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, |
17a52670 AS |
802 | int size) |
803 | { | |
638f5b90 AS |
804 | struct bpf_reg_state *regs = cur_regs(env); |
805 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 806 | |
5722569b | 807 | if (off < 0 || size <= 0 || off + size > map->value_size) { |
61bd5218 | 808 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
809 | map->value_size, off, size); |
810 | return -EACCES; | |
811 | } | |
812 | return 0; | |
813 | } | |
814 | ||
f1174f77 EC |
815 | /* check read/write into a map element with possible variable offset */ |
816 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
638f5b90 | 817 | int off, int size) |
dbcfe5f7 | 818 | { |
638f5b90 | 819 | struct bpf_verifier_state *state = env->cur_state; |
dbcfe5f7 GB |
820 | struct bpf_reg_state *reg = &state->regs[regno]; |
821 | int err; | |
822 | ||
f1174f77 EC |
823 | /* We may have adjusted the register to this map value, so we |
824 | * need to try adding each of min_value and max_value to off | |
825 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 826 | */ |
61bd5218 JK |
827 | if (env->log.level) |
828 | print_verifier_state(env, state); | |
dbcfe5f7 GB |
829 | /* The minimum value is only important with signed |
830 | * comparisons where we can't assume the floor of a | |
831 | * value is 0. If we are using signed variables for our | |
832 | * index'es we need to make sure that whatever we use | |
833 | * will have a set floor within our range. | |
834 | */ | |
b03c9f9f | 835 | if (reg->smin_value < 0) { |
61bd5218 | 836 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
837 | regno); |
838 | return -EACCES; | |
839 | } | |
b03c9f9f | 840 | err = __check_map_access(env, regno, reg->smin_value + off, size); |
dbcfe5f7 | 841 | if (err) { |
61bd5218 JK |
842 | verbose(env, "R%d min value is outside of the array range\n", |
843 | regno); | |
dbcfe5f7 GB |
844 | return err; |
845 | } | |
846 | ||
b03c9f9f EC |
847 | /* If we haven't set a max value then we need to bail since we can't be |
848 | * sure we won't do bad things. | |
849 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 850 | */ |
b03c9f9f | 851 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 852 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
853 | regno); |
854 | return -EACCES; | |
855 | } | |
b03c9f9f | 856 | err = __check_map_access(env, regno, reg->umax_value + off, size); |
f1174f77 | 857 | if (err) |
61bd5218 JK |
858 | verbose(env, "R%d max value is outside of the array range\n", |
859 | regno); | |
f1174f77 | 860 | return err; |
dbcfe5f7 GB |
861 | } |
862 | ||
969bf05e AS |
863 | #define MAX_PACKET_OFF 0xffff |
864 | ||
58e2af8b | 865 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
866 | const struct bpf_call_arg_meta *meta, |
867 | enum bpf_access_type t) | |
4acf6c0b | 868 | { |
36bbef52 | 869 | switch (env->prog->type) { |
3a0af8fd TG |
870 | case BPF_PROG_TYPE_LWT_IN: |
871 | case BPF_PROG_TYPE_LWT_OUT: | |
872 | /* dst_input() and dst_output() can't write for now */ | |
873 | if (t == BPF_WRITE) | |
874 | return false; | |
7e57fbb2 | 875 | /* fallthrough */ |
36bbef52 DB |
876 | case BPF_PROG_TYPE_SCHED_CLS: |
877 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 878 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 879 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 880 | case BPF_PROG_TYPE_SK_SKB: |
36bbef52 DB |
881 | if (meta) |
882 | return meta->pkt_access; | |
883 | ||
884 | env->seen_direct_write = true; | |
4acf6c0b BB |
885 | return true; |
886 | default: | |
887 | return false; | |
888 | } | |
889 | } | |
890 | ||
f1174f77 EC |
891 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
892 | int off, int size) | |
969bf05e | 893 | { |
638f5b90 | 894 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 895 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 896 | |
f1174f77 | 897 | if (off < 0 || size <= 0 || (u64)off + size > reg->range) { |
61bd5218 | 898 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 899 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
900 | return -EACCES; |
901 | } | |
902 | return 0; | |
903 | } | |
904 | ||
f1174f77 EC |
905 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
906 | int size) | |
907 | { | |
638f5b90 | 908 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
909 | struct bpf_reg_state *reg = ®s[regno]; |
910 | int err; | |
911 | ||
912 | /* We may have added a variable offset to the packet pointer; but any | |
913 | * reg->range we have comes after that. We are only checking the fixed | |
914 | * offset. | |
915 | */ | |
916 | ||
917 | /* We don't allow negative numbers, because we aren't tracking enough | |
918 | * detail to prove they're safe. | |
919 | */ | |
b03c9f9f | 920 | if (reg->smin_value < 0) { |
61bd5218 | 921 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
922 | regno); |
923 | return -EACCES; | |
924 | } | |
925 | err = __check_packet_access(env, regno, off, size); | |
926 | if (err) { | |
61bd5218 | 927 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
928 | return err; |
929 | } | |
930 | return err; | |
931 | } | |
932 | ||
933 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 934 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 935 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 936 | { |
f96da094 DB |
937 | struct bpf_insn_access_aux info = { |
938 | .reg_type = *reg_type, | |
939 | }; | |
31fd8581 | 940 | |
4f9218aa JK |
941 | if (env->ops->is_valid_access && |
942 | env->ops->is_valid_access(off, size, t, &info)) { | |
f96da094 DB |
943 | /* A non zero info.ctx_field_size indicates that this field is a |
944 | * candidate for later verifier transformation to load the whole | |
945 | * field and then apply a mask when accessed with a narrower | |
946 | * access than actual ctx access size. A zero info.ctx_field_size | |
947 | * will only allow for whole field access and rejects any other | |
948 | * type of narrower access. | |
31fd8581 | 949 | */ |
23994631 | 950 | *reg_type = info.reg_type; |
31fd8581 | 951 | |
4f9218aa JK |
952 | if (env->analyzer_ops) |
953 | return 0; | |
954 | ||
955 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; | |
32bbe007 AS |
956 | /* remember the offset of last byte accessed in ctx */ |
957 | if (env->prog->aux->max_ctx_offset < off + size) | |
958 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 959 | return 0; |
32bbe007 | 960 | } |
17a52670 | 961 | |
61bd5218 | 962 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
963 | return -EACCES; |
964 | } | |
965 | ||
4cabc5b1 DB |
966 | static bool __is_pointer_value(bool allow_ptr_leaks, |
967 | const struct bpf_reg_state *reg) | |
1be7f75d | 968 | { |
4cabc5b1 | 969 | if (allow_ptr_leaks) |
1be7f75d AS |
970 | return false; |
971 | ||
f1174f77 | 972 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
973 | } |
974 | ||
4cabc5b1 DB |
975 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
976 | { | |
638f5b90 | 977 | return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno); |
4cabc5b1 DB |
978 | } |
979 | ||
61bd5218 JK |
980 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
981 | const struct bpf_reg_state *reg, | |
d1174416 | 982 | int off, int size, bool strict) |
969bf05e | 983 | { |
f1174f77 | 984 | struct tnum reg_off; |
e07b98d9 | 985 | int ip_align; |
d1174416 DM |
986 | |
987 | /* Byte size accesses are always allowed. */ | |
988 | if (!strict || size == 1) | |
989 | return 0; | |
990 | ||
e4eda884 DM |
991 | /* For platforms that do not have a Kconfig enabling |
992 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
993 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
994 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
995 | * to this code only in strict mode where we want to emulate | |
996 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
997 | * unconditional IP align value of '2'. | |
e07b98d9 | 998 | */ |
e4eda884 | 999 | ip_align = 2; |
f1174f77 EC |
1000 | |
1001 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1002 | if (!tnum_is_aligned(reg_off, size)) { | |
1003 | char tn_buf[48]; | |
1004 | ||
1005 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1006 | verbose(env, |
1007 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1008 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1009 | return -EACCES; |
1010 | } | |
79adffcd | 1011 | |
969bf05e AS |
1012 | return 0; |
1013 | } | |
1014 | ||
61bd5218 JK |
1015 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1016 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1017 | const char *pointer_desc, |
1018 | int off, int size, bool strict) | |
79adffcd | 1019 | { |
f1174f77 EC |
1020 | struct tnum reg_off; |
1021 | ||
1022 | /* Byte size accesses are always allowed. */ | |
1023 | if (!strict || size == 1) | |
1024 | return 0; | |
1025 | ||
1026 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1027 | if (!tnum_is_aligned(reg_off, size)) { | |
1028 | char tn_buf[48]; | |
1029 | ||
1030 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1031 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1032 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1033 | return -EACCES; |
1034 | } | |
1035 | ||
969bf05e AS |
1036 | return 0; |
1037 | } | |
1038 | ||
e07b98d9 DM |
1039 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
1040 | const struct bpf_reg_state *reg, | |
79adffcd DB |
1041 | int off, int size) |
1042 | { | |
e07b98d9 | 1043 | bool strict = env->strict_alignment; |
f1174f77 | 1044 | const char *pointer_desc = ""; |
d1174416 | 1045 | |
79adffcd DB |
1046 | switch (reg->type) { |
1047 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1048 | case PTR_TO_PACKET_META: |
1049 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1050 | * right in front, treat it the very same way. | |
1051 | */ | |
61bd5218 | 1052 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
f1174f77 EC |
1053 | case PTR_TO_MAP_VALUE: |
1054 | pointer_desc = "value "; | |
1055 | break; | |
1056 | case PTR_TO_CTX: | |
1057 | pointer_desc = "context "; | |
1058 | break; | |
1059 | case PTR_TO_STACK: | |
1060 | pointer_desc = "stack "; | |
1061 | break; | |
79adffcd | 1062 | default: |
f1174f77 | 1063 | break; |
79adffcd | 1064 | } |
61bd5218 JK |
1065 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1066 | strict); | |
79adffcd DB |
1067 | } |
1068 | ||
17a52670 AS |
1069 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1070 | * if t==write, value_regno is a register which value is stored into memory | |
1071 | * if t==read, value_regno is a register which will receive the value from memory | |
1072 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1073 | * if t==read && value_regno==-1, don't care what we read from memory | |
1074 | */ | |
31fd8581 | 1075 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, int off, |
17a52670 AS |
1076 | int bpf_size, enum bpf_access_type t, |
1077 | int value_regno) | |
1078 | { | |
638f5b90 AS |
1079 | struct bpf_verifier_state *state = env->cur_state; |
1080 | struct bpf_reg_state *regs = cur_regs(env); | |
1081 | struct bpf_reg_state *reg = regs + regno; | |
17a52670 AS |
1082 | int size, err = 0; |
1083 | ||
1084 | size = bpf_size_to_bytes(bpf_size); | |
1085 | if (size < 0) | |
1086 | return size; | |
1087 | ||
f1174f77 | 1088 | /* alignment checks will add in reg->off themselves */ |
e07b98d9 | 1089 | err = check_ptr_alignment(env, reg, off, size); |
969bf05e AS |
1090 | if (err) |
1091 | return err; | |
17a52670 | 1092 | |
f1174f77 EC |
1093 | /* for access checks, reg->off is just part of off */ |
1094 | off += reg->off; | |
1095 | ||
1096 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1097 | if (t == BPF_WRITE && value_regno >= 0 && |
1098 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1099 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1100 | return -EACCES; |
1101 | } | |
48461135 | 1102 | |
f1174f77 | 1103 | err = check_map_access(env, regno, off, size); |
17a52670 | 1104 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1105 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1106 | |
1a0dc1ac | 1107 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1108 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1109 | |
1be7f75d AS |
1110 | if (t == BPF_WRITE && value_regno >= 0 && |
1111 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1112 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1113 | return -EACCES; |
1114 | } | |
f1174f77 EC |
1115 | /* ctx accesses must be at a fixed offset, so that we can |
1116 | * determine what type of data were returned. | |
1117 | */ | |
28e33f9d | 1118 | if (reg->off) { |
f8ddadc4 DM |
1119 | verbose(env, |
1120 | "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n", | |
28e33f9d JK |
1121 | regno, reg->off, off - reg->off); |
1122 | return -EACCES; | |
1123 | } | |
1124 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
f1174f77 EC |
1125 | char tn_buf[48]; |
1126 | ||
1127 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1128 | verbose(env, |
1129 | "variable ctx access var_off=%s off=%d size=%d", | |
f1174f77 EC |
1130 | tn_buf, off, size); |
1131 | return -EACCES; | |
1132 | } | |
31fd8581 | 1133 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1134 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1135 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1136 | * PTR_TO_PACKET[_META,_END]. In the latter |
1137 | * case, we know the offset is zero. | |
f1174f77 EC |
1138 | */ |
1139 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1140 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1141 | else |
638f5b90 | 1142 | mark_reg_known_zero(env, regs, |
61bd5218 | 1143 | value_regno); |
638f5b90 AS |
1144 | regs[value_regno].id = 0; |
1145 | regs[value_regno].off = 0; | |
1146 | regs[value_regno].range = 0; | |
1147 | regs[value_regno].type = reg_type; | |
969bf05e | 1148 | } |
17a52670 | 1149 | |
f1174f77 EC |
1150 | } else if (reg->type == PTR_TO_STACK) { |
1151 | /* stack accesses must be at a fixed offset, so that we can | |
1152 | * determine what type of data were returned. | |
1153 | * See check_stack_read(). | |
1154 | */ | |
1155 | if (!tnum_is_const(reg->var_off)) { | |
1156 | char tn_buf[48]; | |
1157 | ||
1158 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1159 | verbose(env, "variable stack access var_off=%s off=%d size=%d", |
f1174f77 EC |
1160 | tn_buf, off, size); |
1161 | return -EACCES; | |
1162 | } | |
1163 | off += reg->var_off.value; | |
17a52670 | 1164 | if (off >= 0 || off < -MAX_BPF_STACK) { |
61bd5218 JK |
1165 | verbose(env, "invalid stack off=%d size=%d\n", off, |
1166 | size); | |
17a52670 AS |
1167 | return -EACCES; |
1168 | } | |
8726679a AS |
1169 | |
1170 | if (env->prog->aux->stack_depth < -off) | |
1171 | env->prog->aux->stack_depth = -off; | |
1172 | ||
638f5b90 | 1173 | if (t == BPF_WRITE) |
61bd5218 JK |
1174 | err = check_stack_write(env, state, off, size, |
1175 | value_regno); | |
638f5b90 | 1176 | else |
61bd5218 JK |
1177 | err = check_stack_read(env, state, off, size, |
1178 | value_regno); | |
de8f3a83 | 1179 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 1180 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 1181 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
1182 | return -EACCES; |
1183 | } | |
4acf6c0b BB |
1184 | if (t == BPF_WRITE && value_regno >= 0 && |
1185 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
1186 | verbose(env, "R%d leaks addr into packet\n", |
1187 | value_regno); | |
4acf6c0b BB |
1188 | return -EACCES; |
1189 | } | |
969bf05e AS |
1190 | err = check_packet_access(env, regno, off, size); |
1191 | if (!err && t == BPF_READ && value_regno >= 0) | |
638f5b90 | 1192 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1193 | } else { |
61bd5218 JK |
1194 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
1195 | reg_type_str[reg->type]); | |
17a52670 AS |
1196 | return -EACCES; |
1197 | } | |
969bf05e | 1198 | |
f1174f77 | 1199 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 1200 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 1201 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
638f5b90 AS |
1202 | regs[value_regno].var_off = |
1203 | tnum_cast(regs[value_regno].var_off, size); | |
1204 | __update_reg_bounds(®s[value_regno]); | |
969bf05e | 1205 | } |
17a52670 AS |
1206 | return err; |
1207 | } | |
1208 | ||
31fd8581 | 1209 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1210 | { |
17a52670 AS |
1211 | int err; |
1212 | ||
1213 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1214 | insn->imm != 0) { | |
61bd5218 | 1215 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
1216 | return -EINVAL; |
1217 | } | |
1218 | ||
1219 | /* check src1 operand */ | |
dc503a8a | 1220 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
1221 | if (err) |
1222 | return err; | |
1223 | ||
1224 | /* check src2 operand */ | |
dc503a8a | 1225 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
1226 | if (err) |
1227 | return err; | |
1228 | ||
6bdf6abc | 1229 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 1230 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
1231 | return -EACCES; |
1232 | } | |
1233 | ||
17a52670 | 1234 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1235 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1236 | BPF_SIZE(insn->code), BPF_READ, -1); |
1237 | if (err) | |
1238 | return err; | |
1239 | ||
1240 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1241 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
1242 | BPF_SIZE(insn->code), BPF_WRITE, -1); |
1243 | } | |
1244 | ||
f1174f77 EC |
1245 | /* Does this register contain a constant zero? */ |
1246 | static bool register_is_null(struct bpf_reg_state reg) | |
1247 | { | |
1248 | return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0); | |
1249 | } | |
1250 | ||
17a52670 AS |
1251 | /* when register 'regno' is passed into function that will read 'access_size' |
1252 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1253 | * and all elements of stack are initialized. |
1254 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1255 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1256 | */ |
58e2af8b | 1257 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1258 | int access_size, bool zero_size_allowed, |
1259 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1260 | { |
638f5b90 | 1261 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1262 | struct bpf_reg_state *regs = state->regs; |
638f5b90 | 1263 | int off, i, slot, spi; |
17a52670 | 1264 | |
8e2fe1d9 | 1265 | if (regs[regno].type != PTR_TO_STACK) { |
f1174f77 | 1266 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1267 | if (zero_size_allowed && access_size == 0 && |
f1174f77 | 1268 | register_is_null(regs[regno])) |
8e2fe1d9 DB |
1269 | return 0; |
1270 | ||
61bd5218 | 1271 | verbose(env, "R%d type=%s expected=%s\n", regno, |
8e2fe1d9 DB |
1272 | reg_type_str[regs[regno].type], |
1273 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 1274 | return -EACCES; |
8e2fe1d9 | 1275 | } |
17a52670 | 1276 | |
f1174f77 EC |
1277 | /* Only allow fixed-offset stack reads */ |
1278 | if (!tnum_is_const(regs[regno].var_off)) { | |
1279 | char tn_buf[48]; | |
1280 | ||
1281 | tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off); | |
61bd5218 | 1282 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 EC |
1283 | regno, tn_buf); |
1284 | } | |
1285 | off = regs[regno].off + regs[regno].var_off.value; | |
17a52670 AS |
1286 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
1287 | access_size <= 0) { | |
61bd5218 | 1288 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
1289 | regno, off, access_size); |
1290 | return -EACCES; | |
1291 | } | |
1292 | ||
8726679a AS |
1293 | if (env->prog->aux->stack_depth < -off) |
1294 | env->prog->aux->stack_depth = -off; | |
1295 | ||
435faee1 DB |
1296 | if (meta && meta->raw_mode) { |
1297 | meta->access_size = access_size; | |
1298 | meta->regno = regno; | |
1299 | return 0; | |
1300 | } | |
1301 | ||
17a52670 | 1302 | for (i = 0; i < access_size; i++) { |
638f5b90 AS |
1303 | slot = -(off + i) - 1; |
1304 | spi = slot / BPF_REG_SIZE; | |
1305 | if (state->allocated_stack <= slot || | |
1306 | state->stack[spi].slot_type[slot % BPF_REG_SIZE] != | |
1307 | STACK_MISC) { | |
61bd5218 | 1308 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", |
17a52670 AS |
1309 | off, i, access_size); |
1310 | return -EACCES; | |
1311 | } | |
1312 | } | |
1313 | return 0; | |
1314 | } | |
1315 | ||
06c1c049 GB |
1316 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1317 | int access_size, bool zero_size_allowed, | |
1318 | struct bpf_call_arg_meta *meta) | |
1319 | { | |
638f5b90 | 1320 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 1321 | |
f1174f77 | 1322 | switch (reg->type) { |
06c1c049 | 1323 | case PTR_TO_PACKET: |
de8f3a83 | 1324 | case PTR_TO_PACKET_META: |
f1174f77 | 1325 | return check_packet_access(env, regno, reg->off, access_size); |
06c1c049 | 1326 | case PTR_TO_MAP_VALUE: |
f1174f77 EC |
1327 | return check_map_access(env, regno, reg->off, access_size); |
1328 | default: /* scalar_value|ptr_to_stack or invalid ptr */ | |
06c1c049 GB |
1329 | return check_stack_boundary(env, regno, access_size, |
1330 | zero_size_allowed, meta); | |
1331 | } | |
1332 | } | |
1333 | ||
58e2af8b | 1334 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1335 | enum bpf_arg_type arg_type, |
1336 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1337 | { |
638f5b90 | 1338 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 1339 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1340 | int err = 0; |
1341 | ||
80f1d68c | 1342 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1343 | return 0; |
1344 | ||
dc503a8a EC |
1345 | err = check_reg_arg(env, regno, SRC_OP); |
1346 | if (err) | |
1347 | return err; | |
17a52670 | 1348 | |
1be7f75d AS |
1349 | if (arg_type == ARG_ANYTHING) { |
1350 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
1351 | verbose(env, "R%d leaks addr into helper function\n", |
1352 | regno); | |
1be7f75d AS |
1353 | return -EACCES; |
1354 | } | |
80f1d68c | 1355 | return 0; |
1be7f75d | 1356 | } |
80f1d68c | 1357 | |
de8f3a83 | 1358 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 1359 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 1360 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
1361 | return -EACCES; |
1362 | } | |
1363 | ||
8e2fe1d9 | 1364 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1365 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1366 | expected_type = PTR_TO_STACK; | |
de8f3a83 DB |
1367 | if (!type_is_pkt_pointer(type) && |
1368 | type != expected_type) | |
6841de8b | 1369 | goto err_type; |
39f19ebb AS |
1370 | } else if (arg_type == ARG_CONST_SIZE || |
1371 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1372 | expected_type = SCALAR_VALUE; |
1373 | if (type != expected_type) | |
6841de8b | 1374 | goto err_type; |
17a52670 AS |
1375 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1376 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1377 | if (type != expected_type) |
1378 | goto err_type; | |
608cd71a AS |
1379 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1380 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1381 | if (type != expected_type) |
1382 | goto err_type; | |
39f19ebb AS |
1383 | } else if (arg_type == ARG_PTR_TO_MEM || |
1384 | arg_type == ARG_PTR_TO_UNINIT_MEM) { | |
8e2fe1d9 DB |
1385 | expected_type = PTR_TO_STACK; |
1386 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1387 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1388 | * happens during stack boundary checking. |
1389 | */ | |
f1174f77 | 1390 | if (register_is_null(*reg)) |
6841de8b | 1391 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
1392 | else if (!type_is_pkt_pointer(type) && |
1393 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 1394 | type != expected_type) |
6841de8b | 1395 | goto err_type; |
39f19ebb | 1396 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 1397 | } else { |
61bd5218 | 1398 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
1399 | return -EFAULT; |
1400 | } | |
1401 | ||
17a52670 AS |
1402 | if (arg_type == ARG_CONST_MAP_PTR) { |
1403 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1404 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1405 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1406 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1407 | * check that [key, key + map->key_size) are within | |
1408 | * stack limits and initialized | |
1409 | */ | |
33ff9823 | 1410 | if (!meta->map_ptr) { |
17a52670 AS |
1411 | /* in function declaration map_ptr must come before |
1412 | * map_key, so that it's verified and known before | |
1413 | * we have to check map_key here. Otherwise it means | |
1414 | * that kernel subsystem misconfigured verifier | |
1415 | */ | |
61bd5218 | 1416 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
1417 | return -EACCES; |
1418 | } | |
de8f3a83 | 1419 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1420 | err = check_packet_access(env, regno, reg->off, |
6841de8b AS |
1421 | meta->map_ptr->key_size); |
1422 | else | |
1423 | err = check_stack_boundary(env, regno, | |
1424 | meta->map_ptr->key_size, | |
1425 | false, NULL); | |
17a52670 AS |
1426 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1427 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1428 | * check [value, value + map->value_size) validity | |
1429 | */ | |
33ff9823 | 1430 | if (!meta->map_ptr) { |
17a52670 | 1431 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 1432 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
1433 | return -EACCES; |
1434 | } | |
de8f3a83 | 1435 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1436 | err = check_packet_access(env, regno, reg->off, |
6841de8b AS |
1437 | meta->map_ptr->value_size); |
1438 | else | |
1439 | err = check_stack_boundary(env, regno, | |
1440 | meta->map_ptr->value_size, | |
1441 | false, NULL); | |
39f19ebb AS |
1442 | } else if (arg_type == ARG_CONST_SIZE || |
1443 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1444 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1445 | |
17a52670 AS |
1446 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1447 | * from stack pointer 'buf'. Check it | |
1448 | * note: regno == len, regno - 1 == buf | |
1449 | */ | |
1450 | if (regno == 0) { | |
1451 | /* kernel subsystem misconfigured verifier */ | |
61bd5218 JK |
1452 | verbose(env, |
1453 | "ARG_CONST_SIZE cannot be first argument\n"); | |
17a52670 AS |
1454 | return -EACCES; |
1455 | } | |
06c1c049 | 1456 | |
f1174f77 EC |
1457 | /* The register is SCALAR_VALUE; the access check |
1458 | * happens using its boundaries. | |
06c1c049 | 1459 | */ |
f1174f77 EC |
1460 | |
1461 | if (!tnum_is_const(reg->var_off)) | |
06c1c049 GB |
1462 | /* For unprivileged variable accesses, disable raw |
1463 | * mode so that the program is required to | |
1464 | * initialize all the memory that the helper could | |
1465 | * just partially fill up. | |
1466 | */ | |
1467 | meta = NULL; | |
1468 | ||
b03c9f9f | 1469 | if (reg->smin_value < 0) { |
61bd5218 | 1470 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
1471 | regno); |
1472 | return -EACCES; | |
1473 | } | |
06c1c049 | 1474 | |
b03c9f9f | 1475 | if (reg->umin_value == 0) { |
f1174f77 EC |
1476 | err = check_helper_mem_access(env, regno - 1, 0, |
1477 | zero_size_allowed, | |
1478 | meta); | |
06c1c049 GB |
1479 | if (err) |
1480 | return err; | |
06c1c049 | 1481 | } |
f1174f77 | 1482 | |
b03c9f9f | 1483 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 1484 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
1485 | regno); |
1486 | return -EACCES; | |
1487 | } | |
1488 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 1489 | reg->umax_value, |
f1174f77 | 1490 | zero_size_allowed, meta); |
17a52670 AS |
1491 | } |
1492 | ||
1493 | return err; | |
6841de8b | 1494 | err_type: |
61bd5218 | 1495 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
1496 | reg_type_str[type], reg_type_str[expected_type]); |
1497 | return -EACCES; | |
17a52670 AS |
1498 | } |
1499 | ||
61bd5218 JK |
1500 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
1501 | struct bpf_map *map, int func_id) | |
35578d79 | 1502 | { |
35578d79 KX |
1503 | if (!map) |
1504 | return 0; | |
1505 | ||
6aff67c8 AS |
1506 | /* We need a two way check, first is from map perspective ... */ |
1507 | switch (map->map_type) { | |
1508 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1509 | if (func_id != BPF_FUNC_tail_call) | |
1510 | goto error; | |
1511 | break; | |
1512 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1513 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
1514 | func_id != BPF_FUNC_perf_event_output && |
1515 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
1516 | goto error; |
1517 | break; | |
1518 | case BPF_MAP_TYPE_STACK_TRACE: | |
1519 | if (func_id != BPF_FUNC_get_stackid) | |
1520 | goto error; | |
1521 | break; | |
4ed8ec52 | 1522 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1523 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1524 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1525 | goto error; |
1526 | break; | |
546ac1ff JF |
1527 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
1528 | * allow to be modified from bpf side. So do not allow lookup elements | |
1529 | * for now. | |
1530 | */ | |
1531 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 1532 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
1533 | goto error; |
1534 | break; | |
6710e112 JDB |
1535 | /* Restrict bpf side of cpumap, open when use-cases appear */ |
1536 | case BPF_MAP_TYPE_CPUMAP: | |
1537 | if (func_id != BPF_FUNC_redirect_map) | |
1538 | goto error; | |
1539 | break; | |
56f668df | 1540 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1541 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1542 | if (func_id != BPF_FUNC_map_lookup_elem) |
1543 | goto error; | |
16a43625 | 1544 | break; |
174a79ff JF |
1545 | case BPF_MAP_TYPE_SOCKMAP: |
1546 | if (func_id != BPF_FUNC_sk_redirect_map && | |
1547 | func_id != BPF_FUNC_sock_map_update && | |
1548 | func_id != BPF_FUNC_map_delete_elem) | |
1549 | goto error; | |
1550 | break; | |
6aff67c8 AS |
1551 | default: |
1552 | break; | |
1553 | } | |
1554 | ||
1555 | /* ... and second from the function itself. */ | |
1556 | switch (func_id) { | |
1557 | case BPF_FUNC_tail_call: | |
1558 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1559 | goto error; | |
1560 | break; | |
1561 | case BPF_FUNC_perf_event_read: | |
1562 | case BPF_FUNC_perf_event_output: | |
908432ca | 1563 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
1564 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
1565 | goto error; | |
1566 | break; | |
1567 | case BPF_FUNC_get_stackid: | |
1568 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1569 | goto error; | |
1570 | break; | |
60d20f91 | 1571 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1572 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1573 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1574 | goto error; | |
1575 | break; | |
97f91a7c | 1576 | case BPF_FUNC_redirect_map: |
9c270af3 JDB |
1577 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
1578 | map->map_type != BPF_MAP_TYPE_CPUMAP) | |
97f91a7c JF |
1579 | goto error; |
1580 | break; | |
174a79ff JF |
1581 | case BPF_FUNC_sk_redirect_map: |
1582 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1583 | goto error; | |
1584 | break; | |
1585 | case BPF_FUNC_sock_map_update: | |
1586 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1587 | goto error; | |
1588 | break; | |
6aff67c8 AS |
1589 | default: |
1590 | break; | |
35578d79 KX |
1591 | } |
1592 | ||
1593 | return 0; | |
6aff67c8 | 1594 | error: |
61bd5218 | 1595 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 1596 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 1597 | return -EINVAL; |
35578d79 KX |
1598 | } |
1599 | ||
435faee1 DB |
1600 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1601 | { | |
1602 | int count = 0; | |
1603 | ||
39f19ebb | 1604 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1605 | count++; |
39f19ebb | 1606 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1607 | count++; |
39f19ebb | 1608 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1609 | count++; |
39f19ebb | 1610 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1611 | count++; |
39f19ebb | 1612 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1613 | count++; |
1614 | ||
1615 | return count > 1 ? -EINVAL : 0; | |
1616 | } | |
1617 | ||
de8f3a83 DB |
1618 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
1619 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 1620 | */ |
58e2af8b | 1621 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1622 | { |
638f5b90 | 1623 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1624 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
1625 | int i; |
1626 | ||
1627 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 1628 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 1629 | mark_reg_unknown(env, regs, i); |
969bf05e | 1630 | |
638f5b90 AS |
1631 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
1632 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 1633 | continue; |
638f5b90 | 1634 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 DB |
1635 | if (reg_is_pkt_pointer_any(reg)) |
1636 | __mark_reg_unknown(reg); | |
969bf05e AS |
1637 | } |
1638 | } | |
1639 | ||
81ed18ab | 1640 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1641 | { |
17a52670 | 1642 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 1643 | struct bpf_reg_state *regs; |
33ff9823 | 1644 | struct bpf_call_arg_meta meta; |
969bf05e | 1645 | bool changes_data; |
17a52670 AS |
1646 | int i, err; |
1647 | ||
1648 | /* find function prototype */ | |
1649 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
1650 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
1651 | func_id); | |
17a52670 AS |
1652 | return -EINVAL; |
1653 | } | |
1654 | ||
00176a34 JK |
1655 | if (env->ops->get_func_proto) |
1656 | fn = env->ops->get_func_proto(func_id); | |
17a52670 AS |
1657 | |
1658 | if (!fn) { | |
61bd5218 JK |
1659 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
1660 | func_id); | |
17a52670 AS |
1661 | return -EINVAL; |
1662 | } | |
1663 | ||
1664 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1665 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
61bd5218 | 1666 | verbose(env, "cannot call GPL only function from proprietary program\n"); |
17a52670 AS |
1667 | return -EINVAL; |
1668 | } | |
1669 | ||
17bedab2 | 1670 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
969bf05e | 1671 | |
33ff9823 | 1672 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1673 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1674 | |
435faee1 DB |
1675 | /* We only support one arg being in raw mode at the moment, which |
1676 | * is sufficient for the helper functions we have right now. | |
1677 | */ | |
1678 | err = check_raw_mode(fn); | |
1679 | if (err) { | |
61bd5218 | 1680 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 1681 | func_id_name(func_id), func_id); |
435faee1 DB |
1682 | return err; |
1683 | } | |
1684 | ||
17a52670 | 1685 | /* check args */ |
33ff9823 | 1686 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1687 | if (err) |
1688 | return err; | |
33ff9823 | 1689 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1690 | if (err) |
1691 | return err; | |
33ff9823 | 1692 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1693 | if (err) |
1694 | return err; | |
33ff9823 | 1695 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1696 | if (err) |
1697 | return err; | |
33ff9823 | 1698 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1699 | if (err) |
1700 | return err; | |
1701 | ||
435faee1 DB |
1702 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1703 | * is inferred from register state. | |
1704 | */ | |
1705 | for (i = 0; i < meta.access_size; i++) { | |
31fd8581 | 1706 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, BPF_WRITE, -1); |
435faee1 DB |
1707 | if (err) |
1708 | return err; | |
1709 | } | |
1710 | ||
638f5b90 | 1711 | regs = cur_regs(env); |
17a52670 | 1712 | /* reset caller saved regs */ |
dc503a8a | 1713 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 1714 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
1715 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
1716 | } | |
17a52670 | 1717 | |
dc503a8a | 1718 | /* update return register (already marked as written above) */ |
17a52670 | 1719 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 1720 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 1721 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
1722 | } else if (fn->ret_type == RET_VOID) { |
1723 | regs[BPF_REG_0].type = NOT_INIT; | |
1724 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
1725 | struct bpf_insn_aux_data *insn_aux; |
1726 | ||
17a52670 | 1727 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
f1174f77 | 1728 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 1729 | mark_reg_known_zero(env, regs, BPF_REG_0); |
f1174f77 | 1730 | regs[BPF_REG_0].off = 0; |
17a52670 AS |
1731 | /* remember map_ptr, so that check_map_access() |
1732 | * can check 'value_size' boundary of memory access | |
1733 | * to map element returned from bpf_map_lookup_elem() | |
1734 | */ | |
33ff9823 | 1735 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
1736 | verbose(env, |
1737 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
1738 | return -EINVAL; |
1739 | } | |
33ff9823 | 1740 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1741 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
1742 | insn_aux = &env->insn_aux_data[insn_idx]; |
1743 | if (!insn_aux->map_ptr) | |
1744 | insn_aux->map_ptr = meta.map_ptr; | |
1745 | else if (insn_aux->map_ptr != meta.map_ptr) | |
1746 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 1747 | } else { |
61bd5218 | 1748 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 1749 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
1750 | return -EINVAL; |
1751 | } | |
04fd61ab | 1752 | |
61bd5218 | 1753 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
1754 | if (err) |
1755 | return err; | |
04fd61ab | 1756 | |
969bf05e AS |
1757 | if (changes_data) |
1758 | clear_all_pkt_pointers(env); | |
1759 | return 0; | |
1760 | } | |
1761 | ||
f1174f77 EC |
1762 | static void coerce_reg_to_32(struct bpf_reg_state *reg) |
1763 | { | |
f1174f77 EC |
1764 | /* clear high 32 bits */ |
1765 | reg->var_off = tnum_cast(reg->var_off, 4); | |
b03c9f9f EC |
1766 | /* Update bounds */ |
1767 | __update_reg_bounds(reg); | |
1768 | } | |
1769 | ||
1770 | static bool signed_add_overflows(s64 a, s64 b) | |
1771 | { | |
1772 | /* Do the add in u64, where overflow is well-defined */ | |
1773 | s64 res = (s64)((u64)a + (u64)b); | |
1774 | ||
1775 | if (b < 0) | |
1776 | return res > a; | |
1777 | return res < a; | |
1778 | } | |
1779 | ||
1780 | static bool signed_sub_overflows(s64 a, s64 b) | |
1781 | { | |
1782 | /* Do the sub in u64, where overflow is well-defined */ | |
1783 | s64 res = (s64)((u64)a - (u64)b); | |
1784 | ||
1785 | if (b < 0) | |
1786 | return res < a; | |
1787 | return res > a; | |
969bf05e AS |
1788 | } |
1789 | ||
f1174f77 | 1790 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
1791 | * Caller should also handle BPF_MOV case separately. |
1792 | * If we return -EACCES, caller may want to try again treating pointer as a | |
1793 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
1794 | */ | |
1795 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
1796 | struct bpf_insn *insn, | |
1797 | const struct bpf_reg_state *ptr_reg, | |
1798 | const struct bpf_reg_state *off_reg) | |
969bf05e | 1799 | { |
638f5b90 | 1800 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg; |
f1174f77 | 1801 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
1802 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
1803 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
1804 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
1805 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
969bf05e | 1806 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 1807 | u32 dst = insn->dst_reg; |
969bf05e | 1808 | |
f1174f77 | 1809 | dst_reg = ®s[dst]; |
969bf05e | 1810 | |
b03c9f9f | 1811 | if (WARN_ON_ONCE(known && (smin_val != smax_val))) { |
638f5b90 | 1812 | print_verifier_state(env, env->cur_state); |
61bd5218 JK |
1813 | verbose(env, |
1814 | "verifier internal error: known but bad sbounds\n"); | |
b03c9f9f EC |
1815 | return -EINVAL; |
1816 | } | |
1817 | if (WARN_ON_ONCE(known && (umin_val != umax_val))) { | |
638f5b90 | 1818 | print_verifier_state(env, env->cur_state); |
61bd5218 JK |
1819 | verbose(env, |
1820 | "verifier internal error: known but bad ubounds\n"); | |
f1174f77 EC |
1821 | return -EINVAL; |
1822 | } | |
1823 | ||
1824 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
1825 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
1826 | if (!env->allow_ptr_leaks) | |
61bd5218 JK |
1827 | verbose(env, |
1828 | "R%d 32-bit pointer arithmetic prohibited\n", | |
f1174f77 EC |
1829 | dst); |
1830 | return -EACCES; | |
969bf05e AS |
1831 | } |
1832 | ||
f1174f77 EC |
1833 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
1834 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1835 | verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", |
f1174f77 EC |
1836 | dst); |
1837 | return -EACCES; | |
1838 | } | |
1839 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
1840 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1841 | verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", |
f1174f77 EC |
1842 | dst); |
1843 | return -EACCES; | |
1844 | } | |
1845 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
1846 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1847 | verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", |
f1174f77 EC |
1848 | dst); |
1849 | return -EACCES; | |
1850 | } | |
1851 | ||
1852 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
1853 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 1854 | */ |
f1174f77 EC |
1855 | dst_reg->type = ptr_reg->type; |
1856 | dst_reg->id = ptr_reg->id; | |
969bf05e | 1857 | |
f1174f77 EC |
1858 | switch (opcode) { |
1859 | case BPF_ADD: | |
1860 | /* We can take a fixed offset as long as it doesn't overflow | |
1861 | * the s32 'off' field | |
969bf05e | 1862 | */ |
b03c9f9f EC |
1863 | if (known && (ptr_reg->off + smin_val == |
1864 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 1865 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
1866 | dst_reg->smin_value = smin_ptr; |
1867 | dst_reg->smax_value = smax_ptr; | |
1868 | dst_reg->umin_value = umin_ptr; | |
1869 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 1870 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 1871 | dst_reg->off = ptr_reg->off + smin_val; |
f1174f77 EC |
1872 | dst_reg->range = ptr_reg->range; |
1873 | break; | |
1874 | } | |
f1174f77 EC |
1875 | /* A new variable offset is created. Note that off_reg->off |
1876 | * == 0, since it's a scalar. | |
1877 | * dst_reg gets the pointer type and since some positive | |
1878 | * integer value was added to the pointer, give it a new 'id' | |
1879 | * if it's a PTR_TO_PACKET. | |
1880 | * this creates a new 'base' pointer, off_reg (variable) gets | |
1881 | * added into the variable offset, and we copy the fixed offset | |
1882 | * from ptr_reg. | |
969bf05e | 1883 | */ |
b03c9f9f EC |
1884 | if (signed_add_overflows(smin_ptr, smin_val) || |
1885 | signed_add_overflows(smax_ptr, smax_val)) { | |
1886 | dst_reg->smin_value = S64_MIN; | |
1887 | dst_reg->smax_value = S64_MAX; | |
1888 | } else { | |
1889 | dst_reg->smin_value = smin_ptr + smin_val; | |
1890 | dst_reg->smax_value = smax_ptr + smax_val; | |
1891 | } | |
1892 | if (umin_ptr + umin_val < umin_ptr || | |
1893 | umax_ptr + umax_val < umax_ptr) { | |
1894 | dst_reg->umin_value = 0; | |
1895 | dst_reg->umax_value = U64_MAX; | |
1896 | } else { | |
1897 | dst_reg->umin_value = umin_ptr + umin_val; | |
1898 | dst_reg->umax_value = umax_ptr + umax_val; | |
1899 | } | |
f1174f77 EC |
1900 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
1901 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 1902 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
1903 | dst_reg->id = ++env->id_gen; |
1904 | /* something was added to pkt_ptr, set range to zero */ | |
1905 | dst_reg->range = 0; | |
1906 | } | |
1907 | break; | |
1908 | case BPF_SUB: | |
1909 | if (dst_reg == off_reg) { | |
1910 | /* scalar -= pointer. Creates an unknown scalar */ | |
1911 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1912 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
f1174f77 EC |
1913 | dst); |
1914 | return -EACCES; | |
1915 | } | |
1916 | /* We don't allow subtraction from FP, because (according to | |
1917 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
1918 | * be able to deal with it. | |
969bf05e | 1919 | */ |
f1174f77 EC |
1920 | if (ptr_reg->type == PTR_TO_STACK) { |
1921 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1922 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
f1174f77 EC |
1923 | dst); |
1924 | return -EACCES; | |
1925 | } | |
b03c9f9f EC |
1926 | if (known && (ptr_reg->off - smin_val == |
1927 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 1928 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
1929 | dst_reg->smin_value = smin_ptr; |
1930 | dst_reg->smax_value = smax_ptr; | |
1931 | dst_reg->umin_value = umin_ptr; | |
1932 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
1933 | dst_reg->var_off = ptr_reg->var_off; |
1934 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 1935 | dst_reg->off = ptr_reg->off - smin_val; |
f1174f77 EC |
1936 | dst_reg->range = ptr_reg->range; |
1937 | break; | |
1938 | } | |
f1174f77 EC |
1939 | /* A new variable offset is created. If the subtrahend is known |
1940 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 1941 | */ |
b03c9f9f EC |
1942 | if (signed_sub_overflows(smin_ptr, smax_val) || |
1943 | signed_sub_overflows(smax_ptr, smin_val)) { | |
1944 | /* Overflow possible, we know nothing */ | |
1945 | dst_reg->smin_value = S64_MIN; | |
1946 | dst_reg->smax_value = S64_MAX; | |
1947 | } else { | |
1948 | dst_reg->smin_value = smin_ptr - smax_val; | |
1949 | dst_reg->smax_value = smax_ptr - smin_val; | |
1950 | } | |
1951 | if (umin_ptr < umax_val) { | |
1952 | /* Overflow possible, we know nothing */ | |
1953 | dst_reg->umin_value = 0; | |
1954 | dst_reg->umax_value = U64_MAX; | |
1955 | } else { | |
1956 | /* Cannot overflow (as long as bounds are consistent) */ | |
1957 | dst_reg->umin_value = umin_ptr - umax_val; | |
1958 | dst_reg->umax_value = umax_ptr - umin_val; | |
1959 | } | |
f1174f77 EC |
1960 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
1961 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 1962 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
1963 | dst_reg->id = ++env->id_gen; |
1964 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 1965 | if (smin_val < 0) |
f1174f77 | 1966 | dst_reg->range = 0; |
43188702 | 1967 | } |
f1174f77 EC |
1968 | break; |
1969 | case BPF_AND: | |
1970 | case BPF_OR: | |
1971 | case BPF_XOR: | |
1972 | /* bitwise ops on pointers are troublesome, prohibit for now. | |
1973 | * (However, in principle we could allow some cases, e.g. | |
1974 | * ptr &= ~3 which would reduce min_value by 3.) | |
1975 | */ | |
1976 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1977 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", |
f1174f77 EC |
1978 | dst, bpf_alu_string[opcode >> 4]); |
1979 | return -EACCES; | |
1980 | default: | |
1981 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
1982 | if (!env->allow_ptr_leaks) | |
61bd5218 | 1983 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
f1174f77 EC |
1984 | dst, bpf_alu_string[opcode >> 4]); |
1985 | return -EACCES; | |
43188702 JF |
1986 | } |
1987 | ||
b03c9f9f EC |
1988 | __update_reg_bounds(dst_reg); |
1989 | __reg_deduce_bounds(dst_reg); | |
1990 | __reg_bound_offset(dst_reg); | |
43188702 JF |
1991 | return 0; |
1992 | } | |
1993 | ||
f1174f77 EC |
1994 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
1995 | struct bpf_insn *insn, | |
1996 | struct bpf_reg_state *dst_reg, | |
1997 | struct bpf_reg_state src_reg) | |
969bf05e | 1998 | { |
638f5b90 | 1999 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 2000 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2001 | bool src_known, dst_known; |
b03c9f9f EC |
2002 | s64 smin_val, smax_val; |
2003 | u64 umin_val, umax_val; | |
48461135 | 2004 | |
f1174f77 EC |
2005 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
2006 | /* 32-bit ALU ops are (32,32)->64 */ | |
2007 | coerce_reg_to_32(dst_reg); | |
2008 | coerce_reg_to_32(&src_reg); | |
9305706c | 2009 | } |
b03c9f9f EC |
2010 | smin_val = src_reg.smin_value; |
2011 | smax_val = src_reg.smax_value; | |
2012 | umin_val = src_reg.umin_value; | |
2013 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
2014 | src_known = tnum_is_const(src_reg.var_off); |
2015 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 2016 | |
48461135 JB |
2017 | switch (opcode) { |
2018 | case BPF_ADD: | |
b03c9f9f EC |
2019 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
2020 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
2021 | dst_reg->smin_value = S64_MIN; | |
2022 | dst_reg->smax_value = S64_MAX; | |
2023 | } else { | |
2024 | dst_reg->smin_value += smin_val; | |
2025 | dst_reg->smax_value += smax_val; | |
2026 | } | |
2027 | if (dst_reg->umin_value + umin_val < umin_val || | |
2028 | dst_reg->umax_value + umax_val < umax_val) { | |
2029 | dst_reg->umin_value = 0; | |
2030 | dst_reg->umax_value = U64_MAX; | |
2031 | } else { | |
2032 | dst_reg->umin_value += umin_val; | |
2033 | dst_reg->umax_value += umax_val; | |
2034 | } | |
f1174f77 | 2035 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2036 | break; |
2037 | case BPF_SUB: | |
b03c9f9f EC |
2038 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
2039 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
2040 | /* Overflow possible, we know nothing */ | |
2041 | dst_reg->smin_value = S64_MIN; | |
2042 | dst_reg->smax_value = S64_MAX; | |
2043 | } else { | |
2044 | dst_reg->smin_value -= smax_val; | |
2045 | dst_reg->smax_value -= smin_val; | |
2046 | } | |
2047 | if (dst_reg->umin_value < umax_val) { | |
2048 | /* Overflow possible, we know nothing */ | |
2049 | dst_reg->umin_value = 0; | |
2050 | dst_reg->umax_value = U64_MAX; | |
2051 | } else { | |
2052 | /* Cannot overflow (as long as bounds are consistent) */ | |
2053 | dst_reg->umin_value -= umax_val; | |
2054 | dst_reg->umax_value -= umin_val; | |
2055 | } | |
f1174f77 | 2056 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2057 | break; |
2058 | case BPF_MUL: | |
b03c9f9f EC |
2059 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
2060 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 2061 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
2062 | __mark_reg_unbounded(dst_reg); |
2063 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2064 | break; |
2065 | } | |
b03c9f9f EC |
2066 | /* Both values are positive, so we can work with unsigned and |
2067 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 2068 | */ |
b03c9f9f EC |
2069 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
2070 | /* Potential overflow, we know nothing */ | |
2071 | __mark_reg_unbounded(dst_reg); | |
2072 | /* (except what we can learn from the var_off) */ | |
2073 | __update_reg_bounds(dst_reg); | |
2074 | break; | |
2075 | } | |
2076 | dst_reg->umin_value *= umin_val; | |
2077 | dst_reg->umax_value *= umax_val; | |
2078 | if (dst_reg->umax_value > S64_MAX) { | |
2079 | /* Overflow possible, we know nothing */ | |
2080 | dst_reg->smin_value = S64_MIN; | |
2081 | dst_reg->smax_value = S64_MAX; | |
2082 | } else { | |
2083 | dst_reg->smin_value = dst_reg->umin_value; | |
2084 | dst_reg->smax_value = dst_reg->umax_value; | |
2085 | } | |
48461135 JB |
2086 | break; |
2087 | case BPF_AND: | |
f1174f77 | 2088 | if (src_known && dst_known) { |
b03c9f9f EC |
2089 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
2090 | src_reg.var_off.value); | |
f1174f77 EC |
2091 | break; |
2092 | } | |
b03c9f9f EC |
2093 | /* We get our minimum from the var_off, since that's inherently |
2094 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 2095 | */ |
f1174f77 | 2096 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
2097 | dst_reg->umin_value = dst_reg->var_off.value; |
2098 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
2099 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2100 | /* Lose signed bounds when ANDing negative numbers, | |
2101 | * ain't nobody got time for that. | |
2102 | */ | |
2103 | dst_reg->smin_value = S64_MIN; | |
2104 | dst_reg->smax_value = S64_MAX; | |
2105 | } else { | |
2106 | /* ANDing two positives gives a positive, so safe to | |
2107 | * cast result into s64. | |
2108 | */ | |
2109 | dst_reg->smin_value = dst_reg->umin_value; | |
2110 | dst_reg->smax_value = dst_reg->umax_value; | |
2111 | } | |
2112 | /* We may learn something more from the var_off */ | |
2113 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2114 | break; |
2115 | case BPF_OR: | |
2116 | if (src_known && dst_known) { | |
b03c9f9f EC |
2117 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
2118 | src_reg.var_off.value); | |
f1174f77 EC |
2119 | break; |
2120 | } | |
b03c9f9f EC |
2121 | /* We get our maximum from the var_off, and our minimum is the |
2122 | * maximum of the operands' minima | |
f1174f77 EC |
2123 | */ |
2124 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
2125 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
2126 | dst_reg->umax_value = dst_reg->var_off.value | | |
2127 | dst_reg->var_off.mask; | |
2128 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2129 | /* Lose signed bounds when ORing negative numbers, | |
2130 | * ain't nobody got time for that. | |
2131 | */ | |
2132 | dst_reg->smin_value = S64_MIN; | |
2133 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2134 | } else { |
b03c9f9f EC |
2135 | /* ORing two positives gives a positive, so safe to |
2136 | * cast result into s64. | |
2137 | */ | |
2138 | dst_reg->smin_value = dst_reg->umin_value; | |
2139 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 2140 | } |
b03c9f9f EC |
2141 | /* We may learn something more from the var_off */ |
2142 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2143 | break; |
2144 | case BPF_LSH: | |
b03c9f9f EC |
2145 | if (umax_val > 63) { |
2146 | /* Shifts greater than 63 are undefined. This includes | |
2147 | * shifts by a negative number. | |
2148 | */ | |
61bd5218 | 2149 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2150 | break; |
2151 | } | |
b03c9f9f EC |
2152 | /* We lose all sign bit information (except what we can pick |
2153 | * up from var_off) | |
48461135 | 2154 | */ |
b03c9f9f EC |
2155 | dst_reg->smin_value = S64_MIN; |
2156 | dst_reg->smax_value = S64_MAX; | |
2157 | /* If we might shift our top bit out, then we know nothing */ | |
2158 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
2159 | dst_reg->umin_value = 0; | |
2160 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 2161 | } else { |
b03c9f9f EC |
2162 | dst_reg->umin_value <<= umin_val; |
2163 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 2164 | } |
b03c9f9f EC |
2165 | if (src_known) |
2166 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); | |
2167 | else | |
2168 | dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val); | |
2169 | /* We may learn something more from the var_off */ | |
2170 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2171 | break; |
2172 | case BPF_RSH: | |
b03c9f9f EC |
2173 | if (umax_val > 63) { |
2174 | /* Shifts greater than 63 are undefined. This includes | |
2175 | * shifts by a negative number. | |
2176 | */ | |
61bd5218 | 2177 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2178 | break; |
2179 | } | |
2180 | /* BPF_RSH is an unsigned shift, so make the appropriate casts */ | |
b03c9f9f EC |
2181 | if (dst_reg->smin_value < 0) { |
2182 | if (umin_val) { | |
f1174f77 | 2183 | /* Sign bit will be cleared */ |
b03c9f9f EC |
2184 | dst_reg->smin_value = 0; |
2185 | } else { | |
2186 | /* Lost sign bit information */ | |
2187 | dst_reg->smin_value = S64_MIN; | |
2188 | dst_reg->smax_value = S64_MAX; | |
2189 | } | |
d1174416 | 2190 | } else { |
b03c9f9f EC |
2191 | dst_reg->smin_value = |
2192 | (u64)(dst_reg->smin_value) >> umax_val; | |
d1174416 | 2193 | } |
f1174f77 | 2194 | if (src_known) |
b03c9f9f EC |
2195 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, |
2196 | umin_val); | |
f1174f77 | 2197 | else |
b03c9f9f EC |
2198 | dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val); |
2199 | dst_reg->umin_value >>= umax_val; | |
2200 | dst_reg->umax_value >>= umin_val; | |
2201 | /* We may learn something more from the var_off */ | |
2202 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2203 | break; |
2204 | default: | |
61bd5218 | 2205 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
2206 | break; |
2207 | } | |
2208 | ||
b03c9f9f EC |
2209 | __reg_deduce_bounds(dst_reg); |
2210 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
2211 | return 0; |
2212 | } | |
2213 | ||
2214 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
2215 | * and var_off. | |
2216 | */ | |
2217 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
2218 | struct bpf_insn *insn) | |
2219 | { | |
638f5b90 | 2220 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg, *src_reg; |
f1174f77 EC |
2221 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
2222 | u8 opcode = BPF_OP(insn->code); | |
2223 | int rc; | |
2224 | ||
2225 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
2226 | src_reg = NULL; |
2227 | if (dst_reg->type != SCALAR_VALUE) | |
2228 | ptr_reg = dst_reg; | |
2229 | if (BPF_SRC(insn->code) == BPF_X) { | |
2230 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
2231 | if (src_reg->type != SCALAR_VALUE) { |
2232 | if (dst_reg->type != SCALAR_VALUE) { | |
2233 | /* Combining two pointers by any ALU op yields | |
2234 | * an arbitrary scalar. | |
2235 | */ | |
2236 | if (!env->allow_ptr_leaks) { | |
61bd5218 | 2237 | verbose(env, "R%d pointer %s pointer prohibited\n", |
f1174f77 EC |
2238 | insn->dst_reg, |
2239 | bpf_alu_string[opcode >> 4]); | |
2240 | return -EACCES; | |
2241 | } | |
61bd5218 | 2242 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2243 | return 0; |
2244 | } else { | |
2245 | /* scalar += pointer | |
2246 | * This is legal, but we have to reverse our | |
2247 | * src/dest handling in computing the range | |
2248 | */ | |
2249 | rc = adjust_ptr_min_max_vals(env, insn, | |
2250 | src_reg, dst_reg); | |
2251 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2252 | /* scalar += unknown scalar */ | |
2253 | __mark_reg_unknown(&off_reg); | |
2254 | return adjust_scalar_min_max_vals( | |
2255 | env, insn, | |
2256 | dst_reg, off_reg); | |
2257 | } | |
2258 | return rc; | |
2259 | } | |
2260 | } else if (ptr_reg) { | |
2261 | /* pointer += scalar */ | |
2262 | rc = adjust_ptr_min_max_vals(env, insn, | |
2263 | dst_reg, src_reg); | |
2264 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2265 | /* unknown scalar += scalar */ | |
2266 | __mark_reg_unknown(dst_reg); | |
2267 | return adjust_scalar_min_max_vals( | |
2268 | env, insn, dst_reg, *src_reg); | |
2269 | } | |
2270 | return rc; | |
2271 | } | |
2272 | } else { | |
2273 | /* Pretend the src is a reg with a known value, since we only | |
2274 | * need to be able to read from this state. | |
2275 | */ | |
2276 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 2277 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 2278 | src_reg = &off_reg; |
f1174f77 EC |
2279 | if (ptr_reg) { /* pointer += K */ |
2280 | rc = adjust_ptr_min_max_vals(env, insn, | |
2281 | ptr_reg, src_reg); | |
2282 | if (rc == -EACCES && env->allow_ptr_leaks) { | |
2283 | /* unknown scalar += K */ | |
2284 | __mark_reg_unknown(dst_reg); | |
2285 | return adjust_scalar_min_max_vals( | |
2286 | env, insn, dst_reg, off_reg); | |
2287 | } | |
2288 | return rc; | |
2289 | } | |
2290 | } | |
2291 | ||
2292 | /* Got here implies adding two SCALAR_VALUEs */ | |
2293 | if (WARN_ON_ONCE(ptr_reg)) { | |
638f5b90 | 2294 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2295 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
2296 | return -EINVAL; |
2297 | } | |
2298 | if (WARN_ON(!src_reg)) { | |
638f5b90 | 2299 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2300 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
2301 | return -EINVAL; |
2302 | } | |
2303 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
2304 | } |
2305 | ||
17a52670 | 2306 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 2307 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2308 | { |
638f5b90 | 2309 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
2310 | u8 opcode = BPF_OP(insn->code); |
2311 | int err; | |
2312 | ||
2313 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
2314 | if (opcode == BPF_NEG) { | |
2315 | if (BPF_SRC(insn->code) != 0 || | |
2316 | insn->src_reg != BPF_REG_0 || | |
2317 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 2318 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
2319 | return -EINVAL; |
2320 | } | |
2321 | } else { | |
2322 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
2323 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
2324 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 2325 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
2326 | return -EINVAL; |
2327 | } | |
2328 | } | |
2329 | ||
2330 | /* check src operand */ | |
dc503a8a | 2331 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2332 | if (err) |
2333 | return err; | |
2334 | ||
1be7f75d | 2335 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 2336 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
2337 | insn->dst_reg); |
2338 | return -EACCES; | |
2339 | } | |
2340 | ||
17a52670 | 2341 | /* check dest operand */ |
dc503a8a | 2342 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2343 | if (err) |
2344 | return err; | |
2345 | ||
2346 | } else if (opcode == BPF_MOV) { | |
2347 | ||
2348 | if (BPF_SRC(insn->code) == BPF_X) { | |
2349 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2350 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2351 | return -EINVAL; |
2352 | } | |
2353 | ||
2354 | /* check src operand */ | |
dc503a8a | 2355 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2356 | if (err) |
2357 | return err; | |
2358 | } else { | |
2359 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2360 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2361 | return -EINVAL; |
2362 | } | |
2363 | } | |
2364 | ||
2365 | /* check dest operand */ | |
dc503a8a | 2366 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2367 | if (err) |
2368 | return err; | |
2369 | ||
2370 | if (BPF_SRC(insn->code) == BPF_X) { | |
2371 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
2372 | /* case: R1 = R2 | |
2373 | * copy register state to dest reg | |
2374 | */ | |
2375 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
8fe2d6cc | 2376 | regs[insn->dst_reg].live |= REG_LIVE_WRITTEN; |
17a52670 | 2377 | } else { |
f1174f77 | 2378 | /* R1 = (u32) R2 */ |
1be7f75d | 2379 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
2380 | verbose(env, |
2381 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
2382 | insn->src_reg); |
2383 | return -EACCES; | |
2384 | } | |
61bd5218 | 2385 | mark_reg_unknown(env, regs, insn->dst_reg); |
b03c9f9f | 2386 | /* high 32 bits are known zero. */ |
f1174f77 EC |
2387 | regs[insn->dst_reg].var_off = tnum_cast( |
2388 | regs[insn->dst_reg].var_off, 4); | |
b03c9f9f | 2389 | __update_reg_bounds(®s[insn->dst_reg]); |
17a52670 AS |
2390 | } |
2391 | } else { | |
2392 | /* case: R = imm | |
2393 | * remember the value we stored into this reg | |
2394 | */ | |
f1174f77 | 2395 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 2396 | __mark_reg_known(regs + insn->dst_reg, insn->imm); |
17a52670 AS |
2397 | } |
2398 | ||
2399 | } else if (opcode > BPF_END) { | |
61bd5218 | 2400 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
2401 | return -EINVAL; |
2402 | ||
2403 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
2404 | ||
17a52670 AS |
2405 | if (BPF_SRC(insn->code) == BPF_X) { |
2406 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2407 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2408 | return -EINVAL; |
2409 | } | |
2410 | /* check src1 operand */ | |
dc503a8a | 2411 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2412 | if (err) |
2413 | return err; | |
2414 | } else { | |
2415 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2416 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2417 | return -EINVAL; |
2418 | } | |
2419 | } | |
2420 | ||
2421 | /* check src2 operand */ | |
dc503a8a | 2422 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2423 | if (err) |
2424 | return err; | |
2425 | ||
2426 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
2427 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 2428 | verbose(env, "div by zero\n"); |
17a52670 AS |
2429 | return -EINVAL; |
2430 | } | |
2431 | ||
229394e8 RV |
2432 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
2433 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
2434 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
2435 | ||
2436 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 2437 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
2438 | return -EINVAL; |
2439 | } | |
2440 | } | |
2441 | ||
1a0dc1ac | 2442 | /* check dest operand */ |
dc503a8a | 2443 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
2444 | if (err) |
2445 | return err; | |
2446 | ||
f1174f77 | 2447 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
2448 | } |
2449 | ||
2450 | return 0; | |
2451 | } | |
2452 | ||
58e2af8b | 2453 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
de8f3a83 | 2454 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 2455 | enum bpf_reg_type type, |
fb2a311a | 2456 | bool range_right_open) |
969bf05e | 2457 | { |
58e2af8b | 2458 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 2459 | u16 new_range; |
969bf05e | 2460 | int i; |
2d2be8ca | 2461 | |
fb2a311a DB |
2462 | if (dst_reg->off < 0 || |
2463 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
2464 | /* This doesn't give us any range */ |
2465 | return; | |
2466 | ||
b03c9f9f EC |
2467 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
2468 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
2469 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
2470 | * than pkt_end, but that's because it's also less than pkt. | |
2471 | */ | |
2472 | return; | |
2473 | ||
fb2a311a DB |
2474 | new_range = dst_reg->off; |
2475 | if (range_right_open) | |
2476 | new_range--; | |
2477 | ||
2478 | /* Examples for register markings: | |
2d2be8ca | 2479 | * |
fb2a311a | 2480 | * pkt_data in dst register: |
2d2be8ca DB |
2481 | * |
2482 | * r2 = r3; | |
2483 | * r2 += 8; | |
2484 | * if (r2 > pkt_end) goto <handle exception> | |
2485 | * <access okay> | |
2486 | * | |
b4e432f1 DB |
2487 | * r2 = r3; |
2488 | * r2 += 8; | |
2489 | * if (r2 < pkt_end) goto <access okay> | |
2490 | * <handle exception> | |
2491 | * | |
2d2be8ca DB |
2492 | * Where: |
2493 | * r2 == dst_reg, pkt_end == src_reg | |
2494 | * r2=pkt(id=n,off=8,r=0) | |
2495 | * r3=pkt(id=n,off=0,r=0) | |
2496 | * | |
fb2a311a | 2497 | * pkt_data in src register: |
2d2be8ca DB |
2498 | * |
2499 | * r2 = r3; | |
2500 | * r2 += 8; | |
2501 | * if (pkt_end >= r2) goto <access okay> | |
2502 | * <handle exception> | |
2503 | * | |
b4e432f1 DB |
2504 | * r2 = r3; |
2505 | * r2 += 8; | |
2506 | * if (pkt_end <= r2) goto <handle exception> | |
2507 | * <access okay> | |
2508 | * | |
2d2be8ca DB |
2509 | * Where: |
2510 | * pkt_end == dst_reg, r2 == src_reg | |
2511 | * r2=pkt(id=n,off=8,r=0) | |
2512 | * r3=pkt(id=n,off=0,r=0) | |
2513 | * | |
2514 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
2515 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
2516 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
2517 | * the check. | |
969bf05e | 2518 | */ |
2d2be8ca | 2519 | |
f1174f77 EC |
2520 | /* If our ids match, then we must have the same max_value. And we |
2521 | * don't care about the other reg's fixed offset, since if it's too big | |
2522 | * the range won't allow anything. | |
2523 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
2524 | */ | |
969bf05e | 2525 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 2526 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 2527 | /* keep the maximum range already checked */ |
fb2a311a | 2528 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 2529 | |
638f5b90 AS |
2530 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2531 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 2532 | continue; |
638f5b90 | 2533 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 | 2534 | if (reg->type == type && reg->id == dst_reg->id) |
b06723da | 2535 | reg->range = max(reg->range, new_range); |
969bf05e AS |
2536 | } |
2537 | } | |
2538 | ||
48461135 JB |
2539 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2540 | * variable register that we are working on, and src_reg is a constant or we're | |
2541 | * simply doing a BPF_K check. | |
f1174f77 | 2542 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
2543 | */ |
2544 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2545 | struct bpf_reg_state *false_reg, u64 val, | |
2546 | u8 opcode) | |
2547 | { | |
f1174f77 EC |
2548 | /* If the dst_reg is a pointer, we can't learn anything about its |
2549 | * variable offset from the compare (unless src_reg were a pointer into | |
2550 | * the same object, but we don't bother with that. | |
2551 | * Since false_reg and true_reg have the same type by construction, we | |
2552 | * only need to check one of them for pointerness. | |
2553 | */ | |
2554 | if (__is_pointer_value(false, false_reg)) | |
2555 | return; | |
4cabc5b1 | 2556 | |
48461135 JB |
2557 | switch (opcode) { |
2558 | case BPF_JEQ: | |
2559 | /* If this is false then we know nothing Jon Snow, but if it is | |
2560 | * true then we know for sure. | |
2561 | */ | |
b03c9f9f | 2562 | __mark_reg_known(true_reg, val); |
48461135 JB |
2563 | break; |
2564 | case BPF_JNE: | |
2565 | /* If this is true we know nothing Jon Snow, but if it is false | |
2566 | * we know the value for sure; | |
2567 | */ | |
b03c9f9f | 2568 | __mark_reg_known(false_reg, val); |
48461135 JB |
2569 | break; |
2570 | case BPF_JGT: | |
b03c9f9f EC |
2571 | false_reg->umax_value = min(false_reg->umax_value, val); |
2572 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2573 | break; | |
48461135 | 2574 | case BPF_JSGT: |
b03c9f9f EC |
2575 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
2576 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 2577 | break; |
b4e432f1 DB |
2578 | case BPF_JLT: |
2579 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2580 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
2581 | break; | |
2582 | case BPF_JSLT: | |
2583 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
2584 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
2585 | break; | |
48461135 | 2586 | case BPF_JGE: |
b03c9f9f EC |
2587 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
2588 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2589 | break; | |
48461135 | 2590 | case BPF_JSGE: |
b03c9f9f EC |
2591 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
2592 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 2593 | break; |
b4e432f1 DB |
2594 | case BPF_JLE: |
2595 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2596 | true_reg->umax_value = min(true_reg->umax_value, val); | |
2597 | break; | |
2598 | case BPF_JSLE: | |
2599 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
2600 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
2601 | break; | |
48461135 JB |
2602 | default: |
2603 | break; | |
2604 | } | |
2605 | ||
b03c9f9f EC |
2606 | __reg_deduce_bounds(false_reg); |
2607 | __reg_deduce_bounds(true_reg); | |
2608 | /* We might have learned some bits from the bounds. */ | |
2609 | __reg_bound_offset(false_reg); | |
2610 | __reg_bound_offset(true_reg); | |
2611 | /* Intersecting with the old var_off might have improved our bounds | |
2612 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2613 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2614 | */ | |
2615 | __update_reg_bounds(false_reg); | |
2616 | __update_reg_bounds(true_reg); | |
48461135 JB |
2617 | } |
2618 | ||
f1174f77 EC |
2619 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
2620 | * the variable reg. | |
48461135 JB |
2621 | */ |
2622 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
2623 | struct bpf_reg_state *false_reg, u64 val, | |
2624 | u8 opcode) | |
2625 | { | |
f1174f77 EC |
2626 | if (__is_pointer_value(false, false_reg)) |
2627 | return; | |
4cabc5b1 | 2628 | |
48461135 JB |
2629 | switch (opcode) { |
2630 | case BPF_JEQ: | |
2631 | /* If this is false then we know nothing Jon Snow, but if it is | |
2632 | * true then we know for sure. | |
2633 | */ | |
b03c9f9f | 2634 | __mark_reg_known(true_reg, val); |
48461135 JB |
2635 | break; |
2636 | case BPF_JNE: | |
2637 | /* If this is true we know nothing Jon Snow, but if it is false | |
2638 | * we know the value for sure; | |
2639 | */ | |
b03c9f9f | 2640 | __mark_reg_known(false_reg, val); |
48461135 JB |
2641 | break; |
2642 | case BPF_JGT: | |
b03c9f9f EC |
2643 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
2644 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2645 | break; | |
48461135 | 2646 | case BPF_JSGT: |
b03c9f9f EC |
2647 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
2648 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 2649 | break; |
b4e432f1 DB |
2650 | case BPF_JLT: |
2651 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2652 | false_reg->umax_value = min(false_reg->umax_value, val); | |
2653 | break; | |
2654 | case BPF_JSLT: | |
2655 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
2656 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
2657 | break; | |
48461135 | 2658 | case BPF_JGE: |
b03c9f9f EC |
2659 | true_reg->umax_value = min(true_reg->umax_value, val); |
2660 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2661 | break; | |
48461135 | 2662 | case BPF_JSGE: |
b03c9f9f EC |
2663 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
2664 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 2665 | break; |
b4e432f1 DB |
2666 | case BPF_JLE: |
2667 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2668 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
2669 | break; | |
2670 | case BPF_JSLE: | |
2671 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
2672 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
2673 | break; | |
48461135 JB |
2674 | default: |
2675 | break; | |
2676 | } | |
2677 | ||
b03c9f9f EC |
2678 | __reg_deduce_bounds(false_reg); |
2679 | __reg_deduce_bounds(true_reg); | |
2680 | /* We might have learned some bits from the bounds. */ | |
2681 | __reg_bound_offset(false_reg); | |
2682 | __reg_bound_offset(true_reg); | |
2683 | /* Intersecting with the old var_off might have improved our bounds | |
2684 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2685 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2686 | */ | |
2687 | __update_reg_bounds(false_reg); | |
2688 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
2689 | } |
2690 | ||
2691 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
2692 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
2693 | struct bpf_reg_state *dst_reg) | |
2694 | { | |
b03c9f9f EC |
2695 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
2696 | dst_reg->umin_value); | |
2697 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
2698 | dst_reg->umax_value); | |
2699 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
2700 | dst_reg->smin_value); | |
2701 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
2702 | dst_reg->smax_value); | |
f1174f77 EC |
2703 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
2704 | dst_reg->var_off); | |
b03c9f9f EC |
2705 | /* We might have learned new bounds from the var_off. */ |
2706 | __update_reg_bounds(src_reg); | |
2707 | __update_reg_bounds(dst_reg); | |
2708 | /* We might have learned something about the sign bit. */ | |
2709 | __reg_deduce_bounds(src_reg); | |
2710 | __reg_deduce_bounds(dst_reg); | |
2711 | /* We might have learned some bits from the bounds. */ | |
2712 | __reg_bound_offset(src_reg); | |
2713 | __reg_bound_offset(dst_reg); | |
2714 | /* Intersecting with the old var_off might have improved our bounds | |
2715 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2716 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2717 | */ | |
2718 | __update_reg_bounds(src_reg); | |
2719 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2720 | } |
2721 | ||
2722 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
2723 | struct bpf_reg_state *true_dst, | |
2724 | struct bpf_reg_state *false_src, | |
2725 | struct bpf_reg_state *false_dst, | |
2726 | u8 opcode) | |
2727 | { | |
2728 | switch (opcode) { | |
2729 | case BPF_JEQ: | |
2730 | __reg_combine_min_max(true_src, true_dst); | |
2731 | break; | |
2732 | case BPF_JNE: | |
2733 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 2734 | break; |
4cabc5b1 | 2735 | } |
48461135 JB |
2736 | } |
2737 | ||
57a09bf0 | 2738 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 2739 | bool is_null) |
57a09bf0 TG |
2740 | { |
2741 | struct bpf_reg_state *reg = ®s[regno]; | |
2742 | ||
2743 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
2744 | /* Old offset (both fixed and variable parts) should |
2745 | * have been known-zero, because we don't allow pointer | |
2746 | * arithmetic on pointers that might be NULL. | |
2747 | */ | |
b03c9f9f EC |
2748 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
2749 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 2750 | reg->off)) { |
b03c9f9f EC |
2751 | __mark_reg_known_zero(reg); |
2752 | reg->off = 0; | |
f1174f77 EC |
2753 | } |
2754 | if (is_null) { | |
2755 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
2756 | } else if (reg->map_ptr->inner_map_meta) { |
2757 | reg->type = CONST_PTR_TO_MAP; | |
2758 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
2759 | } else { | |
f1174f77 | 2760 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 2761 | } |
a08dd0da DB |
2762 | /* We don't need id from this point onwards anymore, thus we |
2763 | * should better reset it, so that state pruning has chances | |
2764 | * to take effect. | |
2765 | */ | |
2766 | reg->id = 0; | |
57a09bf0 TG |
2767 | } |
2768 | } | |
2769 | ||
2770 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
2771 | * be folded together at some point. | |
2772 | */ | |
2773 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
f1174f77 | 2774 | bool is_null) |
57a09bf0 TG |
2775 | { |
2776 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 2777 | u32 id = regs[regno].id; |
57a09bf0 TG |
2778 | int i; |
2779 | ||
2780 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 2781 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 | 2782 | |
638f5b90 AS |
2783 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2784 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
57a09bf0 | 2785 | continue; |
638f5b90 | 2786 | mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null); |
57a09bf0 TG |
2787 | } |
2788 | } | |
2789 | ||
5beca081 DB |
2790 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
2791 | struct bpf_reg_state *dst_reg, | |
2792 | struct bpf_reg_state *src_reg, | |
2793 | struct bpf_verifier_state *this_branch, | |
2794 | struct bpf_verifier_state *other_branch) | |
2795 | { | |
2796 | if (BPF_SRC(insn->code) != BPF_X) | |
2797 | return false; | |
2798 | ||
2799 | switch (BPF_OP(insn->code)) { | |
2800 | case BPF_JGT: | |
2801 | if ((dst_reg->type == PTR_TO_PACKET && | |
2802 | src_reg->type == PTR_TO_PACKET_END) || | |
2803 | (dst_reg->type == PTR_TO_PACKET_META && | |
2804 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2805 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
2806 | find_good_pkt_pointers(this_branch, dst_reg, | |
2807 | dst_reg->type, false); | |
2808 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2809 | src_reg->type == PTR_TO_PACKET) || | |
2810 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2811 | src_reg->type == PTR_TO_PACKET_META)) { | |
2812 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
2813 | find_good_pkt_pointers(other_branch, src_reg, | |
2814 | src_reg->type, true); | |
2815 | } else { | |
2816 | return false; | |
2817 | } | |
2818 | break; | |
2819 | case BPF_JLT: | |
2820 | if ((dst_reg->type == PTR_TO_PACKET && | |
2821 | src_reg->type == PTR_TO_PACKET_END) || | |
2822 | (dst_reg->type == PTR_TO_PACKET_META && | |
2823 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2824 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
2825 | find_good_pkt_pointers(other_branch, dst_reg, | |
2826 | dst_reg->type, true); | |
2827 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2828 | src_reg->type == PTR_TO_PACKET) || | |
2829 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2830 | src_reg->type == PTR_TO_PACKET_META)) { | |
2831 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
2832 | find_good_pkt_pointers(this_branch, src_reg, | |
2833 | src_reg->type, false); | |
2834 | } else { | |
2835 | return false; | |
2836 | } | |
2837 | break; | |
2838 | case BPF_JGE: | |
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, true); | |
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, false); | |
2853 | } else { | |
2854 | return false; | |
2855 | } | |
2856 | break; | |
2857 | case BPF_JLE: | |
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, false); | |
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, true); | |
2872 | } else { | |
2873 | return false; | |
2874 | } | |
2875 | break; | |
2876 | default: | |
2877 | return false; | |
2878 | } | |
2879 | ||
2880 | return true; | |
2881 | } | |
2882 | ||
58e2af8b | 2883 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
2884 | struct bpf_insn *insn, int *insn_idx) |
2885 | { | |
638f5b90 | 2886 | struct bpf_verifier_state *other_branch, *this_branch = env->cur_state; |
58e2af8b | 2887 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; |
17a52670 AS |
2888 | u8 opcode = BPF_OP(insn->code); |
2889 | int err; | |
2890 | ||
b4e432f1 | 2891 | if (opcode > BPF_JSLE) { |
61bd5218 | 2892 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
2893 | return -EINVAL; |
2894 | } | |
2895 | ||
2896 | if (BPF_SRC(insn->code) == BPF_X) { | |
2897 | if (insn->imm != 0) { | |
61bd5218 | 2898 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
2899 | return -EINVAL; |
2900 | } | |
2901 | ||
2902 | /* check src1 operand */ | |
dc503a8a | 2903 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2904 | if (err) |
2905 | return err; | |
1be7f75d AS |
2906 | |
2907 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 2908 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
2909 | insn->src_reg); |
2910 | return -EACCES; | |
2911 | } | |
17a52670 AS |
2912 | } else { |
2913 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 2914 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
2915 | return -EINVAL; |
2916 | } | |
2917 | } | |
2918 | ||
2919 | /* check src2 operand */ | |
dc503a8a | 2920 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2921 | if (err) |
2922 | return err; | |
2923 | ||
1a0dc1ac AS |
2924 | dst_reg = ®s[insn->dst_reg]; |
2925 | ||
17a52670 AS |
2926 | /* detect if R == 0 where R was initialized to zero earlier */ |
2927 | if (BPF_SRC(insn->code) == BPF_K && | |
2928 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
f1174f77 EC |
2929 | dst_reg->type == SCALAR_VALUE && |
2930 | tnum_equals_const(dst_reg->var_off, insn->imm)) { | |
17a52670 AS |
2931 | if (opcode == BPF_JEQ) { |
2932 | /* if (imm == imm) goto pc+off; | |
2933 | * only follow the goto, ignore fall-through | |
2934 | */ | |
2935 | *insn_idx += insn->off; | |
2936 | return 0; | |
2937 | } else { | |
2938 | /* if (imm != imm) goto pc+off; | |
2939 | * only follow fall-through branch, since | |
2940 | * that's where the program will go | |
2941 | */ | |
2942 | return 0; | |
2943 | } | |
2944 | } | |
2945 | ||
2946 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
2947 | if (!other_branch) | |
2948 | return -EFAULT; | |
2949 | ||
48461135 JB |
2950 | /* detect if we are comparing against a constant value so we can adjust |
2951 | * our min/max values for our dst register. | |
f1174f77 EC |
2952 | * this is only legit if both are scalars (or pointers to the same |
2953 | * object, I suppose, but we don't support that right now), because | |
2954 | * otherwise the different base pointers mean the offsets aren't | |
2955 | * comparable. | |
48461135 JB |
2956 | */ |
2957 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
2958 | if (dst_reg->type == SCALAR_VALUE && |
2959 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
2960 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
2961 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2962 | dst_reg, regs[insn->src_reg].var_off.value, | |
2963 | opcode); | |
2964 | else if (tnum_is_const(dst_reg->var_off)) | |
2965 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
2966 | ®s[insn->src_reg], | |
2967 | dst_reg->var_off.value, opcode); | |
2968 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
2969 | /* Comparing for equality, we can combine knowledge */ | |
2970 | reg_combine_min_max(&other_branch->regs[insn->src_reg], | |
2971 | &other_branch->regs[insn->dst_reg], | |
2972 | ®s[insn->src_reg], | |
2973 | ®s[insn->dst_reg], opcode); | |
2974 | } | |
2975 | } else if (dst_reg->type == SCALAR_VALUE) { | |
48461135 JB |
2976 | reg_set_min_max(&other_branch->regs[insn->dst_reg], |
2977 | dst_reg, insn->imm, opcode); | |
2978 | } | |
2979 | ||
58e2af8b | 2980 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 2981 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
2982 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
2983 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
2984 | /* Mark all identical map registers in each branch as either |
2985 | * safe or unknown depending R == 0 or R != 0 conditional. | |
2986 | */ | |
f1174f77 EC |
2987 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
2988 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
5beca081 DB |
2989 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
2990 | this_branch, other_branch) && | |
2991 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
2992 | verbose(env, "R%d pointer comparison prohibited\n", |
2993 | insn->dst_reg); | |
1be7f75d | 2994 | return -EACCES; |
17a52670 | 2995 | } |
61bd5218 JK |
2996 | if (env->log.level) |
2997 | print_verifier_state(env, this_branch); | |
17a52670 AS |
2998 | return 0; |
2999 | } | |
3000 | ||
0246e64d AS |
3001 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
3002 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
3003 | { | |
3004 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
3005 | ||
3006 | return (struct bpf_map *) (unsigned long) imm64; | |
3007 | } | |
3008 | ||
17a52670 | 3009 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 3010 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3011 | { |
638f5b90 | 3012 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3013 | int err; |
3014 | ||
3015 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 3016 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
3017 | return -EINVAL; |
3018 | } | |
3019 | if (insn->off != 0) { | |
61bd5218 | 3020 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
3021 | return -EINVAL; |
3022 | } | |
3023 | ||
dc503a8a | 3024 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3025 | if (err) |
3026 | return err; | |
3027 | ||
6b173873 | 3028 | if (insn->src_reg == 0) { |
6b173873 JK |
3029 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
3030 | ||
f1174f77 | 3031 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 3032 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 3033 | return 0; |
6b173873 | 3034 | } |
17a52670 AS |
3035 | |
3036 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
3037 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
3038 | ||
3039 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
3040 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
3041 | return 0; | |
3042 | } | |
3043 | ||
96be4325 DB |
3044 | static bool may_access_skb(enum bpf_prog_type type) |
3045 | { | |
3046 | switch (type) { | |
3047 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
3048 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 3049 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
3050 | return true; |
3051 | default: | |
3052 | return false; | |
3053 | } | |
3054 | } | |
3055 | ||
ddd872bc AS |
3056 | /* verify safety of LD_ABS|LD_IND instructions: |
3057 | * - they can only appear in the programs where ctx == skb | |
3058 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
3059 | * preserve R6-R9, and store return value into R0 | |
3060 | * | |
3061 | * Implicit input: | |
3062 | * ctx == skb == R6 == CTX | |
3063 | * | |
3064 | * Explicit input: | |
3065 | * SRC == any register | |
3066 | * IMM == 32-bit immediate | |
3067 | * | |
3068 | * Output: | |
3069 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
3070 | */ | |
58e2af8b | 3071 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 3072 | { |
638f5b90 | 3073 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 3074 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
3075 | int i, err; |
3076 | ||
24701ece | 3077 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 3078 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
3079 | return -EINVAL; |
3080 | } | |
3081 | ||
3082 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 3083 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 3084 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 3085 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
3086 | return -EINVAL; |
3087 | } | |
3088 | ||
3089 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 3090 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
3091 | if (err) |
3092 | return err; | |
3093 | ||
3094 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
61bd5218 JK |
3095 | verbose(env, |
3096 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
3097 | return -EINVAL; |
3098 | } | |
3099 | ||
3100 | if (mode == BPF_IND) { | |
3101 | /* check explicit source operand */ | |
dc503a8a | 3102 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
3103 | if (err) |
3104 | return err; | |
3105 | } | |
3106 | ||
3107 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 3108 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3109 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3110 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3111 | } | |
ddd872bc AS |
3112 | |
3113 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
3114 | * the value fetched from the packet. |
3115 | * Already marked as written above. | |
ddd872bc | 3116 | */ |
61bd5218 | 3117 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
3118 | return 0; |
3119 | } | |
3120 | ||
390ee7e2 AS |
3121 | static int check_return_code(struct bpf_verifier_env *env) |
3122 | { | |
3123 | struct bpf_reg_state *reg; | |
3124 | struct tnum range = tnum_range(0, 1); | |
3125 | ||
3126 | switch (env->prog->type) { | |
3127 | case BPF_PROG_TYPE_CGROUP_SKB: | |
3128 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
3129 | case BPF_PROG_TYPE_SOCK_OPS: | |
3130 | break; | |
3131 | default: | |
3132 | return 0; | |
3133 | } | |
3134 | ||
638f5b90 | 3135 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 3136 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 3137 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
3138 | reg_type_str[reg->type]); |
3139 | return -EINVAL; | |
3140 | } | |
3141 | ||
3142 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 3143 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
3144 | if (!tnum_is_unknown(reg->var_off)) { |
3145 | char tn_buf[48]; | |
3146 | ||
3147 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3148 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 3149 | } else { |
61bd5218 | 3150 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 3151 | } |
61bd5218 | 3152 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
3153 | return -EINVAL; |
3154 | } | |
3155 | return 0; | |
3156 | } | |
3157 | ||
475fb78f AS |
3158 | /* non-recursive DFS pseudo code |
3159 | * 1 procedure DFS-iterative(G,v): | |
3160 | * 2 label v as discovered | |
3161 | * 3 let S be a stack | |
3162 | * 4 S.push(v) | |
3163 | * 5 while S is not empty | |
3164 | * 6 t <- S.pop() | |
3165 | * 7 if t is what we're looking for: | |
3166 | * 8 return t | |
3167 | * 9 for all edges e in G.adjacentEdges(t) do | |
3168 | * 10 if edge e is already labelled | |
3169 | * 11 continue with the next edge | |
3170 | * 12 w <- G.adjacentVertex(t,e) | |
3171 | * 13 if vertex w is not discovered and not explored | |
3172 | * 14 label e as tree-edge | |
3173 | * 15 label w as discovered | |
3174 | * 16 S.push(w) | |
3175 | * 17 continue at 5 | |
3176 | * 18 else if vertex w is discovered | |
3177 | * 19 label e as back-edge | |
3178 | * 20 else | |
3179 | * 21 // vertex w is explored | |
3180 | * 22 label e as forward- or cross-edge | |
3181 | * 23 label t as explored | |
3182 | * 24 S.pop() | |
3183 | * | |
3184 | * convention: | |
3185 | * 0x10 - discovered | |
3186 | * 0x11 - discovered and fall-through edge labelled | |
3187 | * 0x12 - discovered and fall-through and branch edges labelled | |
3188 | * 0x20 - explored | |
3189 | */ | |
3190 | ||
3191 | enum { | |
3192 | DISCOVERED = 0x10, | |
3193 | EXPLORED = 0x20, | |
3194 | FALLTHROUGH = 1, | |
3195 | BRANCH = 2, | |
3196 | }; | |
3197 | ||
58e2af8b | 3198 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 3199 | |
475fb78f AS |
3200 | static int *insn_stack; /* stack of insns to process */ |
3201 | static int cur_stack; /* current stack index */ | |
3202 | static int *insn_state; | |
3203 | ||
3204 | /* t, w, e - match pseudo-code above: | |
3205 | * t - index of current instruction | |
3206 | * w - next instruction | |
3207 | * e - edge | |
3208 | */ | |
58e2af8b | 3209 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
3210 | { |
3211 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
3212 | return 0; | |
3213 | ||
3214 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
3215 | return 0; | |
3216 | ||
3217 | if (w < 0 || w >= env->prog->len) { | |
61bd5218 | 3218 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
3219 | return -EINVAL; |
3220 | } | |
3221 | ||
f1bca824 AS |
3222 | if (e == BRANCH) |
3223 | /* mark branch target for state pruning */ | |
3224 | env->explored_states[w] = STATE_LIST_MARK; | |
3225 | ||
475fb78f AS |
3226 | if (insn_state[w] == 0) { |
3227 | /* tree-edge */ | |
3228 | insn_state[t] = DISCOVERED | e; | |
3229 | insn_state[w] = DISCOVERED; | |
3230 | if (cur_stack >= env->prog->len) | |
3231 | return -E2BIG; | |
3232 | insn_stack[cur_stack++] = w; | |
3233 | return 1; | |
3234 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
61bd5218 | 3235 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
3236 | return -EINVAL; |
3237 | } else if (insn_state[w] == EXPLORED) { | |
3238 | /* forward- or cross-edge */ | |
3239 | insn_state[t] = DISCOVERED | e; | |
3240 | } else { | |
61bd5218 | 3241 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
3242 | return -EFAULT; |
3243 | } | |
3244 | return 0; | |
3245 | } | |
3246 | ||
3247 | /* non-recursive depth-first-search to detect loops in BPF program | |
3248 | * loop == back-edge in directed graph | |
3249 | */ | |
58e2af8b | 3250 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
3251 | { |
3252 | struct bpf_insn *insns = env->prog->insnsi; | |
3253 | int insn_cnt = env->prog->len; | |
3254 | int ret = 0; | |
3255 | int i, t; | |
3256 | ||
3257 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3258 | if (!insn_state) | |
3259 | return -ENOMEM; | |
3260 | ||
3261 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3262 | if (!insn_stack) { | |
3263 | kfree(insn_state); | |
3264 | return -ENOMEM; | |
3265 | } | |
3266 | ||
3267 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
3268 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
3269 | cur_stack = 1; | |
3270 | ||
3271 | peek_stack: | |
3272 | if (cur_stack == 0) | |
3273 | goto check_state; | |
3274 | t = insn_stack[cur_stack - 1]; | |
3275 | ||
3276 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
3277 | u8 opcode = BPF_OP(insns[t].code); | |
3278 | ||
3279 | if (opcode == BPF_EXIT) { | |
3280 | goto mark_explored; | |
3281 | } else if (opcode == BPF_CALL) { | |
3282 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3283 | if (ret == 1) | |
3284 | goto peek_stack; | |
3285 | else if (ret < 0) | |
3286 | goto err_free; | |
07016151 DB |
3287 | if (t + 1 < insn_cnt) |
3288 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3289 | } else if (opcode == BPF_JA) { |
3290 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
3291 | ret = -EINVAL; | |
3292 | goto err_free; | |
3293 | } | |
3294 | /* unconditional jump with single edge */ | |
3295 | ret = push_insn(t, t + insns[t].off + 1, | |
3296 | FALLTHROUGH, env); | |
3297 | if (ret == 1) | |
3298 | goto peek_stack; | |
3299 | else if (ret < 0) | |
3300 | goto err_free; | |
f1bca824 AS |
3301 | /* tell verifier to check for equivalent states |
3302 | * after every call and jump | |
3303 | */ | |
c3de6317 AS |
3304 | if (t + 1 < insn_cnt) |
3305 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3306 | } else { |
3307 | /* conditional jump with two edges */ | |
3c2ce60b | 3308 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
3309 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
3310 | if (ret == 1) | |
3311 | goto peek_stack; | |
3312 | else if (ret < 0) | |
3313 | goto err_free; | |
3314 | ||
3315 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
3316 | if (ret == 1) | |
3317 | goto peek_stack; | |
3318 | else if (ret < 0) | |
3319 | goto err_free; | |
3320 | } | |
3321 | } else { | |
3322 | /* all other non-branch instructions with single | |
3323 | * fall-through edge | |
3324 | */ | |
3325 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3326 | if (ret == 1) | |
3327 | goto peek_stack; | |
3328 | else if (ret < 0) | |
3329 | goto err_free; | |
3330 | } | |
3331 | ||
3332 | mark_explored: | |
3333 | insn_state[t] = EXPLORED; | |
3334 | if (cur_stack-- <= 0) { | |
61bd5218 | 3335 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
3336 | ret = -EFAULT; |
3337 | goto err_free; | |
3338 | } | |
3339 | goto peek_stack; | |
3340 | ||
3341 | check_state: | |
3342 | for (i = 0; i < insn_cnt; i++) { | |
3343 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 3344 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
3345 | ret = -EINVAL; |
3346 | goto err_free; | |
3347 | } | |
3348 | } | |
3349 | ret = 0; /* cfg looks good */ | |
3350 | ||
3351 | err_free: | |
3352 | kfree(insn_state); | |
3353 | kfree(insn_stack); | |
3354 | return ret; | |
3355 | } | |
3356 | ||
f1174f77 EC |
3357 | /* check %cur's range satisfies %old's */ |
3358 | static bool range_within(struct bpf_reg_state *old, | |
3359 | struct bpf_reg_state *cur) | |
3360 | { | |
b03c9f9f EC |
3361 | return old->umin_value <= cur->umin_value && |
3362 | old->umax_value >= cur->umax_value && | |
3363 | old->smin_value <= cur->smin_value && | |
3364 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
3365 | } |
3366 | ||
3367 | /* Maximum number of register states that can exist at once */ | |
3368 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
3369 | struct idpair { | |
3370 | u32 old; | |
3371 | u32 cur; | |
3372 | }; | |
3373 | ||
3374 | /* If in the old state two registers had the same id, then they need to have | |
3375 | * the same id in the new state as well. But that id could be different from | |
3376 | * the old state, so we need to track the mapping from old to new ids. | |
3377 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
3378 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
3379 | * regs with a different old id could still have new id 9, we don't care about | |
3380 | * that. | |
3381 | * So we look through our idmap to see if this old id has been seen before. If | |
3382 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 3383 | */ |
f1174f77 | 3384 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 3385 | { |
f1174f77 | 3386 | unsigned int i; |
969bf05e | 3387 | |
f1174f77 EC |
3388 | for (i = 0; i < ID_MAP_SIZE; i++) { |
3389 | if (!idmap[i].old) { | |
3390 | /* Reached an empty slot; haven't seen this id before */ | |
3391 | idmap[i].old = old_id; | |
3392 | idmap[i].cur = cur_id; | |
3393 | return true; | |
3394 | } | |
3395 | if (idmap[i].old == old_id) | |
3396 | return idmap[i].cur == cur_id; | |
3397 | } | |
3398 | /* We ran out of idmap slots, which should be impossible */ | |
3399 | WARN_ON_ONCE(1); | |
3400 | return false; | |
3401 | } | |
3402 | ||
3403 | /* Returns true if (rold safe implies rcur safe) */ | |
1b688a19 EC |
3404 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
3405 | struct idpair *idmap) | |
f1174f77 | 3406 | { |
dc503a8a EC |
3407 | if (!(rold->live & REG_LIVE_READ)) |
3408 | /* explored state didn't use this */ | |
3409 | return true; | |
3410 | ||
3411 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, live)) == 0) | |
969bf05e AS |
3412 | return true; |
3413 | ||
f1174f77 EC |
3414 | if (rold->type == NOT_INIT) |
3415 | /* explored state can't have used this */ | |
969bf05e | 3416 | return true; |
f1174f77 EC |
3417 | if (rcur->type == NOT_INIT) |
3418 | return false; | |
3419 | switch (rold->type) { | |
3420 | case SCALAR_VALUE: | |
3421 | if (rcur->type == SCALAR_VALUE) { | |
3422 | /* new val must satisfy old val knowledge */ | |
3423 | return range_within(rold, rcur) && | |
3424 | tnum_in(rold->var_off, rcur->var_off); | |
3425 | } else { | |
3426 | /* if we knew anything about the old value, we're not | |
3427 | * equal, because we can't know anything about the | |
3428 | * scalar value of the pointer in the new value. | |
3429 | */ | |
b03c9f9f EC |
3430 | return rold->umin_value == 0 && |
3431 | rold->umax_value == U64_MAX && | |
3432 | rold->smin_value == S64_MIN && | |
3433 | rold->smax_value == S64_MAX && | |
f1174f77 EC |
3434 | tnum_is_unknown(rold->var_off); |
3435 | } | |
3436 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
3437 | /* If the new min/max/var_off satisfy the old ones and |
3438 | * everything else matches, we are OK. | |
3439 | * We don't care about the 'id' value, because nothing | |
3440 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
3441 | */ | |
3442 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
3443 | range_within(rold, rcur) && | |
3444 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
3445 | case PTR_TO_MAP_VALUE_OR_NULL: |
3446 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
3447 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
3448 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
3449 | * checked, doing so could have affected others with the same | |
3450 | * id, and we can't check for that because we lost the id when | |
3451 | * we converted to a PTR_TO_MAP_VALUE. | |
3452 | */ | |
3453 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
3454 | return false; | |
3455 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
3456 | return false; | |
3457 | /* Check our ids match any regs they're supposed to */ | |
3458 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 3459 | case PTR_TO_PACKET_META: |
f1174f77 | 3460 | case PTR_TO_PACKET: |
de8f3a83 | 3461 | if (rcur->type != rold->type) |
f1174f77 EC |
3462 | return false; |
3463 | /* We must have at least as much range as the old ptr | |
3464 | * did, so that any accesses which were safe before are | |
3465 | * still safe. This is true even if old range < old off, | |
3466 | * since someone could have accessed through (ptr - k), or | |
3467 | * even done ptr -= k in a register, to get a safe access. | |
3468 | */ | |
3469 | if (rold->range > rcur->range) | |
3470 | return false; | |
3471 | /* If the offsets don't match, we can't trust our alignment; | |
3472 | * nor can we be sure that we won't fall out of range. | |
3473 | */ | |
3474 | if (rold->off != rcur->off) | |
3475 | return false; | |
3476 | /* id relations must be preserved */ | |
3477 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
3478 | return false; | |
3479 | /* new val must satisfy old val knowledge */ | |
3480 | return range_within(rold, rcur) && | |
3481 | tnum_in(rold->var_off, rcur->var_off); | |
3482 | case PTR_TO_CTX: | |
3483 | case CONST_PTR_TO_MAP: | |
3484 | case PTR_TO_STACK: | |
3485 | case PTR_TO_PACKET_END: | |
3486 | /* Only valid matches are exact, which memcmp() above | |
3487 | * would have accepted | |
3488 | */ | |
3489 | default: | |
3490 | /* Don't know what's going on, just say it's not safe */ | |
3491 | return false; | |
3492 | } | |
969bf05e | 3493 | |
f1174f77 EC |
3494 | /* Shouldn't get here; if we do, say it's not safe */ |
3495 | WARN_ON_ONCE(1); | |
969bf05e AS |
3496 | return false; |
3497 | } | |
3498 | ||
638f5b90 AS |
3499 | static bool stacksafe(struct bpf_verifier_state *old, |
3500 | struct bpf_verifier_state *cur, | |
3501 | struct idpair *idmap) | |
3502 | { | |
3503 | int i, spi; | |
3504 | ||
3505 | /* if explored stack has more populated slots than current stack | |
3506 | * such stacks are not equivalent | |
3507 | */ | |
3508 | if (old->allocated_stack > cur->allocated_stack) | |
3509 | return false; | |
3510 | ||
3511 | /* walk slots of the explored stack and ignore any additional | |
3512 | * slots in the current stack, since explored(safe) state | |
3513 | * didn't use them | |
3514 | */ | |
3515 | for (i = 0; i < old->allocated_stack; i++) { | |
3516 | spi = i / BPF_REG_SIZE; | |
3517 | ||
3518 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) | |
3519 | continue; | |
3520 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != | |
3521 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
3522 | /* Ex: old explored (safe) state has STACK_SPILL in | |
3523 | * this stack slot, but current has has STACK_MISC -> | |
3524 | * this verifier states are not equivalent, | |
3525 | * return false to continue verification of this path | |
3526 | */ | |
3527 | return false; | |
3528 | if (i % BPF_REG_SIZE) | |
3529 | continue; | |
3530 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
3531 | continue; | |
3532 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
3533 | &cur->stack[spi].spilled_ptr, | |
3534 | idmap)) | |
3535 | /* when explored and current stack slot are both storing | |
3536 | * spilled registers, check that stored pointers types | |
3537 | * are the same as well. | |
3538 | * Ex: explored safe path could have stored | |
3539 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
3540 | * but current path has stored: | |
3541 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
3542 | * such verifier states are not equivalent. | |
3543 | * return false to continue verification of this path | |
3544 | */ | |
3545 | return false; | |
3546 | } | |
3547 | return true; | |
3548 | } | |
3549 | ||
f1bca824 AS |
3550 | /* compare two verifier states |
3551 | * | |
3552 | * all states stored in state_list are known to be valid, since | |
3553 | * verifier reached 'bpf_exit' instruction through them | |
3554 | * | |
3555 | * this function is called when verifier exploring different branches of | |
3556 | * execution popped from the state stack. If it sees an old state that has | |
3557 | * more strict register state and more strict stack state then this execution | |
3558 | * branch doesn't need to be explored further, since verifier already | |
3559 | * concluded that more strict state leads to valid finish. | |
3560 | * | |
3561 | * Therefore two states are equivalent if register state is more conservative | |
3562 | * and explored stack state is more conservative than the current one. | |
3563 | * Example: | |
3564 | * explored current | |
3565 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
3566 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
3567 | * | |
3568 | * In other words if current stack state (one being explored) has more | |
3569 | * valid slots than old one that already passed validation, it means | |
3570 | * the verifier can stop exploring and conclude that current state is valid too | |
3571 | * | |
3572 | * Similarly with registers. If explored state has register type as invalid | |
3573 | * whereas register type in current state is meaningful, it means that | |
3574 | * the current state will reach 'bpf_exit' instruction safely | |
3575 | */ | |
48461135 JB |
3576 | static bool states_equal(struct bpf_verifier_env *env, |
3577 | struct bpf_verifier_state *old, | |
58e2af8b | 3578 | struct bpf_verifier_state *cur) |
f1bca824 | 3579 | { |
f1174f77 EC |
3580 | struct idpair *idmap; |
3581 | bool ret = false; | |
f1bca824 AS |
3582 | int i; |
3583 | ||
f1174f77 EC |
3584 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
3585 | /* If we failed to allocate the idmap, just say it's not safe */ | |
3586 | if (!idmap) | |
1a0dc1ac | 3587 | return false; |
f1174f77 EC |
3588 | |
3589 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 3590 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 3591 | goto out_free; |
f1bca824 AS |
3592 | } |
3593 | ||
638f5b90 AS |
3594 | if (!stacksafe(old, cur, idmap)) |
3595 | goto out_free; | |
f1174f77 EC |
3596 | ret = true; |
3597 | out_free: | |
3598 | kfree(idmap); | |
3599 | return ret; | |
f1bca824 AS |
3600 | } |
3601 | ||
8e9cd9ce EC |
3602 | /* A write screens off any subsequent reads; but write marks come from the |
3603 | * straight-line code between a state and its parent. When we arrive at a | |
3604 | * jump target (in the first iteration of the propagate_liveness() loop), | |
3605 | * we didn't arrive by the straight-line code, so read marks in state must | |
3606 | * propagate to parent regardless of state's write marks. | |
3607 | */ | |
dc503a8a EC |
3608 | static bool do_propagate_liveness(const struct bpf_verifier_state *state, |
3609 | struct bpf_verifier_state *parent) | |
3610 | { | |
63f45f84 | 3611 | bool writes = parent == state->parent; /* Observe write marks */ |
dc503a8a EC |
3612 | bool touched = false; /* any changes made? */ |
3613 | int i; | |
3614 | ||
3615 | if (!parent) | |
3616 | return touched; | |
3617 | /* Propagate read liveness of registers... */ | |
3618 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
3619 | /* We don't need to worry about FP liveness because it's read-only */ | |
3620 | for (i = 0; i < BPF_REG_FP; i++) { | |
3621 | if (parent->regs[i].live & REG_LIVE_READ) | |
3622 | continue; | |
63f45f84 EC |
3623 | if (writes && (state->regs[i].live & REG_LIVE_WRITTEN)) |
3624 | continue; | |
3625 | if (state->regs[i].live & REG_LIVE_READ) { | |
dc503a8a EC |
3626 | parent->regs[i].live |= REG_LIVE_READ; |
3627 | touched = true; | |
3628 | } | |
3629 | } | |
3630 | /* ... and stack slots */ | |
638f5b90 AS |
3631 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
3632 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3633 | if (parent->stack[i].slot_type[0] != STACK_SPILL) | |
dc503a8a | 3634 | continue; |
638f5b90 | 3635 | if (state->stack[i].slot_type[0] != STACK_SPILL) |
dc503a8a | 3636 | continue; |
638f5b90 | 3637 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
dc503a8a | 3638 | continue; |
638f5b90 AS |
3639 | if (writes && |
3640 | (state->stack[i].spilled_ptr.live & REG_LIVE_WRITTEN)) | |
63f45f84 | 3641 | continue; |
638f5b90 AS |
3642 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) { |
3643 | parent->stack[i].spilled_ptr.live |= REG_LIVE_READ; | |
dc503a8a EC |
3644 | touched = true; |
3645 | } | |
3646 | } | |
3647 | return touched; | |
3648 | } | |
3649 | ||
8e9cd9ce EC |
3650 | /* "parent" is "a state from which we reach the current state", but initially |
3651 | * it is not the state->parent (i.e. "the state whose straight-line code leads | |
3652 | * to the current state"), instead it is the state that happened to arrive at | |
3653 | * a (prunable) equivalent of the current state. See comment above | |
3654 | * do_propagate_liveness() for consequences of this. | |
3655 | * This function is just a more efficient way of calling mark_reg_read() or | |
3656 | * mark_stack_slot_read() on each reg in "parent" that is read in "state", | |
3657 | * though it requires that parent != state->parent in the call arguments. | |
3658 | */ | |
dc503a8a EC |
3659 | static void propagate_liveness(const struct bpf_verifier_state *state, |
3660 | struct bpf_verifier_state *parent) | |
3661 | { | |
3662 | while (do_propagate_liveness(state, parent)) { | |
3663 | /* Something changed, so we need to feed those changes onward */ | |
3664 | state = parent; | |
3665 | parent = state->parent; | |
3666 | } | |
3667 | } | |
3668 | ||
58e2af8b | 3669 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 3670 | { |
58e2af8b JK |
3671 | struct bpf_verifier_state_list *new_sl; |
3672 | struct bpf_verifier_state_list *sl; | |
638f5b90 | 3673 | struct bpf_verifier_state *cur = env->cur_state; |
1969db47 | 3674 | int i, err; |
f1bca824 AS |
3675 | |
3676 | sl = env->explored_states[insn_idx]; | |
3677 | if (!sl) | |
3678 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
3679 | * be doing state search here | |
3680 | */ | |
3681 | return 0; | |
3682 | ||
3683 | while (sl != STATE_LIST_MARK) { | |
638f5b90 | 3684 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 3685 | /* reached equivalent register/stack state, |
dc503a8a EC |
3686 | * prune the search. |
3687 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
3688 | * If we have any write marks in env->cur_state, they |
3689 | * will prevent corresponding reads in the continuation | |
3690 | * from reaching our parent (an explored_state). Our | |
3691 | * own state will get the read marks recorded, but | |
3692 | * they'll be immediately forgotten as we're pruning | |
3693 | * this state and will pop a new one. | |
f1bca824 | 3694 | */ |
638f5b90 | 3695 | propagate_liveness(&sl->state, cur); |
f1bca824 | 3696 | return 1; |
dc503a8a | 3697 | } |
f1bca824 AS |
3698 | sl = sl->next; |
3699 | } | |
3700 | ||
3701 | /* there were no equivalent states, remember current one. | |
3702 | * technically the current state is not proven to be safe yet, | |
3703 | * but it will either reach bpf_exit (which means it's safe) or | |
3704 | * it will be rejected. Since there are no loops, we won't be | |
3705 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
3706 | */ | |
638f5b90 | 3707 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
3708 | if (!new_sl) |
3709 | return -ENOMEM; | |
3710 | ||
3711 | /* add new state to the head of linked list */ | |
1969db47 AS |
3712 | err = copy_verifier_state(&new_sl->state, cur); |
3713 | if (err) { | |
3714 | free_verifier_state(&new_sl->state, false); | |
3715 | kfree(new_sl); | |
3716 | return err; | |
3717 | } | |
f1bca824 AS |
3718 | new_sl->next = env->explored_states[insn_idx]; |
3719 | env->explored_states[insn_idx] = new_sl; | |
dc503a8a | 3720 | /* connect new state to parentage chain */ |
638f5b90 | 3721 | cur->parent = &new_sl->state; |
8e9cd9ce EC |
3722 | /* clear write marks in current state: the writes we did are not writes |
3723 | * our child did, so they don't screen off its reads from us. | |
3724 | * (There are no read marks in current state, because reads always mark | |
3725 | * their parent and current state never has children yet. Only | |
3726 | * explored_states can get read marks.) | |
3727 | */ | |
dc503a8a | 3728 | for (i = 0; i < BPF_REG_FP; i++) |
638f5b90 AS |
3729 | cur->regs[i].live = REG_LIVE_NONE; |
3730 | for (i = 0; i < cur->allocated_stack / BPF_REG_SIZE; i++) | |
3731 | if (cur->stack[i].slot_type[0] == STACK_SPILL) | |
3732 | cur->stack[i].spilled_ptr.live = REG_LIVE_NONE; | |
f1bca824 AS |
3733 | return 0; |
3734 | } | |
3735 | ||
13a27dfc JK |
3736 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
3737 | int insn_idx, int prev_insn_idx) | |
3738 | { | |
ab3f0063 JK |
3739 | if (env->analyzer_ops && env->analyzer_ops->insn_hook) |
3740 | return env->analyzer_ops->insn_hook(env, insn_idx, | |
3741 | prev_insn_idx); | |
3742 | if (env->dev_ops && env->dev_ops->insn_hook) | |
3743 | return env->dev_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
13a27dfc | 3744 | |
ab3f0063 | 3745 | return 0; |
13a27dfc JK |
3746 | } |
3747 | ||
58e2af8b | 3748 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 3749 | { |
638f5b90 | 3750 | struct bpf_verifier_state *state; |
17a52670 | 3751 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 3752 | struct bpf_reg_state *regs; |
17a52670 AS |
3753 | int insn_cnt = env->prog->len; |
3754 | int insn_idx, prev_insn_idx = 0; | |
3755 | int insn_processed = 0; | |
3756 | bool do_print_state = false; | |
3757 | ||
638f5b90 AS |
3758 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
3759 | if (!state) | |
3760 | return -ENOMEM; | |
3761 | env->cur_state = state; | |
3762 | init_reg_state(env, state->regs); | |
dc503a8a | 3763 | state->parent = NULL; |
17a52670 AS |
3764 | insn_idx = 0; |
3765 | for (;;) { | |
3766 | struct bpf_insn *insn; | |
3767 | u8 class; | |
3768 | int err; | |
3769 | ||
3770 | if (insn_idx >= insn_cnt) { | |
61bd5218 | 3771 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
17a52670 AS |
3772 | insn_idx, insn_cnt); |
3773 | return -EFAULT; | |
3774 | } | |
3775 | ||
3776 | insn = &insns[insn_idx]; | |
3777 | class = BPF_CLASS(insn->code); | |
3778 | ||
07016151 | 3779 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
3780 | verbose(env, |
3781 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
3782 | insn_processed); |
3783 | return -E2BIG; | |
3784 | } | |
3785 | ||
f1bca824 AS |
3786 | err = is_state_visited(env, insn_idx); |
3787 | if (err < 0) | |
3788 | return err; | |
3789 | if (err == 1) { | |
3790 | /* found equivalent state, can prune the search */ | |
61bd5218 | 3791 | if (env->log.level) { |
f1bca824 | 3792 | if (do_print_state) |
61bd5218 | 3793 | verbose(env, "\nfrom %d to %d: safe\n", |
f1bca824 AS |
3794 | prev_insn_idx, insn_idx); |
3795 | else | |
61bd5218 | 3796 | verbose(env, "%d: safe\n", insn_idx); |
f1bca824 AS |
3797 | } |
3798 | goto process_bpf_exit; | |
3799 | } | |
3800 | ||
3c2ce60b DB |
3801 | if (need_resched()) |
3802 | cond_resched(); | |
3803 | ||
61bd5218 JK |
3804 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
3805 | if (env->log.level > 1) | |
3806 | verbose(env, "%d:", insn_idx); | |
c5fc9692 | 3807 | else |
61bd5218 | 3808 | verbose(env, "\nfrom %d to %d:", |
c5fc9692 | 3809 | prev_insn_idx, insn_idx); |
638f5b90 | 3810 | print_verifier_state(env, state); |
17a52670 AS |
3811 | do_print_state = false; |
3812 | } | |
3813 | ||
61bd5218 JK |
3814 | if (env->log.level) { |
3815 | verbose(env, "%d: ", insn_idx); | |
f4ac7e0b JK |
3816 | print_bpf_insn(verbose, env, insn, |
3817 | env->allow_ptr_leaks); | |
17a52670 AS |
3818 | } |
3819 | ||
13a27dfc JK |
3820 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
3821 | if (err) | |
3822 | return err; | |
3823 | ||
638f5b90 | 3824 | regs = cur_regs(env); |
17a52670 | 3825 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 3826 | err = check_alu_op(env, insn); |
17a52670 AS |
3827 | if (err) |
3828 | return err; | |
3829 | ||
3830 | } else if (class == BPF_LDX) { | |
3df126f3 | 3831 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
3832 | |
3833 | /* check for reserved fields is already done */ | |
3834 | ||
17a52670 | 3835 | /* check src operand */ |
dc503a8a | 3836 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3837 | if (err) |
3838 | return err; | |
3839 | ||
dc503a8a | 3840 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
3841 | if (err) |
3842 | return err; | |
3843 | ||
725f9dcd AS |
3844 | src_reg_type = regs[insn->src_reg].type; |
3845 | ||
17a52670 AS |
3846 | /* check that memory (src_reg + off) is readable, |
3847 | * the state of dst_reg will be updated by this func | |
3848 | */ | |
31fd8581 | 3849 | err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, |
17a52670 AS |
3850 | BPF_SIZE(insn->code), BPF_READ, |
3851 | insn->dst_reg); | |
3852 | if (err) | |
3853 | return err; | |
3854 | ||
3df126f3 JK |
3855 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
3856 | ||
3857 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
3858 | /* saw a valid insn |
3859 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 3860 | * save type to validate intersecting paths |
9bac3d6d | 3861 | */ |
3df126f3 | 3862 | *prev_src_type = src_reg_type; |
9bac3d6d | 3863 | |
3df126f3 | 3864 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 3865 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 3866 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
3867 | /* ABuser program is trying to use the same insn |
3868 | * dst_reg = *(u32*) (src_reg + off) | |
3869 | * with different pointer types: | |
3870 | * src_reg == ctx in one branch and | |
3871 | * src_reg == stack|map in some other branch. | |
3872 | * Reject it. | |
3873 | */ | |
61bd5218 | 3874 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
3875 | return -EINVAL; |
3876 | } | |
3877 | ||
17a52670 | 3878 | } else if (class == BPF_STX) { |
3df126f3 | 3879 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 3880 | |
17a52670 | 3881 | if (BPF_MODE(insn->code) == BPF_XADD) { |
31fd8581 | 3882 | err = check_xadd(env, insn_idx, insn); |
17a52670 AS |
3883 | if (err) |
3884 | return err; | |
3885 | insn_idx++; | |
3886 | continue; | |
3887 | } | |
3888 | ||
17a52670 | 3889 | /* check src1 operand */ |
dc503a8a | 3890 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3891 | if (err) |
3892 | return err; | |
3893 | /* check src2 operand */ | |
dc503a8a | 3894 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3895 | if (err) |
3896 | return err; | |
3897 | ||
d691f9e8 AS |
3898 | dst_reg_type = regs[insn->dst_reg].type; |
3899 | ||
17a52670 | 3900 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 3901 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3902 | BPF_SIZE(insn->code), BPF_WRITE, |
3903 | insn->src_reg); | |
3904 | if (err) | |
3905 | return err; | |
3906 | ||
3df126f3 JK |
3907 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
3908 | ||
3909 | if (*prev_dst_type == NOT_INIT) { | |
3910 | *prev_dst_type = dst_reg_type; | |
3911 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 3912 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 3913 | *prev_dst_type == PTR_TO_CTX)) { |
61bd5218 | 3914 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
3915 | return -EINVAL; |
3916 | } | |
3917 | ||
17a52670 AS |
3918 | } else if (class == BPF_ST) { |
3919 | if (BPF_MODE(insn->code) != BPF_MEM || | |
3920 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 3921 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
3922 | return -EINVAL; |
3923 | } | |
3924 | /* check src operand */ | |
dc503a8a | 3925 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3926 | if (err) |
3927 | return err; | |
3928 | ||
3929 | /* check that memory (dst_reg + off) is writeable */ | |
31fd8581 | 3930 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 AS |
3931 | BPF_SIZE(insn->code), BPF_WRITE, |
3932 | -1); | |
3933 | if (err) | |
3934 | return err; | |
3935 | ||
3936 | } else if (class == BPF_JMP) { | |
3937 | u8 opcode = BPF_OP(insn->code); | |
3938 | ||
3939 | if (opcode == BPF_CALL) { | |
3940 | if (BPF_SRC(insn->code) != BPF_K || | |
3941 | insn->off != 0 || | |
3942 | insn->src_reg != BPF_REG_0 || | |
3943 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 3944 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
3945 | return -EINVAL; |
3946 | } | |
3947 | ||
81ed18ab | 3948 | err = check_call(env, insn->imm, insn_idx); |
17a52670 AS |
3949 | if (err) |
3950 | return err; | |
3951 | ||
3952 | } else if (opcode == BPF_JA) { | |
3953 | if (BPF_SRC(insn->code) != BPF_K || | |
3954 | insn->imm != 0 || | |
3955 | insn->src_reg != BPF_REG_0 || | |
3956 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 3957 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
3958 | return -EINVAL; |
3959 | } | |
3960 | ||
3961 | insn_idx += insn->off + 1; | |
3962 | continue; | |
3963 | ||
3964 | } else if (opcode == BPF_EXIT) { | |
3965 | if (BPF_SRC(insn->code) != BPF_K || | |
3966 | insn->imm != 0 || | |
3967 | insn->src_reg != BPF_REG_0 || | |
3968 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 3969 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
3970 | return -EINVAL; |
3971 | } | |
3972 | ||
3973 | /* eBPF calling convetion is such that R0 is used | |
3974 | * to return the value from eBPF program. | |
3975 | * Make sure that it's readable at this time | |
3976 | * of bpf_exit, which means that program wrote | |
3977 | * something into it earlier | |
3978 | */ | |
dc503a8a | 3979 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
3980 | if (err) |
3981 | return err; | |
3982 | ||
1be7f75d | 3983 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 3984 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
3985 | return -EACCES; |
3986 | } | |
3987 | ||
390ee7e2 AS |
3988 | err = check_return_code(env); |
3989 | if (err) | |
3990 | return err; | |
f1bca824 | 3991 | process_bpf_exit: |
638f5b90 AS |
3992 | err = pop_stack(env, &prev_insn_idx, &insn_idx); |
3993 | if (err < 0) { | |
3994 | if (err != -ENOENT) | |
3995 | return err; | |
17a52670 AS |
3996 | break; |
3997 | } else { | |
3998 | do_print_state = true; | |
3999 | continue; | |
4000 | } | |
4001 | } else { | |
4002 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
4003 | if (err) | |
4004 | return err; | |
4005 | } | |
4006 | } else if (class == BPF_LD) { | |
4007 | u8 mode = BPF_MODE(insn->code); | |
4008 | ||
4009 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
4010 | err = check_ld_abs(env, insn); |
4011 | if (err) | |
4012 | return err; | |
4013 | ||
17a52670 AS |
4014 | } else if (mode == BPF_IMM) { |
4015 | err = check_ld_imm(env, insn); | |
4016 | if (err) | |
4017 | return err; | |
4018 | ||
4019 | insn_idx++; | |
4020 | } else { | |
61bd5218 | 4021 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
4022 | return -EINVAL; |
4023 | } | |
4024 | } else { | |
61bd5218 | 4025 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
4026 | return -EINVAL; |
4027 | } | |
4028 | ||
4029 | insn_idx++; | |
4030 | } | |
4031 | ||
61bd5218 JK |
4032 | verbose(env, "processed %d insns, stack depth %d\n", insn_processed, |
4033 | env->prog->aux->stack_depth); | |
17a52670 AS |
4034 | return 0; |
4035 | } | |
4036 | ||
56f668df MKL |
4037 | static int check_map_prealloc(struct bpf_map *map) |
4038 | { | |
4039 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
4040 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
4041 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
4042 | !(map->map_flags & BPF_F_NO_PREALLOC); |
4043 | } | |
4044 | ||
61bd5218 JK |
4045 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
4046 | struct bpf_map *map, | |
fdc15d38 AS |
4047 | struct bpf_prog *prog) |
4048 | ||
4049 | { | |
56f668df MKL |
4050 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
4051 | * preallocated hash maps, since doing memory allocation | |
4052 | * in overflow_handler can crash depending on where nmi got | |
4053 | * triggered. | |
4054 | */ | |
4055 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
4056 | if (!check_map_prealloc(map)) { | |
61bd5218 | 4057 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
4058 | return -EINVAL; |
4059 | } | |
4060 | if (map->inner_map_meta && | |
4061 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 4062 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
4063 | return -EINVAL; |
4064 | } | |
fdc15d38 AS |
4065 | } |
4066 | return 0; | |
4067 | } | |
4068 | ||
0246e64d AS |
4069 | /* look for pseudo eBPF instructions that access map FDs and |
4070 | * replace them with actual map pointers | |
4071 | */ | |
58e2af8b | 4072 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
4073 | { |
4074 | struct bpf_insn *insn = env->prog->insnsi; | |
4075 | int insn_cnt = env->prog->len; | |
fdc15d38 | 4076 | int i, j, err; |
0246e64d | 4077 | |
f1f7714e | 4078 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
4079 | if (err) |
4080 | return err; | |
4081 | ||
0246e64d | 4082 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 4083 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 4084 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 4085 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
4086 | return -EINVAL; |
4087 | } | |
4088 | ||
d691f9e8 AS |
4089 | if (BPF_CLASS(insn->code) == BPF_STX && |
4090 | ((BPF_MODE(insn->code) != BPF_MEM && | |
4091 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 4092 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
4093 | return -EINVAL; |
4094 | } | |
4095 | ||
0246e64d AS |
4096 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
4097 | struct bpf_map *map; | |
4098 | struct fd f; | |
4099 | ||
4100 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
4101 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
4102 | insn[1].off != 0) { | |
61bd5218 | 4103 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
4104 | return -EINVAL; |
4105 | } | |
4106 | ||
4107 | if (insn->src_reg == 0) | |
4108 | /* valid generic load 64-bit imm */ | |
4109 | goto next_insn; | |
4110 | ||
4111 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
4112 | verbose(env, |
4113 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
4114 | return -EINVAL; |
4115 | } | |
4116 | ||
4117 | f = fdget(insn->imm); | |
c2101297 | 4118 | map = __bpf_map_get(f); |
0246e64d | 4119 | if (IS_ERR(map)) { |
61bd5218 | 4120 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 4121 | insn->imm); |
0246e64d AS |
4122 | return PTR_ERR(map); |
4123 | } | |
4124 | ||
61bd5218 | 4125 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
4126 | if (err) { |
4127 | fdput(f); | |
4128 | return err; | |
4129 | } | |
4130 | ||
0246e64d AS |
4131 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
4132 | insn[0].imm = (u32) (unsigned long) map; | |
4133 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
4134 | ||
4135 | /* check whether we recorded this map already */ | |
4136 | for (j = 0; j < env->used_map_cnt; j++) | |
4137 | if (env->used_maps[j] == map) { | |
4138 | fdput(f); | |
4139 | goto next_insn; | |
4140 | } | |
4141 | ||
4142 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
4143 | fdput(f); | |
4144 | return -E2BIG; | |
4145 | } | |
4146 | ||
0246e64d AS |
4147 | /* hold the map. If the program is rejected by verifier, |
4148 | * the map will be released by release_maps() or it | |
4149 | * will be used by the valid program until it's unloaded | |
4150 | * and all maps are released in free_bpf_prog_info() | |
4151 | */ | |
92117d84 AS |
4152 | map = bpf_map_inc(map, false); |
4153 | if (IS_ERR(map)) { | |
4154 | fdput(f); | |
4155 | return PTR_ERR(map); | |
4156 | } | |
4157 | env->used_maps[env->used_map_cnt++] = map; | |
4158 | ||
0246e64d AS |
4159 | fdput(f); |
4160 | next_insn: | |
4161 | insn++; | |
4162 | i++; | |
4163 | } | |
4164 | } | |
4165 | ||
4166 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
4167 | * 'struct bpf_map *' into a register instead of user map_fd. | |
4168 | * These pointers will be used later by verifier to validate map access. | |
4169 | */ | |
4170 | return 0; | |
4171 | } | |
4172 | ||
4173 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 4174 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
4175 | { |
4176 | int i; | |
4177 | ||
4178 | for (i = 0; i < env->used_map_cnt; i++) | |
4179 | bpf_map_put(env->used_maps[i]); | |
4180 | } | |
4181 | ||
4182 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 4183 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
4184 | { |
4185 | struct bpf_insn *insn = env->prog->insnsi; | |
4186 | int insn_cnt = env->prog->len; | |
4187 | int i; | |
4188 | ||
4189 | for (i = 0; i < insn_cnt; i++, insn++) | |
4190 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
4191 | insn->src_reg = 0; | |
4192 | } | |
4193 | ||
8041902d AS |
4194 | /* single env->prog->insni[off] instruction was replaced with the range |
4195 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
4196 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
4197 | */ | |
4198 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
4199 | u32 off, u32 cnt) | |
4200 | { | |
4201 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
4202 | ||
4203 | if (cnt == 1) | |
4204 | return 0; | |
4205 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
4206 | if (!new_data) | |
4207 | return -ENOMEM; | |
4208 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
4209 | memcpy(new_data + off + cnt - 1, old_data + off, | |
4210 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
4211 | env->insn_aux_data = new_data; | |
4212 | vfree(old_data); | |
4213 | return 0; | |
4214 | } | |
4215 | ||
4216 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
4217 | const struct bpf_insn *patch, u32 len) | |
4218 | { | |
4219 | struct bpf_prog *new_prog; | |
4220 | ||
4221 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4222 | if (!new_prog) | |
4223 | return NULL; | |
4224 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
4225 | return NULL; | |
4226 | return new_prog; | |
4227 | } | |
4228 | ||
9bac3d6d AS |
4229 | /* convert load instructions that access fields of 'struct __sk_buff' |
4230 | * into sequence of instructions that access fields of 'struct sk_buff' | |
4231 | */ | |
58e2af8b | 4232 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 4233 | { |
00176a34 | 4234 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 4235 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 4236 | const int insn_cnt = env->prog->len; |
36bbef52 | 4237 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 4238 | struct bpf_prog *new_prog; |
d691f9e8 | 4239 | enum bpf_access_type type; |
f96da094 DB |
4240 | bool is_narrower_load; |
4241 | u32 target_size; | |
9bac3d6d | 4242 | |
36bbef52 DB |
4243 | if (ops->gen_prologue) { |
4244 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
4245 | env->prog); | |
4246 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4247 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
4248 | return -EINVAL; |
4249 | } else if (cnt) { | |
8041902d | 4250 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
4251 | if (!new_prog) |
4252 | return -ENOMEM; | |
8041902d | 4253 | |
36bbef52 | 4254 | env->prog = new_prog; |
3df126f3 | 4255 | delta += cnt - 1; |
36bbef52 DB |
4256 | } |
4257 | } | |
4258 | ||
4259 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
4260 | return 0; |
4261 | ||
3df126f3 | 4262 | insn = env->prog->insnsi + delta; |
36bbef52 | 4263 | |
9bac3d6d | 4264 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
4265 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
4266 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
4267 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4268 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 4269 | type = BPF_READ; |
62c7989b DB |
4270 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
4271 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
4272 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4273 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
4274 | type = BPF_WRITE; |
4275 | else | |
9bac3d6d AS |
4276 | continue; |
4277 | ||
8041902d | 4278 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 4279 | continue; |
9bac3d6d | 4280 | |
31fd8581 | 4281 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 4282 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
4283 | |
4284 | /* If the read access is a narrower load of the field, | |
4285 | * convert to a 4/8-byte load, to minimum program type specific | |
4286 | * convert_ctx_access changes. If conversion is successful, | |
4287 | * we will apply proper mask to the result. | |
4288 | */ | |
f96da094 | 4289 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 4290 | if (is_narrower_load) { |
f96da094 DB |
4291 | u32 off = insn->off; |
4292 | u8 size_code; | |
4293 | ||
4294 | if (type == BPF_WRITE) { | |
61bd5218 | 4295 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
4296 | return -EINVAL; |
4297 | } | |
31fd8581 | 4298 | |
f96da094 | 4299 | size_code = BPF_H; |
31fd8581 YS |
4300 | if (ctx_field_size == 4) |
4301 | size_code = BPF_W; | |
4302 | else if (ctx_field_size == 8) | |
4303 | size_code = BPF_DW; | |
f96da094 | 4304 | |
31fd8581 YS |
4305 | insn->off = off & ~(ctx_field_size - 1); |
4306 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
4307 | } | |
f96da094 DB |
4308 | |
4309 | target_size = 0; | |
4310 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
4311 | &target_size); | |
4312 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
4313 | (ctx_field_size && !target_size)) { | |
61bd5218 | 4314 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
4315 | return -EINVAL; |
4316 | } | |
f96da094 DB |
4317 | |
4318 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
4319 | if (ctx_field_size <= 4) |
4320 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4321 | (1 << size * 8) - 1); |
31fd8581 YS |
4322 | else |
4323 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4324 | (1 << size * 8) - 1); |
31fd8581 | 4325 | } |
9bac3d6d | 4326 | |
8041902d | 4327 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
4328 | if (!new_prog) |
4329 | return -ENOMEM; | |
4330 | ||
3df126f3 | 4331 | delta += cnt - 1; |
9bac3d6d AS |
4332 | |
4333 | /* keep walking new program and skip insns we just inserted */ | |
4334 | env->prog = new_prog; | |
3df126f3 | 4335 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
4336 | } |
4337 | ||
4338 | return 0; | |
4339 | } | |
4340 | ||
79741b3b | 4341 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 4342 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
4343 | * |
4344 | * this function is called after eBPF program passed verification | |
4345 | */ | |
79741b3b | 4346 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 4347 | { |
79741b3b AS |
4348 | struct bpf_prog *prog = env->prog; |
4349 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 4350 | const struct bpf_func_proto *fn; |
79741b3b | 4351 | const int insn_cnt = prog->len; |
81ed18ab AS |
4352 | struct bpf_insn insn_buf[16]; |
4353 | struct bpf_prog *new_prog; | |
4354 | struct bpf_map *map_ptr; | |
4355 | int i, cnt, delta = 0; | |
e245c5c6 | 4356 | |
79741b3b AS |
4357 | for (i = 0; i < insn_cnt; i++, insn++) { |
4358 | if (insn->code != (BPF_JMP | BPF_CALL)) | |
4359 | continue; | |
e245c5c6 | 4360 | |
79741b3b AS |
4361 | if (insn->imm == BPF_FUNC_get_route_realm) |
4362 | prog->dst_needed = 1; | |
4363 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
4364 | bpf_user_rnd_init_once(); | |
79741b3b | 4365 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
4366 | /* If we tail call into other programs, we |
4367 | * cannot make any assumptions since they can | |
4368 | * be replaced dynamically during runtime in | |
4369 | * the program array. | |
4370 | */ | |
4371 | prog->cb_access = 1; | |
80a58d02 | 4372 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 4373 | |
79741b3b AS |
4374 | /* mark bpf_tail_call as different opcode to avoid |
4375 | * conditional branch in the interpeter for every normal | |
4376 | * call and to prevent accidental JITing by JIT compiler | |
4377 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 4378 | */ |
79741b3b | 4379 | insn->imm = 0; |
71189fa9 | 4380 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
79741b3b AS |
4381 | continue; |
4382 | } | |
e245c5c6 | 4383 | |
89c63074 DB |
4384 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
4385 | * handlers are currently limited to 64 bit only. | |
4386 | */ | |
4387 | if (ebpf_jit_enabled() && BITS_PER_LONG == 64 && | |
4388 | insn->imm == BPF_FUNC_map_lookup_elem) { | |
81ed18ab | 4389 | map_ptr = env->insn_aux_data[i + delta].map_ptr; |
fad73a1a MKL |
4390 | if (map_ptr == BPF_MAP_PTR_POISON || |
4391 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
4392 | goto patch_call_imm; |
4393 | ||
4394 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
4395 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4396 | verbose(env, "bpf verifier is misconfigured\n"); |
81ed18ab AS |
4397 | return -EINVAL; |
4398 | } | |
4399 | ||
4400 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
4401 | cnt); | |
4402 | if (!new_prog) | |
4403 | return -ENOMEM; | |
4404 | ||
4405 | delta += cnt - 1; | |
4406 | ||
4407 | /* keep walking new program and skip insns we just inserted */ | |
4408 | env->prog = prog = new_prog; | |
4409 | insn = new_prog->insnsi + i + delta; | |
4410 | continue; | |
4411 | } | |
4412 | ||
109980b8 | 4413 | if (insn->imm == BPF_FUNC_redirect_map) { |
7c300131 DB |
4414 | /* Note, we cannot use prog directly as imm as subsequent |
4415 | * rewrites would still change the prog pointer. The only | |
4416 | * stable address we can use is aux, which also works with | |
4417 | * prog clones during blinding. | |
4418 | */ | |
4419 | u64 addr = (unsigned long)prog->aux; | |
109980b8 DB |
4420 | struct bpf_insn r4_ld[] = { |
4421 | BPF_LD_IMM64(BPF_REG_4, addr), | |
4422 | *insn, | |
4423 | }; | |
4424 | cnt = ARRAY_SIZE(r4_ld); | |
4425 | ||
4426 | new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt); | |
4427 | if (!new_prog) | |
4428 | return -ENOMEM; | |
4429 | ||
4430 | delta += cnt - 1; | |
4431 | env->prog = prog = new_prog; | |
4432 | insn = new_prog->insnsi + i + delta; | |
4433 | } | |
81ed18ab | 4434 | patch_call_imm: |
00176a34 | 4435 | fn = env->ops->get_func_proto(insn->imm); |
79741b3b AS |
4436 | /* all functions that have prototype and verifier allowed |
4437 | * programs to call them, must be real in-kernel functions | |
4438 | */ | |
4439 | if (!fn->func) { | |
61bd5218 JK |
4440 | verbose(env, |
4441 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
4442 | func_id_name(insn->imm), insn->imm); |
4443 | return -EFAULT; | |
e245c5c6 | 4444 | } |
79741b3b | 4445 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 4446 | } |
e245c5c6 | 4447 | |
79741b3b AS |
4448 | return 0; |
4449 | } | |
e245c5c6 | 4450 | |
58e2af8b | 4451 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 4452 | { |
58e2af8b | 4453 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
4454 | int i; |
4455 | ||
4456 | if (!env->explored_states) | |
4457 | return; | |
4458 | ||
4459 | for (i = 0; i < env->prog->len; i++) { | |
4460 | sl = env->explored_states[i]; | |
4461 | ||
4462 | if (sl) | |
4463 | while (sl != STATE_LIST_MARK) { | |
4464 | sln = sl->next; | |
1969db47 | 4465 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
4466 | kfree(sl); |
4467 | sl = sln; | |
4468 | } | |
4469 | } | |
4470 | ||
4471 | kfree(env->explored_states); | |
4472 | } | |
4473 | ||
9bac3d6d | 4474 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 4475 | { |
58e2af8b | 4476 | struct bpf_verifier_env *env; |
61bd5218 | 4477 | struct bpf_verifer_log *log; |
51580e79 AS |
4478 | int ret = -EINVAL; |
4479 | ||
eba0c929 AB |
4480 | /* no program is valid */ |
4481 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
4482 | return -EINVAL; | |
4483 | ||
58e2af8b | 4484 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
4485 | * allocate/free it every time bpf_check() is called |
4486 | */ | |
58e2af8b | 4487 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
4488 | if (!env) |
4489 | return -ENOMEM; | |
61bd5218 | 4490 | log = &env->log; |
cbd35700 | 4491 | |
3df126f3 JK |
4492 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
4493 | (*prog)->len); | |
4494 | ret = -ENOMEM; | |
4495 | if (!env->insn_aux_data) | |
4496 | goto err_free_env; | |
9bac3d6d | 4497 | env->prog = *prog; |
00176a34 | 4498 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 4499 | |
cbd35700 AS |
4500 | /* grab the mutex to protect few globals used by verifier */ |
4501 | mutex_lock(&bpf_verifier_lock); | |
4502 | ||
4503 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
4504 | /* user requested verbose verifier output | |
4505 | * and supplied buffer to store the verification trace | |
4506 | */ | |
e7bf8249 JK |
4507 | log->level = attr->log_level; |
4508 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
4509 | log->len_total = attr->log_size; | |
cbd35700 AS |
4510 | |
4511 | ret = -EINVAL; | |
e7bf8249 JK |
4512 | /* log attributes have to be sane */ |
4513 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
4514 | !log->level || !log->ubuf) | |
3df126f3 | 4515 | goto err_unlock; |
cbd35700 | 4516 | } |
1ad2f583 DB |
4517 | |
4518 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
4519 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 4520 | env->strict_alignment = true; |
cbd35700 | 4521 | |
ab3f0063 JK |
4522 | if (env->prog->aux->offload) { |
4523 | ret = bpf_prog_offload_verifier_prep(env); | |
4524 | if (ret) | |
4525 | goto err_unlock; | |
4526 | } | |
4527 | ||
0246e64d AS |
4528 | ret = replace_map_fd_with_map_ptr(env); |
4529 | if (ret < 0) | |
4530 | goto skip_full_check; | |
4531 | ||
9bac3d6d | 4532 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 4533 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
4534 | GFP_USER); |
4535 | ret = -ENOMEM; | |
4536 | if (!env->explored_states) | |
4537 | goto skip_full_check; | |
4538 | ||
475fb78f AS |
4539 | ret = check_cfg(env); |
4540 | if (ret < 0) | |
4541 | goto skip_full_check; | |
4542 | ||
1be7f75d AS |
4543 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
4544 | ||
17a52670 | 4545 | ret = do_check(env); |
8c01c4f8 CG |
4546 | if (env->cur_state) { |
4547 | free_verifier_state(env->cur_state, true); | |
4548 | env->cur_state = NULL; | |
4549 | } | |
cbd35700 | 4550 | |
0246e64d | 4551 | skip_full_check: |
638f5b90 | 4552 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 4553 | free_states(env); |
0246e64d | 4554 | |
9bac3d6d AS |
4555 | if (ret == 0) |
4556 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
4557 | ret = convert_ctx_accesses(env); | |
4558 | ||
e245c5c6 | 4559 | if (ret == 0) |
79741b3b | 4560 | ret = fixup_bpf_calls(env); |
e245c5c6 | 4561 | |
a2a7d570 | 4562 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 4563 | ret = -ENOSPC; |
a2a7d570 | 4564 | if (log->level && !log->ubuf) { |
cbd35700 | 4565 | ret = -EFAULT; |
a2a7d570 | 4566 | goto err_release_maps; |
cbd35700 AS |
4567 | } |
4568 | ||
0246e64d AS |
4569 | if (ret == 0 && env->used_map_cnt) { |
4570 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
4571 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
4572 | sizeof(env->used_maps[0]), | |
4573 | GFP_KERNEL); | |
0246e64d | 4574 | |
9bac3d6d | 4575 | if (!env->prog->aux->used_maps) { |
0246e64d | 4576 | ret = -ENOMEM; |
a2a7d570 | 4577 | goto err_release_maps; |
0246e64d AS |
4578 | } |
4579 | ||
9bac3d6d | 4580 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 4581 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 4582 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
4583 | |
4584 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
4585 | * bpf_ld_imm64 instructions | |
4586 | */ | |
4587 | convert_pseudo_ld_imm64(env); | |
4588 | } | |
cbd35700 | 4589 | |
a2a7d570 | 4590 | err_release_maps: |
9bac3d6d | 4591 | if (!env->prog->aux->used_maps) |
0246e64d AS |
4592 | /* if we didn't copy map pointers into bpf_prog_info, release |
4593 | * them now. Otherwise free_bpf_prog_info() will release them. | |
4594 | */ | |
4595 | release_maps(env); | |
9bac3d6d | 4596 | *prog = env->prog; |
3df126f3 | 4597 | err_unlock: |
cbd35700 | 4598 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
4599 | vfree(env->insn_aux_data); |
4600 | err_free_env: | |
4601 | kfree(env); | |
51580e79 AS |
4602 | return ret; |
4603 | } | |
13a27dfc | 4604 | |
4f9218aa | 4605 | static const struct bpf_verifier_ops * const bpf_analyzer_ops[] = { |
7cce782e | 4606 | #ifdef CONFIG_NET |
4f9218aa JK |
4607 | [BPF_PROG_TYPE_XDP] = &xdp_analyzer_ops, |
4608 | [BPF_PROG_TYPE_SCHED_CLS] = &tc_cls_act_analyzer_ops, | |
7cce782e | 4609 | #endif |
4f9218aa JK |
4610 | }; |
4611 | ||
13a27dfc JK |
4612 | int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, |
4613 | void *priv) | |
4614 | { | |
4615 | struct bpf_verifier_env *env; | |
4616 | int ret; | |
4617 | ||
4f9218aa JK |
4618 | if (prog->type >= ARRAY_SIZE(bpf_analyzer_ops) || |
4619 | !bpf_analyzer_ops[prog->type]) | |
4620 | return -EOPNOTSUPP; | |
4621 | ||
13a27dfc JK |
4622 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
4623 | if (!env) | |
4624 | return -ENOMEM; | |
4625 | ||
4626 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * | |
4627 | prog->len); | |
4628 | ret = -ENOMEM; | |
4629 | if (!env->insn_aux_data) | |
4630 | goto err_free_env; | |
4631 | env->prog = prog; | |
4f9218aa | 4632 | env->ops = bpf_analyzer_ops[env->prog->type]; |
13a27dfc JK |
4633 | env->analyzer_ops = ops; |
4634 | env->analyzer_priv = priv; | |
4635 | ||
4636 | /* grab the mutex to protect few globals used by verifier */ | |
4637 | mutex_lock(&bpf_verifier_lock); | |
4638 | ||
e07b98d9 | 4639 | env->strict_alignment = false; |
1ad2f583 DB |
4640 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) |
4641 | env->strict_alignment = true; | |
13a27dfc JK |
4642 | |
4643 | env->explored_states = kcalloc(env->prog->len, | |
4644 | sizeof(struct bpf_verifier_state_list *), | |
4645 | GFP_KERNEL); | |
4646 | ret = -ENOMEM; | |
4647 | if (!env->explored_states) | |
4648 | goto skip_full_check; | |
4649 | ||
4650 | ret = check_cfg(env); | |
4651 | if (ret < 0) | |
4652 | goto skip_full_check; | |
4653 | ||
4654 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); | |
4655 | ||
4656 | ret = do_check(env); | |
8c01c4f8 CG |
4657 | if (env->cur_state) { |
4658 | free_verifier_state(env->cur_state, true); | |
4659 | env->cur_state = NULL; | |
4660 | } | |
13a27dfc JK |
4661 | |
4662 | skip_full_check: | |
638f5b90 | 4663 | while (!pop_stack(env, NULL, NULL)); |
13a27dfc JK |
4664 | free_states(env); |
4665 | ||
4666 | mutex_unlock(&bpf_verifier_lock); | |
4667 | vfree(env->insn_aux_data); | |
4668 | err_free_env: | |
4669 | kfree(env); | |
4670 | return ret; | |
4671 | } | |
4672 | EXPORT_SYMBOL_GPL(bpf_analyzer); |