823 | if (stype[0] == STACK_SPILL) { |
9c399760 | 824 | if (size != BPF_REG_SIZE) { |
61bd5218 | 825 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
826 | return -EACCES; |
827 | } | |
9c399760 | 828 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 829 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 830 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
831 | return -EACCES; |
832 | } | |
833 | } | |
834 | ||
dc503a8a | 835 | if (value_regno >= 0) { |
17a52670 | 836 | /* restore register state from stack */ |
638f5b90 | 837 | state->regs[value_regno] = state->stack[spi].spilled_ptr; |
dc503a8a EC |
838 | mark_stack_slot_read(state, spi); |
839 | } | |
17a52670 AS |
840 | return 0; |
841 | } else { | |
842 | for (i = 0; i < size; i++) { | |
638f5b90 | 843 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_MISC) { |
61bd5218 | 844 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
17a52670 AS |
845 | off, i, size); |
846 | return -EACCES; | |
847 | } | |
848 | } | |
849 | if (value_regno >= 0) | |
850 | /* have read misc data from the stack */ | |
61bd5218 | 851 | mark_reg_unknown(env, state->regs, value_regno); |
17a52670 AS |
852 | return 0; |
853 | } | |
854 | } | |
855 | ||
4e28ad40 DB |
856 | static int check_stack_access(struct bpf_verifier_env *env, |
857 | const struct bpf_reg_state *reg, | |
858 | int off, int size) | |
859 | { | |
860 | /* Stack accesses must be at a fixed offset, so that we | |
861 | * can determine what type of data were returned. See | |
862 | * check_stack_read(). | |
863 | */ | |
864 | if (!tnum_is_const(reg->var_off)) { | |
865 | char tn_buf[48]; | |
866 | ||
867 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
a79a9b1c | 868 | verbose(env, "variable stack access var_off=%s off=%d size=%d\n", |
4e28ad40 DB |
869 | tn_buf, off, size); |
870 | return -EACCES; | |
871 | } | |
872 | ||
873 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
874 | verbose(env, "invalid stack off=%d size=%d\n", off, size); | |
875 | return -EACCES; | |
876 | } | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
17a52670 | 881 | /* check read/write into map element returned by bpf_map_lookup_elem() */ |
f1174f77 | 882 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 883 | int size, bool zero_size_allowed) |
17a52670 | 884 | { |
638f5b90 AS |
885 | struct bpf_reg_state *regs = cur_regs(env); |
886 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 887 | |
9fd29c08 YS |
888 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
889 | off + size > map->value_size) { | |
61bd5218 | 890 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
891 | map->value_size, off, size); |
892 | return -EACCES; | |
893 | } | |
894 | return 0; | |
895 | } | |
896 | ||
f1174f77 EC |
897 | /* check read/write into a map element with possible variable offset */ |
898 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 899 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 900 | { |
638f5b90 | 901 | struct bpf_verifier_state *state = env->cur_state; |
dbcfe5f7 GB |
902 | struct bpf_reg_state *reg = &state->regs[regno]; |
903 | int err; | |
904 | ||
f1174f77 EC |
905 | /* We may have adjusted the register to this map value, so we |
906 | * need to try adding each of min_value and max_value to off | |
907 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 908 | */ |
61bd5218 JK |
909 | if (env->log.level) |
910 | print_verifier_state(env, state); | |
f02d76c1 | 911 | |
dbcfe5f7 GB |
912 | /* The minimum value is only important with signed |
913 | * comparisons where we can't assume the floor of a | |
914 | * value is 0. If we are using signed variables for our | |
915 | * index'es we need to make sure that whatever we use | |
916 | * will have a set floor within our range. | |
917 | */ | |
f02d76c1 DB |
918 | if (reg->smin_value < 0 && |
919 | (reg->smin_value == S64_MIN || | |
920 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
921 | reg->smin_value + off < 0)) { | |
61bd5218 | 922 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
923 | regno); |
924 | return -EACCES; | |
925 | } | |
9fd29c08 YS |
926 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
927 | zero_size_allowed); | |
dbcfe5f7 | 928 | if (err) { |
61bd5218 JK |
929 | verbose(env, "R%d min value is outside of the array range\n", |
930 | regno); | |
dbcfe5f7 GB |
931 | return err; |
932 | } | |
933 | ||
b03c9f9f EC |
934 | /* If we haven't set a max value then we need to bail since we can't be |
935 | * sure we won't do bad things. | |
936 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 937 | */ |
b03c9f9f | 938 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 939 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
940 | regno); |
941 | return -EACCES; | |
942 | } | |
9fd29c08 YS |
943 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
944 | zero_size_allowed); | |
f1174f77 | 945 | if (err) |
61bd5218 JK |
946 | verbose(env, "R%d max value is outside of the array range\n", |
947 | regno); | |
f1174f77 | 948 | return err; |
dbcfe5f7 GB |
949 | } |
950 | ||
969bf05e AS |
951 | #define MAX_PACKET_OFF 0xffff |
952 | ||
58e2af8b | 953 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
954 | const struct bpf_call_arg_meta *meta, |
955 | enum bpf_access_type t) | |
4acf6c0b | 956 | { |
36bbef52 | 957 | switch (env->prog->type) { |
3a0af8fd TG |
958 | case BPF_PROG_TYPE_LWT_IN: |
959 | case BPF_PROG_TYPE_LWT_OUT: | |
960 | /* dst_input() and dst_output() can't write for now */ | |
961 | if (t == BPF_WRITE) | |
962 | return false; | |
7e57fbb2 | 963 | /* fallthrough */ |
36bbef52 DB |
964 | case BPF_PROG_TYPE_SCHED_CLS: |
965 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 966 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 967 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 968 | case BPF_PROG_TYPE_SK_SKB: |
36bbef52 DB |
969 | if (meta) |
970 | return meta->pkt_access; | |
971 | ||
972 | env->seen_direct_write = true; | |
4acf6c0b BB |
973 | return true; |
974 | default: | |
975 | return false; | |
976 | } | |
977 | } | |
978 | ||
f1174f77 | 979 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 980 | int off, int size, bool zero_size_allowed) |
969bf05e | 981 | { |
638f5b90 | 982 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 983 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 984 | |
9fd29c08 YS |
985 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
986 | (u64)off + size > reg->range) { | |
61bd5218 | 987 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 988 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
989 | return -EACCES; |
990 | } | |
991 | return 0; | |
992 | } | |
993 | ||
f1174f77 | 994 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 995 | int size, bool zero_size_allowed) |
f1174f77 | 996 | { |
638f5b90 | 997 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
998 | struct bpf_reg_state *reg = ®s[regno]; |
999 | int err; | |
1000 | ||
1001 | /* We may have added a variable offset to the packet pointer; but any | |
1002 | * reg->range we have comes after that. We are only checking the fixed | |
1003 | * offset. | |
1004 | */ | |
1005 | ||
1006 | /* We don't allow negative numbers, because we aren't tracking enough | |
1007 | * detail to prove they're safe. | |
1008 | */ | |
b03c9f9f | 1009 | if (reg->smin_value < 0) { |
61bd5218 | 1010 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
1011 | regno); |
1012 | return -EACCES; | |
1013 | } | |
9fd29c08 | 1014 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 1015 | if (err) { |
61bd5218 | 1016 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
1017 | return err; |
1018 | } | |
1019 | return err; | |
1020 | } | |
1021 | ||
1022 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 1023 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 1024 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 1025 | { |
f96da094 DB |
1026 | struct bpf_insn_access_aux info = { |
1027 | .reg_type = *reg_type, | |
1028 | }; | |
31fd8581 | 1029 | |
4f9218aa JK |
1030 | if (env->ops->is_valid_access && |
1031 | env->ops->is_valid_access(off, size, t, &info)) { | |
f96da094 DB |
1032 | /* A non zero info.ctx_field_size indicates that this field is a |
1033 | * candidate for later verifier transformation to load the whole | |
1034 | * field and then apply a mask when accessed with a narrower | |
1035 | * access than actual ctx access size. A zero info.ctx_field_size | |
1036 | * will only allow for whole field access and rejects any other | |
1037 | * type of narrower access. | |
31fd8581 | 1038 | */ |
23994631 | 1039 | *reg_type = info.reg_type; |
31fd8581 | 1040 | |
4f9218aa | 1041 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
1042 | /* remember the offset of last byte accessed in ctx */ |
1043 | if (env->prog->aux->max_ctx_offset < off + size) | |
1044 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 1045 | return 0; |
32bbe007 | 1046 | } |
17a52670 | 1047 | |
61bd5218 | 1048 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
1049 | return -EACCES; |
1050 | } | |
1051 | ||
4cabc5b1 DB |
1052 | static bool __is_pointer_value(bool allow_ptr_leaks, |
1053 | const struct bpf_reg_state *reg) | |
1be7f75d | 1054 | { |
4cabc5b1 | 1055 | if (allow_ptr_leaks) |
1be7f75d AS |
1056 | return false; |
1057 | ||
f1174f77 | 1058 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
1059 | } |
1060 | ||
4cabc5b1 DB |
1061 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1062 | { | |
638f5b90 | 1063 | return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno); |
4cabc5b1 DB |
1064 | } |
1065 | ||
f37a8cb8 DB |
1066 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
1067 | { | |
1068 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1069 | ||
1070 | return reg->type == PTR_TO_CTX; | |
1071 | } | |
1072 | ||
57b4a36e DB |
1073 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
1074 | { | |
1075 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1076 | ||
1077 | return type_is_pkt_pointer(reg->type); | |
1078 | } | |
1079 | ||
61bd5218 JK |
1080 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
1081 | const struct bpf_reg_state *reg, | |
d1174416 | 1082 | int off, int size, bool strict) |
969bf05e | 1083 | { |
f1174f77 | 1084 | struct tnum reg_off; |
e07b98d9 | 1085 | int ip_align; |
d1174416 DM |
1086 | |
1087 | /* Byte size accesses are always allowed. */ | |
1088 | if (!strict || size == 1) | |
1089 | return 0; | |
1090 | ||
e4eda884 DM |
1091 | /* For platforms that do not have a Kconfig enabling |
1092 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
1093 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
1094 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
1095 | * to this code only in strict mode where we want to emulate | |
1096 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
1097 | * unconditional IP align value of '2'. | |
e07b98d9 | 1098 | */ |
e4eda884 | 1099 | ip_align = 2; |
f1174f77 EC |
1100 | |
1101 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1102 | if (!tnum_is_aligned(reg_off, size)) { | |
1103 | char tn_buf[48]; | |
1104 | ||
1105 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1106 | verbose(env, |
1107 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1108 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1109 | return -EACCES; |
1110 | } | |
79adffcd | 1111 | |
969bf05e AS |
1112 | return 0; |
1113 | } | |
1114 | ||
61bd5218 JK |
1115 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1116 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1117 | const char *pointer_desc, |
1118 | int off, int size, bool strict) | |
79adffcd | 1119 | { |
f1174f77 EC |
1120 | struct tnum reg_off; |
1121 | ||
1122 | /* Byte size accesses are always allowed. */ | |
1123 | if (!strict || size == 1) | |
1124 | return 0; | |
1125 | ||
1126 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1127 | if (!tnum_is_aligned(reg_off, size)) { | |
1128 | char tn_buf[48]; | |
1129 | ||
1130 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1131 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1132 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1133 | return -EACCES; |
1134 | } | |
1135 | ||
969bf05e AS |
1136 | return 0; |
1137 | } | |
1138 | ||
e07b98d9 | 1139 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
57b4a36e DB |
1140 | const struct bpf_reg_state *reg, int off, |
1141 | int size, bool strict_alignment_once) | |
79adffcd | 1142 | { |
57b4a36e | 1143 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 1144 | const char *pointer_desc = ""; |
d1174416 | 1145 | |
79adffcd DB |
1146 | switch (reg->type) { |
1147 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1148 | case PTR_TO_PACKET_META: |
1149 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1150 | * right in front, treat it the very same way. | |
1151 | */ | |
61bd5218 | 1152 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
f1174f77 EC |
1153 | case PTR_TO_MAP_VALUE: |
1154 | pointer_desc = "value "; | |
1155 | break; | |
1156 | case PTR_TO_CTX: | |
1157 | pointer_desc = "context "; | |
1158 | break; | |
1159 | case PTR_TO_STACK: | |
1160 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
1161 | /* The stack spill tracking logic in check_stack_write() |
1162 | * and check_stack_read() relies on stack accesses being | |
1163 | * aligned. | |
1164 | */ | |
1165 | strict = true; | |
f1174f77 | 1166 | break; |
79adffcd | 1167 | default: |
f1174f77 | 1168 | break; |
79adffcd | 1169 | } |
61bd5218 JK |
1170 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1171 | strict); | |
79adffcd DB |
1172 | } |
1173 | ||
173abb86 DB |
1174 | static int check_ctx_reg(struct bpf_verifier_env *env, |
1175 | const struct bpf_reg_state *reg, int regno) | |
1176 | { | |
1177 | /* Access to ctx or passing it to a helper is only allowed in | |
1178 | * its original, unmodified form. | |
1179 | */ | |
1180 | ||
1181 | if (reg->off) { | |
1182 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
1183 | regno, reg->off); | |
1184 | return -EACCES; | |
1185 | } | |
1186 | ||
1187 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
1188 | char tn_buf[48]; | |
1189 | ||
1190 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1191 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
1192 | return -EACCES; | |
1193 | } | |
1194 | ||
1195 | return 0; | |
1196 | } | |
1197 | ||
0c17d1d2 JH |
1198 | /* truncate register to smaller size (in bytes) |
1199 | * must be called with size < BPF_REG_SIZE | |
1200 | */ | |
1201 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
1202 | { | |
1203 | u64 mask; | |
1204 | ||
1205 | /* clear high bits in bit representation */ | |
1206 | reg->var_off = tnum_cast(reg->var_off, size); | |
1207 | ||
1208 | /* fix arithmetic bounds */ | |
1209 | mask = ((u64)1 << (size * 8)) - 1; | |
1210 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
1211 | reg->umin_value &= mask; | |
1212 | reg->umax_value &= mask; | |
1213 | } else { | |
1214 | reg->umin_value = 0; | |
1215 | reg->umax_value = mask; | |
1216 | } | |
1217 | reg->smin_value = reg->umin_value; | |
1218 | reg->smax_value = reg->umax_value; | |
1219 | } | |
1220 | ||
17a52670 AS |
1221 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1222 | * if t==write, value_regno is a register which value is stored into memory | |
1223 | * if t==read, value_regno is a register which will receive the value from memory | |
1224 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1225 | * if t==read && value_regno==-1, don't care what we read from memory | |
1226 | */ | |
57b4a36e DB |
1227 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
1228 | int off, int bpf_size, enum bpf_access_type t, | |
1229 | int value_regno, bool strict_alignment_once) | |
17a52670 | 1230 | { |
638f5b90 AS |
1231 | struct bpf_verifier_state *state = env->cur_state; |
1232 | struct bpf_reg_state *regs = cur_regs(env); | |
1233 | struct bpf_reg_state *reg = regs + regno; | |
17a52670 AS |
1234 | int size, err = 0; |
1235 | ||
1236 | size = bpf_size_to_bytes(bpf_size); | |
1237 | if (size < 0) | |
1238 | return size; | |
1239 | ||
f1174f77 | 1240 | /* alignment checks will add in reg->off themselves */ |
57b4a36e | 1241 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
1242 | if (err) |
1243 | return err; | |
17a52670 | 1244 | |
f1174f77 EC |
1245 | /* for access checks, reg->off is just part of off */ |
1246 | off += reg->off; | |
1247 | ||
1248 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1249 | if (t == BPF_WRITE && value_regno >= 0 && |
1250 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1251 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1252 | return -EACCES; |
1253 | } | |
48461135 | 1254 | |
9fd29c08 | 1255 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 1256 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1257 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1258 | |
1a0dc1ac | 1259 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1260 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1261 | |
1be7f75d AS |
1262 | if (t == BPF_WRITE && value_regno >= 0 && |
1263 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1264 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1265 | return -EACCES; |
1266 | } | |
f1174f77 | 1267 | |
173abb86 DB |
1268 | err = check_ctx_reg(env, reg, regno); |
1269 | if (err < 0) | |
1270 | return err; | |
1271 | ||
31fd8581 | 1272 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1273 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1274 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1275 | * PTR_TO_PACKET[_META,_END]. In the latter |
1276 | * case, we know the offset is zero. | |
f1174f77 EC |
1277 | */ |
1278 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1279 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1280 | else |
638f5b90 | 1281 | mark_reg_known_zero(env, regs, |
61bd5218 | 1282 | value_regno); |
638f5b90 | 1283 | regs[value_regno].type = reg_type; |
969bf05e | 1284 | } |
17a52670 | 1285 | |
f1174f77 | 1286 | } else if (reg->type == PTR_TO_STACK) { |
f1174f77 | 1287 | off += reg->var_off.value; |
4e28ad40 DB |
1288 | err = check_stack_access(env, reg, off, size); |
1289 | if (err) | |
1290 | return err; | |
8726679a AS |
1291 | |
1292 | if (env->prog->aux->stack_depth < -off) | |
1293 | env->prog->aux->stack_depth = -off; | |
1294 | ||
638f5b90 | 1295 | if (t == BPF_WRITE) |
61bd5218 | 1296 | err = check_stack_write(env, state, off, size, |
abd098e0 | 1297 | value_regno, insn_idx); |
638f5b90 | 1298 | else |
61bd5218 JK |
1299 | err = check_stack_read(env, state, off, size, |
1300 | value_regno); | |
de8f3a83 | 1301 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 1302 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 1303 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
1304 | return -EACCES; |
1305 | } | |
4acf6c0b BB |
1306 | if (t == BPF_WRITE && value_regno >= 0 && |
1307 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
1308 | verbose(env, "R%d leaks addr into packet\n", |
1309 | value_regno); | |
4acf6c0b BB |
1310 | return -EACCES; |
1311 | } | |
9fd29c08 | 1312 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 1313 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1314 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1315 | } else { |
61bd5218 JK |
1316 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
1317 | reg_type_str[reg->type]); | |
17a52670 AS |
1318 | return -EACCES; |
1319 | } | |
969bf05e | 1320 | |
f1174f77 | 1321 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 1322 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 1323 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 1324 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 1325 | } |
17a52670 AS |
1326 | return err; |
1327 | } | |
1328 | ||
31fd8581 | 1329 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1330 | { |
17a52670 AS |
1331 | int err; |
1332 | ||
1333 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1334 | insn->imm != 0) { | |
61bd5218 | 1335 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
1336 | return -EINVAL; |
1337 | } | |
1338 | ||
1339 | /* check src1 operand */ | |
dc503a8a | 1340 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
1341 | if (err) |
1342 | return err; | |
1343 | ||
1344 | /* check src2 operand */ | |
dc503a8a | 1345 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
1346 | if (err) |
1347 | return err; | |
1348 | ||
6bdf6abc | 1349 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 1350 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
1351 | return -EACCES; |
1352 | } | |
1353 | ||
57b4a36e DB |
1354 | if (is_ctx_reg(env, insn->dst_reg) || |
1355 | is_pkt_reg(env, insn->dst_reg)) { | |
1356 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", | |
1357 | insn->dst_reg, is_ctx_reg(env, insn->dst_reg) ? | |
1358 | "context" : "packet"); | |
f37a8cb8 DB |
1359 | return -EACCES; |
1360 | } | |
1361 | ||
17a52670 | 1362 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1363 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
57b4a36e | 1364 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
1365 | if (err) |
1366 | return err; | |
1367 | ||
1368 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1369 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
57b4a36e | 1370 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
1371 | } |
1372 | ||
f1174f77 EC |
1373 | /* Does this register contain a constant zero? */ |
1374 | static bool register_is_null(struct bpf_reg_state reg) | |
1375 | { | |
1376 | return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0); | |
1377 | } | |
1378 | ||
17a52670 AS |
1379 | /* when register 'regno' is passed into function that will read 'access_size' |
1380 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1381 | * and all elements of stack are initialized. |
1382 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1383 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1384 | */ |
58e2af8b | 1385 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1386 | int access_size, bool zero_size_allowed, |
1387 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1388 | { |
638f5b90 | 1389 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1390 | struct bpf_reg_state *regs = state->regs; |
638f5b90 | 1391 | int off, i, slot, spi; |
17a52670 | 1392 | |
8e2fe1d9 | 1393 | if (regs[regno].type != PTR_TO_STACK) { |
f1174f77 | 1394 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1395 | if (zero_size_allowed && access_size == 0 && |
f1174f77 | 1396 | register_is_null(regs[regno])) |
8e2fe1d9 DB |
1397 | return 0; |
1398 | ||
61bd5218 | 1399 | verbose(env, "R%d type=%s expected=%s\n", regno, |
8e2fe1d9 DB |
1400 | reg_type_str[regs[regno].type], |
1401 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 1402 | return -EACCES; |
8e2fe1d9 | 1403 | } |
17a52670 | 1404 | |
f1174f77 EC |
1405 | /* Only allow fixed-offset stack reads */ |
1406 | if (!tnum_is_const(regs[regno].var_off)) { | |
1407 | char tn_buf[48]; | |
1408 | ||
1409 | tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off); | |
61bd5218 | 1410 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 | 1411 | regno, tn_buf); |
ea25f914 | 1412 | return -EACCES; |
f1174f77 EC |
1413 | } |
1414 | off = regs[regno].off + regs[regno].var_off.value; | |
17a52670 | 1415 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
9fd29c08 | 1416 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { |
61bd5218 | 1417 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
1418 | regno, off, access_size); |
1419 | return -EACCES; | |
1420 | } | |
1421 | ||
8726679a AS |
1422 | if (env->prog->aux->stack_depth < -off) |
1423 | env->prog->aux->stack_depth = -off; | |
1424 | ||
435faee1 DB |
1425 | if (meta && meta->raw_mode) { |
1426 | meta->access_size = access_size; | |
1427 | meta->regno = regno; | |
1428 | return 0; | |
1429 | } | |
1430 | ||
17a52670 | 1431 | for (i = 0; i < access_size; i++) { |
638f5b90 AS |
1432 | slot = -(off + i) - 1; |
1433 | spi = slot / BPF_REG_SIZE; | |
1434 | if (state->allocated_stack <= slot || | |
1435 | state->stack[spi].slot_type[slot % BPF_REG_SIZE] != | |
1436 | STACK_MISC) { | |
61bd5218 | 1437 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", |
17a52670 AS |
1438 | off, i, access_size); |
1439 | return -EACCES; | |
1440 | } | |
1441 | } | |
1442 | return 0; | |
1443 | } | |
1444 | ||
06c1c049 GB |
1445 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1446 | int access_size, bool zero_size_allowed, | |
1447 | struct bpf_call_arg_meta *meta) | |
1448 | { | |
638f5b90 | 1449 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 1450 | |
f1174f77 | 1451 | switch (reg->type) { |
06c1c049 | 1452 | case PTR_TO_PACKET: |
de8f3a83 | 1453 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
1454 | return check_packet_access(env, regno, reg->off, access_size, |
1455 | zero_size_allowed); | |
06c1c049 | 1456 | case PTR_TO_MAP_VALUE: |
9fd29c08 YS |
1457 | return check_map_access(env, regno, reg->off, access_size, |
1458 | zero_size_allowed); | |
f1174f77 | 1459 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
1460 | return check_stack_boundary(env, regno, access_size, |
1461 | zero_size_allowed, meta); | |
1462 | } | |
1463 | } | |
1464 | ||
58e2af8b | 1465 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1466 | enum bpf_arg_type arg_type, |
1467 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1468 | { |
638f5b90 | 1469 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 1470 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1471 | int err = 0; |
1472 | ||
80f1d68c | 1473 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1474 | return 0; |
1475 | ||
dc503a8a EC |
1476 | err = check_reg_arg(env, regno, SRC_OP); |
1477 | if (err) | |
1478 | return err; | |
17a52670 | 1479 | |
1be7f75d AS |
1480 | if (arg_type == ARG_ANYTHING) { |
1481 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
1482 | verbose(env, "R%d leaks addr into helper function\n", |
1483 | regno); | |
1be7f75d AS |
1484 | return -EACCES; |
1485 | } | |
80f1d68c | 1486 | return 0; |
1be7f75d | 1487 | } |
80f1d68c | 1488 | |
de8f3a83 | 1489 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 1490 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 1491 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
1492 | return -EACCES; |
1493 | } | |
1494 | ||
8e2fe1d9 | 1495 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1496 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1497 | expected_type = PTR_TO_STACK; | |
de8f3a83 DB |
1498 | if (!type_is_pkt_pointer(type) && |
1499 | type != expected_type) | |
6841de8b | 1500 | goto err_type; |
39f19ebb AS |
1501 | } else if (arg_type == ARG_CONST_SIZE || |
1502 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1503 | expected_type = SCALAR_VALUE; |
1504 | if (type != expected_type) | |
6841de8b | 1505 | goto err_type; |
17a52670 AS |
1506 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1507 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1508 | if (type != expected_type) |
1509 | goto err_type; | |
608cd71a AS |
1510 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1511 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1512 | if (type != expected_type) |
1513 | goto err_type; | |
173abb86 DB |
1514 | err = check_ctx_reg(env, reg, regno); |
1515 | if (err < 0) | |
1516 | return err; | |
39f19ebb | 1517 | } else if (arg_type == ARG_PTR_TO_MEM || |
db1ac496 | 1518 | arg_type == ARG_PTR_TO_MEM_OR_NULL || |
39f19ebb | 1519 | arg_type == ARG_PTR_TO_UNINIT_MEM) { |
8e2fe1d9 DB |
1520 | expected_type = PTR_TO_STACK; |
1521 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1522 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1523 | * happens during stack boundary checking. |
1524 | */ | |
db1ac496 GB |
1525 | if (register_is_null(*reg) && |
1526 | arg_type == ARG_PTR_TO_MEM_OR_NULL) | |
6841de8b | 1527 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
1528 | else if (!type_is_pkt_pointer(type) && |
1529 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 1530 | type != expected_type) |
6841de8b | 1531 | goto err_type; |
39f19ebb | 1532 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 1533 | } else { |
61bd5218 | 1534 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
1535 | return -EFAULT; |
1536 | } | |
1537 | ||
17a52670 AS |
1538 | if (arg_type == ARG_CONST_MAP_PTR) { |
1539 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1540 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1541 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1542 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1543 | * check that [key, key + map->key_size) are within | |
1544 | * stack limits and initialized | |
1545 | */ | |
33ff9823 | 1546 | if (!meta->map_ptr) { |
17a52670 AS |
1547 | /* in function declaration map_ptr must come before |
1548 | * map_key, so that it's verified and known before | |
1549 | * we have to check map_key here. Otherwise it means | |
1550 | * that kernel subsystem misconfigured verifier | |
1551 | */ | |
61bd5218 | 1552 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
1553 | return -EACCES; |
1554 | } | |
de8f3a83 | 1555 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1556 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1557 | meta->map_ptr->key_size, |
1558 | false); | |
6841de8b AS |
1559 | else |
1560 | err = check_stack_boundary(env, regno, | |
1561 | meta->map_ptr->key_size, | |
1562 | false, NULL); | |
17a52670 AS |
1563 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1564 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1565 | * check [value, value + map->value_size) validity | |
1566 | */ | |
33ff9823 | 1567 | if (!meta->map_ptr) { |
17a52670 | 1568 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 1569 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
1570 | return -EACCES; |
1571 | } | |
de8f3a83 | 1572 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1573 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1574 | meta->map_ptr->value_size, |
1575 | false); | |
6841de8b AS |
1576 | else |
1577 | err = check_stack_boundary(env, regno, | |
1578 | meta->map_ptr->value_size, | |
1579 | false, NULL); | |
39f19ebb AS |
1580 | } else if (arg_type == ARG_CONST_SIZE || |
1581 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1582 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1583 | |
17a52670 AS |
1584 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1585 | * from stack pointer 'buf'. Check it | |
1586 | * note: regno == len, regno - 1 == buf | |
1587 | */ | |
1588 | if (regno == 0) { | |
1589 | /* kernel subsystem misconfigured verifier */ | |
61bd5218 JK |
1590 | verbose(env, |
1591 | "ARG_CONST_SIZE cannot be first argument\n"); | |
17a52670 AS |
1592 | return -EACCES; |
1593 | } | |
06c1c049 | 1594 | |
f1174f77 EC |
1595 | /* The register is SCALAR_VALUE; the access check |
1596 | * happens using its boundaries. | |
06c1c049 | 1597 | */ |
f1174f77 EC |
1598 | |
1599 | if (!tnum_is_const(reg->var_off)) | |
06c1c049 GB |
1600 | /* For unprivileged variable accesses, disable raw |
1601 | * mode so that the program is required to | |
1602 | * initialize all the memory that the helper could | |
1603 | * just partially fill up. | |
1604 | */ | |
1605 | meta = NULL; | |
1606 | ||
b03c9f9f | 1607 | if (reg->smin_value < 0) { |
61bd5218 | 1608 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
1609 | regno); |
1610 | return -EACCES; | |
1611 | } | |
06c1c049 | 1612 | |
b03c9f9f | 1613 | if (reg->umin_value == 0) { |
f1174f77 EC |
1614 | err = check_helper_mem_access(env, regno - 1, 0, |
1615 | zero_size_allowed, | |
1616 | meta); | |
06c1c049 GB |
1617 | if (err) |
1618 | return err; | |
06c1c049 | 1619 | } |
f1174f77 | 1620 | |
b03c9f9f | 1621 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 1622 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
1623 | regno); |
1624 | return -EACCES; | |
1625 | } | |
1626 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 1627 | reg->umax_value, |
f1174f77 | 1628 | zero_size_allowed, meta); |
17a52670 AS |
1629 | } |
1630 | ||
1631 | return err; | |
6841de8b | 1632 | err_type: |
61bd5218 | 1633 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
1634 | reg_type_str[type], reg_type_str[expected_type]); |
1635 | return -EACCES; | |
17a52670 AS |
1636 | } |
1637 | ||
61bd5218 JK |
1638 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
1639 | struct bpf_map *map, int func_id) | |
35578d79 | 1640 | { |
35578d79 KX |
1641 | if (!map) |
1642 | return 0; | |
1643 | ||
6aff67c8 AS |
1644 | /* We need a two way check, first is from map perspective ... */ |
1645 | switch (map->map_type) { | |
1646 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1647 | if (func_id != BPF_FUNC_tail_call) | |
1648 | goto error; | |
1649 | break; | |
1650 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1651 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
1652 | func_id != BPF_FUNC_perf_event_output && |
1653 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
1654 | goto error; |
1655 | break; | |
1656 | case BPF_MAP_TYPE_STACK_TRACE: | |
1657 | if (func_id != BPF_FUNC_get_stackid) | |
1658 | goto error; | |
1659 | break; | |
4ed8ec52 | 1660 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1661 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1662 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1663 | goto error; |
1664 | break; | |
546ac1ff JF |
1665 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
1666 | * allow to be modified from bpf side. So do not allow lookup elements | |
1667 | * for now. | |
1668 | */ | |
1669 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 1670 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
1671 | goto error; |
1672 | break; | |
6710e112 JDB |
1673 | /* Restrict bpf side of cpumap, open when use-cases appear */ |
1674 | case BPF_MAP_TYPE_CPUMAP: | |
1675 | if (func_id != BPF_FUNC_redirect_map) | |
1676 | goto error; | |
1677 | break; | |
56f668df | 1678 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1679 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1680 | if (func_id != BPF_FUNC_map_lookup_elem) |
1681 | goto error; | |
16a43625 | 1682 | break; |
174a79ff JF |
1683 | case BPF_MAP_TYPE_SOCKMAP: |
1684 | if (func_id != BPF_FUNC_sk_redirect_map && | |
1685 | func_id != BPF_FUNC_sock_map_update && | |
1686 | func_id != BPF_FUNC_map_delete_elem) | |
1687 | goto error; | |
1688 | break; | |
6aff67c8 AS |
1689 | default: |
1690 | break; | |
1691 | } | |
1692 | ||
1693 | /* ... and second from the function itself. */ | |
1694 | switch (func_id) { | |
1695 | case BPF_FUNC_tail_call: | |
1696 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1697 | goto error; | |
1698 | break; | |
1699 | case BPF_FUNC_perf_event_read: | |
1700 | case BPF_FUNC_perf_event_output: | |
908432ca | 1701 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
1702 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
1703 | goto error; | |
1704 | break; | |
1705 | case BPF_FUNC_get_stackid: | |
1706 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1707 | goto error; | |
1708 | break; | |
60d20f91 | 1709 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1710 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1711 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1712 | goto error; | |
1713 | break; | |
97f91a7c | 1714 | case BPF_FUNC_redirect_map: |
9c270af3 JDB |
1715 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
1716 | map->map_type != BPF_MAP_TYPE_CPUMAP) | |
97f91a7c JF |
1717 | goto error; |
1718 | break; | |
174a79ff JF |
1719 | case BPF_FUNC_sk_redirect_map: |
1720 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1721 | goto error; | |
1722 | break; | |
1723 | case BPF_FUNC_sock_map_update: | |
1724 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1725 | goto error; | |
1726 | break; | |
6aff67c8 AS |
1727 | default: |
1728 | break; | |
35578d79 KX |
1729 | } |
1730 | ||
1731 | return 0; | |
6aff67c8 | 1732 | error: |
61bd5218 | 1733 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 1734 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 1735 | return -EINVAL; |
35578d79 KX |
1736 | } |
1737 | ||
435faee1 DB |
1738 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1739 | { | |
1740 | int count = 0; | |
1741 | ||
39f19ebb | 1742 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1743 | count++; |
39f19ebb | 1744 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1745 | count++; |
39f19ebb | 1746 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1747 | count++; |
39f19ebb | 1748 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1749 | count++; |
39f19ebb | 1750 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1751 | count++; |
1752 | ||
1753 | return count > 1 ? -EINVAL : 0; | |
1754 | } | |
1755 | ||
de8f3a83 DB |
1756 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
1757 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 1758 | */ |
58e2af8b | 1759 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1760 | { |
638f5b90 | 1761 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1762 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
1763 | int i; |
1764 | ||
1765 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 1766 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 1767 | mark_reg_unknown(env, regs, i); |
969bf05e | 1768 | |
638f5b90 AS |
1769 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
1770 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 1771 | continue; |
638f5b90 | 1772 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 DB |
1773 | if (reg_is_pkt_pointer_any(reg)) |
1774 | __mark_reg_unknown(reg); | |
969bf05e AS |
1775 | } |
1776 | } | |
1777 | ||
3f62bcc9 DB |
1778 | static int |
1779 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
1780 | int func_id, int insn_idx) | |
1781 | { | |
1782 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
1783 | ||
1784 | if (func_id != BPF_FUNC_tail_call && | |
1785 | func_id != BPF_FUNC_map_lookup_elem) | |
1786 | return 0; | |
1787 | if (meta->map_ptr == NULL) { | |
1788 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
1789 | return -EINVAL; | |
1790 | } | |
1791 | ||
1792 | if (!BPF_MAP_PTR(aux->map_state)) | |
1793 | bpf_map_ptr_store(aux, meta->map_ptr, | |
1794 | meta->map_ptr->unpriv_array); | |
1795 | else if (BPF_MAP_PTR(aux->map_state) != meta->map_ptr) | |
1796 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, | |
1797 | meta->map_ptr->unpriv_array); | |
1798 | return 0; | |
1799 | } | |
1800 | ||
81ed18ab | 1801 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1802 | { |
17a52670 | 1803 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 1804 | struct bpf_reg_state *regs; |
33ff9823 | 1805 | struct bpf_call_arg_meta meta; |
969bf05e | 1806 | bool changes_data; |
17a52670 AS |
1807 | int i, err; |
1808 | ||
1809 | /* find function prototype */ | |
1810 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
1811 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
1812 | func_id); | |
17a52670 AS |
1813 | return -EINVAL; |
1814 | } | |
1815 | ||
00176a34 JK |
1816 | if (env->ops->get_func_proto) |
1817 | fn = env->ops->get_func_proto(func_id); | |
17a52670 AS |
1818 | |
1819 | if (!fn) { | |
61bd5218 JK |
1820 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
1821 | func_id); | |
17a52670 AS |
1822 | return -EINVAL; |
1823 | } | |
1824 | ||
1825 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1826 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
61bd5218 | 1827 | verbose(env, "cannot call GPL only function from proprietary program\n"); |
17a52670 AS |
1828 | return -EINVAL; |
1829 | } | |
1830 | ||
04514d13 | 1831 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 1832 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
1833 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
1834 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
1835 | func_id_name(func_id), func_id); | |
1836 | return -EINVAL; | |
1837 | } | |
969bf05e | 1838 | |
33ff9823 | 1839 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1840 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1841 | |
435faee1 DB |
1842 | /* We only support one arg being in raw mode at the moment, which |
1843 | * is sufficient for the helper functions we have right now. | |
1844 | */ | |
1845 | err = check_raw_mode(fn); | |
1846 | if (err) { | |
61bd5218 | 1847 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 1848 | func_id_name(func_id), func_id); |
435faee1 DB |
1849 | return err; |
1850 | } | |
1851 | ||
17a52670 | 1852 | /* check args */ |
33ff9823 | 1853 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1854 | if (err) |
1855 | return err; | |
33ff9823 | 1856 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1857 | if (err) |
1858 | return err; | |
33ff9823 | 1859 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1860 | if (err) |
1861 | return err; | |
33ff9823 | 1862 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1863 | if (err) |
1864 | return err; | |
33ff9823 | 1865 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1866 | if (err) |
1867 | return err; | |
1868 | ||
3f62bcc9 DB |
1869 | err = record_func_map(env, &meta, func_id, insn_idx); |
1870 | if (err) | |
1871 | return err; | |
1872 | ||
435faee1 DB |
1873 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1874 | * is inferred from register state. | |
1875 | */ | |
1876 | for (i = 0; i < meta.access_size; i++) { | |
57b4a36e DB |
1877 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
1878 | BPF_WRITE, -1, false); | |
435faee1 DB |
1879 | if (err) |
1880 | return err; | |
1881 | } | |
1882 | ||
638f5b90 | 1883 | regs = cur_regs(env); |
17a52670 | 1884 | /* reset caller saved regs */ |
dc503a8a | 1885 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 1886 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
1887 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
1888 | } | |
17a52670 | 1889 | |
dc503a8a | 1890 | /* update return register (already marked as written above) */ |
17a52670 | 1891 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 1892 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 1893 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
1894 | } else if (fn->ret_type == RET_VOID) { |
1895 | regs[BPF_REG_0].type = NOT_INIT; | |
1896 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
1897 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
f1174f77 | 1898 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 1899 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
1900 | /* remember map_ptr, so that check_map_access() |
1901 | * can check 'value_size' boundary of memory access | |
1902 | * to map element returned from bpf_map_lookup_elem() | |
1903 | */ | |
33ff9823 | 1904 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
1905 | verbose(env, |
1906 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
1907 | return -EINVAL; |
1908 | } | |
33ff9823 | 1909 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1910 | regs[BPF_REG_0].id = ++env->id_gen; |
17a52670 | 1911 | } else { |
61bd5218 | 1912 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 1913 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
1914 | return -EINVAL; |
1915 | } | |
04fd61ab | 1916 | |
61bd5218 | 1917 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
1918 | if (err) |
1919 | return err; | |
04fd61ab | 1920 | |
969bf05e AS |
1921 | if (changes_data) |
1922 | clear_all_pkt_pointers(env); | |
1923 | return 0; | |
1924 | } | |
1925 | ||
b03c9f9f EC |
1926 | static bool signed_add_overflows(s64 a, s64 b) |
1927 | { | |
1928 | /* Do the add in u64, where overflow is well-defined */ | |
1929 | s64 res = (s64)((u64)a + (u64)b); | |
1930 | ||
1931 | if (b < 0) | |
1932 | return res > a; | |
1933 | return res < a; | |
1934 | } | |
1935 | ||
1936 | static bool signed_sub_overflows(s64 a, s64 b) | |
1937 | { | |
1938 | /* Do the sub in u64, where overflow is well-defined */ | |
1939 | s64 res = (s64)((u64)a - (u64)b); | |
1940 | ||
1941 | if (b < 0) | |
1942 | return res < a; | |
1943 | return res > a; | |
969bf05e AS |
1944 | } |
1945 | ||
bb7f0f98 AS |
1946 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
1947 | const struct bpf_reg_state *reg, | |
1948 | enum bpf_reg_type type) | |
1949 | { | |
1950 | bool known = tnum_is_const(reg->var_off); | |
1951 | s64 val = reg->var_off.value; | |
1952 | s64 smin = reg->smin_value; | |
1953 | ||
1954 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
1955 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
1956 | reg_type_str[type], val); | |
1957 | return false; | |
1958 | } | |
1959 | ||
1960 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
1961 | verbose(env, "%s pointer offset %d is not allowed\n", | |
1962 | reg_type_str[type], reg->off); | |
1963 | return false; | |
1964 | } | |
1965 | ||
1966 | if (smin == S64_MIN) { | |
1967 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
1968 | reg_type_str[type]); | |
1969 | return false; | |
1970 | } | |
1971 | ||
1972 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
1973 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
1974 | smin, reg_type_str[type]); | |
1975 | return false; | |
1976 | } | |
1977 | ||
1978 | return true; | |
1979 | } | |
1980 | ||
010150ff DB |
1981 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
1982 | { | |
1983 | return &env->insn_aux_data[env->insn_idx]; | |
1984 | } | |
1985 | ||
1986 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, | |
1987 | u32 *ptr_limit, u8 opcode, bool off_is_neg) | |
1988 | { | |
1989 | bool mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
1990 | (opcode == BPF_SUB && !off_is_neg); | |
1991 | u32 off; | |
1992 | ||
1993 | switch (ptr_reg->type) { | |
1994 | case PTR_TO_STACK: | |
1995 | off = ptr_reg->off + ptr_reg->var_off.value; | |
1996 | if (mask_to_left) | |
1997 | *ptr_limit = MAX_BPF_STACK + off; | |
1998 | else | |
1999 | *ptr_limit = -off; | |
2000 | return 0; | |
2001 | case PTR_TO_MAP_VALUE: | |
2002 | if (mask_to_left) { | |
2003 | *ptr_limit = ptr_reg->umax_value + ptr_reg->off; | |
2004 | } else { | |
2005 | off = ptr_reg->smin_value + ptr_reg->off; | |
2006 | *ptr_limit = ptr_reg->map_ptr->value_size - off; | |
2007 | } | |
2008 | return 0; | |
2009 | default: | |
2010 | return -EINVAL; | |
2011 | } | |
2012 | } | |
2013 | ||
d51999cb DB |
2014 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
2015 | const struct bpf_insn *insn) | |
2016 | { | |
2017 | return env->allow_ptr_leaks || BPF_SRC(insn->code) == BPF_K; | |
2018 | } | |
2019 | ||
2020 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
2021 | u32 alu_state, u32 alu_limit) | |
2022 | { | |
2023 | /* If we arrived here from different branches with different | |
2024 | * state or limits to sanitize, then this won't work. | |
2025 | */ | |
2026 | if (aux->alu_state && | |
2027 | (aux->alu_state != alu_state || | |
2028 | aux->alu_limit != alu_limit)) | |
2029 | return -EACCES; | |
2030 | ||
2031 | /* Corresponding fixup done in fixup_bpf_calls(). */ | |
2032 | aux->alu_state = alu_state; | |
2033 | aux->alu_limit = alu_limit; | |
2034 | return 0; | |
2035 | } | |
2036 | ||
2037 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
2038 | struct bpf_insn *insn) | |
2039 | { | |
2040 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
2041 | ||
2042 | if (can_skip_alu_sanitation(env, insn)) | |
2043 | return 0; | |
2044 | ||
2045 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
2046 | } | |
2047 | ||
010150ff DB |
2048 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
2049 | struct bpf_insn *insn, | |
2050 | const struct bpf_reg_state *ptr_reg, | |
2051 | struct bpf_reg_state *dst_reg, | |
2052 | bool off_is_neg) | |
2053 | { | |
2054 | struct bpf_verifier_state *vstate = env->cur_state; | |
2055 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
2056 | bool ptr_is_dst_reg = ptr_reg == dst_reg; | |
2057 | u8 opcode = BPF_OP(insn->code); | |
2058 | u32 alu_state, alu_limit; | |
2059 | struct bpf_reg_state tmp; | |
2060 | bool ret; | |
2061 | ||
d51999cb | 2062 | if (can_skip_alu_sanitation(env, insn)) |
010150ff DB |
2063 | return 0; |
2064 | ||
2065 | /* We already marked aux for masking from non-speculative | |
2066 | * paths, thus we got here in the first place. We only care | |
2067 | * to explore bad access from here. | |
2068 | */ | |
2069 | if (vstate->speculative) | |
2070 | goto do_sim; | |
2071 | ||
2072 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
2073 | alu_state |= ptr_is_dst_reg ? | |
2074 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
2075 | ||
2076 | if (retrieve_ptr_limit(ptr_reg, &alu_limit, opcode, off_is_neg)) | |
2077 | return 0; | |
d51999cb | 2078 | if (update_alu_sanitation_state(aux, alu_state, alu_limit)) |
010150ff | 2079 | return -EACCES; |
010150ff DB |
2080 | do_sim: |
2081 | /* Simulate and find potential out-of-bounds access under | |
2082 | * speculative execution from truncation as a result of | |
2083 | * masking when off was not within expected range. If off | |
2084 | * sits in dst, then we temporarily need to move ptr there | |
2085 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
2086 | * for cases where we use K-based arithmetic in one direction | |
2087 | * and truncated reg-based in the other in order to explore | |
2088 | * bad access. | |
2089 | */ | |
2090 | if (!ptr_is_dst_reg) { | |
2091 | tmp = *dst_reg; | |
2092 | *dst_reg = *ptr_reg; | |
2093 | } | |
2094 | ret = push_stack(env, env->insn_idx + 1, env->insn_idx, true); | |
9cb89e91 | 2095 | if (!ptr_is_dst_reg && ret) |
010150ff DB |
2096 | *dst_reg = tmp; |
2097 | return !ret ? -EFAULT : 0; | |
2098 | } | |
2099 | ||
f1174f77 | 2100 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
2101 | * Caller should also handle BPF_MOV case separately. |
2102 | * If we return -EACCES, caller may want to try again treating pointer as a | |
2103 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
2104 | */ | |
2105 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
2106 | struct bpf_insn *insn, | |
2107 | const struct bpf_reg_state *ptr_reg, | |
2108 | const struct bpf_reg_state *off_reg) | |
969bf05e | 2109 | { |
638f5b90 | 2110 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg; |
f1174f77 | 2111 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
2112 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
2113 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
2114 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
2115 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
1bbe41a8 | 2116 | u32 dst = insn->dst_reg, src = insn->src_reg; |
969bf05e | 2117 | u8 opcode = BPF_OP(insn->code); |
010150ff | 2118 | int ret; |
969bf05e | 2119 | |
f1174f77 | 2120 | dst_reg = ®s[dst]; |
969bf05e | 2121 | |
6f16101e DB |
2122 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
2123 | smin_val > smax_val || umin_val > umax_val) { | |
2124 | /* Taint dst register if offset had invalid bounds derived from | |
2125 | * e.g. dead branches. | |
2126 | */ | |
2127 | __mark_reg_unknown(dst_reg); | |
2128 | return 0; | |
f1174f77 EC |
2129 | } |
2130 | ||
2131 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
2132 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
2133 | verbose(env, |
2134 | "R%d 32-bit pointer arithmetic prohibited\n", | |
2135 | dst); | |
f1174f77 | 2136 | return -EACCES; |
969bf05e AS |
2137 | } |
2138 | ||
f1174f77 | 2139 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
82abbf8d AS |
2140 | verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", |
2141 | dst); | |
f1174f77 EC |
2142 | return -EACCES; |
2143 | } | |
2144 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
82abbf8d AS |
2145 | verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", |
2146 | dst); | |
f1174f77 EC |
2147 | return -EACCES; |
2148 | } | |
2149 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
82abbf8d AS |
2150 | verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", |
2151 | dst); | |
f1174f77 EC |
2152 | return -EACCES; |
2153 | } | |
1bbe41a8 DB |
2154 | if (ptr_reg->type == PTR_TO_MAP_VALUE) { |
2155 | if (!env->allow_ptr_leaks && !known && (smin_val < 0) != (smax_val < 0)) { | |
2156 | verbose(env, "R%d has unknown scalar with mixed signed bounds, pointer arithmetic with it prohibited for !root\n", | |
2157 | off_reg == dst_reg ? dst : src); | |
2158 | return -EACCES; | |
2159 | } | |
2160 | } | |
f1174f77 EC |
2161 | |
2162 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
2163 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 2164 | */ |
f1174f77 EC |
2165 | dst_reg->type = ptr_reg->type; |
2166 | dst_reg->id = ptr_reg->id; | |
969bf05e | 2167 | |
bb7f0f98 AS |
2168 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
2169 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
2170 | return -EINVAL; | |
2171 | ||
f1174f77 EC |
2172 | switch (opcode) { |
2173 | case BPF_ADD: | |
010150ff DB |
2174 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
2175 | if (ret < 0) { | |
2176 | verbose(env, "R%d tried to add from different maps or paths\n", dst); | |
2177 | return ret; | |
2178 | } | |
f1174f77 EC |
2179 | /* We can take a fixed offset as long as it doesn't overflow |
2180 | * the s32 'off' field | |
969bf05e | 2181 | */ |
b03c9f9f EC |
2182 | if (known && (ptr_reg->off + smin_val == |
2183 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 2184 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
2185 | dst_reg->smin_value = smin_ptr; |
2186 | dst_reg->smax_value = smax_ptr; | |
2187 | dst_reg->umin_value = umin_ptr; | |
2188 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 2189 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 2190 | dst_reg->off = ptr_reg->off + smin_val; |
9f4f72e7 | 2191 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
2192 | break; |
2193 | } | |
f1174f77 EC |
2194 | /* A new variable offset is created. Note that off_reg->off |
2195 | * == 0, since it's a scalar. | |
2196 | * dst_reg gets the pointer type and since some positive | |
2197 | * integer value was added to the pointer, give it a new 'id' | |
2198 | * if it's a PTR_TO_PACKET. | |
2199 | * this creates a new 'base' pointer, off_reg (variable) gets | |
2200 | * added into the variable offset, and we copy the fixed offset | |
2201 | * from ptr_reg. | |
969bf05e | 2202 | */ |
b03c9f9f EC |
2203 | if (signed_add_overflows(smin_ptr, smin_val) || |
2204 | signed_add_overflows(smax_ptr, smax_val)) { | |
2205 | dst_reg->smin_value = S64_MIN; | |
2206 | dst_reg->smax_value = S64_MAX; | |
2207 | } else { | |
2208 | dst_reg->smin_value = smin_ptr + smin_val; | |
2209 | dst_reg->smax_value = smax_ptr + smax_val; | |
2210 | } | |
2211 | if (umin_ptr + umin_val < umin_ptr || | |
2212 | umax_ptr + umax_val < umax_ptr) { | |
2213 | dst_reg->umin_value = 0; | |
2214 | dst_reg->umax_value = U64_MAX; | |
2215 | } else { | |
2216 | dst_reg->umin_value = umin_ptr + umin_val; | |
2217 | dst_reg->umax_value = umax_ptr + umax_val; | |
2218 | } | |
f1174f77 EC |
2219 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
2220 | dst_reg->off = ptr_reg->off; | |
9f4f72e7 | 2221 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 2222 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2223 | dst_reg->id = ++env->id_gen; |
2224 | /* something was added to pkt_ptr, set range to zero */ | |
9f4f72e7 | 2225 | dst_reg->raw = 0; |
f1174f77 EC |
2226 | } |
2227 | break; | |
2228 | case BPF_SUB: | |
010150ff DB |
2229 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
2230 | if (ret < 0) { | |
2231 | verbose(env, "R%d tried to sub from different maps or paths\n", dst); | |
2232 | return ret; | |
2233 | } | |
f1174f77 EC |
2234 | if (dst_reg == off_reg) { |
2235 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
2236 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
2237 | dst); | |
f1174f77 EC |
2238 | return -EACCES; |
2239 | } | |
2240 | /* We don't allow subtraction from FP, because (according to | |
2241 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
2242 | * be able to deal with it. | |
969bf05e | 2243 | */ |
f1174f77 | 2244 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
2245 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
2246 | dst); | |
f1174f77 EC |
2247 | return -EACCES; |
2248 | } | |
b03c9f9f EC |
2249 | if (known && (ptr_reg->off - smin_val == |
2250 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 2251 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
2252 | dst_reg->smin_value = smin_ptr; |
2253 | dst_reg->smax_value = smax_ptr; | |
2254 | dst_reg->umin_value = umin_ptr; | |
2255 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
2256 | dst_reg->var_off = ptr_reg->var_off; |
2257 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 2258 | dst_reg->off = ptr_reg->off - smin_val; |
9f4f72e7 | 2259 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
2260 | break; |
2261 | } | |
f1174f77 EC |
2262 | /* A new variable offset is created. If the subtrahend is known |
2263 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 2264 | */ |
b03c9f9f EC |
2265 | if (signed_sub_overflows(smin_ptr, smax_val) || |
2266 | signed_sub_overflows(smax_ptr, smin_val)) { | |
2267 | /* Overflow possible, we know nothing */ | |
2268 | dst_reg->smin_value = S64_MIN; | |
2269 | dst_reg->smax_value = S64_MAX; | |
2270 | } else { | |
2271 | dst_reg->smin_value = smin_ptr - smax_val; | |
2272 | dst_reg->smax_value = smax_ptr - smin_val; | |
2273 | } | |
2274 | if (umin_ptr < umax_val) { | |
2275 | /* Overflow possible, we know nothing */ | |
2276 | dst_reg->umin_value = 0; | |
2277 | dst_reg->umax_value = U64_MAX; | |
2278 | } else { | |
2279 | /* Cannot overflow (as long as bounds are consistent) */ | |
2280 | dst_reg->umin_value = umin_ptr - umax_val; | |
2281 | dst_reg->umax_value = umax_ptr - umin_val; | |
2282 | } | |
f1174f77 EC |
2283 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
2284 | dst_reg->off = ptr_reg->off; | |
9f4f72e7 | 2285 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 2286 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2287 | dst_reg->id = ++env->id_gen; |
2288 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 2289 | if (smin_val < 0) |
9f4f72e7 | 2290 | dst_reg->raw = 0; |
43188702 | 2291 | } |
f1174f77 EC |
2292 | break; |
2293 | case BPF_AND: | |
2294 | case BPF_OR: | |
2295 | case BPF_XOR: | |
82abbf8d AS |
2296 | /* bitwise ops on pointers are troublesome, prohibit. */ |
2297 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
2298 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
2299 | return -EACCES; |
2300 | default: | |
2301 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
2302 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
2303 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 2304 | return -EACCES; |
43188702 JF |
2305 | } |
2306 | ||
bb7f0f98 AS |
2307 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
2308 | return -EINVAL; | |
2309 | ||
b03c9f9f EC |
2310 | __update_reg_bounds(dst_reg); |
2311 | __reg_deduce_bounds(dst_reg); | |
2312 | __reg_bound_offset(dst_reg); | |
d25f7129 DB |
2313 | |
2314 | /* For unprivileged we require that resulting offset must be in bounds | |
2315 | * in order to be able to sanitize access later on. | |
2316 | */ | |
4e28ad40 DB |
2317 | if (!env->allow_ptr_leaks) { |
2318 | if (dst_reg->type == PTR_TO_MAP_VALUE && | |
2319 | check_map_access(env, dst, dst_reg->off, 1, false)) { | |
2320 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
2321 | "prohibited for !root\n", dst); | |
2322 | return -EACCES; | |
2323 | } else if (dst_reg->type == PTR_TO_STACK && | |
2324 | check_stack_access(env, dst_reg, dst_reg->off + | |
2325 | dst_reg->var_off.value, 1)) { | |
2326 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
2327 | "prohibited for !root\n", dst); | |
2328 | return -EACCES; | |
2329 | } | |
d25f7129 DB |
2330 | } |
2331 | ||
43188702 JF |
2332 | return 0; |
2333 | } | |
2334 | ||
468f6eaf JH |
2335 | /* WARNING: This function does calculations on 64-bit values, but the actual |
2336 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
2337 | * need extra checks in the 32-bit case. | |
2338 | */ | |
f1174f77 EC |
2339 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
2340 | struct bpf_insn *insn, | |
2341 | struct bpf_reg_state *dst_reg, | |
2342 | struct bpf_reg_state src_reg) | |
969bf05e | 2343 | { |
638f5b90 | 2344 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 2345 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2346 | bool src_known, dst_known; |
b03c9f9f EC |
2347 | s64 smin_val, smax_val; |
2348 | u64 umin_val, umax_val; | |
468f6eaf | 2349 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
d51999cb DB |
2350 | u32 dst = insn->dst_reg; |
2351 | int ret; | |
48461135 | 2352 | |
4c6d059c JH |
2353 | if (insn_bitness == 32) { |
2354 | /* Relevant for 32-bit RSH: Information can propagate towards | |
2355 | * LSB, so it isn't sufficient to only truncate the output to | |
2356 | * 32 bits. | |
2357 | */ | |
2358 | coerce_reg_to_size(dst_reg, 4); | |
2359 | coerce_reg_to_size(&src_reg, 4); | |
2360 | } | |
2361 | ||
b03c9f9f EC |
2362 | smin_val = src_reg.smin_value; |
2363 | smax_val = src_reg.smax_value; | |
2364 | umin_val = src_reg.umin_value; | |
2365 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
2366 | src_known = tnum_is_const(src_reg.var_off); |
2367 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 2368 | |
6f16101e DB |
2369 | if ((src_known && (smin_val != smax_val || umin_val != umax_val)) || |
2370 | smin_val > smax_val || umin_val > umax_val) { | |
2371 | /* Taint dst register if offset had invalid bounds derived from | |
2372 | * e.g. dead branches. | |
2373 | */ | |
2374 | __mark_reg_unknown(dst_reg); | |
2375 | return 0; | |
2376 | } | |
2377 | ||
bb7f0f98 AS |
2378 | if (!src_known && |
2379 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
2380 | __mark_reg_unknown(dst_reg); | |
2381 | return 0; | |
2382 | } | |
2383 | ||
48461135 JB |
2384 | switch (opcode) { |
2385 | case BPF_ADD: | |
d51999cb DB |
2386 | ret = sanitize_val_alu(env, insn); |
2387 | if (ret < 0) { | |
2388 | verbose(env, "R%d tried to add from different pointers or scalars\n", dst); | |
2389 | return ret; | |
2390 | } | |
b03c9f9f EC |
2391 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
2392 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
2393 | dst_reg->smin_value = S64_MIN; | |
2394 | dst_reg->smax_value = S64_MAX; | |
2395 | } else { | |
2396 | dst_reg->smin_value += smin_val; | |
2397 | dst_reg->smax_value += smax_val; | |
2398 | } | |
2399 | if (dst_reg->umin_value + umin_val < umin_val || | |
2400 | dst_reg->umax_value + umax_val < umax_val) { | |
2401 | dst_reg->umin_value = 0; | |
2402 | dst_reg->umax_value = U64_MAX; | |
2403 | } else { | |
2404 | dst_reg->umin_value += umin_val; | |
2405 | dst_reg->umax_value += umax_val; | |
2406 | } | |
f1174f77 | 2407 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2408 | break; |
2409 | case BPF_SUB: | |
d51999cb DB |
2410 | ret = sanitize_val_alu(env, insn); |
2411 | if (ret < 0) { | |
2412 | verbose(env, "R%d tried to sub from different pointers or scalars\n", dst); | |
2413 | return ret; | |
2414 | } | |
b03c9f9f EC |
2415 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
2416 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
2417 | /* Overflow possible, we know nothing */ | |
2418 | dst_reg->smin_value = S64_MIN; | |
2419 | dst_reg->smax_value = S64_MAX; | |
2420 | } else { | |
2421 | dst_reg->smin_value -= smax_val; | |
2422 | dst_reg->smax_value -= smin_val; | |
2423 | } | |
2424 | if (dst_reg->umin_value < umax_val) { | |
2425 | /* Overflow possible, we know nothing */ | |
2426 | dst_reg->umin_value = 0; | |
2427 | dst_reg->umax_value = U64_MAX; | |
2428 | } else { | |
2429 | /* Cannot overflow (as long as bounds are consistent) */ | |
2430 | dst_reg->umin_value -= umax_val; | |
2431 | dst_reg->umax_value -= umin_val; | |
2432 | } | |
f1174f77 | 2433 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2434 | break; |
2435 | case BPF_MUL: | |
b03c9f9f EC |
2436 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
2437 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 2438 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
2439 | __mark_reg_unbounded(dst_reg); |
2440 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2441 | break; |
2442 | } | |
b03c9f9f EC |
2443 | /* Both values are positive, so we can work with unsigned and |
2444 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 2445 | */ |
b03c9f9f EC |
2446 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
2447 | /* Potential overflow, we know nothing */ | |
2448 | __mark_reg_unbounded(dst_reg); | |
2449 | /* (except what we can learn from the var_off) */ | |
2450 | __update_reg_bounds(dst_reg); | |
2451 | break; | |
2452 | } | |
2453 | dst_reg->umin_value *= umin_val; | |
2454 | dst_reg->umax_value *= umax_val; | |
2455 | if (dst_reg->umax_value > S64_MAX) { | |
2456 | /* Overflow possible, we know nothing */ | |
2457 | dst_reg->smin_value = S64_MIN; | |
2458 | dst_reg->smax_value = S64_MAX; | |
2459 | } else { | |
2460 | dst_reg->smin_value = dst_reg->umin_value; | |
2461 | dst_reg->smax_value = dst_reg->umax_value; | |
2462 | } | |
48461135 JB |
2463 | break; |
2464 | case BPF_AND: | |
f1174f77 | 2465 | if (src_known && dst_known) { |
b03c9f9f EC |
2466 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
2467 | src_reg.var_off.value); | |
f1174f77 EC |
2468 | break; |
2469 | } | |
b03c9f9f EC |
2470 | /* We get our minimum from the var_off, since that's inherently |
2471 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 2472 | */ |
f1174f77 | 2473 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
2474 | dst_reg->umin_value = dst_reg->var_off.value; |
2475 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
2476 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2477 | /* Lose signed bounds when ANDing negative numbers, | |
2478 | * ain't nobody got time for that. | |
2479 | */ | |
2480 | dst_reg->smin_value = S64_MIN; | |
2481 | dst_reg->smax_value = S64_MAX; | |
2482 | } else { | |
2483 | /* ANDing two positives gives a positive, so safe to | |
2484 | * cast result into s64. | |
2485 | */ | |
2486 | dst_reg->smin_value = dst_reg->umin_value; | |
2487 | dst_reg->smax_value = dst_reg->umax_value; | |
2488 | } | |
2489 | /* We may learn something more from the var_off */ | |
2490 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2491 | break; |
2492 | case BPF_OR: | |
2493 | if (src_known && dst_known) { | |
b03c9f9f EC |
2494 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
2495 | src_reg.var_off.value); | |
f1174f77 EC |
2496 | break; |
2497 | } | |
b03c9f9f EC |
2498 | /* We get our maximum from the var_off, and our minimum is the |
2499 | * maximum of the operands' minima | |
f1174f77 EC |
2500 | */ |
2501 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
2502 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
2503 | dst_reg->umax_value = dst_reg->var_off.value | | |
2504 | dst_reg->var_off.mask; | |
2505 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2506 | /* Lose signed bounds when ORing negative numbers, | |
2507 | * ain't nobody got time for that. | |
2508 | */ | |
2509 | dst_reg->smin_value = S64_MIN; | |
2510 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2511 | } else { |
b03c9f9f EC |
2512 | /* ORing two positives gives a positive, so safe to |
2513 | * cast result into s64. | |
2514 | */ | |
2515 | dst_reg->smin_value = dst_reg->umin_value; | |
2516 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 2517 | } |
b03c9f9f EC |
2518 | /* We may learn something more from the var_off */ |
2519 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2520 | break; |
2521 | case BPF_LSH: | |
468f6eaf JH |
2522 | if (umax_val >= insn_bitness) { |
2523 | /* Shifts greater than 31 or 63 are undefined. | |
2524 | * This includes shifts by a negative number. | |
b03c9f9f | 2525 | */ |
61bd5218 | 2526 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2527 | break; |
2528 | } | |
b03c9f9f EC |
2529 | /* We lose all sign bit information (except what we can pick |
2530 | * up from var_off) | |
48461135 | 2531 | */ |
b03c9f9f EC |
2532 | dst_reg->smin_value = S64_MIN; |
2533 | dst_reg->smax_value = S64_MAX; | |
2534 | /* If we might shift our top bit out, then we know nothing */ | |
2535 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
2536 | dst_reg->umin_value = 0; | |
2537 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 2538 | } else { |
b03c9f9f EC |
2539 | dst_reg->umin_value <<= umin_val; |
2540 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 2541 | } |
b03c9f9f EC |
2542 | if (src_known) |
2543 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); | |
2544 | else | |
2545 | dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val); | |
2546 | /* We may learn something more from the var_off */ | |
2547 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2548 | break; |
2549 | case BPF_RSH: | |
468f6eaf JH |
2550 | if (umax_val >= insn_bitness) { |
2551 | /* Shifts greater than 31 or 63 are undefined. | |
2552 | * This includes shifts by a negative number. | |
b03c9f9f | 2553 | */ |
61bd5218 | 2554 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2555 | break; |
2556 | } | |
4374f256 EC |
2557 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
2558 | * be negative, then either: | |
2559 | * 1) src_reg might be zero, so the sign bit of the result is | |
2560 | * unknown, so we lose our signed bounds | |
2561 | * 2) it's known negative, thus the unsigned bounds capture the | |
2562 | * signed bounds | |
2563 | * 3) the signed bounds cross zero, so they tell us nothing | |
2564 | * about the result | |
2565 | * If the value in dst_reg is known nonnegative, then again the | |
2566 | * unsigned bounts capture the signed bounds. | |
2567 | * Thus, in all cases it suffices to blow away our signed bounds | |
2568 | * and rely on inferring new ones from the unsigned bounds and | |
2569 | * var_off of the result. | |
2570 | */ | |
2571 | dst_reg->smin_value = S64_MIN; | |
2572 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2573 | if (src_known) |
b03c9f9f EC |
2574 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, |
2575 | umin_val); | |
f1174f77 | 2576 | else |
b03c9f9f EC |
2577 | dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val); |
2578 | dst_reg->umin_value >>= umax_val; | |
2579 | dst_reg->umax_value >>= umin_val; | |
2580 | /* We may learn something more from the var_off */ | |
2581 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2582 | break; |
2583 | default: | |
61bd5218 | 2584 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
2585 | break; |
2586 | } | |
2587 | ||
468f6eaf JH |
2588 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
2589 | /* 32-bit ALU ops are (32,32)->32 */ | |
2590 | coerce_reg_to_size(dst_reg, 4); | |
468f6eaf JH |
2591 | } |
2592 | ||
b03c9f9f EC |
2593 | __reg_deduce_bounds(dst_reg); |
2594 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
2595 | return 0; |
2596 | } | |
2597 | ||
2598 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
2599 | * and var_off. | |
2600 | */ | |
2601 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
2602 | struct bpf_insn *insn) | |
2603 | { | |
638f5b90 | 2604 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg, *src_reg; |
f1174f77 EC |
2605 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
2606 | u8 opcode = BPF_OP(insn->code); | |
f1174f77 EC |
2607 | |
2608 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
2609 | src_reg = NULL; |
2610 | if (dst_reg->type != SCALAR_VALUE) | |
2611 | ptr_reg = dst_reg; | |
2612 | if (BPF_SRC(insn->code) == BPF_X) { | |
2613 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
2614 | if (src_reg->type != SCALAR_VALUE) { |
2615 | if (dst_reg->type != SCALAR_VALUE) { | |
2616 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
2617 | * an arbitrary scalar. Disallow all math except |
2618 | * pointer subtraction | |
f1174f77 | 2619 | */ |
e60e198a | 2620 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
2621 | mark_reg_unknown(env, regs, insn->dst_reg); |
2622 | return 0; | |
f1174f77 | 2623 | } |
82abbf8d AS |
2624 | verbose(env, "R%d pointer %s pointer prohibited\n", |
2625 | insn->dst_reg, | |
2626 | bpf_alu_string[opcode >> 4]); | |
2627 | return -EACCES; | |
f1174f77 EC |
2628 | } else { |
2629 | /* scalar += pointer | |
2630 | * This is legal, but we have to reverse our | |
2631 | * src/dest handling in computing the range | |
2632 | */ | |
82abbf8d AS |
2633 | return adjust_ptr_min_max_vals(env, insn, |
2634 | src_reg, dst_reg); | |
f1174f77 EC |
2635 | } |
2636 | } else if (ptr_reg) { | |
2637 | /* pointer += scalar */ | |
82abbf8d AS |
2638 | return adjust_ptr_min_max_vals(env, insn, |
2639 | dst_reg, src_reg); | |
f1174f77 EC |
2640 | } |
2641 | } else { | |
2642 | /* Pretend the src is a reg with a known value, since we only | |
2643 | * need to be able to read from this state. | |
2644 | */ | |
2645 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 2646 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 2647 | src_reg = &off_reg; |
82abbf8d AS |
2648 | if (ptr_reg) /* pointer += K */ |
2649 | return adjust_ptr_min_max_vals(env, insn, | |
2650 | ptr_reg, src_reg); | |
f1174f77 EC |
2651 | } |
2652 | ||
2653 | /* Got here implies adding two SCALAR_VALUEs */ | |
2654 | if (WARN_ON_ONCE(ptr_reg)) { | |
638f5b90 | 2655 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2656 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
2657 | return -EINVAL; |
2658 | } | |
2659 | if (WARN_ON(!src_reg)) { | |
638f5b90 | 2660 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2661 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
2662 | return -EINVAL; |
2663 | } | |
2664 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
2665 | } |
2666 | ||
17a52670 | 2667 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 2668 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2669 | { |
638f5b90 | 2670 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
2671 | u8 opcode = BPF_OP(insn->code); |
2672 | int err; | |
2673 | ||
2674 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
2675 | if (opcode == BPF_NEG) { | |
2676 | if (BPF_SRC(insn->code) != 0 || | |
2677 | insn->src_reg != BPF_REG_0 || | |
2678 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 2679 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
2680 | return -EINVAL; |
2681 | } | |
2682 | } else { | |
2683 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
2684 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
2685 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 2686 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
2687 | return -EINVAL; |
2688 | } | |
2689 | } | |
2690 | ||
2691 | /* check src operand */ | |
dc503a8a | 2692 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2693 | if (err) |
2694 | return err; | |
2695 | ||
1be7f75d | 2696 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 2697 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
2698 | insn->dst_reg); |
2699 | return -EACCES; | |
2700 | } | |
2701 | ||
17a52670 | 2702 | /* check dest operand */ |
dc503a8a | 2703 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2704 | if (err) |
2705 | return err; | |
2706 | ||
2707 | } else if (opcode == BPF_MOV) { | |
2708 | ||
2709 | if (BPF_SRC(insn->code) == BPF_X) { | |
2710 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2711 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2712 | return -EINVAL; |
2713 | } | |
2714 | ||
2715 | /* check src operand */ | |
dc503a8a | 2716 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2717 | if (err) |
2718 | return err; | |
2719 | } else { | |
2720 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2721 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2722 | return -EINVAL; |
2723 | } | |
2724 | } | |
2725 | ||
2726 | /* check dest operand */ | |
dc503a8a | 2727 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2728 | if (err) |
2729 | return err; | |
2730 | ||
2731 | if (BPF_SRC(insn->code) == BPF_X) { | |
a25d38b5 JW |
2732 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
2733 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
2734 | ||
17a52670 AS |
2735 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
2736 | /* case: R1 = R2 | |
2737 | * copy register state to dest reg | |
2738 | */ | |
a25d38b5 JW |
2739 | *dst_reg = *src_reg; |
2740 | dst_reg->live |= REG_LIVE_WRITTEN; | |
17a52670 | 2741 | } else { |
f1174f77 | 2742 | /* R1 = (u32) R2 */ |
1be7f75d | 2743 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
2744 | verbose(env, |
2745 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
2746 | insn->src_reg); |
2747 | return -EACCES; | |
a25d38b5 JW |
2748 | } else if (src_reg->type == SCALAR_VALUE) { |
2749 | *dst_reg = *src_reg; | |
2750 | dst_reg->live |= REG_LIVE_WRITTEN; | |
2751 | } else { | |
2752 | mark_reg_unknown(env, regs, | |
2753 | insn->dst_reg); | |
1be7f75d | 2754 | } |
a25d38b5 | 2755 | coerce_reg_to_size(dst_reg, 4); |
17a52670 AS |
2756 | } |
2757 | } else { | |
2758 | /* case: R = imm | |
2759 | * remember the value we stored into this reg | |
2760 | */ | |
f1174f77 | 2761 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
2762 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
2763 | __mark_reg_known(regs + insn->dst_reg, | |
2764 | insn->imm); | |
2765 | } else { | |
2766 | __mark_reg_known(regs + insn->dst_reg, | |
2767 | (u32)insn->imm); | |
2768 | } | |
17a52670 AS |
2769 | } |
2770 | ||
2771 | } else if (opcode > BPF_END) { | |
61bd5218 | 2772 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
2773 | return -EINVAL; |
2774 | ||
2775 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
2776 | ||
17a52670 AS |
2777 | if (BPF_SRC(insn->code) == BPF_X) { |
2778 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2779 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2780 | return -EINVAL; |
2781 | } | |
2782 | /* check src1 operand */ | |
dc503a8a | 2783 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2784 | if (err) |
2785 | return err; | |
2786 | } else { | |
2787 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2788 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2789 | return -EINVAL; |
2790 | } | |
2791 | } | |
2792 | ||
2793 | /* check src2 operand */ | |
dc503a8a | 2794 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2795 | if (err) |
2796 | return err; | |
2797 | ||
2798 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
2799 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 2800 | verbose(env, "div by zero\n"); |
17a52670 AS |
2801 | return -EINVAL; |
2802 | } | |
2803 | ||
7891a87e DB |
2804 | if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) { |
2805 | verbose(env, "BPF_ARSH not supported for 32 bit ALU\n"); | |
2806 | return -EINVAL; | |
2807 | } | |
2808 | ||
229394e8 RV |
2809 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
2810 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
2811 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
2812 | ||
2813 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 2814 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
2815 | return -EINVAL; |
2816 | } | |
2817 | } | |
2818 | ||
1a0dc1ac | 2819 | /* check dest operand */ |
dc503a8a | 2820 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
2821 | if (err) |
2822 | return err; | |
2823 | ||
f1174f77 | 2824 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
2825 | } |
2826 | ||
2827 | return 0; | |
2828 | } | |
2829 | ||
58e2af8b | 2830 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
de8f3a83 | 2831 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 2832 | enum bpf_reg_type type, |
fb2a311a | 2833 | bool range_right_open) |
969bf05e | 2834 | { |
58e2af8b | 2835 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 2836 | u16 new_range; |
969bf05e | 2837 | int i; |
2d2be8ca | 2838 | |
fb2a311a DB |
2839 | if (dst_reg->off < 0 || |
2840 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
2841 | /* This doesn't give us any range */ |
2842 | return; | |
2843 | ||
b03c9f9f EC |
2844 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
2845 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
2846 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
2847 | * than pkt_end, but that's because it's also less than pkt. | |
2848 | */ | |
2849 | return; | |
2850 | ||
fb2a311a DB |
2851 | new_range = dst_reg->off; |
2852 | if (range_right_open) | |
2853 | new_range--; | |
2854 | ||
2855 | /* Examples for register markings: | |
2d2be8ca | 2856 | * |
fb2a311a | 2857 | * pkt_data in dst register: |
2d2be8ca DB |
2858 | * |
2859 | * r2 = r3; | |
2860 | * r2 += 8; | |
2861 | * if (r2 > pkt_end) goto <handle exception> | |
2862 | * <access okay> | |
2863 | * | |
b4e432f1 DB |
2864 | * r2 = r3; |
2865 | * r2 += 8; | |
2866 | * if (r2 < pkt_end) goto <access okay> | |
2867 | * <handle exception> | |
2868 | * | |
2d2be8ca DB |
2869 | * Where: |
2870 | * r2 == dst_reg, pkt_end == src_reg | |
2871 | * r2=pkt(id=n,off=8,r=0) | |
2872 | * r3=pkt(id=n,off=0,r=0) | |
2873 | * | |
fb2a311a | 2874 | * pkt_data in src register: |
2d2be8ca DB |
2875 | * |
2876 | * r2 = r3; | |
2877 | * r2 += 8; | |
2878 | * if (pkt_end >= r2) goto <access okay> | |
2879 | * <handle exception> | |
2880 | * | |
b4e432f1 DB |
2881 | * r2 = r3; |
2882 | * r2 += 8; | |
2883 | * if (pkt_end <= r2) goto <handle exception> | |
2884 | * <access okay> | |
2885 | * | |
2d2be8ca DB |
2886 | * Where: |
2887 | * pkt_end == dst_reg, r2 == src_reg | |
2888 | * r2=pkt(id=n,off=8,r=0) | |
2889 | * r3=pkt(id=n,off=0,r=0) | |
2890 | * | |
2891 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
2892 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
2893 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
2894 | * the check. | |
969bf05e | 2895 | */ |
2d2be8ca | 2896 | |
f1174f77 EC |
2897 | /* If our ids match, then we must have the same max_value. And we |
2898 | * don't care about the other reg's fixed offset, since if it's too big | |
2899 | * the range won't allow anything. | |
2900 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
2901 | */ | |
969bf05e | 2902 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 2903 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 2904 | /* keep the maximum range already checked */ |
fb2a311a | 2905 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 2906 | |
638f5b90 AS |
2907 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2908 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 2909 | continue; |
638f5b90 | 2910 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 | 2911 | if (reg->type == type && reg->id == dst_reg->id) |
b06723da | 2912 | reg->range = max(reg->range, new_range); |
969bf05e AS |
2913 | } |
2914 | } | |
2915 | ||
c779b6c6 AS |
2916 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
2917 | * and return: | |
2918 | * 1 - branch will be taken and "goto target" will be executed | |
2919 | * 0 - branch will not be taken and fall-through to next insn | |
2920 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value range [0,10] | |
2921 | */ | |
2922 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
2923 | { | |
2924 | if (__is_pointer_value(false, reg)) | |
2925 | return -1; | |
2926 | ||
2927 | switch (opcode) { | |
2928 | case BPF_JEQ: | |
2929 | if (tnum_is_const(reg->var_off)) | |
2930 | return !!tnum_equals_const(reg->var_off, val); | |
2931 | break; | |
2932 | case BPF_JNE: | |
2933 | if (tnum_is_const(reg->var_off)) | |
2934 | return !tnum_equals_const(reg->var_off, val); | |
2935 | break; | |
2936 | case BPF_JGT: | |
2937 | if (reg->umin_value > val) | |
2938 | return 1; | |
2939 | else if (reg->umax_value <= val) | |
2940 | return 0; | |
2941 | break; | |
2942 | case BPF_JSGT: | |
2943 | if (reg->smin_value > (s64)val) | |
2944 | return 1; | |
2945 | else if (reg->smax_value < (s64)val) | |
2946 | return 0; | |
2947 | break; | |
2948 | case BPF_JLT: | |
2949 | if (reg->umax_value < val) | |
2950 | return 1; | |
2951 | else if (reg->umin_value >= val) | |
2952 | return 0; | |
2953 | break; | |
2954 | case BPF_JSLT: | |
2955 | if (reg->smax_value < (s64)val) | |
2956 | return 1; | |
2957 | else if (reg->smin_value >= (s64)val) | |
2958 | return 0; | |
2959 | break; | |
2960 | case BPF_JGE: | |
2961 | if (reg->umin_value >= val) | |
2962 | return 1; | |
2963 | else if (reg->umax_value < val) | |
2964 | return 0; | |
2965 | break; | |
2966 | case BPF_JSGE: | |
2967 | if (reg->smin_value >= (s64)val) | |
2968 | return 1; | |
2969 | else if (reg->smax_value < (s64)val) | |
2970 | return 0; | |
2971 | break; | |
2972 | case BPF_JLE: | |
2973 | if (reg->umax_value <= val) | |
2974 | return 1; | |
2975 | else if (reg->umin_value > val) | |
2976 | return 0; | |
2977 | break; | |
2978 | case BPF_JSLE: | |
2979 | if (reg->smax_value <= (s64)val) | |
2980 | return 1; | |
2981 | else if (reg->smin_value > (s64)val) | |
2982 | return 0; | |
2983 | break; | |
2984 | } | |
2985 | ||
2986 | return -1; | |
2987 | } | |
2988 | ||
48461135 JB |
2989 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2990 | * variable register that we are working on, and src_reg is a constant or we're | |
2991 | * simply doing a BPF_K check. | |
f1174f77 | 2992 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
2993 | */ |
2994 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2995 | struct bpf_reg_state *false_reg, u64 val, | |
2996 | u8 opcode) | |
2997 | { | |
f1174f77 EC |
2998 | /* If the dst_reg is a pointer, we can't learn anything about its |
2999 | * variable offset from the compare (unless src_reg were a pointer into | |
3000 | * the same object, but we don't bother with that. | |
3001 | * Since false_reg and true_reg have the same type by construction, we | |
3002 | * only need to check one of them for pointerness. | |
3003 | */ | |
3004 | if (__is_pointer_value(false, false_reg)) | |
3005 | return; | |
4cabc5b1 | 3006 | |
48461135 JB |
3007 | switch (opcode) { |
3008 | case BPF_JEQ: | |
3009 | /* If this is false then we know nothing Jon Snow, but if it is | |
3010 | * true then we know for sure. | |
3011 | */ | |
b03c9f9f | 3012 | __mark_reg_known(true_reg, val); |
48461135 JB |
3013 | break; |
3014 | case BPF_JNE: | |
3015 | /* If this is true we know nothing Jon Snow, but if it is false | |
3016 | * we know the value for sure; | |
3017 | */ | |
b03c9f9f | 3018 | __mark_reg_known(false_reg, val); |
48461135 JB |
3019 | break; |
3020 | case BPF_JGT: | |
b03c9f9f EC |
3021 | false_reg->umax_value = min(false_reg->umax_value, val); |
3022 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
3023 | break; | |
48461135 | 3024 | case BPF_JSGT: |
b03c9f9f EC |
3025 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
3026 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 3027 | break; |
b4e432f1 DB |
3028 | case BPF_JLT: |
3029 | false_reg->umin_value = max(false_reg->umin_value, val); | |
3030 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
3031 | break; | |
3032 | case BPF_JSLT: | |
3033 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
3034 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
3035 | break; | |
48461135 | 3036 | case BPF_JGE: |
b03c9f9f EC |
3037 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
3038 | true_reg->umin_value = max(true_reg->umin_value, val); | |
3039 | break; | |
48461135 | 3040 | case BPF_JSGE: |
b03c9f9f EC |
3041 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
3042 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 3043 | break; |
b4e432f1 DB |
3044 | case BPF_JLE: |
3045 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
3046 | true_reg->umax_value = min(true_reg->umax_value, val); | |
3047 | break; | |
3048 | case BPF_JSLE: | |
3049 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
3050 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
3051 | break; | |
48461135 JB |
3052 | default: |
3053 | break; | |
3054 | } | |
3055 | ||
b03c9f9f EC |
3056 | __reg_deduce_bounds(false_reg); |
3057 | __reg_deduce_bounds(true_reg); | |
3058 | /* We might have learned some bits from the bounds. */ | |
3059 | __reg_bound_offset(false_reg); | |
3060 | __reg_bound_offset(true_reg); | |
3061 | /* Intersecting with the old var_off might have improved our bounds | |
3062 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3063 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3064 | */ | |
3065 | __update_reg_bounds(false_reg); | |
3066 | __update_reg_bounds(true_reg); | |
48461135 JB |
3067 | } |
3068 | ||
f1174f77 EC |
3069 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
3070 | * the variable reg. | |
48461135 JB |
3071 | */ |
3072 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3073 | struct bpf_reg_state *false_reg, u64 val, | |
3074 | u8 opcode) | |
3075 | { | |
f1174f77 EC |
3076 | if (__is_pointer_value(false, false_reg)) |
3077 | return; | |
4cabc5b1 | 3078 | |
48461135 JB |
3079 | switch (opcode) { |
3080 | case BPF_JEQ: | |
3081 | /* If this is false then we know nothing Jon Snow, but if it is | |
3082 | * true then we know for sure. | |
3083 | */ | |
b03c9f9f | 3084 | __mark_reg_known(true_reg, val); |
48461135 JB |
3085 | break; |
3086 | case BPF_JNE: | |
3087 | /* If this is true we know nothing Jon Snow, but if it is false | |
3088 | * we know the value for sure; | |
3089 | */ | |
b03c9f9f | 3090 | __mark_reg_known(false_reg, val); |
48461135 JB |
3091 | break; |
3092 | case BPF_JGT: | |
b03c9f9f EC |
3093 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
3094 | false_reg->umin_value = max(false_reg->umin_value, val); | |
3095 | break; | |
48461135 | 3096 | case BPF_JSGT: |
b03c9f9f EC |
3097 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
3098 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 3099 | break; |
b4e432f1 DB |
3100 | case BPF_JLT: |
3101 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
3102 | false_reg->umax_value = min(false_reg->umax_value, val); | |
3103 | break; | |
3104 | case BPF_JSLT: | |
3105 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
3106 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
3107 | break; | |
48461135 | 3108 | case BPF_JGE: |
b03c9f9f EC |
3109 | true_reg->umax_value = min(true_reg->umax_value, val); |
3110 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
3111 | break; | |
48461135 | 3112 | case BPF_JSGE: |
b03c9f9f EC |
3113 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
3114 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 3115 | break; |
b4e432f1 DB |
3116 | case BPF_JLE: |
3117 | true_reg->umin_value = max(true_reg->umin_value, val); | |
3118 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
3119 | break; | |
3120 | case BPF_JSLE: | |
3121 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
3122 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
3123 | break; | |
48461135 JB |
3124 | default: |
3125 | break; | |
3126 | } | |
3127 | ||
b03c9f9f EC |
3128 | __reg_deduce_bounds(false_reg); |
3129 | __reg_deduce_bounds(true_reg); | |
3130 | /* We might have learned some bits from the bounds. */ | |
3131 | __reg_bound_offset(false_reg); | |
3132 | __reg_bound_offset(true_reg); | |
3133 | /* Intersecting with the old var_off might have improved our bounds | |
3134 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3135 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3136 | */ | |
3137 | __update_reg_bounds(false_reg); | |
3138 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
3139 | } |
3140 | ||
3141 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
3142 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
3143 | struct bpf_reg_state *dst_reg) | |
3144 | { | |
b03c9f9f EC |
3145 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
3146 | dst_reg->umin_value); | |
3147 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
3148 | dst_reg->umax_value); | |
3149 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
3150 | dst_reg->smin_value); | |
3151 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
3152 | dst_reg->smax_value); | |
f1174f77 EC |
3153 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
3154 | dst_reg->var_off); | |
b03c9f9f EC |
3155 | /* We might have learned new bounds from the var_off. */ |
3156 | __update_reg_bounds(src_reg); | |
3157 | __update_reg_bounds(dst_reg); | |
3158 | /* We might have learned something about the sign bit. */ | |
3159 | __reg_deduce_bounds(src_reg); | |
3160 | __reg_deduce_bounds(dst_reg); | |
3161 | /* We might have learned some bits from the bounds. */ | |
3162 | __reg_bound_offset(src_reg); | |
3163 | __reg_bound_offset(dst_reg); | |
3164 | /* Intersecting with the old var_off might have improved our bounds | |
3165 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3166 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3167 | */ | |
3168 | __update_reg_bounds(src_reg); | |
3169 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
3170 | } |
3171 | ||
3172 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
3173 | struct bpf_reg_state *true_dst, | |
3174 | struct bpf_reg_state *false_src, | |
3175 | struct bpf_reg_state *false_dst, | |
3176 | u8 opcode) | |
3177 | { | |
3178 | switch (opcode) { | |
3179 | case BPF_JEQ: | |
3180 | __reg_combine_min_max(true_src, true_dst); | |
3181 | break; | |
3182 | case BPF_JNE: | |
3183 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 3184 | break; |
4cabc5b1 | 3185 | } |
48461135 JB |
3186 | } |
3187 | ||
57a09bf0 | 3188 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 3189 | bool is_null) |
57a09bf0 TG |
3190 | { |
3191 | struct bpf_reg_state *reg = ®s[regno]; | |
3192 | ||
3193 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
3194 | /* Old offset (both fixed and variable parts) should |
3195 | * have been known-zero, because we don't allow pointer | |
3196 | * arithmetic on pointers that might be NULL. | |
3197 | */ | |
b03c9f9f EC |
3198 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
3199 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 3200 | reg->off)) { |
b03c9f9f EC |
3201 | __mark_reg_known_zero(reg); |
3202 | reg->off = 0; | |
f1174f77 EC |
3203 | } |
3204 | if (is_null) { | |
3205 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
3206 | } else if (reg->map_ptr->inner_map_meta) { |
3207 | reg->type = CONST_PTR_TO_MAP; | |
3208 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
3209 | } else { | |
f1174f77 | 3210 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 3211 | } |
a08dd0da DB |
3212 | /* We don't need id from this point onwards anymore, thus we |
3213 | * should better reset it, so that state pruning has chances | |
3214 | * to take effect. | |
3215 | */ | |
3216 | reg->id = 0; | |
57a09bf0 TG |
3217 | } |
3218 | } | |
3219 | ||
3220 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
3221 | * be folded together at some point. | |
3222 | */ | |
3223 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
f1174f77 | 3224 | bool is_null) |
57a09bf0 TG |
3225 | { |
3226 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 3227 | u32 id = regs[regno].id; |
57a09bf0 TG |
3228 | int i; |
3229 | ||
3230 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 3231 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 | 3232 | |
638f5b90 AS |
3233 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
3234 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
57a09bf0 | 3235 | continue; |
638f5b90 | 3236 | mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null); |
57a09bf0 TG |
3237 | } |
3238 | } | |
3239 | ||
5beca081 DB |
3240 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
3241 | struct bpf_reg_state *dst_reg, | |
3242 | struct bpf_reg_state *src_reg, | |
3243 | struct bpf_verifier_state *this_branch, | |
3244 | struct bpf_verifier_state *other_branch) | |
3245 | { | |
3246 | if (BPF_SRC(insn->code) != BPF_X) | |
3247 | return false; | |
3248 | ||
3249 | switch (BPF_OP(insn->code)) { | |
3250 | case BPF_JGT: | |
3251 | if ((dst_reg->type == PTR_TO_PACKET && | |
3252 | src_reg->type == PTR_TO_PACKET_END) || | |
3253 | (dst_reg->type == PTR_TO_PACKET_META && | |
3254 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3255 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
3256 | find_good_pkt_pointers(this_branch, dst_reg, | |
3257 | dst_reg->type, false); | |
3258 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3259 | src_reg->type == PTR_TO_PACKET) || | |
3260 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3261 | src_reg->type == PTR_TO_PACKET_META)) { | |
3262 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
3263 | find_good_pkt_pointers(other_branch, src_reg, | |
3264 | src_reg->type, true); | |
3265 | } else { | |
3266 | return false; | |
3267 | } | |
3268 | break; | |
3269 | case BPF_JLT: | |
3270 | if ((dst_reg->type == PTR_TO_PACKET && | |
3271 | src_reg->type == PTR_TO_PACKET_END) || | |
3272 | (dst_reg->type == PTR_TO_PACKET_META && | |
3273 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3274 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
3275 | find_good_pkt_pointers(other_branch, dst_reg, | |
3276 | dst_reg->type, true); | |
3277 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3278 | src_reg->type == PTR_TO_PACKET) || | |
3279 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3280 | src_reg->type == PTR_TO_PACKET_META)) { | |
3281 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
3282 | find_good_pkt_pointers(this_branch, src_reg, | |
3283 | src_reg->type, false); | |
3284 | } else { | |
3285 | return false; | |
3286 | } | |
3287 | break; | |
3288 | case BPF_JGE: | |
3289 | if ((dst_reg->type == PTR_TO_PACKET && | |
3290 | src_reg->type == PTR_TO_PACKET_END) || | |
3291 | (dst_reg->type == PTR_TO_PACKET_META && | |
3292 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3293 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
3294 | find_good_pkt_pointers(this_branch, dst_reg, | |
3295 | dst_reg->type, true); | |
3296 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3297 | src_reg->type == PTR_TO_PACKET) || | |
3298 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3299 | src_reg->type == PTR_TO_PACKET_META)) { | |
3300 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
3301 | find_good_pkt_pointers(other_branch, src_reg, | |
3302 | src_reg->type, false); | |
3303 | } else { | |
3304 | return false; | |
3305 | } | |
3306 | break; | |
3307 | case BPF_JLE: | |
3308 | if ((dst_reg->type == PTR_TO_PACKET && | |
3309 | src_reg->type == PTR_TO_PACKET_END) || | |
3310 | (dst_reg->type == PTR_TO_PACKET_META && | |
3311 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3312 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
3313 | find_good_pkt_pointers(other_branch, dst_reg, | |
3314 | dst_reg->type, false); | |
3315 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3316 | src_reg->type == PTR_TO_PACKET) || | |
3317 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3318 | src_reg->type == PTR_TO_PACKET_META)) { | |
3319 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
3320 | find_good_pkt_pointers(this_branch, src_reg, | |
3321 | src_reg->type, true); | |
3322 | } else { | |
3323 | return false; | |
3324 | } | |
3325 | break; | |
3326 | default: | |
3327 | return false; | |
3328 | } | |
3329 | ||
3330 | return true; | |
3331 | } | |
3332 | ||
58e2af8b | 3333 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
3334 | struct bpf_insn *insn, int *insn_idx) |
3335 | { | |
638f5b90 | 3336 | struct bpf_verifier_state *other_branch, *this_branch = env->cur_state; |
58e2af8b | 3337 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; |
17a52670 AS |
3338 | u8 opcode = BPF_OP(insn->code); |
3339 | int err; | |
3340 | ||
b4e432f1 | 3341 | if (opcode > BPF_JSLE) { |
61bd5218 | 3342 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
3343 | return -EINVAL; |
3344 | } | |
3345 | ||
3346 | if (BPF_SRC(insn->code) == BPF_X) { | |
3347 | if (insn->imm != 0) { | |
61bd5218 | 3348 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3349 | return -EINVAL; |
3350 | } | |
3351 | ||
3352 | /* check src1 operand */ | |
dc503a8a | 3353 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3354 | if (err) |
3355 | return err; | |
1be7f75d AS |
3356 | |
3357 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 3358 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
3359 | insn->src_reg); |
3360 | return -EACCES; | |
3361 | } | |
17a52670 AS |
3362 | } else { |
3363 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 3364 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3365 | return -EINVAL; |
3366 | } | |
3367 | } | |
3368 | ||
3369 | /* check src2 operand */ | |
dc503a8a | 3370 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3371 | if (err) |
3372 | return err; | |
3373 | ||
1a0dc1ac AS |
3374 | dst_reg = ®s[insn->dst_reg]; |
3375 | ||
c779b6c6 AS |
3376 | if (BPF_SRC(insn->code) == BPF_K) { |
3377 | int pred = is_branch_taken(dst_reg, insn->imm, opcode); | |
3378 | ||
3379 | if (pred == 1) { | |
3380 | /* only follow the goto, ignore fall-through */ | |
17a52670 AS |
3381 | *insn_idx += insn->off; |
3382 | return 0; | |
c779b6c6 AS |
3383 | } else if (pred == 0) { |
3384 | /* only follow fall-through branch, since | |
17a52670 AS |
3385 | * that's where the program will go |
3386 | */ | |
3387 | return 0; | |
3388 | } | |
3389 | } | |
3390 | ||
010150ff DB |
3391 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
3392 | false); | |
17a52670 AS |
3393 | if (!other_branch) |
3394 | return -EFAULT; | |
3395 | ||
48461135 JB |
3396 | /* detect if we are comparing against a constant value so we can adjust |
3397 | * our min/max values for our dst register. | |
f1174f77 EC |
3398 | * this is only legit if both are scalars (or pointers to the same |
3399 | * object, I suppose, but we don't support that right now), because | |
3400 | * otherwise the different base pointers mean the offsets aren't | |
3401 | * comparable. | |
48461135 JB |
3402 | */ |
3403 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
3404 | if (dst_reg->type == SCALAR_VALUE && |
3405 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
3406 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
3407 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
3408 | dst_reg, regs[insn->src_reg].var_off.value, | |
3409 | opcode); | |
3410 | else if (tnum_is_const(dst_reg->var_off)) | |
3411 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
3412 | ®s[insn->src_reg], | |
3413 | dst_reg->var_off.value, opcode); | |
3414 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
3415 | /* Comparing for equality, we can combine knowledge */ | |
3416 | reg_combine_min_max(&other_branch->regs[insn->src_reg], | |
3417 | &other_branch->regs[insn->dst_reg], | |
3418 | ®s[insn->src_reg], | |
3419 | ®s[insn->dst_reg], opcode); | |
3420 | } | |
3421 | } else if (dst_reg->type == SCALAR_VALUE) { | |
48461135 JB |
3422 | reg_set_min_max(&other_branch->regs[insn->dst_reg], |
3423 | dst_reg, insn->imm, opcode); | |
3424 | } | |
3425 | ||
58e2af8b | 3426 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 3427 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
3428 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
3429 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
3430 | /* Mark all identical map registers in each branch as either |
3431 | * safe or unknown depending R == 0 or R != 0 conditional. | |
3432 | */ | |
f1174f77 EC |
3433 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
3434 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
5beca081 DB |
3435 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
3436 | this_branch, other_branch) && | |
3437 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
3438 | verbose(env, "R%d pointer comparison prohibited\n", |
3439 | insn->dst_reg); | |
1be7f75d | 3440 | return -EACCES; |
17a52670 | 3441 | } |
61bd5218 JK |
3442 | if (env->log.level) |
3443 | print_verifier_state(env, this_branch); | |
17a52670 AS |
3444 | return 0; |
3445 | } | |
3446 | ||
0246e64d AS |
3447 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
3448 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
3449 | { | |
3450 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
3451 | ||
3452 | return (struct bpf_map *) (unsigned long) imm64; | |
3453 | } | |
3454 | ||
17a52670 | 3455 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 3456 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3457 | { |
638f5b90 | 3458 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3459 | int err; |
3460 | ||
3461 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 3462 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
3463 | return -EINVAL; |
3464 | } | |
3465 | if (insn->off != 0) { | |
61bd5218 | 3466 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
3467 | return -EINVAL; |
3468 | } | |
3469 | ||
dc503a8a | 3470 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3471 | if (err) |
3472 | return err; | |
3473 | ||
6b173873 | 3474 | if (insn->src_reg == 0) { |
6b173873 JK |
3475 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
3476 | ||
f1174f77 | 3477 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 3478 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 3479 | return 0; |
6b173873 | 3480 | } |
17a52670 AS |
3481 | |
3482 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
3483 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
3484 | ||
3485 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
3486 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
3487 | return 0; | |
3488 | } | |
3489 | ||
96be4325 DB |
3490 | static bool may_access_skb(enum bpf_prog_type type) |
3491 | { | |
3492 | switch (type) { | |
3493 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
3494 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 3495 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
3496 | return true; |
3497 | default: | |
3498 | return false; | |
3499 | } | |
3500 | } | |
3501 | ||
ddd872bc AS |
3502 | /* verify safety of LD_ABS|LD_IND instructions: |
3503 | * - they can only appear in the programs where ctx == skb | |
3504 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
3505 | * preserve R6-R9, and store return value into R0 | |
3506 | * | |
3507 | * Implicit input: | |
3508 | * ctx == skb == R6 == CTX | |
3509 | * | |
3510 | * Explicit input: | |
3511 | * SRC == any register | |
3512 | * IMM == 32-bit immediate | |
3513 | * | |
3514 | * Output: | |
3515 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
3516 | */ | |
58e2af8b | 3517 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 3518 | { |
638f5b90 | 3519 | struct bpf_reg_state *regs = cur_regs(env); |
ff467ce7 | 3520 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 3521 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
3522 | int i, err; |
3523 | ||
24701ece | 3524 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 3525 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
3526 | return -EINVAL; |
3527 | } | |
3528 | ||
3529 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 3530 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 3531 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 3532 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
3533 | return -EINVAL; |
3534 | } | |
3535 | ||
3536 | /* check whether implicit source operand (register R6) is readable */ | |
ff467ce7 | 3537 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
3538 | if (err) |
3539 | return err; | |
3540 | ||
ff467ce7 | 3541 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
3542 | verbose(env, |
3543 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
3544 | return -EINVAL; |
3545 | } | |
3546 | ||
3547 | if (mode == BPF_IND) { | |
3548 | /* check explicit source operand */ | |
dc503a8a | 3549 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
3550 | if (err) |
3551 | return err; | |
3552 | } | |
3553 | ||
ff467ce7 DB |
3554 | err = check_ctx_reg(env, ®s[ctx_reg], ctx_reg); |
3555 | if (err < 0) | |
3556 | return err; | |
3557 | ||
ddd872bc | 3558 | /* reset caller saved regs to unreadable */ |
dc503a8a | 3559 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3560 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3561 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3562 | } | |
ddd872bc AS |
3563 | |
3564 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
3565 | * the value fetched from the packet. |
3566 | * Already marked as written above. | |
ddd872bc | 3567 | */ |
61bd5218 | 3568 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
3569 | return 0; |
3570 | } | |
3571 | ||
390ee7e2 AS |
3572 | static int check_return_code(struct bpf_verifier_env *env) |
3573 | { | |
3574 | struct bpf_reg_state *reg; | |
3575 | struct tnum range = tnum_range(0, 1); | |
3576 | ||
3577 | switch (env->prog->type) { | |
3578 | case BPF_PROG_TYPE_CGROUP_SKB: | |
3579 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
3580 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 3581 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
390ee7e2 AS |
3582 | break; |
3583 | default: | |
3584 | return 0; | |
3585 | } | |
3586 | ||
638f5b90 | 3587 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 3588 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 3589 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
3590 | reg_type_str[reg->type]); |
3591 | return -EINVAL; | |
3592 | } | |
3593 | ||
3594 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 3595 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
3596 | if (!tnum_is_unknown(reg->var_off)) { |
3597 | char tn_buf[48]; | |
3598 | ||
3599 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3600 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 3601 | } else { |
61bd5218 | 3602 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 3603 | } |
61bd5218 | 3604 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
3605 | return -EINVAL; |
3606 | } | |
3607 | return 0; | |
3608 | } | |
3609 | ||
475fb78f AS |
3610 | /* non-recursive DFS pseudo code |
3611 | * 1 procedure DFS-iterative(G,v): | |
3612 | * 2 label v as discovered | |
3613 | * 3 let S be a stack | |
3614 | * 4 S.push(v) | |
3615 | * 5 while S is not empty | |
3616 | * 6 t <- S.pop() | |
3617 | * 7 if t is what we're looking for: | |
3618 | * 8 return t | |
3619 | * 9 for all edges e in G.adjacentEdges(t) do | |
3620 | * 10 if edge e is already labelled | |
3621 | * 11 continue with the next edge | |
3622 | * 12 w <- G.adjacentVertex(t,e) | |
3623 | * 13 if vertex w is not discovered and not explored | |
3624 | * 14 label e as tree-edge | |
3625 | * 15 label w as discovered | |
3626 | * 16 S.push(w) | |
3627 | * 17 continue at 5 | |
3628 | * 18 else if vertex w is discovered | |
3629 | * 19 label e as back-edge | |
3630 | * 20 else | |
3631 | * 21 // vertex w is explored | |
3632 | * 22 label e as forward- or cross-edge | |
3633 | * 23 label t as explored | |
3634 | * 24 S.pop() | |
3635 | * | |
3636 | * convention: | |
3637 | * 0x10 - discovered | |
3638 | * 0x11 - discovered and fall-through edge labelled | |
3639 | * 0x12 - discovered and fall-through and branch edges labelled | |
3640 | * 0x20 - explored | |
3641 | */ | |
3642 | ||
3643 | enum { | |
3644 | DISCOVERED = 0x10, | |
3645 | EXPLORED = 0x20, | |
3646 | FALLTHROUGH = 1, | |
3647 | BRANCH = 2, | |
3648 | }; | |
3649 | ||
58e2af8b | 3650 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 3651 | |
475fb78f AS |
3652 | static int *insn_stack; /* stack of insns to process */ |
3653 | static int cur_stack; /* current stack index */ | |
3654 | static int *insn_state; | |
3655 | ||
3656 | /* t, w, e - match pseudo-code above: | |
3657 | * t - index of current instruction | |
3658 | * w - next instruction | |
3659 | * e - edge | |
3660 | */ | |
58e2af8b | 3661 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
3662 | { |
3663 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
3664 | return 0; | |
3665 | ||
3666 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
3667 | return 0; | |
3668 | ||
3669 | if (w < 0 || w >= env->prog->len) { | |
61bd5218 | 3670 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
3671 | return -EINVAL; |
3672 | } | |
3673 | ||
f1bca824 AS |
3674 | if (e == BRANCH) |
3675 | /* mark branch target for state pruning */ | |
3676 | env->explored_states[w] = STATE_LIST_MARK; | |
3677 | ||
475fb78f AS |
3678 | if (insn_state[w] == 0) { |
3679 | /* tree-edge */ | |
3680 | insn_state[t] = DISCOVERED | e; | |
3681 | insn_state[w] = DISCOVERED; | |
3682 | if (cur_stack >= env->prog->len) | |
3683 | return -E2BIG; | |
3684 | insn_stack[cur_stack++] = w; | |
3685 | return 1; | |
3686 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
61bd5218 | 3687 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
3688 | return -EINVAL; |
3689 | } else if (insn_state[w] == EXPLORED) { | |
3690 | /* forward- or cross-edge */ | |
3691 | insn_state[t] = DISCOVERED | e; | |
3692 | } else { | |
61bd5218 | 3693 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
3694 | return -EFAULT; |
3695 | } | |
3696 | return 0; | |
3697 | } | |
3698 | ||
3699 | /* non-recursive depth-first-search to detect loops in BPF program | |
3700 | * loop == back-edge in directed graph | |
3701 | */ | |
58e2af8b | 3702 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
3703 | { |
3704 | struct bpf_insn *insns = env->prog->insnsi; | |
3705 | int insn_cnt = env->prog->len; | |
3706 | int ret = 0; | |
3707 | int i, t; | |
3708 | ||
3709 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3710 | if (!insn_state) | |
3711 | return -ENOMEM; | |
3712 | ||
3713 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3714 | if (!insn_stack) { | |
3715 | kfree(insn_state); | |
3716 | return -ENOMEM; | |
3717 | } | |
3718 | ||
3719 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
3720 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
3721 | cur_stack = 1; | |
3722 | ||
3723 | peek_stack: | |
3724 | if (cur_stack == 0) | |
3725 | goto check_state; | |
3726 | t = insn_stack[cur_stack - 1]; | |
3727 | ||
3728 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
3729 | u8 opcode = BPF_OP(insns[t].code); | |
3730 | ||
3731 | if (opcode == BPF_EXIT) { | |
3732 | goto mark_explored; | |
3733 | } else if (opcode == BPF_CALL) { | |
3734 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3735 | if (ret == 1) | |
3736 | goto peek_stack; | |
3737 | else if (ret < 0) | |
3738 | goto err_free; | |
07016151 DB |
3739 | if (t + 1 < insn_cnt) |
3740 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3741 | } else if (opcode == BPF_JA) { |
3742 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
3743 | ret = -EINVAL; | |
3744 | goto err_free; | |
3745 | } | |
3746 | /* unconditional jump with single edge */ | |
3747 | ret = push_insn(t, t + insns[t].off + 1, | |
3748 | FALLTHROUGH, env); | |
3749 | if (ret == 1) | |
3750 | goto peek_stack; | |
3751 | else if (ret < 0) | |
3752 | goto err_free; | |
f1bca824 AS |
3753 | /* tell verifier to check for equivalent states |
3754 | * after every call and jump | |
3755 | */ | |
c3de6317 AS |
3756 | if (t + 1 < insn_cnt) |
3757 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3758 | } else { |
3759 | /* conditional jump with two edges */ | |
3c2ce60b | 3760 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
3761 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
3762 | if (ret == 1) | |
3763 | goto peek_stack; | |
3764 | else if (ret < 0) | |
3765 | goto err_free; | |
3766 | ||
3767 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
3768 | if (ret == 1) | |
3769 | goto peek_stack; | |
3770 | else if (ret < 0) | |
3771 | goto err_free; | |
3772 | } | |
3773 | } else { | |
3774 | /* all other non-branch instructions with single | |
3775 | * fall-through edge | |
3776 | */ | |
3777 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3778 | if (ret == 1) | |
3779 | goto peek_stack; | |
3780 | else if (ret < 0) | |
3781 | goto err_free; | |
3782 | } | |
3783 | ||
3784 | mark_explored: | |
3785 | insn_state[t] = EXPLORED; | |
3786 | if (cur_stack-- <= 0) { | |
61bd5218 | 3787 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
3788 | ret = -EFAULT; |
3789 | goto err_free; | |
3790 | } | |
3791 | goto peek_stack; | |
3792 | ||
3793 | check_state: | |
3794 | for (i = 0; i < insn_cnt; i++) { | |
3795 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 3796 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
3797 | ret = -EINVAL; |
3798 | goto err_free; | |
3799 | } | |
3800 | } | |
3801 | ret = 0; /* cfg looks good */ | |
3802 | ||
3803 | err_free: | |
3804 | kfree(insn_state); | |
3805 | kfree(insn_stack); | |
3806 | return ret; | |
3807 | } | |
3808 | ||
f1174f77 EC |
3809 | /* check %cur's range satisfies %old's */ |
3810 | static bool range_within(struct bpf_reg_state *old, | |
3811 | struct bpf_reg_state *cur) | |
3812 | { | |
b03c9f9f EC |
3813 | return old->umin_value <= cur->umin_value && |
3814 | old->umax_value >= cur->umax_value && | |
3815 | old->smin_value <= cur->smin_value && | |
3816 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
3817 | } |
3818 | ||
3819 | /* Maximum number of register states that can exist at once */ | |
3820 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
3821 | struct idpair { | |
3822 | u32 old; | |
3823 | u32 cur; | |
3824 | }; | |
3825 | ||
3826 | /* If in the old state two registers had the same id, then they need to have | |
3827 | * the same id in the new state as well. But that id could be different from | |
3828 | * the old state, so we need to track the mapping from old to new ids. | |
3829 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
3830 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
3831 | * regs with a different old id could still have new id 9, we don't care about | |
3832 | * that. | |
3833 | * So we look through our idmap to see if this old id has been seen before. If | |
3834 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 3835 | */ |
f1174f77 | 3836 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 3837 | { |
f1174f77 | 3838 | unsigned int i; |
969bf05e | 3839 | |
f1174f77 EC |
3840 | for (i = 0; i < ID_MAP_SIZE; i++) { |
3841 | if (!idmap[i].old) { | |
3842 | /* Reached an empty slot; haven't seen this id before */ | |
3843 | idmap[i].old = old_id; | |
3844 | idmap[i].cur = cur_id; | |
3845 | return true; | |
3846 | } | |
3847 | if (idmap[i].old == old_id) | |
3848 | return idmap[i].cur == cur_id; | |
3849 | } | |
3850 | /* We ran out of idmap slots, which should be impossible */ | |
3851 | WARN_ON_ONCE(1); | |
3852 | return false; | |
3853 | } | |
3854 | ||
3855 | /* Returns true if (rold safe implies rcur safe) */ | |
1b688a19 EC |
3856 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
3857 | struct idpair *idmap) | |
f1174f77 | 3858 | { |
dc503a8a EC |
3859 | if (!(rold->live & REG_LIVE_READ)) |
3860 | /* explored state didn't use this */ | |
3861 | return true; | |
3862 | ||
3863 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, live)) == 0) | |
969bf05e AS |
3864 | return true; |
3865 | ||
f1174f77 EC |
3866 | if (rold->type == NOT_INIT) |
3867 | /* explored state can't have used this */ | |
969bf05e | 3868 | return true; |
f1174f77 EC |
3869 | if (rcur->type == NOT_INIT) |
3870 | return false; | |
3871 | switch (rold->type) { | |
3872 | case SCALAR_VALUE: | |
3873 | if (rcur->type == SCALAR_VALUE) { | |
3874 | /* new val must satisfy old val knowledge */ | |
3875 | return range_within(rold, rcur) && | |
3876 | tnum_in(rold->var_off, rcur->var_off); | |
3877 | } else { | |
179d1c56 JH |
3878 | /* We're trying to use a pointer in place of a scalar. |
3879 | * Even if the scalar was unbounded, this could lead to | |
3880 | * pointer leaks because scalars are allowed to leak | |
3881 | * while pointers are not. We could make this safe in | |
3882 | * special cases if root is calling us, but it's | |
3883 | * probably not worth the hassle. | |
f1174f77 | 3884 | */ |
179d1c56 | 3885 | return false; |
f1174f77 EC |
3886 | } |
3887 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
3888 | /* If the new min/max/var_off satisfy the old ones and |
3889 | * everything else matches, we are OK. | |
3890 | * We don't care about the 'id' value, because nothing | |
3891 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
3892 | */ | |
3893 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
3894 | range_within(rold, rcur) && | |
3895 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
3896 | case PTR_TO_MAP_VALUE_OR_NULL: |
3897 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
3898 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
3899 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
3900 | * checked, doing so could have affected others with the same | |
3901 | * id, and we can't check for that because we lost the id when | |
3902 | * we converted to a PTR_TO_MAP_VALUE. | |
3903 | */ | |
3904 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
3905 | return false; | |
3906 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
3907 | return false; | |
3908 | /* Check our ids match any regs they're supposed to */ | |
3909 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 3910 | case PTR_TO_PACKET_META: |
f1174f77 | 3911 | case PTR_TO_PACKET: |
de8f3a83 | 3912 | if (rcur->type != rold->type) |
f1174f77 EC |
3913 | return false; |
3914 | /* We must have at least as much range as the old ptr | |
3915 | * did, so that any accesses which were safe before are | |
3916 | * still safe. This is true even if old range < old off, | |
3917 | * since someone could have accessed through (ptr - k), or | |
3918 | * even done ptr -= k in a register, to get a safe access. | |
3919 | */ | |
3920 | if (rold->range > rcur->range) | |
3921 | return false; | |
3922 | /* If the offsets don't match, we can't trust our alignment; | |
3923 | * nor can we be sure that we won't fall out of range. | |
3924 | */ | |
3925 | if (rold->off != rcur->off) | |
3926 | return false; | |
3927 | /* id relations must be preserved */ | |
3928 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
3929 | return false; | |
3930 | /* new val must satisfy old val knowledge */ | |
3931 | return range_within(rold, rcur) && | |
3932 | tnum_in(rold->var_off, rcur->var_off); | |
3933 | case PTR_TO_CTX: | |
3934 | case CONST_PTR_TO_MAP: | |
3935 | case PTR_TO_STACK: | |
3936 | case PTR_TO_PACKET_END: | |
3937 | /* Only valid matches are exact, which memcmp() above | |
3938 | * would have accepted | |
3939 | */ | |
3940 | default: | |
3941 | /* Don't know what's going on, just say it's not safe */ | |
3942 | return false; | |
3943 | } | |
969bf05e | 3944 | |
f1174f77 EC |
3945 | /* Shouldn't get here; if we do, say it's not safe */ |
3946 | WARN_ON_ONCE(1); | |
969bf05e AS |
3947 | return false; |
3948 | } | |
3949 | ||
638f5b90 AS |
3950 | static bool stacksafe(struct bpf_verifier_state *old, |
3951 | struct bpf_verifier_state *cur, | |
3952 | struct idpair *idmap) | |
3953 | { | |
3954 | int i, spi; | |
3955 | ||
3956 | /* if explored stack has more populated slots than current stack | |
3957 | * such stacks are not equivalent | |
3958 | */ | |
3959 | if (old->allocated_stack > cur->allocated_stack) | |
3960 | return false; | |
3961 | ||
3962 | /* walk slots of the explored stack and ignore any additional | |
3963 | * slots in the current stack, since explored(safe) state | |
3964 | * didn't use them | |
3965 | */ | |
3966 | for (i = 0; i < old->allocated_stack; i++) { | |
3967 | spi = i / BPF_REG_SIZE; | |
3968 | ||
3969 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) | |
3970 | continue; | |
3971 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != | |
3972 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
3973 | /* Ex: old explored (safe) state has STACK_SPILL in | |
3974 | * this stack slot, but current has has STACK_MISC -> | |
3975 | * this verifier states are not equivalent, | |
3976 | * return false to continue verification of this path | |
3977 | */ | |
3978 | return false; | |
3979 | if (i % BPF_REG_SIZE) | |
3980 | continue; | |
3981 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
3982 | continue; | |
3983 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
3984 | &cur->stack[spi].spilled_ptr, | |
3985 | idmap)) | |
3986 | /* when explored and current stack slot are both storing | |
3987 | * spilled registers, check that stored pointers types | |
3988 | * are the same as well. | |
3989 | * Ex: explored safe path could have stored | |
3990 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
3991 | * but current path has stored: | |
3992 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
3993 | * such verifier states are not equivalent. | |
3994 | * return false to continue verification of this path | |
3995 | */ | |
3996 | return false; | |
3997 | } | |
3998 | return true; | |
3999 | } | |
4000 | ||
f1bca824 AS |
4001 | /* compare two verifier states |
4002 | * | |
4003 | * all states stored in state_list are known to be valid, since | |
4004 | * verifier reached 'bpf_exit' instruction through them | |
4005 | * | |
4006 | * this function is called when verifier exploring different branches of | |
4007 | * execution popped from the state stack. If it sees an old state that has | |
4008 | * more strict register state and more strict stack state then this execution | |
4009 | * branch doesn't need to be explored further, since verifier already | |
4010 | * concluded that more strict state leads to valid finish. | |
4011 | * | |
4012 | * Therefore two states are equivalent if register state is more conservative | |
4013 | * and explored stack state is more conservative than the current one. | |
4014 | * Example: | |
4015 | * explored current | |
4016 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
4017 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
4018 | * | |
4019 | * In other words if current stack state (one being explored) has more | |
4020 | * valid slots than old one that already passed validation, it means | |
4021 | * the verifier can stop exploring and conclude that current state is valid too | |
4022 | * | |
4023 | * Similarly with registers. If explored state has register type as invalid | |
4024 | * whereas register type in current state is meaningful, it means that | |
4025 | * the current state will reach 'bpf_exit' instruction safely | |
4026 | */ | |
48461135 JB |
4027 | static bool states_equal(struct bpf_verifier_env *env, |
4028 | struct bpf_verifier_state *old, | |
58e2af8b | 4029 | struct bpf_verifier_state *cur) |
f1bca824 | 4030 | { |
f1174f77 EC |
4031 | struct idpair *idmap; |
4032 | bool ret = false; | |
f1bca824 AS |
4033 | int i; |
4034 | ||
010150ff DB |
4035 | /* Verification state from speculative execution simulation |
4036 | * must never prune a non-speculative execution one. | |
4037 | */ | |
4038 | if (old->speculative && !cur->speculative) | |
4039 | return false; | |
4040 | ||
f1174f77 EC |
4041 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
4042 | /* If we failed to allocate the idmap, just say it's not safe */ | |
4043 | if (!idmap) | |
1a0dc1ac | 4044 | return false; |
f1174f77 EC |
4045 | |
4046 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 4047 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 4048 | goto out_free; |
f1bca824 AS |
4049 | } |
4050 | ||
638f5b90 AS |
4051 | if (!stacksafe(old, cur, idmap)) |
4052 | goto out_free; | |
f1174f77 EC |
4053 | ret = true; |
4054 | out_free: | |
4055 | kfree(idmap); | |
4056 | return ret; | |
f1bca824 AS |
4057 | } |
4058 | ||
8e9cd9ce EC |
4059 | /* A write screens off any subsequent reads; but write marks come from the |
4060 | * straight-line code between a state and its parent. When we arrive at a | |
4061 | * jump target (in the first iteration of the propagate_liveness() loop), | |
4062 | * we didn't arrive by the straight-line code, so read marks in state must | |
4063 | * propagate to parent regardless of state's write marks. | |
4064 | */ | |
dc503a8a EC |
4065 | static bool do_propagate_liveness(const struct bpf_verifier_state *state, |
4066 | struct bpf_verifier_state *parent) | |
4067 | { | |
63f45f84 | 4068 | bool writes = parent == state->parent; /* Observe write marks */ |
dc503a8a EC |
4069 | bool touched = false; /* any changes made? */ |
4070 | int i; | |
4071 | ||
4072 | if (!parent) | |
4073 | return touched; | |
4074 | /* Propagate read liveness of registers... */ | |
4075 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
4076 | /* We don't need to worry about FP liveness because it's read-only */ | |
4077 | for (i = 0; i < BPF_REG_FP; i++) { | |
4078 | if (parent->regs[i].live & REG_LIVE_READ) | |
4079 | continue; | |
63f45f84 EC |
4080 | if (writes && (state->regs[i].live & REG_LIVE_WRITTEN)) |
4081 | continue; | |
4082 | if (state->regs[i].live & REG_LIVE_READ) { | |
dc503a8a EC |
4083 | parent->regs[i].live |= REG_LIVE_READ; |
4084 | touched = true; | |
4085 | } | |
4086 | } | |
4087 | /* ... and stack slots */ | |
638f5b90 AS |
4088 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
4089 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
4090 | if (parent->stack[i].slot_type[0] != STACK_SPILL) | |
dc503a8a | 4091 | continue; |
638f5b90 | 4092 | if (state->stack[i].slot_type[0] != STACK_SPILL) |
dc503a8a | 4093 | continue; |
638f5b90 | 4094 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
dc503a8a | 4095 | continue; |
638f5b90 AS |
4096 | if (writes && |
4097 | (state->stack[i].spilled_ptr.live & REG_LIVE_WRITTEN)) | |
63f45f84 | 4098 | continue; |
638f5b90 AS |
4099 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) { |
4100 | parent->stack[i].spilled_ptr.live |= REG_LIVE_READ; | |
dc503a8a EC |
4101 | touched = true; |
4102 | } | |
4103 | } | |
4104 | return touched; | |
4105 | } | |
4106 | ||
8e9cd9ce EC |
4107 | /* "parent" is "a state from which we reach the current state", but initially |
4108 | * it is not the state->parent (i.e. "the state whose straight-line code leads | |
4109 | * to the current state"), instead it is the state that happened to arrive at | |
4110 | * a (prunable) equivalent of the current state. See comment above | |
4111 | * do_propagate_liveness() for consequences of this. | |
4112 | * This function is just a more efficient way of calling mark_reg_read() or | |
4113 | * mark_stack_slot_read() on each reg in "parent" that is read in "state", | |
4114 | * though it requires that parent != state->parent in the call arguments. | |
4115 | */ | |
dc503a8a EC |
4116 | static void propagate_liveness(const struct bpf_verifier_state *state, |
4117 | struct bpf_verifier_state *parent) | |
4118 | { | |
4119 | while (do_propagate_liveness(state, parent)) { | |
4120 | /* Something changed, so we need to feed those changes onward */ | |
4121 | state = parent; | |
4122 | parent = state->parent; | |
4123 | } | |
4124 | } | |
4125 | ||
58e2af8b | 4126 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 4127 | { |
58e2af8b JK |
4128 | struct bpf_verifier_state_list *new_sl; |
4129 | struct bpf_verifier_state_list *sl; | |
638f5b90 | 4130 | struct bpf_verifier_state *cur = env->cur_state; |
adec32a5 | 4131 | int i, err, states_cnt = 0; |
f1bca824 AS |
4132 | |
4133 | sl = env->explored_states[insn_idx]; | |
4134 | if (!sl) | |
4135 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
4136 | * be doing state search here | |
4137 | */ | |
4138 | return 0; | |
4139 | ||
4140 | while (sl != STATE_LIST_MARK) { | |
638f5b90 | 4141 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 4142 | /* reached equivalent register/stack state, |
dc503a8a EC |
4143 | * prune the search. |
4144 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
4145 | * If we have any write marks in env->cur_state, they |
4146 | * will prevent corresponding reads in the continuation | |
4147 | * from reaching our parent (an explored_state). Our | |
4148 | * own state will get the read marks recorded, but | |
4149 | * they'll be immediately forgotten as we're pruning | |
4150 | * this state and will pop a new one. | |
f1bca824 | 4151 | */ |
638f5b90 | 4152 | propagate_liveness(&sl->state, cur); |
f1bca824 | 4153 | return 1; |
dc503a8a | 4154 | } |
f1bca824 | 4155 | sl = sl->next; |
adec32a5 | 4156 | states_cnt++; |
f1bca824 AS |
4157 | } |
4158 | ||
adec32a5 AS |
4159 | if (!env->allow_ptr_leaks && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
4160 | return 0; | |
4161 | ||
f1bca824 AS |
4162 | /* there were no equivalent states, remember current one. |
4163 | * technically the current state is not proven to be safe yet, | |
4164 | * but it will either reach bpf_exit (which means it's safe) or | |
4165 | * it will be rejected. Since there are no loops, we won't be | |
4166 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
4167 | */ | |
638f5b90 | 4168 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
4169 | if (!new_sl) |
4170 | return -ENOMEM; | |
4171 | ||
4172 | /* add new state to the head of linked list */ | |
1969db47 AS |
4173 | err = copy_verifier_state(&new_sl->state, cur); |
4174 | if (err) { | |
4175 | free_verifier_state(&new_sl->state, false); | |
4176 | kfree(new_sl); | |
4177 | return err; | |
4178 | } | |
f1bca824 AS |
4179 | new_sl->next = env->explored_states[insn_idx]; |
4180 | env->explored_states[insn_idx] = new_sl; | |
dc503a8a | 4181 | /* connect new state to parentage chain */ |
638f5b90 | 4182 | cur->parent = &new_sl->state; |
8e9cd9ce EC |
4183 | /* clear write marks in current state: the writes we did are not writes |
4184 | * our child did, so they don't screen off its reads from us. | |
4185 | * (There are no read marks in current state, because reads always mark | |
4186 | * their parent and current state never has children yet. Only | |
4187 | * explored_states can get read marks.) | |
4188 | */ | |
dc503a8a | 4189 | for (i = 0; i < BPF_REG_FP; i++) |
638f5b90 AS |
4190 | cur->regs[i].live = REG_LIVE_NONE; |
4191 | for (i = 0; i < cur->allocated_stack / BPF_REG_SIZE; i++) | |
4192 | if (cur->stack[i].slot_type[0] == STACK_SPILL) | |
4193 | cur->stack[i].spilled_ptr.live = REG_LIVE_NONE; | |
f1bca824 AS |
4194 | return 0; |
4195 | } | |
4196 | ||
13a27dfc JK |
4197 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
4198 | int insn_idx, int prev_insn_idx) | |
4199 | { | |
ab3f0063 JK |
4200 | if (env->dev_ops && env->dev_ops->insn_hook) |
4201 | return env->dev_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
13a27dfc | 4202 | |
ab3f0063 | 4203 | return 0; |
13a27dfc JK |
4204 | } |
4205 | ||
58e2af8b | 4206 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 4207 | { |
638f5b90 | 4208 | struct bpf_verifier_state *state; |
17a52670 | 4209 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 4210 | struct bpf_reg_state *regs; |
17a52670 | 4211 | int insn_cnt = env->prog->len; |
17a52670 AS |
4212 | int insn_processed = 0; |
4213 | bool do_print_state = false; | |
4214 | ||
638f5b90 AS |
4215 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
4216 | if (!state) | |
4217 | return -ENOMEM; | |
4218 | env->cur_state = state; | |
4219 | init_reg_state(env, state->regs); | |
dc503a8a | 4220 | state->parent = NULL; |
010150ff | 4221 | state->speculative = false; |
f93a623f | 4222 | |
17a52670 AS |
4223 | for (;;) { |
4224 | struct bpf_insn *insn; | |
4225 | u8 class; | |
4226 | int err; | |
4227 | ||
f93a623f | 4228 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 4229 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
f93a623f | 4230 | env->insn_idx, insn_cnt); |
17a52670 AS |
4231 | return -EFAULT; |
4232 | } | |
4233 | ||
f93a623f | 4234 | insn = &insns[env->insn_idx]; |
17a52670 AS |
4235 | class = BPF_CLASS(insn->code); |
4236 | ||
07016151 | 4237 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
4238 | verbose(env, |
4239 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
4240 | insn_processed); |
4241 | return -E2BIG; | |
4242 | } | |
4243 | ||
f93a623f | 4244 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
4245 | if (err < 0) |
4246 | return err; | |
4247 | if (err == 1) { | |
4248 | /* found equivalent state, can prune the search */ | |
61bd5218 | 4249 | if (env->log.level) { |
f1bca824 | 4250 | if (do_print_state) |
010150ff DB |
4251 | verbose(env, "\nfrom %d to %d%s: safe\n", |
4252 | env->prev_insn_idx, env->insn_idx, | |
4253 | env->cur_state->speculative ? | |
4254 | " (speculative execution)" : ""); | |
f1bca824 | 4255 | else |
f93a623f | 4256 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
4257 | } |
4258 | goto process_bpf_exit; | |
4259 | } | |
4260 | ||
1ec7c3bd AS |
4261 | if (signal_pending(current)) |
4262 | return -EAGAIN; | |
4263 | ||
3c2ce60b DB |
4264 | if (need_resched()) |
4265 | cond_resched(); | |
4266 | ||
61bd5218 JK |
4267 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
4268 | if (env->log.level > 1) | |
f93a623f | 4269 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 4270 | else |
010150ff DB |
4271 | verbose(env, "\nfrom %d to %d%s:", |
4272 | env->prev_insn_idx, env->insn_idx, | |
4273 | env->cur_state->speculative ? | |
4274 | " (speculative execution)" : ""); | |
638f5b90 | 4275 | print_verifier_state(env, state); |
17a52670 AS |
4276 | do_print_state = false; |
4277 | } | |
4278 | ||
61bd5218 | 4279 | if (env->log.level) { |
f93a623f | 4280 | verbose(env, "%d: ", env->insn_idx); |
f4ac7e0b JK |
4281 | print_bpf_insn(verbose, env, insn, |
4282 | env->allow_ptr_leaks); | |
17a52670 AS |
4283 | } |
4284 | ||
f93a623f | 4285 | err = ext_analyzer_insn_hook(env, env->insn_idx, env->prev_insn_idx); |
13a27dfc JK |
4286 | if (err) |
4287 | return err; | |
4288 | ||
638f5b90 | 4289 | regs = cur_regs(env); |
f93a623f DB |
4290 | env->insn_aux_data[env->insn_idx].seen = true; |
4291 | ||
17a52670 | 4292 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 4293 | err = check_alu_op(env, insn); |
17a52670 AS |
4294 | if (err) |
4295 | return err; | |
4296 | ||
4297 | } else if (class == BPF_LDX) { | |
3df126f3 | 4298 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
4299 | |
4300 | /* check for reserved fields is already done */ | |
4301 | ||
17a52670 | 4302 | /* check src operand */ |
dc503a8a | 4303 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4304 | if (err) |
4305 | return err; | |
4306 | ||
dc503a8a | 4307 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
4308 | if (err) |
4309 | return err; | |
4310 | ||
725f9dcd AS |
4311 | src_reg_type = regs[insn->src_reg].type; |
4312 | ||
17a52670 AS |
4313 | /* check that memory (src_reg + off) is readable, |
4314 | * the state of dst_reg will be updated by this func | |
4315 | */ | |
f93a623f DB |
4316 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
4317 | insn->off, BPF_SIZE(insn->code), | |
4318 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
4319 | if (err) |
4320 | return err; | |
4321 | ||
f93a623f | 4322 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
4323 | |
4324 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
4325 | /* saw a valid insn |
4326 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 4327 | * save type to validate intersecting paths |
9bac3d6d | 4328 | */ |
3df126f3 | 4329 | *prev_src_type = src_reg_type; |
9bac3d6d | 4330 | |
3df126f3 | 4331 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 4332 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 4333 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
4334 | /* ABuser program is trying to use the same insn |
4335 | * dst_reg = *(u32*) (src_reg + off) | |
4336 | * with different pointer types: | |
4337 | * src_reg == ctx in one branch and | |
4338 | * src_reg == stack|map in some other branch. | |
4339 | * Reject it. | |
4340 | */ | |
61bd5218 | 4341 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
4342 | return -EINVAL; |
4343 | } | |
4344 | ||
17a52670 | 4345 | } else if (class == BPF_STX) { |
3df126f3 | 4346 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 4347 | |
17a52670 | 4348 | if (BPF_MODE(insn->code) == BPF_XADD) { |
f93a623f | 4349 | err = check_xadd(env, env->insn_idx, insn); |
17a52670 AS |
4350 | if (err) |
4351 | return err; | |
f93a623f | 4352 | env->insn_idx++; |
17a52670 AS |
4353 | continue; |
4354 | } | |
4355 | ||
17a52670 | 4356 | /* check src1 operand */ |
dc503a8a | 4357 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4358 | if (err) |
4359 | return err; | |
4360 | /* check src2 operand */ | |
dc503a8a | 4361 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4362 | if (err) |
4363 | return err; | |
4364 | ||
d691f9e8 AS |
4365 | dst_reg_type = regs[insn->dst_reg].type; |
4366 | ||
17a52670 | 4367 | /* check that memory (dst_reg + off) is writeable */ |
f93a623f DB |
4368 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
4369 | insn->off, BPF_SIZE(insn->code), | |
4370 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
4371 | if (err) |
4372 | return err; | |
4373 | ||
f93a623f | 4374 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
4375 | |
4376 | if (*prev_dst_type == NOT_INIT) { | |
4377 | *prev_dst_type = dst_reg_type; | |
4378 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 4379 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 4380 | *prev_dst_type == PTR_TO_CTX)) { |
61bd5218 | 4381 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
4382 | return -EINVAL; |
4383 | } | |
4384 | ||
17a52670 AS |
4385 | } else if (class == BPF_ST) { |
4386 | if (BPF_MODE(insn->code) != BPF_MEM || | |
4387 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 4388 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
4389 | return -EINVAL; |
4390 | } | |
4391 | /* check src operand */ | |
dc503a8a | 4392 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4393 | if (err) |
4394 | return err; | |
4395 | ||
f37a8cb8 DB |
4396 | if (is_ctx_reg(env, insn->dst_reg)) { |
4397 | verbose(env, "BPF_ST stores into R%d context is not allowed\n", | |
4398 | insn->dst_reg); | |
4399 | return -EACCES; | |
4400 | } | |
4401 | ||
17a52670 | 4402 | /* check that memory (dst_reg + off) is writeable */ |
f93a623f DB |
4403 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
4404 | insn->off, BPF_SIZE(insn->code), | |
4405 | BPF_WRITE, -1, false); | |
17a52670 AS |
4406 | if (err) |
4407 | return err; | |
4408 | ||
4409 | } else if (class == BPF_JMP) { | |
4410 | u8 opcode = BPF_OP(insn->code); | |
4411 | ||
4412 | if (opcode == BPF_CALL) { | |
4413 | if (BPF_SRC(insn->code) != BPF_K || | |
4414 | insn->off != 0 || | |
4415 | insn->src_reg != BPF_REG_0 || | |
4416 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4417 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
4418 | return -EINVAL; |
4419 | } | |
4420 | ||
f93a623f | 4421 | err = check_call(env, insn->imm, env->insn_idx); |
17a52670 AS |
4422 | if (err) |
4423 | return err; | |
4424 | ||
4425 | } else if (opcode == BPF_JA) { | |
4426 | if (BPF_SRC(insn->code) != BPF_K || | |
4427 | insn->imm != 0 || | |
4428 | insn->src_reg != BPF_REG_0 || | |
4429 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4430 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
4431 | return -EINVAL; |
4432 | } | |
4433 | ||
f93a623f | 4434 | env->insn_idx += insn->off + 1; |
17a52670 AS |
4435 | continue; |
4436 | ||
4437 | } else if (opcode == BPF_EXIT) { | |
4438 | if (BPF_SRC(insn->code) != BPF_K || | |
4439 | insn->imm != 0 || | |
4440 | insn->src_reg != BPF_REG_0 || | |
4441 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4442 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
4443 | return -EINVAL; |
4444 | } | |
4445 | ||
4446 | /* eBPF calling convetion is such that R0 is used | |
4447 | * to return the value from eBPF program. | |
4448 | * Make sure that it's readable at this time | |
4449 | * of bpf_exit, which means that program wrote | |
4450 | * something into it earlier | |
4451 | */ | |
dc503a8a | 4452 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
4453 | if (err) |
4454 | return err; | |
4455 | ||
1be7f75d | 4456 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 4457 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
4458 | return -EACCES; |
4459 | } | |
4460 | ||
390ee7e2 AS |
4461 | err = check_return_code(env); |
4462 | if (err) | |
4463 | return err; | |
f1bca824 | 4464 | process_bpf_exit: |
f93a623f DB |
4465 | err = pop_stack(env, &env->prev_insn_idx, |
4466 | &env->insn_idx); | |
638f5b90 AS |
4467 | if (err < 0) { |
4468 | if (err != -ENOENT) | |
4469 | return err; | |
17a52670 AS |
4470 | break; |
4471 | } else { | |
4472 | do_print_state = true; | |
4473 | continue; | |
4474 | } | |
4475 | } else { | |
f93a623f | 4476 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
4477 | if (err) |
4478 | return err; | |
4479 | } | |
4480 | } else if (class == BPF_LD) { | |
4481 | u8 mode = BPF_MODE(insn->code); | |
4482 | ||
4483 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
4484 | err = check_ld_abs(env, insn); |
4485 | if (err) | |
4486 | return err; | |
4487 | ||
17a52670 AS |
4488 | } else if (mode == BPF_IMM) { |
4489 | err = check_ld_imm(env, insn); | |
4490 | if (err) | |
4491 | return err; | |
4492 | ||
f93a623f DB |
4493 | env->insn_idx++; |
4494 | env->insn_aux_data[env->insn_idx].seen = true; | |
17a52670 | 4495 | } else { |
61bd5218 | 4496 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
4497 | return -EINVAL; |
4498 | } | |
4499 | } else { | |
61bd5218 | 4500 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
4501 | return -EINVAL; |
4502 | } | |
4503 | ||
f93a623f | 4504 | env->insn_idx++; |
17a52670 AS |
4505 | } |
4506 | ||
61bd5218 JK |
4507 | verbose(env, "processed %d insns, stack depth %d\n", insn_processed, |
4508 | env->prog->aux->stack_depth); | |
17a52670 AS |
4509 | return 0; |
4510 | } | |
4511 | ||
56f668df MKL |
4512 | static int check_map_prealloc(struct bpf_map *map) |
4513 | { | |
4514 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
4515 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
4516 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
4517 | !(map->map_flags & BPF_F_NO_PREALLOC); |
4518 | } | |
4519 | ||
61bd5218 JK |
4520 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
4521 | struct bpf_map *map, | |
fdc15d38 AS |
4522 | struct bpf_prog *prog) |
4523 | ||
4524 | { | |
56f668df MKL |
4525 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
4526 | * preallocated hash maps, since doing memory allocation | |
4527 | * in overflow_handler can crash depending on where nmi got | |
4528 | * triggered. | |
4529 | */ | |
4530 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
4531 | if (!check_map_prealloc(map)) { | |
61bd5218 | 4532 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
4533 | return -EINVAL; |
4534 | } | |
4535 | if (map->inner_map_meta && | |
4536 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 4537 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
4538 | return -EINVAL; |
4539 | } | |
fdc15d38 AS |
4540 | } |
4541 | return 0; | |
4542 | } | |
4543 | ||
0246e64d AS |
4544 | /* look for pseudo eBPF instructions that access map FDs and |
4545 | * replace them with actual map pointers | |
4546 | */ | |
58e2af8b | 4547 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
4548 | { |
4549 | struct bpf_insn *insn = env->prog->insnsi; | |
4550 | int insn_cnt = env->prog->len; | |
fdc15d38 | 4551 | int i, j, err; |
0246e64d | 4552 | |
f1f7714e | 4553 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
4554 | if (err) |
4555 | return err; | |
4556 | ||
0246e64d | 4557 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 4558 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 4559 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 4560 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
4561 | return -EINVAL; |
4562 | } | |
4563 | ||
d691f9e8 AS |
4564 | if (BPF_CLASS(insn->code) == BPF_STX && |
4565 | ((BPF_MODE(insn->code) != BPF_MEM && | |
4566 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 4567 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
4568 | return -EINVAL; |
4569 | } | |
4570 | ||
0246e64d AS |
4571 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
4572 | struct bpf_map *map; | |
4573 | struct fd f; | |
4574 | ||
4575 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
4576 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
4577 | insn[1].off != 0) { | |
61bd5218 | 4578 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
4579 | return -EINVAL; |
4580 | } | |
4581 | ||
4582 | if (insn->src_reg == 0) | |
4583 | /* valid generic load 64-bit imm */ | |
4584 | goto next_insn; | |
4585 | ||
4586 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
4587 | verbose(env, |
4588 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
4589 | return -EINVAL; |
4590 | } | |
4591 | ||
4592 | f = fdget(insn->imm); | |
c2101297 | 4593 | map = __bpf_map_get(f); |
0246e64d | 4594 | if (IS_ERR(map)) { |
61bd5218 | 4595 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 4596 | insn->imm); |
0246e64d AS |
4597 | return PTR_ERR(map); |
4598 | } | |
4599 | ||
61bd5218 | 4600 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
4601 | if (err) { |
4602 | fdput(f); | |
4603 | return err; | |
4604 | } | |
4605 | ||
0246e64d AS |
4606 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
4607 | insn[0].imm = (u32) (unsigned long) map; | |
4608 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
4609 | ||
4610 | /* check whether we recorded this map already */ | |
4611 | for (j = 0; j < env->used_map_cnt; j++) | |
4612 | if (env->used_maps[j] == map) { | |
4613 | fdput(f); | |
4614 | goto next_insn; | |
4615 | } | |
4616 | ||
4617 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
4618 | fdput(f); | |
4619 | return -E2BIG; | |
4620 | } | |
4621 | ||
0246e64d AS |
4622 | /* hold the map. If the program is rejected by verifier, |
4623 | * the map will be released by release_maps() or it | |
4624 | * will be used by the valid program until it's unloaded | |
d6f1a482 | 4625 | * and all maps are released in free_used_maps() |
0246e64d | 4626 | */ |
92117d84 AS |
4627 | map = bpf_map_inc(map, false); |
4628 | if (IS_ERR(map)) { | |
4629 | fdput(f); | |
4630 | return PTR_ERR(map); | |
4631 | } | |
4632 | env->used_maps[env->used_map_cnt++] = map; | |
4633 | ||
0246e64d AS |
4634 | fdput(f); |
4635 | next_insn: | |
4636 | insn++; | |
4637 | i++; | |
4638 | } | |
4639 | } | |
4640 | ||
4641 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
4642 | * 'struct bpf_map *' into a register instead of user map_fd. | |
4643 | * These pointers will be used later by verifier to validate map access. | |
4644 | */ | |
4645 | return 0; | |
4646 | } | |
4647 | ||
4648 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 4649 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
4650 | { |
4651 | int i; | |
4652 | ||
4653 | for (i = 0; i < env->used_map_cnt; i++) | |
4654 | bpf_map_put(env->used_maps[i]); | |
4655 | } | |
4656 | ||
4657 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 4658 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
4659 | { |
4660 | struct bpf_insn *insn = env->prog->insnsi; | |
4661 | int insn_cnt = env->prog->len; | |
4662 | int i; | |
4663 | ||
4664 | for (i = 0; i < insn_cnt; i++, insn++) | |
4665 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
4666 | insn->src_reg = 0; | |
4667 | } | |
4668 | ||
8041902d AS |
4669 | /* single env->prog->insni[off] instruction was replaced with the range |
4670 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
4671 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
4672 | */ | |
4673 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
4674 | u32 off, u32 cnt) | |
4675 | { | |
4676 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
c131187d | 4677 | int i; |
8041902d AS |
4678 | |
4679 | if (cnt == 1) | |
4680 | return 0; | |
4681 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
4682 | if (!new_data) | |
4683 | return -ENOMEM; | |
4684 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
4685 | memcpy(new_data + off + cnt - 1, old_data + off, | |
4686 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
c131187d AS |
4687 | for (i = off; i < off + cnt - 1; i++) |
4688 | new_data[i].seen = true; | |
8041902d AS |
4689 | env->insn_aux_data = new_data; |
4690 | vfree(old_data); | |
4691 | return 0; | |
4692 | } | |
4693 | ||
4694 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
4695 | const struct bpf_insn *patch, u32 len) | |
4696 | { | |
4697 | struct bpf_prog *new_prog; | |
4698 | ||
4699 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4700 | if (!new_prog) | |
4701 | return NULL; | |
4702 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
4703 | return NULL; | |
4704 | return new_prog; | |
4705 | } | |
4706 | ||
c131187d AS |
4707 | /* The verifier does more data flow analysis than llvm and will not explore |
4708 | * branches that are dead at run time. Malicious programs can have dead code | |
4709 | * too. Therefore replace all dead at-run-time code with nops. | |
4710 | */ | |
4711 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
4712 | { | |
4713 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
4714 | struct bpf_insn nop = BPF_MOV64_REG(BPF_REG_0, BPF_REG_0); | |
4715 | struct bpf_insn *insn = env->prog->insnsi; | |
4716 | const int insn_cnt = env->prog->len; | |
4717 | int i; | |
4718 | ||
4719 | for (i = 0; i < insn_cnt; i++) { | |
4720 | if (aux_data[i].seen) | |
4721 | continue; | |
4722 | memcpy(insn + i, &nop, sizeof(nop)); | |
4723 | } | |
4724 | } | |
4725 | ||
9bac3d6d AS |
4726 | /* convert load instructions that access fields of 'struct __sk_buff' |
4727 | * into sequence of instructions that access fields of 'struct sk_buff' | |
4728 | */ | |
58e2af8b | 4729 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 4730 | { |
00176a34 | 4731 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 4732 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 4733 | const int insn_cnt = env->prog->len; |
36bbef52 | 4734 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 4735 | struct bpf_prog *new_prog; |
d691f9e8 | 4736 | enum bpf_access_type type; |
f96da094 DB |
4737 | bool is_narrower_load; |
4738 | u32 target_size; | |
9bac3d6d | 4739 | |
36bbef52 DB |
4740 | if (ops->gen_prologue) { |
4741 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
4742 | env->prog); | |
4743 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4744 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
4745 | return -EINVAL; |
4746 | } else if (cnt) { | |
8041902d | 4747 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
4748 | if (!new_prog) |
4749 | return -ENOMEM; | |
8041902d | 4750 | |
36bbef52 | 4751 | env->prog = new_prog; |
3df126f3 | 4752 | delta += cnt - 1; |
36bbef52 DB |
4753 | } |
4754 | } | |
4755 | ||
4756 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
4757 | return 0; |
4758 | ||
3df126f3 | 4759 | insn = env->prog->insnsi + delta; |
36bbef52 | 4760 | |
9bac3d6d | 4761 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
4762 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
4763 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
4764 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4765 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 4766 | type = BPF_READ; |
62c7989b DB |
4767 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
4768 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
4769 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4770 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
4771 | type = BPF_WRITE; |
4772 | else | |
9bac3d6d AS |
4773 | continue; |
4774 | ||
abd098e0 AS |
4775 | if (type == BPF_WRITE && |
4776 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
4777 | struct bpf_insn patch[] = { | |
4778 | /* Sanitize suspicious stack slot with zero. | |
4779 | * There are no memory dependencies for this store, | |
4780 | * since it's only using frame pointer and immediate | |
4781 | * constant of zero | |
4782 | */ | |
4783 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
4784 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
4785 | 0), | |
4786 | /* the original STX instruction will immediately | |
4787 | * overwrite the same stack slot with appropriate value | |
4788 | */ | |
4789 | *insn, | |
4790 | }; | |
4791 | ||
4792 | cnt = ARRAY_SIZE(patch); | |
4793 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
4794 | if (!new_prog) | |
4795 | return -ENOMEM; | |
4796 | ||
4797 | delta += cnt - 1; | |
4798 | env->prog = new_prog; | |
4799 | insn = new_prog->insnsi + i + delta; | |
4800 | continue; | |
4801 | } | |
4802 | ||
8041902d | 4803 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 4804 | continue; |
9bac3d6d | 4805 | |
31fd8581 | 4806 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 4807 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
4808 | |
4809 | /* If the read access is a narrower load of the field, | |
4810 | * convert to a 4/8-byte load, to minimum program type specific | |
4811 | * convert_ctx_access changes. If conversion is successful, | |
4812 | * we will apply proper mask to the result. | |
4813 | */ | |
f96da094 | 4814 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 4815 | if (is_narrower_load) { |
f96da094 DB |
4816 | u32 off = insn->off; |
4817 | u8 size_code; | |
4818 | ||
4819 | if (type == BPF_WRITE) { | |
61bd5218 | 4820 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
4821 | return -EINVAL; |
4822 | } | |
31fd8581 | 4823 | |
f96da094 | 4824 | size_code = BPF_H; |
31fd8581 YS |
4825 | if (ctx_field_size == 4) |
4826 | size_code = BPF_W; | |
4827 | else if (ctx_field_size == 8) | |
4828 | size_code = BPF_DW; | |
f96da094 | 4829 | |
31fd8581 YS |
4830 | insn->off = off & ~(ctx_field_size - 1); |
4831 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
4832 | } | |
f96da094 DB |
4833 | |
4834 | target_size = 0; | |
4835 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
4836 | &target_size); | |
4837 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
4838 | (ctx_field_size && !target_size)) { | |
61bd5218 | 4839 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
4840 | return -EINVAL; |
4841 | } | |
f96da094 DB |
4842 | |
4843 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
4844 | if (ctx_field_size <= 4) |
4845 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4846 | (1 << size * 8) - 1); |
31fd8581 YS |
4847 | else |
4848 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
93539bbb | 4849 | (1ULL << size * 8) - 1); |
31fd8581 | 4850 | } |
9bac3d6d | 4851 | |
8041902d | 4852 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
4853 | if (!new_prog) |
4854 | return -ENOMEM; | |
4855 | ||
3df126f3 | 4856 | delta += cnt - 1; |
9bac3d6d AS |
4857 | |
4858 | /* keep walking new program and skip insns we just inserted */ | |
4859 | env->prog = new_prog; | |
3df126f3 | 4860 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
4861 | } |
4862 | ||
4863 | return 0; | |
4864 | } | |
4865 | ||
79741b3b | 4866 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 4867 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
4868 | * |
4869 | * this function is called after eBPF program passed verification | |
4870 | */ | |
79741b3b | 4871 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 4872 | { |
79741b3b AS |
4873 | struct bpf_prog *prog = env->prog; |
4874 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 4875 | const struct bpf_func_proto *fn; |
79741b3b | 4876 | const int insn_cnt = prog->len; |
3f62bcc9 | 4877 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
4878 | struct bpf_insn insn_buf[16]; |
4879 | struct bpf_prog *new_prog; | |
4880 | struct bpf_map *map_ptr; | |
4881 | int i, cnt, delta = 0; | |
e245c5c6 | 4882 | |
79741b3b | 4883 | for (i = 0; i < insn_cnt; i++, insn++) { |
68fda450 AS |
4884 | if (insn->code == (BPF_ALU | BPF_MOD | BPF_X) || |
4885 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
4886 | /* due to JIT bugs clear upper 32-bits of src register | |
4887 | * before div/mod operation | |
4888 | */ | |
4889 | insn_buf[0] = BPF_MOV32_REG(insn->src_reg, insn->src_reg); | |
4890 | insn_buf[1] = *insn; | |
4891 | cnt = 2; | |
4892 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
4893 | if (!new_prog) | |
4894 | return -ENOMEM; | |
4895 | ||
4896 | delta += cnt - 1; | |
4897 | env->prog = prog = new_prog; | |
4898 | insn = new_prog->insnsi + i + delta; | |
4899 | continue; | |
4900 | } | |
4901 | ||
010150ff DB |
4902 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
4903 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
4904 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
4905 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
4906 | struct bpf_insn insn_buf[16]; | |
4907 | struct bpf_insn *patch = &insn_buf[0]; | |
4908 | bool issrc, isneg; | |
4909 | u32 off_reg; | |
4910 | ||
4911 | aux = &env->insn_aux_data[i + delta]; | |
f0ee01bb DB |
4912 | if (!aux->alu_state || |
4913 | aux->alu_state == BPF_ALU_NON_POINTER) | |
010150ff DB |
4914 | continue; |
4915 | ||
4916 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
4917 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
4918 | BPF_ALU_SANITIZE_SRC; | |
4919 | ||
4920 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
4921 | if (isneg) | |
4922 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
4923 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit - 1); | |
4924 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
4925 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
4926 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
4927 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
4928 | if (issrc) { | |
4929 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, | |
4930 | off_reg); | |
4931 | insn->src_reg = BPF_REG_AX; | |
4932 | } else { | |
4933 | *patch++ = BPF_ALU64_REG(BPF_AND, off_reg, | |
4934 | BPF_REG_AX); | |
4935 | } | |
4936 | if (isneg) | |
4937 | insn->code = insn->code == code_add ? | |
4938 | code_sub : code_add; | |
4939 | *patch++ = *insn; | |
4940 | if (issrc && isneg) | |
4941 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
4942 | cnt = patch - insn_buf; | |
4943 | ||
4944 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
4945 | if (!new_prog) | |
4946 | return -ENOMEM; | |
4947 | ||
4948 | delta += cnt - 1; | |
4949 | env->prog = prog = new_prog; | |
4950 | insn = new_prog->insnsi + i + delta; | |
4951 | continue; | |
4952 | } | |
4953 | ||
79741b3b AS |
4954 | if (insn->code != (BPF_JMP | BPF_CALL)) |
4955 | continue; | |
e245c5c6 | 4956 | |
79741b3b AS |
4957 | if (insn->imm == BPF_FUNC_get_route_realm) |
4958 | prog->dst_needed = 1; | |
4959 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
4960 | bpf_user_rnd_init_once(); | |
79741b3b | 4961 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
4962 | /* If we tail call into other programs, we |
4963 | * cannot make any assumptions since they can | |
4964 | * be replaced dynamically during runtime in | |
4965 | * the program array. | |
4966 | */ | |
4967 | prog->cb_access = 1; | |
80a58d02 | 4968 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 4969 | |
79741b3b AS |
4970 | /* mark bpf_tail_call as different opcode to avoid |
4971 | * conditional branch in the interpeter for every normal | |
4972 | * call and to prevent accidental JITing by JIT compiler | |
4973 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 4974 | */ |
79741b3b | 4975 | insn->imm = 0; |
71189fa9 | 4976 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 4977 | |
3f62bcc9 DB |
4978 | aux = &env->insn_aux_data[i + delta]; |
4979 | if (!bpf_map_ptr_unpriv(aux)) | |
4980 | continue; | |
4981 | ||
b2157399 AS |
4982 | /* instead of changing every JIT dealing with tail_call |
4983 | * emit two extra insns: | |
4984 | * if (index >= max_entries) goto out; | |
4985 | * index &= array->index_mask; | |
4986 | * to avoid out-of-bounds cpu speculation | |
4987 | */ | |
3f62bcc9 | 4988 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 4989 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
4990 | return -EINVAL; |
4991 | } | |
3f62bcc9 DB |
4992 | |
4993 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
b2157399 AS |
4994 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
4995 | map_ptr->max_entries, 2); | |
4996 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
4997 | container_of(map_ptr, | |
4998 | struct bpf_array, | |
4999 | map)->index_mask); | |
5000 | insn_buf[2] = *insn; | |
5001 | cnt = 3; | |
5002 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
5003 | if (!new_prog) | |
5004 | return -ENOMEM; | |
5005 | ||
5006 | delta += cnt - 1; | |
5007 | env->prog = prog = new_prog; | |
5008 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
5009 | continue; |
5010 | } | |
e245c5c6 | 5011 | |
89c63074 DB |
5012 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
5013 | * handlers are currently limited to 64 bit only. | |
5014 | */ | |
5015 | if (ebpf_jit_enabled() && BITS_PER_LONG == 64 && | |
5016 | insn->imm == BPF_FUNC_map_lookup_elem) { | |
3f62bcc9 DB |
5017 | aux = &env->insn_aux_data[i + delta]; |
5018 | if (bpf_map_ptr_poisoned(aux)) | |
5019 | goto patch_call_imm; | |
5020 | ||
5021 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
5022 | if (!map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
5023 | goto patch_call_imm; |
5024 | ||
5025 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
5026 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 5027 | verbose(env, "bpf verifier is misconfigured\n"); |
81ed18ab AS |
5028 | return -EINVAL; |
5029 | } | |
5030 | ||
5031 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
5032 | cnt); | |
5033 | if (!new_prog) | |
5034 | return -ENOMEM; | |
5035 | ||
5036 | delta += cnt - 1; | |
5037 | ||
5038 | /* keep walking new program and skip insns we just inserted */ | |
5039 | env->prog = prog = new_prog; | |
5040 | insn = new_prog->insnsi + i + delta; | |
5041 | continue; | |
5042 | } | |
5043 | ||
109980b8 | 5044 | if (insn->imm == BPF_FUNC_redirect_map) { |
7c300131 DB |
5045 | /* Note, we cannot use prog directly as imm as subsequent |
5046 | * rewrites would still change the prog pointer. The only | |
5047 | * stable address we can use is aux, which also works with | |
5048 | * prog clones during blinding. | |
5049 | */ | |
5050 | u64 addr = (unsigned long)prog->aux; | |
109980b8 DB |
5051 | struct bpf_insn r4_ld[] = { |
5052 | BPF_LD_IMM64(BPF_REG_4, addr), | |
5053 | *insn, | |
5054 | }; | |
5055 | cnt = ARRAY_SIZE(r4_ld); | |
5056 | ||
5057 | new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt); | |
5058 | if (!new_prog) | |
5059 | return -ENOMEM; | |
5060 | ||
5061 | delta += cnt - 1; | |
5062 | env->prog = prog = new_prog; | |
5063 | insn = new_prog->insnsi + i + delta; | |
5064 | } | |
81ed18ab | 5065 | patch_call_imm: |
00176a34 | 5066 | fn = env->ops->get_func_proto(insn->imm); |
79741b3b AS |
5067 | /* all functions that have prototype and verifier allowed |
5068 | * programs to call them, must be real in-kernel functions | |
5069 | */ | |
5070 | if (!fn->func) { | |
61bd5218 JK |
5071 | verbose(env, |
5072 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
5073 | func_id_name(insn->imm), insn->imm); |
5074 | return -EFAULT; | |
e245c5c6 | 5075 | } |
79741b3b | 5076 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 5077 | } |
e245c5c6 | 5078 | |
79741b3b AS |
5079 | return 0; |
5080 | } | |
e245c5c6 | 5081 | |
58e2af8b | 5082 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 5083 | { |
58e2af8b | 5084 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
5085 | int i; |
5086 | ||
5087 | if (!env->explored_states) | |
5088 | return; | |
5089 | ||
5090 | for (i = 0; i < env->prog->len; i++) { | |
5091 | sl = env->explored_states[i]; | |
5092 | ||
5093 | if (sl) | |
5094 | while (sl != STATE_LIST_MARK) { | |
5095 | sln = sl->next; | |
1969db47 | 5096 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
5097 | kfree(sl); |
5098 | sl = sln; | |
5099 | } | |
5100 | } | |
5101 | ||
5102 | kfree(env->explored_states); | |
5103 | } | |
5104 | ||
9bac3d6d | 5105 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 5106 | { |
58e2af8b | 5107 | struct bpf_verifier_env *env; |
61bd5218 | 5108 | struct bpf_verifer_log *log; |
51580e79 AS |
5109 | int ret = -EINVAL; |
5110 | ||
eba0c929 AB |
5111 | /* no program is valid */ |
5112 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
5113 | return -EINVAL; | |
5114 | ||
58e2af8b | 5115 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
5116 | * allocate/free it every time bpf_check() is called |
5117 | */ | |
58e2af8b | 5118 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
5119 | if (!env) |
5120 | return -ENOMEM; | |
61bd5218 | 5121 | log = &env->log; |
cbd35700 | 5122 | |
3df126f3 JK |
5123 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
5124 | (*prog)->len); | |
5125 | ret = -ENOMEM; | |
5126 | if (!env->insn_aux_data) | |
5127 | goto err_free_env; | |
9bac3d6d | 5128 | env->prog = *prog; |
00176a34 | 5129 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 5130 | |
cbd35700 AS |
5131 | /* grab the mutex to protect few globals used by verifier */ |
5132 | mutex_lock(&bpf_verifier_lock); | |
5133 | ||
5134 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
5135 | /* user requested verbose verifier output | |
5136 | * and supplied buffer to store the verification trace | |
5137 | */ | |
e7bf8249 JK |
5138 | log->level = attr->log_level; |
5139 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
5140 | log->len_total = attr->log_size; | |
cbd35700 AS |
5141 | |
5142 | ret = -EINVAL; | |
e7bf8249 JK |
5143 | /* log attributes have to be sane */ |
5144 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
5145 | !log->level || !log->ubuf) | |
3df126f3 | 5146 | goto err_unlock; |
cbd35700 | 5147 | } |
1ad2f583 DB |
5148 | |
5149 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
5150 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 5151 | env->strict_alignment = true; |
cbd35700 | 5152 | |
ab3f0063 JK |
5153 | if (env->prog->aux->offload) { |
5154 | ret = bpf_prog_offload_verifier_prep(env); | |
5155 | if (ret) | |
5156 | goto err_unlock; | |
5157 | } | |
5158 | ||
0246e64d AS |
5159 | ret = replace_map_fd_with_map_ptr(env); |
5160 | if (ret < 0) | |
5161 | goto skip_full_check; | |
5162 | ||
9bac3d6d | 5163 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 5164 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
5165 | GFP_USER); |
5166 | ret = -ENOMEM; | |
5167 | if (!env->explored_states) | |
5168 | goto skip_full_check; | |
5169 | ||
475fb78f AS |
5170 | ret = check_cfg(env); |
5171 | if (ret < 0) | |
5172 | goto skip_full_check; | |
5173 | ||
1be7f75d AS |
5174 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
5175 | ||
17a52670 | 5176 | ret = do_check(env); |
8c01c4f8 CG |
5177 | if (env->cur_state) { |
5178 | free_verifier_state(env->cur_state, true); | |
5179 | env->cur_state = NULL; | |
5180 | } | |
cbd35700 | 5181 | |
0246e64d | 5182 | skip_full_check: |
638f5b90 | 5183 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 5184 | free_states(env); |
0246e64d | 5185 | |
c131187d AS |
5186 | if (ret == 0) |
5187 | sanitize_dead_code(env); | |
5188 | ||
9bac3d6d AS |
5189 | if (ret == 0) |
5190 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
5191 | ret = convert_ctx_accesses(env); | |
5192 | ||
e245c5c6 | 5193 | if (ret == 0) |
79741b3b | 5194 | ret = fixup_bpf_calls(env); |
e245c5c6 | 5195 | |
a2a7d570 | 5196 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 5197 | ret = -ENOSPC; |
a2a7d570 | 5198 | if (log->level && !log->ubuf) { |
cbd35700 | 5199 | ret = -EFAULT; |
a2a7d570 | 5200 | goto err_release_maps; |
cbd35700 AS |
5201 | } |
5202 | ||
0246e64d AS |
5203 | if (ret == 0 && env->used_map_cnt) { |
5204 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
5205 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
5206 | sizeof(env->used_maps[0]), | |
5207 | GFP_KERNEL); | |
0246e64d | 5208 | |
9bac3d6d | 5209 | if (!env->prog->aux->used_maps) { |
0246e64d | 5210 | ret = -ENOMEM; |
a2a7d570 | 5211 | goto err_release_maps; |
0246e64d AS |
5212 | } |
5213 | ||
9bac3d6d | 5214 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 5215 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 5216 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
5217 | |
5218 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
5219 | * bpf_ld_imm64 instructions | |
5220 | */ | |
5221 | convert_pseudo_ld_imm64(env); | |
5222 | } | |
cbd35700 | 5223 | |
a2a7d570 | 5224 | err_release_maps: |
9bac3d6d | 5225 | if (!env->prog->aux->used_maps) |
0246e64d | 5226 | /* if we didn't copy map pointers into bpf_prog_info, release |
d6f1a482 | 5227 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
5228 | */ |
5229 | release_maps(env); | |
9bac3d6d | 5230 | *prog = env->prog; |
3df126f3 | 5231 | err_unlock: |
cbd35700 | 5232 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
5233 | vfree(env->insn_aux_data); |
5234 | err_free_env: | |
5235 | kfree(env); | |
51580e79 AS |
5236 | return ret; |
5237 | } |