]>
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", | |
e8988995 | 282 | (-i - 1) * BPF_REG_SIZE, |
638f5b90 | 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, | |
abd098e0 | 694 | int size, int value_regno, int insn_idx) |
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 | |
abd098e0 AS |
725 | for (i = 0; i < BPF_REG_SIZE; i++) { |
726 | if (state->stack[spi].slot_type[i] == STACK_MISC && | |
727 | !env->allow_ptr_leaks) { | |
728 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; | |
729 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
730 | ||
731 | /* detected reuse of integer stack slot with a pointer | |
732 | * which means either llvm is reusing stack slot or | |
733 | * an attacker is trying to exploit CVE-2018-3639 | |
734 | * (speculative store bypass) | |
735 | * Have to sanitize that slot with preemptive | |
736 | * store of zero. | |
737 | */ | |
738 | if (*poff && *poff != soff) { | |
739 | /* disallow programs where single insn stores | |
740 | * into two different stack slots, since verifier | |
741 | * cannot sanitize them | |
742 | */ | |
743 | verbose(env, | |
744 | "insn %d cannot access two stack slots fp%d and fp%d", | |
745 | insn_idx, *poff, soff); | |
746 | return -EINVAL; | |
747 | } | |
748 | *poff = soff; | |
749 | } | |
638f5b90 | 750 | state->stack[spi].slot_type[i] = STACK_SPILL; |
abd098e0 | 751 | } |
9c399760 | 752 | } else { |
17a52670 | 753 | /* regular write of data into stack */ |
638f5b90 | 754 | state->stack[spi].spilled_ptr = (struct bpf_reg_state) {}; |
9c399760 AS |
755 | |
756 | for (i = 0; i < size; i++) | |
638f5b90 AS |
757 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
758 | STACK_MISC; | |
17a52670 AS |
759 | } |
760 | return 0; | |
761 | } | |
762 | ||
dc503a8a EC |
763 | static void mark_stack_slot_read(const struct bpf_verifier_state *state, int slot) |
764 | { | |
765 | struct bpf_verifier_state *parent = state->parent; | |
766 | ||
767 | while (parent) { | |
768 | /* if read wasn't screened by an earlier write ... */ | |
638f5b90 | 769 | if (state->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN) |
dc503a8a EC |
770 | break; |
771 | /* ... then we depend on parent's value */ | |
638f5b90 | 772 | parent->stack[slot].spilled_ptr.live |= REG_LIVE_READ; |
dc503a8a EC |
773 | state = parent; |
774 | parent = state->parent; | |
775 | } | |
776 | } | |
777 | ||
61bd5218 JK |
778 | static int check_stack_read(struct bpf_verifier_env *env, |
779 | struct bpf_verifier_state *state, int off, int size, | |
17a52670 AS |
780 | int value_regno) |
781 | { | |
638f5b90 AS |
782 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
783 | u8 *stype; | |
17a52670 | 784 | |
638f5b90 AS |
785 | if (state->allocated_stack <= slot) { |
786 | verbose(env, "invalid read from stack off %d+0 size %d\n", | |
787 | off, size); | |
788 | return -EACCES; | |
789 | } | |
790 | stype = state->stack[spi].slot_type; | |
17a52670 | 791 | |
638f5b90 | 792 | if (stype[0] == STACK_SPILL) { |
9c399760 | 793 | if (size != BPF_REG_SIZE) { |
61bd5218 | 794 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
795 | return -EACCES; |
796 | } | |
9c399760 | 797 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 798 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 799 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
800 | return -EACCES; |
801 | } | |
802 | } | |
803 | ||
dc503a8a | 804 | if (value_regno >= 0) { |
17a52670 | 805 | /* restore register state from stack */ |
638f5b90 | 806 | state->regs[value_regno] = state->stack[spi].spilled_ptr; |
dc503a8a EC |
807 | mark_stack_slot_read(state, spi); |
808 | } | |
17a52670 AS |
809 | return 0; |
810 | } else { | |
811 | for (i = 0; i < size; i++) { | |
638f5b90 | 812 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_MISC) { |
61bd5218 | 813 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
17a52670 AS |
814 | off, i, size); |
815 | return -EACCES; | |
816 | } | |
817 | } | |
818 | if (value_regno >= 0) | |
819 | /* have read misc data from the stack */ | |
61bd5218 | 820 | mark_reg_unknown(env, state->regs, value_regno); |
17a52670 AS |
821 | return 0; |
822 | } | |
823 | } | |
824 | ||
825 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
f1174f77 | 826 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 827 | int size, bool zero_size_allowed) |
17a52670 | 828 | { |
638f5b90 AS |
829 | struct bpf_reg_state *regs = cur_regs(env); |
830 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 831 | |
9fd29c08 YS |
832 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
833 | off + size > map->value_size) { | |
61bd5218 | 834 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
835 | map->value_size, off, size); |
836 | return -EACCES; | |
837 | } | |
838 | return 0; | |
839 | } | |
840 | ||
f1174f77 EC |
841 | /* check read/write into a map element with possible variable offset */ |
842 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 843 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 844 | { |
638f5b90 | 845 | struct bpf_verifier_state *state = env->cur_state; |
dbcfe5f7 GB |
846 | struct bpf_reg_state *reg = &state->regs[regno]; |
847 | int err; | |
848 | ||
f1174f77 EC |
849 | /* We may have adjusted the register to this map value, so we |
850 | * need to try adding each of min_value and max_value to off | |
851 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 852 | */ |
61bd5218 JK |
853 | if (env->log.level) |
854 | print_verifier_state(env, state); | |
dbcfe5f7 GB |
855 | /* The minimum value is only important with signed |
856 | * comparisons where we can't assume the floor of a | |
857 | * value is 0. If we are using signed variables for our | |
858 | * index'es we need to make sure that whatever we use | |
859 | * will have a set floor within our range. | |
860 | */ | |
b03c9f9f | 861 | if (reg->smin_value < 0) { |
61bd5218 | 862 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
863 | regno); |
864 | return -EACCES; | |
865 | } | |
9fd29c08 YS |
866 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
867 | zero_size_allowed); | |
dbcfe5f7 | 868 | if (err) { |
61bd5218 JK |
869 | verbose(env, "R%d min value is outside of the array range\n", |
870 | regno); | |
dbcfe5f7 GB |
871 | return err; |
872 | } | |
873 | ||
b03c9f9f EC |
874 | /* If we haven't set a max value then we need to bail since we can't be |
875 | * sure we won't do bad things. | |
876 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 877 | */ |
b03c9f9f | 878 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 879 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
880 | regno); |
881 | return -EACCES; | |
882 | } | |
9fd29c08 YS |
883 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
884 | zero_size_allowed); | |
f1174f77 | 885 | if (err) |
61bd5218 JK |
886 | verbose(env, "R%d max value is outside of the array range\n", |
887 | regno); | |
f1174f77 | 888 | return err; |
dbcfe5f7 GB |
889 | } |
890 | ||
969bf05e AS |
891 | #define MAX_PACKET_OFF 0xffff |
892 | ||
58e2af8b | 893 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
894 | const struct bpf_call_arg_meta *meta, |
895 | enum bpf_access_type t) | |
4acf6c0b | 896 | { |
36bbef52 | 897 | switch (env->prog->type) { |
3a0af8fd TG |
898 | case BPF_PROG_TYPE_LWT_IN: |
899 | case BPF_PROG_TYPE_LWT_OUT: | |
900 | /* dst_input() and dst_output() can't write for now */ | |
901 | if (t == BPF_WRITE) | |
902 | return false; | |
7e57fbb2 | 903 | /* fallthrough */ |
36bbef52 DB |
904 | case BPF_PROG_TYPE_SCHED_CLS: |
905 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 906 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 907 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 908 | case BPF_PROG_TYPE_SK_SKB: |
36bbef52 DB |
909 | if (meta) |
910 | return meta->pkt_access; | |
911 | ||
912 | env->seen_direct_write = true; | |
4acf6c0b BB |
913 | return true; |
914 | default: | |
915 | return false; | |
916 | } | |
917 | } | |
918 | ||
f1174f77 | 919 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 920 | int off, int size, bool zero_size_allowed) |
969bf05e | 921 | { |
638f5b90 | 922 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 923 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 924 | |
9fd29c08 YS |
925 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
926 | (u64)off + size > reg->range) { | |
61bd5218 | 927 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 928 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
929 | return -EACCES; |
930 | } | |
931 | return 0; | |
932 | } | |
933 | ||
f1174f77 | 934 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 935 | int size, bool zero_size_allowed) |
f1174f77 | 936 | { |
638f5b90 | 937 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
938 | struct bpf_reg_state *reg = ®s[regno]; |
939 | int err; | |
940 | ||
941 | /* We may have added a variable offset to the packet pointer; but any | |
942 | * reg->range we have comes after that. We are only checking the fixed | |
943 | * offset. | |
944 | */ | |
945 | ||
946 | /* We don't allow negative numbers, because we aren't tracking enough | |
947 | * detail to prove they're safe. | |
948 | */ | |
b03c9f9f | 949 | if (reg->smin_value < 0) { |
61bd5218 | 950 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
951 | regno); |
952 | return -EACCES; | |
953 | } | |
9fd29c08 | 954 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 955 | if (err) { |
61bd5218 | 956 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
957 | return err; |
958 | } | |
959 | return err; | |
960 | } | |
961 | ||
962 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 963 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 964 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 965 | { |
f96da094 DB |
966 | struct bpf_insn_access_aux info = { |
967 | .reg_type = *reg_type, | |
968 | }; | |
31fd8581 | 969 | |
4f9218aa JK |
970 | if (env->ops->is_valid_access && |
971 | env->ops->is_valid_access(off, size, t, &info)) { | |
f96da094 DB |
972 | /* A non zero info.ctx_field_size indicates that this field is a |
973 | * candidate for later verifier transformation to load the whole | |
974 | * field and then apply a mask when accessed with a narrower | |
975 | * access than actual ctx access size. A zero info.ctx_field_size | |
976 | * will only allow for whole field access and rejects any other | |
977 | * type of narrower access. | |
31fd8581 | 978 | */ |
23994631 | 979 | *reg_type = info.reg_type; |
31fd8581 | 980 | |
4f9218aa | 981 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
982 | /* remember the offset of last byte accessed in ctx */ |
983 | if (env->prog->aux->max_ctx_offset < off + size) | |
984 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 985 | return 0; |
32bbe007 | 986 | } |
17a52670 | 987 | |
61bd5218 | 988 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
989 | return -EACCES; |
990 | } | |
991 | ||
4cabc5b1 DB |
992 | static bool __is_pointer_value(bool allow_ptr_leaks, |
993 | const struct bpf_reg_state *reg) | |
1be7f75d | 994 | { |
4cabc5b1 | 995 | if (allow_ptr_leaks) |
1be7f75d AS |
996 | return false; |
997 | ||
f1174f77 | 998 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
999 | } |
1000 | ||
4cabc5b1 DB |
1001 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1002 | { | |
638f5b90 | 1003 | return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno); |
4cabc5b1 DB |
1004 | } |
1005 | ||
f37a8cb8 DB |
1006 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
1007 | { | |
1008 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1009 | ||
1010 | return reg->type == PTR_TO_CTX; | |
1011 | } | |
1012 | ||
57b4a36e DB |
1013 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
1014 | { | |
1015 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1016 | ||
1017 | return type_is_pkt_pointer(reg->type); | |
1018 | } | |
1019 | ||
61bd5218 JK |
1020 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
1021 | const struct bpf_reg_state *reg, | |
d1174416 | 1022 | int off, int size, bool strict) |
969bf05e | 1023 | { |
f1174f77 | 1024 | struct tnum reg_off; |
e07b98d9 | 1025 | int ip_align; |
d1174416 DM |
1026 | |
1027 | /* Byte size accesses are always allowed. */ | |
1028 | if (!strict || size == 1) | |
1029 | return 0; | |
1030 | ||
e4eda884 DM |
1031 | /* For platforms that do not have a Kconfig enabling |
1032 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
1033 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
1034 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
1035 | * to this code only in strict mode where we want to emulate | |
1036 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
1037 | * unconditional IP align value of '2'. | |
e07b98d9 | 1038 | */ |
e4eda884 | 1039 | ip_align = 2; |
f1174f77 EC |
1040 | |
1041 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1042 | if (!tnum_is_aligned(reg_off, size)) { | |
1043 | char tn_buf[48]; | |
1044 | ||
1045 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1046 | verbose(env, |
1047 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1048 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1049 | return -EACCES; |
1050 | } | |
79adffcd | 1051 | |
969bf05e AS |
1052 | return 0; |
1053 | } | |
1054 | ||
61bd5218 JK |
1055 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1056 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1057 | const char *pointer_desc, |
1058 | int off, int size, bool strict) | |
79adffcd | 1059 | { |
f1174f77 EC |
1060 | struct tnum reg_off; |
1061 | ||
1062 | /* Byte size accesses are always allowed. */ | |
1063 | if (!strict || size == 1) | |
1064 | return 0; | |
1065 | ||
1066 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1067 | if (!tnum_is_aligned(reg_off, size)) { | |
1068 | char tn_buf[48]; | |
1069 | ||
1070 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1071 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1072 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1073 | return -EACCES; |
1074 | } | |
1075 | ||
969bf05e AS |
1076 | return 0; |
1077 | } | |
1078 | ||
e07b98d9 | 1079 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
57b4a36e DB |
1080 | const struct bpf_reg_state *reg, int off, |
1081 | int size, bool strict_alignment_once) | |
79adffcd | 1082 | { |
57b4a36e | 1083 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 1084 | const char *pointer_desc = ""; |
d1174416 | 1085 | |
79adffcd DB |
1086 | switch (reg->type) { |
1087 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1088 | case PTR_TO_PACKET_META: |
1089 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1090 | * right in front, treat it the very same way. | |
1091 | */ | |
61bd5218 | 1092 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
f1174f77 EC |
1093 | case PTR_TO_MAP_VALUE: |
1094 | pointer_desc = "value "; | |
1095 | break; | |
1096 | case PTR_TO_CTX: | |
1097 | pointer_desc = "context "; | |
1098 | break; | |
1099 | case PTR_TO_STACK: | |
1100 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
1101 | /* The stack spill tracking logic in check_stack_write() |
1102 | * and check_stack_read() relies on stack accesses being | |
1103 | * aligned. | |
1104 | */ | |
1105 | strict = true; | |
f1174f77 | 1106 | break; |
79adffcd | 1107 | default: |
f1174f77 | 1108 | break; |
79adffcd | 1109 | } |
61bd5218 JK |
1110 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1111 | strict); | |
79adffcd DB |
1112 | } |
1113 | ||
0c17d1d2 JH |
1114 | /* truncate register to smaller size (in bytes) |
1115 | * must be called with size < BPF_REG_SIZE | |
1116 | */ | |
1117 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
1118 | { | |
1119 | u64 mask; | |
1120 | ||
1121 | /* clear high bits in bit representation */ | |
1122 | reg->var_off = tnum_cast(reg->var_off, size); | |
1123 | ||
1124 | /* fix arithmetic bounds */ | |
1125 | mask = ((u64)1 << (size * 8)) - 1; | |
1126 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
1127 | reg->umin_value &= mask; | |
1128 | reg->umax_value &= mask; | |
1129 | } else { | |
1130 | reg->umin_value = 0; | |
1131 | reg->umax_value = mask; | |
1132 | } | |
1133 | reg->smin_value = reg->umin_value; | |
1134 | reg->smax_value = reg->umax_value; | |
1135 | } | |
1136 | ||
17a52670 AS |
1137 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1138 | * if t==write, value_regno is a register which value is stored into memory | |
1139 | * if t==read, value_regno is a register which will receive the value from memory | |
1140 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1141 | * if t==read && value_regno==-1, don't care what we read from memory | |
1142 | */ | |
57b4a36e DB |
1143 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
1144 | int off, int bpf_size, enum bpf_access_type t, | |
1145 | int value_regno, bool strict_alignment_once) | |
17a52670 | 1146 | { |
638f5b90 AS |
1147 | struct bpf_verifier_state *state = env->cur_state; |
1148 | struct bpf_reg_state *regs = cur_regs(env); | |
1149 | struct bpf_reg_state *reg = regs + regno; | |
17a52670 AS |
1150 | int size, err = 0; |
1151 | ||
1152 | size = bpf_size_to_bytes(bpf_size); | |
1153 | if (size < 0) | |
1154 | return size; | |
1155 | ||
f1174f77 | 1156 | /* alignment checks will add in reg->off themselves */ |
57b4a36e | 1157 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
1158 | if (err) |
1159 | return err; | |
17a52670 | 1160 | |
f1174f77 EC |
1161 | /* for access checks, reg->off is just part of off */ |
1162 | off += reg->off; | |
1163 | ||
1164 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1165 | if (t == BPF_WRITE && value_regno >= 0 && |
1166 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1167 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1168 | return -EACCES; |
1169 | } | |
48461135 | 1170 | |
9fd29c08 | 1171 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 1172 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1173 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1174 | |
1a0dc1ac | 1175 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1176 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1177 | |
1be7f75d AS |
1178 | if (t == BPF_WRITE && value_regno >= 0 && |
1179 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1180 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1181 | return -EACCES; |
1182 | } | |
f1174f77 EC |
1183 | /* ctx accesses must be at a fixed offset, so that we can |
1184 | * determine what type of data were returned. | |
1185 | */ | |
28e33f9d | 1186 | if (reg->off) { |
f8ddadc4 DM |
1187 | verbose(env, |
1188 | "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n", | |
28e33f9d JK |
1189 | regno, reg->off, off - reg->off); |
1190 | return -EACCES; | |
1191 | } | |
1192 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
f1174f77 EC |
1193 | char tn_buf[48]; |
1194 | ||
1195 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1196 | verbose(env, |
1197 | "variable ctx access var_off=%s off=%d size=%d", | |
f1174f77 EC |
1198 | tn_buf, off, size); |
1199 | return -EACCES; | |
1200 | } | |
31fd8581 | 1201 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1202 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1203 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1204 | * PTR_TO_PACKET[_META,_END]. In the latter |
1205 | * case, we know the offset is zero. | |
f1174f77 EC |
1206 | */ |
1207 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1208 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1209 | else |
638f5b90 | 1210 | mark_reg_known_zero(env, regs, |
61bd5218 | 1211 | value_regno); |
638f5b90 AS |
1212 | regs[value_regno].id = 0; |
1213 | regs[value_regno].off = 0; | |
1214 | regs[value_regno].range = 0; | |
1215 | regs[value_regno].type = reg_type; | |
969bf05e | 1216 | } |
17a52670 | 1217 | |
f1174f77 EC |
1218 | } else if (reg->type == PTR_TO_STACK) { |
1219 | /* stack accesses must be at a fixed offset, so that we can | |
1220 | * determine what type of data were returned. | |
1221 | * See check_stack_read(). | |
1222 | */ | |
1223 | if (!tnum_is_const(reg->var_off)) { | |
1224 | char tn_buf[48]; | |
1225 | ||
1226 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1227 | verbose(env, "variable stack access var_off=%s off=%d size=%d", |
f1174f77 EC |
1228 | tn_buf, off, size); |
1229 | return -EACCES; | |
1230 | } | |
1231 | off += reg->var_off.value; | |
17a52670 | 1232 | if (off >= 0 || off < -MAX_BPF_STACK) { |
61bd5218 JK |
1233 | verbose(env, "invalid stack off=%d size=%d\n", off, |
1234 | size); | |
17a52670 AS |
1235 | return -EACCES; |
1236 | } | |
8726679a AS |
1237 | |
1238 | if (env->prog->aux->stack_depth < -off) | |
1239 | env->prog->aux->stack_depth = -off; | |
1240 | ||
638f5b90 | 1241 | if (t == BPF_WRITE) |
61bd5218 | 1242 | err = check_stack_write(env, state, off, size, |
abd098e0 | 1243 | value_regno, insn_idx); |
638f5b90 | 1244 | else |
61bd5218 JK |
1245 | err = check_stack_read(env, state, off, size, |
1246 | value_regno); | |
de8f3a83 | 1247 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 1248 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 1249 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
1250 | return -EACCES; |
1251 | } | |
4acf6c0b BB |
1252 | if (t == BPF_WRITE && value_regno >= 0 && |
1253 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
1254 | verbose(env, "R%d leaks addr into packet\n", |
1255 | value_regno); | |
4acf6c0b BB |
1256 | return -EACCES; |
1257 | } | |
9fd29c08 | 1258 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 1259 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1260 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1261 | } else { |
61bd5218 JK |
1262 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
1263 | reg_type_str[reg->type]); | |
17a52670 AS |
1264 | return -EACCES; |
1265 | } | |
969bf05e | 1266 | |
f1174f77 | 1267 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 1268 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 1269 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 1270 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 1271 | } |
17a52670 AS |
1272 | return err; |
1273 | } | |
1274 | ||
31fd8581 | 1275 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1276 | { |
17a52670 AS |
1277 | int err; |
1278 | ||
1279 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1280 | insn->imm != 0) { | |
61bd5218 | 1281 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
1282 | return -EINVAL; |
1283 | } | |
1284 | ||
1285 | /* check src1 operand */ | |
dc503a8a | 1286 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
1287 | if (err) |
1288 | return err; | |
1289 | ||
1290 | /* check src2 operand */ | |
dc503a8a | 1291 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
1292 | if (err) |
1293 | return err; | |
1294 | ||
6bdf6abc | 1295 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 1296 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
1297 | return -EACCES; |
1298 | } | |
1299 | ||
57b4a36e DB |
1300 | if (is_ctx_reg(env, insn->dst_reg) || |
1301 | is_pkt_reg(env, insn->dst_reg)) { | |
1302 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", | |
1303 | insn->dst_reg, is_ctx_reg(env, insn->dst_reg) ? | |
1304 | "context" : "packet"); | |
f37a8cb8 DB |
1305 | return -EACCES; |
1306 | } | |
1307 | ||
17a52670 | 1308 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1309 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
57b4a36e | 1310 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
1311 | if (err) |
1312 | return err; | |
1313 | ||
1314 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1315 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
57b4a36e | 1316 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
1317 | } |
1318 | ||
f1174f77 EC |
1319 | /* Does this register contain a constant zero? */ |
1320 | static bool register_is_null(struct bpf_reg_state reg) | |
1321 | { | |
1322 | return reg.type == SCALAR_VALUE && tnum_equals_const(reg.var_off, 0); | |
1323 | } | |
1324 | ||
17a52670 AS |
1325 | /* when register 'regno' is passed into function that will read 'access_size' |
1326 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1327 | * and all elements of stack are initialized. |
1328 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1329 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1330 | */ |
58e2af8b | 1331 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1332 | int access_size, bool zero_size_allowed, |
1333 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1334 | { |
638f5b90 | 1335 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1336 | struct bpf_reg_state *regs = state->regs; |
638f5b90 | 1337 | int off, i, slot, spi; |
17a52670 | 1338 | |
8e2fe1d9 | 1339 | if (regs[regno].type != PTR_TO_STACK) { |
f1174f77 | 1340 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1341 | if (zero_size_allowed && access_size == 0 && |
f1174f77 | 1342 | register_is_null(regs[regno])) |
8e2fe1d9 DB |
1343 | return 0; |
1344 | ||
61bd5218 | 1345 | verbose(env, "R%d type=%s expected=%s\n", regno, |
8e2fe1d9 DB |
1346 | reg_type_str[regs[regno].type], |
1347 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 1348 | return -EACCES; |
8e2fe1d9 | 1349 | } |
17a52670 | 1350 | |
f1174f77 EC |
1351 | /* Only allow fixed-offset stack reads */ |
1352 | if (!tnum_is_const(regs[regno].var_off)) { | |
1353 | char tn_buf[48]; | |
1354 | ||
1355 | tnum_strn(tn_buf, sizeof(tn_buf), regs[regno].var_off); | |
61bd5218 | 1356 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 | 1357 | regno, tn_buf); |
ea25f914 | 1358 | return -EACCES; |
f1174f77 EC |
1359 | } |
1360 | off = regs[regno].off + regs[regno].var_off.value; | |
17a52670 | 1361 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
9fd29c08 | 1362 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { |
61bd5218 | 1363 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
1364 | regno, off, access_size); |
1365 | return -EACCES; | |
1366 | } | |
1367 | ||
8726679a AS |
1368 | if (env->prog->aux->stack_depth < -off) |
1369 | env->prog->aux->stack_depth = -off; | |
1370 | ||
435faee1 DB |
1371 | if (meta && meta->raw_mode) { |
1372 | meta->access_size = access_size; | |
1373 | meta->regno = regno; | |
1374 | return 0; | |
1375 | } | |
1376 | ||
17a52670 | 1377 | for (i = 0; i < access_size; i++) { |
638f5b90 AS |
1378 | slot = -(off + i) - 1; |
1379 | spi = slot / BPF_REG_SIZE; | |
1380 | if (state->allocated_stack <= slot || | |
1381 | state->stack[spi].slot_type[slot % BPF_REG_SIZE] != | |
1382 | STACK_MISC) { | |
61bd5218 | 1383 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", |
17a52670 AS |
1384 | off, i, access_size); |
1385 | return -EACCES; | |
1386 | } | |
1387 | } | |
1388 | return 0; | |
1389 | } | |
1390 | ||
06c1c049 GB |
1391 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1392 | int access_size, bool zero_size_allowed, | |
1393 | struct bpf_call_arg_meta *meta) | |
1394 | { | |
638f5b90 | 1395 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 1396 | |
f1174f77 | 1397 | switch (reg->type) { |
06c1c049 | 1398 | case PTR_TO_PACKET: |
de8f3a83 | 1399 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
1400 | return check_packet_access(env, regno, reg->off, access_size, |
1401 | zero_size_allowed); | |
06c1c049 | 1402 | case PTR_TO_MAP_VALUE: |
9fd29c08 YS |
1403 | return check_map_access(env, regno, reg->off, access_size, |
1404 | zero_size_allowed); | |
f1174f77 | 1405 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
1406 | return check_stack_boundary(env, regno, access_size, |
1407 | zero_size_allowed, meta); | |
1408 | } | |
1409 | } | |
1410 | ||
58e2af8b | 1411 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1412 | enum bpf_arg_type arg_type, |
1413 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1414 | { |
638f5b90 | 1415 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 1416 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1417 | int err = 0; |
1418 | ||
80f1d68c | 1419 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1420 | return 0; |
1421 | ||
dc503a8a EC |
1422 | err = check_reg_arg(env, regno, SRC_OP); |
1423 | if (err) | |
1424 | return err; | |
17a52670 | 1425 | |
1be7f75d AS |
1426 | if (arg_type == ARG_ANYTHING) { |
1427 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
1428 | verbose(env, "R%d leaks addr into helper function\n", |
1429 | regno); | |
1be7f75d AS |
1430 | return -EACCES; |
1431 | } | |
80f1d68c | 1432 | return 0; |
1be7f75d | 1433 | } |
80f1d68c | 1434 | |
de8f3a83 | 1435 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 1436 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 1437 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
1438 | return -EACCES; |
1439 | } | |
1440 | ||
8e2fe1d9 | 1441 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1442 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1443 | expected_type = PTR_TO_STACK; | |
de8f3a83 DB |
1444 | if (!type_is_pkt_pointer(type) && |
1445 | type != expected_type) | |
6841de8b | 1446 | goto err_type; |
39f19ebb AS |
1447 | } else if (arg_type == ARG_CONST_SIZE || |
1448 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1449 | expected_type = SCALAR_VALUE; |
1450 | if (type != expected_type) | |
6841de8b | 1451 | goto err_type; |
17a52670 AS |
1452 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1453 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1454 | if (type != expected_type) |
1455 | goto err_type; | |
608cd71a AS |
1456 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1457 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1458 | if (type != expected_type) |
1459 | goto err_type; | |
39f19ebb | 1460 | } else if (arg_type == ARG_PTR_TO_MEM || |
db1ac496 | 1461 | arg_type == ARG_PTR_TO_MEM_OR_NULL || |
39f19ebb | 1462 | arg_type == ARG_PTR_TO_UNINIT_MEM) { |
8e2fe1d9 DB |
1463 | expected_type = PTR_TO_STACK; |
1464 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1465 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1466 | * happens during stack boundary checking. |
1467 | */ | |
db1ac496 GB |
1468 | if (register_is_null(*reg) && |
1469 | arg_type == ARG_PTR_TO_MEM_OR_NULL) | |
6841de8b | 1470 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
1471 | else if (!type_is_pkt_pointer(type) && |
1472 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 1473 | type != expected_type) |
6841de8b | 1474 | goto err_type; |
39f19ebb | 1475 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 1476 | } else { |
61bd5218 | 1477 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
1478 | return -EFAULT; |
1479 | } | |
1480 | ||
17a52670 AS |
1481 | if (arg_type == ARG_CONST_MAP_PTR) { |
1482 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1483 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1484 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1485 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1486 | * check that [key, key + map->key_size) are within | |
1487 | * stack limits and initialized | |
1488 | */ | |
33ff9823 | 1489 | if (!meta->map_ptr) { |
17a52670 AS |
1490 | /* in function declaration map_ptr must come before |
1491 | * map_key, so that it's verified and known before | |
1492 | * we have to check map_key here. Otherwise it means | |
1493 | * that kernel subsystem misconfigured verifier | |
1494 | */ | |
61bd5218 | 1495 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
1496 | return -EACCES; |
1497 | } | |
de8f3a83 | 1498 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1499 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1500 | meta->map_ptr->key_size, |
1501 | false); | |
6841de8b AS |
1502 | else |
1503 | err = check_stack_boundary(env, regno, | |
1504 | meta->map_ptr->key_size, | |
1505 | false, NULL); | |
17a52670 AS |
1506 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1507 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1508 | * check [value, value + map->value_size) validity | |
1509 | */ | |
33ff9823 | 1510 | if (!meta->map_ptr) { |
17a52670 | 1511 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 1512 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
1513 | return -EACCES; |
1514 | } | |
de8f3a83 | 1515 | if (type_is_pkt_pointer(type)) |
f1174f77 | 1516 | err = check_packet_access(env, regno, reg->off, |
9fd29c08 YS |
1517 | meta->map_ptr->value_size, |
1518 | false); | |
6841de8b AS |
1519 | else |
1520 | err = check_stack_boundary(env, regno, | |
1521 | meta->map_ptr->value_size, | |
1522 | false, NULL); | |
39f19ebb AS |
1523 | } else if (arg_type == ARG_CONST_SIZE || |
1524 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1525 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1526 | |
17a52670 AS |
1527 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1528 | * from stack pointer 'buf'. Check it | |
1529 | * note: regno == len, regno - 1 == buf | |
1530 | */ | |
1531 | if (regno == 0) { | |
1532 | /* kernel subsystem misconfigured verifier */ | |
61bd5218 JK |
1533 | verbose(env, |
1534 | "ARG_CONST_SIZE cannot be first argument\n"); | |
17a52670 AS |
1535 | return -EACCES; |
1536 | } | |
06c1c049 | 1537 | |
f1174f77 EC |
1538 | /* The register is SCALAR_VALUE; the access check |
1539 | * happens using its boundaries. | |
06c1c049 | 1540 | */ |
f1174f77 EC |
1541 | |
1542 | if (!tnum_is_const(reg->var_off)) | |
06c1c049 GB |
1543 | /* For unprivileged variable accesses, disable raw |
1544 | * mode so that the program is required to | |
1545 | * initialize all the memory that the helper could | |
1546 | * just partially fill up. | |
1547 | */ | |
1548 | meta = NULL; | |
1549 | ||
b03c9f9f | 1550 | if (reg->smin_value < 0) { |
61bd5218 | 1551 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
1552 | regno); |
1553 | return -EACCES; | |
1554 | } | |
06c1c049 | 1555 | |
b03c9f9f | 1556 | if (reg->umin_value == 0) { |
f1174f77 EC |
1557 | err = check_helper_mem_access(env, regno - 1, 0, |
1558 | zero_size_allowed, | |
1559 | meta); | |
06c1c049 GB |
1560 | if (err) |
1561 | return err; | |
06c1c049 | 1562 | } |
f1174f77 | 1563 | |
b03c9f9f | 1564 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 1565 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
1566 | regno); |
1567 | return -EACCES; | |
1568 | } | |
1569 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 1570 | reg->umax_value, |
f1174f77 | 1571 | zero_size_allowed, meta); |
17a52670 AS |
1572 | } |
1573 | ||
1574 | return err; | |
6841de8b | 1575 | err_type: |
61bd5218 | 1576 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
1577 | reg_type_str[type], reg_type_str[expected_type]); |
1578 | return -EACCES; | |
17a52670 AS |
1579 | } |
1580 | ||
61bd5218 JK |
1581 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
1582 | struct bpf_map *map, int func_id) | |
35578d79 | 1583 | { |
35578d79 KX |
1584 | if (!map) |
1585 | return 0; | |
1586 | ||
6aff67c8 AS |
1587 | /* We need a two way check, first is from map perspective ... */ |
1588 | switch (map->map_type) { | |
1589 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1590 | if (func_id != BPF_FUNC_tail_call) | |
1591 | goto error; | |
1592 | break; | |
1593 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1594 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
1595 | func_id != BPF_FUNC_perf_event_output && |
1596 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
1597 | goto error; |
1598 | break; | |
1599 | case BPF_MAP_TYPE_STACK_TRACE: | |
1600 | if (func_id != BPF_FUNC_get_stackid) | |
1601 | goto error; | |
1602 | break; | |
4ed8ec52 | 1603 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1604 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1605 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1606 | goto error; |
1607 | break; | |
546ac1ff JF |
1608 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
1609 | * allow to be modified from bpf side. So do not allow lookup elements | |
1610 | * for now. | |
1611 | */ | |
1612 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 1613 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
1614 | goto error; |
1615 | break; | |
6710e112 JDB |
1616 | /* Restrict bpf side of cpumap, open when use-cases appear */ |
1617 | case BPF_MAP_TYPE_CPUMAP: | |
1618 | if (func_id != BPF_FUNC_redirect_map) | |
1619 | goto error; | |
1620 | break; | |
56f668df | 1621 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1622 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1623 | if (func_id != BPF_FUNC_map_lookup_elem) |
1624 | goto error; | |
16a43625 | 1625 | break; |
174a79ff JF |
1626 | case BPF_MAP_TYPE_SOCKMAP: |
1627 | if (func_id != BPF_FUNC_sk_redirect_map && | |
1628 | func_id != BPF_FUNC_sock_map_update && | |
1629 | func_id != BPF_FUNC_map_delete_elem) | |
1630 | goto error; | |
1631 | break; | |
6aff67c8 AS |
1632 | default: |
1633 | break; | |
1634 | } | |
1635 | ||
1636 | /* ... and second from the function itself. */ | |
1637 | switch (func_id) { | |
1638 | case BPF_FUNC_tail_call: | |
1639 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1640 | goto error; | |
1641 | break; | |
1642 | case BPF_FUNC_perf_event_read: | |
1643 | case BPF_FUNC_perf_event_output: | |
908432ca | 1644 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
1645 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
1646 | goto error; | |
1647 | break; | |
1648 | case BPF_FUNC_get_stackid: | |
1649 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1650 | goto error; | |
1651 | break; | |
60d20f91 | 1652 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1653 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1654 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1655 | goto error; | |
1656 | break; | |
97f91a7c | 1657 | case BPF_FUNC_redirect_map: |
9c270af3 JDB |
1658 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
1659 | map->map_type != BPF_MAP_TYPE_CPUMAP) | |
97f91a7c JF |
1660 | goto error; |
1661 | break; | |
174a79ff JF |
1662 | case BPF_FUNC_sk_redirect_map: |
1663 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1664 | goto error; | |
1665 | break; | |
1666 | case BPF_FUNC_sock_map_update: | |
1667 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
1668 | goto error; | |
1669 | break; | |
6aff67c8 AS |
1670 | default: |
1671 | break; | |
35578d79 KX |
1672 | } |
1673 | ||
1674 | return 0; | |
6aff67c8 | 1675 | error: |
61bd5218 | 1676 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 1677 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 1678 | return -EINVAL; |
35578d79 KX |
1679 | } |
1680 | ||
435faee1 DB |
1681 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1682 | { | |
1683 | int count = 0; | |
1684 | ||
39f19ebb | 1685 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1686 | count++; |
39f19ebb | 1687 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1688 | count++; |
39f19ebb | 1689 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1690 | count++; |
39f19ebb | 1691 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1692 | count++; |
39f19ebb | 1693 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1694 | count++; |
1695 | ||
1696 | return count > 1 ? -EINVAL : 0; | |
1697 | } | |
1698 | ||
de8f3a83 DB |
1699 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
1700 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 1701 | */ |
58e2af8b | 1702 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1703 | { |
638f5b90 | 1704 | struct bpf_verifier_state *state = env->cur_state; |
58e2af8b | 1705 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
1706 | int i; |
1707 | ||
1708 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 1709 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 1710 | mark_reg_unknown(env, regs, i); |
969bf05e | 1711 | |
638f5b90 AS |
1712 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
1713 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 1714 | continue; |
638f5b90 | 1715 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 DB |
1716 | if (reg_is_pkt_pointer_any(reg)) |
1717 | __mark_reg_unknown(reg); | |
969bf05e AS |
1718 | } |
1719 | } | |
1720 | ||
81ed18ab | 1721 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1722 | { |
17a52670 | 1723 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 1724 | struct bpf_reg_state *regs; |
33ff9823 | 1725 | struct bpf_call_arg_meta meta; |
969bf05e | 1726 | bool changes_data; |
17a52670 AS |
1727 | int i, err; |
1728 | ||
1729 | /* find function prototype */ | |
1730 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
1731 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
1732 | func_id); | |
17a52670 AS |
1733 | return -EINVAL; |
1734 | } | |
1735 | ||
00176a34 JK |
1736 | if (env->ops->get_func_proto) |
1737 | fn = env->ops->get_func_proto(func_id); | |
17a52670 AS |
1738 | |
1739 | if (!fn) { | |
61bd5218 JK |
1740 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
1741 | func_id); | |
17a52670 AS |
1742 | return -EINVAL; |
1743 | } | |
1744 | ||
1745 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1746 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
61bd5218 | 1747 | verbose(env, "cannot call GPL only function from proprietary program\n"); |
17a52670 AS |
1748 | return -EINVAL; |
1749 | } | |
1750 | ||
04514d13 | 1751 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 1752 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
1753 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
1754 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
1755 | func_id_name(func_id), func_id); | |
1756 | return -EINVAL; | |
1757 | } | |
969bf05e | 1758 | |
33ff9823 | 1759 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1760 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1761 | |
435faee1 DB |
1762 | /* We only support one arg being in raw mode at the moment, which |
1763 | * is sufficient for the helper functions we have right now. | |
1764 | */ | |
1765 | err = check_raw_mode(fn); | |
1766 | if (err) { | |
61bd5218 | 1767 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 1768 | func_id_name(func_id), func_id); |
435faee1 DB |
1769 | return err; |
1770 | } | |
1771 | ||
17a52670 | 1772 | /* check args */ |
33ff9823 | 1773 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1774 | if (err) |
1775 | return err; | |
33ff9823 | 1776 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1777 | if (err) |
1778 | return err; | |
b2157399 AS |
1779 | if (func_id == BPF_FUNC_tail_call) { |
1780 | if (meta.map_ptr == NULL) { | |
1781 | verbose(env, "verifier bug\n"); | |
1782 | return -EINVAL; | |
1783 | } | |
1784 | env->insn_aux_data[insn_idx].map_ptr = meta.map_ptr; | |
1785 | } | |
33ff9823 | 1786 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1787 | if (err) |
1788 | return err; | |
33ff9823 | 1789 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1790 | if (err) |
1791 | return err; | |
33ff9823 | 1792 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1793 | if (err) |
1794 | return err; | |
1795 | ||
435faee1 DB |
1796 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1797 | * is inferred from register state. | |
1798 | */ | |
1799 | for (i = 0; i < meta.access_size; i++) { | |
57b4a36e DB |
1800 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
1801 | BPF_WRITE, -1, false); | |
435faee1 DB |
1802 | if (err) |
1803 | return err; | |
1804 | } | |
1805 | ||
638f5b90 | 1806 | regs = cur_regs(env); |
17a52670 | 1807 | /* reset caller saved regs */ |
dc503a8a | 1808 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 1809 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
1810 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
1811 | } | |
17a52670 | 1812 | |
dc503a8a | 1813 | /* update return register (already marked as written above) */ |
17a52670 | 1814 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 1815 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 1816 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
1817 | } else if (fn->ret_type == RET_VOID) { |
1818 | regs[BPF_REG_0].type = NOT_INIT; | |
1819 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
1820 | struct bpf_insn_aux_data *insn_aux; |
1821 | ||
17a52670 | 1822 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
f1174f77 | 1823 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 1824 | mark_reg_known_zero(env, regs, BPF_REG_0); |
f1174f77 | 1825 | regs[BPF_REG_0].off = 0; |
17a52670 AS |
1826 | /* remember map_ptr, so that check_map_access() |
1827 | * can check 'value_size' boundary of memory access | |
1828 | * to map element returned from bpf_map_lookup_elem() | |
1829 | */ | |
33ff9823 | 1830 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
1831 | verbose(env, |
1832 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
1833 | return -EINVAL; |
1834 | } | |
33ff9823 | 1835 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1836 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
1837 | insn_aux = &env->insn_aux_data[insn_idx]; |
1838 | if (!insn_aux->map_ptr) | |
1839 | insn_aux->map_ptr = meta.map_ptr; | |
1840 | else if (insn_aux->map_ptr != meta.map_ptr) | |
1841 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 1842 | } else { |
61bd5218 | 1843 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 1844 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
1845 | return -EINVAL; |
1846 | } | |
04fd61ab | 1847 | |
61bd5218 | 1848 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
1849 | if (err) |
1850 | return err; | |
04fd61ab | 1851 | |
969bf05e AS |
1852 | if (changes_data) |
1853 | clear_all_pkt_pointers(env); | |
1854 | return 0; | |
1855 | } | |
1856 | ||
b03c9f9f EC |
1857 | static bool signed_add_overflows(s64 a, s64 b) |
1858 | { | |
1859 | /* Do the add in u64, where overflow is well-defined */ | |
1860 | s64 res = (s64)((u64)a + (u64)b); | |
1861 | ||
1862 | if (b < 0) | |
1863 | return res > a; | |
1864 | return res < a; | |
1865 | } | |
1866 | ||
1867 | static bool signed_sub_overflows(s64 a, s64 b) | |
1868 | { | |
1869 | /* Do the sub in u64, where overflow is well-defined */ | |
1870 | s64 res = (s64)((u64)a - (u64)b); | |
1871 | ||
1872 | if (b < 0) | |
1873 | return res < a; | |
1874 | return res > a; | |
969bf05e AS |
1875 | } |
1876 | ||
bb7f0f98 AS |
1877 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
1878 | const struct bpf_reg_state *reg, | |
1879 | enum bpf_reg_type type) | |
1880 | { | |
1881 | bool known = tnum_is_const(reg->var_off); | |
1882 | s64 val = reg->var_off.value; | |
1883 | s64 smin = reg->smin_value; | |
1884 | ||
1885 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
1886 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
1887 | reg_type_str[type], val); | |
1888 | return false; | |
1889 | } | |
1890 | ||
1891 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
1892 | verbose(env, "%s pointer offset %d is not allowed\n", | |
1893 | reg_type_str[type], reg->off); | |
1894 | return false; | |
1895 | } | |
1896 | ||
1897 | if (smin == S64_MIN) { | |
1898 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
1899 | reg_type_str[type]); | |
1900 | return false; | |
1901 | } | |
1902 | ||
1903 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
1904 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
1905 | smin, reg_type_str[type]); | |
1906 | return false; | |
1907 | } | |
1908 | ||
1909 | return true; | |
1910 | } | |
1911 | ||
f1174f77 | 1912 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
1913 | * Caller should also handle BPF_MOV case separately. |
1914 | * If we return -EACCES, caller may want to try again treating pointer as a | |
1915 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
1916 | */ | |
1917 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
1918 | struct bpf_insn *insn, | |
1919 | const struct bpf_reg_state *ptr_reg, | |
1920 | const struct bpf_reg_state *off_reg) | |
969bf05e | 1921 | { |
638f5b90 | 1922 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg; |
f1174f77 | 1923 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
1924 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
1925 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
1926 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
1927 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
969bf05e | 1928 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 1929 | u32 dst = insn->dst_reg; |
969bf05e | 1930 | |
f1174f77 | 1931 | dst_reg = ®s[dst]; |
969bf05e | 1932 | |
6f16101e DB |
1933 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
1934 | smin_val > smax_val || umin_val > umax_val) { | |
1935 | /* Taint dst register if offset had invalid bounds derived from | |
1936 | * e.g. dead branches. | |
1937 | */ | |
1938 | __mark_reg_unknown(dst_reg); | |
1939 | return 0; | |
f1174f77 EC |
1940 | } |
1941 | ||
1942 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
1943 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
1944 | verbose(env, |
1945 | "R%d 32-bit pointer arithmetic prohibited\n", | |
1946 | dst); | |
f1174f77 | 1947 | return -EACCES; |
969bf05e AS |
1948 | } |
1949 | ||
f1174f77 | 1950 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
82abbf8d AS |
1951 | verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", |
1952 | dst); | |
f1174f77 EC |
1953 | return -EACCES; |
1954 | } | |
1955 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
82abbf8d AS |
1956 | verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", |
1957 | dst); | |
f1174f77 EC |
1958 | return -EACCES; |
1959 | } | |
1960 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
82abbf8d AS |
1961 | verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", |
1962 | dst); | |
f1174f77 EC |
1963 | return -EACCES; |
1964 | } | |
1965 | ||
1966 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
1967 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 1968 | */ |
f1174f77 EC |
1969 | dst_reg->type = ptr_reg->type; |
1970 | dst_reg->id = ptr_reg->id; | |
969bf05e | 1971 | |
bb7f0f98 AS |
1972 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
1973 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
1974 | return -EINVAL; | |
1975 | ||
f1174f77 EC |
1976 | switch (opcode) { |
1977 | case BPF_ADD: | |
1978 | /* We can take a fixed offset as long as it doesn't overflow | |
1979 | * the s32 'off' field | |
969bf05e | 1980 | */ |
b03c9f9f EC |
1981 | if (known && (ptr_reg->off + smin_val == |
1982 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 1983 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
1984 | dst_reg->smin_value = smin_ptr; |
1985 | dst_reg->smax_value = smax_ptr; | |
1986 | dst_reg->umin_value = umin_ptr; | |
1987 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 1988 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 1989 | dst_reg->off = ptr_reg->off + smin_val; |
f1174f77 EC |
1990 | dst_reg->range = ptr_reg->range; |
1991 | break; | |
1992 | } | |
f1174f77 EC |
1993 | /* A new variable offset is created. Note that off_reg->off |
1994 | * == 0, since it's a scalar. | |
1995 | * dst_reg gets the pointer type and since some positive | |
1996 | * integer value was added to the pointer, give it a new 'id' | |
1997 | * if it's a PTR_TO_PACKET. | |
1998 | * this creates a new 'base' pointer, off_reg (variable) gets | |
1999 | * added into the variable offset, and we copy the fixed offset | |
2000 | * from ptr_reg. | |
969bf05e | 2001 | */ |
b03c9f9f EC |
2002 | if (signed_add_overflows(smin_ptr, smin_val) || |
2003 | signed_add_overflows(smax_ptr, smax_val)) { | |
2004 | dst_reg->smin_value = S64_MIN; | |
2005 | dst_reg->smax_value = S64_MAX; | |
2006 | } else { | |
2007 | dst_reg->smin_value = smin_ptr + smin_val; | |
2008 | dst_reg->smax_value = smax_ptr + smax_val; | |
2009 | } | |
2010 | if (umin_ptr + umin_val < umin_ptr || | |
2011 | umax_ptr + umax_val < umax_ptr) { | |
2012 | dst_reg->umin_value = 0; | |
2013 | dst_reg->umax_value = U64_MAX; | |
2014 | } else { | |
2015 | dst_reg->umin_value = umin_ptr + umin_val; | |
2016 | dst_reg->umax_value = umax_ptr + umax_val; | |
2017 | } | |
f1174f77 EC |
2018 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
2019 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 2020 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2021 | dst_reg->id = ++env->id_gen; |
2022 | /* something was added to pkt_ptr, set range to zero */ | |
2023 | dst_reg->range = 0; | |
2024 | } | |
2025 | break; | |
2026 | case BPF_SUB: | |
2027 | if (dst_reg == off_reg) { | |
2028 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
2029 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
2030 | dst); | |
f1174f77 EC |
2031 | return -EACCES; |
2032 | } | |
2033 | /* We don't allow subtraction from FP, because (according to | |
2034 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
2035 | * be able to deal with it. | |
969bf05e | 2036 | */ |
f1174f77 | 2037 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
2038 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
2039 | dst); | |
f1174f77 EC |
2040 | return -EACCES; |
2041 | } | |
b03c9f9f EC |
2042 | if (known && (ptr_reg->off - smin_val == |
2043 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 2044 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
2045 | dst_reg->smin_value = smin_ptr; |
2046 | dst_reg->smax_value = smax_ptr; | |
2047 | dst_reg->umin_value = umin_ptr; | |
2048 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
2049 | dst_reg->var_off = ptr_reg->var_off; |
2050 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 2051 | dst_reg->off = ptr_reg->off - smin_val; |
f1174f77 EC |
2052 | dst_reg->range = ptr_reg->range; |
2053 | break; | |
2054 | } | |
f1174f77 EC |
2055 | /* A new variable offset is created. If the subtrahend is known |
2056 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 2057 | */ |
b03c9f9f EC |
2058 | if (signed_sub_overflows(smin_ptr, smax_val) || |
2059 | signed_sub_overflows(smax_ptr, smin_val)) { | |
2060 | /* Overflow possible, we know nothing */ | |
2061 | dst_reg->smin_value = S64_MIN; | |
2062 | dst_reg->smax_value = S64_MAX; | |
2063 | } else { | |
2064 | dst_reg->smin_value = smin_ptr - smax_val; | |
2065 | dst_reg->smax_value = smax_ptr - smin_val; | |
2066 | } | |
2067 | if (umin_ptr < umax_val) { | |
2068 | /* Overflow possible, we know nothing */ | |
2069 | dst_reg->umin_value = 0; | |
2070 | dst_reg->umax_value = U64_MAX; | |
2071 | } else { | |
2072 | /* Cannot overflow (as long as bounds are consistent) */ | |
2073 | dst_reg->umin_value = umin_ptr - umax_val; | |
2074 | dst_reg->umax_value = umax_ptr - umin_val; | |
2075 | } | |
f1174f77 EC |
2076 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
2077 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 2078 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2079 | dst_reg->id = ++env->id_gen; |
2080 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 2081 | if (smin_val < 0) |
f1174f77 | 2082 | dst_reg->range = 0; |
43188702 | 2083 | } |
f1174f77 EC |
2084 | break; |
2085 | case BPF_AND: | |
2086 | case BPF_OR: | |
2087 | case BPF_XOR: | |
82abbf8d AS |
2088 | /* bitwise ops on pointers are troublesome, prohibit. */ |
2089 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
2090 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
2091 | return -EACCES; |
2092 | default: | |
2093 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
2094 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
2095 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 2096 | return -EACCES; |
43188702 JF |
2097 | } |
2098 | ||
bb7f0f98 AS |
2099 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
2100 | return -EINVAL; | |
2101 | ||
b03c9f9f EC |
2102 | __update_reg_bounds(dst_reg); |
2103 | __reg_deduce_bounds(dst_reg); | |
2104 | __reg_bound_offset(dst_reg); | |
43188702 JF |
2105 | return 0; |
2106 | } | |
2107 | ||
468f6eaf JH |
2108 | /* WARNING: This function does calculations on 64-bit values, but the actual |
2109 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
2110 | * need extra checks in the 32-bit case. | |
2111 | */ | |
f1174f77 EC |
2112 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
2113 | struct bpf_insn *insn, | |
2114 | struct bpf_reg_state *dst_reg, | |
2115 | struct bpf_reg_state src_reg) | |
969bf05e | 2116 | { |
638f5b90 | 2117 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 2118 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2119 | bool src_known, dst_known; |
b03c9f9f EC |
2120 | s64 smin_val, smax_val; |
2121 | u64 umin_val, umax_val; | |
468f6eaf | 2122 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
48461135 | 2123 | |
4c6d059c JH |
2124 | if (insn_bitness == 32) { |
2125 | /* Relevant for 32-bit RSH: Information can propagate towards | |
2126 | * LSB, so it isn't sufficient to only truncate the output to | |
2127 | * 32 bits. | |
2128 | */ | |
2129 | coerce_reg_to_size(dst_reg, 4); | |
2130 | coerce_reg_to_size(&src_reg, 4); | |
2131 | } | |
2132 | ||
b03c9f9f EC |
2133 | smin_val = src_reg.smin_value; |
2134 | smax_val = src_reg.smax_value; | |
2135 | umin_val = src_reg.umin_value; | |
2136 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
2137 | src_known = tnum_is_const(src_reg.var_off); |
2138 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 2139 | |
6f16101e DB |
2140 | if ((src_known && (smin_val != smax_val || umin_val != umax_val)) || |
2141 | smin_val > smax_val || umin_val > umax_val) { | |
2142 | /* Taint dst register if offset had invalid bounds derived from | |
2143 | * e.g. dead branches. | |
2144 | */ | |
2145 | __mark_reg_unknown(dst_reg); | |
2146 | return 0; | |
2147 | } | |
2148 | ||
bb7f0f98 AS |
2149 | if (!src_known && |
2150 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
2151 | __mark_reg_unknown(dst_reg); | |
2152 | return 0; | |
2153 | } | |
2154 | ||
48461135 JB |
2155 | switch (opcode) { |
2156 | case BPF_ADD: | |
b03c9f9f EC |
2157 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
2158 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
2159 | dst_reg->smin_value = S64_MIN; | |
2160 | dst_reg->smax_value = S64_MAX; | |
2161 | } else { | |
2162 | dst_reg->smin_value += smin_val; | |
2163 | dst_reg->smax_value += smax_val; | |
2164 | } | |
2165 | if (dst_reg->umin_value + umin_val < umin_val || | |
2166 | dst_reg->umax_value + umax_val < umax_val) { | |
2167 | dst_reg->umin_value = 0; | |
2168 | dst_reg->umax_value = U64_MAX; | |
2169 | } else { | |
2170 | dst_reg->umin_value += umin_val; | |
2171 | dst_reg->umax_value += umax_val; | |
2172 | } | |
f1174f77 | 2173 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2174 | break; |
2175 | case BPF_SUB: | |
b03c9f9f EC |
2176 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
2177 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
2178 | /* Overflow possible, we know nothing */ | |
2179 | dst_reg->smin_value = S64_MIN; | |
2180 | dst_reg->smax_value = S64_MAX; | |
2181 | } else { | |
2182 | dst_reg->smin_value -= smax_val; | |
2183 | dst_reg->smax_value -= smin_val; | |
2184 | } | |
2185 | if (dst_reg->umin_value < umax_val) { | |
2186 | /* Overflow possible, we know nothing */ | |
2187 | dst_reg->umin_value = 0; | |
2188 | dst_reg->umax_value = U64_MAX; | |
2189 | } else { | |
2190 | /* Cannot overflow (as long as bounds are consistent) */ | |
2191 | dst_reg->umin_value -= umax_val; | |
2192 | dst_reg->umax_value -= umin_val; | |
2193 | } | |
f1174f77 | 2194 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2195 | break; |
2196 | case BPF_MUL: | |
b03c9f9f EC |
2197 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
2198 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 2199 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
2200 | __mark_reg_unbounded(dst_reg); |
2201 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2202 | break; |
2203 | } | |
b03c9f9f EC |
2204 | /* Both values are positive, so we can work with unsigned and |
2205 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 2206 | */ |
b03c9f9f EC |
2207 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
2208 | /* Potential overflow, we know nothing */ | |
2209 | __mark_reg_unbounded(dst_reg); | |
2210 | /* (except what we can learn from the var_off) */ | |
2211 | __update_reg_bounds(dst_reg); | |
2212 | break; | |
2213 | } | |
2214 | dst_reg->umin_value *= umin_val; | |
2215 | dst_reg->umax_value *= umax_val; | |
2216 | if (dst_reg->umax_value > S64_MAX) { | |
2217 | /* Overflow possible, we know nothing */ | |
2218 | dst_reg->smin_value = S64_MIN; | |
2219 | dst_reg->smax_value = S64_MAX; | |
2220 | } else { | |
2221 | dst_reg->smin_value = dst_reg->umin_value; | |
2222 | dst_reg->smax_value = dst_reg->umax_value; | |
2223 | } | |
48461135 JB |
2224 | break; |
2225 | case BPF_AND: | |
f1174f77 | 2226 | if (src_known && dst_known) { |
b03c9f9f EC |
2227 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
2228 | src_reg.var_off.value); | |
f1174f77 EC |
2229 | break; |
2230 | } | |
b03c9f9f EC |
2231 | /* We get our minimum from the var_off, since that's inherently |
2232 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 2233 | */ |
f1174f77 | 2234 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
2235 | dst_reg->umin_value = dst_reg->var_off.value; |
2236 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
2237 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2238 | /* Lose signed bounds when ANDing negative numbers, | |
2239 | * ain't nobody got time for that. | |
2240 | */ | |
2241 | dst_reg->smin_value = S64_MIN; | |
2242 | dst_reg->smax_value = S64_MAX; | |
2243 | } else { | |
2244 | /* ANDing two positives gives a positive, so safe to | |
2245 | * cast result into s64. | |
2246 | */ | |
2247 | dst_reg->smin_value = dst_reg->umin_value; | |
2248 | dst_reg->smax_value = dst_reg->umax_value; | |
2249 | } | |
2250 | /* We may learn something more from the var_off */ | |
2251 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2252 | break; |
2253 | case BPF_OR: | |
2254 | if (src_known && dst_known) { | |
b03c9f9f EC |
2255 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
2256 | src_reg.var_off.value); | |
f1174f77 EC |
2257 | break; |
2258 | } | |
b03c9f9f EC |
2259 | /* We get our maximum from the var_off, and our minimum is the |
2260 | * maximum of the operands' minima | |
f1174f77 EC |
2261 | */ |
2262 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
2263 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
2264 | dst_reg->umax_value = dst_reg->var_off.value | | |
2265 | dst_reg->var_off.mask; | |
2266 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2267 | /* Lose signed bounds when ORing negative numbers, | |
2268 | * ain't nobody got time for that. | |
2269 | */ | |
2270 | dst_reg->smin_value = S64_MIN; | |
2271 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2272 | } else { |
b03c9f9f EC |
2273 | /* ORing two positives gives a positive, so safe to |
2274 | * cast result into s64. | |
2275 | */ | |
2276 | dst_reg->smin_value = dst_reg->umin_value; | |
2277 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 2278 | } |
b03c9f9f EC |
2279 | /* We may learn something more from the var_off */ |
2280 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2281 | break; |
2282 | case BPF_LSH: | |
468f6eaf JH |
2283 | if (umax_val >= insn_bitness) { |
2284 | /* Shifts greater than 31 or 63 are undefined. | |
2285 | * This includes shifts by a negative number. | |
b03c9f9f | 2286 | */ |
61bd5218 | 2287 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2288 | break; |
2289 | } | |
b03c9f9f EC |
2290 | /* We lose all sign bit information (except what we can pick |
2291 | * up from var_off) | |
48461135 | 2292 | */ |
b03c9f9f EC |
2293 | dst_reg->smin_value = S64_MIN; |
2294 | dst_reg->smax_value = S64_MAX; | |
2295 | /* If we might shift our top bit out, then we know nothing */ | |
2296 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
2297 | dst_reg->umin_value = 0; | |
2298 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 2299 | } else { |
b03c9f9f EC |
2300 | dst_reg->umin_value <<= umin_val; |
2301 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 2302 | } |
b03c9f9f EC |
2303 | if (src_known) |
2304 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); | |
2305 | else | |
2306 | dst_reg->var_off = tnum_lshift(tnum_unknown, umin_val); | |
2307 | /* We may learn something more from the var_off */ | |
2308 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2309 | break; |
2310 | case BPF_RSH: | |
468f6eaf JH |
2311 | if (umax_val >= insn_bitness) { |
2312 | /* Shifts greater than 31 or 63 are undefined. | |
2313 | * This includes shifts by a negative number. | |
b03c9f9f | 2314 | */ |
61bd5218 | 2315 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2316 | break; |
2317 | } | |
4374f256 EC |
2318 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
2319 | * be negative, then either: | |
2320 | * 1) src_reg might be zero, so the sign bit of the result is | |
2321 | * unknown, so we lose our signed bounds | |
2322 | * 2) it's known negative, thus the unsigned bounds capture the | |
2323 | * signed bounds | |
2324 | * 3) the signed bounds cross zero, so they tell us nothing | |
2325 | * about the result | |
2326 | * If the value in dst_reg is known nonnegative, then again the | |
2327 | * unsigned bounts capture the signed bounds. | |
2328 | * Thus, in all cases it suffices to blow away our signed bounds | |
2329 | * and rely on inferring new ones from the unsigned bounds and | |
2330 | * var_off of the result. | |
2331 | */ | |
2332 | dst_reg->smin_value = S64_MIN; | |
2333 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2334 | if (src_known) |
b03c9f9f EC |
2335 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, |
2336 | umin_val); | |
f1174f77 | 2337 | else |
b03c9f9f EC |
2338 | dst_reg->var_off = tnum_rshift(tnum_unknown, umin_val); |
2339 | dst_reg->umin_value >>= umax_val; | |
2340 | dst_reg->umax_value >>= umin_val; | |
2341 | /* We may learn something more from the var_off */ | |
2342 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2343 | break; |
2344 | default: | |
61bd5218 | 2345 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
2346 | break; |
2347 | } | |
2348 | ||
468f6eaf JH |
2349 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
2350 | /* 32-bit ALU ops are (32,32)->32 */ | |
2351 | coerce_reg_to_size(dst_reg, 4); | |
468f6eaf JH |
2352 | } |
2353 | ||
b03c9f9f EC |
2354 | __reg_deduce_bounds(dst_reg); |
2355 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
2356 | return 0; |
2357 | } | |
2358 | ||
2359 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
2360 | * and var_off. | |
2361 | */ | |
2362 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
2363 | struct bpf_insn *insn) | |
2364 | { | |
638f5b90 | 2365 | struct bpf_reg_state *regs = cur_regs(env), *dst_reg, *src_reg; |
f1174f77 EC |
2366 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
2367 | u8 opcode = BPF_OP(insn->code); | |
f1174f77 EC |
2368 | |
2369 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
2370 | src_reg = NULL; |
2371 | if (dst_reg->type != SCALAR_VALUE) | |
2372 | ptr_reg = dst_reg; | |
2373 | if (BPF_SRC(insn->code) == BPF_X) { | |
2374 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
2375 | if (src_reg->type != SCALAR_VALUE) { |
2376 | if (dst_reg->type != SCALAR_VALUE) { | |
2377 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
2378 | * an arbitrary scalar. Disallow all math except |
2379 | * pointer subtraction | |
f1174f77 | 2380 | */ |
82abbf8d AS |
2381 | if (opcode == BPF_SUB){ |
2382 | mark_reg_unknown(env, regs, insn->dst_reg); | |
2383 | return 0; | |
f1174f77 | 2384 | } |
82abbf8d AS |
2385 | verbose(env, "R%d pointer %s pointer prohibited\n", |
2386 | insn->dst_reg, | |
2387 | bpf_alu_string[opcode >> 4]); | |
2388 | return -EACCES; | |
f1174f77 EC |
2389 | } else { |
2390 | /* scalar += pointer | |
2391 | * This is legal, but we have to reverse our | |
2392 | * src/dest handling in computing the range | |
2393 | */ | |
82abbf8d AS |
2394 | return adjust_ptr_min_max_vals(env, insn, |
2395 | src_reg, dst_reg); | |
f1174f77 EC |
2396 | } |
2397 | } else if (ptr_reg) { | |
2398 | /* pointer += scalar */ | |
82abbf8d AS |
2399 | return adjust_ptr_min_max_vals(env, insn, |
2400 | dst_reg, src_reg); | |
f1174f77 EC |
2401 | } |
2402 | } else { | |
2403 | /* Pretend the src is a reg with a known value, since we only | |
2404 | * need to be able to read from this state. | |
2405 | */ | |
2406 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 2407 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 2408 | src_reg = &off_reg; |
82abbf8d AS |
2409 | if (ptr_reg) /* pointer += K */ |
2410 | return adjust_ptr_min_max_vals(env, insn, | |
2411 | ptr_reg, src_reg); | |
f1174f77 EC |
2412 | } |
2413 | ||
2414 | /* Got here implies adding two SCALAR_VALUEs */ | |
2415 | if (WARN_ON_ONCE(ptr_reg)) { | |
638f5b90 | 2416 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2417 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
2418 | return -EINVAL; |
2419 | } | |
2420 | if (WARN_ON(!src_reg)) { | |
638f5b90 | 2421 | print_verifier_state(env, env->cur_state); |
61bd5218 | 2422 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
2423 | return -EINVAL; |
2424 | } | |
2425 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
2426 | } |
2427 | ||
17a52670 | 2428 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 2429 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2430 | { |
638f5b90 | 2431 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
2432 | u8 opcode = BPF_OP(insn->code); |
2433 | int err; | |
2434 | ||
2435 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
2436 | if (opcode == BPF_NEG) { | |
2437 | if (BPF_SRC(insn->code) != 0 || | |
2438 | insn->src_reg != BPF_REG_0 || | |
2439 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 2440 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
2441 | return -EINVAL; |
2442 | } | |
2443 | } else { | |
2444 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
2445 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
2446 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 2447 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
2448 | return -EINVAL; |
2449 | } | |
2450 | } | |
2451 | ||
2452 | /* check src operand */ | |
dc503a8a | 2453 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2454 | if (err) |
2455 | return err; | |
2456 | ||
1be7f75d | 2457 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 2458 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
2459 | insn->dst_reg); |
2460 | return -EACCES; | |
2461 | } | |
2462 | ||
17a52670 | 2463 | /* check dest operand */ |
dc503a8a | 2464 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2465 | if (err) |
2466 | return err; | |
2467 | ||
2468 | } else if (opcode == BPF_MOV) { | |
2469 | ||
2470 | if (BPF_SRC(insn->code) == BPF_X) { | |
2471 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2472 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2473 | return -EINVAL; |
2474 | } | |
2475 | ||
2476 | /* check src operand */ | |
dc503a8a | 2477 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2478 | if (err) |
2479 | return err; | |
2480 | } else { | |
2481 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2482 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
2483 | return -EINVAL; |
2484 | } | |
2485 | } | |
2486 | ||
2487 | /* check dest operand */ | |
dc503a8a | 2488 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
2489 | if (err) |
2490 | return err; | |
2491 | ||
2492 | if (BPF_SRC(insn->code) == BPF_X) { | |
2493 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
2494 | /* case: R1 = R2 | |
2495 | * copy register state to dest reg | |
2496 | */ | |
2497 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
8fe2d6cc | 2498 | regs[insn->dst_reg].live |= REG_LIVE_WRITTEN; |
17a52670 | 2499 | } else { |
f1174f77 | 2500 | /* R1 = (u32) R2 */ |
1be7f75d | 2501 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
2502 | verbose(env, |
2503 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
2504 | insn->src_reg); |
2505 | return -EACCES; | |
2506 | } | |
61bd5218 | 2507 | mark_reg_unknown(env, regs, insn->dst_reg); |
0c17d1d2 | 2508 | coerce_reg_to_size(®s[insn->dst_reg], 4); |
17a52670 AS |
2509 | } |
2510 | } else { | |
2511 | /* case: R = imm | |
2512 | * remember the value we stored into this reg | |
2513 | */ | |
f1174f77 | 2514 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
2515 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
2516 | __mark_reg_known(regs + insn->dst_reg, | |
2517 | insn->imm); | |
2518 | } else { | |
2519 | __mark_reg_known(regs + insn->dst_reg, | |
2520 | (u32)insn->imm); | |
2521 | } | |
17a52670 AS |
2522 | } |
2523 | ||
2524 | } else if (opcode > BPF_END) { | |
61bd5218 | 2525 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
2526 | return -EINVAL; |
2527 | ||
2528 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
2529 | ||
17a52670 AS |
2530 | if (BPF_SRC(insn->code) == BPF_X) { |
2531 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 2532 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2533 | return -EINVAL; |
2534 | } | |
2535 | /* check src1 operand */ | |
dc503a8a | 2536 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2537 | if (err) |
2538 | return err; | |
2539 | } else { | |
2540 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 2541 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
2542 | return -EINVAL; |
2543 | } | |
2544 | } | |
2545 | ||
2546 | /* check src2 operand */ | |
dc503a8a | 2547 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2548 | if (err) |
2549 | return err; | |
2550 | ||
2551 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
2552 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 2553 | verbose(env, "div by zero\n"); |
17a52670 AS |
2554 | return -EINVAL; |
2555 | } | |
2556 | ||
7891a87e DB |
2557 | if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) { |
2558 | verbose(env, "BPF_ARSH not supported for 32 bit ALU\n"); | |
2559 | return -EINVAL; | |
2560 | } | |
2561 | ||
229394e8 RV |
2562 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
2563 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
2564 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
2565 | ||
2566 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 2567 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
2568 | return -EINVAL; |
2569 | } | |
2570 | } | |
2571 | ||
1a0dc1ac | 2572 | /* check dest operand */ |
dc503a8a | 2573 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
2574 | if (err) |
2575 | return err; | |
2576 | ||
f1174f77 | 2577 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
2578 | } |
2579 | ||
2580 | return 0; | |
2581 | } | |
2582 | ||
58e2af8b | 2583 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
de8f3a83 | 2584 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 2585 | enum bpf_reg_type type, |
fb2a311a | 2586 | bool range_right_open) |
969bf05e | 2587 | { |
58e2af8b | 2588 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 2589 | u16 new_range; |
969bf05e | 2590 | int i; |
2d2be8ca | 2591 | |
fb2a311a DB |
2592 | if (dst_reg->off < 0 || |
2593 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
2594 | /* This doesn't give us any range */ |
2595 | return; | |
2596 | ||
b03c9f9f EC |
2597 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
2598 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
2599 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
2600 | * than pkt_end, but that's because it's also less than pkt. | |
2601 | */ | |
2602 | return; | |
2603 | ||
fb2a311a DB |
2604 | new_range = dst_reg->off; |
2605 | if (range_right_open) | |
2606 | new_range--; | |
2607 | ||
2608 | /* Examples for register markings: | |
2d2be8ca | 2609 | * |
fb2a311a | 2610 | * pkt_data in dst register: |
2d2be8ca DB |
2611 | * |
2612 | * r2 = r3; | |
2613 | * r2 += 8; | |
2614 | * if (r2 > pkt_end) goto <handle exception> | |
2615 | * <access okay> | |
2616 | * | |
b4e432f1 DB |
2617 | * r2 = r3; |
2618 | * r2 += 8; | |
2619 | * if (r2 < pkt_end) goto <access okay> | |
2620 | * <handle exception> | |
2621 | * | |
2d2be8ca DB |
2622 | * Where: |
2623 | * r2 == dst_reg, pkt_end == src_reg | |
2624 | * r2=pkt(id=n,off=8,r=0) | |
2625 | * r3=pkt(id=n,off=0,r=0) | |
2626 | * | |
fb2a311a | 2627 | * pkt_data in src register: |
2d2be8ca DB |
2628 | * |
2629 | * r2 = r3; | |
2630 | * r2 += 8; | |
2631 | * if (pkt_end >= r2) goto <access okay> | |
2632 | * <handle exception> | |
2633 | * | |
b4e432f1 DB |
2634 | * r2 = r3; |
2635 | * r2 += 8; | |
2636 | * if (pkt_end <= r2) goto <handle exception> | |
2637 | * <access okay> | |
2638 | * | |
2d2be8ca DB |
2639 | * Where: |
2640 | * pkt_end == dst_reg, r2 == src_reg | |
2641 | * r2=pkt(id=n,off=8,r=0) | |
2642 | * r3=pkt(id=n,off=0,r=0) | |
2643 | * | |
2644 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
2645 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
2646 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
2647 | * the check. | |
969bf05e | 2648 | */ |
2d2be8ca | 2649 | |
f1174f77 EC |
2650 | /* If our ids match, then we must have the same max_value. And we |
2651 | * don't care about the other reg's fixed offset, since if it's too big | |
2652 | * the range won't allow anything. | |
2653 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
2654 | */ | |
969bf05e | 2655 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 2656 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 2657 | /* keep the maximum range already checked */ |
fb2a311a | 2658 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 2659 | |
638f5b90 AS |
2660 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2661 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 2662 | continue; |
638f5b90 | 2663 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 | 2664 | if (reg->type == type && reg->id == dst_reg->id) |
b06723da | 2665 | reg->range = max(reg->range, new_range); |
969bf05e AS |
2666 | } |
2667 | } | |
2668 | ||
48461135 JB |
2669 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2670 | * variable register that we are working on, and src_reg is a constant or we're | |
2671 | * simply doing a BPF_K check. | |
f1174f77 | 2672 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
2673 | */ |
2674 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2675 | struct bpf_reg_state *false_reg, u64 val, | |
2676 | u8 opcode) | |
2677 | { | |
f1174f77 EC |
2678 | /* If the dst_reg is a pointer, we can't learn anything about its |
2679 | * variable offset from the compare (unless src_reg were a pointer into | |
2680 | * the same object, but we don't bother with that. | |
2681 | * Since false_reg and true_reg have the same type by construction, we | |
2682 | * only need to check one of them for pointerness. | |
2683 | */ | |
2684 | if (__is_pointer_value(false, false_reg)) | |
2685 | return; | |
4cabc5b1 | 2686 | |
48461135 JB |
2687 | switch (opcode) { |
2688 | case BPF_JEQ: | |
2689 | /* If this is false then we know nothing Jon Snow, but if it is | |
2690 | * true then we know for sure. | |
2691 | */ | |
b03c9f9f | 2692 | __mark_reg_known(true_reg, val); |
48461135 JB |
2693 | break; |
2694 | case BPF_JNE: | |
2695 | /* If this is true we know nothing Jon Snow, but if it is false | |
2696 | * we know the value for sure; | |
2697 | */ | |
b03c9f9f | 2698 | __mark_reg_known(false_reg, val); |
48461135 JB |
2699 | break; |
2700 | case BPF_JGT: | |
b03c9f9f EC |
2701 | false_reg->umax_value = min(false_reg->umax_value, val); |
2702 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2703 | break; | |
48461135 | 2704 | case BPF_JSGT: |
b03c9f9f EC |
2705 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
2706 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 2707 | break; |
b4e432f1 DB |
2708 | case BPF_JLT: |
2709 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2710 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
2711 | break; | |
2712 | case BPF_JSLT: | |
2713 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
2714 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
2715 | break; | |
48461135 | 2716 | case BPF_JGE: |
b03c9f9f EC |
2717 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
2718 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2719 | break; | |
48461135 | 2720 | case BPF_JSGE: |
b03c9f9f EC |
2721 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
2722 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 2723 | break; |
b4e432f1 DB |
2724 | case BPF_JLE: |
2725 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2726 | true_reg->umax_value = min(true_reg->umax_value, val); | |
2727 | break; | |
2728 | case BPF_JSLE: | |
2729 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
2730 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
2731 | break; | |
48461135 JB |
2732 | default: |
2733 | break; | |
2734 | } | |
2735 | ||
b03c9f9f EC |
2736 | __reg_deduce_bounds(false_reg); |
2737 | __reg_deduce_bounds(true_reg); | |
2738 | /* We might have learned some bits from the bounds. */ | |
2739 | __reg_bound_offset(false_reg); | |
2740 | __reg_bound_offset(true_reg); | |
2741 | /* Intersecting with the old var_off might have improved our bounds | |
2742 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2743 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2744 | */ | |
2745 | __update_reg_bounds(false_reg); | |
2746 | __update_reg_bounds(true_reg); | |
48461135 JB |
2747 | } |
2748 | ||
f1174f77 EC |
2749 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
2750 | * the variable reg. | |
48461135 JB |
2751 | */ |
2752 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
2753 | struct bpf_reg_state *false_reg, u64 val, | |
2754 | u8 opcode) | |
2755 | { | |
f1174f77 EC |
2756 | if (__is_pointer_value(false, false_reg)) |
2757 | return; | |
4cabc5b1 | 2758 | |
48461135 JB |
2759 | switch (opcode) { |
2760 | case BPF_JEQ: | |
2761 | /* If this is false then we know nothing Jon Snow, but if it is | |
2762 | * true then we know for sure. | |
2763 | */ | |
b03c9f9f | 2764 | __mark_reg_known(true_reg, val); |
48461135 JB |
2765 | break; |
2766 | case BPF_JNE: | |
2767 | /* If this is true we know nothing Jon Snow, but if it is false | |
2768 | * we know the value for sure; | |
2769 | */ | |
b03c9f9f | 2770 | __mark_reg_known(false_reg, val); |
48461135 JB |
2771 | break; |
2772 | case BPF_JGT: | |
b03c9f9f EC |
2773 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
2774 | false_reg->umin_value = max(false_reg->umin_value, val); | |
2775 | break; | |
48461135 | 2776 | case BPF_JSGT: |
b03c9f9f EC |
2777 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
2778 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 2779 | break; |
b4e432f1 DB |
2780 | case BPF_JLT: |
2781 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
2782 | false_reg->umax_value = min(false_reg->umax_value, val); | |
2783 | break; | |
2784 | case BPF_JSLT: | |
2785 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
2786 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
2787 | break; | |
48461135 | 2788 | case BPF_JGE: |
b03c9f9f EC |
2789 | true_reg->umax_value = min(true_reg->umax_value, val); |
2790 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
2791 | break; | |
48461135 | 2792 | case BPF_JSGE: |
b03c9f9f EC |
2793 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
2794 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 2795 | break; |
b4e432f1 DB |
2796 | case BPF_JLE: |
2797 | true_reg->umin_value = max(true_reg->umin_value, val); | |
2798 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
2799 | break; | |
2800 | case BPF_JSLE: | |
2801 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
2802 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
2803 | break; | |
48461135 JB |
2804 | default: |
2805 | break; | |
2806 | } | |
2807 | ||
b03c9f9f EC |
2808 | __reg_deduce_bounds(false_reg); |
2809 | __reg_deduce_bounds(true_reg); | |
2810 | /* We might have learned some bits from the bounds. */ | |
2811 | __reg_bound_offset(false_reg); | |
2812 | __reg_bound_offset(true_reg); | |
2813 | /* Intersecting with the old var_off might have improved our bounds | |
2814 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2815 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2816 | */ | |
2817 | __update_reg_bounds(false_reg); | |
2818 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
2819 | } |
2820 | ||
2821 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
2822 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
2823 | struct bpf_reg_state *dst_reg) | |
2824 | { | |
b03c9f9f EC |
2825 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
2826 | dst_reg->umin_value); | |
2827 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
2828 | dst_reg->umax_value); | |
2829 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
2830 | dst_reg->smin_value); | |
2831 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
2832 | dst_reg->smax_value); | |
f1174f77 EC |
2833 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
2834 | dst_reg->var_off); | |
b03c9f9f EC |
2835 | /* We might have learned new bounds from the var_off. */ |
2836 | __update_reg_bounds(src_reg); | |
2837 | __update_reg_bounds(dst_reg); | |
2838 | /* We might have learned something about the sign bit. */ | |
2839 | __reg_deduce_bounds(src_reg); | |
2840 | __reg_deduce_bounds(dst_reg); | |
2841 | /* We might have learned some bits from the bounds. */ | |
2842 | __reg_bound_offset(src_reg); | |
2843 | __reg_bound_offset(dst_reg); | |
2844 | /* Intersecting with the old var_off might have improved our bounds | |
2845 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
2846 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
2847 | */ | |
2848 | __update_reg_bounds(src_reg); | |
2849 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2850 | } |
2851 | ||
2852 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
2853 | struct bpf_reg_state *true_dst, | |
2854 | struct bpf_reg_state *false_src, | |
2855 | struct bpf_reg_state *false_dst, | |
2856 | u8 opcode) | |
2857 | { | |
2858 | switch (opcode) { | |
2859 | case BPF_JEQ: | |
2860 | __reg_combine_min_max(true_src, true_dst); | |
2861 | break; | |
2862 | case BPF_JNE: | |
2863 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 2864 | break; |
4cabc5b1 | 2865 | } |
48461135 JB |
2866 | } |
2867 | ||
57a09bf0 | 2868 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 2869 | bool is_null) |
57a09bf0 TG |
2870 | { |
2871 | struct bpf_reg_state *reg = ®s[regno]; | |
2872 | ||
2873 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
2874 | /* Old offset (both fixed and variable parts) should |
2875 | * have been known-zero, because we don't allow pointer | |
2876 | * arithmetic on pointers that might be NULL. | |
2877 | */ | |
b03c9f9f EC |
2878 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
2879 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 2880 | reg->off)) { |
b03c9f9f EC |
2881 | __mark_reg_known_zero(reg); |
2882 | reg->off = 0; | |
f1174f77 EC |
2883 | } |
2884 | if (is_null) { | |
2885 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
2886 | } else if (reg->map_ptr->inner_map_meta) { |
2887 | reg->type = CONST_PTR_TO_MAP; | |
2888 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
2889 | } else { | |
f1174f77 | 2890 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 2891 | } |
a08dd0da DB |
2892 | /* We don't need id from this point onwards anymore, thus we |
2893 | * should better reset it, so that state pruning has chances | |
2894 | * to take effect. | |
2895 | */ | |
2896 | reg->id = 0; | |
57a09bf0 TG |
2897 | } |
2898 | } | |
2899 | ||
2900 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
2901 | * be folded together at some point. | |
2902 | */ | |
2903 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
f1174f77 | 2904 | bool is_null) |
57a09bf0 TG |
2905 | { |
2906 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 2907 | u32 id = regs[regno].id; |
57a09bf0 TG |
2908 | int i; |
2909 | ||
2910 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 2911 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 | 2912 | |
638f5b90 AS |
2913 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2914 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
57a09bf0 | 2915 | continue; |
638f5b90 | 2916 | mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null); |
57a09bf0 TG |
2917 | } |
2918 | } | |
2919 | ||
5beca081 DB |
2920 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
2921 | struct bpf_reg_state *dst_reg, | |
2922 | struct bpf_reg_state *src_reg, | |
2923 | struct bpf_verifier_state *this_branch, | |
2924 | struct bpf_verifier_state *other_branch) | |
2925 | { | |
2926 | if (BPF_SRC(insn->code) != BPF_X) | |
2927 | return false; | |
2928 | ||
2929 | switch (BPF_OP(insn->code)) { | |
2930 | case BPF_JGT: | |
2931 | if ((dst_reg->type == PTR_TO_PACKET && | |
2932 | src_reg->type == PTR_TO_PACKET_END) || | |
2933 | (dst_reg->type == PTR_TO_PACKET_META && | |
2934 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2935 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
2936 | find_good_pkt_pointers(this_branch, dst_reg, | |
2937 | dst_reg->type, false); | |
2938 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2939 | src_reg->type == PTR_TO_PACKET) || | |
2940 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2941 | src_reg->type == PTR_TO_PACKET_META)) { | |
2942 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
2943 | find_good_pkt_pointers(other_branch, src_reg, | |
2944 | src_reg->type, true); | |
2945 | } else { | |
2946 | return false; | |
2947 | } | |
2948 | break; | |
2949 | case BPF_JLT: | |
2950 | if ((dst_reg->type == PTR_TO_PACKET && | |
2951 | src_reg->type == PTR_TO_PACKET_END) || | |
2952 | (dst_reg->type == PTR_TO_PACKET_META && | |
2953 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2954 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
2955 | find_good_pkt_pointers(other_branch, dst_reg, | |
2956 | dst_reg->type, true); | |
2957 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2958 | src_reg->type == PTR_TO_PACKET) || | |
2959 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2960 | src_reg->type == PTR_TO_PACKET_META)) { | |
2961 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
2962 | find_good_pkt_pointers(this_branch, src_reg, | |
2963 | src_reg->type, false); | |
2964 | } else { | |
2965 | return false; | |
2966 | } | |
2967 | break; | |
2968 | case BPF_JGE: | |
2969 | if ((dst_reg->type == PTR_TO_PACKET && | |
2970 | src_reg->type == PTR_TO_PACKET_END) || | |
2971 | (dst_reg->type == PTR_TO_PACKET_META && | |
2972 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2973 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
2974 | find_good_pkt_pointers(this_branch, dst_reg, | |
2975 | dst_reg->type, true); | |
2976 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2977 | src_reg->type == PTR_TO_PACKET) || | |
2978 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2979 | src_reg->type == PTR_TO_PACKET_META)) { | |
2980 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
2981 | find_good_pkt_pointers(other_branch, src_reg, | |
2982 | src_reg->type, false); | |
2983 | } else { | |
2984 | return false; | |
2985 | } | |
2986 | break; | |
2987 | case BPF_JLE: | |
2988 | if ((dst_reg->type == PTR_TO_PACKET && | |
2989 | src_reg->type == PTR_TO_PACKET_END) || | |
2990 | (dst_reg->type == PTR_TO_PACKET_META && | |
2991 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
2992 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
2993 | find_good_pkt_pointers(other_branch, dst_reg, | |
2994 | dst_reg->type, false); | |
2995 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
2996 | src_reg->type == PTR_TO_PACKET) || | |
2997 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
2998 | src_reg->type == PTR_TO_PACKET_META)) { | |
2999 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
3000 | find_good_pkt_pointers(this_branch, src_reg, | |
3001 | src_reg->type, true); | |
3002 | } else { | |
3003 | return false; | |
3004 | } | |
3005 | break; | |
3006 | default: | |
3007 | return false; | |
3008 | } | |
3009 | ||
3010 | return true; | |
3011 | } | |
3012 | ||
58e2af8b | 3013 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
3014 | struct bpf_insn *insn, int *insn_idx) |
3015 | { | |
638f5b90 | 3016 | struct bpf_verifier_state *other_branch, *this_branch = env->cur_state; |
58e2af8b | 3017 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; |
17a52670 AS |
3018 | u8 opcode = BPF_OP(insn->code); |
3019 | int err; | |
3020 | ||
b4e432f1 | 3021 | if (opcode > BPF_JSLE) { |
61bd5218 | 3022 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
3023 | return -EINVAL; |
3024 | } | |
3025 | ||
3026 | if (BPF_SRC(insn->code) == BPF_X) { | |
3027 | if (insn->imm != 0) { | |
61bd5218 | 3028 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3029 | return -EINVAL; |
3030 | } | |
3031 | ||
3032 | /* check src1 operand */ | |
dc503a8a | 3033 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3034 | if (err) |
3035 | return err; | |
1be7f75d AS |
3036 | |
3037 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 3038 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
3039 | insn->src_reg); |
3040 | return -EACCES; | |
3041 | } | |
17a52670 AS |
3042 | } else { |
3043 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 3044 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3045 | return -EINVAL; |
3046 | } | |
3047 | } | |
3048 | ||
3049 | /* check src2 operand */ | |
dc503a8a | 3050 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3051 | if (err) |
3052 | return err; | |
3053 | ||
1a0dc1ac AS |
3054 | dst_reg = ®s[insn->dst_reg]; |
3055 | ||
17a52670 AS |
3056 | /* detect if R == 0 where R was initialized to zero earlier */ |
3057 | if (BPF_SRC(insn->code) == BPF_K && | |
3058 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
f1174f77 EC |
3059 | dst_reg->type == SCALAR_VALUE && |
3060 | tnum_equals_const(dst_reg->var_off, insn->imm)) { | |
17a52670 AS |
3061 | if (opcode == BPF_JEQ) { |
3062 | /* if (imm == imm) goto pc+off; | |
3063 | * only follow the goto, ignore fall-through | |
3064 | */ | |
3065 | *insn_idx += insn->off; | |
3066 | return 0; | |
3067 | } else { | |
3068 | /* if (imm != imm) goto pc+off; | |
3069 | * only follow fall-through branch, since | |
3070 | * that's where the program will go | |
3071 | */ | |
3072 | return 0; | |
3073 | } | |
3074 | } | |
3075 | ||
3076 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
3077 | if (!other_branch) | |
3078 | return -EFAULT; | |
3079 | ||
48461135 JB |
3080 | /* detect if we are comparing against a constant value so we can adjust |
3081 | * our min/max values for our dst register. | |
f1174f77 EC |
3082 | * this is only legit if both are scalars (or pointers to the same |
3083 | * object, I suppose, but we don't support that right now), because | |
3084 | * otherwise the different base pointers mean the offsets aren't | |
3085 | * comparable. | |
48461135 JB |
3086 | */ |
3087 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
3088 | if (dst_reg->type == SCALAR_VALUE && |
3089 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
3090 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
3091 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
3092 | dst_reg, regs[insn->src_reg].var_off.value, | |
3093 | opcode); | |
3094 | else if (tnum_is_const(dst_reg->var_off)) | |
3095 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
3096 | ®s[insn->src_reg], | |
3097 | dst_reg->var_off.value, opcode); | |
3098 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
3099 | /* Comparing for equality, we can combine knowledge */ | |
3100 | reg_combine_min_max(&other_branch->regs[insn->src_reg], | |
3101 | &other_branch->regs[insn->dst_reg], | |
3102 | ®s[insn->src_reg], | |
3103 | ®s[insn->dst_reg], opcode); | |
3104 | } | |
3105 | } else if (dst_reg->type == SCALAR_VALUE) { | |
48461135 JB |
3106 | reg_set_min_max(&other_branch->regs[insn->dst_reg], |
3107 | dst_reg, insn->imm, opcode); | |
3108 | } | |
3109 | ||
58e2af8b | 3110 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 3111 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
3112 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
3113 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
3114 | /* Mark all identical map registers in each branch as either |
3115 | * safe or unknown depending R == 0 or R != 0 conditional. | |
3116 | */ | |
f1174f77 EC |
3117 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
3118 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
5beca081 DB |
3119 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
3120 | this_branch, other_branch) && | |
3121 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
3122 | verbose(env, "R%d pointer comparison prohibited\n", |
3123 | insn->dst_reg); | |
1be7f75d | 3124 | return -EACCES; |
17a52670 | 3125 | } |
61bd5218 JK |
3126 | if (env->log.level) |
3127 | print_verifier_state(env, this_branch); | |
17a52670 AS |
3128 | return 0; |
3129 | } | |
3130 | ||
0246e64d AS |
3131 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
3132 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
3133 | { | |
3134 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
3135 | ||
3136 | return (struct bpf_map *) (unsigned long) imm64; | |
3137 | } | |
3138 | ||
17a52670 | 3139 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 3140 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3141 | { |
638f5b90 | 3142 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3143 | int err; |
3144 | ||
3145 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 3146 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
3147 | return -EINVAL; |
3148 | } | |
3149 | if (insn->off != 0) { | |
61bd5218 | 3150 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
3151 | return -EINVAL; |
3152 | } | |
3153 | ||
dc503a8a | 3154 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3155 | if (err) |
3156 | return err; | |
3157 | ||
6b173873 | 3158 | if (insn->src_reg == 0) { |
6b173873 JK |
3159 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
3160 | ||
f1174f77 | 3161 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 3162 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 3163 | return 0; |
6b173873 | 3164 | } |
17a52670 AS |
3165 | |
3166 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
3167 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
3168 | ||
3169 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
3170 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
3171 | return 0; | |
3172 | } | |
3173 | ||
96be4325 DB |
3174 | static bool may_access_skb(enum bpf_prog_type type) |
3175 | { | |
3176 | switch (type) { | |
3177 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
3178 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 3179 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
3180 | return true; |
3181 | default: | |
3182 | return false; | |
3183 | } | |
3184 | } | |
3185 | ||
ddd872bc AS |
3186 | /* verify safety of LD_ABS|LD_IND instructions: |
3187 | * - they can only appear in the programs where ctx == skb | |
3188 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
3189 | * preserve R6-R9, and store return value into R0 | |
3190 | * | |
3191 | * Implicit input: | |
3192 | * ctx == skb == R6 == CTX | |
3193 | * | |
3194 | * Explicit input: | |
3195 | * SRC == any register | |
3196 | * IMM == 32-bit immediate | |
3197 | * | |
3198 | * Output: | |
3199 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
3200 | */ | |
58e2af8b | 3201 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 3202 | { |
638f5b90 | 3203 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 3204 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
3205 | int i, err; |
3206 | ||
24701ece | 3207 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 3208 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
3209 | return -EINVAL; |
3210 | } | |
3211 | ||
3212 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 3213 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 3214 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 3215 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
3216 | return -EINVAL; |
3217 | } | |
3218 | ||
3219 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 3220 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
3221 | if (err) |
3222 | return err; | |
3223 | ||
3224 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
61bd5218 JK |
3225 | verbose(env, |
3226 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
3227 | return -EINVAL; |
3228 | } | |
3229 | ||
3230 | if (mode == BPF_IND) { | |
3231 | /* check explicit source operand */ | |
dc503a8a | 3232 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
3233 | if (err) |
3234 | return err; | |
3235 | } | |
3236 | ||
3237 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 3238 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3239 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3240 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3241 | } | |
ddd872bc AS |
3242 | |
3243 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
3244 | * the value fetched from the packet. |
3245 | * Already marked as written above. | |
ddd872bc | 3246 | */ |
61bd5218 | 3247 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
3248 | return 0; |
3249 | } | |
3250 | ||
390ee7e2 AS |
3251 | static int check_return_code(struct bpf_verifier_env *env) |
3252 | { | |
3253 | struct bpf_reg_state *reg; | |
3254 | struct tnum range = tnum_range(0, 1); | |
3255 | ||
3256 | switch (env->prog->type) { | |
3257 | case BPF_PROG_TYPE_CGROUP_SKB: | |
3258 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
3259 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 3260 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
390ee7e2 AS |
3261 | break; |
3262 | default: | |
3263 | return 0; | |
3264 | } | |
3265 | ||
638f5b90 | 3266 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 3267 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 3268 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
3269 | reg_type_str[reg->type]); |
3270 | return -EINVAL; | |
3271 | } | |
3272 | ||
3273 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 3274 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
3275 | if (!tnum_is_unknown(reg->var_off)) { |
3276 | char tn_buf[48]; | |
3277 | ||
3278 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3279 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 3280 | } else { |
61bd5218 | 3281 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 3282 | } |
61bd5218 | 3283 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
3284 | return -EINVAL; |
3285 | } | |
3286 | return 0; | |
3287 | } | |
3288 | ||
475fb78f AS |
3289 | /* non-recursive DFS pseudo code |
3290 | * 1 procedure DFS-iterative(G,v): | |
3291 | * 2 label v as discovered | |
3292 | * 3 let S be a stack | |
3293 | * 4 S.push(v) | |
3294 | * 5 while S is not empty | |
3295 | * 6 t <- S.pop() | |
3296 | * 7 if t is what we're looking for: | |
3297 | * 8 return t | |
3298 | * 9 for all edges e in G.adjacentEdges(t) do | |
3299 | * 10 if edge e is already labelled | |
3300 | * 11 continue with the next edge | |
3301 | * 12 w <- G.adjacentVertex(t,e) | |
3302 | * 13 if vertex w is not discovered and not explored | |
3303 | * 14 label e as tree-edge | |
3304 | * 15 label w as discovered | |
3305 | * 16 S.push(w) | |
3306 | * 17 continue at 5 | |
3307 | * 18 else if vertex w is discovered | |
3308 | * 19 label e as back-edge | |
3309 | * 20 else | |
3310 | * 21 // vertex w is explored | |
3311 | * 22 label e as forward- or cross-edge | |
3312 | * 23 label t as explored | |
3313 | * 24 S.pop() | |
3314 | * | |
3315 | * convention: | |
3316 | * 0x10 - discovered | |
3317 | * 0x11 - discovered and fall-through edge labelled | |
3318 | * 0x12 - discovered and fall-through and branch edges labelled | |
3319 | * 0x20 - explored | |
3320 | */ | |
3321 | ||
3322 | enum { | |
3323 | DISCOVERED = 0x10, | |
3324 | EXPLORED = 0x20, | |
3325 | FALLTHROUGH = 1, | |
3326 | BRANCH = 2, | |
3327 | }; | |
3328 | ||
58e2af8b | 3329 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 3330 | |
475fb78f AS |
3331 | static int *insn_stack; /* stack of insns to process */ |
3332 | static int cur_stack; /* current stack index */ | |
3333 | static int *insn_state; | |
3334 | ||
3335 | /* t, w, e - match pseudo-code above: | |
3336 | * t - index of current instruction | |
3337 | * w - next instruction | |
3338 | * e - edge | |
3339 | */ | |
58e2af8b | 3340 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
3341 | { |
3342 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
3343 | return 0; | |
3344 | ||
3345 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
3346 | return 0; | |
3347 | ||
3348 | if (w < 0 || w >= env->prog->len) { | |
61bd5218 | 3349 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
3350 | return -EINVAL; |
3351 | } | |
3352 | ||
f1bca824 AS |
3353 | if (e == BRANCH) |
3354 | /* mark branch target for state pruning */ | |
3355 | env->explored_states[w] = STATE_LIST_MARK; | |
3356 | ||
475fb78f AS |
3357 | if (insn_state[w] == 0) { |
3358 | /* tree-edge */ | |
3359 | insn_state[t] = DISCOVERED | e; | |
3360 | insn_state[w] = DISCOVERED; | |
3361 | if (cur_stack >= env->prog->len) | |
3362 | return -E2BIG; | |
3363 | insn_stack[cur_stack++] = w; | |
3364 | return 1; | |
3365 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
61bd5218 | 3366 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
3367 | return -EINVAL; |
3368 | } else if (insn_state[w] == EXPLORED) { | |
3369 | /* forward- or cross-edge */ | |
3370 | insn_state[t] = DISCOVERED | e; | |
3371 | } else { | |
61bd5218 | 3372 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
3373 | return -EFAULT; |
3374 | } | |
3375 | return 0; | |
3376 | } | |
3377 | ||
3378 | /* non-recursive depth-first-search to detect loops in BPF program | |
3379 | * loop == back-edge in directed graph | |
3380 | */ | |
58e2af8b | 3381 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
3382 | { |
3383 | struct bpf_insn *insns = env->prog->insnsi; | |
3384 | int insn_cnt = env->prog->len; | |
3385 | int ret = 0; | |
3386 | int i, t; | |
3387 | ||
3388 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3389 | if (!insn_state) | |
3390 | return -ENOMEM; | |
3391 | ||
3392 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
3393 | if (!insn_stack) { | |
3394 | kfree(insn_state); | |
3395 | return -ENOMEM; | |
3396 | } | |
3397 | ||
3398 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
3399 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
3400 | cur_stack = 1; | |
3401 | ||
3402 | peek_stack: | |
3403 | if (cur_stack == 0) | |
3404 | goto check_state; | |
3405 | t = insn_stack[cur_stack - 1]; | |
3406 | ||
3407 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
3408 | u8 opcode = BPF_OP(insns[t].code); | |
3409 | ||
3410 | if (opcode == BPF_EXIT) { | |
3411 | goto mark_explored; | |
3412 | } else if (opcode == BPF_CALL) { | |
3413 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3414 | if (ret == 1) | |
3415 | goto peek_stack; | |
3416 | else if (ret < 0) | |
3417 | goto err_free; | |
07016151 DB |
3418 | if (t + 1 < insn_cnt) |
3419 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3420 | } else if (opcode == BPF_JA) { |
3421 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
3422 | ret = -EINVAL; | |
3423 | goto err_free; | |
3424 | } | |
3425 | /* unconditional jump with single edge */ | |
3426 | ret = push_insn(t, t + insns[t].off + 1, | |
3427 | FALLTHROUGH, env); | |
3428 | if (ret == 1) | |
3429 | goto peek_stack; | |
3430 | else if (ret < 0) | |
3431 | goto err_free; | |
f1bca824 AS |
3432 | /* tell verifier to check for equivalent states |
3433 | * after every call and jump | |
3434 | */ | |
c3de6317 AS |
3435 | if (t + 1 < insn_cnt) |
3436 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
3437 | } else { |
3438 | /* conditional jump with two edges */ | |
3c2ce60b | 3439 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
3440 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
3441 | if (ret == 1) | |
3442 | goto peek_stack; | |
3443 | else if (ret < 0) | |
3444 | goto err_free; | |
3445 | ||
3446 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
3447 | if (ret == 1) | |
3448 | goto peek_stack; | |
3449 | else if (ret < 0) | |
3450 | goto err_free; | |
3451 | } | |
3452 | } else { | |
3453 | /* all other non-branch instructions with single | |
3454 | * fall-through edge | |
3455 | */ | |
3456 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
3457 | if (ret == 1) | |
3458 | goto peek_stack; | |
3459 | else if (ret < 0) | |
3460 | goto err_free; | |
3461 | } | |
3462 | ||
3463 | mark_explored: | |
3464 | insn_state[t] = EXPLORED; | |
3465 | if (cur_stack-- <= 0) { | |
61bd5218 | 3466 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
3467 | ret = -EFAULT; |
3468 | goto err_free; | |
3469 | } | |
3470 | goto peek_stack; | |
3471 | ||
3472 | check_state: | |
3473 | for (i = 0; i < insn_cnt; i++) { | |
3474 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 3475 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
3476 | ret = -EINVAL; |
3477 | goto err_free; | |
3478 | } | |
3479 | } | |
3480 | ret = 0; /* cfg looks good */ | |
3481 | ||
3482 | err_free: | |
3483 | kfree(insn_state); | |
3484 | kfree(insn_stack); | |
3485 | return ret; | |
3486 | } | |
3487 | ||
f1174f77 EC |
3488 | /* check %cur's range satisfies %old's */ |
3489 | static bool range_within(struct bpf_reg_state *old, | |
3490 | struct bpf_reg_state *cur) | |
3491 | { | |
b03c9f9f EC |
3492 | return old->umin_value <= cur->umin_value && |
3493 | old->umax_value >= cur->umax_value && | |
3494 | old->smin_value <= cur->smin_value && | |
3495 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
3496 | } |
3497 | ||
3498 | /* Maximum number of register states that can exist at once */ | |
3499 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
3500 | struct idpair { | |
3501 | u32 old; | |
3502 | u32 cur; | |
3503 | }; | |
3504 | ||
3505 | /* If in the old state two registers had the same id, then they need to have | |
3506 | * the same id in the new state as well. But that id could be different from | |
3507 | * the old state, so we need to track the mapping from old to new ids. | |
3508 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
3509 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
3510 | * regs with a different old id could still have new id 9, we don't care about | |
3511 | * that. | |
3512 | * So we look through our idmap to see if this old id has been seen before. If | |
3513 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 3514 | */ |
f1174f77 | 3515 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 3516 | { |
f1174f77 | 3517 | unsigned int i; |
969bf05e | 3518 | |
f1174f77 EC |
3519 | for (i = 0; i < ID_MAP_SIZE; i++) { |
3520 | if (!idmap[i].old) { | |
3521 | /* Reached an empty slot; haven't seen this id before */ | |
3522 | idmap[i].old = old_id; | |
3523 | idmap[i].cur = cur_id; | |
3524 | return true; | |
3525 | } | |
3526 | if (idmap[i].old == old_id) | |
3527 | return idmap[i].cur == cur_id; | |
3528 | } | |
3529 | /* We ran out of idmap slots, which should be impossible */ | |
3530 | WARN_ON_ONCE(1); | |
3531 | return false; | |
3532 | } | |
3533 | ||
3534 | /* Returns true if (rold safe implies rcur safe) */ | |
1b688a19 EC |
3535 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
3536 | struct idpair *idmap) | |
f1174f77 | 3537 | { |
dc503a8a EC |
3538 | if (!(rold->live & REG_LIVE_READ)) |
3539 | /* explored state didn't use this */ | |
3540 | return true; | |
3541 | ||
3542 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, live)) == 0) | |
969bf05e AS |
3543 | return true; |
3544 | ||
f1174f77 EC |
3545 | if (rold->type == NOT_INIT) |
3546 | /* explored state can't have used this */ | |
969bf05e | 3547 | return true; |
f1174f77 EC |
3548 | if (rcur->type == NOT_INIT) |
3549 | return false; | |
3550 | switch (rold->type) { | |
3551 | case SCALAR_VALUE: | |
3552 | if (rcur->type == SCALAR_VALUE) { | |
3553 | /* new val must satisfy old val knowledge */ | |
3554 | return range_within(rold, rcur) && | |
3555 | tnum_in(rold->var_off, rcur->var_off); | |
3556 | } else { | |
179d1c56 JH |
3557 | /* We're trying to use a pointer in place of a scalar. |
3558 | * Even if the scalar was unbounded, this could lead to | |
3559 | * pointer leaks because scalars are allowed to leak | |
3560 | * while pointers are not. We could make this safe in | |
3561 | * special cases if root is calling us, but it's | |
3562 | * probably not worth the hassle. | |
f1174f77 | 3563 | */ |
179d1c56 | 3564 | return false; |
f1174f77 EC |
3565 | } |
3566 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
3567 | /* If the new min/max/var_off satisfy the old ones and |
3568 | * everything else matches, we are OK. | |
3569 | * We don't care about the 'id' value, because nothing | |
3570 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
3571 | */ | |
3572 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
3573 | range_within(rold, rcur) && | |
3574 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
3575 | case PTR_TO_MAP_VALUE_OR_NULL: |
3576 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
3577 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
3578 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
3579 | * checked, doing so could have affected others with the same | |
3580 | * id, and we can't check for that because we lost the id when | |
3581 | * we converted to a PTR_TO_MAP_VALUE. | |
3582 | */ | |
3583 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
3584 | return false; | |
3585 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
3586 | return false; | |
3587 | /* Check our ids match any regs they're supposed to */ | |
3588 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 3589 | case PTR_TO_PACKET_META: |
f1174f77 | 3590 | case PTR_TO_PACKET: |
de8f3a83 | 3591 | if (rcur->type != rold->type) |
f1174f77 EC |
3592 | return false; |
3593 | /* We must have at least as much range as the old ptr | |
3594 | * did, so that any accesses which were safe before are | |
3595 | * still safe. This is true even if old range < old off, | |
3596 | * since someone could have accessed through (ptr - k), or | |
3597 | * even done ptr -= k in a register, to get a safe access. | |
3598 | */ | |
3599 | if (rold->range > rcur->range) | |
3600 | return false; | |
3601 | /* If the offsets don't match, we can't trust our alignment; | |
3602 | * nor can we be sure that we won't fall out of range. | |
3603 | */ | |
3604 | if (rold->off != rcur->off) | |
3605 | return false; | |
3606 | /* id relations must be preserved */ | |
3607 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
3608 | return false; | |
3609 | /* new val must satisfy old val knowledge */ | |
3610 | return range_within(rold, rcur) && | |
3611 | tnum_in(rold->var_off, rcur->var_off); | |
3612 | case PTR_TO_CTX: | |
3613 | case CONST_PTR_TO_MAP: | |
3614 | case PTR_TO_STACK: | |
3615 | case PTR_TO_PACKET_END: | |
3616 | /* Only valid matches are exact, which memcmp() above | |
3617 | * would have accepted | |
3618 | */ | |
3619 | default: | |
3620 | /* Don't know what's going on, just say it's not safe */ | |
3621 | return false; | |
3622 | } | |
969bf05e | 3623 | |
f1174f77 EC |
3624 | /* Shouldn't get here; if we do, say it's not safe */ |
3625 | WARN_ON_ONCE(1); | |
969bf05e AS |
3626 | return false; |
3627 | } | |
3628 | ||
638f5b90 AS |
3629 | static bool stacksafe(struct bpf_verifier_state *old, |
3630 | struct bpf_verifier_state *cur, | |
3631 | struct idpair *idmap) | |
3632 | { | |
3633 | int i, spi; | |
3634 | ||
3635 | /* if explored stack has more populated slots than current stack | |
3636 | * such stacks are not equivalent | |
3637 | */ | |
3638 | if (old->allocated_stack > cur->allocated_stack) | |
3639 | return false; | |
3640 | ||
3641 | /* walk slots of the explored stack and ignore any additional | |
3642 | * slots in the current stack, since explored(safe) state | |
3643 | * didn't use them | |
3644 | */ | |
3645 | for (i = 0; i < old->allocated_stack; i++) { | |
3646 | spi = i / BPF_REG_SIZE; | |
3647 | ||
3648 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) | |
3649 | continue; | |
3650 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != | |
3651 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
3652 | /* Ex: old explored (safe) state has STACK_SPILL in | |
3653 | * this stack slot, but current has has STACK_MISC -> | |
3654 | * this verifier states are not equivalent, | |
3655 | * return false to continue verification of this path | |
3656 | */ | |
3657 | return false; | |
3658 | if (i % BPF_REG_SIZE) | |
3659 | continue; | |
3660 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
3661 | continue; | |
3662 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
3663 | &cur->stack[spi].spilled_ptr, | |
3664 | idmap)) | |
3665 | /* when explored and current stack slot are both storing | |
3666 | * spilled registers, check that stored pointers types | |
3667 | * are the same as well. | |
3668 | * Ex: explored safe path could have stored | |
3669 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
3670 | * but current path has stored: | |
3671 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
3672 | * such verifier states are not equivalent. | |
3673 | * return false to continue verification of this path | |
3674 | */ | |
3675 | return false; | |
3676 | } | |
3677 | return true; | |
3678 | } | |
3679 | ||
f1bca824 AS |
3680 | /* compare two verifier states |
3681 | * | |
3682 | * all states stored in state_list are known to be valid, since | |
3683 | * verifier reached 'bpf_exit' instruction through them | |
3684 | * | |
3685 | * this function is called when verifier exploring different branches of | |
3686 | * execution popped from the state stack. If it sees an old state that has | |
3687 | * more strict register state and more strict stack state then this execution | |
3688 | * branch doesn't need to be explored further, since verifier already | |
3689 | * concluded that more strict state leads to valid finish. | |
3690 | * | |
3691 | * Therefore two states are equivalent if register state is more conservative | |
3692 | * and explored stack state is more conservative than the current one. | |
3693 | * Example: | |
3694 | * explored current | |
3695 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
3696 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
3697 | * | |
3698 | * In other words if current stack state (one being explored) has more | |
3699 | * valid slots than old one that already passed validation, it means | |
3700 | * the verifier can stop exploring and conclude that current state is valid too | |
3701 | * | |
3702 | * Similarly with registers. If explored state has register type as invalid | |
3703 | * whereas register type in current state is meaningful, it means that | |
3704 | * the current state will reach 'bpf_exit' instruction safely | |
3705 | */ | |
48461135 JB |
3706 | static bool states_equal(struct bpf_verifier_env *env, |
3707 | struct bpf_verifier_state *old, | |
58e2af8b | 3708 | struct bpf_verifier_state *cur) |
f1bca824 | 3709 | { |
f1174f77 EC |
3710 | struct idpair *idmap; |
3711 | bool ret = false; | |
f1bca824 AS |
3712 | int i; |
3713 | ||
f1174f77 EC |
3714 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
3715 | /* If we failed to allocate the idmap, just say it's not safe */ | |
3716 | if (!idmap) | |
1a0dc1ac | 3717 | return false; |
f1174f77 EC |
3718 | |
3719 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 3720 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 3721 | goto out_free; |
f1bca824 AS |
3722 | } |
3723 | ||
638f5b90 AS |
3724 | if (!stacksafe(old, cur, idmap)) |
3725 | goto out_free; | |
f1174f77 EC |
3726 | ret = true; |
3727 | out_free: | |
3728 | kfree(idmap); | |
3729 | return ret; | |
f1bca824 AS |
3730 | } |
3731 | ||
8e9cd9ce EC |
3732 | /* A write screens off any subsequent reads; but write marks come from the |
3733 | * straight-line code between a state and its parent. When we arrive at a | |
3734 | * jump target (in the first iteration of the propagate_liveness() loop), | |
3735 | * we didn't arrive by the straight-line code, so read marks in state must | |
3736 | * propagate to parent regardless of state's write marks. | |
3737 | */ | |
dc503a8a EC |
3738 | static bool do_propagate_liveness(const struct bpf_verifier_state *state, |
3739 | struct bpf_verifier_state *parent) | |
3740 | { | |
63f45f84 | 3741 | bool writes = parent == state->parent; /* Observe write marks */ |
dc503a8a EC |
3742 | bool touched = false; /* any changes made? */ |
3743 | int i; | |
3744 | ||
3745 | if (!parent) | |
3746 | return touched; | |
3747 | /* Propagate read liveness of registers... */ | |
3748 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
3749 | /* We don't need to worry about FP liveness because it's read-only */ | |
3750 | for (i = 0; i < BPF_REG_FP; i++) { | |
3751 | if (parent->regs[i].live & REG_LIVE_READ) | |
3752 | continue; | |
63f45f84 EC |
3753 | if (writes && (state->regs[i].live & REG_LIVE_WRITTEN)) |
3754 | continue; | |
3755 | if (state->regs[i].live & REG_LIVE_READ) { | |
dc503a8a EC |
3756 | parent->regs[i].live |= REG_LIVE_READ; |
3757 | touched = true; | |
3758 | } | |
3759 | } | |
3760 | /* ... and stack slots */ | |
638f5b90 AS |
3761 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
3762 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3763 | if (parent->stack[i].slot_type[0] != STACK_SPILL) | |
dc503a8a | 3764 | continue; |
638f5b90 | 3765 | if (state->stack[i].slot_type[0] != STACK_SPILL) |
dc503a8a | 3766 | continue; |
638f5b90 | 3767 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
dc503a8a | 3768 | continue; |
638f5b90 AS |
3769 | if (writes && |
3770 | (state->stack[i].spilled_ptr.live & REG_LIVE_WRITTEN)) | |
63f45f84 | 3771 | continue; |
638f5b90 AS |
3772 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) { |
3773 | parent->stack[i].spilled_ptr.live |= REG_LIVE_READ; | |
dc503a8a EC |
3774 | touched = true; |
3775 | } | |
3776 | } | |
3777 | return touched; | |
3778 | } | |
3779 | ||
8e9cd9ce EC |
3780 | /* "parent" is "a state from which we reach the current state", but initially |
3781 | * it is not the state->parent (i.e. "the state whose straight-line code leads | |
3782 | * to the current state"), instead it is the state that happened to arrive at | |
3783 | * a (prunable) equivalent of the current state. See comment above | |
3784 | * do_propagate_liveness() for consequences of this. | |
3785 | * This function is just a more efficient way of calling mark_reg_read() or | |
3786 | * mark_stack_slot_read() on each reg in "parent" that is read in "state", | |
3787 | * though it requires that parent != state->parent in the call arguments. | |
3788 | */ | |
dc503a8a EC |
3789 | static void propagate_liveness(const struct bpf_verifier_state *state, |
3790 | struct bpf_verifier_state *parent) | |
3791 | { | |
3792 | while (do_propagate_liveness(state, parent)) { | |
3793 | /* Something changed, so we need to feed those changes onward */ | |
3794 | state = parent; | |
3795 | parent = state->parent; | |
3796 | } | |
3797 | } | |
3798 | ||
58e2af8b | 3799 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 3800 | { |
58e2af8b JK |
3801 | struct bpf_verifier_state_list *new_sl; |
3802 | struct bpf_verifier_state_list *sl; | |
638f5b90 | 3803 | struct bpf_verifier_state *cur = env->cur_state; |
1969db47 | 3804 | int i, err; |
f1bca824 AS |
3805 | |
3806 | sl = env->explored_states[insn_idx]; | |
3807 | if (!sl) | |
3808 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
3809 | * be doing state search here | |
3810 | */ | |
3811 | return 0; | |
3812 | ||
3813 | while (sl != STATE_LIST_MARK) { | |
638f5b90 | 3814 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 3815 | /* reached equivalent register/stack state, |
dc503a8a EC |
3816 | * prune the search. |
3817 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
3818 | * If we have any write marks in env->cur_state, they |
3819 | * will prevent corresponding reads in the continuation | |
3820 | * from reaching our parent (an explored_state). Our | |
3821 | * own state will get the read marks recorded, but | |
3822 | * they'll be immediately forgotten as we're pruning | |
3823 | * this state and will pop a new one. | |
f1bca824 | 3824 | */ |
638f5b90 | 3825 | propagate_liveness(&sl->state, cur); |
f1bca824 | 3826 | return 1; |
dc503a8a | 3827 | } |
f1bca824 AS |
3828 | sl = sl->next; |
3829 | } | |
3830 | ||
3831 | /* there were no equivalent states, remember current one. | |
3832 | * technically the current state is not proven to be safe yet, | |
3833 | * but it will either reach bpf_exit (which means it's safe) or | |
3834 | * it will be rejected. Since there are no loops, we won't be | |
3835 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
3836 | */ | |
638f5b90 | 3837 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
3838 | if (!new_sl) |
3839 | return -ENOMEM; | |
3840 | ||
3841 | /* add new state to the head of linked list */ | |
1969db47 AS |
3842 | err = copy_verifier_state(&new_sl->state, cur); |
3843 | if (err) { | |
3844 | free_verifier_state(&new_sl->state, false); | |
3845 | kfree(new_sl); | |
3846 | return err; | |
3847 | } | |
f1bca824 AS |
3848 | new_sl->next = env->explored_states[insn_idx]; |
3849 | env->explored_states[insn_idx] = new_sl; | |
dc503a8a | 3850 | /* connect new state to parentage chain */ |
638f5b90 | 3851 | cur->parent = &new_sl->state; |
8e9cd9ce EC |
3852 | /* clear write marks in current state: the writes we did are not writes |
3853 | * our child did, so they don't screen off its reads from us. | |
3854 | * (There are no read marks in current state, because reads always mark | |
3855 | * their parent and current state never has children yet. Only | |
3856 | * explored_states can get read marks.) | |
3857 | */ | |
dc503a8a | 3858 | for (i = 0; i < BPF_REG_FP; i++) |
638f5b90 AS |
3859 | cur->regs[i].live = REG_LIVE_NONE; |
3860 | for (i = 0; i < cur->allocated_stack / BPF_REG_SIZE; i++) | |
3861 | if (cur->stack[i].slot_type[0] == STACK_SPILL) | |
3862 | cur->stack[i].spilled_ptr.live = REG_LIVE_NONE; | |
f1bca824 AS |
3863 | return 0; |
3864 | } | |
3865 | ||
13a27dfc JK |
3866 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
3867 | int insn_idx, int prev_insn_idx) | |
3868 | { | |
ab3f0063 JK |
3869 | if (env->dev_ops && env->dev_ops->insn_hook) |
3870 | return env->dev_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
13a27dfc | 3871 | |
ab3f0063 | 3872 | return 0; |
13a27dfc JK |
3873 | } |
3874 | ||
58e2af8b | 3875 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 3876 | { |
638f5b90 | 3877 | struct bpf_verifier_state *state; |
17a52670 | 3878 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 3879 | struct bpf_reg_state *regs; |
17a52670 AS |
3880 | int insn_cnt = env->prog->len; |
3881 | int insn_idx, prev_insn_idx = 0; | |
3882 | int insn_processed = 0; | |
3883 | bool do_print_state = false; | |
3884 | ||
638f5b90 AS |
3885 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
3886 | if (!state) | |
3887 | return -ENOMEM; | |
3888 | env->cur_state = state; | |
3889 | init_reg_state(env, state->regs); | |
dc503a8a | 3890 | state->parent = NULL; |
17a52670 AS |
3891 | insn_idx = 0; |
3892 | for (;;) { | |
3893 | struct bpf_insn *insn; | |
3894 | u8 class; | |
3895 | int err; | |
3896 | ||
3897 | if (insn_idx >= insn_cnt) { | |
61bd5218 | 3898 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
17a52670 AS |
3899 | insn_idx, insn_cnt); |
3900 | return -EFAULT; | |
3901 | } | |
3902 | ||
3903 | insn = &insns[insn_idx]; | |
3904 | class = BPF_CLASS(insn->code); | |
3905 | ||
07016151 | 3906 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
3907 | verbose(env, |
3908 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
3909 | insn_processed); |
3910 | return -E2BIG; | |
3911 | } | |
3912 | ||
f1bca824 AS |
3913 | err = is_state_visited(env, insn_idx); |
3914 | if (err < 0) | |
3915 | return err; | |
3916 | if (err == 1) { | |
3917 | /* found equivalent state, can prune the search */ | |
61bd5218 | 3918 | if (env->log.level) { |
f1bca824 | 3919 | if (do_print_state) |
61bd5218 | 3920 | verbose(env, "\nfrom %d to %d: safe\n", |
f1bca824 AS |
3921 | prev_insn_idx, insn_idx); |
3922 | else | |
61bd5218 | 3923 | verbose(env, "%d: safe\n", insn_idx); |
f1bca824 AS |
3924 | } |
3925 | goto process_bpf_exit; | |
3926 | } | |
3927 | ||
3c2ce60b DB |
3928 | if (need_resched()) |
3929 | cond_resched(); | |
3930 | ||
61bd5218 JK |
3931 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
3932 | if (env->log.level > 1) | |
3933 | verbose(env, "%d:", insn_idx); | |
c5fc9692 | 3934 | else |
61bd5218 | 3935 | verbose(env, "\nfrom %d to %d:", |
c5fc9692 | 3936 | prev_insn_idx, insn_idx); |
638f5b90 | 3937 | print_verifier_state(env, state); |
17a52670 AS |
3938 | do_print_state = false; |
3939 | } | |
3940 | ||
61bd5218 JK |
3941 | if (env->log.level) { |
3942 | verbose(env, "%d: ", insn_idx); | |
f4ac7e0b JK |
3943 | print_bpf_insn(verbose, env, insn, |
3944 | env->allow_ptr_leaks); | |
17a52670 AS |
3945 | } |
3946 | ||
13a27dfc JK |
3947 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
3948 | if (err) | |
3949 | return err; | |
3950 | ||
638f5b90 | 3951 | regs = cur_regs(env); |
c131187d | 3952 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 3953 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 3954 | err = check_alu_op(env, insn); |
17a52670 AS |
3955 | if (err) |
3956 | return err; | |
3957 | ||
3958 | } else if (class == BPF_LDX) { | |
3df126f3 | 3959 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
3960 | |
3961 | /* check for reserved fields is already done */ | |
3962 | ||
17a52670 | 3963 | /* check src operand */ |
dc503a8a | 3964 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3965 | if (err) |
3966 | return err; | |
3967 | ||
dc503a8a | 3968 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
3969 | if (err) |
3970 | return err; | |
3971 | ||
725f9dcd AS |
3972 | src_reg_type = regs[insn->src_reg].type; |
3973 | ||
17a52670 AS |
3974 | /* check that memory (src_reg + off) is readable, |
3975 | * the state of dst_reg will be updated by this func | |
3976 | */ | |
31fd8581 | 3977 | err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, |
17a52670 | 3978 | BPF_SIZE(insn->code), BPF_READ, |
57b4a36e | 3979 | insn->dst_reg, false); |
17a52670 AS |
3980 | if (err) |
3981 | return err; | |
3982 | ||
3df126f3 JK |
3983 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
3984 | ||
3985 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
3986 | /* saw a valid insn |
3987 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 3988 | * save type to validate intersecting paths |
9bac3d6d | 3989 | */ |
3df126f3 | 3990 | *prev_src_type = src_reg_type; |
9bac3d6d | 3991 | |
3df126f3 | 3992 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 3993 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 3994 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
3995 | /* ABuser program is trying to use the same insn |
3996 | * dst_reg = *(u32*) (src_reg + off) | |
3997 | * with different pointer types: | |
3998 | * src_reg == ctx in one branch and | |
3999 | * src_reg == stack|map in some other branch. | |
4000 | * Reject it. | |
4001 | */ | |
61bd5218 | 4002 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
4003 | return -EINVAL; |
4004 | } | |
4005 | ||
17a52670 | 4006 | } else if (class == BPF_STX) { |
3df126f3 | 4007 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 4008 | |
17a52670 | 4009 | if (BPF_MODE(insn->code) == BPF_XADD) { |
31fd8581 | 4010 | err = check_xadd(env, insn_idx, insn); |
17a52670 AS |
4011 | if (err) |
4012 | return err; | |
4013 | insn_idx++; | |
4014 | continue; | |
4015 | } | |
4016 | ||
17a52670 | 4017 | /* check src1 operand */ |
dc503a8a | 4018 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4019 | if (err) |
4020 | return err; | |
4021 | /* check src2 operand */ | |
dc503a8a | 4022 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4023 | if (err) |
4024 | return err; | |
4025 | ||
d691f9e8 AS |
4026 | dst_reg_type = regs[insn->dst_reg].type; |
4027 | ||
17a52670 | 4028 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 4029 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 | 4030 | BPF_SIZE(insn->code), BPF_WRITE, |
57b4a36e | 4031 | insn->src_reg, false); |
17a52670 AS |
4032 | if (err) |
4033 | return err; | |
4034 | ||
3df126f3 JK |
4035 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
4036 | ||
4037 | if (*prev_dst_type == NOT_INIT) { | |
4038 | *prev_dst_type = dst_reg_type; | |
4039 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 4040 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 4041 | *prev_dst_type == PTR_TO_CTX)) { |
61bd5218 | 4042 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
4043 | return -EINVAL; |
4044 | } | |
4045 | ||
17a52670 AS |
4046 | } else if (class == BPF_ST) { |
4047 | if (BPF_MODE(insn->code) != BPF_MEM || | |
4048 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 4049 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
4050 | return -EINVAL; |
4051 | } | |
4052 | /* check src operand */ | |
dc503a8a | 4053 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4054 | if (err) |
4055 | return err; | |
4056 | ||
f37a8cb8 DB |
4057 | if (is_ctx_reg(env, insn->dst_reg)) { |
4058 | verbose(env, "BPF_ST stores into R%d context is not allowed\n", | |
4059 | insn->dst_reg); | |
4060 | return -EACCES; | |
4061 | } | |
4062 | ||
17a52670 | 4063 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 4064 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 | 4065 | BPF_SIZE(insn->code), BPF_WRITE, |
57b4a36e | 4066 | -1, false); |
17a52670 AS |
4067 | if (err) |
4068 | return err; | |
4069 | ||
4070 | } else if (class == BPF_JMP) { | |
4071 | u8 opcode = BPF_OP(insn->code); | |
4072 | ||
4073 | if (opcode == BPF_CALL) { | |
4074 | if (BPF_SRC(insn->code) != BPF_K || | |
4075 | insn->off != 0 || | |
4076 | insn->src_reg != BPF_REG_0 || | |
4077 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4078 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
4079 | return -EINVAL; |
4080 | } | |
4081 | ||
81ed18ab | 4082 | err = check_call(env, insn->imm, insn_idx); |
17a52670 AS |
4083 | if (err) |
4084 | return err; | |
4085 | ||
4086 | } else if (opcode == BPF_JA) { | |
4087 | if (BPF_SRC(insn->code) != BPF_K || | |
4088 | insn->imm != 0 || | |
4089 | insn->src_reg != BPF_REG_0 || | |
4090 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4091 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
4092 | return -EINVAL; |
4093 | } | |
4094 | ||
4095 | insn_idx += insn->off + 1; | |
4096 | continue; | |
4097 | ||
4098 | } else if (opcode == BPF_EXIT) { | |
4099 | if (BPF_SRC(insn->code) != BPF_K || | |
4100 | insn->imm != 0 || | |
4101 | insn->src_reg != BPF_REG_0 || | |
4102 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4103 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
4104 | return -EINVAL; |
4105 | } | |
4106 | ||
4107 | /* eBPF calling convetion is such that R0 is used | |
4108 | * to return the value from eBPF program. | |
4109 | * Make sure that it's readable at this time | |
4110 | * of bpf_exit, which means that program wrote | |
4111 | * something into it earlier | |
4112 | */ | |
dc503a8a | 4113 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
4114 | if (err) |
4115 | return err; | |
4116 | ||
1be7f75d | 4117 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 4118 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
4119 | return -EACCES; |
4120 | } | |
4121 | ||
390ee7e2 AS |
4122 | err = check_return_code(env); |
4123 | if (err) | |
4124 | return err; | |
f1bca824 | 4125 | process_bpf_exit: |
638f5b90 AS |
4126 | err = pop_stack(env, &prev_insn_idx, &insn_idx); |
4127 | if (err < 0) { | |
4128 | if (err != -ENOENT) | |
4129 | return err; | |
17a52670 AS |
4130 | break; |
4131 | } else { | |
4132 | do_print_state = true; | |
4133 | continue; | |
4134 | } | |
4135 | } else { | |
4136 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
4137 | if (err) | |
4138 | return err; | |
4139 | } | |
4140 | } else if (class == BPF_LD) { | |
4141 | u8 mode = BPF_MODE(insn->code); | |
4142 | ||
4143 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
4144 | err = check_ld_abs(env, insn); |
4145 | if (err) | |
4146 | return err; | |
4147 | ||
17a52670 AS |
4148 | } else if (mode == BPF_IMM) { |
4149 | err = check_ld_imm(env, insn); | |
4150 | if (err) | |
4151 | return err; | |
4152 | ||
4153 | insn_idx++; | |
c131187d | 4154 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 4155 | } else { |
61bd5218 | 4156 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
4157 | return -EINVAL; |
4158 | } | |
4159 | } else { | |
61bd5218 | 4160 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
4161 | return -EINVAL; |
4162 | } | |
4163 | ||
4164 | insn_idx++; | |
4165 | } | |
4166 | ||
61bd5218 JK |
4167 | verbose(env, "processed %d insns, stack depth %d\n", insn_processed, |
4168 | env->prog->aux->stack_depth); | |
17a52670 AS |
4169 | return 0; |
4170 | } | |
4171 | ||
56f668df MKL |
4172 | static int check_map_prealloc(struct bpf_map *map) |
4173 | { | |
4174 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
4175 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
4176 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
4177 | !(map->map_flags & BPF_F_NO_PREALLOC); |
4178 | } | |
4179 | ||
61bd5218 JK |
4180 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
4181 | struct bpf_map *map, | |
fdc15d38 AS |
4182 | struct bpf_prog *prog) |
4183 | ||
4184 | { | |
56f668df MKL |
4185 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
4186 | * preallocated hash maps, since doing memory allocation | |
4187 | * in overflow_handler can crash depending on where nmi got | |
4188 | * triggered. | |
4189 | */ | |
4190 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
4191 | if (!check_map_prealloc(map)) { | |
61bd5218 | 4192 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
4193 | return -EINVAL; |
4194 | } | |
4195 | if (map->inner_map_meta && | |
4196 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 4197 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
4198 | return -EINVAL; |
4199 | } | |
fdc15d38 AS |
4200 | } |
4201 | return 0; | |
4202 | } | |
4203 | ||
0246e64d AS |
4204 | /* look for pseudo eBPF instructions that access map FDs and |
4205 | * replace them with actual map pointers | |
4206 | */ | |
58e2af8b | 4207 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
4208 | { |
4209 | struct bpf_insn *insn = env->prog->insnsi; | |
4210 | int insn_cnt = env->prog->len; | |
fdc15d38 | 4211 | int i, j, err; |
0246e64d | 4212 | |
f1f7714e | 4213 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
4214 | if (err) |
4215 | return err; | |
4216 | ||
0246e64d | 4217 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 4218 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 4219 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 4220 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
4221 | return -EINVAL; |
4222 | } | |
4223 | ||
d691f9e8 AS |
4224 | if (BPF_CLASS(insn->code) == BPF_STX && |
4225 | ((BPF_MODE(insn->code) != BPF_MEM && | |
4226 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 4227 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
4228 | return -EINVAL; |
4229 | } | |
4230 | ||
0246e64d AS |
4231 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
4232 | struct bpf_map *map; | |
4233 | struct fd f; | |
4234 | ||
4235 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
4236 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
4237 | insn[1].off != 0) { | |
61bd5218 | 4238 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
4239 | return -EINVAL; |
4240 | } | |
4241 | ||
4242 | if (insn->src_reg == 0) | |
4243 | /* valid generic load 64-bit imm */ | |
4244 | goto next_insn; | |
4245 | ||
4246 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
4247 | verbose(env, |
4248 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
4249 | return -EINVAL; |
4250 | } | |
4251 | ||
4252 | f = fdget(insn->imm); | |
c2101297 | 4253 | map = __bpf_map_get(f); |
0246e64d | 4254 | if (IS_ERR(map)) { |
61bd5218 | 4255 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 4256 | insn->imm); |
0246e64d AS |
4257 | return PTR_ERR(map); |
4258 | } | |
4259 | ||
61bd5218 | 4260 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
4261 | if (err) { |
4262 | fdput(f); | |
4263 | return err; | |
4264 | } | |
4265 | ||
0246e64d AS |
4266 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
4267 | insn[0].imm = (u32) (unsigned long) map; | |
4268 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
4269 | ||
4270 | /* check whether we recorded this map already */ | |
4271 | for (j = 0; j < env->used_map_cnt; j++) | |
4272 | if (env->used_maps[j] == map) { | |
4273 | fdput(f); | |
4274 | goto next_insn; | |
4275 | } | |
4276 | ||
4277 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
4278 | fdput(f); | |
4279 | return -E2BIG; | |
4280 | } | |
4281 | ||
0246e64d AS |
4282 | /* hold the map. If the program is rejected by verifier, |
4283 | * the map will be released by release_maps() or it | |
4284 | * will be used by the valid program until it's unloaded | |
4285 | * and all maps are released in free_bpf_prog_info() | |
4286 | */ | |
92117d84 AS |
4287 | map = bpf_map_inc(map, false); |
4288 | if (IS_ERR(map)) { | |
4289 | fdput(f); | |
4290 | return PTR_ERR(map); | |
4291 | } | |
4292 | env->used_maps[env->used_map_cnt++] = map; | |
4293 | ||
0246e64d AS |
4294 | fdput(f); |
4295 | next_insn: | |
4296 | insn++; | |
4297 | i++; | |
4298 | } | |
4299 | } | |
4300 | ||
4301 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
4302 | * 'struct bpf_map *' into a register instead of user map_fd. | |
4303 | * These pointers will be used later by verifier to validate map access. | |
4304 | */ | |
4305 | return 0; | |
4306 | } | |
4307 | ||
4308 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 4309 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
4310 | { |
4311 | int i; | |
4312 | ||
4313 | for (i = 0; i < env->used_map_cnt; i++) | |
4314 | bpf_map_put(env->used_maps[i]); | |
4315 | } | |
4316 | ||
4317 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 4318 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
4319 | { |
4320 | struct bpf_insn *insn = env->prog->insnsi; | |
4321 | int insn_cnt = env->prog->len; | |
4322 | int i; | |
4323 | ||
4324 | for (i = 0; i < insn_cnt; i++, insn++) | |
4325 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
4326 | insn->src_reg = 0; | |
4327 | } | |
4328 | ||
8041902d AS |
4329 | /* single env->prog->insni[off] instruction was replaced with the range |
4330 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
4331 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
4332 | */ | |
4333 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
4334 | u32 off, u32 cnt) | |
4335 | { | |
4336 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
c131187d | 4337 | int i; |
8041902d AS |
4338 | |
4339 | if (cnt == 1) | |
4340 | return 0; | |
4341 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
4342 | if (!new_data) | |
4343 | return -ENOMEM; | |
4344 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
4345 | memcpy(new_data + off + cnt - 1, old_data + off, | |
4346 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
c131187d AS |
4347 | for (i = off; i < off + cnt - 1; i++) |
4348 | new_data[i].seen = true; | |
8041902d AS |
4349 | env->insn_aux_data = new_data; |
4350 | vfree(old_data); | |
4351 | return 0; | |
4352 | } | |
4353 | ||
4354 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
4355 | const struct bpf_insn *patch, u32 len) | |
4356 | { | |
4357 | struct bpf_prog *new_prog; | |
4358 | ||
4359 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4360 | if (!new_prog) | |
4361 | return NULL; | |
4362 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
4363 | return NULL; | |
4364 | return new_prog; | |
4365 | } | |
4366 | ||
c131187d AS |
4367 | /* The verifier does more data flow analysis than llvm and will not explore |
4368 | * branches that are dead at run time. Malicious programs can have dead code | |
4369 | * too. Therefore replace all dead at-run-time code with nops. | |
4370 | */ | |
4371 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
4372 | { | |
4373 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
4374 | struct bpf_insn nop = BPF_MOV64_REG(BPF_REG_0, BPF_REG_0); | |
4375 | struct bpf_insn *insn = env->prog->insnsi; | |
4376 | const int insn_cnt = env->prog->len; | |
4377 | int i; | |
4378 | ||
4379 | for (i = 0; i < insn_cnt; i++) { | |
4380 | if (aux_data[i].seen) | |
4381 | continue; | |
4382 | memcpy(insn + i, &nop, sizeof(nop)); | |
4383 | } | |
4384 | } | |
4385 | ||
9bac3d6d AS |
4386 | /* convert load instructions that access fields of 'struct __sk_buff' |
4387 | * into sequence of instructions that access fields of 'struct sk_buff' | |
4388 | */ | |
58e2af8b | 4389 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 4390 | { |
00176a34 | 4391 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 4392 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 4393 | const int insn_cnt = env->prog->len; |
36bbef52 | 4394 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 4395 | struct bpf_prog *new_prog; |
d691f9e8 | 4396 | enum bpf_access_type type; |
f96da094 DB |
4397 | bool is_narrower_load; |
4398 | u32 target_size; | |
9bac3d6d | 4399 | |
36bbef52 DB |
4400 | if (ops->gen_prologue) { |
4401 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
4402 | env->prog); | |
4403 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4404 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
4405 | return -EINVAL; |
4406 | } else if (cnt) { | |
8041902d | 4407 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
4408 | if (!new_prog) |
4409 | return -ENOMEM; | |
8041902d | 4410 | |
36bbef52 | 4411 | env->prog = new_prog; |
3df126f3 | 4412 | delta += cnt - 1; |
36bbef52 DB |
4413 | } |
4414 | } | |
4415 | ||
4416 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
4417 | return 0; |
4418 | ||
3df126f3 | 4419 | insn = env->prog->insnsi + delta; |
36bbef52 | 4420 | |
9bac3d6d | 4421 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
4422 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
4423 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
4424 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4425 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 4426 | type = BPF_READ; |
62c7989b DB |
4427 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
4428 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
4429 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 4430 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
4431 | type = BPF_WRITE; |
4432 | else | |
9bac3d6d AS |
4433 | continue; |
4434 | ||
abd098e0 AS |
4435 | if (type == BPF_WRITE && |
4436 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
4437 | struct bpf_insn patch[] = { | |
4438 | /* Sanitize suspicious stack slot with zero. | |
4439 | * There are no memory dependencies for this store, | |
4440 | * since it's only using frame pointer and immediate | |
4441 | * constant of zero | |
4442 | */ | |
4443 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
4444 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
4445 | 0), | |
4446 | /* the original STX instruction will immediately | |
4447 | * overwrite the same stack slot with appropriate value | |
4448 | */ | |
4449 | *insn, | |
4450 | }; | |
4451 | ||
4452 | cnt = ARRAY_SIZE(patch); | |
4453 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
4454 | if (!new_prog) | |
4455 | return -ENOMEM; | |
4456 | ||
4457 | delta += cnt - 1; | |
4458 | env->prog = new_prog; | |
4459 | insn = new_prog->insnsi + i + delta; | |
4460 | continue; | |
4461 | } | |
4462 | ||
8041902d | 4463 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 4464 | continue; |
9bac3d6d | 4465 | |
31fd8581 | 4466 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 4467 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
4468 | |
4469 | /* If the read access is a narrower load of the field, | |
4470 | * convert to a 4/8-byte load, to minimum program type specific | |
4471 | * convert_ctx_access changes. If conversion is successful, | |
4472 | * we will apply proper mask to the result. | |
4473 | */ | |
f96da094 | 4474 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 4475 | if (is_narrower_load) { |
f96da094 DB |
4476 | u32 off = insn->off; |
4477 | u8 size_code; | |
4478 | ||
4479 | if (type == BPF_WRITE) { | |
61bd5218 | 4480 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
4481 | return -EINVAL; |
4482 | } | |
31fd8581 | 4483 | |
f96da094 | 4484 | size_code = BPF_H; |
31fd8581 YS |
4485 | if (ctx_field_size == 4) |
4486 | size_code = BPF_W; | |
4487 | else if (ctx_field_size == 8) | |
4488 | size_code = BPF_DW; | |
f96da094 | 4489 | |
31fd8581 YS |
4490 | insn->off = off & ~(ctx_field_size - 1); |
4491 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
4492 | } | |
f96da094 DB |
4493 | |
4494 | target_size = 0; | |
4495 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
4496 | &target_size); | |
4497 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
4498 | (ctx_field_size && !target_size)) { | |
61bd5218 | 4499 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
4500 | return -EINVAL; |
4501 | } | |
f96da094 DB |
4502 | |
4503 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
4504 | if (ctx_field_size <= 4) |
4505 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4506 | (1 << size * 8) - 1); |
31fd8581 YS |
4507 | else |
4508 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 4509 | (1 << size * 8) - 1); |
31fd8581 | 4510 | } |
9bac3d6d | 4511 | |
8041902d | 4512 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
4513 | if (!new_prog) |
4514 | return -ENOMEM; | |
4515 | ||
3df126f3 | 4516 | delta += cnt - 1; |
9bac3d6d AS |
4517 | |
4518 | /* keep walking new program and skip insns we just inserted */ | |
4519 | env->prog = new_prog; | |
3df126f3 | 4520 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
4521 | } |
4522 | ||
4523 | return 0; | |
4524 | } | |
4525 | ||
79741b3b | 4526 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 4527 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
4528 | * |
4529 | * this function is called after eBPF program passed verification | |
4530 | */ | |
79741b3b | 4531 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 4532 | { |
79741b3b AS |
4533 | struct bpf_prog *prog = env->prog; |
4534 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 4535 | const struct bpf_func_proto *fn; |
79741b3b | 4536 | const int insn_cnt = prog->len; |
81ed18ab AS |
4537 | struct bpf_insn insn_buf[16]; |
4538 | struct bpf_prog *new_prog; | |
4539 | struct bpf_map *map_ptr; | |
4540 | int i, cnt, delta = 0; | |
e245c5c6 | 4541 | |
79741b3b | 4542 | for (i = 0; i < insn_cnt; i++, insn++) { |
68fda450 AS |
4543 | if (insn->code == (BPF_ALU | BPF_MOD | BPF_X) || |
4544 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
4545 | /* due to JIT bugs clear upper 32-bits of src register | |
4546 | * before div/mod operation | |
4547 | */ | |
4548 | insn_buf[0] = BPF_MOV32_REG(insn->src_reg, insn->src_reg); | |
4549 | insn_buf[1] = *insn; | |
4550 | cnt = 2; | |
4551 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
4552 | if (!new_prog) | |
4553 | return -ENOMEM; | |
4554 | ||
4555 | delta += cnt - 1; | |
4556 | env->prog = prog = new_prog; | |
4557 | insn = new_prog->insnsi + i + delta; | |
4558 | continue; | |
4559 | } | |
4560 | ||
79741b3b AS |
4561 | if (insn->code != (BPF_JMP | BPF_CALL)) |
4562 | continue; | |
e245c5c6 | 4563 | |
79741b3b AS |
4564 | if (insn->imm == BPF_FUNC_get_route_realm) |
4565 | prog->dst_needed = 1; | |
4566 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
4567 | bpf_user_rnd_init_once(); | |
79741b3b | 4568 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
4569 | /* If we tail call into other programs, we |
4570 | * cannot make any assumptions since they can | |
4571 | * be replaced dynamically during runtime in | |
4572 | * the program array. | |
4573 | */ | |
4574 | prog->cb_access = 1; | |
80a58d02 | 4575 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 4576 | |
79741b3b AS |
4577 | /* mark bpf_tail_call as different opcode to avoid |
4578 | * conditional branch in the interpeter for every normal | |
4579 | * call and to prevent accidental JITing by JIT compiler | |
4580 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 4581 | */ |
79741b3b | 4582 | insn->imm = 0; |
71189fa9 | 4583 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 AS |
4584 | |
4585 | /* instead of changing every JIT dealing with tail_call | |
4586 | * emit two extra insns: | |
4587 | * if (index >= max_entries) goto out; | |
4588 | * index &= array->index_mask; | |
4589 | * to avoid out-of-bounds cpu speculation | |
4590 | */ | |
4591 | map_ptr = env->insn_aux_data[i + delta].map_ptr; | |
4592 | if (map_ptr == BPF_MAP_PTR_POISON) { | |
40950343 | 4593 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
4594 | return -EINVAL; |
4595 | } | |
4596 | if (!map_ptr->unpriv_array) | |
4597 | continue; | |
4598 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, | |
4599 | map_ptr->max_entries, 2); | |
4600 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
4601 | container_of(map_ptr, | |
4602 | struct bpf_array, | |
4603 | map)->index_mask); | |
4604 | insn_buf[2] = *insn; | |
4605 | cnt = 3; | |
4606 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
4607 | if (!new_prog) | |
4608 | return -ENOMEM; | |
4609 | ||
4610 | delta += cnt - 1; | |
4611 | env->prog = prog = new_prog; | |
4612 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
4613 | continue; |
4614 | } | |
e245c5c6 | 4615 | |
89c63074 DB |
4616 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
4617 | * handlers are currently limited to 64 bit only. | |
4618 | */ | |
4619 | if (ebpf_jit_enabled() && BITS_PER_LONG == 64 && | |
4620 | insn->imm == BPF_FUNC_map_lookup_elem) { | |
81ed18ab | 4621 | map_ptr = env->insn_aux_data[i + delta].map_ptr; |
fad73a1a MKL |
4622 | if (map_ptr == BPF_MAP_PTR_POISON || |
4623 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
4624 | goto patch_call_imm; |
4625 | ||
4626 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
4627 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 4628 | verbose(env, "bpf verifier is misconfigured\n"); |
81ed18ab AS |
4629 | return -EINVAL; |
4630 | } | |
4631 | ||
4632 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
4633 | cnt); | |
4634 | if (!new_prog) | |
4635 | return -ENOMEM; | |
4636 | ||
4637 | delta += cnt - 1; | |
4638 | ||
4639 | /* keep walking new program and skip insns we just inserted */ | |
4640 | env->prog = prog = new_prog; | |
4641 | insn = new_prog->insnsi + i + delta; | |
4642 | continue; | |
4643 | } | |
4644 | ||
109980b8 | 4645 | if (insn->imm == BPF_FUNC_redirect_map) { |
7c300131 DB |
4646 | /* Note, we cannot use prog directly as imm as subsequent |
4647 | * rewrites would still change the prog pointer. The only | |
4648 | * stable address we can use is aux, which also works with | |
4649 | * prog clones during blinding. | |
4650 | */ | |
4651 | u64 addr = (unsigned long)prog->aux; | |
109980b8 DB |
4652 | struct bpf_insn r4_ld[] = { |
4653 | BPF_LD_IMM64(BPF_REG_4, addr), | |
4654 | *insn, | |
4655 | }; | |
4656 | cnt = ARRAY_SIZE(r4_ld); | |
4657 | ||
4658 | new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt); | |
4659 | if (!new_prog) | |
4660 | return -ENOMEM; | |
4661 | ||
4662 | delta += cnt - 1; | |
4663 | env->prog = prog = new_prog; | |
4664 | insn = new_prog->insnsi + i + delta; | |
4665 | } | |
81ed18ab | 4666 | patch_call_imm: |
00176a34 | 4667 | fn = env->ops->get_func_proto(insn->imm); |
79741b3b AS |
4668 | /* all functions that have prototype and verifier allowed |
4669 | * programs to call them, must be real in-kernel functions | |
4670 | */ | |
4671 | if (!fn->func) { | |
61bd5218 JK |
4672 | verbose(env, |
4673 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
4674 | func_id_name(insn->imm), insn->imm); |
4675 | return -EFAULT; | |
e245c5c6 | 4676 | } |
79741b3b | 4677 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 4678 | } |
e245c5c6 | 4679 | |
79741b3b AS |
4680 | return 0; |
4681 | } | |
e245c5c6 | 4682 | |
58e2af8b | 4683 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 4684 | { |
58e2af8b | 4685 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
4686 | int i; |
4687 | ||
4688 | if (!env->explored_states) | |
4689 | return; | |
4690 | ||
4691 | for (i = 0; i < env->prog->len; i++) { | |
4692 | sl = env->explored_states[i]; | |
4693 | ||
4694 | if (sl) | |
4695 | while (sl != STATE_LIST_MARK) { | |
4696 | sln = sl->next; | |
1969db47 | 4697 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
4698 | kfree(sl); |
4699 | sl = sln; | |
4700 | } | |
4701 | } | |
4702 | ||
4703 | kfree(env->explored_states); | |
4704 | } | |
4705 | ||
9bac3d6d | 4706 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 4707 | { |
58e2af8b | 4708 | struct bpf_verifier_env *env; |
61bd5218 | 4709 | struct bpf_verifer_log *log; |
51580e79 AS |
4710 | int ret = -EINVAL; |
4711 | ||
eba0c929 AB |
4712 | /* no program is valid */ |
4713 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
4714 | return -EINVAL; | |
4715 | ||
58e2af8b | 4716 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
4717 | * allocate/free it every time bpf_check() is called |
4718 | */ | |
58e2af8b | 4719 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
4720 | if (!env) |
4721 | return -ENOMEM; | |
61bd5218 | 4722 | log = &env->log; |
cbd35700 | 4723 | |
3df126f3 JK |
4724 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
4725 | (*prog)->len); | |
4726 | ret = -ENOMEM; | |
4727 | if (!env->insn_aux_data) | |
4728 | goto err_free_env; | |
9bac3d6d | 4729 | env->prog = *prog; |
00176a34 | 4730 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 4731 | |
cbd35700 AS |
4732 | /* grab the mutex to protect few globals used by verifier */ |
4733 | mutex_lock(&bpf_verifier_lock); | |
4734 | ||
4735 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
4736 | /* user requested verbose verifier output | |
4737 | * and supplied buffer to store the verification trace | |
4738 | */ | |
e7bf8249 JK |
4739 | log->level = attr->log_level; |
4740 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
4741 | log->len_total = attr->log_size; | |
cbd35700 AS |
4742 | |
4743 | ret = -EINVAL; | |
e7bf8249 JK |
4744 | /* log attributes have to be sane */ |
4745 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
4746 | !log->level || !log->ubuf) | |
3df126f3 | 4747 | goto err_unlock; |
cbd35700 | 4748 | } |
1ad2f583 DB |
4749 | |
4750 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
4751 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 4752 | env->strict_alignment = true; |
cbd35700 | 4753 | |
ab3f0063 JK |
4754 | if (env->prog->aux->offload) { |
4755 | ret = bpf_prog_offload_verifier_prep(env); | |
4756 | if (ret) | |
4757 | goto err_unlock; | |
4758 | } | |
4759 | ||
0246e64d AS |
4760 | ret = replace_map_fd_with_map_ptr(env); |
4761 | if (ret < 0) | |
4762 | goto skip_full_check; | |
4763 | ||
9bac3d6d | 4764 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 4765 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
4766 | GFP_USER); |
4767 | ret = -ENOMEM; | |
4768 | if (!env->explored_states) | |
4769 | goto skip_full_check; | |
4770 | ||
475fb78f AS |
4771 | ret = check_cfg(env); |
4772 | if (ret < 0) | |
4773 | goto skip_full_check; | |
4774 | ||
1be7f75d AS |
4775 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
4776 | ||
17a52670 | 4777 | ret = do_check(env); |
8c01c4f8 CG |
4778 | if (env->cur_state) { |
4779 | free_verifier_state(env->cur_state, true); | |
4780 | env->cur_state = NULL; | |
4781 | } | |
cbd35700 | 4782 | |
0246e64d | 4783 | skip_full_check: |
638f5b90 | 4784 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 4785 | free_states(env); |
0246e64d | 4786 | |
c131187d AS |
4787 | if (ret == 0) |
4788 | sanitize_dead_code(env); | |
4789 | ||
9bac3d6d AS |
4790 | if (ret == 0) |
4791 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
4792 | ret = convert_ctx_accesses(env); | |
4793 | ||
e245c5c6 | 4794 | if (ret == 0) |
79741b3b | 4795 | ret = fixup_bpf_calls(env); |
e245c5c6 | 4796 | |
a2a7d570 | 4797 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 4798 | ret = -ENOSPC; |
a2a7d570 | 4799 | if (log->level && !log->ubuf) { |
cbd35700 | 4800 | ret = -EFAULT; |
a2a7d570 | 4801 | goto err_release_maps; |
cbd35700 AS |
4802 | } |
4803 | ||
0246e64d AS |
4804 | if (ret == 0 && env->used_map_cnt) { |
4805 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
4806 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
4807 | sizeof(env->used_maps[0]), | |
4808 | GFP_KERNEL); | |
0246e64d | 4809 | |
9bac3d6d | 4810 | if (!env->prog->aux->used_maps) { |
0246e64d | 4811 | ret = -ENOMEM; |
a2a7d570 | 4812 | goto err_release_maps; |
0246e64d AS |
4813 | } |
4814 | ||
9bac3d6d | 4815 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 4816 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 4817 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
4818 | |
4819 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
4820 | * bpf_ld_imm64 instructions | |
4821 | */ | |
4822 | convert_pseudo_ld_imm64(env); | |
4823 | } | |
cbd35700 | 4824 | |
a2a7d570 | 4825 | err_release_maps: |
9bac3d6d | 4826 | if (!env->prog->aux->used_maps) |
0246e64d AS |
4827 | /* if we didn't copy map pointers into bpf_prog_info, release |
4828 | * them now. Otherwise free_bpf_prog_info() will release them. | |
4829 | */ | |
4830 | release_maps(env); | |
9bac3d6d | 4831 | *prog = env->prog; |
3df126f3 | 4832 | err_unlock: |
cbd35700 | 4833 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
4834 | vfree(env->insn_aux_data); |
4835 | err_free_env: | |
4836 | kfree(env); | |
51580e79 AS |
4837 | return ret; |
4838 | } |