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