]>
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> |
51580e79 | 23 | |
f4ac7e0b JK |
24 | #include "disasm.h" |
25 | ||
00176a34 JK |
26 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
27 | #define BPF_PROG_TYPE(_id, _name) \ | |
28 | [_id] = & _name ## _verifier_ops, | |
29 | #define BPF_MAP_TYPE(_id, _ops) | |
30 | #include <linux/bpf_types.h> | |
31 | #undef BPF_PROG_TYPE | |
32 | #undef BPF_MAP_TYPE | |
33 | }; | |
34 | ||
51580e79 AS |
35 | /* bpf_check() is a static code analyzer that walks eBPF program |
36 | * instruction by instruction and updates register/stack state. | |
37 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
38 | * | |
39 | * The first pass is depth-first-search to check that the program is a DAG. | |
40 | * It rejects the following programs: | |
41 | * - larger than BPF_MAXINSNS insns | |
42 | * - if loop is present (detected via back-edge) | |
43 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
44 | * - out of bounds or malformed jumps | |
45 | * The second pass is all possible path descent from the 1st insn. | |
46 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 47 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
48 | * insn is less then 4K, but there are too many branches that change stack/regs. |
49 | * Number of 'branches to be analyzed' is limited to 1k | |
50 | * | |
51 | * On entry to each instruction, each register has a type, and the instruction | |
52 | * changes the types of the registers depending on instruction semantics. | |
53 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
54 | * copied to R1. | |
55 | * | |
56 | * All registers are 64-bit. | |
57 | * R0 - return register | |
58 | * R1-R5 argument passing registers | |
59 | * R6-R9 callee saved registers | |
60 | * R10 - frame pointer read-only | |
61 | * | |
62 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
63 | * and has type PTR_TO_CTX. | |
64 | * | |
65 | * Verifier tracks arithmetic operations on pointers in case: | |
66 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
67 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
68 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
69 | * and 2nd arithmetic instruction is pattern matched to recognize | |
70 | * that it wants to construct a pointer to some element within stack. | |
71 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
72 | * (and -20 constant is saved for further stack bounds checking). | |
73 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
74 | * | |
f1174f77 | 75 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 76 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 77 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
78 | * |
79 | * When verifier sees load or store instructions the type of base register | |
f1174f77 | 80 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer |
51580e79 AS |
81 | * types recognized by check_mem_access() function. |
82 | * | |
83 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
84 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
85 | * | |
86 | * registers used to pass values to function calls are checked against | |
87 | * function argument constraints. | |
88 | * | |
89 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
90 | * It means that the register type passed to this function must be | |
91 | * PTR_TO_STACK and it will be used inside the function as | |
92 | * 'pointer to map element key' | |
93 | * | |
94 | * For example the argument constraints for bpf_map_lookup_elem(): | |
95 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
96 | * .arg1_type = ARG_CONST_MAP_PTR, | |
97 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
98 | * | |
99 | * ret_type says that this function returns 'pointer to map elem value or null' | |
100 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
101 | * 2nd argument should be a pointer to stack, which will be used inside | |
102 | * the helper function as a pointer to map element key. | |
103 | * | |
104 | * On the kernel side the helper function looks like: | |
105 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
106 | * { | |
107 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
108 | * void *key = (void *) (unsigned long) r2; | |
109 | * void *value; | |
110 | * | |
111 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
112 | * [key, key + map->key_size) bytes are valid and were initialized on | |
113 | * the stack of eBPF program. | |
114 | * } | |
115 | * | |
116 | * Corresponding eBPF program may look like: | |
117 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
118 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
119 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
120 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
121 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
122 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
123 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
124 | * | |
125 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
126 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
127 | * and were initialized prior to this call. | |
128 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
129 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
130 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
131 | * returns ether pointer to map value or NULL. | |
132 | * | |
133 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
134 | * insn, the register holding that pointer in the true branch changes state to | |
135 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
136 | * branch. See check_cond_jmp_op(). | |
137 | * | |
138 | * After the call R0 is set to return type of the function and registers R1-R5 | |
139 | * are set to NOT_INIT to indicate that they are no longer readable. | |
140 | */ | |
141 | ||
17a52670 | 142 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 143 | struct bpf_verifier_stack_elem { |
17a52670 AS |
144 | /* verifer state is 'st' |
145 | * before processing instruction 'insn_idx' | |
146 | * and after processing instruction 'prev_insn_idx' | |
147 | */ | |
58e2af8b | 148 | struct bpf_verifier_state st; |
17a52670 AS |
149 | int insn_idx; |
150 | int prev_insn_idx; | |
58e2af8b | 151 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
152 | }; |
153 | ||
8e17c1b1 | 154 | #define BPF_COMPLEXITY_LIMIT_INSNS 131072 |
07016151 DB |
155 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 |
156 | ||
fad73a1a MKL |
157 | #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA) |
158 | ||
33ff9823 DB |
159 | struct bpf_call_arg_meta { |
160 | struct bpf_map *map_ptr; | |
435faee1 | 161 | bool raw_mode; |
36bbef52 | 162 | bool pkt_access; |
435faee1 DB |
163 | int regno; |
164 | int access_size; | |
33ff9823 DB |
165 | }; |
166 | ||
cbd35700 AS |
167 | static DEFINE_MUTEX(bpf_verifier_lock); |
168 | ||
169 | /* log_level controls verbosity level of eBPF verifier. | |
170 | * verbose() is used to dump the verification trace to the log, so the user | |
171 | * can figure out what's wrong with the program | |
172 | */ | |
61bd5218 JK |
173 | static __printf(2, 3) void verbose(struct bpf_verifier_env *env, |
174 | const char *fmt, ...) | |
cbd35700 | 175 | { |
61bd5218 | 176 | struct bpf_verifer_log *log = &env->log; |
a2a7d570 | 177 | unsigned int n; |
cbd35700 AS |
178 | va_list args; |
179 | ||
a2a7d570 | 180 | if (!log->level || !log->ubuf || bpf_verifier_log_full(log)) |
cbd35700 AS |
181 | return; |
182 | ||
183 | va_start(args, fmt); | |
a2a7d570 | 184 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
cbd35700 | 185 | va_end(args); |
a2a7d570 JK |
186 | |
187 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
188 | "verifier log line truncated - local buffer too short\n"); | |
189 | ||
190 | n = min(log->len_total - log->len_used - 1, n); | |
191 | log->kbuf[n] = '\0'; | |
192 | ||
193 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) | |
194 | log->len_used += n; | |
195 | else | |
196 | log->ubuf = NULL; | |
cbd35700 AS |
197 | } |
198 | ||
de8f3a83 DB |
199 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
200 | { | |
201 | return type == PTR_TO_PACKET || | |
202 | type == PTR_TO_PACKET_META; | |
203 | } | |
204 | ||
17a52670 AS |
205 | /* string representation of 'enum bpf_reg_type' */ |
206 | static const char * const reg_type_str[] = { | |
207 | [NOT_INIT] = "?", | |
f1174f77 | 208 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
209 | [PTR_TO_CTX] = "ctx", |
210 | [CONST_PTR_TO_MAP] = "map_ptr", | |
211 | [PTR_TO_MAP_VALUE] = "map_value", | |
212 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 213 | [PTR_TO_STACK] = "fp", |
969bf05e | 214 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 215 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 216 | [PTR_TO_PACKET_END] = "pkt_end", |
17a52670 AS |
217 | }; |
218 | ||
61bd5218 JK |
219 | static void print_verifier_state(struct bpf_verifier_env *env, |
220 | struct bpf_verifier_state *state) | |
17a52670 | 221 | { |
58e2af8b | 222 | struct bpf_reg_state *reg; |
17a52670 AS |
223 | enum bpf_reg_type t; |
224 | int i; | |
225 | ||
226 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
227 | reg = &state->regs[i]; |
228 | t = reg->type; | |
17a52670 AS |
229 | if (t == NOT_INIT) |
230 | continue; | |
61bd5218 | 231 | verbose(env, " R%d=%s", i, reg_type_str[t]); |
f1174f77 EC |
232 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
233 | tnum_is_const(reg->var_off)) { | |
234 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 235 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 236 | } else { |
61bd5218 | 237 | verbose(env, "(id=%d", reg->id); |
f1174f77 | 238 | if (t != SCALAR_VALUE) |
61bd5218 | 239 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 240 | if (type_is_pkt_pointer(t)) |
61bd5218 | 241 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
242 | else if (t == CONST_PTR_TO_MAP || |
243 | t == PTR_TO_MAP_VALUE || | |
244 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 245 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
246 | reg->map_ptr->key_size, |
247 | reg->map_ptr->value_size); | |
7d1238f2 EC |
248 | if (tnum_is_const(reg->var_off)) { |
249 | /* Typically an immediate SCALAR_VALUE, but | |
250 | * could be a pointer whose offset is too big | |
251 | * for reg->off | |
252 | */ | |
61bd5218 | 253 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
254 | } else { |
255 | if (reg->smin_value != reg->umin_value && | |
256 | reg->smin_value != S64_MIN) | |
61bd5218 | 257 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
258 | (long long)reg->smin_value); |
259 | if (reg->smax_value != reg->umax_value && | |
260 | reg->smax_value != S64_MAX) | |
61bd5218 | 261 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
262 | (long long)reg->smax_value); |
263 | if (reg->umin_value != 0) | |
61bd5218 | 264 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
265 | (unsigned long long)reg->umin_value); |
266 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 267 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
268 | (unsigned long long)reg->umax_value); |
269 | if (!tnum_is_unknown(reg->var_off)) { | |
270 | char tn_buf[48]; | |
f1174f77 | 271 | |
7d1238f2 | 272 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 273 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 274 | } |
f1174f77 | 275 | } |
61bd5218 | 276 | verbose(env, ")"); |
f1174f77 | 277 | } |
17a52670 | 278 | } |
638f5b90 AS |
279 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
280 | if (state->stack[i].slot_type[0] == STACK_SPILL) | |
281 | verbose(env, " fp%d=%s", | |
282 | -MAX_BPF_STACK + i * BPF_REG_SIZE, | |
283 | reg_type_str[state->stack[i].spilled_ptr.type]); | |
17a52670 | 284 | } |
61bd5218 | 285 | verbose(env, "\n"); |
17a52670 AS |
286 | } |
287 | ||
638f5b90 AS |
288 | static int copy_stack_state(struct bpf_verifier_state *dst, |
289 | const struct bpf_verifier_state *src) | |
17a52670 | 290 | { |
638f5b90 AS |
291 | if (!src->stack) |
292 | return 0; | |
293 | if (WARN_ON_ONCE(dst->allocated_stack < src->allocated_stack)) { | |
294 | /* internal bug, make state invalid to reject the program */ | |
295 | memset(dst, 0, sizeof(*dst)); | |
296 | return -EFAULT; | |
297 | } | |
298 | memcpy(dst->stack, src->stack, | |
299 | sizeof(*src->stack) * (src->allocated_stack / BPF_REG_SIZE)); | |
300 | return 0; | |
301 | } | |
302 | ||
303 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
304 | * make it consume minimal amount of memory. check_stack_write() access from | |
305 | * the program calls into realloc_verifier_state() to grow the stack size. | |
306 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state | |
307 | * which this function copies over. It points to previous bpf_verifier_state | |
308 | * which is never reallocated | |
309 | */ | |
310 | static int realloc_verifier_state(struct bpf_verifier_state *state, int size, | |
311 | bool copy_old) | |
312 | { | |
313 | u32 old_size = state->allocated_stack; | |
314 | struct bpf_stack_state *new_stack; | |
315 | int slot = size / BPF_REG_SIZE; | |
316 | ||
317 | if (size <= old_size || !size) { | |
318 | if (copy_old) | |
319 | return 0; | |
320 | state->allocated_stack = slot * BPF_REG_SIZE; | |
321 | if (!size && old_size) { | |
322 | kfree(state->stack); | |
323 | state->stack = NULL; | |
324 | } | |
325 | return 0; | |
326 | } | |
327 | new_stack = kmalloc_array(slot, sizeof(struct bpf_stack_state), | |
328 | GFP_KERNEL); | |
329 | if (!new_stack) | |
330 | return -ENOMEM; | |
331 | if (copy_old) { | |
332 | if (state->stack) | |
333 | memcpy(new_stack, state->stack, | |
334 | sizeof(*new_stack) * (old_size / BPF_REG_SIZE)); | |
335 | memset(new_stack + old_size / BPF_REG_SIZE, 0, | |
336 | sizeof(*new_stack) * (size - old_size) / BPF_REG_SIZE); | |
337 | } | |
338 | state->allocated_stack = slot * BPF_REG_SIZE; | |
339 | kfree(state->stack); | |
340 | state->stack = new_stack; | |
341 | return 0; | |
342 | } | |
343 | ||
1969db47 AS |
344 | static void free_verifier_state(struct bpf_verifier_state *state, |
345 | bool free_self) | |
638f5b90 AS |
346 | { |
347 | kfree(state->stack); | |
1969db47 AS |
348 | if (free_self) |
349 | kfree(state); | |
638f5b90 AS |
350 | } |
351 | ||
352 | /* copy verifier state from src to dst growing dst stack space | |
353 | * when necessary to accommodate larger src stack | |
354 | */ | |
355 | static int copy_verifier_state(struct bpf_verifier_state *dst, | |
356 | const struct bpf_verifier_state *src) | |
357 | { | |
358 | int err; | |
359 | ||
360 | err = realloc_verifier_state(dst, src->allocated_stack, false); | |
361 | if (err) | |
362 | return err; | |
363 | memcpy(dst, src, offsetof(struct bpf_verifier_state, allocated_stack)); | |
364 | return copy_stack_state(dst, src); | |
365 | } | |
366 | ||
367 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, | |
368 | int *insn_idx) | |
369 | { | |
370 | struct bpf_verifier_state *cur = env->cur_state; | |
371 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
372 | int err; | |
17a52670 AS |
373 | |
374 | if (env->head == NULL) | |
638f5b90 | 375 | return -ENOENT; |
17a52670 | 376 | |
638f5b90 AS |
377 | if (cur) { |
378 | err = copy_verifier_state(cur, &head->st); | |
379 | if (err) | |
380 | return err; | |
381 | } | |
382 | if (insn_idx) | |
383 | *insn_idx = head->insn_idx; | |
17a52670 | 384 | if (prev_insn_idx) |
638f5b90 AS |
385 | *prev_insn_idx = head->prev_insn_idx; |
386 | elem = head->next; | |
1969db47 | 387 | free_verifier_state(&head->st, false); |
638f5b90 | 388 | kfree(head); |
17a52670 AS |
389 | env->head = elem; |
390 | env->stack_size--; | |
638f5b90 | 391 | return 0; |
17a52670 AS |
392 | } |
393 | ||
58e2af8b JK |
394 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
395 | int insn_idx, int prev_insn_idx) | |
17a52670 | 396 | { |
638f5b90 | 397 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 398 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 399 | int err; |
17a52670 | 400 | |
638f5b90 | 401 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
402 | if (!elem) |
403 | goto err; | |
404 | ||
17a52670 AS |
405 | elem->insn_idx = insn_idx; |
406 | elem->prev_insn_idx = prev_insn_idx; | |
407 | elem->next = env->head; | |
408 | env->head = elem; | |
409 | env->stack_size++; | |
1969db47 AS |
410 | err = copy_verifier_state(&elem->st, cur); |
411 | if (err) | |
412 | goto err; | |
07016151 | 413 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
61bd5218 | 414 | verbose(env, "BPF program is too complex\n"); |
17a52670 AS |
415 | goto err; |
416 | } | |
417 | return &elem->st; | |
418 | err: | |
419 | /* pop all elements and return */ | |
638f5b90 | 420 | while (!pop_stack(env, NULL, NULL)); |
17a52670 AS |
421 | return NULL; |
422 | } | |
423 | ||
424 | #define CALLER_SAVED_REGS 6 | |
425 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
426 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
427 | }; | |
428 | ||
f1174f77 EC |
429 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
430 | ||
b03c9f9f EC |
431 | /* Mark the unknown part of a register (variable offset or scalar value) as |
432 | * known to have the value @imm. | |
433 | */ | |
434 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
435 | { | |
436 | reg->id = 0; | |
437 | reg->var_off = tnum_const(imm); | |
438 | reg->smin_value = (s64)imm; | |
439 | reg->smax_value = (s64)imm; | |
440 | reg->umin_value = imm; | |
441 | reg->umax_value = imm; | |
442 | } | |
443 | ||
f1174f77 EC |
444 | /* Mark the 'variable offset' part of a register as zero. This should be |
445 | * used only on registers holding a pointer type. | |
446 | */ | |
447 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 448 | { |
b03c9f9f | 449 | __mark_reg_known(reg, 0); |
f1174f77 | 450 | } |
a9789ef9 | 451 | |
61bd5218 JK |
452 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
453 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
454 | { |
455 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 456 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
457 | /* Something bad happened, let's kill all regs */ |
458 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
459 | __mark_reg_not_init(regs + regno); | |
460 | return; | |
461 | } | |
462 | __mark_reg_known_zero(regs + regno); | |
463 | } | |
464 | ||
de8f3a83 DB |
465 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
466 | { | |
467 | return type_is_pkt_pointer(reg->type); | |
468 | } | |
469 | ||
470 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
471 | { | |
472 | return reg_is_pkt_pointer(reg) || | |
473 | reg->type == PTR_TO_PACKET_END; | |
474 | } | |
475 | ||
476 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
477 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
478 | enum bpf_reg_type which) | |
479 | { | |
480 | /* The register can already have a range from prior markings. | |
481 | * This is fine as long as it hasn't been advanced from its | |
482 | * origin. | |
483 | */ | |
484 | return reg->type == which && | |
485 | reg->id == 0 && | |
486 | reg->off == 0 && | |
487 | tnum_equals_const(reg->var_off, 0); | |
488 | } | |
489 | ||
b03c9f9f EC |
490 | /* Attempts to improve min/max values based on var_off information */ |
491 | static void __update_reg_bounds(struct bpf_reg_state *reg) | |
492 | { | |
493 | /* min signed is max(sign bit) | min(other bits) */ | |
494 | reg->smin_value = max_t(s64, reg->smin_value, | |
495 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
496 | /* max signed is min(sign bit) | max(other bits) */ | |
497 | reg->smax_value = min_t(s64, reg->smax_value, | |
498 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
499 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
500 | reg->umax_value = min(reg->umax_value, | |
501 | reg->var_off.value | reg->var_off.mask); | |
502 | } | |
503 | ||
504 | /* Uses signed min/max values to inform unsigned, and vice-versa */ | |
505 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) | |
506 | { | |
507 | /* Learn sign from signed bounds. | |
508 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
509 | * are the same, so combine. This works even in the negative case, e.g. | |
510 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
511 | */ | |
512 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
513 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
514 | reg->umin_value); | |
515 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
516 | reg->umax_value); | |
517 | return; | |
518 | } | |
519 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
520 | * boundary, so we must be careful. | |
521 | */ | |
522 | if ((s64)reg->umax_value >= 0) { | |
523 | /* Positive. We can't learn anything from the smin, but smax | |
524 | * is positive, hence safe. | |
525 | */ | |
526 | reg->smin_value = reg->umin_value; | |
527 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
528 | reg->umax_value); | |
529 | } else if ((s64)reg->umin_value < 0) { | |
530 | /* Negative. We can't learn anything from the smax, but smin | |
531 | * is negative, hence safe. | |
532 | */ | |
533 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
534 | reg->umin_value); | |
535 | reg->smax_value = reg->umax_value; | |
536 | } | |
537 | } | |
538 | ||
539 | /* Attempts to improve var_off based on unsigned min/max information */ | |
540 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
541 | { | |
542 | reg->var_off = tnum_intersect(reg->var_off, | |
543 | tnum_range(reg->umin_value, | |
544 | reg->umax_value)); | |
545 | } | |
546 | ||
547 | /* Reset the min/max bounds of a register */ | |
548 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
549 | { | |
550 | reg->smin_value = S64_MIN; | |
551 | reg->smax_value = S64_MAX; | |
552 | reg->umin_value = 0; | |
553 | reg->umax_value = U64_MAX; | |
554 | } | |
555 | ||
f1174f77 EC |
556 | /* Mark a register as having a completely unknown (scalar) value. */ |
557 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
558 | { | |
559 | reg->type = SCALAR_VALUE; | |
560 | reg->id = 0; | |
561 | reg->off = 0; | |
562 | reg->var_off = tnum_unknown; | |
b03c9f9f | 563 | __mark_reg_unbounded(reg); |
f1174f77 EC |
564 | } |
565 | ||
61bd5218 JK |
566 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
567 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
568 | { |
569 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 570 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
f1174f77 EC |
571 | /* Something bad happened, let's kill all regs */ |
572 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
573 | __mark_reg_not_init(regs + regno); | |
574 | return; | |
575 | } | |
576 | __mark_reg_unknown(regs + regno); | |
577 | } | |
578 | ||
579 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
580 | { | |
581 | __mark_reg_unknown(reg); | |
582 | reg->type = NOT_INIT; | |
583 | } | |
584 | ||
61bd5218 JK |
585 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
586 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
587 | { |
588 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 589 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
f1174f77 EC |
590 | /* Something bad happened, let's kill all regs */ |
591 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
592 | __mark_reg_not_init(regs + regno); | |
593 | return; | |
594 | } | |
595 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
596 | } |
597 | ||
61bd5218 JK |
598 | static void init_reg_state(struct bpf_verifier_env *env, |
599 | struct bpf_reg_state *regs) | |
17a52670 AS |
600 | { |
601 | int i; | |
602 | ||
dc503a8a | 603 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 604 | mark_reg_not_init(env, regs, i); |
dc503a8a EC |
605 | regs[i].live = REG_LIVE_NONE; |
606 | } | |
17a52670 AS |
607 | |
608 | /* frame pointer */ | |
f1174f77 | 609 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 610 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
17a52670 AS |
611 | |
612 | /* 1st arg to a function */ | |
613 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
61bd5218 | 614 | mark_reg_known_zero(env, regs, BPF_REG_1); |
6760bf2d DB |
615 | } |
616 | ||
17a52670 AS |
617 | enum reg_arg_type { |
618 | SRC_OP, /* register is used as source operand */ | |
619 | DST_OP, /* register is used as destination operand */ | |
620 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
621 | }; | |
622 | ||
dc503a8a EC |
623 | static void mark_reg_read(const struct bpf_verifier_state *state, u32 regno) |
624 | { | |
625 | struct bpf_verifier_state *parent = state->parent; | |
626 | ||
8fe2d6cc AS |
627 | if (regno == BPF_REG_FP) |
628 | /* We don't need to worry about FP liveness because it's read-only */ | |
629 | return; | |
630 | ||
dc503a8a EC |
631 | while (parent) { |
632 | /* if read wasn't screened by an earlier write ... */ | |
633 | if (state->regs[regno].live & REG_LIVE_WRITTEN) | |
634 | break; | |
635 | /* ... then we depend on parent's value */ | |
636 | parent->regs[regno].live |= REG_LIVE_READ; | |
637 | state = parent; | |
638 | parent = state->parent; | |
639 | } | |
640 | } | |
641 | ||
642 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, | |
17a52670 AS |
643 | enum reg_arg_type t) |
644 | { | |
638f5b90 | 645 | struct bpf_reg_state *regs = env->cur_state->regs; |
dc503a8a | 646 | |
17a52670 | 647 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 648 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
649 | return -EINVAL; |
650 | } | |
651 | ||
652 | if (t == SRC_OP) { | |
653 | /* check whether register used as source operand can be read */ | |
654 | if (regs[regno].type == NOT_INIT) { | |
61bd5218 | 655 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
656 | return -EACCES; |
657 | } | |
638f5b90 | 658 | mark_reg_read(env->cur_state, regno); |
17a52670 AS |
659 | } else { |
660 | /* check whether register used as dest operand can be written to */ | |
661 | if (regno == BPF_REG_FP) { | |
61bd5218 | 662 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
663 | return -EACCES; |
664 | } | |
dc503a8a | 665 | regs[regno].live |= REG_LIVE_WRITTEN; |
17a52670 | 666 | if (t == DST_OP) |
61bd5218 | 667 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
668 | } |
669 | return 0; | |
670 | } | |
671 | ||
1be7f75d AS |
672 | static bool is_spillable_regtype(enum bpf_reg_type type) |
673 | { | |
674 | switch (type) { | |
675 | case PTR_TO_MAP_VALUE: | |
676 | case PTR_TO_MAP_VALUE_OR_NULL: | |
677 | case PTR_TO_STACK: | |
678 | case PTR_TO_CTX: | |
969bf05e | 679 | case PTR_TO_PACKET: |
de8f3a83 | 680 | case PTR_TO_PACKET_META: |
969bf05e | 681 | case PTR_TO_PACKET_END: |
1be7f75d AS |
682 | case CONST_PTR_TO_MAP: |
683 | return true; | |
684 | default: | |
685 | return false; | |
686 | } | |
687 | } | |
688 | ||
17a52670 AS |
689 | /* check_stack_read/write functions track spill/fill of registers, |
690 | * stack boundary and alignment are checked in check_mem_access() | |
691 | */ | |
61bd5218 JK |
692 | static int check_stack_write(struct bpf_verifier_env *env, |
693 | struct bpf_verifier_state *state, int off, | |
58e2af8b | 694 | int size, int value_regno) |
17a52670 | 695 | { |
638f5b90 AS |
696 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
697 | ||
698 | err = realloc_verifier_state(state, round_up(slot + 1, BPF_REG_SIZE), | |
699 | true); | |
700 | if (err) | |
701 | return err; | |
9c399760 AS |
702 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
703 | * so it's aligned access and [off, off + size) are within stack limits | |
704 | */ | |
638f5b90 AS |
705 | if (!env->allow_ptr_leaks && |
706 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
707 | size != BPF_REG_SIZE) { | |
708 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
709 | return -EACCES; | |
710 | } | |
17a52670 AS |
711 | |
712 | if (value_regno >= 0 && | |
1be7f75d | 713 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
714 | |
715 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 716 | if (size != BPF_REG_SIZE) { |
61bd5218 | 717 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
718 | return -EACCES; |
719 | } | |
720 | ||
17a52670 | 721 | /* save register state */ |
638f5b90 AS |
722 | state->stack[spi].spilled_ptr = state->regs[value_regno]; |
723 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
17a52670 | 724 | |
9c399760 | 725 | for (i = 0; i < BPF_REG_SIZE; i++) |
638f5b90 | 726 | state->stack[spi].slot_type[i] = STACK_SPILL; |
9c399760 | 727 | } else { |
17a52670 | 728 | /* regular write of data into stack */ |
638f5b90 | 729 | state->stack[spi].spilled_ptr = (struct bpf_reg_state) {}; |
9c399760 AS |
730 | |
731 | for (i = 0; i < size; i++) | |
638f5b90 AS |
732 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
733 | STACK_MISC; | |
17a52670 AS |
734 | } |
735 | return 0; | |
736 | } | |
737 | ||
dc503a8a EC |
738 | static void mark_stack_slot_read(const struct bpf_verifier_state *state, int slot) |
739 | { | |
740 | struct bpf_verifier_state *parent = state->parent; | |
741 | ||
742 | while (parent) { | |
743 | /* if read wasn't screened by an earlier write ... */ | |
638f5b90 | 744 | if (state->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN) |
dc503a8a EC |
745 | break; |
746 | /* ... then we depend on parent's value */ | |
638f5b90 | 747 | parent->stack[slot].spilled_ptr.live |= REG_LIVE_READ; |
dc503a8a EC |
748 | state = parent; |
749 | parent = state->parent; | |
750 | } | |
751 | } | |
752 | ||
61bd5218 JK |
753 | static int check_stack_read(struct bpf_verifier_env *env, |
754 | struct bpf_verifier_state *state, int off, int size, | |
17a52670 AS |
755 | int value_regno) |
756 | { | |
638f5b90 AS |
757 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
758 | u8 *stype; | |
17a52670 | 759 | |
638f5b90 AS |
760 | if (state->allocated_stack <= slot) { |
761 | verbose(env, "invalid read from stack off %d+0 size %d\n", | |
762 | off, size); | |
763 | return -EACCES; | |
764 | } | |
765 | stype = state->stack[spi].slot_type; | |
17a52670 | 766 | |
638f5b90 | 767 | if (stype[0] == STACK_SPILL) { |
9c399760 | 768 | if (size != BPF_REG_SIZE) { |
61bd5218 | 769 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
770 | return -EACCES; |
771 | } | |
9c399760 | 772 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 773 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 774 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
775 | return -EACCES; |
776 | } | |
777 | } | |
778 | ||
dc503a8a | 779 | if (value_regno >= 0) { |
17a52670 | 780 | /* restore register state from stack */ |
638f5b90 | 781 | state->regs[value_regno] = state->stack[spi].spilled_ptr; |
dc503a8a EC |
782 | mark_stack_slot_read(state, spi); |
783 | } | |
17a52670 AS |
784 | return 0; |
785 | } else { | |
786 | for (i = 0; i < size; i++) { | |
638f5b90 | 787 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_MISC) { |
61bd5218 | 788 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
17a52670 AS |
789 | off, i, size); |
790 | return -EACCES; | |
791 | } | |
792 | } | |
793 | if (value_regno >= 0) | |
794 | /* have read misc data from the stack */ | |
61bd5218 | 795 | mark_reg_unknown(env, state->regs, value_regno); |
17a52670 AS |
796 | return 0; |
797 | } | |
798 | } | |
799 | ||
800 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
f1174f77 | 801 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 802 | int size, bool zero_size_allowed) |
17a52670 | 803 | { |
638f5b90 AS |
804 | struct bpf_reg_state *regs = cur_regs(env); |
805 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 806 | |
9fd29c08 YS |
807 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
808 | off + size > map->value_size) { | |
61bd5218 | 809 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
810 | map->value_size, off, size); |
811 | return -EACCES; | |
812 | } | |
813 | return 0; | |
814 | } | |
815 | ||
f1174f77 EC |
816 | /* check read/write into a map element with possible variable offset */ |
817 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 818 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 819 | { |
638f5b90 | 820 | struct bpf_verifier_state *state = env->cur_state; |
dbcfe5f7 GB |
821 | struct bpf_reg_state *reg = &state->regs[regno]; |
822 | int err; | |
823 | ||
f1174f77 EC |
824 | /* We may have adjusted the register to this map value, so we |
825 | * need to try adding each of min_value and max_value to off | |
826 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 827 | */ |
61bd5218 JK |
828 | if (env->log.level) |
829 | print_verifier_state(env, state); | |
dbcfe5f7 GB |
830 | /* The minimum value is only important with signed |
831 | * comparisons where we can't assume the floor of a | |
832 | * value is 0. If we are using signed variables for our | |
833 | * index'es we need to make sure that whatever we use | |
834 | * will have a set floor within our range. | |
835 | */ | |
b03c9f9f | 836 | if (reg->smin_value < 0) { |
61bd5218 | 837 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
838 | regno); |
839 | return -EACCES; | |
840 | } | |
9fd29c08 YS |
841 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
842 | zero_size_allowed); | |
dbcfe5f7 | 843 | if (err) { |
61bd5218 JK |
844 | verbose(env, "R%d min value is outside of the array range\n", |
845 | regno); | |
dbcfe5f7 GB |
846 | return err; |
847 | } | |
848 | ||
b03c9f9f EC |
849 | /* If we haven't set a max value then we need to bail since we can't be |
850 | * sure we won't do bad things. | |
851 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 852 | */ |
b03c9f9f | 853 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 854 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
855 | regno); |
856 | return -EACCES; | |
857 | } | |
9fd29c08 YS |
858 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
859 | zero_size_allowed); | |
f1174f77 | 860 | if (err) |
61bd5218 JK |
861 | verbose(env, "R%d max value is outside of the array range\n", |
862 | regno); | |
f1174f77 | 863 | return err; |
dbcfe5f7 GB |
864 | } |
865 | ||
969bf05e AS |
866 | #define MAX_PACKET_OFF 0xffff |
867 | ||
58e2af8b | 868 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
869 | const struct bpf_call_arg_meta *meta, |
870 | enum bpf_access_type t) | |
4acf6c0b | 871 | { |
36bbef52 | 872 | switch (env->prog->type) { |
3a0af8fd TG |
873 | case BPF_PROG_TYPE_LWT_IN: |
874 | case BPF_PROG_TYPE_LWT_OUT: | |
875 | /* dst_input() and dst_output() can't write for now */ | |
876 | if (t == BPF_WRITE) | |
877 | return false; | |
7e57fbb2 | 878 | /* fallthrough */ |
36bbef52 DB |
879 | case BPF_PROG_TYPE_SCHED_CLS: |
880 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 881 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 882 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 883 | case BPF_PROG_TYPE_SK_SKB: |
36bbef52 DB |
884 | if (meta) |
885 | return meta->pkt_access; | |
886 | ||
887 | env->seen_direct_write = true; | |
4acf6c0b BB |
888 | return true; |
889 | default: | |
890 | return false; | |
891 | } | |
892 | } | |
893 | ||
f1174f77 | 894 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 895 | int off, int size, bool zero_size_allowed) |
969bf05e | 896 | { |
638f5b90 | 897 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 898 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 899 | |
9fd29c08 YS |
900 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
901 | (u64)off + size > reg->range) { | |
61bd5218 | 902 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 903 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
904 | return -EACCES; |
905 | } | |
906 | return 0; | |
907 | } | |
908 | ||
f1174f77 | 909 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 910 | int size, bool zero_size_allowed) |
f1174f77 | 911 | { |
638f5b90 | 912 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
913 | struct bpf_reg_state *reg = ®s[regno]; |
914 | int err; | |
915 | ||
916 | /* We may have added a variable offset to the packet pointer; but any | |
917 | * reg->range we have comes after that. We are only checking the fixed | |
918 | * offset. | |
919 | */ | |
920 | ||
921 | /* We don't allow negative numbers, because we aren't tracking enough | |
922 | * detail to prove they're safe. | |
923 | */ | |
b03c9f9f | 924 | if (reg->smin_value < 0) { |
61bd5218 | 925 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
926 | regno); |
927 | return -EACCES; | |
928 | } | |
9fd29c08 | 929 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 930 | if (err) { |
61bd5218 | 931 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
932 | return err; |
933 | } | |
934 | return err; | |
935 | } | |
936 | ||
937 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 938 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 939 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 940 | { |
f96da094 DB |
941 | struct bpf_insn_access_aux info = { |
942 | .reg_type = *reg_type, | |
943 | }; | |
31fd8581 | 944 | |
4f9218aa JK |
945 | if (env->ops->is_valid_access && |
946 | env->ops->is_valid_access(off, size, t, &info)) { | |
f96da094 DB |
947 | /* A non zero info.ctx_field_size indicates that this field is a |
948 | * candidate for later verifier transformation to load the whole | |
949 | * field and then apply a mask when accessed with a narrower | |
950 | * access than actual ctx access size. A zero info.ctx_field_size | |
951 | * will only allow for whole field access and rejects any other | |
952 | * type of narrower access. | |
31fd8581 | 953 | */ |
23994631 | 954 | *reg_type = info.reg_type; |
31fd8581 | 955 | |
4f9218aa | 956 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
957 | /* remember the offset of last byte accessed in ctx */ |
958 | if (env->prog->aux->max_ctx_offset < off + size) | |
959 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 960 | return 0; |
32bbe007 | 961 | } |
17a52670 | 962 | |
61bd5218 | 963 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
964 | return -EACCES; |
965 | } | |
966 | ||
4cabc5b1 DB |
967 | static bool __is_pointer_value(bool allow_ptr_leaks, |
968 | const struct bpf_reg_state *reg) | |
1be7f75d | 969 | { |
4cabc5b1 | 970 | if (allow_ptr_leaks) |
1be7f75d AS |
971 | return false; |
972 | ||
f1174f77 | 973 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
974 | } |
975 | ||
4cabc5b1 DB |
976 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
977 | { | |
638f5b90 | 978 | return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno); |
4cabc5b1 DB |
979 | } |
980 | ||
61bd5218 JK |
981 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
982 | const struct bpf_reg_state *reg, | |
d1174416 | 983 | int off, int size, bool strict) |
969bf05e | 984 | { |
f1174f77 | 985 | struct tnum reg_off; |
e07b98d9 | 986 | int ip_align; |
d1174416 DM |
987 | |
988 | /* Byte size accesses are always allowed. */ | |
989 | if (!strict || size == 1) | |
990 | return 0; | |
991 | ||
e4eda884 DM |
992 | /* For platforms that do not have a Kconfig enabling |
993 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
994 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
995 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
996 | * to this code only in strict mode where we want to emulate | |
997 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
998 | * unconditional IP align value of '2'. | |
e07b98d9 | 999 | */ |
e4eda884 | 1000 | ip_align = 2; |
f1174f77 EC |
1001 | |
1002 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1003 | if (!tnum_is_aligned(reg_off, size)) { | |
1004 | char tn_buf[48]; | |
1005 | ||
1006 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1007 | verbose(env, |
1008 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1009 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1010 | return -EACCES; |
1011 | } | |
79adffcd | 1012 | |
969bf05e AS |
1013 | return 0; |
1014 | } | |
1015 | ||
61bd5218 JK |
1016 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1017 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1018 | const char *pointer_desc, |
1019 | int off, int size, bool strict) | |
79adffcd | 1020 | { |
f1174f77 EC |
1021 | struct tnum reg_off; |
1022 | ||
1023 | /* Byte size accesses are always allowed. */ | |
1024 | if (!strict || size == 1) | |
1025 | return 0; | |
1026 | ||
1027 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1028 | if (!tnum_is_aligned(reg_off, size)) { | |
1029 | char tn_buf[48]; | |
1030 | ||
1031 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1032 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1033 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1034 | return -EACCES; |
1035 | } | |
1036 | ||
969bf05e AS |
1037 | return 0; |
1038 | } | |
1039 | ||
e07b98d9 DM |
1040 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
1041 | const struct bpf_reg_state *reg, | |
79adffcd DB |
1042 | int off, int size) |
1043 | { | |
e07b98d9 | 1044 | bool strict = env->strict_alignment; |
f1174f77 | 1045 | const char *pointer_desc = ""; |
d1174416 | 1046 | |
79adffcd DB |
1047 | switch (reg->type) { |
1048 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1049 | case PTR_TO_PACKET_META: |
1050 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1051 | * right in front, treat it the very same way. | |
1052 | */ | |
61bd5218 | 1053 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
f1174f77 EC |
1054 | case PTR_TO_MAP_VALUE: |
1055 | pointer_desc = "value "; | |
1056 | break; | |
1057 | case PTR_TO_CTX: | |
1058 | pointer_desc = "context "; | |
1059 | break; | |
1060 | case PTR_TO_STACK: | |
1061 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
1062 | /* The stack spill tracking logic in check_stack_write() |
1063 | * and check_stack_read() relies on stack accesses being | |
1064 | * aligned. | |
1065 | */ | |
1066 | strict = true; | |
f1174f77 | 1067 | break; |
79adffcd | 1068 | default: |
f1174f77 | 1069 | break; |
79adffcd | 1070 | } |
61bd5218 JK |
1071 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1072 | strict); | |
79adffcd DB |
1073 | } |
1074 | ||
0c17d1d2 JH |
1075 | /* truncate register to smaller size (in bytes) |
1076 | * must be called with size < BPF_REG_SIZE | |
1077 | */ | |
1078 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
1079 | { | |
1080 | u64 mask; | |
1081 | ||
1082 | /* clear high bits in bit representation */ | |
1083 | reg->var_off = tnum_cast(reg->var_off, size); | |
1084 | ||
1085 | /* fix arithmetic bounds */ | |
1086 | mask = ((u64)1 << (size * 8)) - 1; | |
1087 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
1088 | reg->umin_value &= mask; | |
1089 | reg->umax_value &= mask; | |
1090 | } else { | |
1091 | reg->umin_value = 0; | |
1092 | reg->umax_value = mask; | |
1093 | } | |
1094 | reg->smin_value = reg->umin_value; | |
1095 | reg->smax_value = reg->umax_value; | |
1096 | } | |
1097 | ||
17a52670 AS |
1098 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1099 | * if t==write, value_regno is a register which value is stored into memory | |
1100 | * if t==read, value_regno is a register which will receive the value from memory | |
1101 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1102 | * if t==read && value_regno==-1, don't care what we read from memory | |
1103 | */ | |
31fd8581 | 1104 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off, |
17a52670 AS |
1105 | int bpf_size, enum bpf_access_type t, |
1106 | int value_regno) | |
1107 | { | |
638f5b90 AS |
1108 | struct bpf_verifier_state *state = env->cur_state; |
1109 | struct bpf_reg_state *regs = cur_regs(env); | |
1110 | struct bpf_reg_state *reg = regs + regno; | |
17a52670 AS |
1111 | int size, err = 0; |
1112 | ||
1113 | size = bpf_size_to_bytes(bpf_size); | |
1114 | if (size < 0) | |
1115 | return size; | |
1116 | ||
f1174f77 | 1117 | /* alignment checks will add in reg->off themselves */ |
e07b98d9 | 1118 | err = check_ptr_alignment(env, reg, off, size); |
969bf05e AS |
1119 | if (err) |
1120 | return err; | |
17a52670 | 1121 | |
f1174f77 EC |
1122 | /* for access checks, reg->off is just part of off */ |
1123 | off += reg->off; | |
1124 | ||
1125 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1126 | if (t == BPF_WRITE && value_regno >= 0 && |
1127 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1128 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1129 | return -EACCES; |
1130 | } | |
48461135 | 1131 | |
9fd29c08 | 1132 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 1133 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1134 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1135 | |
1a0dc1ac | 1136 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1137 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1138 | |
1be7f75d AS |
1139 | if (t == BPF_WRITE && value_regno >= 0 && |
1140 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1141 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1142 | return -EACCES; |
1143 | } | |
f1174f77 EC |
1144 | /* ctx accesses must be at a fixed offset, so that we can |
1145 | * determine what type of data were returned. | |
1146 | */ | |
28e33f9d | 1147 | if (reg->off) { |
f8ddadc4 DM |
1148 | verbose(env, |
1149 | "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n", | |
28e33f9d JK |
1150 | regno, reg->off, off - reg->off); |
1151 | return -EACCES; | |
1152 | } | |
1153 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
f1174f77 EC |
1154 | char tn_buf[48]; |
1155 | ||
1156 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1157 | verbose(env, |
1158 | "variable ctx access var_off=%s off=%d size=%d", | |
f1174f77 EC |
1159 | tn_buf, off, size); |
1160 | return -EACCES; | |
1161 | } | |
31fd8581 | 1162 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1163 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1164 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1165 | * PTR_TO_PACKET[_META,_END]. In the latter |
1166 | * case, we know the offset is zero. | |
f1174f77 EC |
1167 | */ |
1168 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1169 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1170 | else |
638f5b90 | 1171 | mark_reg_known_zero(env, regs, |
61bd5218 | 1172 | value_regno); |
638f5b90 AS |
1173 | regs[value_regno].id = 0; |
1174 | regs[value_regno].off = 0; | |
1175 | regs[value_regno].range = 0; | |
1176 | regs[value_regno].type = reg_type; | |
969bf05e | 1177 | } |
17a52670 | 1178 | |
f1174f77 EC |
1179 | } else if (reg->type == PTR_TO_STACK) { |
1180 | /* stack accesses must be at a fixed offset, so that we can | |
1181 | * determine what type of data were returned. | |
1182 | * See check_stack_read(). | |
1183 | */ | |
1184 | if (!tnum_is_const(reg->var_off)) { | |
1185 | char tn_buf[48]; | |
1186 | ||
1187 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1188 | verbose(env, "variable stack access var_off=%s off=%d size=%d", |
f1174f77 EC |
1189 | tn_buf, off, size); |
1190 | return -EACCES; | |
1191 | } | |
1192 | off += reg->var_off.value; | |
17a52670 | 1193 | if (off >= 0 || off < -MAX_BPF_STACK) { |
61bd5218 JK |
1194 | verbose(env, "invalid stack off=%d size=%d\n", off, |
1195 | size); | |
17a52670 AS |
1196 | return -EACCES; |
1197 | } | |
8726679a AS |
1198 | |
1199 | if (env->prog->aux->stack_depth < -off) | |
1200 | env->prog->aux->stack_depth = -off; | |
1201 | ||
638f5b90 | 1202 | if (t == BPF_WRITE) |
61bd5218 JK |
1203 | err = check_stack_write(env, state, off, size, |
1204 | value_regno); | |
638f5b90 | 1205 | else |
61bd5218 JK |
1206 | err = check_stack_read(env, state, off, size, |
1207 | value_regno); | |
de8f3a83 | 1208 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 1209 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 1210 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
1211 | return -EACCES; |
1212 | } | |
4acf6c0b BB |
1213 | if (t == BPF_WRITE && value_regno >= 0 && |
1214 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
1215 | verbose(env, "R%d leaks addr into packet\n", |
1216 | value_regno); | |
4acf6c0b BB |
1217 | return -EACCES; |
1218 | } | |
9fd29c08 | 1219 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 1220 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1221 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1222 | } else { |
61bd5218 JK |
1223 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
1224 | reg_type_str[reg->type]); | |
17a52670 AS |
1225 | return -EACCES; |
1226 | } | |
969bf05e | 1227 | |
f1174f77 | 1228 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 1229 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 1230 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 1231 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 1232 | } |
17a52670 AS |
1233 | return err; |
1234 | } | |
1235 | ||
31fd8581 | 1236 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1237 | { |
17a52670 AS |
1238 | int err; |
1239 | ||
1240 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1241 | insn->imm != 0) { | |
61bd5218 | 1242 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
1243 | return -EINVAL; |
1244 | } | |
1245 | ||
1246 | /* check src1 operand */ | |
dc503a8a | 1247 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
1248 | if (err) |
1249 | return err; | |
1250 | ||
1251 | /* check src2 operand */ | |
dc503a8a | 1252 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
1253 | if (err) |
1254 | return err; | |
1255 | ||
6bdf6abc | 1256 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 1257 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
1258 | return -EACCES; |
1259 | } | |
1260 | ||
17a52670 | 1261 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1262 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1263 | BPF_SIZE(insn->code), BPF_READ, -1); |
1264 | if (err) | |
1265 | return err; | |
1266 | ||
1267 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1268 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1269 | BPF_SIZE(insn->code), BPF_WRITE, -1); |
1270 | } | |
1271 | ||
f1174f77 EC |
1272 | /* Does this register contain a constant zero? */ |
1273 | static bool register_is_null(struct bpf_reg_state reg) | |
1274 | { | |
1275 | return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0); | |
1276 | } | |
1277 | ||
17a52670 AS |
1278 | /* when register 'regno' is passed into function that will read 'access_size' |
1279 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1280 | * and all elements of stack are initialized. |
1281 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1282 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1283 | */ |
58e2af8b | 1284 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1285 | int access_size, bool zero_size_allowed, |
1286 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1287 | { |
638f5b90 | 1288 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1289 | struct bpf_reg_state *regs = state->regs; |
638f5b90 | 1290 | int off, i, slot, spi; |
17a52670 | 1291 | |
8e2fe1d9 | 1292 | if (regs[regno].type != PTR_TO_STACK) { |
f1174f77 | 1293 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1294 | if (zero_size_allowed && access_size == 0 && |
f1174f77 | 1295 | register_is_null(regs[regno])) |
8e2fe1d9 DB |
1296 | return 0; |
1297 | ||
61bd5218 | 1298 | verbose(env, "R%d type=%s expected=%s\n", regno, |
8e2fe1d9 DB |
1299 | reg_type_str[regs[regno].type], |
1300 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 1301 | return -EACCES; |
8e2fe1d9 | 1302 | } |
17a52670 | 1303 | |
f1174f77 EC |
1304 | /* Only allow fixed-offset stack reads */ |
1305 | if (!tnum_is_const(regs[regno].var_off)) { | |
1306 | char tn_buf[48]; | |
1307 | ||
1308 | tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off); | |
61bd5218 | 1309 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 | 1310 | regno, tn_buf); |
ea25f914 | 1311 | return -EACCES; |
f1174f77 EC |
1312 | } |
1313 | off = regs[regno].off + regs[regno].var_off.value; | |
17a52670 | 1314 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
9fd29c08 | 1315 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { |
61bd5218 | 1316 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
1317 | regno, off, access_size); |
1318 | return -EACCES; | |
1319 | } | |
1320 | ||
8726679a AS |
1321 | if (env->prog->aux->stack_depth < -off) |
1322 | env->prog->aux->stack_depth = -off; | |
1323 | ||
435faee1 DB |
1324 | if (meta && meta->raw_mode) { |
1325 | meta->access_size = access_size; | |
1326 | meta->regno = regno; | |
1327 | return 0; | |
1328 | } | |
1329 | ||
17a52670 | 1330 | for (i = 0; i < access_size; i++) { |
638f5b90 AS |
1331 | slot = -(off + i) - 1; |
1332 | spi = slot / BPF_REG_SIZE; | |
1333 | if (state->allocated_stack <= slot || | |
1334 | state->stack[spi].slot_type[slot % BPF_REG_SIZE] != | |
1335 | STACK_MISC) { | |
61bd5218 | 1336 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", |
17a52670 AS |
1337 | off, i, access_size); |
1338 | return -EACCES; | |
1339 | } | |
1340 | } | |
1341 | return 0; | |
1342 | } | |
1343 | ||
06c1c049 GB |
1344 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1345 | int access_size, bool zero_size_allowed, | |
1346 | struct bpf_call_arg_meta *meta) | |
1347 | { | |
638f5b90 | 1348 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 1349 | |
f1174f77 | 1350 | switch (reg->type) { |
06c1c049 | 1351 | case PTR_TO_PACKET: |
de8f3a83 | 1352 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
1353 | return check_packet_access(env, regno, reg->off, access_size, |
1354 | zero_size_allowed); | |
06c1c049 | 1355 | case PTR_TO_MAP_VALUE: |
9fd29c08 YS |
1356 | return check_map_access(env, regno, reg->off, access_size, |
1357 | zero_size_allowed); | |
f1174f77 | 1358 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
1359 | return check_stack_boundary(env, regno, access_size, |
1360 | zero_size_allowed, meta); | |
1361 | } | |
1362 | } | |
1363 | ||
58e2af8b | 1364 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1365 | enum bpf_arg_type arg_type, |
1366 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1367 | { |
638f5b90 | 1368 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 1369 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1370 | int err = 0; |
1371 | ||
80f1d68c | 1372 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1373 | return 0; |
1374 | ||
dc503a8a EC |
1375 | err = check_reg_arg(env, regno, SRC_OP); |
1376 | if (err) | |
1377 | return err; | |
17a52670 | 1378 | |
1be7f75d AS |
1379 | if (arg_type == ARG_ANYTHING) { |
1380 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
1381 | verbose(env, "R%d leaks addr into helper function\n", |
1382 | regno); | |
1be7f75d AS |
1383 | return -EACCES; |
1384 | } | |
80f1d68c | 1385 | return 0; |
1be7f75d | 1386 | } |
80f1d68c | 1387 | |
de8f3a83 | 1388 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 1389 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 1390 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
1391 | return -EACCES; |
1392 | } | |
1393 | ||
8e2fe1d9 | 1394 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1395 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1396 | expected_type = PTR_TO_STACK; | |
de8f3a83 DB |
1397 | if (!type_is_pkt_pointer(type) && |
1398 | type != expected_type) | |
6841de8b | 1399 | goto err_type; |
39f19ebb AS |
1400 | } else if (arg_type == ARG_CONST_SIZE || |
1401 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1402 | expected_type = SCALAR_VALUE; |
1403 | if (type != expected_type) | |
6841de8b | 1404 | goto err_type; |
17a52670 AS |
1405 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1406 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1407 | if (type != expected_type) |
1408 | goto err_type; | |
608cd71a AS |
1409 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1410 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1411 | if (type != expected_type) |
1412 | goto err_type; | |
39f19ebb | 1413 | } else if (arg_type == ARG_PTR_TO_MEM || |
db1ac496 | 1414 | arg_type == ARG_PTR_TO_MEM_OR_NULL || |
39f19ebb | 1415 | arg_type == ARG_PTR_TO_UNINIT_MEM) { |
8e2fe1d9 DB |
1416 | expected_type = PTR_TO_STACK; |
1417 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1418 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1419 | * happens during stack boundary checking. |
1420 | */ | |
db1ac496 GB |
1421 | if (register_is_null(*reg) && |
1422 | arg_type == ARG_PTR_TO_MEM_OR_NULL) | |
6841de8b | 1423 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
1424 | else if (!type_is_pkt_pointer(type) && |
1425 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 1426 | type != expected_type) |
6841de8b | 1427 | goto err_type; |
39f19ebb | 1428 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 1429 | } else { |
61bd5218 | 1430 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
1431 | return -EFAULT; |
1432 | } | |
1433 | ||
17a52670 AS |
1434 | if (arg_type == ARG_CONST_MAP_PTR) { |
1435 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1436 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1437 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1438 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1439 | * check that [key, key + map->key_size) are within | |
1440 | * stack limits and initialized | |
1441 | */ | |
33ff9823 | 1442 | if (!meta->map_ptr) { |
17a52670 AS |
1443 | /* in function declaration map_ptr must come before |
1444 | * map_key, so that it's verified and known before | |
1445 | * we have to check map_key here. Otherwise it means | |
1446 | * that kernel subsystem misconfigured verifier | |
1447 | */ | |
61bd5218 | 1448 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
1449 | return -EACCES; |
1450 | } | |
de8f3a83 | 1451 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1452 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1453 | meta->map_ptr->key_size, |
1454 | false); | |
6841de8b AS |
1455 | else |
1456 | err = check_stack_boundary(env, regno, | |
1457 | meta->map_ptr->key_size, | |
1458 | false, NULL); | |
17a52670 AS |
1459 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1460 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1461 | * check [value, value + map->value_size) validity | |
1462 | */ | |
33ff9823 | 1463 | if (!meta->map_ptr) { |
17a52670 | 1464 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 1465 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
1466 | return -EACCES; |
1467 | } | |
de8f3a83 | 1468 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1469 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1470 | meta->map_ptr->value_size, |
1471 | false); | |
6841de8b AS |
1472 | else |
1473 | err = check_stack_boundary(env, regno, | |
1474 | meta->map_ptr->value_size, | |
1475 | false, NULL); | |
39f19ebb AS |
1476 | } else if (arg_type == ARG_CONST_SIZE || |
1477 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1478 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1479 | |
17a52670 AS |
1480 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1481 | * from stack pointer 'buf'. Check it | |
1482 | * note: regno == len, regno - 1 == buf | |
1483 | */ | |
1484 | if (regno == 0) { | |
1485 | /* kernel subsystem misconfigured verifier */ | |
61bd5218 JK |
1486 | verbose(env, |
1487 | "ARG_CONST_SIZE cannot be first argument\n"); | |
17a52670 AS |
1488 | return -EACCES; |
1489 | } | |
06c1c049 | 1490 | |
f1174f77 EC |
1491 | /* The register is SCALAR_VALUE; the access check |
1492 | * happens using its boundaries. | |
06c1c049 | 1493 | */ |
f1174f77 EC |
1494 | |
1495 | if (!tnum_is_const(reg->var_off)) | |
06c1c049 GB |
1496 | /* For unprivileged variable accesses, disable raw |
1497 | * mode so that the program is required to | |
1498 | * initialize all the memory that the helper could | |
1499 | * just partially fill up. | |
1500 | */ | |
1501 | meta = NULL; | |
1502 | ||
b03c9f9f | 1503 | if (reg->smin_value < 0) { |
61bd5218 | 1504 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
1505 | regno); |
1506 | return -EACCES; | |
1507 | } | |
06c1c049 | 1508 | |
b03c9f9f | 1509 | if (reg->umin_value == 0) { |
f1174f77 EC |
1510 | err = check_helper_mem_access(env, regno - 1, 0, |
1511 | zero_size_allowed, | |
1512 | meta); | |
06c1c049 GB |
1513 | if (err) |
1514 | return err; | |
06c1c049 | 1515 | } |
f1174f77 | 1516 | |
b03c9f9f | 1517 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 1518 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
1519 | regno); |
1520 | return -EACCES; | |
1521 | } | |
1522 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 1523 | reg->umax_value, |
f1174f77 | 1524 | zero_size_allowed, meta); |
17a52670 AS |
1525 | } |
1526 | ||
1527 | return err; | |
6841de8b | 1528 | err_type: |
61bd5218 | 1529 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
1530 | reg_type_str[type], reg_type_str[expected_type]); |
1531 | return -EACCES; | |
17a52670 AS |
1532 | } |
1533 | ||
61bd5218 JK |
1534 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
1535 | struct bpf_map *map, int func_id) | |
35578d79 | 1536 | { |
35578d79 KX |
1537 | if (!map) |
1538 | return 0; | |
1539 | ||
6aff67c8 AS |
1540 | /* We need a two way check, first is from map perspective ... */ |
1541 | switch (map->map_type) { | |
1542 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1543 | if (func_id != BPF_FUNC_tail_call) | |
1544 | goto error; | |
1545 | break; | |
1546 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1547 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
1548 | func_id != BPF_FUNC_perf_event_output && |
1549 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
1550 | goto error; |
1551 | break; | |
1552 | case BPF_MAP_TYPE_STACK_TRACE: | |
1553 | if (func_id != BPF_FUNC_get_stackid) | |
1554 | goto error; | |
1555 | break; | |
4ed8ec52 | 1556 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1557 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1558 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1559 | goto error; |
1560 | break; | |
546ac1ff JF |
1561 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
1562 | * allow to be modified from bpf side. So do not allow lookup elements | |
1563 | * for now. | |
1564 | */ | |
1565 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 1566 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
1567 | goto error; |
1568 | break; | |
6710e112 JDB |
1569 | /* Restrict bpf side of cpumap, open when use-cases appear */ |
1570 | case BPF_MAP_TYPE_CPUMAP: | |
1571 | if (func_id != BPF_FUNC_redirect_map) | |
1572 | goto error; | |
1573 | break; | |
56f668df | 1574 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1575 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1576 | if (func_id != BPF_FUNC_map_lookup_elem) |
1577 | goto error; | |
16a43625 | 1578 | break; |
174a79ff JF |
1579 | case BPF_MAP_TYPE_SOCKMAP: |
1580 | if (func_id != BPF_FUNC_sk_redirect_map && | |
1581 | func_id != BPF_FUNC_sock_map_update && | |
1582 | func_id != BPF_FUNC_map_delete_elem) | |
1583 | goto error; | |
1584 | break; | |
6aff67c8 AS |
1585 | default: |
1586 | break; | |
1587 | } | |
1588 | ||
1589 | /* ... and second from the function itself. */ | |
1590 | switch (func_id) { | |
1591 | case BPF_FUNC_tail_call: | |
1592 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1593 | goto error; | |
1594 | break; | |
1595 | case BPF_FUNC_perf_event_read: | |
1596 | case BPF_FUNC_perf_event_output: | |
908432ca | 1597 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
1598 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
1599 | goto error; | |
1600 | break; | |
1601 | case BPF_FUNC_get_stackid: | |
1602 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1603 | goto error; | |
1604 | break; | |
60d20f91 | 1605 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1606 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1607 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1608 | goto error; | |
1609 | break; | |
97f91a7c | 1610 | case BPF_FUNC_redirect_map: |
9c270af3 JDB |
1611 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
1612 | map->map_type != BPF_MAP_TYPE_CPUMAP) | |
97f91a7c JF |
1613 | goto error; |
1614 | break; | |
174a79ff JF |
1615 | case BPF_FUNC_sk_redirect_map: |
1616 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1617 | goto error; | |
1618 | break; | |
1619 | case BPF_FUNC_sock_map_update: | |
1620 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1621 | goto error; | |
1622 | break; | |
6aff67c8 AS |
1623 | default: |
1624 | break; | |
35578d79 KX |
1625 | } |
1626 | ||
1627 | return 0; | |
6aff67c8 | 1628 | error: |
61bd5218 | 1629 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 1630 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 1631 | return -EINVAL; |
35578d79 KX |
1632 | } |
1633 | ||
435faee1 DB |
1634 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1635 | { | |
1636 | int count = 0; | |
1637 | ||
39f19ebb | 1638 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1639 | count++; |
39f19ebb | 1640 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1641 | count++; |
39f19ebb | 1642 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1643 | count++; |
39f19ebb | 1644 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1645 | count++; |
39f19ebb | 1646 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1647 | count++; |
1648 | ||
1649 | return count > 1 ? -EINVAL : 0; | |
1650 | } | |
1651 | ||
de8f3a83 DB |
1652 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
1653 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 1654 | */ |
58e2af8b | 1655 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1656 | { |
638f5b90 | 1657 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1658 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
1659 | int i; |
1660 | ||
1661 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 1662 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 1663 | mark_reg_unknown(env, regs, i); |
969bf05e | 1664 | |
638f5b90 AS |
1665 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
1666 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 1667 | continue; |
638f5b90 | 1668 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 DB |
1669 | if (reg_is_pkt_pointer_any(reg)) |
1670 | __mark_reg_unknown(reg); | |
969bf05e AS |
1671 | } |
1672 | } | |
1673 | ||
81ed18ab | 1674 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1675 | { |
17a52670 | 1676 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 1677 | struct bpf_reg_state *regs; |
33ff9823 | 1678 | struct bpf_call_arg_meta meta; |
969bf05e | 1679 | bool changes_data; |
17a52670 AS |
1680 | int i, err; |
1681 | ||
1682 | /* find function prototype */ | |
1683 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
1684 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
1685 | func_id); | |
17a52670 AS |
1686 | return -EINVAL; |
1687 | } | |
1688 | ||
00176a34 JK |
1689 | if (env->ops->get_func_proto) |
1690 | fn = env->ops->get_func_proto(func_id); | |
17a52670 AS |
1691 | |
1692 | if (!fn) { | |
61bd5218 JK |
1693 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
1694 | func_id); | |
17a52670 AS |
1695 | return -EINVAL; |
1696 | } | |
1697 | ||
1698 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1699 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
61bd5218 | 1700 | verbose(env, "cannot call GPL only function from proprietary program\n"); |
17a52670 AS |
1701 | return -EINVAL; |
1702 | } | |
1703 | ||
04514d13 | 1704 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 1705 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
1706 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
1707 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
1708 | func_id_name(func_id), func_id); | |
1709 | return -EINVAL; | |
1710 | } | |
969bf05e | 1711 | |
33ff9823 | 1712 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1713 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1714 | |
435faee1 DB |
1715 | /* We only support one arg being in raw mode at the moment, which |
1716 | * is sufficient for the helper functions we have right now. | |
1717 | */ | |
1718 | err = check_raw_mode(fn); | |
1719 | if (err) { | |
61bd5218 | 1720 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 1721 | func_id_name(func_id), func_id); |
435faee1 DB |
1722 | return err; |
1723 | } | |
1724 | ||
17a52670 | 1725 | /* check args */ |
33ff9823 | 1726 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1727 | if (err) |
1728 | return err; | |
33ff9823 | 1729 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1730 | if (err) |
1731 | return err; | |
33ff9823 | 1732 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1733 | if (err) |
1734 | return err; | |
33ff9823 | 1735 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1736 | if (err) |
1737 | return err; | |
33ff9823 | 1738 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1739 | if (err) |
1740 | return err; | |
1741 | ||
435faee1 DB |
1742 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1743 | * is inferred from register state. | |
1744 | */ | |
1745 | for (i = 0; i < meta.access_size; i++) { | |
31fd8581 | 1746 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1); |
435faee1 DB |
1747 | if (err) |
1748 | return err; | |
1749 | } | |
1750 | ||
638f5b90 | 1751 | regs = cur_regs(env); |
17a52670 | 1752 | /* reset caller saved regs */ |
dc503a8a | 1753 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 1754 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
1755 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
1756 | } | |
17a52670 | 1757 | |
dc503a8a | 1758 | /* update return register (already marked as written above) */ |
17a52670 | 1759 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 1760 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 1761 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
1762 | } else if (fn->ret_type == RET_VOID) { |
1763 | regs[BPF_REG_0].type = NOT_INIT; | |
1764 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
1765 | struct bpf_insn_aux_data *insn_aux; |
1766 | ||
17a52670 | 1767 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
f1174f77 | 1768 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 1769 | mark_reg_known_zero(env, regs, BPF_REG_0); |
f1174f77 | 1770 | regs[BPF_REG_0].off = 0; |
17a52670 AS |
1771 | /* remember map_ptr, so that check_map_access() |
1772 | * can check 'value_size' boundary of memory access | |
1773 | * to map element returned from bpf_map_lookup_elem() | |
1774 | */ | |
33ff9823 | 1775 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
1776 | verbose(env, |
1777 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
1778 | return -EINVAL; |
1779 | } | |
33ff9823 | 1780 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1781 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
1782 | insn_aux = &env->insn_aux_data[insn_idx]; |
1783 | if (!insn_aux->map_ptr) | |
1784 | insn_aux->map_ptr = meta.map_ptr; | |
1785 | else if (insn_aux->map_ptr != meta.map_ptr) | |
1786 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 1787 | } else { |
61bd5218 | 1788 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 1789 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
1790 | return -EINVAL; |
1791 | } | |
04fd61ab | 1792 | |
61bd5218 | 1793 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
1794 | if (err) |
1795 | return err; | |
04fd61ab | 1796 | |
969bf05e AS |
1797 | if (changes_data) |
1798 | clear_all_pkt_pointers(env); | |
1799 | return 0; | |
1800 | } | |
1801 | ||
b03c9f9f EC |
1802 | static bool signed_add_overflows(s64 a, s64 b) |
1803 | { | |
1804 | /* Do the add in u64, where overflow is well-defined */ | |
1805 | s64 res = (s64)((u64)a + (u64)b); | |
1806 | ||
1807 | if (b < 0) | |
1808 | return res > a; | |
1809 | return res < a; | |
1810 | } | |
1811 | ||
1812 | static bool signed_sub_overflows(s64 a, s64 b) | |
1813 | { | |
1814 | /* Do the sub in u64, where overflow is well-defined */ | |
1815 | s64 res = (s64)((u64)a - (u64)b); | |
1816 | ||
1817 | if (b < 0) | |
1818 | return res < a; | |
1819 | return res > a; | |
969bf05e AS |
1820 | } |
1821 | ||
f1174f77 | 1822 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
1823 | * Caller should also handle BPF_MOV case separately. |
1824 | * If we return -EACCES, caller may want to try again treating pointer as a | |
1825 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
1826 | */ | |
1827 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
1828 | struct bpf_insn *insn, | |
1829 | const struct bpf_reg_state *ptr_reg, | |
1830 | const struct bpf_reg_state *off_reg) | |
969bf05e | 1831 | { |
638f5b90 | 1832 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg; |
f1174f77 | 1833 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
1834 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
1835 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
1836 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
1837 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
969bf05e | 1838 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 1839 | u32 dst = insn->dst_reg; |
969bf05e | 1840 | |
f1174f77 | 1841 | dst_reg = ®s[dst]; |
969bf05e | 1842 | |
b03c9f9f | 1843 | if (WARN_ON_ONCE(known && (smin_val != smax_val))) { |
638f5b90 | 1844 | print_verifier_state(env, env->cur_state); |
61bd5218 JK |
1845 | verbose(env, |
1846 | "verifier internal error: known but bad sbounds\n"); | |
b03c9f9f EC |
1847 | return -EINVAL; |
1848 | } | |
1849 | if (WARN_ON_ONCE(known && (umin_val != umax_val))) { | |
638f5b90 | 1850 | print_verifier_state(env, env->cur_state); |
61bd5218 JK |
1851 | verbose(env, |
1852 | "verifier internal error: known but bad ubounds\n"); | |
f1174f77 EC |
1853 | return -EINVAL; |
1854 | } | |
1855 | ||
1856 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
1857 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
1858 | if (!env->allow_ptr_leaks) | |
61bd5218 JK |
1859 | verbose(env, |
1860 | "R%d 32-bit pointer arithmetic prohibited\n", | |
f1174f77 EC |
1861 | dst); |
1862 | return -EACCES; | |
969bf05e AS |
1863 | } |
1864 | ||
f1174f77 EC |
1865 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
1866 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1867 | verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", |
f1174f77 EC |
1868 | dst); |
1869 | return -EACCES; | |
1870 | } | |
1871 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
1872 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1873 | verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", |
f1174f77 EC |
1874 | dst); |
1875 | return -EACCES; | |
1876 | } | |
1877 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
1878 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1879 | verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", |
f1174f77 EC |
1880 | dst); |
1881 | return -EACCES; | |
1882 | } | |
1883 | ||
1884 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
1885 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 1886 | */ |
f1174f77 EC |
1887 | dst_reg->type = ptr_reg->type; |
1888 | dst_reg->id = ptr_reg->id; | |
969bf05e | 1889 | |
f1174f77 EC |
1890 | switch (opcode) { |
1891 | case BPF_ADD: | |
1892 | /* We can take a fixed offset as long as it doesn't overflow | |
1893 | * the s32 'off' field | |
969bf05e | 1894 | */ |
b03c9f9f EC |
1895 | if (known && (ptr_reg->off + smin_val == |
1896 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 1897 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
1898 | dst_reg->smin_value = smin_ptr; |
1899 | dst_reg->smax_value = smax_ptr; | |
1900 | dst_reg->umin_value = umin_ptr; | |
1901 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 1902 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 1903 | dst_reg->off = ptr_reg->off + smin_val; |
f1174f77 EC |
1904 | dst_reg->range = ptr_reg->range; |
1905 | break; | |
1906 | } | |
f1174f77 EC |
1907 | /* A new variable offset is created. Note that off_reg->off |
1908 | * == 0, since it's a scalar. | |
1909 | * dst_reg gets the pointer type and since some positive | |
1910 | * integer value was added to the pointer, give it a new 'id' | |
1911 | * if it's a PTR_TO_PACKET. | |
1912 | * this creates a new 'base' pointer, off_reg (variable) gets | |
1913 | * added into the variable offset, and we copy the fixed offset | |
1914 | * from ptr_reg. | |
969bf05e | 1915 | */ |
b03c9f9f EC |
1916 | if (signed_add_overflows(smin_ptr, smin_val) || |
1917 | signed_add_overflows(smax_ptr, smax_val)) { | |
1918 | dst_reg->smin_value = S64_MIN; | |
1919 | dst_reg->smax_value = S64_MAX; | |
1920 | } else { | |
1921 | dst_reg->smin_value = smin_ptr + smin_val; | |
1922 | dst_reg->smax_value = smax_ptr + smax_val; | |
1923 | } | |
1924 | if (umin_ptr + umin_val < umin_ptr || | |
1925 | umax_ptr + umax_val < umax_ptr) { | |
1926 | dst_reg->umin_value = 0; | |
1927 | dst_reg->umax_value = U64_MAX; | |
1928 | } else { | |
1929 | dst_reg->umin_value = umin_ptr + umin_val; | |
1930 | dst_reg->umax_value = umax_ptr + umax_val; | |
1931 | } | |
f1174f77 EC |
1932 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
1933 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 1934 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
1935 | dst_reg->id = ++env->id_gen; |
1936 | /* something was added to pkt_ptr, set range to zero */ | |
1937 | dst_reg->range = 0; | |
1938 | } | |
1939 | break; | |
1940 | case BPF_SUB: | |
1941 | if (dst_reg == off_reg) { | |
1942 | /* scalar -= pointer. Creates an unknown scalar */ | |
1943 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1944 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
f1174f77 EC |
1945 | dst); |
1946 | return -EACCES; | |
1947 | } | |
1948 | /* We don't allow subtraction from FP, because (according to | |
1949 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
1950 | * be able to deal with it. | |
969bf05e | 1951 | */ |
f1174f77 EC |
1952 | if (ptr_reg->type == PTR_TO_STACK) { |
1953 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1954 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
f1174f77 EC |
1955 | dst); |
1956 | return -EACCES; | |
1957 | } | |
b03c9f9f EC |
1958 | if (known && (ptr_reg->off - smin_val == |
1959 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 1960 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
1961 | dst_reg->smin_value = smin_ptr; |
1962 | dst_reg->smax_value = smax_ptr; | |
1963 | dst_reg->umin_value = umin_ptr; | |
1964 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
1965 | dst_reg->var_off = ptr_reg->var_off; |
1966 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 1967 | dst_reg->off = ptr_reg->off - smin_val; |
f1174f77 EC |
1968 | dst_reg->range = ptr_reg->range; |
1969 | break; | |
1970 | } | |
f1174f77 EC |
1971 | /* A new variable offset is created. If the subtrahend is known |
1972 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 1973 | */ |
b03c9f9f EC |
1974 | if (signed_sub_overflows(smin_ptr, smax_val) || |
1975 | signed_sub_overflows(smax_ptr, smin_val)) { | |
1976 | /* Overflow possible, we know nothing */ | |
1977 | dst_reg->smin_value = S64_MIN; | |
1978 | dst_reg->smax_value = S64_MAX; | |
1979 | } else { | |
1980 | dst_reg->smin_value = smin_ptr - smax_val; | |
1981 | dst_reg->smax_value = smax_ptr - smin_val; | |
1982 | } | |
1983 | if (umin_ptr < umax_val) { | |
1984 | /* Overflow possible, we know nothing */ | |
1985 | dst_reg->umin_value = 0; | |
1986 | dst_reg->umax_value = U64_MAX; | |
1987 | } else { | |
1988 | /* Cannot overflow (as long as bounds are consistent) */ | |
1989 | dst_reg->umin_value = umin_ptr - umax_val; | |
1990 | dst_reg->umax_value = umax_ptr - umin_val; | |
1991 | } | |
f1174f77 EC |
1992 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
1993 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 1994 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
1995 | dst_reg->id = ++env->id_gen; |
1996 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 1997 | if (smin_val < 0) |
f1174f77 | 1998 | dst_reg->range = 0; |
43188702 | 1999 | } |
f1174f77 EC |
2000 | break; |
2001 | case BPF_AND: | |
2002 | case BPF_OR: | |
2003 | case BPF_XOR: | |
2004 | /* bitwise ops on pointers are troublesome, prohibit for now. | |
2005 | * (However, in principle we could allow some cases, e.g. | |
2006 | * ptr &= ~3 which would reduce min_value by 3.) | |
2007 | */ | |
2008 | if (!env->allow_ptr_leaks) | |
61bd5218 | 2009 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", |
f1174f77 EC |
2010 | dst, bpf_alu_string[opcode >> 4]); |
2011 | return -EACCES; | |
2012 | default: | |
2013 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
2014 | if (!env->allow_ptr_leaks) | |
61bd5218 | 2015 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
f1174f77 EC |
2016 | dst, bpf_alu_string[opcode >> 4]); |
2017 | return -EACCES; | |
43188702 JF |
2018 | } |
2019 | ||
b03c9f9f EC |
2020 | __update_reg_bounds(dst_reg); |
2021 | __reg_deduce_bounds(dst_reg); | |
2022 | __reg_bound_offset(dst_reg); | |
43188702 JF |
2023 | return 0; |
2024 | } | |
2025 | ||
468f6eaf JH |
2026 | /* WARNING: This function does calculations on 64-bit values, but the actual |
2027 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
2028 | * need extra checks in the 32-bit case. | |
2029 | */ | |
f1174f77 EC |
2030 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
2031 | struct bpf_insn *insn, | |
2032 | struct bpf_reg_state *dst_reg, | |
2033 | struct bpf_reg_state src_reg) | |
969bf05e | 2034 | { |
638f5b90 | 2035 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 2036 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2037 | bool src_known, dst_known; |
b03c9f9f EC |
2038 | s64 smin_val, smax_val; |
2039 | u64 umin_val, umax_val; | |
468f6eaf | 2040 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
48461135 | 2041 | |
b03c9f9f EC |
2042 | smin_val = src_reg.smin_value; |
2043 | smax_val = src_reg.smax_value; | |
2044 | umin_val = src_reg.umin_value; | |
2045 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
2046 | src_known = tnum_is_const(src_reg.var_off); |
2047 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 2048 | |
48461135 JB |
2049 | switch (opcode) { |
2050 | case BPF_ADD: | |
b03c9f9f EC |
2051 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
2052 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
2053 | dst_reg->smin_value = S64_MIN; | |
2054 | dst_reg->smax_value = S64_MAX; | |
2055 | } else { | |
2056 | dst_reg->smin_value += smin_val; | |
2057 | dst_reg->smax_value += smax_val; | |
2058 | } | |
2059 | if (dst_reg->umin_value + umin_val < umin_val || | |
2060 | dst_reg->umax_value + umax_val < umax_val) { | |
2061 | dst_reg->umin_value = 0; | |
2062 | dst_reg->umax_value = U64_MAX; | |
2063 | } else { | |
2064 | dst_reg->umin_value += umin_val; | |
2065 | dst_reg->umax_value += umax_val; | |
2066 | } | |
f1174f77 | 2067 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2068 | break; |
2069 | case BPF_SUB: | |
b03c9f9f EC |
2070 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
2071 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
2072 | /* Overflow possible, we know nothing */ | |
2073 | dst_reg->smin_value = S64_MIN; | |
2074 | dst_reg->smax_value = S64_MAX; | |
2075 | } else { | |
2076 | dst_reg->smin_value -= smax_val; | |
2077 | dst_reg->smax_value -= smin_val; | |
2078 | } | |
2079 | if (dst_reg->umin_value < umax_val) { | |
2080 | /* Overflow possible, we know nothing */ | |
2081 | dst_reg->umin_value = 0; | |
2082 | dst_reg->umax_value = U64_MAX; | |
2083 | } else { | |
2084 | /* Cannot overflow (as long as bounds are consistent) */ | |
2085 | dst_reg->umin_value -= umax_val; | |
2086 | dst_reg->umax_value -= umin_val; | |
2087 | } | |
f1174f77 | 2088 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2089 | break; |
2090 | case BPF_MUL: | |
b03c9f9f EC |
2091 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
2092 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 2093 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
2094 | __mark_reg_unbounded(dst_reg); |
2095 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2096 | break; |
2097 | } | |
b03c9f9f EC |
2098 | /* Both values are positive, so we can work with unsigned and |
2099 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 2100 | */ |
b03c9f9f EC |
2101 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
2102 | /* Potential overflow, we know nothing */ | |
2103 | __mark_reg_unbounded(dst_reg); | |
2104 | /* (except what we can learn from the var_off) */ | |
2105 | __update_reg_bounds(dst_reg); | |
2106 | break; | |
2107 | } | |
2108 | dst_reg->umin_value *= umin_val; | |
2109 | dst_reg->umax_value *= umax_val; | |
2110 | if (dst_reg->umax_value > S64_MAX) { | |
2111 | /* Overflow possible, we know nothing */ | |
2112 | dst_reg->smin_value = S64_MIN; | |
2113 | dst_reg->smax_value = S64_MAX; | |
2114 | } else { | |
2115 | dst_reg->smin_value = dst_reg->umin_value; | |
2116 | dst_reg->smax_value = dst_reg->umax_value; | |
2117 | } | |
48461135 JB |
2118 | break; |
2119 | case BPF_AND: | |
f1174f77 | 2120 | if (src_known && dst_known) { |
b03c9f9f EC |
2121 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
2122 | src_reg.var_off.value); | |
f1174f77 EC |
2123 | break; |
2124 | } | |
b03c9f9f EC |
2125 | /* We get our minimum from the var_off, since that's inherently |
2126 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 2127 | */ |
f1174f77 | 2128 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
2129 | dst_reg->umin_value = dst_reg->var_off.value; |
2130 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
2131 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2132 | /* Lose signed bounds when ANDing negative numbers, | |
2133 | * ain't nobody got time for that. | |
2134 | */ | |
2135 | dst_reg->smin_value = S64_MIN; | |
2136 | dst_reg->smax_value = S64_MAX; | |
2137 | } else { | |
2138 | /* ANDing two positives gives a positive, so safe to | |
2139 | * cast result into s64. | |
2140 | */ | |
2141 | dst_reg->smin_value = dst_reg->umin_value; | |
2142 | dst_reg->smax_value = dst_reg->umax_value; | |
2143 | } | |
2144 | /* We may learn something more from the var_off */ | |
2145 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2146 | break; |
2147 | case BPF_OR: | |
2148 | if (src_known && dst_known) { | |
b03c9f9f EC |
2149 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
2150 | src_reg.var_off.value); | |
f1174f77 EC |
2151 | break; |
2152 | } | |
b03c9f9f EC |
2153 | /* We get our maximum from the var_off, and our minimum is the |
2154 | * maximum of the operands' minima | |
f1174f77 EC |
2155 | */ |
2156 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
2157 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
2158 | dst_reg->umax_value = dst_reg->var_off.value | | |
2159 | dst_reg->var_off.mask; | |
2160 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2161 | /* Lose signed bounds when ORing negative numbers, | |
2162 | * ain't nobody got time for that. | |
2163 | */ | |
2164 | dst_reg->smin_value = S64_MIN; | |
2165 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2166 | } else { |
b03c9f9f EC |
2167 | /* ORing two positives gives a positive, so safe to |
2168 | * cast result into s64. | |
2169 | */ | |
2170 | dst_reg->smin_value = dst_reg->umin_value; | |
2171 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 2172 | } |
b03c9f9f EC |
2173 | /* We may learn something more from the var_off */ |
2174 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2175 | break; |
2176 | case BPF_LSH: | |
468f6eaf JH |
2177 | if (umax_val >= insn_bitness) { |
2178 | /* Shifts greater than 31 or 63 are undefined. | |
2179 | * This includes shifts by a negative number. | |
b03c9f9f | 2180 | */ |
61bd5218 | 2181 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2182 | break; |
2183 | } | |
b03c9f9f EC |
2184 | /* We lose all sign bit information (except what we can pick |
2185 | * up from var_off) | |
48461135 | 2186 | */ |
b03c9f9f EC |
2187 | dst_reg->smin_value = S64_MIN; |
2188 | dst_reg->smax_value = S64_MAX; | |
2189 | /* If we might shift our top bit out, then we know nothing */ | |
2190 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
2191 | dst_reg->umin_value = 0; | |
2192 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 2193 | } else { |
b03c9f9f EC |
2194 | dst_reg->umin_value <<= umin_val; |
2195 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 2196 | } |
b03c9f9f EC |
2197 | if (src_known) |
2198 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); | |
2199 | else | |
2200 | dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val); | |
2201 | /* We may learn something more from the var_off */ | |
2202 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2203 | break; |
2204 | case BPF_RSH: | |
468f6eaf JH |
2205 | if (umax_val >= insn_bitness) { |
2206 | /* Shifts greater than 31 or 63 are undefined. | |
2207 | * This includes shifts by a negative number. | |
b03c9f9f | 2208 | */ |
61bd5218 | 2209 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2210 | break; |
2211 | } | |
4374f256 EC |
2212 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
2213 | * be negative, then either: | |
2214 | * 1) src_reg might be zero, so the sign bit of the result is | |
2215 | * unknown, so we lose our signed bounds | |
2216 | * 2) it's known negative, thus the unsigned bounds capture the | |
2217 | * signed bounds | |
2218 | * 3) the signed bounds cross zero, so they tell us nothing | |
2219 | * about the result | |
2220 | * If the value in dst_reg is known nonnegative, then again the | |
2221 | * unsigned bounts capture the signed bounds. | |
2222 | * Thus, in all cases it suffices to blow away our signed bounds | |
2223 | * and rely on inferring new ones from the unsigned bounds and | |
2224 | * var_off of the result. | |
2225 | */ | |
2226 | dst_reg->smin_value = S64_MIN; | |
2227 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2228 | if (src_known) |
b03c9f9f EC |
2229 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, |
2230 | umin_val); | |
f1174f77 | 2231 | else |
b03c9f9f EC |
2232 | dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val); |
2233 | dst_reg->umin_value >>= umax_val; | |
2234 | dst_reg->umax_value >>= umin_val; | |
2235 | /* We may learn something more from the var_off */ | |
2236 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2237 | break; |
2238 | default: | |
61bd5218 | 2239 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
2240 | break; |
2241 | } | |
2242 | ||
468f6eaf JH |
2243 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
2244 | /* 32-bit ALU ops are (32,32)->32 */ | |
2245 | coerce_reg_to_size(dst_reg, 4); | |
2246 | coerce_reg_to_size(&src_reg, 4); | |
2247 | } | |
2248 | ||
b03c9f9f EC |
2249 | __reg_deduce_bounds(dst_reg); |
2250 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
2251 | return 0; |
2252 | } | |
2253 | ||
2254 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
2255 | * and var_off. | |
2256 | */ | |
2257 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
2258 | struct bpf_insn *insn) | |
2259 | { | |
638f5b90 | 2260 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg, *src_reg; |
f1174f77 EC |
2261 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
2262 | u8 opcode = BPF_OP(insn->code); | |
2263 | int rc; | |
2264 | ||
2265 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
2266 | src_reg = NULL; |
2267 | if (dst_reg->type != SCALAR_VALUE) | |
2268 | ptr_reg = dst_reg; | |
2269 | if (BPF_SRC(insn->code) == BPF_X) { | |
2270 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
2271 | if (src_reg->type != SCALAR_VALUE) { |
2272 | if (dst_reg->type != SCALAR_VALUE) { | |
2273 | /* Combining two pointers by any ALU op yields | |
2274 | * an arbitrary scalar. | |
2275 | */ | |
2276 | if (!env->allow_ptr_leaks) { | |
61bd5218 | 2277 | verbose(env, "R%d pointer %s pointer prohibited\n", |
f1174f77 EC |
2278 | insn->dst_reg, |
2279 | bpf_alu_string[opcode >> 4]); | |
2280 | return -EACCES; | |
2281 | } | |
61bd5218 | 2282 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2283 | return 0; |
2284 | } else { | |
2285 | /* scalar += pointer | |
2286 | * This is legal, but we have to reverse our | |
2287 | * src/dest handling in computing the range | |
2288 | */ | |
2289 | rc = adjust_ptr_min_max_vals(env, insn, | |
2290 | src_reg, dst_reg); | |
2291 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2292 | /* scalar += unknown scalar */ | |
2293 | __mark_reg_unknown(&off_reg); | |
2294 | return adjust_scalar_min_max_vals( | |
2295 | env, insn, | |
2296 | dst_reg, off_reg); | |
2297 | } | |
2298 | return rc; | |
2299 | } | |
2300 | } else if (ptr_reg) { | |
2301 | /* pointer += scalar */ | |
2302 | rc = adjust_ptr_min_max_vals(env, insn, | |
2303 | dst_reg, src_reg); | |
2304 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2305 | /* unknown scalar += scalar */ | |
2306 | __mark_reg_unknown(dst_reg); | |
2307 | return adjust_scalar_min_max_vals( | |
2308 | env, insn, dst_reg, *src_reg); | |
2309 | } | |
2310 | return rc; | |
2311 | } | |
2312 | } else { | |
2313 | /* Pretend the src is a reg with a known value, since we only | |
2314 | * need to be able to read from this state. | |
2315 | */ | |
2316 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 2317 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 2318 | src_reg = &off_reg; |
f1174f77 EC |
2319 | if (ptr_reg) { /* pointer += K */ |
2320 | rc = adjust_ptr_min_max_vals(env, insn, | |
2321 | ptr_reg, src_reg); | |
2322 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2323 | /* unknown scalar += K */ | |
2324 | __mark_reg_unknown(dst_reg); | |
2325 | return adjust_scalar_min_max_vals( | |
2326 | env, insn, dst_reg, off_reg); | |
2327 | } | |
2328 | return rc; | |
2329 | } | |
2330 | } | |
2331 | ||
2332 | /* Got here implies adding two SCALAR_VALUEs */ | |
2333 | if (WARN_ON_ONCE(ptr_reg)) { | |
638f5b90 | 2334 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2335 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
2336 | return -EINVAL; |
2337 | } | |
2338 | if (WARN_ON(!src_reg)) { | |
638f5b90 | 2339 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2340 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
2341 | return -EINVAL; |
2342 | } | |
2343 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
2344 | } |
2345 | ||
17a52670 | 2346 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 2347 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2348 | { |
638f5b90 | 2349 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
2350 | u8 opcode = BPF_OP(insn->code); |
2351 | int err; | |
2352 | ||
2353 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
2354 | if (opcode == BPF_NEG) { | |
2355 | if (BPF_SRC(insn->code) != 0 || | |
2356 | insn->src_reg != BPF_REG_0 || | |
2357 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 2358 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
2359 | return -EINVAL; |
2360 | } | |
2361 | } else { | |
2362 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
2363 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
2364 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 2365 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
2366 | return -EINVAL; |
2367 | } | |
2368 | } | |
2369 | ||
2370 | /* check src operand */ | |
dc503a8a | 2371 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2372 | if (err) |
2373 | return err; | |
2374 | ||
1be7f75d | 2375 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 2376 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
2377 | insn->dst_reg); |
2378 | return -EACCES; | |
2379 | } | |
2380 | ||
17a52670 | 2381 | /* check dest operand */ |
dc503a8a | 2382 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2383 | if (err) |
2384 | return err; | |
2385 | ||
2386 | } else if (opcode == BPF_MOV) { | |
2387 | ||
2388 | if (BPF_SRC(insn->code) == BPF_X) { | |
2389 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2390 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2391 | return -EINVAL; |
2392 | } | |
2393 | ||
2394 | /* check src operand */ | |
dc503a8a | 2395 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2396 | if (err) |
2397 | return err; | |
2398 | } else { | |
2399 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2400 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2401 | return -EINVAL; |
2402 | } | |
2403 | } | |
2404 | ||
2405 | /* check dest operand */ | |
dc503a8a | 2406 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2407 | if (err) |
2408 | return err; | |
2409 | ||
2410 | if (BPF_SRC(insn->code) == BPF_X) { | |
2411 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
2412 | /* case: R1 = R2 | |
2413 | * copy register state to dest reg | |
2414 | */ | |
2415 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
8fe2d6cc | 2416 | regs[insn->dst_reg].live |= REG_LIVE_WRITTEN; |
17a52670 | 2417 | } else { |
f1174f77 | 2418 | /* R1 = (u32) R2 */ |
1be7f75d | 2419 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
2420 | verbose(env, |
2421 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
2422 | insn->src_reg); |
2423 | return -EACCES; | |
2424 | } | |
61bd5218 | 2425 | mark_reg_unknown(env, regs, insn->dst_reg); |
0c17d1d2 | 2426 | coerce_reg_to_size(®s[insn->dst_reg], 4); |
17a52670 AS |
2427 | } |
2428 | } else { | |
2429 | /* case: R = imm | |
2430 | * remember the value we stored into this reg | |
2431 | */ | |
f1174f77 | 2432 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
2433 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
2434 | __mark_reg_known(regs + insn->dst_reg, | |
2435 | insn->imm); | |
2436 | } else { | |
2437 | __mark_reg_known(regs + insn->dst_reg, | |
2438 | (u32)insn->imm); | |
2439 | } | |
17a52670 AS |
2440 | } |
2441 | ||
2442 | } else if (opcode > BPF_END) { | |
61bd5218 | 2443 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
2444 | return -EINVAL; |
2445 | ||
2446 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
2447 | ||
17a52670 AS |
2448 | if (BPF_SRC(insn->code) == BPF_X) { |
2449 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2450 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2451 | return -EINVAL; |
2452 | } | |
2453 | /* check src1 operand */ | |
dc503a8a | 2454 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2455 | if (err) |
2456 | return err; | |
2457 | } else { | |
2458 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2459 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2460 | return -EINVAL; |
2461 | } | |
2462 | } | |
2463 | ||
2464 | /* check src2 operand */ | |
dc503a8a | 2465 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2466 | if (err) |
2467 | return err; | |
2468 | ||
2469 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
2470 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 2471 | verbose(env, "div by zero\n"); |
17a52670 AS |
2472 | return -EINVAL; |
2473 | } | |
2474 | ||
229394e8 RV |
2475 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
2476 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
2477 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
2478 | ||
2479 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 2480 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
2481 | return -EINVAL; |
2482 | } | |
2483 | } | |
2484 | ||
1a0dc1ac | 2485 | /* check dest operand */ |
dc503a8a | 2486 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
2487 | if (err) |
2488 | return err; | |
2489 | ||
f1174f77 | 2490 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
2491 | } |
2492 | ||
2493 | return 0; | |
2494 | } | |
2495 | ||
58e2af8b | 2496 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
de8f3a83 | 2497 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 2498 | enum bpf_reg_type type, |
fb2a311a | 2499 | bool range_right_open) |
969bf05e | 2500 | { |
58e2af8b | 2501 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 2502 | u16 new_range; |
969bf05e | 2503 | int i; |
2d2be8ca | 2504 | |
fb2a311a DB |
2505 | if (dst_reg->off < 0 || |
2506 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
2507 | /* This doesn't give us any range */ |
2508 | return; | |
2509 | ||
b03c9f9f EC |
2510 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
2511 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
2512 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
2513 | * than pkt_end, but that's because it's also less than pkt. | |
2514 | */ | |
2515 | return; | |
2516 | ||
fb2a311a DB |
2517 | new_range = dst_reg->off; |
2518 | if (range_right_open) | |
2519 | new_range--; | |
2520 | ||
2521 | /* Examples for register markings: | |
2d2be8ca | 2522 | * |
fb2a311a | 2523 | * pkt_data in dst register: |
2d2be8ca DB |
2524 | * |
2525 | * r2 = r3; | |
2526 | * r2 += 8; | |
2527 | * if (r2 > pkt_end) goto <handle exception> | |
2528 | * <access okay> | |
2529 | * | |
b4e432f1 DB |
2530 | * r2 = r3; |
2531 | * r2 += 8; | |
2532 | * if (r2 < pkt_end) goto <access okay> | |
2533 | * <handle exception> | |
2534 | * | |
2d2be8ca DB |
2535 | * Where: |
2536 | * r2 == dst_reg, pkt_end == src_reg | |
2537 | * r2=pkt(id=n,off=8,r=0) | |
2538 | * r3=pkt(id=n,off=0,r=0) | |
2539 | * | |
fb2a311a | 2540 | * pkt_data in src register: |
2d2be8ca DB |
2541 | * |
2542 | * r2 = r3; | |
2543 | * r2 += 8; | |
2544 | * if (pkt_end >= r2) goto <access okay> | |
2545 | * <handle exception> | |
2546 | * | |
b4e432f1 DB |
2547 | * r2 = r3; |
2548 | * r2 += 8; | |
2549 | * if (pkt_end <= r2) goto <handle exception> | |
2550 | * <access okay> | |
2551 | * | |
2d2be8ca DB |
2552 | * Where: |
2553 | * pkt_end == dst_reg, r2 == src_reg | |
2554 | * r2=pkt(id=n,off=8,r=0) | |
2555 | * r3=pkt(id=n,off=0,r=0) | |
2556 | * | |
2557 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
2558 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
2559 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
2560 | * the check. | |
969bf05e | 2561 | */ |
2d2be8ca | 2562 | |
f1174f77 EC |
2563 | /* If our ids match, then we must have the same max_value. And we |
2564 | * don't care about the other reg's fixed offset, since if it's too big | |
2565 | * the range won't allow anything. | |
2566 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
2567 | */ | |
969bf05e | 2568 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 2569 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 2570 | /* keep the maximum range already checked */ |
fb2a311a | 2571 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 2572 | |
638f5b90 AS |
2573 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2574 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 2575 | continue; |
638f5b90 | 2576 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 | 2577 | if (reg->type == type && reg->id == dst_reg->id) |
b06723da | 2578 | reg->range = max(reg->range, new_range); |
969bf05e AS |
2579 | } |
2580 | } | |
2581 | ||
48461135 JB |
2582 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2583 | * variable register that we are working on, and src_reg is a constant or we're | |
2584 | * simply doing a BPF_K check. | |
f1174f77 | 2585 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
2586 | */ |
2587 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2588 | struct bpf_reg_state *false_reg, u64 val, | |
2589 | u8 opcode) | |
2590 | { | |
f1174f77 EC |
2591 | /* If the dst_reg is a pointer, we can't learn anything about its |
2592 | * variable offset from the compare (unless src_reg were a pointer into | |
2593 | * the same object, but we don't bother with that. | |
2594 | * Since false_reg and true_reg have the same type by construction, we | |
2595 | * only need to check one of them for pointerness. | |
2596 | */ | |
2597 | if (__is_pointer_value(false, false_reg)) | |
2598 | return; | |
4cabc5b1 | 2599 | |
48461135 JB |
2600 | switch (opcode) { |
2601 | case BPF_JEQ: | |
2602 | /* If this is false then we know nothing Jon Snow, but if it is | |
2603 | * true then we know for sure. | |
2604 | */ | |
b03c9f9f | 2605 | __mark_reg_known(true_reg, val); |
48461135 JB |
2606 | break; |
2607 | case BPF_JNE: | |
2608 | /* If this is true we know nothing Jon Snow, but if it is false | |
2609 | * we know the value for sure; | |
2610 | */ | |
b03c9f9f | 2611 | __mark_reg_known(false_reg, val); |
48461135 JB |
2612 | break; |
2613 | case BPF_JGT: | |
b03c9f9f EC |
2614 | false_reg->umax_value = min(false_reg->umax_value, val); |
2615 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2616 | break; | |
48461135 | 2617 | case BPF_JSGT: |
b03c9f9f EC |
2618 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
2619 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 2620 | break; |
b4e432f1 DB |
2621 | case BPF_JLT: |
2622 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2623 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
2624 | break; | |
2625 | case BPF_JSLT: | |
2626 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
2627 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
2628 | break; | |
48461135 | 2629 | case BPF_JGE: |
b03c9f9f EC |
2630 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
2631 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2632 | break; | |
48461135 | 2633 | case BPF_JSGE: |
b03c9f9f EC |
2634 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
2635 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 2636 | break; |
b4e432f1 DB |
2637 | case BPF_JLE: |
2638 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2639 | true_reg->umax_value = min(true_reg->umax_value, val); | |
2640 | break; | |
2641 | case BPF_JSLE: | |
2642 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
2643 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
2644 | break; | |
48461135 JB |
2645 | default: |
2646 | break; | |
2647 | } | |
2648 | ||
b03c9f9f EC |
2649 | __reg_deduce_bounds(false_reg); |
2650 | __reg_deduce_bounds(true_reg); | |
2651 | /* We might have learned some bits from the bounds. */ | |
2652 | __reg_bound_offset(false_reg); | |
2653 | __reg_bound_offset(true_reg); | |
2654 | /* Intersecting with the old var_off might have improved our bounds | |
2655 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2656 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2657 | */ | |
2658 | __update_reg_bounds(false_reg); | |
2659 | __update_reg_bounds(true_reg); | |
48461135 JB |
2660 | } |
2661 | ||
f1174f77 EC |
2662 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
2663 | * the variable reg. | |
48461135 JB |
2664 | */ |
2665 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
2666 | struct bpf_reg_state *false_reg, u64 val, | |
2667 | u8 opcode) | |
2668 | { | |
f1174f77 EC |
2669 | if (__is_pointer_value(false, false_reg)) |
2670 | return; | |
4cabc5b1 | 2671 | |
48461135 JB |
2672 | switch (opcode) { |
2673 | case BPF_JEQ: | |
2674 | /* If this is false then we know nothing Jon Snow, but if it is | |
2675 | * true then we know for sure. | |
2676 | */ | |
b03c9f9f | 2677 | __mark_reg_known(true_reg, val); |
48461135 JB |
2678 | break; |
2679 | case BPF_JNE: | |
2680 | /* If this is true we know nothing Jon Snow, but if it is false | |
2681 | * we know the value for sure; | |
2682 | */ | |
b03c9f9f | 2683 | __mark_reg_known(false_reg, val); |
48461135 JB |
2684 | break; |
2685 | case BPF_JGT: | |
b03c9f9f EC |
2686 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
2687 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2688 | break; | |
48461135 | 2689 | case BPF_JSGT: |
b03c9f9f EC |
2690 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
2691 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 2692 | break; |
b4e432f1 DB |
2693 | case BPF_JLT: |
2694 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2695 | false_reg->umax_value = min(false_reg->umax_value, val); | |
2696 | break; | |
2697 | case BPF_JSLT: | |
2698 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
2699 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
2700 | break; | |
48461135 | 2701 | case BPF_JGE: |
b03c9f9f EC |
2702 | true_reg->umax_value = min(true_reg->umax_value, val); |
2703 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2704 | break; | |
48461135 | 2705 | case BPF_JSGE: |
b03c9f9f EC |
2706 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
2707 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 2708 | break; |
b4e432f1 DB |
2709 | case BPF_JLE: |
2710 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2711 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
2712 | break; | |
2713 | case BPF_JSLE: | |
2714 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
2715 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
2716 | break; | |
48461135 JB |
2717 | default: |
2718 | break; | |
2719 | } | |
2720 | ||
b03c9f9f EC |
2721 | __reg_deduce_bounds(false_reg); |
2722 | __reg_deduce_bounds(true_reg); | |
2723 | /* We might have learned some bits from the bounds. */ | |
2724 | __reg_bound_offset(false_reg); | |
2725 | __reg_bound_offset(true_reg); | |
2726 | /* Intersecting with the old var_off might have improved our bounds | |
2727 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2728 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2729 | */ | |
2730 | __update_reg_bounds(false_reg); | |
2731 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
2732 | } |
2733 | ||
2734 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
2735 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
2736 | struct bpf_reg_state *dst_reg) | |
2737 | { | |
b03c9f9f EC |
2738 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
2739 | dst_reg->umin_value); | |
2740 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
2741 | dst_reg->umax_value); | |
2742 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
2743 | dst_reg->smin_value); | |
2744 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
2745 | dst_reg->smax_value); | |
f1174f77 EC |
2746 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
2747 | dst_reg->var_off); | |
b03c9f9f EC |
2748 | /* We might have learned new bounds from the var_off. */ |
2749 | __update_reg_bounds(src_reg); | |
2750 | __update_reg_bounds(dst_reg); | |
2751 | /* We might have learned something about the sign bit. */ | |
2752 | __reg_deduce_bounds(src_reg); | |
2753 | __reg_deduce_bounds(dst_reg); | |
2754 | /* We might have learned some bits from the bounds. */ | |
2755 | __reg_bound_offset(src_reg); | |
2756 | __reg_bound_offset(dst_reg); | |
2757 | /* Intersecting with the old var_off might have improved our bounds | |
2758 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2759 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2760 | */ | |
2761 | __update_reg_bounds(src_reg); | |
2762 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2763 | } |
2764 | ||
2765 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
2766 | struct bpf_reg_state *true_dst, | |
2767 | struct bpf_reg_state *false_src, | |
2768 | struct bpf_reg_state *false_dst, | |
2769 | u8 opcode) | |
2770 | { | |
2771 | switch (opcode) { | |
2772 | case BPF_JEQ: | |
2773 | __reg_combine_min_max(true_src, true_dst); | |
2774 | break; | |
2775 | case BPF_JNE: | |
2776 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 2777 | break; |
4cabc5b1 | 2778 | } |
48461135 JB |
2779 | } |
2780 | ||
57a09bf0 | 2781 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 2782 | bool is_null) |
57a09bf0 TG |
2783 | { |
2784 | struct bpf_reg_state *reg = ®s[regno]; | |
2785 | ||
2786 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
2787 | /* Old offset (both fixed and variable parts) should |
2788 | * have been known-zero, because we don't allow pointer | |
2789 | * arithmetic on pointers that might be NULL. | |
2790 | */ | |
b03c9f9f EC |
2791 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
2792 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 2793 | reg->off)) { |
b03c9f9f EC |
2794 | __mark_reg_known_zero(reg); |
2795 | reg->off = 0; | |
f1174f77 EC |
2796 | } |
2797 | if (is_null) { | |
2798 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
2799 | } else if (reg->map_ptr->inner_map_meta) { |
2800 | reg->type = CONST_PTR_TO_MAP; | |
2801 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
2802 | } else { | |
f1174f77 | 2803 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 2804 | } |
a08dd0da DB |
2805 | /* We don't need id from this point onwards anymore, thus we |
2806 | * should better reset it, so that state pruning has chances | |
2807 | * to take effect. | |
2808 | */ | |
2809 | reg->id = 0; | |
57a09bf0 TG |
2810 | } |
2811 | } | |
2812 | ||
2813 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
2814 | * be folded together at some point. | |
2815 | */ | |
2816 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
f1174f77 | 2817 | bool is_null) |
57a09bf0 TG |
2818 | { |
2819 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 2820 | u32 id = regs[regno].id; |
57a09bf0 TG |
2821 | int i; |
2822 | ||
2823 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 2824 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 | 2825 | |
638f5b90 AS |
2826 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2827 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
57a09bf0 | 2828 | continue; |
638f5b90 | 2829 | mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null); |
57a09bf0 TG |
2830 | } |
2831 | } | |
2832 | ||
5beca081 DB |
2833 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
2834 | struct bpf_reg_state *dst_reg, | |
2835 | struct bpf_reg_state *src_reg, | |
2836 | struct bpf_verifier_state *this_branch, | |
2837 | struct bpf_verifier_state *other_branch) | |
2838 | { | |
2839 | if (BPF_SRC(insn->code) != BPF_X) | |
2840 | return false; | |
2841 | ||
2842 | switch (BPF_OP(insn->code)) { | |
2843 | case BPF_JGT: | |
2844 | if ((dst_reg->type == PTR_TO_PACKET && | |
2845 | src_reg->type == PTR_TO_PACKET_END) || | |
2846 | (dst_reg->type == PTR_TO_PACKET_META && | |
2847 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2848 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
2849 | find_good_pkt_pointers(this_branch, dst_reg, | |
2850 | dst_reg->type, false); | |
2851 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2852 | src_reg->type == PTR_TO_PACKET) || | |
2853 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2854 | src_reg->type == PTR_TO_PACKET_META)) { | |
2855 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
2856 | find_good_pkt_pointers(other_branch, src_reg, | |
2857 | src_reg->type, true); | |
2858 | } else { | |
2859 | return false; | |
2860 | } | |
2861 | break; | |
2862 | case BPF_JLT: | |
2863 | if ((dst_reg->type == PTR_TO_PACKET && | |
2864 | src_reg->type == PTR_TO_PACKET_END) || | |
2865 | (dst_reg->type == PTR_TO_PACKET_META && | |
2866 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2867 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
2868 | find_good_pkt_pointers(other_branch, dst_reg, | |
2869 | dst_reg->type, true); | |
2870 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2871 | src_reg->type == PTR_TO_PACKET) || | |
2872 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2873 | src_reg->type == PTR_TO_PACKET_META)) { | |
2874 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
2875 | find_good_pkt_pointers(this_branch, src_reg, | |
2876 | src_reg->type, false); | |
2877 | } else { | |
2878 | return false; | |
2879 | } | |
2880 | break; | |
2881 | case BPF_JGE: | |
2882 | if ((dst_reg->type == PTR_TO_PACKET && | |
2883 | src_reg->type == PTR_TO_PACKET_END) || | |
2884 | (dst_reg->type == PTR_TO_PACKET_META && | |
2885 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2886 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
2887 | find_good_pkt_pointers(this_branch, dst_reg, | |
2888 | dst_reg->type, true); | |
2889 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2890 | src_reg->type == PTR_TO_PACKET) || | |
2891 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2892 | src_reg->type == PTR_TO_PACKET_META)) { | |
2893 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
2894 | find_good_pkt_pointers(other_branch, src_reg, | |
2895 | src_reg->type, false); | |
2896 | } else { | |
2897 | return false; | |
2898 | } | |
2899 | break; | |
2900 | case BPF_JLE: | |
2901 | if ((dst_reg->type == PTR_TO_PACKET && | |
2902 | src_reg->type == PTR_TO_PACKET_END) || | |
2903 | (dst_reg->type == PTR_TO_PACKET_META && | |
2904 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2905 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
2906 | find_good_pkt_pointers(other_branch, dst_reg, | |
2907 | dst_reg->type, false); | |
2908 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2909 | src_reg->type == PTR_TO_PACKET) || | |
2910 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2911 | src_reg->type == PTR_TO_PACKET_META)) { | |
2912 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
2913 | find_good_pkt_pointers(this_branch, src_reg, | |
2914 | src_reg->type, true); | |
2915 | } else { | |
2916 | return false; | |
2917 | } | |
2918 | break; | |
2919 | default: | |
2920 | return false; | |
2921 | } | |
2922 | ||
2923 | return true; | |
2924 | } | |
2925 | ||
58e2af8b | 2926 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
2927 | struct bpf_insn *insn, int *insn_idx) |
2928 | { | |
638f5b90 | 2929 | struct bpf_verifier_state *other_branch, *this_branch = env->cur_state; |
58e2af8b | 2930 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; |
17a52670 AS |
2931 | u8 opcode = BPF_OP(insn->code); |
2932 | int err; | |
2933 | ||
b4e432f1 | 2934 | if (opcode > BPF_JSLE) { |
61bd5218 | 2935 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
2936 | return -EINVAL; |
2937 | } | |
2938 | ||
2939 | if (BPF_SRC(insn->code) == BPF_X) { | |
2940 | if (insn->imm != 0) { | |
61bd5218 | 2941 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
2942 | return -EINVAL; |
2943 | } | |
2944 | ||
2945 | /* check src1 operand */ | |
dc503a8a | 2946 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2947 | if (err) |
2948 | return err; | |
1be7f75d AS |
2949 | |
2950 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 2951 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
2952 | insn->src_reg); |
2953 | return -EACCES; | |
2954 | } | |
17a52670 AS |
2955 | } else { |
2956 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 2957 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
2958 | return -EINVAL; |
2959 | } | |
2960 | } | |
2961 | ||
2962 | /* check src2 operand */ | |
dc503a8a | 2963 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2964 | if (err) |
2965 | return err; | |
2966 | ||
1a0dc1ac AS |
2967 | dst_reg = ®s[insn->dst_reg]; |
2968 | ||
17a52670 AS |
2969 | /* detect if R == 0 where R was initialized to zero earlier */ |
2970 | if (BPF_SRC(insn->code) == BPF_K && | |
2971 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
f1174f77 EC |
2972 | dst_reg->type == SCALAR_VALUE && |
2973 | tnum_equals_const(dst_reg->var_off, insn->imm)) { | |
17a52670 AS |
2974 | if (opcode == BPF_JEQ) { |
2975 | /* if (imm == imm) goto pc+off; | |
2976 | * only follow the goto, ignore fall-through | |
2977 | */ | |
2978 | *insn_idx += insn->off; | |
2979 | return 0; | |
2980 | } else { | |
2981 | /* if (imm != imm) goto pc+off; | |
2982 | * only follow fall-through branch, since | |
2983 | * that's where the program will go | |
2984 | */ | |
2985 | return 0; | |
2986 | } | |
2987 | } | |
2988 | ||
2989 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
2990 | if (!other_branch) | |
2991 | return -EFAULT; | |
2992 | ||
48461135 JB |
2993 | /* detect if we are comparing against a constant value so we can adjust |
2994 | * our min/max values for our dst register. | |
f1174f77 EC |
2995 | * this is only legit if both are scalars (or pointers to the same |
2996 | * object, I suppose, but we don't support that right now), because | |
2997 | * otherwise the different base pointers mean the offsets aren't | |
2998 | * comparable. | |
48461135 JB |
2999 | */ |
3000 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
3001 | if (dst_reg->type == SCALAR_VALUE && |
3002 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
3003 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
3004 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
3005 | dst_reg, regs[insn->src_reg].var_off.value, | |
3006 | opcode); | |
3007 | else if (tnum_is_const(dst_reg->var_off)) | |
3008 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
3009 | ®s[insn->src_reg], | |
3010 | dst_reg->var_off.value, opcode); | |
3011 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
3012 | /* Comparing for equality, we can combine knowledge */ | |
3013 | reg_combine_min_max(&other_branch->regs[insn->src_reg], | |
3014 | &other_branch->regs[insn->dst_reg], | |
3015 | ®s[insn->src_reg], | |
3016 | ®s[insn->dst_reg], opcode); | |
3017 | } | |
3018 | } else if (dst_reg->type == SCALAR_VALUE) { | |
48461135 JB |
3019 | reg_set_min_max(&other_branch->regs[insn->dst_reg], |
3020 | dst_reg, insn->imm, opcode); | |
3021 | } | |
3022 | ||
58e2af8b | 3023 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 3024 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
3025 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
3026 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
3027 | /* Mark all identical map registers in each branch as either |
3028 | * safe or unknown depending R == 0 or R != 0 conditional. | |
3029 | */ | |
f1174f77 EC |
3030 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
3031 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
5beca081 DB |
3032 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
3033 | this_branch, other_branch) && | |
3034 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
3035 | verbose(env, "R%d pointer comparison prohibited\n", |
3036 | insn->dst_reg); | |
1be7f75d | 3037 | return -EACCES; |
17a52670 | 3038 | } |
61bd5218 JK |
3039 | if (env->log.level) |
3040 | print_verifier_state(env, this_branch); | |
17a52670 AS |
3041 | return 0; |
3042 | } | |
3043 | ||
0246e64d AS |
3044 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
3045 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
3046 | { | |
3047 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
3048 | ||
3049 | return (struct bpf_map *) (unsigned long) imm64; | |
3050 | } | |
3051 | ||
17a52670 | 3052 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 3053 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3054 | { |
638f5b90 | 3055 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3056 | int err; |
3057 | ||
3058 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 3059 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
3060 | return -EINVAL; |
3061 | } | |
3062 | if (insn->off != 0) { | |
61bd5218 | 3063 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
3064 | return -EINVAL; |
3065 | } | |
3066 | ||
dc503a8a | 3067 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3068 | if (err) |
3069 | return err; | |
3070 | ||
6b173873 | 3071 | if (insn->src_reg == 0) { |
6b173873 JK |
3072 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
3073 | ||
f1174f77 | 3074 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 3075 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 3076 | return 0; |
6b173873 | 3077 | } |
17a52670 AS |
3078 | |
3079 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
3080 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
3081 | ||
3082 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
3083 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
3084 | return 0; | |
3085 | } | |
3086 | ||
96be4325 DB |
3087 | static bool may_access_skb(enum bpf_prog_type type) |
3088 | { | |
3089 | switch (type) { | |
3090 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
3091 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 3092 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
3093 | return true; |
3094 | default: | |
3095 | return false; | |
3096 | } | |
3097 | } | |
3098 | ||
ddd872bc AS |
3099 | /* verify safety of LD_ABS|LD_IND instructions: |
3100 | * - they can only appear in the programs where ctx == skb | |
3101 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
3102 | * preserve R6-R9, and store return value into R0 | |
3103 | * | |
3104 | * Implicit input: | |
3105 | * ctx == skb == R6 == CTX | |
3106 | * | |
3107 | * Explicit input: | |
3108 | * SRC == any register | |
3109 | * IMM == 32-bit immediate | |
3110 | * | |
3111 | * Output: | |
3112 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
3113 | */ | |
58e2af8b | 3114 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 3115 | { |
638f5b90 | 3116 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 3117 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
3118 | int i, err; |
3119 | ||
24701ece | 3120 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 3121 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
3122 | return -EINVAL; |
3123 | } | |
3124 | ||
3125 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 3126 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 3127 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 3128 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
3129 | return -EINVAL; |
3130 | } | |
3131 | ||
3132 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 3133 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
3134 | if (err) |
3135 | return err; | |
3136 | ||
3137 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
61bd5218 JK |
3138 | verbose(env, |
3139 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
3140 | return -EINVAL; |
3141 | } | |
3142 | ||
3143 | if (mode == BPF_IND) { | |
3144 | /* check explicit source operand */ | |
dc503a8a | 3145 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
3146 | if (err) |
3147 | return err; | |
3148 | } | |
3149 | ||
3150 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 3151 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3152 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3153 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3154 | } | |
ddd872bc AS |
3155 | |
3156 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
3157 | * the value fetched from the packet. |
3158 | * Already marked as written above. | |
ddd872bc | 3159 | */ |
61bd5218 | 3160 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
3161 | return 0; |
3162 | } | |
3163 | ||
390ee7e2 AS |
3164 | static int check_return_code(struct bpf_verifier_env *env) |
3165 | { | |
3166 | struct bpf_reg_state *reg; | |
3167 | struct tnum range = tnum_range(0, 1); | |
3168 | ||
3169 | switch (env->prog->type) { | |
3170 | case BPF_PROG_TYPE_CGROUP_SKB: | |
3171 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
3172 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 3173 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
390ee7e2 AS |
3174 | break; |
3175 | default: | |
3176 | return 0; | |
3177 | } | |
3178 | ||
638f5b90 | 3179 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 3180 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 3181 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
3182 | reg_type_str[reg->type]); |
3183 | return -EINVAL; | |
3184 | } | |
3185 | ||
3186 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 3187 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
3188 | if (!tnum_is_unknown(reg->var_off)) { |
3189 | char tn_buf[48]; | |
3190 | ||
3191 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3192 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 3193 | } else { |
61bd5218 | 3194 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 3195 | } |
61bd5218 | 3196 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
3197 | return -EINVAL; |
3198 | } | |
3199 | return 0; | |
3200 | } | |
3201 | ||
475fb78f AS |
3202 | /* non-recursive DFS pseudo code |
3203 | * 1 procedure DFS-iterative(G,v): | |
3204 | * 2 label v as discovered | |
3205 | * 3 let S be a stack | |
3206 | * 4 S.push(v) | |
3207 | * 5 while S is not empty | |
3208 | * 6 t <- S.pop() | |
3209 | * 7 if t is what we're looking for: | |
3210 | * 8 return t | |
3211 | * 9 for all edges e in G.adjacentEdges(t) do | |
3212 | * 10 if edge e is already labelled | |
3213 | * 11 continue with the next edge | |
3214 | * 12 w <- G.adjacentVertex(t,e) | |
3215 | * 13 if vertex w is not discovered and not explored | |
3216 | * 14 label e as tree-edge | |
3217 | * 15 label w as discovered | |
3218 | * 16 S.push(w) | |
3219 | * 17 continue at 5 | |
3220 | * 18 else if vertex w is discovered | |
3221 | * 19 label e as back-edge | |
3222 | * 20 else | |
3223 | * 21 // vertex w is explored | |
3224 | * 22 label e as forward- or cross-edge | |
3225 | * 23 label t as explored | |
3226 | * 24 S.pop() | |
3227 | * | |
3228 | * convention: | |
3229 | * 0x10 - discovered | |
3230 | * 0x11 - discovered and fall-through edge labelled | |
3231 | * 0x12 - discovered and fall-through and branch edges labelled | |
3232 | * 0x20 - explored | |
3233 | */ | |
3234 | ||
3235 | enum { | |
3236 | DISCOVERED = 0x10, | |
3237 | EXPLORED = 0x20, | |
3238 | FALLTHROUGH = 1, | |
3239 | BRANCH = 2, | |
3240 | }; | |
3241 | ||
58e2af8b | 3242 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 3243 | |
475fb78f AS |
3244 | static int *insn_stack; /* stack of insns to process */ |
3245 | static int cur_stack; /* current stack index */ | |
3246 | static int *insn_state; | |
3247 | ||
3248 | /* t, w, e - match pseudo-code above: | |
3249 | * t - index of current instruction | |
3250 | * w - next instruction | |
3251 | * e - edge | |
3252 | */ | |
58e2af8b | 3253 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
3254 | { |
3255 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
3256 | return 0; | |
3257 | ||
3258 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
3259 | return 0; | |
3260 | ||
3261 | if (w < 0 || w >= env->prog->len) { | |
61bd5218 | 3262 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
3263 | return -EINVAL; |
3264 | } | |
3265 | ||
f1bca824 AS |
3266 | if (e == BRANCH) |
3267 | /* mark branch target for state pruning */ | |
3268 | env->explored_states[w] = STATE_LIST_MARK; | |
3269 | ||
475fb78f AS |
3270 | if (insn_state[w] == 0) { |
3271 | /* tree-edge */ | |
3272 | insn_state[t] = DISCOVERED | e; | |
3273 | insn_state[w] = DISCOVERED; | |
3274 | if (cur_stack >= env->prog->len) | |
3275 | return -E2BIG; | |
3276 | insn_stack[cur_stack++] = w; | |
3277 | return 1; | |
3278 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
61bd5218 | 3279 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
3280 | return -EINVAL; |
3281 | } else if (insn_state[w] == EXPLORED) { | |
3282 | /* forward- or cross-edge */ | |
3283 | insn_state[t] = DISCOVERED | e; | |
3284 | } else { | |
61bd5218 | 3285 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
3286 | return -EFAULT; |
3287 | } | |
3288 | return 0; | |
3289 | } | |
3290 | ||
3291 | /* non-recursive depth-first-search to detect loops in BPF program | |
3292 | * loop == back-edge in directed graph | |
3293 | */ | |
58e2af8b | 3294 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
3295 | { |
3296 | struct bpf_insn *insns = env->prog->insnsi; | |
3297 | int insn_cnt = env->prog->len; | |
3298 | int ret = 0; | |
3299 | int i, t; | |
3300 | ||
3301 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3302 | if (!insn_state) | |
3303 | return -ENOMEM; | |
3304 | ||
3305 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3306 | if (!insn_stack) { | |
3307 | kfree(insn_state); | |
3308 | return -ENOMEM; | |
3309 | } | |
3310 | ||
3311 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
3312 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
3313 | cur_stack = 1; | |
3314 | ||
3315 | peek_stack: | |
3316 | if (cur_stack == 0) | |
3317 | goto check_state; | |
3318 | t = insn_stack[cur_stack - 1]; | |
3319 | ||
3320 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
3321 | u8 opcode = BPF_OP(insns[t].code); | |
3322 | ||
3323 | if (opcode == BPF_EXIT) { | |
3324 | goto mark_explored; | |
3325 | } else if (opcode == BPF_CALL) { | |
3326 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3327 | if (ret == 1) | |
3328 | goto peek_stack; | |
3329 | else if (ret < 0) | |
3330 | goto err_free; | |
07016151 DB |
3331 | if (t + 1 < insn_cnt) |
3332 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3333 | } else if (opcode == BPF_JA) { |
3334 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
3335 | ret = -EINVAL; | |
3336 | goto err_free; | |
3337 | } | |
3338 | /* unconditional jump with single edge */ | |
3339 | ret = push_insn(t, t + insns[t].off + 1, | |
3340 | FALLTHROUGH, env); | |
3341 | if (ret == 1) | |
3342 | goto peek_stack; | |
3343 | else if (ret < 0) | |
3344 | goto err_free; | |
f1bca824 AS |
3345 | /* tell verifier to check for equivalent states |
3346 | * after every call and jump | |
3347 | */ | |
c3de6317 AS |
3348 | if (t + 1 < insn_cnt) |
3349 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3350 | } else { |
3351 | /* conditional jump with two edges */ | |
3c2ce60b | 3352 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
3353 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
3354 | if (ret == 1) | |
3355 | goto peek_stack; | |
3356 | else if (ret < 0) | |
3357 | goto err_free; | |
3358 | ||
3359 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
3360 | if (ret == 1) | |
3361 | goto peek_stack; | |
3362 | else if (ret < 0) | |
3363 | goto err_free; | |
3364 | } | |
3365 | } else { | |
3366 | /* all other non-branch instructions with single | |
3367 | * fall-through edge | |
3368 | */ | |
3369 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3370 | if (ret == 1) | |
3371 | goto peek_stack; | |
3372 | else if (ret < 0) | |
3373 | goto err_free; | |
3374 | } | |
3375 | ||
3376 | mark_explored: | |
3377 | insn_state[t] = EXPLORED; | |
3378 | if (cur_stack-- <= 0) { | |
61bd5218 | 3379 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
3380 | ret = -EFAULT; |
3381 | goto err_free; | |
3382 | } | |
3383 | goto peek_stack; | |
3384 | ||
3385 | check_state: | |
3386 | for (i = 0; i < insn_cnt; i++) { | |
3387 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 3388 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
3389 | ret = -EINVAL; |
3390 | goto err_free; | |
3391 | } | |
3392 | } | |
3393 | ret = 0; /* cfg looks good */ | |
3394 | ||
3395 | err_free: | |
3396 | kfree(insn_state); | |
3397 | kfree(insn_stack); | |
3398 | return ret; | |
3399 | } | |
3400 | ||
f1174f77 EC |
3401 | /* check %cur's range satisfies %old's */ |
3402 | static bool range_within(struct bpf_reg_state *old, | |
3403 | struct bpf_reg_state *cur) | |
3404 | { | |
b03c9f9f EC |
3405 | return old->umin_value <= cur->umin_value && |
3406 | old->umax_value >= cur->umax_value && | |
3407 | old->smin_value <= cur->smin_value && | |
3408 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
3409 | } |
3410 | ||
3411 | /* Maximum number of register states that can exist at once */ | |
3412 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
3413 | struct idpair { | |
3414 | u32 old; | |
3415 | u32 cur; | |
3416 | }; | |
3417 | ||
3418 | /* If in the old state two registers had the same id, then they need to have | |
3419 | * the same id in the new state as well. But that id could be different from | |
3420 | * the old state, so we need to track the mapping from old to new ids. | |
3421 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
3422 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
3423 | * regs with a different old id could still have new id 9, we don't care about | |
3424 | * that. | |
3425 | * So we look through our idmap to see if this old id has been seen before. If | |
3426 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 3427 | */ |
f1174f77 | 3428 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 3429 | { |
f1174f77 | 3430 | unsigned int i; |
969bf05e | 3431 | |
f1174f77 EC |
3432 | for (i = 0; i < ID_MAP_SIZE; i++) { |
3433 | if (!idmap[i].old) { | |
3434 | /* Reached an empty slot; haven't seen this id before */ | |
3435 | idmap[i].old = old_id; | |
3436 | idmap[i].cur = cur_id; | |
3437 | return true; | |
3438 | } | |
3439 | if (idmap[i].old == old_id) | |
3440 | return idmap[i].cur == cur_id; | |
3441 | } | |
3442 | /* We ran out of idmap slots, which should be impossible */ | |
3443 | WARN_ON_ONCE(1); | |
3444 | return false; | |
3445 | } | |
3446 | ||
3447 | /* Returns true if (rold safe implies rcur safe) */ | |
1b688a19 EC |
3448 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
3449 | struct idpair *idmap) | |
f1174f77 | 3450 | { |
dc503a8a EC |
3451 | if (!(rold->live & REG_LIVE_READ)) |
3452 | /* explored state didn't use this */ | |
3453 | return true; | |
3454 | ||
3455 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, live)) == 0) | |
969bf05e AS |
3456 | return true; |
3457 | ||
f1174f77 EC |
3458 | if (rold->type == NOT_INIT) |
3459 | /* explored state can't have used this */ | |
969bf05e | 3460 | return true; |
f1174f77 EC |
3461 | if (rcur->type == NOT_INIT) |
3462 | return false; | |
3463 | switch (rold->type) { | |
3464 | case SCALAR_VALUE: | |
3465 | if (rcur->type == SCALAR_VALUE) { | |
3466 | /* new val must satisfy old val knowledge */ | |
3467 | return range_within(rold, rcur) && | |
3468 | tnum_in(rold->var_off, rcur->var_off); | |
3469 | } else { | |
3470 | /* if we knew anything about the old value, we're not | |
3471 | * equal, because we can't know anything about the | |
3472 | * scalar value of the pointer in the new value. | |
3473 | */ | |
b03c9f9f EC |
3474 | return rold->umin_value == 0 && |
3475 | rold->umax_value == U64_MAX && | |
3476 | rold->smin_value == S64_MIN && | |
3477 | rold->smax_value == S64_MAX && | |
f1174f77 EC |
3478 | tnum_is_unknown(rold->var_off); |
3479 | } | |
3480 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
3481 | /* If the new min/max/var_off satisfy the old ones and |
3482 | * everything else matches, we are OK. | |
3483 | * We don't care about the 'id' value, because nothing | |
3484 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
3485 | */ | |
3486 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
3487 | range_within(rold, rcur) && | |
3488 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
3489 | case PTR_TO_MAP_VALUE_OR_NULL: |
3490 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
3491 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
3492 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
3493 | * checked, doing so could have affected others with the same | |
3494 | * id, and we can't check for that because we lost the id when | |
3495 | * we converted to a PTR_TO_MAP_VALUE. | |
3496 | */ | |
3497 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
3498 | return false; | |
3499 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
3500 | return false; | |
3501 | /* Check our ids match any regs they're supposed to */ | |
3502 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 3503 | case PTR_TO_PACKET_META: |
f1174f77 | 3504 | case PTR_TO_PACKET: |
de8f3a83 | 3505 | if (rcur->type != rold->type) |
f1174f77 EC |
3506 | return false; |
3507 | /* We must have at least as much range as the old ptr | |
3508 | * did, so that any accesses which were safe before are | |
3509 | * still safe. This is true even if old range < old off, | |
3510 | * since someone could have accessed through (ptr - k), or | |
3511 | * even done ptr -= k in a register, to get a safe access. | |
3512 | */ | |
3513 | if (rold->range > rcur->range) | |
3514 | return false; | |
3515 | /* If the offsets don't match, we can't trust our alignment; | |
3516 | * nor can we be sure that we won't fall out of range. | |
3517 | */ | |
3518 | if (rold->off != rcur->off) | |
3519 | return false; | |
3520 | /* id relations must be preserved */ | |
3521 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
3522 | return false; | |
3523 | /* new val must satisfy old val knowledge */ | |
3524 | return range_within(rold, rcur) && | |
3525 | tnum_in(rold->var_off, rcur->var_off); | |
3526 | case PTR_TO_CTX: | |
3527 | case CONST_PTR_TO_MAP: | |
3528 | case PTR_TO_STACK: | |
3529 | case PTR_TO_PACKET_END: | |
3530 | /* Only valid matches are exact, which memcmp() above | |
3531 | * would have accepted | |
3532 | */ | |
3533 | default: | |
3534 | /* Don't know what's going on, just say it's not safe */ | |
3535 | return false; | |
3536 | } | |
969bf05e | 3537 | |
f1174f77 EC |
3538 | /* Shouldn't get here; if we do, say it's not safe */ |
3539 | WARN_ON_ONCE(1); | |
969bf05e AS |
3540 | return false; |
3541 | } | |
3542 | ||
638f5b90 AS |
3543 | static bool stacksafe(struct bpf_verifier_state *old, |
3544 | struct bpf_verifier_state *cur, | |
3545 | struct idpair *idmap) | |
3546 | { | |
3547 | int i, spi; | |
3548 | ||
3549 | /* if explored stack has more populated slots than current stack | |
3550 | * such stacks are not equivalent | |
3551 | */ | |
3552 | if (old->allocated_stack > cur->allocated_stack) | |
3553 | return false; | |
3554 | ||
3555 | /* walk slots of the explored stack and ignore any additional | |
3556 | * slots in the current stack, since explored(safe) state | |
3557 | * didn't use them | |
3558 | */ | |
3559 | for (i = 0; i < old->allocated_stack; i++) { | |
3560 | spi = i / BPF_REG_SIZE; | |
3561 | ||
3562 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) | |
3563 | continue; | |
3564 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != | |
3565 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
3566 | /* Ex: old explored (safe) state has STACK_SPILL in | |
3567 | * this stack slot, but current has has STACK_MISC -> | |
3568 | * this verifier states are not equivalent, | |
3569 | * return false to continue verification of this path | |
3570 | */ | |
3571 | return false; | |
3572 | if (i % BPF_REG_SIZE) | |
3573 | continue; | |
3574 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
3575 | continue; | |
3576 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
3577 | &cur->stack[spi].spilled_ptr, | |
3578 | idmap)) | |
3579 | /* when explored and current stack slot are both storing | |
3580 | * spilled registers, check that stored pointers types | |
3581 | * are the same as well. | |
3582 | * Ex: explored safe path could have stored | |
3583 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
3584 | * but current path has stored: | |
3585 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
3586 | * such verifier states are not equivalent. | |
3587 | * return false to continue verification of this path | |
3588 | */ | |
3589 | return false; | |
3590 | } | |
3591 | return true; | |
3592 | } | |
3593 | ||
f1bca824 AS |
3594 | /* compare two verifier states |
3595 | * | |
3596 | * all states stored in state_list are known to be valid, since | |
3597 | * verifier reached 'bpf_exit' instruction through them | |
3598 | * | |
3599 | * this function is called when verifier exploring different branches of | |
3600 | * execution popped from the state stack. If it sees an old state that has | |
3601 | * more strict register state and more strict stack state then this execution | |
3602 | * branch doesn't need to be explored further, since verifier already | |
3603 | * concluded that more strict state leads to valid finish. | |
3604 | * | |
3605 | * Therefore two states are equivalent if register state is more conservative | |
3606 | * and explored stack state is more conservative than the current one. | |
3607 | * Example: | |
3608 | * explored current | |
3609 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
3610 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
3611 | * | |
3612 | * In other words if current stack state (one being explored) has more | |
3613 | * valid slots than old one that already passed validation, it means | |
3614 | * the verifier can stop exploring and conclude that current state is valid too | |
3615 | * | |
3616 | * Similarly with registers. If explored state has register type as invalid | |
3617 | * whereas register type in current state is meaningful, it means that | |
3618 | * the current state will reach 'bpf_exit' instruction safely | |
3619 | */ | |
48461135 JB |
3620 | static bool states_equal(struct bpf_verifier_env *env, |
3621 | struct bpf_verifier_state *old, | |
58e2af8b | 3622 | struct bpf_verifier_state *cur) |
f1bca824 | 3623 | { |
f1174f77 EC |
3624 | struct idpair *idmap; |
3625 | bool ret = false; | |
f1bca824 AS |
3626 | int i; |
3627 | ||
f1174f77 EC |
3628 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
3629 | /* If we failed to allocate the idmap, just say it's not safe */ | |
3630 | if (!idmap) | |
1a0dc1ac | 3631 | return false; |
f1174f77 EC |
3632 | |
3633 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 3634 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 3635 | goto out_free; |
f1bca824 AS |
3636 | } |
3637 | ||
638f5b90 AS |
3638 | if (!stacksafe(old, cur, idmap)) |
3639 | goto out_free; | |
f1174f77 EC |
3640 | ret = true; |
3641 | out_free: | |
3642 | kfree(idmap); | |
3643 | return ret; | |
f1bca824 AS |
3644 | } |
3645 | ||
8e9cd9ce EC |
3646 | /* A write screens off any subsequent reads; but write marks come from the |
3647 | * straight-line code between a state and its parent. When we arrive at a | |
3648 | * jump target (in the first iteration of the propagate_liveness() loop), | |
3649 | * we didn't arrive by the straight-line code, so read marks in state must | |
3650 | * propagate to parent regardless of state's write marks. | |
3651 | */ | |
dc503a8a EC |
3652 | static bool do_propagate_liveness(const struct bpf_verifier_state *state, |
3653 | struct bpf_verifier_state *parent) | |
3654 | { | |
63f45f84 | 3655 | bool writes = parent == state->parent; /* Observe write marks */ |
dc503a8a EC |
3656 | bool touched = false; /* any changes made? */ |
3657 | int i; | |
3658 | ||
3659 | if (!parent) | |
3660 | return touched; | |
3661 | /* Propagate read liveness of registers... */ | |
3662 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
3663 | /* We don't need to worry about FP liveness because it's read-only */ | |
3664 | for (i = 0; i < BPF_REG_FP; i++) { | |
3665 | if (parent->regs[i].live & REG_LIVE_READ) | |
3666 | continue; | |
63f45f84 EC |
3667 | if (writes && (state->regs[i].live & REG_LIVE_WRITTEN)) |
3668 | continue; | |
3669 | if (state->regs[i].live & REG_LIVE_READ) { | |
dc503a8a EC |
3670 | parent->regs[i].live |= REG_LIVE_READ; |
3671 | touched = true; | |
3672 | } | |
3673 | } | |
3674 | /* ... and stack slots */ | |
638f5b90 AS |
3675 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
3676 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3677 | if (parent->stack[i].slot_type[0] != STACK_SPILL) | |
dc503a8a | 3678 | continue; |
638f5b90 | 3679 | if (state->stack[i].slot_type[0] != STACK_SPILL) |
dc503a8a | 3680 | continue; |
638f5b90 | 3681 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
dc503a8a | 3682 | continue; |
638f5b90 AS |
3683 | if (writes && |
3684 | (state->stack[i].spilled_ptr.live & REG_LIVE_WRITTEN)) | |
63f45f84 | 3685 | continue; |
638f5b90 AS |
3686 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) { |
3687 | parent->stack[i].spilled_ptr.live |= REG_LIVE_READ; | |
dc503a8a EC |
3688 | touched = true; |
3689 | } | |
3690 | } | |
3691 | return touched; | |
3692 | } | |
3693 | ||
8e9cd9ce EC |
3694 | /* "parent" is "a state from which we reach the current state", but initially |
3695 | * it is not the state->parent (i.e. "the state whose straight-line code leads | |
3696 | * to the current state"), instead it is the state that happened to arrive at | |
3697 | * a (prunable) equivalent of the current state. See comment above | |
3698 | * do_propagate_liveness() for consequences of this. | |
3699 | * This function is just a more efficient way of calling mark_reg_read() or | |
3700 | * mark_stack_slot_read() on each reg in "parent" that is read in "state", | |
3701 | * though it requires that parent != state->parent in the call arguments. | |
3702 | */ | |
dc503a8a EC |
3703 | static void propagate_liveness(const struct bpf_verifier_state *state, |
3704 | struct bpf_verifier_state *parent) | |
3705 | { | |
3706 | while (do_propagate_liveness(state, parent)) { | |
3707 | /* Something changed, so we need to feed those changes onward */ | |
3708 | state = parent; | |
3709 | parent = state->parent; | |
3710 | } | |
3711 | } | |
3712 | ||
58e2af8b | 3713 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 3714 | { |
58e2af8b JK |
3715 | struct bpf_verifier_state_list *new_sl; |
3716 | struct bpf_verifier_state_list *sl; | |
638f5b90 | 3717 | struct bpf_verifier_state *cur = env->cur_state; |
1969db47 | 3718 | int i, err; |
f1bca824 AS |
3719 | |
3720 | sl = env->explored_states[insn_idx]; | |
3721 | if (!sl) | |
3722 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
3723 | * be doing state search here | |
3724 | */ | |
3725 | return 0; | |
3726 | ||
3727 | while (sl != STATE_LIST_MARK) { | |
638f5b90 | 3728 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 3729 | /* reached equivalent register/stack state, |
dc503a8a EC |
3730 | * prune the search. |
3731 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
3732 | * If we have any write marks in env->cur_state, they |
3733 | * will prevent corresponding reads in the continuation | |
3734 | * from reaching our parent (an explored_state). Our | |
3735 | * own state will get the read marks recorded, but | |
3736 | * they'll be immediately forgotten as we're pruning | |
3737 | * this state and will pop a new one. | |
f1bca824 | 3738 | */ |
638f5b90 | 3739 | propagate_liveness(&sl->state, cur); |
f1bca824 | 3740 | return 1; |
dc503a8a | 3741 | } |
f1bca824 AS |
3742 | sl = sl->next; |
3743 | } | |
3744 | ||
3745 | /* there were no equivalent states, remember current one. | |
3746 | * technically the current state is not proven to be safe yet, | |
3747 | * but it will either reach bpf_exit (which means it's safe) or | |
3748 | * it will be rejected. Since there are no loops, we won't be | |
3749 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
3750 | */ | |
638f5b90 | 3751 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
3752 | if (!new_sl) |
3753 | return -ENOMEM; | |
3754 | ||
3755 | /* add new state to the head of linked list */ | |
1969db47 AS |
3756 | err = copy_verifier_state(&new_sl->state, cur); |
3757 | if (err) { | |
3758 | free_verifier_state(&new_sl->state, false); | |
3759 | kfree(new_sl); | |
3760 | return err; | |
3761 | } | |
f1bca824 AS |
3762 | new_sl->next = env->explored_states[insn_idx]; |
3763 | env->explored_states[insn_idx] = new_sl; | |
dc503a8a | 3764 | /* connect new state to parentage chain */ |
638f5b90 | 3765 | cur->parent = &new_sl->state; |
8e9cd9ce EC |
3766 | /* clear write marks in current state: the writes we did are not writes |
3767 | * our child did, so they don't screen off its reads from us. | |
3768 | * (There are no read marks in current state, because reads always mark | |
3769 | * their parent and current state never has children yet. Only | |
3770 | * explored_states can get read marks.) | |
3771 | */ | |
dc503a8a | 3772 | for (i = 0; i < BPF_REG_FP; i++) |
638f5b90 AS |
3773 | cur->regs[i].live = REG_LIVE_NONE; |
3774 | for (i = 0; i < cur->allocated_stack / BPF_REG_SIZE; i++) | |
3775 | if (cur->stack[i].slot_type[0] == STACK_SPILL) | |
3776 | cur->stack[i].spilled_ptr.live = REG_LIVE_NONE; | |
f1bca824 AS |
3777 | return 0; |
3778 | } | |
3779 | ||
13a27dfc JK |
3780 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
3781 | int insn_idx, int prev_insn_idx) | |
3782 | { | |
ab3f0063 JK |
3783 | if (env->dev_ops && env->dev_ops->insn_hook) |
3784 | return env->dev_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
13a27dfc | 3785 | |
ab3f0063 | 3786 | return 0; |
13a27dfc JK |
3787 | } |
3788 | ||
58e2af8b | 3789 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 3790 | { |
638f5b90 | 3791 | struct bpf_verifier_state *state; |
17a52670 | 3792 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 3793 | struct bpf_reg_state *regs; |
17a52670 AS |
3794 | int insn_cnt = env->prog->len; |
3795 | int insn_idx, prev_insn_idx = 0; | |
3796 | int insn_processed = 0; | |
3797 | bool do_print_state = false; | |
3798 | ||
638f5b90 AS |
3799 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
3800 | if (!state) | |
3801 | return -ENOMEM; | |
3802 | env->cur_state = state; | |
3803 | init_reg_state(env, state->regs); | |
dc503a8a | 3804 | state->parent = NULL; |
17a52670 AS |
3805 | insn_idx = 0; |
3806 | for (;;) { | |
3807 | struct bpf_insn *insn; | |
3808 | u8 class; | |
3809 | int err; | |
3810 | ||
3811 | if (insn_idx >= insn_cnt) { | |
61bd5218 | 3812 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
17a52670 AS |
3813 | insn_idx, insn_cnt); |
3814 | return -EFAULT; | |
3815 | } | |
3816 | ||
3817 | insn = &insns[insn_idx]; | |
3818 | class = BPF_CLASS(insn->code); | |
3819 | ||
07016151 | 3820 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
3821 | verbose(env, |
3822 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
3823 | insn_processed); |
3824 | return -E2BIG; | |
3825 | } | |
3826 | ||
f1bca824 AS |
3827 | err = is_state_visited(env, insn_idx); |
3828 | if (err < 0) | |
3829 | return err; | |
3830 | if (err == 1) { | |
3831 | /* found equivalent state, can prune the search */ | |
61bd5218 | 3832 | if (env->log.level) { |
f1bca824 | 3833 | if (do_print_state) |
61bd5218 | 3834 | verbose(env, "\nfrom %d to %d: safe\n", |
f1bca824 AS |
3835 | prev_insn_idx, insn_idx); |
3836 | else | |
61bd5218 | 3837 | verbose(env, "%d: safe\n", insn_idx); |
f1bca824 AS |
3838 | } |
3839 | goto process_bpf_exit; | |
3840 | } | |
3841 | ||
3c2ce60b DB |
3842 | if (need_resched()) |
3843 | cond_resched(); | |
3844 | ||
61bd5218 JK |
3845 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
3846 | if (env->log.level > 1) | |
3847 | verbose(env, "%d:", insn_idx); | |
c5fc9692 | 3848 | else |
61bd5218 | 3849 | verbose(env, "\nfrom %d to %d:", |
c5fc9692 | 3850 | prev_insn_idx, insn_idx); |
638f5b90 | 3851 | print_verifier_state(env, state); |
17a52670 AS |
3852 | do_print_state = false; |
3853 | } | |
3854 | ||
61bd5218 JK |
3855 | if (env->log.level) { |
3856 | verbose(env, "%d: ", insn_idx); | |
f4ac7e0b JK |
3857 | print_bpf_insn(verbose, env, insn, |
3858 | env->allow_ptr_leaks); | |
17a52670 AS |
3859 | } |
3860 | ||
13a27dfc JK |
3861 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
3862 | if (err) | |
3863 | return err; | |
3864 | ||
638f5b90 | 3865 | regs = cur_regs(env); |
c131187d | 3866 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 3867 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 3868 | err = check_alu_op(env, insn); |
17a52670 AS |
3869 | if (err) |
3870 | return err; | |
3871 | ||
3872 | } else if (class == BPF_LDX) { | |
3df126f3 | 3873 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
3874 | |
3875 | /* check for reserved fields is already done */ | |
3876 | ||
17a52670 | 3877 | /* check src operand */ |
dc503a8a | 3878 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3879 | if (err) |
3880 | return err; | |
3881 | ||
dc503a8a | 3882 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
3883 | if (err) |
3884 | return err; | |
3885 | ||
725f9dcd AS |
3886 | src_reg_type = regs[insn->src_reg].type; |
3887 | ||
17a52670 AS |
3888 | /* check that memory (src_reg + off) is readable, |
3889 | * the state of dst_reg will be updated by this func | |
3890 | */ | |
31fd8581 | 3891 | err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, |
17a52670 AS |
3892 | BPF_SIZE(insn->code), BPF_READ, |
3893 | insn->dst_reg); | |
3894 | if (err) | |
3895 | return err; | |
3896 | ||
3df126f3 JK |
3897 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
3898 | ||
3899 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
3900 | /* saw a valid insn |
3901 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 3902 | * save type to validate intersecting paths |
9bac3d6d | 3903 | */ |
3df126f3 | 3904 | *prev_src_type = src_reg_type; |
9bac3d6d | 3905 | |
3df126f3 | 3906 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 3907 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 3908 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
3909 | /* ABuser program is trying to use the same insn |
3910 | * dst_reg = *(u32*) (src_reg + off) | |
3911 | * with different pointer types: | |
3912 | * src_reg == ctx in one branch and | |
3913 | * src_reg == stack|map in some other branch. | |
3914 | * Reject it. | |
3915 | */ | |
61bd5218 | 3916 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
3917 | return -EINVAL; |
3918 | } | |
3919 | ||
17a52670 | 3920 | } else if (class == BPF_STX) { |
3df126f3 | 3921 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 3922 | |
17a52670 | 3923 | if (BPF_MODE(insn->code) == BPF_XADD) { |
31fd8581 | 3924 | err = check_xadd(env, insn_idx, insn); |
17a52670 AS |
3925 | if (err) |
3926 | return err; | |
3927 | insn_idx++; | |
3928 | continue; | |
3929 | } | |
3930 | ||
17a52670 | 3931 | /* check src1 operand */ |
dc503a8a | 3932 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3933 | if (err) |
3934 | return err; | |
3935 | /* check src2 operand */ | |
dc503a8a | 3936 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3937 | if (err) |
3938 | return err; | |
3939 | ||
d691f9e8 AS |
3940 | dst_reg_type = regs[insn->dst_reg].type; |
3941 | ||
17a52670 | 3942 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 3943 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3944 | BPF_SIZE(insn->code), BPF_WRITE, |
3945 | insn->src_reg); | |
3946 | if (err) | |
3947 | return err; | |
3948 | ||
3df126f3 JK |
3949 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
3950 | ||
3951 | if (*prev_dst_type == NOT_INIT) { | |
3952 | *prev_dst_type = dst_reg_type; | |
3953 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 3954 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 3955 | *prev_dst_type == PTR_TO_CTX)) { |
61bd5218 | 3956 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
3957 | return -EINVAL; |
3958 | } | |
3959 | ||
17a52670 AS |
3960 | } else if (class == BPF_ST) { |
3961 | if (BPF_MODE(insn->code) != BPF_MEM || | |
3962 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 3963 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
3964 | return -EINVAL; |
3965 | } | |
3966 | /* check src operand */ | |
dc503a8a | 3967 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3968 | if (err) |
3969 | return err; | |
3970 | ||
3971 | /* check that memory (dst_reg + off) is writeable */ | |
31fd8581 | 3972 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3973 | BPF_SIZE(insn->code), BPF_WRITE, |
3974 | -1); | |
3975 | if (err) | |
3976 | return err; | |
3977 | ||
3978 | } else if (class == BPF_JMP) { | |
3979 | u8 opcode = BPF_OP(insn->code); | |
3980 | ||
3981 | if (opcode == BPF_CALL) { | |
3982 | if (BPF_SRC(insn->code) != BPF_K || | |
3983 | insn->off != 0 || | |
3984 | insn->src_reg != BPF_REG_0 || | |
3985 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 3986 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
3987 | return -EINVAL; |
3988 | } | |
3989 | ||
81ed18ab | 3990 | err = check_call(env, insn->imm, insn_idx); |
17a52670 AS |
3991 | if (err) |
3992 | return err; | |
3993 | ||
3994 | } else if (opcode == BPF_JA) { | |
3995 | if (BPF_SRC(insn->code) != BPF_K || | |
3996 | insn->imm != 0 || | |
3997 | insn->src_reg != BPF_REG_0 || | |
3998 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 3999 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
4000 | return -EINVAL; |
4001 | } | |
4002 | ||
4003 | insn_idx += insn->off + 1; | |
4004 | continue; | |
4005 | ||
4006 | } else if (opcode == BPF_EXIT) { | |
4007 | if (BPF_SRC(insn->code) != BPF_K || | |
4008 | insn->imm != 0 || | |
4009 | insn->src_reg != BPF_REG_0 || | |
4010 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4011 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
4012 | return -EINVAL; |
4013 | } | |
4014 | ||
4015 | /* eBPF calling convetion is such that R0 is used | |
4016 | * to return the value from eBPF program. | |
4017 | * Make sure that it's readable at this time | |
4018 | * of bpf_exit, which means that program wrote | |
4019 | * something into it earlier | |
4020 | */ | |
dc503a8a | 4021 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
4022 | if (err) |
4023 | return err; | |
4024 | ||
1be7f75d | 4025 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 4026 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
4027 | return -EACCES; |
4028 | } | |
4029 | ||
390ee7e2 AS |
4030 | err = check_return_code(env); |
4031 | if (err) | |
4032 | return err; | |
f1bca824 | 4033 | process_bpf_exit: |
638f5b90 AS |
4034 | err = pop_stack(env, &prev_insn_idx, &insn_idx); |
4035 | if (err < 0) { | |
4036 | if (err != -ENOENT) | |
4037 | return err; | |
17a52670 AS |
4038 | break; |
4039 | } else { | |
4040 | do_print_state = true; | |
4041 | continue; | |
4042 | } | |
4043 | } else { | |
4044 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
4045 | if (err) | |
4046 | return err; | |
4047 | } | |
4048 | } else if (class == BPF_LD) { | |
4049 | u8 mode = BPF_MODE(insn->code); | |
4050 | ||
4051 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
4052 | err = check_ld_abs(env, insn); |
4053 | if (err) | |
4054 | return err; | |
4055 | ||
17a52670 AS |
4056 | } else if (mode == BPF_IMM) { |
4057 | err = check_ld_imm(env, insn); | |
4058 | if (err) | |
4059 | return err; | |
4060 | ||
4061 | insn_idx++; | |
c131187d | 4062 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 4063 | } else { |
61bd5218 | 4064 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
4065 | return -EINVAL; |
4066 | } | |
4067 | } else { | |
61bd5218 | 4068 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
4069 | return -EINVAL; |
4070 | } | |
4071 | ||
4072 | insn_idx++; | |
4073 | } | |
4074 | ||
61bd5218 JK |
4075 | verbose(env, "processed %d insns, stack depth %d\n", insn_processed, |
4076 | env->prog->aux->stack_depth); | |
17a52670 AS |
4077 | return 0; |
4078 | } | |
4079 | ||
56f668df MKL |
4080 | static int check_map_prealloc(struct bpf_map *map) |
4081 | { | |
4082 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
4083 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
4084 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
4085 | !(map->map_flags & BPF_F_NO_PREALLOC); |
4086 | } | |
4087 | ||
61bd5218 JK |
4088 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
4089 | struct bpf_map *map, | |
fdc15d38 AS |
4090 | struct bpf_prog *prog) |
4091 | ||
4092 | { | |
56f668df MKL |
4093 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
4094 | * preallocated hash maps, since doing memory allocation | |
4095 | * in overflow_handler can crash depending on where nmi got | |
4096 | * triggered. | |
4097 | */ | |
4098 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
4099 | if (!check_map_prealloc(map)) { | |
61bd5218 | 4100 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
4101 | return -EINVAL; |
4102 | } | |
4103 | if (map->inner_map_meta && | |
4104 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 4105 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
4106 | return -EINVAL; |
4107 | } | |
fdc15d38 AS |
4108 | } |
4109 | return 0; | |
4110 | } | |
4111 | ||
0246e64d AS |
4112 | /* look for pseudo eBPF instructions that access map FDs and |
4113 | * replace them with actual map pointers | |
4114 | */ | |
58e2af8b | 4115 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
4116 | { |
4117 | struct bpf_insn *insn = env->prog->insnsi; | |
4118 | int insn_cnt = env->prog->len; | |
fdc15d38 | 4119 | int i, j, err; |
0246e64d | 4120 | |
f1f7714e | 4121 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
4122 | if (err) |
4123 | return err; | |
4124 | ||
0246e64d | 4125 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 4126 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 4127 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 4128 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
4129 | return -EINVAL; |
4130 | } | |
4131 | ||
d691f9e8 AS |
4132 | if (BPF_CLASS(insn->code) == BPF_STX && |
4133 | ((BPF_MODE(insn->code) != BPF_MEM && | |
4134 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 4135 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
4136 | return -EINVAL; |
4137 | } | |
4138 | ||
0246e64d AS |
4139 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
4140 | struct bpf_map *map; | |
4141 | struct fd f; | |
4142 | ||
4143 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
4144 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
4145 | insn[1].off != 0) { | |
61bd5218 | 4146 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
4147 | return -EINVAL; |
4148 | } | |
4149 | ||
4150 | if (insn->src_reg == 0) | |
4151 | /* valid generic load 64-bit imm */ | |
4152 | goto next_insn; | |
4153 | ||
4154 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
4155 | verbose(env, |
4156 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
4157 | return -EINVAL; |
4158 | } | |
4159 | ||
4160 | f = fdget(insn->imm); | |
c2101297 | 4161 | map = __bpf_map_get(f); |
0246e64d | 4162 | if (IS_ERR(map)) { |
61bd5218 | 4163 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 4164 | insn->imm); |
0246e64d AS |
4165 | return PTR_ERR(map); |
4166 | } | |
4167 | ||
61bd5218 | 4168 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
4169 | if (err) { |
4170 | fdput(f); | |
4171 | return err; | |
4172 | } | |
4173 | ||
0246e64d AS |
4174 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
4175 | insn[0].imm = (u32) (unsigned long) map; | |
4176 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
4177 | ||
4178 | /* check whether we recorded this map already */ | |
4179 | for (j = 0; j < env->used_map_cnt; j++) | |
4180 | if (env->used_maps[j] == map) { | |
4181 | fdput(f); | |
4182 | goto next_insn; | |
4183 | } | |
4184 | ||
4185 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
4186 | fdput(f); | |
4187 | return -E2BIG; | |
4188 | } | |
4189 | ||
0246e64d AS |
4190 | /* hold the map. If the program is rejected by verifier, |
4191 | * the map will be released by release_maps() or it | |
4192 | * will be used by the valid program until it's unloaded | |
4193 | * and all maps are released in free_bpf_prog_info() | |
4194 | */ | |
92117d84 AS |
4195 | map = bpf_map_inc(map, false); |
4196 | if (IS_ERR(map)) { | |
4197 | fdput(f); | |
4198 | return PTR_ERR(map); | |
4199 | } | |
4200 | env->used_maps[env->used_map_cnt++] = map; | |
4201 | ||
0246e64d AS |
4202 | fdput(f); |
4203 | next_insn: | |
4204 | insn++; | |
4205 | i++; | |
4206 | } | |
4207 | } | |
4208 | ||
4209 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
4210 | * 'struct bpf_map *' into a register instead of user map_fd. | |
4211 | * These pointers will be used later by verifier to validate map access. | |
4212 | */ | |
4213 | return 0; | |
4214 | } | |
4215 | ||
4216 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 4217 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
4218 | { |
4219 | int i; | |
4220 | ||
4221 | for (i = 0; i < env->used_map_cnt; i++) | |
4222 | bpf_map_put(env->used_maps[i]); | |
4223 | } | |
4224 | ||
4225 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 4226 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
4227 | { |
4228 | struct bpf_insn *insn = env->prog->insnsi; | |
4229 | int insn_cnt = env->prog->len; | |
4230 | int i; | |
4231 | ||
4232 | for (i = 0; i < insn_cnt; i++, insn++) | |
4233 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
4234 | insn->src_reg = 0; | |
4235 | } | |
4236 | ||
8041902d AS |
4237 | /* single env->prog->insni[off] instruction was replaced with the range |
4238 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
4239 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
4240 | */ | |
4241 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
4242 | u32 off, u32 cnt) | |
4243 | { | |
4244 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
c131187d | 4245 | int i; |
8041902d AS |
4246 | |
4247 | if (cnt == 1) | |
4248 | return 0; | |
4249 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
4250 | if (!new_data) | |
4251 | return -ENOMEM; | |
4252 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
4253 | memcpy(new_data + off + cnt - 1, old_data + off, | |
4254 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
c131187d AS |
4255 | for (i = off; i < off + cnt - 1; i++) |
4256 | new_data[i].seen = true; | |
8041902d AS |
4257 | env->insn_aux_data = new_data; |
4258 | vfree(old_data); | |
4259 | return 0; | |
4260 | } | |
4261 | ||
4262 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
4263 | const struct bpf_insn *patch, u32 len) | |
4264 | { | |
4265 | struct bpf_prog *new_prog; | |
4266 | ||
4267 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4268 | if (!new_prog) | |
4269 | return NULL; | |
4270 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
4271 | return NULL; | |
4272 | return new_prog; | |
4273 | } | |
4274 | ||
c131187d AS |
4275 | /* The verifier does more data flow analysis than llvm and will not explore |
4276 | * branches that are dead at run time. Malicious programs can have dead code | |
4277 | * too. Therefore replace all dead at-run-time code with nops. | |
4278 | */ | |
4279 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
4280 | { | |
4281 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
4282 | struct bpf_insn nop = BPF_MOV64_REG(BPF_REG_0, BPF_REG_0); | |
4283 | struct bpf_insn *insn = env->prog->insnsi; | |
4284 | const int insn_cnt = env->prog->len; | |
4285 | int i; | |
4286 | ||
4287 | for (i = 0; i < insn_cnt; i++) { | |
4288 | if (aux_data[i].seen) | |
4289 | continue; | |
4290 | memcpy(insn + i, &nop, sizeof(nop)); | |
4291 | } | |
4292 | } | |
4293 | ||
9bac3d6d AS |
4294 | /* convert load instructions that access fields of 'struct __sk_buff' |
4295 | * into sequence of instructions that access fields of 'struct sk_buff' | |
4296 | */ | |
58e2af8b | 4297 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 4298 | { |
00176a34 | 4299 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 4300 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 4301 | const int insn_cnt = env->prog->len; |
36bbef52 | 4302 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 4303 | struct bpf_prog *new_prog; |
d691f9e8 | 4304 | enum bpf_access_type type; |
f96da094 DB |
4305 | bool is_narrower_load; |
4306 | u32 target_size; | |
9bac3d6d | 4307 | |
36bbef52 DB |
4308 | if (ops->gen_prologue) { |
4309 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
4310 | env->prog); | |
4311 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4312 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
4313 | return -EINVAL; |
4314 | } else if (cnt) { | |
8041902d | 4315 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
4316 | if (!new_prog) |
4317 | return -ENOMEM; | |
8041902d | 4318 | |
36bbef52 | 4319 | env->prog = new_prog; |
3df126f3 | 4320 | delta += cnt - 1; |
36bbef52 DB |
4321 | } |
4322 | } | |
4323 | ||
4324 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
4325 | return 0; |
4326 | ||
3df126f3 | 4327 | insn = env->prog->insnsi + delta; |
36bbef52 | 4328 | |
9bac3d6d | 4329 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
4330 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
4331 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
4332 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4333 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 4334 | type = BPF_READ; |
62c7989b DB |
4335 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
4336 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
4337 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4338 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
4339 | type = BPF_WRITE; |
4340 | else | |
9bac3d6d AS |
4341 | continue; |
4342 | ||
8041902d | 4343 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 4344 | continue; |
9bac3d6d | 4345 | |
31fd8581 | 4346 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 4347 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
4348 | |
4349 | /* If the read access is a narrower load of the field, | |
4350 | * convert to a 4/8-byte load, to minimum program type specific | |
4351 | * convert_ctx_access changes. If conversion is successful, | |
4352 | * we will apply proper mask to the result. | |
4353 | */ | |
f96da094 | 4354 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 4355 | if (is_narrower_load) { |
f96da094 DB |
4356 | u32 off = insn->off; |
4357 | u8 size_code; | |
4358 | ||
4359 | if (type == BPF_WRITE) { | |
61bd5218 | 4360 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
4361 | return -EINVAL; |
4362 | } | |
31fd8581 | 4363 | |
f96da094 | 4364 | size_code = BPF_H; |
31fd8581 YS |
4365 | if (ctx_field_size == 4) |
4366 | size_code = BPF_W; | |
4367 | else if (ctx_field_size == 8) | |
4368 | size_code = BPF_DW; | |
f96da094 | 4369 | |
31fd8581 YS |
4370 | insn->off = off & ~(ctx_field_size - 1); |
4371 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
4372 | } | |
f96da094 DB |
4373 | |
4374 | target_size = 0; | |
4375 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
4376 | &target_size); | |
4377 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
4378 | (ctx_field_size && !target_size)) { | |
61bd5218 | 4379 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
4380 | return -EINVAL; |
4381 | } | |
f96da094 DB |
4382 | |
4383 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
4384 | if (ctx_field_size <= 4) |
4385 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4386 | (1 << size * 8) - 1); |
31fd8581 YS |
4387 | else |
4388 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4389 | (1 << size * 8) - 1); |
31fd8581 | 4390 | } |
9bac3d6d | 4391 | |
8041902d | 4392 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
4393 | if (!new_prog) |
4394 | return -ENOMEM; | |
4395 | ||
3df126f3 | 4396 | delta += cnt - 1; |
9bac3d6d AS |
4397 | |
4398 | /* keep walking new program and skip insns we just inserted */ | |
4399 | env->prog = new_prog; | |
3df126f3 | 4400 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
4401 | } |
4402 | ||
4403 | return 0; | |
4404 | } | |
4405 | ||
79741b3b | 4406 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 4407 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
4408 | * |
4409 | * this function is called after eBPF program passed verification | |
4410 | */ | |
79741b3b | 4411 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 4412 | { |
79741b3b AS |
4413 | struct bpf_prog *prog = env->prog; |
4414 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 4415 | const struct bpf_func_proto *fn; |
79741b3b | 4416 | const int insn_cnt = prog->len; |
81ed18ab AS |
4417 | struct bpf_insn insn_buf[16]; |
4418 | struct bpf_prog *new_prog; | |
4419 | struct bpf_map *map_ptr; | |
4420 | int i, cnt, delta = 0; | |
e245c5c6 | 4421 | |
79741b3b AS |
4422 | for (i = 0; i < insn_cnt; i++, insn++) { |
4423 | if (insn->code != (BPF_JMP | BPF_CALL)) | |
4424 | continue; | |
e245c5c6 | 4425 | |
79741b3b AS |
4426 | if (insn->imm == BPF_FUNC_get_route_realm) |
4427 | prog->dst_needed = 1; | |
4428 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
4429 | bpf_user_rnd_init_once(); | |
79741b3b | 4430 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
4431 | /* If we tail call into other programs, we |
4432 | * cannot make any assumptions since they can | |
4433 | * be replaced dynamically during runtime in | |
4434 | * the program array. | |
4435 | */ | |
4436 | prog->cb_access = 1; | |
80a58d02 | 4437 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 4438 | |
79741b3b AS |
4439 | /* mark bpf_tail_call as different opcode to avoid |
4440 | * conditional branch in the interpeter for every normal | |
4441 | * call and to prevent accidental JITing by JIT compiler | |
4442 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 4443 | */ |
79741b3b | 4444 | insn->imm = 0; |
71189fa9 | 4445 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
79741b3b AS |
4446 | continue; |
4447 | } | |
e245c5c6 | 4448 | |
89c63074 DB |
4449 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
4450 | * handlers are currently limited to 64 bit only. | |
4451 | */ | |
4452 | if (ebpf_jit_enabled() && BITS_PER_LONG == 64 && | |
4453 | insn->imm == BPF_FUNC_map_lookup_elem) { | |
81ed18ab | 4454 | map_ptr = env->insn_aux_data[i + delta].map_ptr; |
fad73a1a MKL |
4455 | if (map_ptr == BPF_MAP_PTR_POISON || |
4456 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
4457 | goto patch_call_imm; |
4458 | ||
4459 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
4460 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4461 | verbose(env, "bpf verifier is misconfigured\n"); |
81ed18ab AS |
4462 | return -EINVAL; |
4463 | } | |
4464 | ||
4465 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
4466 | cnt); | |
4467 | if (!new_prog) | |
4468 | return -ENOMEM; | |
4469 | ||
4470 | delta += cnt - 1; | |
4471 | ||
4472 | /* keep walking new program and skip insns we just inserted */ | |
4473 | env->prog = prog = new_prog; | |
4474 | insn = new_prog->insnsi + i + delta; | |
4475 | continue; | |
4476 | } | |
4477 | ||
109980b8 | 4478 | if (insn->imm == BPF_FUNC_redirect_map) { |
7c300131 DB |
4479 | /* Note, we cannot use prog directly as imm as subsequent |
4480 | * rewrites would still change the prog pointer. The only | |
4481 | * stable address we can use is aux, which also works with | |
4482 | * prog clones during blinding. | |
4483 | */ | |
4484 | u64 addr = (unsigned long)prog->aux; | |
109980b8 DB |
4485 | struct bpf_insn r4_ld[] = { |
4486 | BPF_LD_IMM64(BPF_REG_4, addr), | |
4487 | *insn, | |
4488 | }; | |
4489 | cnt = ARRAY_SIZE(r4_ld); | |
4490 | ||
4491 | new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt); | |
4492 | if (!new_prog) | |
4493 | return -ENOMEM; | |
4494 | ||
4495 | delta += cnt - 1; | |
4496 | env->prog = prog = new_prog; | |
4497 | insn = new_prog->insnsi + i + delta; | |
4498 | } | |
81ed18ab | 4499 | patch_call_imm: |
00176a34 | 4500 | fn = env->ops->get_func_proto(insn->imm); |
79741b3b AS |
4501 | /* all functions that have prototype and verifier allowed |
4502 | * programs to call them, must be real in-kernel functions | |
4503 | */ | |
4504 | if (!fn->func) { | |
61bd5218 JK |
4505 | verbose(env, |
4506 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
4507 | func_id_name(insn->imm), insn->imm); |
4508 | return -EFAULT; | |
e245c5c6 | 4509 | } |
79741b3b | 4510 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 4511 | } |
e245c5c6 | 4512 | |
79741b3b AS |
4513 | return 0; |
4514 | } | |
e245c5c6 | 4515 | |
58e2af8b | 4516 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 4517 | { |
58e2af8b | 4518 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
4519 | int i; |
4520 | ||
4521 | if (!env->explored_states) | |
4522 | return; | |
4523 | ||
4524 | for (i = 0; i < env->prog->len; i++) { | |
4525 | sl = env->explored_states[i]; | |
4526 | ||
4527 | if (sl) | |
4528 | while (sl != STATE_LIST_MARK) { | |
4529 | sln = sl->next; | |
1969db47 | 4530 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
4531 | kfree(sl); |
4532 | sl = sln; | |
4533 | } | |
4534 | } | |
4535 | ||
4536 | kfree(env->explored_states); | |
4537 | } | |
4538 | ||
9bac3d6d | 4539 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 4540 | { |
58e2af8b | 4541 | struct bpf_verifier_env *env; |
61bd5218 | 4542 | struct bpf_verifer_log *log; |
51580e79 AS |
4543 | int ret = -EINVAL; |
4544 | ||
eba0c929 AB |
4545 | /* no program is valid */ |
4546 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
4547 | return -EINVAL; | |
4548 | ||
58e2af8b | 4549 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
4550 | * allocate/free it every time bpf_check() is called |
4551 | */ | |
58e2af8b | 4552 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
4553 | if (!env) |
4554 | return -ENOMEM; | |
61bd5218 | 4555 | log = &env->log; |
cbd35700 | 4556 | |
3df126f3 JK |
4557 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
4558 | (*prog)->len); | |
4559 | ret = -ENOMEM; | |
4560 | if (!env->insn_aux_data) | |
4561 | goto err_free_env; | |
9bac3d6d | 4562 | env->prog = *prog; |
00176a34 | 4563 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 4564 | |
cbd35700 AS |
4565 | /* grab the mutex to protect few globals used by verifier */ |
4566 | mutex_lock(&bpf_verifier_lock); | |
4567 | ||
4568 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
4569 | /* user requested verbose verifier output | |
4570 | * and supplied buffer to store the verification trace | |
4571 | */ | |
e7bf8249 JK |
4572 | log->level = attr->log_level; |
4573 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
4574 | log->len_total = attr->log_size; | |
cbd35700 AS |
4575 | |
4576 | ret = -EINVAL; | |
e7bf8249 JK |
4577 | /* log attributes have to be sane */ |
4578 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
4579 | !log->level || !log->ubuf) | |
3df126f3 | 4580 | goto err_unlock; |
cbd35700 | 4581 | } |
1ad2f583 DB |
4582 | |
4583 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
4584 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 4585 | env->strict_alignment = true; |
cbd35700 | 4586 | |
ab3f0063 JK |
4587 | if (env->prog->aux->offload) { |
4588 | ret = bpf_prog_offload_verifier_prep(env); | |
4589 | if (ret) | |
4590 | goto err_unlock; | |
4591 | } | |
4592 | ||
0246e64d AS |
4593 | ret = replace_map_fd_with_map_ptr(env); |
4594 | if (ret < 0) | |
4595 | goto skip_full_check; | |
4596 | ||
9bac3d6d | 4597 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 4598 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
4599 | GFP_USER); |
4600 | ret = -ENOMEM; | |
4601 | if (!env->explored_states) | |
4602 | goto skip_full_check; | |
4603 | ||
475fb78f AS |
4604 | ret = check_cfg(env); |
4605 | if (ret < 0) | |
4606 | goto skip_full_check; | |
4607 | ||
1be7f75d AS |
4608 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
4609 | ||
17a52670 | 4610 | ret = do_check(env); |
8c01c4f8 CG |
4611 | if (env->cur_state) { |
4612 | free_verifier_state(env->cur_state, true); | |
4613 | env->cur_state = NULL; | |
4614 | } | |
cbd35700 | 4615 | |
0246e64d | 4616 | skip_full_check: |
638f5b90 | 4617 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 4618 | free_states(env); |
0246e64d | 4619 | |
c131187d AS |
4620 | if (ret == 0) |
4621 | sanitize_dead_code(env); | |
4622 | ||
9bac3d6d AS |
4623 | if (ret == 0) |
4624 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
4625 | ret = convert_ctx_accesses(env); | |
4626 | ||
e245c5c6 | 4627 | if (ret == 0) |
79741b3b | 4628 | ret = fixup_bpf_calls(env); |
e245c5c6 | 4629 | |
a2a7d570 | 4630 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 4631 | ret = -ENOSPC; |
a2a7d570 | 4632 | if (log->level && !log->ubuf) { |
cbd35700 | 4633 | ret = -EFAULT; |
a2a7d570 | 4634 | goto err_release_maps; |
cbd35700 AS |
4635 | } |
4636 | ||
0246e64d AS |
4637 | if (ret == 0 && env->used_map_cnt) { |
4638 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
4639 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
4640 | sizeof(env->used_maps[0]), | |
4641 | GFP_KERNEL); | |
0246e64d | 4642 | |
9bac3d6d | 4643 | if (!env->prog->aux->used_maps) { |
0246e64d | 4644 | ret = -ENOMEM; |
a2a7d570 | 4645 | goto err_release_maps; |
0246e64d AS |
4646 | } |
4647 | ||
9bac3d6d | 4648 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 4649 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 4650 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
4651 | |
4652 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
4653 | * bpf_ld_imm64 instructions | |
4654 | */ | |
4655 | convert_pseudo_ld_imm64(env); | |
4656 | } | |
cbd35700 | 4657 | |
a2a7d570 | 4658 | err_release_maps: |
9bac3d6d | 4659 | if (!env->prog->aux->used_maps) |
0246e64d AS |
4660 | /* if we didn't copy map pointers into bpf_prog_info, release |
4661 | * them now. Otherwise free_bpf_prog_info() will release them. | |
4662 | */ | |
4663 | release_maps(env); | |
9bac3d6d | 4664 | *prog = env->prog; |
3df126f3 | 4665 | err_unlock: |
cbd35700 | 4666 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
4667 | vfree(env->insn_aux_data); |
4668 | err_free_env: | |
4669 | kfree(env); | |
51580e79 AS |
4670 | return ret; |
4671 | } |