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> |
cc8b0b92 AS |
23 | #include <linux/bsearch.h> |
24 | #include <linux/sort.h> | |
c195651e | 25 | #include <linux/perf_event.h> |
51580e79 | 26 | |
f4ac7e0b JK |
27 | #include "disasm.h" |
28 | ||
00176a34 JK |
29 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
30 | #define BPF_PROG_TYPE(_id, _name) \ | |
31 | [_id] = & _name ## _verifier_ops, | |
32 | #define BPF_MAP_TYPE(_id, _ops) | |
33 | #include <linux/bpf_types.h> | |
34 | #undef BPF_PROG_TYPE | |
35 | #undef BPF_MAP_TYPE | |
36 | }; | |
37 | ||
51580e79 AS |
38 | /* bpf_check() is a static code analyzer that walks eBPF program |
39 | * instruction by instruction and updates register/stack state. | |
40 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
41 | * | |
42 | * The first pass is depth-first-search to check that the program is a DAG. | |
43 | * It rejects the following programs: | |
44 | * - larger than BPF_MAXINSNS insns | |
45 | * - if loop is present (detected via back-edge) | |
46 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
47 | * - out of bounds or malformed jumps | |
48 | * The second pass is all possible path descent from the 1st insn. | |
49 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 50 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
51 | * insn is less then 4K, but there are too many branches that change stack/regs. |
52 | * Number of 'branches to be analyzed' is limited to 1k | |
53 | * | |
54 | * On entry to each instruction, each register has a type, and the instruction | |
55 | * changes the types of the registers depending on instruction semantics. | |
56 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
57 | * copied to R1. | |
58 | * | |
59 | * All registers are 64-bit. | |
60 | * R0 - return register | |
61 | * R1-R5 argument passing registers | |
62 | * R6-R9 callee saved registers | |
63 | * R10 - frame pointer read-only | |
64 | * | |
65 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
66 | * and has type PTR_TO_CTX. | |
67 | * | |
68 | * Verifier tracks arithmetic operations on pointers in case: | |
69 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
70 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
71 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
72 | * and 2nd arithmetic instruction is pattern matched to recognize | |
73 | * that it wants to construct a pointer to some element within stack. | |
74 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
75 | * (and -20 constant is saved for further stack bounds checking). | |
76 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
77 | * | |
f1174f77 | 78 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 79 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 80 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
81 | * |
82 | * When verifier sees load or store instructions the type of base register | |
f1174f77 | 83 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK. These are three pointer |
51580e79 AS |
84 | * types recognized by check_mem_access() function. |
85 | * | |
86 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
87 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
88 | * | |
89 | * registers used to pass values to function calls are checked against | |
90 | * function argument constraints. | |
91 | * | |
92 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
93 | * It means that the register type passed to this function must be | |
94 | * PTR_TO_STACK and it will be used inside the function as | |
95 | * 'pointer to map element key' | |
96 | * | |
97 | * For example the argument constraints for bpf_map_lookup_elem(): | |
98 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
99 | * .arg1_type = ARG_CONST_MAP_PTR, | |
100 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
101 | * | |
102 | * ret_type says that this function returns 'pointer to map elem value or null' | |
103 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
104 | * 2nd argument should be a pointer to stack, which will be used inside | |
105 | * the helper function as a pointer to map element key. | |
106 | * | |
107 | * On the kernel side the helper function looks like: | |
108 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
109 | * { | |
110 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
111 | * void *key = (void *) (unsigned long) r2; | |
112 | * void *value; | |
113 | * | |
114 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
115 | * [key, key + map->key_size) bytes are valid and were initialized on | |
116 | * the stack of eBPF program. | |
117 | * } | |
118 | * | |
119 | * Corresponding eBPF program may look like: | |
120 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
121 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
122 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
123 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
124 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
125 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
126 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
127 | * | |
128 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
129 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
130 | * and were initialized prior to this call. | |
131 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
132 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
133 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
134 | * returns ether pointer to map value or NULL. | |
135 | * | |
136 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
137 | * insn, the register holding that pointer in the true branch changes state to | |
138 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
139 | * branch. See check_cond_jmp_op(). | |
140 | * | |
141 | * After the call R0 is set to return type of the function and registers R1-R5 | |
142 | * are set to NOT_INIT to indicate that they are no longer readable. | |
143 | */ | |
144 | ||
17a52670 | 145 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 146 | struct bpf_verifier_stack_elem { |
17a52670 AS |
147 | /* verifer state is 'st' |
148 | * before processing instruction 'insn_idx' | |
149 | * and after processing instruction 'prev_insn_idx' | |
150 | */ | |
58e2af8b | 151 | struct bpf_verifier_state st; |
17a52670 AS |
152 | int insn_idx; |
153 | int prev_insn_idx; | |
58e2af8b | 154 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
155 | }; |
156 | ||
8e17c1b1 | 157 | #define BPF_COMPLEXITY_LIMIT_INSNS 131072 |
07016151 DB |
158 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 |
159 | ||
fad73a1a MKL |
160 | #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA) |
161 | ||
33ff9823 DB |
162 | struct bpf_call_arg_meta { |
163 | struct bpf_map *map_ptr; | |
435faee1 | 164 | bool raw_mode; |
36bbef52 | 165 | bool pkt_access; |
435faee1 DB |
166 | int regno; |
167 | int access_size; | |
849fa506 YS |
168 | s64 msize_smax_value; |
169 | u64 msize_umax_value; | |
33ff9823 DB |
170 | }; |
171 | ||
cbd35700 AS |
172 | static DEFINE_MUTEX(bpf_verifier_lock); |
173 | ||
77d2e05a MKL |
174 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
175 | va_list args) | |
cbd35700 | 176 | { |
a2a7d570 | 177 | unsigned int n; |
cbd35700 | 178 | |
a2a7d570 | 179 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
180 | |
181 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
182 | "verifier log line truncated - local buffer too short\n"); | |
183 | ||
184 | n = min(log->len_total - log->len_used - 1, n); | |
185 | log->kbuf[n] = '\0'; | |
186 | ||
187 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) | |
188 | log->len_used += n; | |
189 | else | |
190 | log->ubuf = NULL; | |
cbd35700 | 191 | } |
abe08840 JO |
192 | |
193 | /* log_level controls verbosity level of eBPF verifier. | |
194 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
195 | * so the user can figure out what's wrong with the program | |
430e68d1 | 196 | */ |
abe08840 JO |
197 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
198 | const char *fmt, ...) | |
199 | { | |
200 | va_list args; | |
201 | ||
77d2e05a MKL |
202 | if (!bpf_verifier_log_needed(&env->log)) |
203 | return; | |
204 | ||
abe08840 | 205 | va_start(args, fmt); |
77d2e05a | 206 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
207 | va_end(args); |
208 | } | |
209 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
210 | ||
211 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
212 | { | |
77d2e05a | 213 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
214 | va_list args; |
215 | ||
77d2e05a MKL |
216 | if (!bpf_verifier_log_needed(&env->log)) |
217 | return; | |
218 | ||
abe08840 | 219 | va_start(args, fmt); |
77d2e05a | 220 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
221 | va_end(args); |
222 | } | |
cbd35700 | 223 | |
de8f3a83 DB |
224 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
225 | { | |
226 | return type == PTR_TO_PACKET || | |
227 | type == PTR_TO_PACKET_META; | |
228 | } | |
229 | ||
17a52670 AS |
230 | /* string representation of 'enum bpf_reg_type' */ |
231 | static const char * const reg_type_str[] = { | |
232 | [NOT_INIT] = "?", | |
f1174f77 | 233 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
234 | [PTR_TO_CTX] = "ctx", |
235 | [CONST_PTR_TO_MAP] = "map_ptr", | |
236 | [PTR_TO_MAP_VALUE] = "map_value", | |
237 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 238 | [PTR_TO_STACK] = "fp", |
969bf05e | 239 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 240 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 241 | [PTR_TO_PACKET_END] = "pkt_end", |
17a52670 AS |
242 | }; |
243 | ||
4e92024a AS |
244 | static void print_liveness(struct bpf_verifier_env *env, |
245 | enum bpf_reg_liveness live) | |
246 | { | |
247 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN)) | |
248 | verbose(env, "_"); | |
249 | if (live & REG_LIVE_READ) | |
250 | verbose(env, "r"); | |
251 | if (live & REG_LIVE_WRITTEN) | |
252 | verbose(env, "w"); | |
253 | } | |
254 | ||
f4d7e40a AS |
255 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
256 | const struct bpf_reg_state *reg) | |
257 | { | |
258 | struct bpf_verifier_state *cur = env->cur_state; | |
259 | ||
260 | return cur->frame[reg->frameno]; | |
261 | } | |
262 | ||
61bd5218 | 263 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 264 | const struct bpf_func_state *state) |
17a52670 | 265 | { |
f4d7e40a | 266 | const struct bpf_reg_state *reg; |
17a52670 AS |
267 | enum bpf_reg_type t; |
268 | int i; | |
269 | ||
f4d7e40a AS |
270 | if (state->frameno) |
271 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 272 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
273 | reg = &state->regs[i]; |
274 | t = reg->type; | |
17a52670 AS |
275 | if (t == NOT_INIT) |
276 | continue; | |
4e92024a AS |
277 | verbose(env, " R%d", i); |
278 | print_liveness(env, reg->live); | |
279 | verbose(env, "=%s", reg_type_str[t]); | |
f1174f77 EC |
280 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
281 | tnum_is_const(reg->var_off)) { | |
282 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 283 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f4d7e40a AS |
284 | if (t == PTR_TO_STACK) |
285 | verbose(env, ",call_%d", func(env, reg)->callsite); | |
f1174f77 | 286 | } else { |
61bd5218 | 287 | verbose(env, "(id=%d", reg->id); |
f1174f77 | 288 | if (t != SCALAR_VALUE) |
61bd5218 | 289 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 290 | if (type_is_pkt_pointer(t)) |
61bd5218 | 291 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
292 | else if (t == CONST_PTR_TO_MAP || |
293 | t == PTR_TO_MAP_VALUE || | |
294 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 295 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
296 | reg->map_ptr->key_size, |
297 | reg->map_ptr->value_size); | |
7d1238f2 EC |
298 | if (tnum_is_const(reg->var_off)) { |
299 | /* Typically an immediate SCALAR_VALUE, but | |
300 | * could be a pointer whose offset is too big | |
301 | * for reg->off | |
302 | */ | |
61bd5218 | 303 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
304 | } else { |
305 | if (reg->smin_value != reg->umin_value && | |
306 | reg->smin_value != S64_MIN) | |
61bd5218 | 307 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
308 | (long long)reg->smin_value); |
309 | if (reg->smax_value != reg->umax_value && | |
310 | reg->smax_value != S64_MAX) | |
61bd5218 | 311 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
312 | (long long)reg->smax_value); |
313 | if (reg->umin_value != 0) | |
61bd5218 | 314 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
315 | (unsigned long long)reg->umin_value); |
316 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 317 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
318 | (unsigned long long)reg->umax_value); |
319 | if (!tnum_is_unknown(reg->var_off)) { | |
320 | char tn_buf[48]; | |
f1174f77 | 321 | |
7d1238f2 | 322 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 323 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 324 | } |
f1174f77 | 325 | } |
61bd5218 | 326 | verbose(env, ")"); |
f1174f77 | 327 | } |
17a52670 | 328 | } |
638f5b90 | 329 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
4e92024a AS |
330 | if (state->stack[i].slot_type[0] == STACK_SPILL) { |
331 | verbose(env, " fp%d", | |
332 | (-i - 1) * BPF_REG_SIZE); | |
333 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
334 | verbose(env, "=%s", | |
638f5b90 | 335 | reg_type_str[state->stack[i].spilled_ptr.type]); |
4e92024a | 336 | } |
cc2b14d5 AS |
337 | if (state->stack[i].slot_type[0] == STACK_ZERO) |
338 | verbose(env, " fp%d=0", (-i - 1) * BPF_REG_SIZE); | |
17a52670 | 339 | } |
61bd5218 | 340 | verbose(env, "\n"); |
17a52670 AS |
341 | } |
342 | ||
f4d7e40a AS |
343 | static int copy_stack_state(struct bpf_func_state *dst, |
344 | const struct bpf_func_state *src) | |
17a52670 | 345 | { |
638f5b90 AS |
346 | if (!src->stack) |
347 | return 0; | |
348 | if (WARN_ON_ONCE(dst->allocated_stack < src->allocated_stack)) { | |
349 | /* internal bug, make state invalid to reject the program */ | |
350 | memset(dst, 0, sizeof(*dst)); | |
351 | return -EFAULT; | |
352 | } | |
353 | memcpy(dst->stack, src->stack, | |
354 | sizeof(*src->stack) * (src->allocated_stack / BPF_REG_SIZE)); | |
355 | return 0; | |
356 | } | |
357 | ||
358 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
359 | * make it consume minimal amount of memory. check_stack_write() access from | |
f4d7e40a | 360 | * the program calls into realloc_func_state() to grow the stack size. |
638f5b90 AS |
361 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state |
362 | * which this function copies over. It points to previous bpf_verifier_state | |
363 | * which is never reallocated | |
364 | */ | |
f4d7e40a AS |
365 | static int realloc_func_state(struct bpf_func_state *state, int size, |
366 | bool copy_old) | |
638f5b90 AS |
367 | { |
368 | u32 old_size = state->allocated_stack; | |
369 | struct bpf_stack_state *new_stack; | |
370 | int slot = size / BPF_REG_SIZE; | |
371 | ||
372 | if (size <= old_size || !size) { | |
373 | if (copy_old) | |
374 | return 0; | |
375 | state->allocated_stack = slot * BPF_REG_SIZE; | |
376 | if (!size && old_size) { | |
377 | kfree(state->stack); | |
378 | state->stack = NULL; | |
379 | } | |
380 | return 0; | |
381 | } | |
382 | new_stack = kmalloc_array(slot, sizeof(struct bpf_stack_state), | |
383 | GFP_KERNEL); | |
384 | if (!new_stack) | |
385 | return -ENOMEM; | |
386 | if (copy_old) { | |
387 | if (state->stack) | |
388 | memcpy(new_stack, state->stack, | |
389 | sizeof(*new_stack) * (old_size / BPF_REG_SIZE)); | |
390 | memset(new_stack + old_size / BPF_REG_SIZE, 0, | |
391 | sizeof(*new_stack) * (size - old_size) / BPF_REG_SIZE); | |
392 | } | |
393 | state->allocated_stack = slot * BPF_REG_SIZE; | |
394 | kfree(state->stack); | |
395 | state->stack = new_stack; | |
396 | return 0; | |
397 | } | |
398 | ||
f4d7e40a AS |
399 | static void free_func_state(struct bpf_func_state *state) |
400 | { | |
5896351e AS |
401 | if (!state) |
402 | return; | |
f4d7e40a AS |
403 | kfree(state->stack); |
404 | kfree(state); | |
405 | } | |
406 | ||
1969db47 AS |
407 | static void free_verifier_state(struct bpf_verifier_state *state, |
408 | bool free_self) | |
638f5b90 | 409 | { |
f4d7e40a AS |
410 | int i; |
411 | ||
412 | for (i = 0; i <= state->curframe; i++) { | |
413 | free_func_state(state->frame[i]); | |
414 | state->frame[i] = NULL; | |
415 | } | |
1969db47 AS |
416 | if (free_self) |
417 | kfree(state); | |
638f5b90 AS |
418 | } |
419 | ||
420 | /* copy verifier state from src to dst growing dst stack space | |
421 | * when necessary to accommodate larger src stack | |
422 | */ | |
f4d7e40a AS |
423 | static int copy_func_state(struct bpf_func_state *dst, |
424 | const struct bpf_func_state *src) | |
638f5b90 AS |
425 | { |
426 | int err; | |
427 | ||
f4d7e40a | 428 | err = realloc_func_state(dst, src->allocated_stack, false); |
638f5b90 AS |
429 | if (err) |
430 | return err; | |
f4d7e40a | 431 | memcpy(dst, src, offsetof(struct bpf_func_state, allocated_stack)); |
638f5b90 AS |
432 | return copy_stack_state(dst, src); |
433 | } | |
434 | ||
f4d7e40a AS |
435 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
436 | const struct bpf_verifier_state *src) | |
437 | { | |
438 | struct bpf_func_state *dst; | |
439 | int i, err; | |
440 | ||
441 | /* if dst has more stack frames then src frame, free them */ | |
442 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
443 | free_func_state(dst_state->frame[i]); | |
444 | dst_state->frame[i] = NULL; | |
445 | } | |
446 | dst_state->curframe = src->curframe; | |
447 | dst_state->parent = src->parent; | |
448 | for (i = 0; i <= src->curframe; i++) { | |
449 | dst = dst_state->frame[i]; | |
450 | if (!dst) { | |
451 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
452 | if (!dst) | |
453 | return -ENOMEM; | |
454 | dst_state->frame[i] = dst; | |
455 | } | |
456 | err = copy_func_state(dst, src->frame[i]); | |
457 | if (err) | |
458 | return err; | |
459 | } | |
460 | return 0; | |
461 | } | |
462 | ||
638f5b90 AS |
463 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
464 | int *insn_idx) | |
465 | { | |
466 | struct bpf_verifier_state *cur = env->cur_state; | |
467 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
468 | int err; | |
17a52670 AS |
469 | |
470 | if (env->head == NULL) | |
638f5b90 | 471 | return -ENOENT; |
17a52670 | 472 | |
638f5b90 AS |
473 | if (cur) { |
474 | err = copy_verifier_state(cur, &head->st); | |
475 | if (err) | |
476 | return err; | |
477 | } | |
478 | if (insn_idx) | |
479 | *insn_idx = head->insn_idx; | |
17a52670 | 480 | if (prev_insn_idx) |
638f5b90 AS |
481 | *prev_insn_idx = head->prev_insn_idx; |
482 | elem = head->next; | |
1969db47 | 483 | free_verifier_state(&head->st, false); |
638f5b90 | 484 | kfree(head); |
17a52670 AS |
485 | env->head = elem; |
486 | env->stack_size--; | |
638f5b90 | 487 | return 0; |
17a52670 AS |
488 | } |
489 | ||
58e2af8b JK |
490 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
491 | int insn_idx, int prev_insn_idx) | |
17a52670 | 492 | { |
638f5b90 | 493 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 494 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 495 | int err; |
17a52670 | 496 | |
638f5b90 | 497 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
498 | if (!elem) |
499 | goto err; | |
500 | ||
17a52670 AS |
501 | elem->insn_idx = insn_idx; |
502 | elem->prev_insn_idx = prev_insn_idx; | |
503 | elem->next = env->head; | |
504 | env->head = elem; | |
505 | env->stack_size++; | |
1969db47 AS |
506 | err = copy_verifier_state(&elem->st, cur); |
507 | if (err) | |
508 | goto err; | |
07016151 | 509 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
61bd5218 | 510 | verbose(env, "BPF program is too complex\n"); |
17a52670 AS |
511 | goto err; |
512 | } | |
513 | return &elem->st; | |
514 | err: | |
5896351e AS |
515 | free_verifier_state(env->cur_state, true); |
516 | env->cur_state = NULL; | |
17a52670 | 517 | /* pop all elements and return */ |
638f5b90 | 518 | while (!pop_stack(env, NULL, NULL)); |
17a52670 AS |
519 | return NULL; |
520 | } | |
521 | ||
522 | #define CALLER_SAVED_REGS 6 | |
523 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
524 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
525 | }; | |
526 | ||
f1174f77 EC |
527 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
528 | ||
b03c9f9f EC |
529 | /* Mark the unknown part of a register (variable offset or scalar value) as |
530 | * known to have the value @imm. | |
531 | */ | |
532 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
533 | { | |
534 | reg->id = 0; | |
535 | reg->var_off = tnum_const(imm); | |
536 | reg->smin_value = (s64)imm; | |
537 | reg->smax_value = (s64)imm; | |
538 | reg->umin_value = imm; | |
539 | reg->umax_value = imm; | |
540 | } | |
541 | ||
f1174f77 EC |
542 | /* Mark the 'variable offset' part of a register as zero. This should be |
543 | * used only on registers holding a pointer type. | |
544 | */ | |
545 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 546 | { |
b03c9f9f | 547 | __mark_reg_known(reg, 0); |
f1174f77 | 548 | } |
a9789ef9 | 549 | |
cc2b14d5 AS |
550 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
551 | { | |
552 | __mark_reg_known(reg, 0); | |
553 | reg->off = 0; | |
554 | reg->type = SCALAR_VALUE; | |
555 | } | |
556 | ||
61bd5218 JK |
557 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
558 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
559 | { |
560 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 561 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
562 | /* Something bad happened, let's kill all regs */ |
563 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
564 | __mark_reg_not_init(regs + regno); | |
565 | return; | |
566 | } | |
567 | __mark_reg_known_zero(regs + regno); | |
568 | } | |
569 | ||
de8f3a83 DB |
570 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
571 | { | |
572 | return type_is_pkt_pointer(reg->type); | |
573 | } | |
574 | ||
575 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
576 | { | |
577 | return reg_is_pkt_pointer(reg) || | |
578 | reg->type == PTR_TO_PACKET_END; | |
579 | } | |
580 | ||
581 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
582 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
583 | enum bpf_reg_type which) | |
584 | { | |
585 | /* The register can already have a range from prior markings. | |
586 | * This is fine as long as it hasn't been advanced from its | |
587 | * origin. | |
588 | */ | |
589 | return reg->type == which && | |
590 | reg->id == 0 && | |
591 | reg->off == 0 && | |
592 | tnum_equals_const(reg->var_off, 0); | |
593 | } | |
594 | ||
b03c9f9f EC |
595 | /* Attempts to improve min/max values based on var_off information */ |
596 | static void __update_reg_bounds(struct bpf_reg_state *reg) | |
597 | { | |
598 | /* min signed is max(sign bit) | min(other bits) */ | |
599 | reg->smin_value = max_t(s64, reg->smin_value, | |
600 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
601 | /* max signed is min(sign bit) | max(other bits) */ | |
602 | reg->smax_value = min_t(s64, reg->smax_value, | |
603 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
604 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
605 | reg->umax_value = min(reg->umax_value, | |
606 | reg->var_off.value | reg->var_off.mask); | |
607 | } | |
608 | ||
609 | /* Uses signed min/max values to inform unsigned, and vice-versa */ | |
610 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) | |
611 | { | |
612 | /* Learn sign from signed bounds. | |
613 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
614 | * are the same, so combine. This works even in the negative case, e.g. | |
615 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
616 | */ | |
617 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
618 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
619 | reg->umin_value); | |
620 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
621 | reg->umax_value); | |
622 | return; | |
623 | } | |
624 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
625 | * boundary, so we must be careful. | |
626 | */ | |
627 | if ((s64)reg->umax_value >= 0) { | |
628 | /* Positive. We can't learn anything from the smin, but smax | |
629 | * is positive, hence safe. | |
630 | */ | |
631 | reg->smin_value = reg->umin_value; | |
632 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
633 | reg->umax_value); | |
634 | } else if ((s64)reg->umin_value < 0) { | |
635 | /* Negative. We can't learn anything from the smax, but smin | |
636 | * is negative, hence safe. | |
637 | */ | |
638 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
639 | reg->umin_value); | |
640 | reg->smax_value = reg->umax_value; | |
641 | } | |
642 | } | |
643 | ||
644 | /* Attempts to improve var_off based on unsigned min/max information */ | |
645 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
646 | { | |
647 | reg->var_off = tnum_intersect(reg->var_off, | |
648 | tnum_range(reg->umin_value, | |
649 | reg->umax_value)); | |
650 | } | |
651 | ||
652 | /* Reset the min/max bounds of a register */ | |
653 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
654 | { | |
655 | reg->smin_value = S64_MIN; | |
656 | reg->smax_value = S64_MAX; | |
657 | reg->umin_value = 0; | |
658 | reg->umax_value = U64_MAX; | |
659 | } | |
660 | ||
f1174f77 EC |
661 | /* Mark a register as having a completely unknown (scalar) value. */ |
662 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
663 | { | |
664 | reg->type = SCALAR_VALUE; | |
665 | reg->id = 0; | |
666 | reg->off = 0; | |
667 | reg->var_off = tnum_unknown; | |
f4d7e40a | 668 | reg->frameno = 0; |
b03c9f9f | 669 | __mark_reg_unbounded(reg); |
f1174f77 EC |
670 | } |
671 | ||
61bd5218 JK |
672 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
673 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
674 | { |
675 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 676 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
677 | /* Something bad happened, let's kill all regs except FP */ |
678 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
679 | __mark_reg_not_init(regs + regno); |
680 | return; | |
681 | } | |
682 | __mark_reg_unknown(regs + regno); | |
683 | } | |
684 | ||
685 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
686 | { | |
687 | __mark_reg_unknown(reg); | |
688 | reg->type = NOT_INIT; | |
689 | } | |
690 | ||
61bd5218 JK |
691 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
692 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
693 | { |
694 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 695 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
696 | /* Something bad happened, let's kill all regs except FP */ |
697 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
698 | __mark_reg_not_init(regs + regno); |
699 | return; | |
700 | } | |
701 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
702 | } |
703 | ||
61bd5218 | 704 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 705 | struct bpf_func_state *state) |
17a52670 | 706 | { |
f4d7e40a | 707 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
708 | int i; |
709 | ||
dc503a8a | 710 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 711 | mark_reg_not_init(env, regs, i); |
dc503a8a EC |
712 | regs[i].live = REG_LIVE_NONE; |
713 | } | |
17a52670 AS |
714 | |
715 | /* frame pointer */ | |
f1174f77 | 716 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 717 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 718 | regs[BPF_REG_FP].frameno = state->frameno; |
17a52670 AS |
719 | |
720 | /* 1st arg to a function */ | |
721 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
61bd5218 | 722 | mark_reg_known_zero(env, regs, BPF_REG_1); |
6760bf2d DB |
723 | } |
724 | ||
f4d7e40a AS |
725 | #define BPF_MAIN_FUNC (-1) |
726 | static void init_func_state(struct bpf_verifier_env *env, | |
727 | struct bpf_func_state *state, | |
728 | int callsite, int frameno, int subprogno) | |
729 | { | |
730 | state->callsite = callsite; | |
731 | state->frameno = frameno; | |
732 | state->subprogno = subprogno; | |
733 | init_reg_state(env, state); | |
734 | } | |
735 | ||
17a52670 AS |
736 | enum reg_arg_type { |
737 | SRC_OP, /* register is used as source operand */ | |
738 | DST_OP, /* register is used as destination operand */ | |
739 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
740 | }; | |
741 | ||
cc8b0b92 AS |
742 | static int cmp_subprogs(const void *a, const void *b) |
743 | { | |
744 | return *(int *)a - *(int *)b; | |
745 | } | |
746 | ||
747 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
748 | { | |
749 | u32 *p; | |
750 | ||
751 | p = bsearch(&off, env->subprog_starts, env->subprog_cnt, | |
752 | sizeof(env->subprog_starts[0]), cmp_subprogs); | |
753 | if (!p) | |
754 | return -ENOENT; | |
755 | return p - env->subprog_starts; | |
756 | ||
757 | } | |
758 | ||
759 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
760 | { | |
761 | int insn_cnt = env->prog->len; | |
762 | int ret; | |
763 | ||
764 | if (off >= insn_cnt || off < 0) { | |
765 | verbose(env, "call to invalid destination\n"); | |
766 | return -EINVAL; | |
767 | } | |
768 | ret = find_subprog(env, off); | |
769 | if (ret >= 0) | |
770 | return 0; | |
771 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { | |
772 | verbose(env, "too many subprograms\n"); | |
773 | return -E2BIG; | |
774 | } | |
775 | env->subprog_starts[env->subprog_cnt++] = off; | |
776 | sort(env->subprog_starts, env->subprog_cnt, | |
777 | sizeof(env->subprog_starts[0]), cmp_subprogs, NULL); | |
778 | return 0; | |
779 | } | |
780 | ||
781 | static int check_subprogs(struct bpf_verifier_env *env) | |
782 | { | |
783 | int i, ret, subprog_start, subprog_end, off, cur_subprog = 0; | |
784 | struct bpf_insn *insn = env->prog->insnsi; | |
785 | int insn_cnt = env->prog->len; | |
786 | ||
787 | /* determine subprog starts. The end is one before the next starts */ | |
788 | for (i = 0; i < insn_cnt; i++) { | |
789 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
790 | continue; | |
791 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
792 | continue; | |
793 | if (!env->allow_ptr_leaks) { | |
794 | verbose(env, "function calls to other bpf functions are allowed for root only\n"); | |
795 | return -EPERM; | |
796 | } | |
797 | if (bpf_prog_is_dev_bound(env->prog->aux)) { | |
e90004d5 | 798 | verbose(env, "function calls in offloaded programs are not supported yet\n"); |
cc8b0b92 AS |
799 | return -EINVAL; |
800 | } | |
801 | ret = add_subprog(env, i + insn[i].imm + 1); | |
802 | if (ret < 0) | |
803 | return ret; | |
804 | } | |
805 | ||
806 | if (env->log.level > 1) | |
807 | for (i = 0; i < env->subprog_cnt; i++) | |
808 | verbose(env, "func#%d @%d\n", i, env->subprog_starts[i]); | |
809 | ||
810 | /* now check that all jumps are within the same subprog */ | |
811 | subprog_start = 0; | |
812 | if (env->subprog_cnt == cur_subprog) | |
813 | subprog_end = insn_cnt; | |
814 | else | |
815 | subprog_end = env->subprog_starts[cur_subprog++]; | |
816 | for (i = 0; i < insn_cnt; i++) { | |
817 | u8 code = insn[i].code; | |
818 | ||
819 | if (BPF_CLASS(code) != BPF_JMP) | |
820 | goto next; | |
821 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
822 | goto next; | |
823 | off = i + insn[i].off + 1; | |
824 | if (off < subprog_start || off >= subprog_end) { | |
825 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
826 | return -EINVAL; | |
827 | } | |
828 | next: | |
829 | if (i == subprog_end - 1) { | |
830 | /* to avoid fall-through from one subprog into another | |
831 | * the last insn of the subprog should be either exit | |
832 | * or unconditional jump back | |
833 | */ | |
834 | if (code != (BPF_JMP | BPF_EXIT) && | |
835 | code != (BPF_JMP | BPF_JA)) { | |
836 | verbose(env, "last insn is not an exit or jmp\n"); | |
837 | return -EINVAL; | |
838 | } | |
839 | subprog_start = subprog_end; | |
840 | if (env->subprog_cnt == cur_subprog) | |
841 | subprog_end = insn_cnt; | |
842 | else | |
843 | subprog_end = env->subprog_starts[cur_subprog++]; | |
844 | } | |
845 | } | |
846 | return 0; | |
847 | } | |
848 | ||
fa2d41ad | 849 | static |
f4d7e40a AS |
850 | struct bpf_verifier_state *skip_callee(struct bpf_verifier_env *env, |
851 | const struct bpf_verifier_state *state, | |
852 | struct bpf_verifier_state *parent, | |
853 | u32 regno) | |
dc503a8a | 854 | { |
f4d7e40a AS |
855 | struct bpf_verifier_state *tmp = NULL; |
856 | ||
857 | /* 'parent' could be a state of caller and | |
858 | * 'state' could be a state of callee. In such case | |
859 | * parent->curframe < state->curframe | |
860 | * and it's ok for r1 - r5 registers | |
861 | * | |
862 | * 'parent' could be a callee's state after it bpf_exit-ed. | |
863 | * In such case parent->curframe > state->curframe | |
864 | * and it's ok for r0 only | |
865 | */ | |
866 | if (parent->curframe == state->curframe || | |
867 | (parent->curframe < state->curframe && | |
868 | regno >= BPF_REG_1 && regno <= BPF_REG_5) || | |
869 | (parent->curframe > state->curframe && | |
870 | regno == BPF_REG_0)) | |
871 | return parent; | |
872 | ||
873 | if (parent->curframe > state->curframe && | |
874 | regno >= BPF_REG_6) { | |
875 | /* for callee saved regs we have to skip the whole chain | |
876 | * of states that belong to callee and mark as LIVE_READ | |
877 | * the registers before the call | |
878 | */ | |
879 | tmp = parent; | |
880 | while (tmp && tmp->curframe != state->curframe) { | |
881 | tmp = tmp->parent; | |
882 | } | |
883 | if (!tmp) | |
884 | goto bug; | |
885 | parent = tmp; | |
886 | } else { | |
887 | goto bug; | |
888 | } | |
889 | return parent; | |
890 | bug: | |
891 | verbose(env, "verifier bug regno %d tmp %p\n", regno, tmp); | |
892 | verbose(env, "regno %d parent frame %d current frame %d\n", | |
893 | regno, parent->curframe, state->curframe); | |
fa2d41ad | 894 | return NULL; |
f4d7e40a AS |
895 | } |
896 | ||
897 | static int mark_reg_read(struct bpf_verifier_env *env, | |
898 | const struct bpf_verifier_state *state, | |
899 | struct bpf_verifier_state *parent, | |
900 | u32 regno) | |
901 | { | |
902 | bool writes = parent == state->parent; /* Observe write marks */ | |
dc503a8a | 903 | |
8fe2d6cc AS |
904 | if (regno == BPF_REG_FP) |
905 | /* We don't need to worry about FP liveness because it's read-only */ | |
f4d7e40a | 906 | return 0; |
8fe2d6cc | 907 | |
dc503a8a EC |
908 | while (parent) { |
909 | /* if read wasn't screened by an earlier write ... */ | |
f4d7e40a | 910 | if (writes && state->frame[state->curframe]->regs[regno].live & REG_LIVE_WRITTEN) |
dc503a8a | 911 | break; |
f4d7e40a AS |
912 | parent = skip_callee(env, state, parent, regno); |
913 | if (!parent) | |
914 | return -EFAULT; | |
dc503a8a | 915 | /* ... then we depend on parent's value */ |
f4d7e40a | 916 | parent->frame[parent->curframe]->regs[regno].live |= REG_LIVE_READ; |
dc503a8a EC |
917 | state = parent; |
918 | parent = state->parent; | |
f4d7e40a | 919 | writes = true; |
dc503a8a | 920 | } |
f4d7e40a | 921 | return 0; |
dc503a8a EC |
922 | } |
923 | ||
924 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, | |
17a52670 AS |
925 | enum reg_arg_type t) |
926 | { | |
f4d7e40a AS |
927 | struct bpf_verifier_state *vstate = env->cur_state; |
928 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
929 | struct bpf_reg_state *regs = state->regs; | |
dc503a8a | 930 | |
17a52670 | 931 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 932 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
933 | return -EINVAL; |
934 | } | |
935 | ||
936 | if (t == SRC_OP) { | |
937 | /* check whether register used as source operand can be read */ | |
938 | if (regs[regno].type == NOT_INIT) { | |
61bd5218 | 939 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
940 | return -EACCES; |
941 | } | |
f4d7e40a | 942 | return mark_reg_read(env, vstate, vstate->parent, regno); |
17a52670 AS |
943 | } else { |
944 | /* check whether register used as dest operand can be written to */ | |
945 | if (regno == BPF_REG_FP) { | |
61bd5218 | 946 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
947 | return -EACCES; |
948 | } | |
dc503a8a | 949 | regs[regno].live |= REG_LIVE_WRITTEN; |
17a52670 | 950 | if (t == DST_OP) |
61bd5218 | 951 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
952 | } |
953 | return 0; | |
954 | } | |
955 | ||
1be7f75d AS |
956 | static bool is_spillable_regtype(enum bpf_reg_type type) |
957 | { | |
958 | switch (type) { | |
959 | case PTR_TO_MAP_VALUE: | |
960 | case PTR_TO_MAP_VALUE_OR_NULL: | |
961 | case PTR_TO_STACK: | |
962 | case PTR_TO_CTX: | |
969bf05e | 963 | case PTR_TO_PACKET: |
de8f3a83 | 964 | case PTR_TO_PACKET_META: |
969bf05e | 965 | case PTR_TO_PACKET_END: |
1be7f75d AS |
966 | case CONST_PTR_TO_MAP: |
967 | return true; | |
968 | default: | |
969 | return false; | |
970 | } | |
971 | } | |
972 | ||
cc2b14d5 AS |
973 | /* Does this register contain a constant zero? */ |
974 | static bool register_is_null(struct bpf_reg_state *reg) | |
975 | { | |
976 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
977 | } | |
978 | ||
17a52670 AS |
979 | /* check_stack_read/write functions track spill/fill of registers, |
980 | * stack boundary and alignment are checked in check_mem_access() | |
981 | */ | |
61bd5218 | 982 | static int check_stack_write(struct bpf_verifier_env *env, |
f4d7e40a AS |
983 | struct bpf_func_state *state, /* func where register points to */ |
984 | int off, int size, int value_regno) | |
17a52670 | 985 | { |
f4d7e40a | 986 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 987 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
f4d7e40a | 988 | enum bpf_reg_type type; |
638f5b90 | 989 | |
f4d7e40a AS |
990 | err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE), |
991 | true); | |
638f5b90 AS |
992 | if (err) |
993 | return err; | |
9c399760 AS |
994 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
995 | * so it's aligned access and [off, off + size) are within stack limits | |
996 | */ | |
638f5b90 AS |
997 | if (!env->allow_ptr_leaks && |
998 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
999 | size != BPF_REG_SIZE) { | |
1000 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
1001 | return -EACCES; | |
1002 | } | |
17a52670 | 1003 | |
f4d7e40a | 1004 | cur = env->cur_state->frame[env->cur_state->curframe]; |
17a52670 | 1005 | if (value_regno >= 0 && |
f4d7e40a | 1006 | is_spillable_regtype((type = cur->regs[value_regno].type))) { |
17a52670 AS |
1007 | |
1008 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 1009 | if (size != BPF_REG_SIZE) { |
61bd5218 | 1010 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
1011 | return -EACCES; |
1012 | } | |
1013 | ||
f4d7e40a AS |
1014 | if (state != cur && type == PTR_TO_STACK) { |
1015 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); | |
1016 | return -EINVAL; | |
1017 | } | |
1018 | ||
17a52670 | 1019 | /* save register state */ |
f4d7e40a | 1020 | state->stack[spi].spilled_ptr = cur->regs[value_regno]; |
638f5b90 | 1021 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; |
17a52670 | 1022 | |
9c399760 | 1023 | for (i = 0; i < BPF_REG_SIZE; i++) |
638f5b90 | 1024 | state->stack[spi].slot_type[i] = STACK_SPILL; |
9c399760 | 1025 | } else { |
cc2b14d5 AS |
1026 | u8 type = STACK_MISC; |
1027 | ||
17a52670 | 1028 | /* regular write of data into stack */ |
638f5b90 | 1029 | state->stack[spi].spilled_ptr = (struct bpf_reg_state) {}; |
9c399760 | 1030 | |
cc2b14d5 AS |
1031 | /* only mark the slot as written if all 8 bytes were written |
1032 | * otherwise read propagation may incorrectly stop too soon | |
1033 | * when stack slots are partially written. | |
1034 | * This heuristic means that read propagation will be | |
1035 | * conservative, since it will add reg_live_read marks | |
1036 | * to stack slots all the way to first state when programs | |
1037 | * writes+reads less than 8 bytes | |
1038 | */ | |
1039 | if (size == BPF_REG_SIZE) | |
1040 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1041 | ||
1042 | /* when we zero initialize stack slots mark them as such */ | |
1043 | if (value_regno >= 0 && | |
1044 | register_is_null(&cur->regs[value_regno])) | |
1045 | type = STACK_ZERO; | |
1046 | ||
9c399760 | 1047 | for (i = 0; i < size; i++) |
638f5b90 | 1048 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 1049 | type; |
17a52670 AS |
1050 | } |
1051 | return 0; | |
1052 | } | |
1053 | ||
f4d7e40a AS |
1054 | /* registers of every function are unique and mark_reg_read() propagates |
1055 | * the liveness in the following cases: | |
1056 | * - from callee into caller for R1 - R5 that were used as arguments | |
1057 | * - from caller into callee for R0 that used as result of the call | |
1058 | * - from caller to the same caller skipping states of the callee for R6 - R9, | |
1059 | * since R6 - R9 are callee saved by implicit function prologue and | |
1060 | * caller's R6 != callee's R6, so when we propagate liveness up to | |
1061 | * parent states we need to skip callee states for R6 - R9. | |
1062 | * | |
1063 | * stack slot marking is different, since stacks of caller and callee are | |
1064 | * accessible in both (since caller can pass a pointer to caller's stack to | |
1065 | * callee which can pass it to another function), hence mark_stack_slot_read() | |
1066 | * has to propagate the stack liveness to all parent states at given frame number. | |
1067 | * Consider code: | |
1068 | * f1() { | |
1069 | * ptr = fp - 8; | |
1070 | * *ptr = ctx; | |
1071 | * call f2 { | |
1072 | * .. = *ptr; | |
1073 | * } | |
1074 | * .. = *ptr; | |
1075 | * } | |
1076 | * First *ptr is reading from f1's stack and mark_stack_slot_read() has | |
1077 | * to mark liveness at the f1's frame and not f2's frame. | |
1078 | * Second *ptr is also reading from f1's stack and mark_stack_slot_read() has | |
1079 | * to propagate liveness to f2 states at f1's frame level and further into | |
1080 | * f1 states at f1's frame level until write into that stack slot | |
1081 | */ | |
1082 | static void mark_stack_slot_read(struct bpf_verifier_env *env, | |
1083 | const struct bpf_verifier_state *state, | |
1084 | struct bpf_verifier_state *parent, | |
1085 | int slot, int frameno) | |
dc503a8a | 1086 | { |
f4d7e40a | 1087 | bool writes = parent == state->parent; /* Observe write marks */ |
dc503a8a EC |
1088 | |
1089 | while (parent) { | |
cc2b14d5 AS |
1090 | if (parent->frame[frameno]->allocated_stack <= slot * BPF_REG_SIZE) |
1091 | /* since LIVE_WRITTEN mark is only done for full 8-byte | |
1092 | * write the read marks are conservative and parent | |
1093 | * state may not even have the stack allocated. In such case | |
1094 | * end the propagation, since the loop reached beginning | |
1095 | * of the function | |
1096 | */ | |
1097 | break; | |
dc503a8a | 1098 | /* if read wasn't screened by an earlier write ... */ |
f4d7e40a | 1099 | if (writes && state->frame[frameno]->stack[slot].spilled_ptr.live & REG_LIVE_WRITTEN) |
dc503a8a EC |
1100 | break; |
1101 | /* ... then we depend on parent's value */ | |
f4d7e40a | 1102 | parent->frame[frameno]->stack[slot].spilled_ptr.live |= REG_LIVE_READ; |
dc503a8a EC |
1103 | state = parent; |
1104 | parent = state->parent; | |
f4d7e40a | 1105 | writes = true; |
dc503a8a EC |
1106 | } |
1107 | } | |
1108 | ||
61bd5218 | 1109 | static int check_stack_read(struct bpf_verifier_env *env, |
f4d7e40a AS |
1110 | struct bpf_func_state *reg_state /* func where register points to */, |
1111 | int off, int size, int value_regno) | |
17a52670 | 1112 | { |
f4d7e40a AS |
1113 | struct bpf_verifier_state *vstate = env->cur_state; |
1114 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 AS |
1115 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
1116 | u8 *stype; | |
17a52670 | 1117 | |
f4d7e40a | 1118 | if (reg_state->allocated_stack <= slot) { |
638f5b90 AS |
1119 | verbose(env, "invalid read from stack off %d+0 size %d\n", |
1120 | off, size); | |
1121 | return -EACCES; | |
1122 | } | |
f4d7e40a | 1123 | stype = reg_state->stack[spi].slot_type; |
17a52670 | 1124 | |
638f5b90 | 1125 | if (stype[0] == STACK_SPILL) { |
9c399760 | 1126 | if (size != BPF_REG_SIZE) { |
61bd5218 | 1127 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
1128 | return -EACCES; |
1129 | } | |
9c399760 | 1130 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 1131 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 1132 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
1133 | return -EACCES; |
1134 | } | |
1135 | } | |
1136 | ||
dc503a8a | 1137 | if (value_regno >= 0) { |
17a52670 | 1138 | /* restore register state from stack */ |
f4d7e40a | 1139 | state->regs[value_regno] = reg_state->stack[spi].spilled_ptr; |
2f18f62e AS |
1140 | /* mark reg as written since spilled pointer state likely |
1141 | * has its liveness marks cleared by is_state_visited() | |
1142 | * which resets stack/reg liveness for state transitions | |
1143 | */ | |
1144 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
dc503a8a | 1145 | } |
cc2b14d5 AS |
1146 | mark_stack_slot_read(env, vstate, vstate->parent, spi, |
1147 | reg_state->frameno); | |
17a52670 AS |
1148 | return 0; |
1149 | } else { | |
cc2b14d5 AS |
1150 | int zeros = 0; |
1151 | ||
17a52670 | 1152 | for (i = 0; i < size; i++) { |
cc2b14d5 AS |
1153 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_MISC) |
1154 | continue; | |
1155 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_ZERO) { | |
1156 | zeros++; | |
1157 | continue; | |
17a52670 | 1158 | } |
cc2b14d5 AS |
1159 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
1160 | off, i, size); | |
1161 | return -EACCES; | |
1162 | } | |
1163 | mark_stack_slot_read(env, vstate, vstate->parent, spi, | |
1164 | reg_state->frameno); | |
1165 | if (value_regno >= 0) { | |
1166 | if (zeros == size) { | |
1167 | /* any size read into register is zero extended, | |
1168 | * so the whole register == const_zero | |
1169 | */ | |
1170 | __mark_reg_const_zero(&state->regs[value_regno]); | |
1171 | } else { | |
1172 | /* have read misc data from the stack */ | |
1173 | mark_reg_unknown(env, state->regs, value_regno); | |
1174 | } | |
1175 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
17a52670 | 1176 | } |
17a52670 AS |
1177 | return 0; |
1178 | } | |
1179 | } | |
1180 | ||
1181 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
f1174f77 | 1182 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1183 | int size, bool zero_size_allowed) |
17a52670 | 1184 | { |
638f5b90 AS |
1185 | struct bpf_reg_state *regs = cur_regs(env); |
1186 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 1187 | |
9fd29c08 YS |
1188 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1189 | off + size > map->value_size) { | |
61bd5218 | 1190 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
1191 | map->value_size, off, size); |
1192 | return -EACCES; | |
1193 | } | |
1194 | return 0; | |
1195 | } | |
1196 | ||
f1174f77 EC |
1197 | /* check read/write into a map element with possible variable offset */ |
1198 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 1199 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 1200 | { |
f4d7e40a AS |
1201 | struct bpf_verifier_state *vstate = env->cur_state; |
1202 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
1203 | struct bpf_reg_state *reg = &state->regs[regno]; |
1204 | int err; | |
1205 | ||
f1174f77 EC |
1206 | /* We may have adjusted the register to this map value, so we |
1207 | * need to try adding each of min_value and max_value to off | |
1208 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 1209 | */ |
61bd5218 JK |
1210 | if (env->log.level) |
1211 | print_verifier_state(env, state); | |
dbcfe5f7 GB |
1212 | /* The minimum value is only important with signed |
1213 | * comparisons where we can't assume the floor of a | |
1214 | * value is 0. If we are using signed variables for our | |
1215 | * index'es we need to make sure that whatever we use | |
1216 | * will have a set floor within our range. | |
1217 | */ | |
b03c9f9f | 1218 | if (reg->smin_value < 0) { |
61bd5218 | 1219 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
1220 | regno); |
1221 | return -EACCES; | |
1222 | } | |
9fd29c08 YS |
1223 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
1224 | zero_size_allowed); | |
dbcfe5f7 | 1225 | if (err) { |
61bd5218 JK |
1226 | verbose(env, "R%d min value is outside of the array range\n", |
1227 | regno); | |
dbcfe5f7 GB |
1228 | return err; |
1229 | } | |
1230 | ||
b03c9f9f EC |
1231 | /* If we haven't set a max value then we need to bail since we can't be |
1232 | * sure we won't do bad things. | |
1233 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 1234 | */ |
b03c9f9f | 1235 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 1236 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
1237 | regno); |
1238 | return -EACCES; | |
1239 | } | |
9fd29c08 YS |
1240 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
1241 | zero_size_allowed); | |
f1174f77 | 1242 | if (err) |
61bd5218 JK |
1243 | verbose(env, "R%d max value is outside of the array range\n", |
1244 | regno); | |
f1174f77 | 1245 | return err; |
dbcfe5f7 GB |
1246 | } |
1247 | ||
969bf05e AS |
1248 | #define MAX_PACKET_OFF 0xffff |
1249 | ||
58e2af8b | 1250 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
1251 | const struct bpf_call_arg_meta *meta, |
1252 | enum bpf_access_type t) | |
4acf6c0b | 1253 | { |
36bbef52 | 1254 | switch (env->prog->type) { |
3a0af8fd TG |
1255 | case BPF_PROG_TYPE_LWT_IN: |
1256 | case BPF_PROG_TYPE_LWT_OUT: | |
1257 | /* dst_input() and dst_output() can't write for now */ | |
1258 | if (t == BPF_WRITE) | |
1259 | return false; | |
7e57fbb2 | 1260 | /* fallthrough */ |
36bbef52 DB |
1261 | case BPF_PROG_TYPE_SCHED_CLS: |
1262 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 1263 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 1264 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 1265 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 1266 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
1267 | if (meta) |
1268 | return meta->pkt_access; | |
1269 | ||
1270 | env->seen_direct_write = true; | |
4acf6c0b BB |
1271 | return true; |
1272 | default: | |
1273 | return false; | |
1274 | } | |
1275 | } | |
1276 | ||
f1174f77 | 1277 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 1278 | int off, int size, bool zero_size_allowed) |
969bf05e | 1279 | { |
638f5b90 | 1280 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 1281 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 1282 | |
9fd29c08 YS |
1283 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1284 | (u64)off + size > reg->range) { | |
61bd5218 | 1285 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 1286 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
1287 | return -EACCES; |
1288 | } | |
1289 | return 0; | |
1290 | } | |
1291 | ||
f1174f77 | 1292 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1293 | int size, bool zero_size_allowed) |
f1174f77 | 1294 | { |
638f5b90 | 1295 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
1296 | struct bpf_reg_state *reg = ®s[regno]; |
1297 | int err; | |
1298 | ||
1299 | /* We may have added a variable offset to the packet pointer; but any | |
1300 | * reg->range we have comes after that. We are only checking the fixed | |
1301 | * offset. | |
1302 | */ | |
1303 | ||
1304 | /* We don't allow negative numbers, because we aren't tracking enough | |
1305 | * detail to prove they're safe. | |
1306 | */ | |
b03c9f9f | 1307 | if (reg->smin_value < 0) { |
61bd5218 | 1308 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
1309 | regno); |
1310 | return -EACCES; | |
1311 | } | |
9fd29c08 | 1312 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 1313 | if (err) { |
61bd5218 | 1314 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
1315 | return err; |
1316 | } | |
1317 | return err; | |
1318 | } | |
1319 | ||
1320 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 1321 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 1322 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 1323 | { |
f96da094 DB |
1324 | struct bpf_insn_access_aux info = { |
1325 | .reg_type = *reg_type, | |
1326 | }; | |
31fd8581 | 1327 | |
4f9218aa | 1328 | if (env->ops->is_valid_access && |
5e43f899 | 1329 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
1330 | /* A non zero info.ctx_field_size indicates that this field is a |
1331 | * candidate for later verifier transformation to load the whole | |
1332 | * field and then apply a mask when accessed with a narrower | |
1333 | * access than actual ctx access size. A zero info.ctx_field_size | |
1334 | * will only allow for whole field access and rejects any other | |
1335 | * type of narrower access. | |
31fd8581 | 1336 | */ |
23994631 | 1337 | *reg_type = info.reg_type; |
31fd8581 | 1338 | |
4f9218aa | 1339 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
1340 | /* remember the offset of last byte accessed in ctx */ |
1341 | if (env->prog->aux->max_ctx_offset < off + size) | |
1342 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 1343 | return 0; |
32bbe007 | 1344 | } |
17a52670 | 1345 | |
61bd5218 | 1346 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
1347 | return -EACCES; |
1348 | } | |
1349 | ||
4cabc5b1 DB |
1350 | static bool __is_pointer_value(bool allow_ptr_leaks, |
1351 | const struct bpf_reg_state *reg) | |
1be7f75d | 1352 | { |
4cabc5b1 | 1353 | if (allow_ptr_leaks) |
1be7f75d AS |
1354 | return false; |
1355 | ||
f1174f77 | 1356 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
1357 | } |
1358 | ||
4cabc5b1 DB |
1359 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1360 | { | |
638f5b90 | 1361 | return __is_pointer_value(env->allow_ptr_leaks, cur_regs(env) + regno); |
4cabc5b1 DB |
1362 | } |
1363 | ||
f37a8cb8 DB |
1364 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
1365 | { | |
1366 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1367 | ||
1368 | return reg->type == PTR_TO_CTX; | |
1369 | } | |
1370 | ||
ca369602 DB |
1371 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
1372 | { | |
1373 | const struct bpf_reg_state *reg = cur_regs(env) + regno; | |
1374 | ||
1375 | return type_is_pkt_pointer(reg->type); | |
1376 | } | |
1377 | ||
61bd5218 JK |
1378 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
1379 | const struct bpf_reg_state *reg, | |
d1174416 | 1380 | int off, int size, bool strict) |
969bf05e | 1381 | { |
f1174f77 | 1382 | struct tnum reg_off; |
e07b98d9 | 1383 | int ip_align; |
d1174416 DM |
1384 | |
1385 | /* Byte size accesses are always allowed. */ | |
1386 | if (!strict || size == 1) | |
1387 | return 0; | |
1388 | ||
e4eda884 DM |
1389 | /* For platforms that do not have a Kconfig enabling |
1390 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
1391 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
1392 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
1393 | * to this code only in strict mode where we want to emulate | |
1394 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
1395 | * unconditional IP align value of '2'. | |
e07b98d9 | 1396 | */ |
e4eda884 | 1397 | ip_align = 2; |
f1174f77 EC |
1398 | |
1399 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1400 | if (!tnum_is_aligned(reg_off, size)) { | |
1401 | char tn_buf[48]; | |
1402 | ||
1403 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1404 | verbose(env, |
1405 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1406 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1407 | return -EACCES; |
1408 | } | |
79adffcd | 1409 | |
969bf05e AS |
1410 | return 0; |
1411 | } | |
1412 | ||
61bd5218 JK |
1413 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1414 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1415 | const char *pointer_desc, |
1416 | int off, int size, bool strict) | |
79adffcd | 1417 | { |
f1174f77 EC |
1418 | struct tnum reg_off; |
1419 | ||
1420 | /* Byte size accesses are always allowed. */ | |
1421 | if (!strict || size == 1) | |
1422 | return 0; | |
1423 | ||
1424 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1425 | if (!tnum_is_aligned(reg_off, size)) { | |
1426 | char tn_buf[48]; | |
1427 | ||
1428 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1429 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1430 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1431 | return -EACCES; |
1432 | } | |
1433 | ||
969bf05e AS |
1434 | return 0; |
1435 | } | |
1436 | ||
e07b98d9 | 1437 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
1438 | const struct bpf_reg_state *reg, int off, |
1439 | int size, bool strict_alignment_once) | |
79adffcd | 1440 | { |
ca369602 | 1441 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 1442 | const char *pointer_desc = ""; |
d1174416 | 1443 | |
79adffcd DB |
1444 | switch (reg->type) { |
1445 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1446 | case PTR_TO_PACKET_META: |
1447 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1448 | * right in front, treat it the very same way. | |
1449 | */ | |
61bd5218 | 1450 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
f1174f77 EC |
1451 | case PTR_TO_MAP_VALUE: |
1452 | pointer_desc = "value "; | |
1453 | break; | |
1454 | case PTR_TO_CTX: | |
1455 | pointer_desc = "context "; | |
1456 | break; | |
1457 | case PTR_TO_STACK: | |
1458 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
1459 | /* The stack spill tracking logic in check_stack_write() |
1460 | * and check_stack_read() relies on stack accesses being | |
1461 | * aligned. | |
1462 | */ | |
1463 | strict = true; | |
f1174f77 | 1464 | break; |
79adffcd | 1465 | default: |
f1174f77 | 1466 | break; |
79adffcd | 1467 | } |
61bd5218 JK |
1468 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1469 | strict); | |
79adffcd DB |
1470 | } |
1471 | ||
f4d7e40a AS |
1472 | static int update_stack_depth(struct bpf_verifier_env *env, |
1473 | const struct bpf_func_state *func, | |
1474 | int off) | |
1475 | { | |
70a87ffe | 1476 | u16 stack = env->subprog_stack_depth[func->subprogno]; |
f4d7e40a AS |
1477 | |
1478 | if (stack >= -off) | |
1479 | return 0; | |
1480 | ||
1481 | /* update known max for given subprogram */ | |
1482 | env->subprog_stack_depth[func->subprogno] = -off; | |
70a87ffe AS |
1483 | return 0; |
1484 | } | |
f4d7e40a | 1485 | |
70a87ffe AS |
1486 | /* starting from main bpf function walk all instructions of the function |
1487 | * and recursively walk all callees that given function can call. | |
1488 | * Ignore jump and exit insns. | |
1489 | * Since recursion is prevented by check_cfg() this algorithm | |
1490 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
1491 | */ | |
1492 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
1493 | { | |
1494 | int depth = 0, frame = 0, subprog = 0, i = 0, subprog_end; | |
1495 | struct bpf_insn *insn = env->prog->insnsi; | |
1496 | int insn_cnt = env->prog->len; | |
1497 | int ret_insn[MAX_CALL_FRAMES]; | |
1498 | int ret_prog[MAX_CALL_FRAMES]; | |
f4d7e40a | 1499 | |
70a87ffe AS |
1500 | process_func: |
1501 | /* round up to 32-bytes, since this is granularity | |
1502 | * of interpreter stack size | |
1503 | */ | |
1504 | depth += round_up(max_t(u32, env->subprog_stack_depth[subprog], 1), 32); | |
1505 | if (depth > MAX_BPF_STACK) { | |
f4d7e40a | 1506 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 1507 | frame + 1, depth); |
f4d7e40a AS |
1508 | return -EACCES; |
1509 | } | |
70a87ffe AS |
1510 | continue_func: |
1511 | if (env->subprog_cnt == subprog) | |
1512 | subprog_end = insn_cnt; | |
1513 | else | |
1514 | subprog_end = env->subprog_starts[subprog]; | |
1515 | for (; i < subprog_end; i++) { | |
1516 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
1517 | continue; | |
1518 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
1519 | continue; | |
1520 | /* remember insn and function to return to */ | |
1521 | ret_insn[frame] = i + 1; | |
1522 | ret_prog[frame] = subprog; | |
1523 | ||
1524 | /* find the callee */ | |
1525 | i = i + insn[i].imm + 1; | |
1526 | subprog = find_subprog(env, i); | |
1527 | if (subprog < 0) { | |
1528 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
1529 | i); | |
1530 | return -EFAULT; | |
1531 | } | |
1532 | subprog++; | |
1533 | frame++; | |
1534 | if (frame >= MAX_CALL_FRAMES) { | |
1535 | WARN_ONCE(1, "verifier bug. Call stack is too deep\n"); | |
1536 | return -EFAULT; | |
1537 | } | |
1538 | goto process_func; | |
1539 | } | |
1540 | /* end of for() loop means the last insn of the 'subprog' | |
1541 | * was reached. Doesn't matter whether it was JA or EXIT | |
1542 | */ | |
1543 | if (frame == 0) | |
1544 | return 0; | |
1545 | depth -= round_up(max_t(u32, env->subprog_stack_depth[subprog], 1), 32); | |
1546 | frame--; | |
1547 | i = ret_insn[frame]; | |
1548 | subprog = ret_prog[frame]; | |
1549 | goto continue_func; | |
f4d7e40a AS |
1550 | } |
1551 | ||
19d28fbd | 1552 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
1553 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
1554 | const struct bpf_insn *insn, int idx) | |
1555 | { | |
1556 | int start = idx + insn->imm + 1, subprog; | |
1557 | ||
1558 | subprog = find_subprog(env, start); | |
1559 | if (subprog < 0) { | |
1560 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
1561 | start); | |
1562 | return -EFAULT; | |
1563 | } | |
1564 | subprog++; | |
1565 | return env->subprog_stack_depth[subprog]; | |
1566 | } | |
19d28fbd | 1567 | #endif |
1ea47e01 | 1568 | |
0c17d1d2 JH |
1569 | /* truncate register to smaller size (in bytes) |
1570 | * must be called with size < BPF_REG_SIZE | |
1571 | */ | |
1572 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
1573 | { | |
1574 | u64 mask; | |
1575 | ||
1576 | /* clear high bits in bit representation */ | |
1577 | reg->var_off = tnum_cast(reg->var_off, size); | |
1578 | ||
1579 | /* fix arithmetic bounds */ | |
1580 | mask = ((u64)1 << (size * 8)) - 1; | |
1581 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
1582 | reg->umin_value &= mask; | |
1583 | reg->umax_value &= mask; | |
1584 | } else { | |
1585 | reg->umin_value = 0; | |
1586 | reg->umax_value = mask; | |
1587 | } | |
1588 | reg->smin_value = reg->umin_value; | |
1589 | reg->smax_value = reg->umax_value; | |
1590 | } | |
1591 | ||
17a52670 AS |
1592 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1593 | * if t==write, value_regno is a register which value is stored into memory | |
1594 | * if t==read, value_regno is a register which will receive the value from memory | |
1595 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1596 | * if t==read && value_regno==-1, don't care what we read from memory | |
1597 | */ | |
ca369602 DB |
1598 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
1599 | int off, int bpf_size, enum bpf_access_type t, | |
1600 | int value_regno, bool strict_alignment_once) | |
17a52670 | 1601 | { |
638f5b90 AS |
1602 | struct bpf_reg_state *regs = cur_regs(env); |
1603 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 1604 | struct bpf_func_state *state; |
17a52670 AS |
1605 | int size, err = 0; |
1606 | ||
1607 | size = bpf_size_to_bytes(bpf_size); | |
1608 | if (size < 0) | |
1609 | return size; | |
1610 | ||
f1174f77 | 1611 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 1612 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
1613 | if (err) |
1614 | return err; | |
17a52670 | 1615 | |
f1174f77 EC |
1616 | /* for access checks, reg->off is just part of off */ |
1617 | off += reg->off; | |
1618 | ||
1619 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1620 | if (t == BPF_WRITE && value_regno >= 0 && |
1621 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1622 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1623 | return -EACCES; |
1624 | } | |
48461135 | 1625 | |
9fd29c08 | 1626 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 1627 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1628 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1629 | |
1a0dc1ac | 1630 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1631 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1632 | |
1be7f75d AS |
1633 | if (t == BPF_WRITE && value_regno >= 0 && |
1634 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1635 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1636 | return -EACCES; |
1637 | } | |
f1174f77 EC |
1638 | /* ctx accesses must be at a fixed offset, so that we can |
1639 | * determine what type of data were returned. | |
1640 | */ | |
28e33f9d | 1641 | if (reg->off) { |
f8ddadc4 DM |
1642 | verbose(env, |
1643 | "dereference of modified ctx ptr R%d off=%d+%d, ctx+const is allowed, ctx+const+const is not\n", | |
28e33f9d JK |
1644 | regno, reg->off, off - reg->off); |
1645 | return -EACCES; | |
1646 | } | |
1647 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
f1174f77 EC |
1648 | char tn_buf[48]; |
1649 | ||
1650 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1651 | verbose(env, |
1652 | "variable ctx access var_off=%s off=%d size=%d", | |
f1174f77 EC |
1653 | tn_buf, off, size); |
1654 | return -EACCES; | |
1655 | } | |
31fd8581 | 1656 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1657 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1658 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1659 | * PTR_TO_PACKET[_META,_END]. In the latter |
1660 | * case, we know the offset is zero. | |
f1174f77 EC |
1661 | */ |
1662 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1663 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1664 | else |
638f5b90 | 1665 | mark_reg_known_zero(env, regs, |
61bd5218 | 1666 | value_regno); |
638f5b90 AS |
1667 | regs[value_regno].id = 0; |
1668 | regs[value_regno].off = 0; | |
1669 | regs[value_regno].range = 0; | |
1670 | regs[value_regno].type = reg_type; | |
969bf05e | 1671 | } |
17a52670 | 1672 | |
f1174f77 EC |
1673 | } else if (reg->type == PTR_TO_STACK) { |
1674 | /* stack accesses must be at a fixed offset, so that we can | |
1675 | * determine what type of data were returned. | |
1676 | * See check_stack_read(). | |
1677 | */ | |
1678 | if (!tnum_is_const(reg->var_off)) { | |
1679 | char tn_buf[48]; | |
1680 | ||
1681 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1682 | verbose(env, "variable stack access var_off=%s off=%d size=%d", |
f1174f77 EC |
1683 | tn_buf, off, size); |
1684 | return -EACCES; | |
1685 | } | |
1686 | off += reg->var_off.value; | |
17a52670 | 1687 | if (off >= 0 || off < -MAX_BPF_STACK) { |
61bd5218 JK |
1688 | verbose(env, "invalid stack off=%d size=%d\n", off, |
1689 | size); | |
17a52670 AS |
1690 | return -EACCES; |
1691 | } | |
8726679a | 1692 | |
f4d7e40a AS |
1693 | state = func(env, reg); |
1694 | err = update_stack_depth(env, state, off); | |
1695 | if (err) | |
1696 | return err; | |
8726679a | 1697 | |
638f5b90 | 1698 | if (t == BPF_WRITE) |
61bd5218 JK |
1699 | err = check_stack_write(env, state, off, size, |
1700 | value_regno); | |
638f5b90 | 1701 | else |
61bd5218 JK |
1702 | err = check_stack_read(env, state, off, size, |
1703 | value_regno); | |
de8f3a83 | 1704 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 1705 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 1706 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
1707 | return -EACCES; |
1708 | } | |
4acf6c0b BB |
1709 | if (t == BPF_WRITE && value_regno >= 0 && |
1710 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
1711 | verbose(env, "R%d leaks addr into packet\n", |
1712 | value_regno); | |
4acf6c0b BB |
1713 | return -EACCES; |
1714 | } | |
9fd29c08 | 1715 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 1716 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1717 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1718 | } else { |
61bd5218 JK |
1719 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
1720 | reg_type_str[reg->type]); | |
17a52670 AS |
1721 | return -EACCES; |
1722 | } | |
969bf05e | 1723 | |
f1174f77 | 1724 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 1725 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 1726 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 1727 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 1728 | } |
17a52670 AS |
1729 | return err; |
1730 | } | |
1731 | ||
31fd8581 | 1732 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 1733 | { |
17a52670 AS |
1734 | int err; |
1735 | ||
1736 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
1737 | insn->imm != 0) { | |
61bd5218 | 1738 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
1739 | return -EINVAL; |
1740 | } | |
1741 | ||
1742 | /* check src1 operand */ | |
dc503a8a | 1743 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
1744 | if (err) |
1745 | return err; | |
1746 | ||
1747 | /* check src2 operand */ | |
dc503a8a | 1748 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
1749 | if (err) |
1750 | return err; | |
1751 | ||
6bdf6abc | 1752 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 1753 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
1754 | return -EACCES; |
1755 | } | |
1756 | ||
ca369602 DB |
1757 | if (is_ctx_reg(env, insn->dst_reg) || |
1758 | is_pkt_reg(env, insn->dst_reg)) { | |
1759 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", | |
1760 | insn->dst_reg, is_ctx_reg(env, insn->dst_reg) ? | |
1761 | "context" : "packet"); | |
f37a8cb8 DB |
1762 | return -EACCES; |
1763 | } | |
1764 | ||
17a52670 | 1765 | /* check whether atomic_add can read the memory */ |
31fd8581 | 1766 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 1767 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
1768 | if (err) |
1769 | return err; | |
1770 | ||
1771 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 1772 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 1773 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
1774 | } |
1775 | ||
1776 | /* when register 'regno' is passed into function that will read 'access_size' | |
1777 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
1778 | * and all elements of stack are initialized. |
1779 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
1780 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 1781 | */ |
58e2af8b | 1782 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
1783 | int access_size, bool zero_size_allowed, |
1784 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1785 | { |
914cb781 | 1786 | struct bpf_reg_state *reg = cur_regs(env) + regno; |
f4d7e40a | 1787 | struct bpf_func_state *state = func(env, reg); |
638f5b90 | 1788 | int off, i, slot, spi; |
17a52670 | 1789 | |
914cb781 | 1790 | if (reg->type != PTR_TO_STACK) { |
f1174f77 | 1791 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 1792 | if (zero_size_allowed && access_size == 0 && |
914cb781 | 1793 | register_is_null(reg)) |
8e2fe1d9 DB |
1794 | return 0; |
1795 | ||
61bd5218 | 1796 | verbose(env, "R%d type=%s expected=%s\n", regno, |
914cb781 | 1797 | reg_type_str[reg->type], |
8e2fe1d9 | 1798 | reg_type_str[PTR_TO_STACK]); |
17a52670 | 1799 | return -EACCES; |
8e2fe1d9 | 1800 | } |
17a52670 | 1801 | |
f1174f77 | 1802 | /* Only allow fixed-offset stack reads */ |
914cb781 | 1803 | if (!tnum_is_const(reg->var_off)) { |
f1174f77 EC |
1804 | char tn_buf[48]; |
1805 | ||
914cb781 | 1806 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 1807 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 | 1808 | regno, tn_buf); |
ea25f914 | 1809 | return -EACCES; |
f1174f77 | 1810 | } |
914cb781 | 1811 | off = reg->off + reg->var_off.value; |
17a52670 | 1812 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
9fd29c08 | 1813 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { |
61bd5218 | 1814 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
1815 | regno, off, access_size); |
1816 | return -EACCES; | |
1817 | } | |
1818 | ||
435faee1 DB |
1819 | if (meta && meta->raw_mode) { |
1820 | meta->access_size = access_size; | |
1821 | meta->regno = regno; | |
1822 | return 0; | |
1823 | } | |
1824 | ||
17a52670 | 1825 | for (i = 0; i < access_size; i++) { |
cc2b14d5 AS |
1826 | u8 *stype; |
1827 | ||
638f5b90 AS |
1828 | slot = -(off + i) - 1; |
1829 | spi = slot / BPF_REG_SIZE; | |
cc2b14d5 AS |
1830 | if (state->allocated_stack <= slot) |
1831 | goto err; | |
1832 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
1833 | if (*stype == STACK_MISC) | |
1834 | goto mark; | |
1835 | if (*stype == STACK_ZERO) { | |
1836 | /* helper can write anything into the stack */ | |
1837 | *stype = STACK_MISC; | |
1838 | goto mark; | |
17a52670 | 1839 | } |
cc2b14d5 AS |
1840 | err: |
1841 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", | |
1842 | off, i, access_size); | |
1843 | return -EACCES; | |
1844 | mark: | |
1845 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
1846 | * the whole slot to be marked as 'read' | |
1847 | */ | |
1848 | mark_stack_slot_read(env, env->cur_state, env->cur_state->parent, | |
1849 | spi, state->frameno); | |
17a52670 | 1850 | } |
f4d7e40a | 1851 | return update_stack_depth(env, state, off); |
17a52670 AS |
1852 | } |
1853 | ||
06c1c049 GB |
1854 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1855 | int access_size, bool zero_size_allowed, | |
1856 | struct bpf_call_arg_meta *meta) | |
1857 | { | |
638f5b90 | 1858 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 1859 | |
f1174f77 | 1860 | switch (reg->type) { |
06c1c049 | 1861 | case PTR_TO_PACKET: |
de8f3a83 | 1862 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
1863 | return check_packet_access(env, regno, reg->off, access_size, |
1864 | zero_size_allowed); | |
06c1c049 | 1865 | case PTR_TO_MAP_VALUE: |
9fd29c08 YS |
1866 | return check_map_access(env, regno, reg->off, access_size, |
1867 | zero_size_allowed); | |
f1174f77 | 1868 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
1869 | return check_stack_boundary(env, regno, access_size, |
1870 | zero_size_allowed, meta); | |
1871 | } | |
1872 | } | |
1873 | ||
90133415 DB |
1874 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
1875 | { | |
1876 | return type == ARG_PTR_TO_MEM || | |
1877 | type == ARG_PTR_TO_MEM_OR_NULL || | |
1878 | type == ARG_PTR_TO_UNINIT_MEM; | |
1879 | } | |
1880 | ||
1881 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
1882 | { | |
1883 | return type == ARG_CONST_SIZE || | |
1884 | type == ARG_CONST_SIZE_OR_ZERO; | |
1885 | } | |
1886 | ||
58e2af8b | 1887 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1888 | enum bpf_arg_type arg_type, |
1889 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1890 | { |
638f5b90 | 1891 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 1892 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1893 | int err = 0; |
1894 | ||
80f1d68c | 1895 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1896 | return 0; |
1897 | ||
dc503a8a EC |
1898 | err = check_reg_arg(env, regno, SRC_OP); |
1899 | if (err) | |
1900 | return err; | |
17a52670 | 1901 | |
1be7f75d AS |
1902 | if (arg_type == ARG_ANYTHING) { |
1903 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
1904 | verbose(env, "R%d leaks addr into helper function\n", |
1905 | regno); | |
1be7f75d AS |
1906 | return -EACCES; |
1907 | } | |
80f1d68c | 1908 | return 0; |
1be7f75d | 1909 | } |
80f1d68c | 1910 | |
de8f3a83 | 1911 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 1912 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 1913 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
1914 | return -EACCES; |
1915 | } | |
1916 | ||
8e2fe1d9 | 1917 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1918 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1919 | expected_type = PTR_TO_STACK; | |
d71962f3 | 1920 | if (!type_is_pkt_pointer(type) && type != PTR_TO_MAP_VALUE && |
de8f3a83 | 1921 | type != expected_type) |
6841de8b | 1922 | goto err_type; |
39f19ebb AS |
1923 | } else if (arg_type == ARG_CONST_SIZE || |
1924 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
1925 | expected_type = SCALAR_VALUE; |
1926 | if (type != expected_type) | |
6841de8b | 1927 | goto err_type; |
17a52670 AS |
1928 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1929 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1930 | if (type != expected_type) |
1931 | goto err_type; | |
608cd71a AS |
1932 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1933 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1934 | if (type != expected_type) |
1935 | goto err_type; | |
90133415 | 1936 | } else if (arg_type_is_mem_ptr(arg_type)) { |
8e2fe1d9 DB |
1937 | expected_type = PTR_TO_STACK; |
1938 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 1939 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
1940 | * happens during stack boundary checking. |
1941 | */ | |
914cb781 | 1942 | if (register_is_null(reg) && |
db1ac496 | 1943 | arg_type == ARG_PTR_TO_MEM_OR_NULL) |
6841de8b | 1944 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
1945 | else if (!type_is_pkt_pointer(type) && |
1946 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 1947 | type != expected_type) |
6841de8b | 1948 | goto err_type; |
39f19ebb | 1949 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 1950 | } else { |
61bd5218 | 1951 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
1952 | return -EFAULT; |
1953 | } | |
1954 | ||
17a52670 AS |
1955 | if (arg_type == ARG_CONST_MAP_PTR) { |
1956 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1957 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1958 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1959 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1960 | * check that [key, key + map->key_size) are within | |
1961 | * stack limits and initialized | |
1962 | */ | |
33ff9823 | 1963 | if (!meta->map_ptr) { |
17a52670 AS |
1964 | /* in function declaration map_ptr must come before |
1965 | * map_key, so that it's verified and known before | |
1966 | * we have to check map_key here. Otherwise it means | |
1967 | * that kernel subsystem misconfigured verifier | |
1968 | */ | |
61bd5218 | 1969 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
1970 | return -EACCES; |
1971 | } | |
d71962f3 PC |
1972 | err = check_helper_mem_access(env, regno, |
1973 | meta->map_ptr->key_size, false, | |
1974 | NULL); | |
17a52670 AS |
1975 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1976 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1977 | * check [value, value + map->value_size) validity | |
1978 | */ | |
33ff9823 | 1979 | if (!meta->map_ptr) { |
17a52670 | 1980 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 1981 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
1982 | return -EACCES; |
1983 | } | |
d71962f3 PC |
1984 | err = check_helper_mem_access(env, regno, |
1985 | meta->map_ptr->value_size, false, | |
1986 | NULL); | |
90133415 | 1987 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 1988 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 1989 | |
849fa506 YS |
1990 | /* remember the mem_size which may be used later |
1991 | * to refine return values. | |
1992 | */ | |
1993 | meta->msize_smax_value = reg->smax_value; | |
1994 | meta->msize_umax_value = reg->umax_value; | |
1995 | ||
f1174f77 EC |
1996 | /* The register is SCALAR_VALUE; the access check |
1997 | * happens using its boundaries. | |
06c1c049 | 1998 | */ |
f1174f77 | 1999 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
2000 | /* For unprivileged variable accesses, disable raw |
2001 | * mode so that the program is required to | |
2002 | * initialize all the memory that the helper could | |
2003 | * just partially fill up. | |
2004 | */ | |
2005 | meta = NULL; | |
2006 | ||
b03c9f9f | 2007 | if (reg->smin_value < 0) { |
61bd5218 | 2008 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
2009 | regno); |
2010 | return -EACCES; | |
2011 | } | |
06c1c049 | 2012 | |
b03c9f9f | 2013 | if (reg->umin_value == 0) { |
f1174f77 EC |
2014 | err = check_helper_mem_access(env, regno - 1, 0, |
2015 | zero_size_allowed, | |
2016 | meta); | |
06c1c049 GB |
2017 | if (err) |
2018 | return err; | |
06c1c049 | 2019 | } |
f1174f77 | 2020 | |
b03c9f9f | 2021 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 2022 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
2023 | regno); |
2024 | return -EACCES; | |
2025 | } | |
2026 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 2027 | reg->umax_value, |
f1174f77 | 2028 | zero_size_allowed, meta); |
17a52670 AS |
2029 | } |
2030 | ||
2031 | return err; | |
6841de8b | 2032 | err_type: |
61bd5218 | 2033 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
2034 | reg_type_str[type], reg_type_str[expected_type]); |
2035 | return -EACCES; | |
17a52670 AS |
2036 | } |
2037 | ||
61bd5218 JK |
2038 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
2039 | struct bpf_map *map, int func_id) | |
35578d79 | 2040 | { |
35578d79 KX |
2041 | if (!map) |
2042 | return 0; | |
2043 | ||
6aff67c8 AS |
2044 | /* We need a two way check, first is from map perspective ... */ |
2045 | switch (map->map_type) { | |
2046 | case BPF_MAP_TYPE_PROG_ARRAY: | |
2047 | if (func_id != BPF_FUNC_tail_call) | |
2048 | goto error; | |
2049 | break; | |
2050 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
2051 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
2052 | func_id != BPF_FUNC_perf_event_output && |
2053 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
2054 | goto error; |
2055 | break; | |
2056 | case BPF_MAP_TYPE_STACK_TRACE: | |
2057 | if (func_id != BPF_FUNC_get_stackid) | |
2058 | goto error; | |
2059 | break; | |
4ed8ec52 | 2060 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 2061 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 2062 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
2063 | goto error; |
2064 | break; | |
546ac1ff JF |
2065 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
2066 | * allow to be modified from bpf side. So do not allow lookup elements | |
2067 | * for now. | |
2068 | */ | |
2069 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 2070 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
2071 | goto error; |
2072 | break; | |
6710e112 JDB |
2073 | /* Restrict bpf side of cpumap, open when use-cases appear */ |
2074 | case BPF_MAP_TYPE_CPUMAP: | |
2075 | if (func_id != BPF_FUNC_redirect_map) | |
2076 | goto error; | |
2077 | break; | |
56f668df | 2078 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 2079 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
2080 | if (func_id != BPF_FUNC_map_lookup_elem) |
2081 | goto error; | |
16a43625 | 2082 | break; |
174a79ff JF |
2083 | case BPF_MAP_TYPE_SOCKMAP: |
2084 | if (func_id != BPF_FUNC_sk_redirect_map && | |
2085 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb JF |
2086 | func_id != BPF_FUNC_map_delete_elem && |
2087 | func_id != BPF_FUNC_msg_redirect_map) | |
174a79ff JF |
2088 | goto error; |
2089 | break; | |
6aff67c8 AS |
2090 | default: |
2091 | break; | |
2092 | } | |
2093 | ||
2094 | /* ... and second from the function itself. */ | |
2095 | switch (func_id) { | |
2096 | case BPF_FUNC_tail_call: | |
2097 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
2098 | goto error; | |
f4d7e40a AS |
2099 | if (env->subprog_cnt) { |
2100 | verbose(env, "tail_calls are not allowed in programs with bpf-to-bpf calls\n"); | |
2101 | return -EINVAL; | |
2102 | } | |
6aff67c8 AS |
2103 | break; |
2104 | case BPF_FUNC_perf_event_read: | |
2105 | case BPF_FUNC_perf_event_output: | |
908432ca | 2106 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
2107 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
2108 | goto error; | |
2109 | break; | |
2110 | case BPF_FUNC_get_stackid: | |
2111 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
2112 | goto error; | |
2113 | break; | |
60d20f91 | 2114 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 2115 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
2116 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
2117 | goto error; | |
2118 | break; | |
97f91a7c | 2119 | case BPF_FUNC_redirect_map: |
9c270af3 JDB |
2120 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
2121 | map->map_type != BPF_MAP_TYPE_CPUMAP) | |
97f91a7c JF |
2122 | goto error; |
2123 | break; | |
174a79ff | 2124 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 2125 | case BPF_FUNC_msg_redirect_map: |
174a79ff JF |
2126 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
2127 | goto error; | |
2128 | break; | |
2129 | case BPF_FUNC_sock_map_update: | |
2130 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) | |
2131 | goto error; | |
2132 | break; | |
6aff67c8 AS |
2133 | default: |
2134 | break; | |
35578d79 KX |
2135 | } |
2136 | ||
2137 | return 0; | |
6aff67c8 | 2138 | error: |
61bd5218 | 2139 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 2140 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 2141 | return -EINVAL; |
35578d79 KX |
2142 | } |
2143 | ||
90133415 | 2144 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
2145 | { |
2146 | int count = 0; | |
2147 | ||
39f19ebb | 2148 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2149 | count++; |
39f19ebb | 2150 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2151 | count++; |
39f19ebb | 2152 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2153 | count++; |
39f19ebb | 2154 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2155 | count++; |
39f19ebb | 2156 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
2157 | count++; |
2158 | ||
90133415 DB |
2159 | /* We only support one arg being in raw mode at the moment, |
2160 | * which is sufficient for the helper functions we have | |
2161 | * right now. | |
2162 | */ | |
2163 | return count <= 1; | |
2164 | } | |
2165 | ||
2166 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
2167 | enum bpf_arg_type arg_next) | |
2168 | { | |
2169 | return (arg_type_is_mem_ptr(arg_curr) && | |
2170 | !arg_type_is_mem_size(arg_next)) || | |
2171 | (!arg_type_is_mem_ptr(arg_curr) && | |
2172 | arg_type_is_mem_size(arg_next)); | |
2173 | } | |
2174 | ||
2175 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
2176 | { | |
2177 | /* bpf_xxx(..., buf, len) call will access 'len' | |
2178 | * bytes from memory 'buf'. Both arg types need | |
2179 | * to be paired, so make sure there's no buggy | |
2180 | * helper function specification. | |
2181 | */ | |
2182 | if (arg_type_is_mem_size(fn->arg1_type) || | |
2183 | arg_type_is_mem_ptr(fn->arg5_type) || | |
2184 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
2185 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
2186 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
2187 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
2188 | return false; | |
2189 | ||
2190 | return true; | |
2191 | } | |
2192 | ||
2193 | static int check_func_proto(const struct bpf_func_proto *fn) | |
2194 | { | |
2195 | return check_raw_mode_ok(fn) && | |
2196 | check_arg_pair_ok(fn) ? 0 : -EINVAL; | |
435faee1 DB |
2197 | } |
2198 | ||
de8f3a83 DB |
2199 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
2200 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 2201 | */ |
f4d7e40a AS |
2202 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
2203 | struct bpf_func_state *state) | |
969bf05e | 2204 | { |
58e2af8b | 2205 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
2206 | int i; |
2207 | ||
2208 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 2209 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 2210 | mark_reg_unknown(env, regs, i); |
969bf05e | 2211 | |
638f5b90 AS |
2212 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
2213 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
969bf05e | 2214 | continue; |
638f5b90 | 2215 | reg = &state->stack[i].spilled_ptr; |
de8f3a83 DB |
2216 | if (reg_is_pkt_pointer_any(reg)) |
2217 | __mark_reg_unknown(reg); | |
969bf05e AS |
2218 | } |
2219 | } | |
2220 | ||
f4d7e40a AS |
2221 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
2222 | { | |
2223 | struct bpf_verifier_state *vstate = env->cur_state; | |
2224 | int i; | |
2225 | ||
2226 | for (i = 0; i <= vstate->curframe; i++) | |
2227 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
2228 | } | |
2229 | ||
2230 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
2231 | int *insn_idx) | |
2232 | { | |
2233 | struct bpf_verifier_state *state = env->cur_state; | |
2234 | struct bpf_func_state *caller, *callee; | |
2235 | int i, subprog, target_insn; | |
2236 | ||
aada9ce6 | 2237 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 2238 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 2239 | state->curframe + 2); |
f4d7e40a AS |
2240 | return -E2BIG; |
2241 | } | |
2242 | ||
2243 | target_insn = *insn_idx + insn->imm; | |
2244 | subprog = find_subprog(env, target_insn + 1); | |
2245 | if (subprog < 0) { | |
2246 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
2247 | target_insn + 1); | |
2248 | return -EFAULT; | |
2249 | } | |
2250 | ||
2251 | caller = state->frame[state->curframe]; | |
2252 | if (state->frame[state->curframe + 1]) { | |
2253 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
2254 | state->curframe + 1); | |
2255 | return -EFAULT; | |
2256 | } | |
2257 | ||
2258 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); | |
2259 | if (!callee) | |
2260 | return -ENOMEM; | |
2261 | state->frame[state->curframe + 1] = callee; | |
2262 | ||
2263 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
2264 | * into its own stack before reading from it. | |
2265 | * callee can read/write into caller's stack | |
2266 | */ | |
2267 | init_func_state(env, callee, | |
2268 | /* remember the callsite, it will be used by bpf_exit */ | |
2269 | *insn_idx /* callsite */, | |
2270 | state->curframe + 1 /* frameno within this callchain */, | |
2271 | subprog + 1 /* subprog number within this prog */); | |
2272 | ||
2273 | /* copy r1 - r5 args that callee can access */ | |
2274 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) | |
2275 | callee->regs[i] = caller->regs[i]; | |
2276 | ||
2277 | /* after the call regsiters r0 - r5 were scratched */ | |
2278 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
2279 | mark_reg_not_init(env, caller->regs, caller_saved[i]); | |
2280 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
2281 | } | |
2282 | ||
2283 | /* only increment it after check_reg_arg() finished */ | |
2284 | state->curframe++; | |
2285 | ||
2286 | /* and go analyze first insn of the callee */ | |
2287 | *insn_idx = target_insn; | |
2288 | ||
2289 | if (env->log.level) { | |
2290 | verbose(env, "caller:\n"); | |
2291 | print_verifier_state(env, caller); | |
2292 | verbose(env, "callee:\n"); | |
2293 | print_verifier_state(env, callee); | |
2294 | } | |
2295 | return 0; | |
2296 | } | |
2297 | ||
2298 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) | |
2299 | { | |
2300 | struct bpf_verifier_state *state = env->cur_state; | |
2301 | struct bpf_func_state *caller, *callee; | |
2302 | struct bpf_reg_state *r0; | |
2303 | ||
2304 | callee = state->frame[state->curframe]; | |
2305 | r0 = &callee->regs[BPF_REG_0]; | |
2306 | if (r0->type == PTR_TO_STACK) { | |
2307 | /* technically it's ok to return caller's stack pointer | |
2308 | * (or caller's caller's pointer) back to the caller, | |
2309 | * since these pointers are valid. Only current stack | |
2310 | * pointer will be invalid as soon as function exits, | |
2311 | * but let's be conservative | |
2312 | */ | |
2313 | verbose(env, "cannot return stack pointer to the caller\n"); | |
2314 | return -EINVAL; | |
2315 | } | |
2316 | ||
2317 | state->curframe--; | |
2318 | caller = state->frame[state->curframe]; | |
2319 | /* return to the caller whatever r0 had in the callee */ | |
2320 | caller->regs[BPF_REG_0] = *r0; | |
2321 | ||
2322 | *insn_idx = callee->callsite + 1; | |
2323 | if (env->log.level) { | |
2324 | verbose(env, "returning from callee:\n"); | |
2325 | print_verifier_state(env, callee); | |
2326 | verbose(env, "to caller at %d:\n", *insn_idx); | |
2327 | print_verifier_state(env, caller); | |
2328 | } | |
2329 | /* clear everything in the callee */ | |
2330 | free_func_state(callee); | |
2331 | state->frame[state->curframe + 1] = NULL; | |
2332 | return 0; | |
2333 | } | |
2334 | ||
849fa506 YS |
2335 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
2336 | int func_id, | |
2337 | struct bpf_call_arg_meta *meta) | |
2338 | { | |
2339 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
2340 | ||
2341 | if (ret_type != RET_INTEGER || | |
2342 | (func_id != BPF_FUNC_get_stack && | |
2343 | func_id != BPF_FUNC_probe_read_str)) | |
2344 | return; | |
2345 | ||
2346 | ret_reg->smax_value = meta->msize_smax_value; | |
2347 | ret_reg->umax_value = meta->msize_umax_value; | |
2348 | __reg_deduce_bounds(ret_reg); | |
2349 | __reg_bound_offset(ret_reg); | |
2350 | } | |
2351 | ||
f4d7e40a | 2352 | static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 2353 | { |
17a52670 | 2354 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 2355 | struct bpf_reg_state *regs; |
33ff9823 | 2356 | struct bpf_call_arg_meta meta; |
969bf05e | 2357 | bool changes_data; |
17a52670 AS |
2358 | int i, err; |
2359 | ||
2360 | /* find function prototype */ | |
2361 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
2362 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
2363 | func_id); | |
17a52670 AS |
2364 | return -EINVAL; |
2365 | } | |
2366 | ||
00176a34 | 2367 | if (env->ops->get_func_proto) |
5e43f899 | 2368 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 2369 | if (!fn) { |
61bd5218 JK |
2370 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
2371 | func_id); | |
17a52670 AS |
2372 | return -EINVAL; |
2373 | } | |
2374 | ||
2375 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 2376 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
61bd5218 | 2377 | verbose(env, "cannot call GPL only function from proprietary program\n"); |
17a52670 AS |
2378 | return -EINVAL; |
2379 | } | |
2380 | ||
04514d13 | 2381 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 2382 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
2383 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
2384 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
2385 | func_id_name(func_id), func_id); | |
2386 | return -EINVAL; | |
2387 | } | |
969bf05e | 2388 | |
33ff9823 | 2389 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 2390 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 2391 | |
90133415 | 2392 | err = check_func_proto(fn); |
435faee1 | 2393 | if (err) { |
61bd5218 | 2394 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 2395 | func_id_name(func_id), func_id); |
435faee1 DB |
2396 | return err; |
2397 | } | |
2398 | ||
17a52670 | 2399 | /* check args */ |
33ff9823 | 2400 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
2401 | if (err) |
2402 | return err; | |
33ff9823 | 2403 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
2404 | if (err) |
2405 | return err; | |
b2157399 AS |
2406 | if (func_id == BPF_FUNC_tail_call) { |
2407 | if (meta.map_ptr == NULL) { | |
2408 | verbose(env, "verifier bug\n"); | |
2409 | return -EINVAL; | |
2410 | } | |
2411 | env->insn_aux_data[insn_idx].map_ptr = meta.map_ptr; | |
2412 | } | |
33ff9823 | 2413 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
2414 | if (err) |
2415 | return err; | |
33ff9823 | 2416 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
2417 | if (err) |
2418 | return err; | |
33ff9823 | 2419 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
2420 | if (err) |
2421 | return err; | |
2422 | ||
435faee1 DB |
2423 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
2424 | * is inferred from register state. | |
2425 | */ | |
2426 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
2427 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
2428 | BPF_WRITE, -1, false); | |
435faee1 DB |
2429 | if (err) |
2430 | return err; | |
2431 | } | |
2432 | ||
638f5b90 | 2433 | regs = cur_regs(env); |
17a52670 | 2434 | /* reset caller saved regs */ |
dc503a8a | 2435 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 2436 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
2437 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
2438 | } | |
17a52670 | 2439 | |
dc503a8a | 2440 | /* update return register (already marked as written above) */ |
17a52670 | 2441 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 2442 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 2443 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
2444 | } else if (fn->ret_type == RET_VOID) { |
2445 | regs[BPF_REG_0].type = NOT_INIT; | |
2446 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
2447 | struct bpf_insn_aux_data *insn_aux; |
2448 | ||
17a52670 | 2449 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
f1174f77 | 2450 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 2451 | mark_reg_known_zero(env, regs, BPF_REG_0); |
f1174f77 | 2452 | regs[BPF_REG_0].off = 0; |
17a52670 AS |
2453 | /* remember map_ptr, so that check_map_access() |
2454 | * can check 'value_size' boundary of memory access | |
2455 | * to map element returned from bpf_map_lookup_elem() | |
2456 | */ | |
33ff9823 | 2457 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
2458 | verbose(env, |
2459 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
2460 | return -EINVAL; |
2461 | } | |
33ff9823 | 2462 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 2463 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
2464 | insn_aux = &env->insn_aux_data[insn_idx]; |
2465 | if (!insn_aux->map_ptr) | |
2466 | insn_aux->map_ptr = meta.map_ptr; | |
2467 | else if (insn_aux->map_ptr != meta.map_ptr) | |
2468 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 2469 | } else { |
61bd5218 | 2470 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 2471 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
2472 | return -EINVAL; |
2473 | } | |
04fd61ab | 2474 | |
849fa506 YS |
2475 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
2476 | ||
61bd5218 | 2477 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
2478 | if (err) |
2479 | return err; | |
04fd61ab | 2480 | |
c195651e YS |
2481 | if (func_id == BPF_FUNC_get_stack && !env->prog->has_callchain_buf) { |
2482 | const char *err_str; | |
2483 | ||
2484 | #ifdef CONFIG_PERF_EVENTS | |
2485 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
2486 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
2487 | #else | |
2488 | err = -ENOTSUPP; | |
2489 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
2490 | #endif | |
2491 | if (err) { | |
2492 | verbose(env, err_str, func_id_name(func_id), func_id); | |
2493 | return err; | |
2494 | } | |
2495 | ||
2496 | env->prog->has_callchain_buf = true; | |
2497 | } | |
2498 | ||
969bf05e AS |
2499 | if (changes_data) |
2500 | clear_all_pkt_pointers(env); | |
2501 | return 0; | |
2502 | } | |
2503 | ||
b03c9f9f EC |
2504 | static bool signed_add_overflows(s64 a, s64 b) |
2505 | { | |
2506 | /* Do the add in u64, where overflow is well-defined */ | |
2507 | s64 res = (s64)((u64)a + (u64)b); | |
2508 | ||
2509 | if (b < 0) | |
2510 | return res > a; | |
2511 | return res < a; | |
2512 | } | |
2513 | ||
2514 | static bool signed_sub_overflows(s64 a, s64 b) | |
2515 | { | |
2516 | /* Do the sub in u64, where overflow is well-defined */ | |
2517 | s64 res = (s64)((u64)a - (u64)b); | |
2518 | ||
2519 | if (b < 0) | |
2520 | return res < a; | |
2521 | return res > a; | |
969bf05e AS |
2522 | } |
2523 | ||
bb7f0f98 AS |
2524 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
2525 | const struct bpf_reg_state *reg, | |
2526 | enum bpf_reg_type type) | |
2527 | { | |
2528 | bool known = tnum_is_const(reg->var_off); | |
2529 | s64 val = reg->var_off.value; | |
2530 | s64 smin = reg->smin_value; | |
2531 | ||
2532 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
2533 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
2534 | reg_type_str[type], val); | |
2535 | return false; | |
2536 | } | |
2537 | ||
2538 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
2539 | verbose(env, "%s pointer offset %d is not allowed\n", | |
2540 | reg_type_str[type], reg->off); | |
2541 | return false; | |
2542 | } | |
2543 | ||
2544 | if (smin == S64_MIN) { | |
2545 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
2546 | reg_type_str[type]); | |
2547 | return false; | |
2548 | } | |
2549 | ||
2550 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
2551 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
2552 | smin, reg_type_str[type]); | |
2553 | return false; | |
2554 | } | |
2555 | ||
2556 | return true; | |
2557 | } | |
2558 | ||
f1174f77 | 2559 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
2560 | * Caller should also handle BPF_MOV case separately. |
2561 | * If we return -EACCES, caller may want to try again treating pointer as a | |
2562 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
2563 | */ | |
2564 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
2565 | struct bpf_insn *insn, | |
2566 | const struct bpf_reg_state *ptr_reg, | |
2567 | const struct bpf_reg_state *off_reg) | |
969bf05e | 2568 | { |
f4d7e40a AS |
2569 | struct bpf_verifier_state *vstate = env->cur_state; |
2570 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
2571 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 2572 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
2573 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
2574 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
2575 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
2576 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
969bf05e | 2577 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2578 | u32 dst = insn->dst_reg; |
969bf05e | 2579 | |
f1174f77 | 2580 | dst_reg = ®s[dst]; |
969bf05e | 2581 | |
6f16101e DB |
2582 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
2583 | smin_val > smax_val || umin_val > umax_val) { | |
2584 | /* Taint dst register if offset had invalid bounds derived from | |
2585 | * e.g. dead branches. | |
2586 | */ | |
2587 | __mark_reg_unknown(dst_reg); | |
2588 | return 0; | |
f1174f77 EC |
2589 | } |
2590 | ||
2591 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
2592 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
2593 | verbose(env, |
2594 | "R%d 32-bit pointer arithmetic prohibited\n", | |
2595 | dst); | |
f1174f77 | 2596 | return -EACCES; |
969bf05e AS |
2597 | } |
2598 | ||
f1174f77 | 2599 | if (ptr_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
82abbf8d AS |
2600 | verbose(env, "R%d pointer arithmetic on PTR_TO_MAP_VALUE_OR_NULL prohibited, null-check it first\n", |
2601 | dst); | |
f1174f77 EC |
2602 | return -EACCES; |
2603 | } | |
2604 | if (ptr_reg->type == CONST_PTR_TO_MAP) { | |
82abbf8d AS |
2605 | verbose(env, "R%d pointer arithmetic on CONST_PTR_TO_MAP prohibited\n", |
2606 | dst); | |
f1174f77 EC |
2607 | return -EACCES; |
2608 | } | |
2609 | if (ptr_reg->type == PTR_TO_PACKET_END) { | |
82abbf8d AS |
2610 | verbose(env, "R%d pointer arithmetic on PTR_TO_PACKET_END prohibited\n", |
2611 | dst); | |
f1174f77 EC |
2612 | return -EACCES; |
2613 | } | |
2614 | ||
2615 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
2616 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 2617 | */ |
f1174f77 EC |
2618 | dst_reg->type = ptr_reg->type; |
2619 | dst_reg->id = ptr_reg->id; | |
969bf05e | 2620 | |
bb7f0f98 AS |
2621 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
2622 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
2623 | return -EINVAL; | |
2624 | ||
f1174f77 EC |
2625 | switch (opcode) { |
2626 | case BPF_ADD: | |
2627 | /* We can take a fixed offset as long as it doesn't overflow | |
2628 | * the s32 'off' field | |
969bf05e | 2629 | */ |
b03c9f9f EC |
2630 | if (known && (ptr_reg->off + smin_val == |
2631 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 2632 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
2633 | dst_reg->smin_value = smin_ptr; |
2634 | dst_reg->smax_value = smax_ptr; | |
2635 | dst_reg->umin_value = umin_ptr; | |
2636 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 2637 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 2638 | dst_reg->off = ptr_reg->off + smin_val; |
f1174f77 EC |
2639 | dst_reg->range = ptr_reg->range; |
2640 | break; | |
2641 | } | |
f1174f77 EC |
2642 | /* A new variable offset is created. Note that off_reg->off |
2643 | * == 0, since it's a scalar. | |
2644 | * dst_reg gets the pointer type and since some positive | |
2645 | * integer value was added to the pointer, give it a new 'id' | |
2646 | * if it's a PTR_TO_PACKET. | |
2647 | * this creates a new 'base' pointer, off_reg (variable) gets | |
2648 | * added into the variable offset, and we copy the fixed offset | |
2649 | * from ptr_reg. | |
969bf05e | 2650 | */ |
b03c9f9f EC |
2651 | if (signed_add_overflows(smin_ptr, smin_val) || |
2652 | signed_add_overflows(smax_ptr, smax_val)) { | |
2653 | dst_reg->smin_value = S64_MIN; | |
2654 | dst_reg->smax_value = S64_MAX; | |
2655 | } else { | |
2656 | dst_reg->smin_value = smin_ptr + smin_val; | |
2657 | dst_reg->smax_value = smax_ptr + smax_val; | |
2658 | } | |
2659 | if (umin_ptr + umin_val < umin_ptr || | |
2660 | umax_ptr + umax_val < umax_ptr) { | |
2661 | dst_reg->umin_value = 0; | |
2662 | dst_reg->umax_value = U64_MAX; | |
2663 | } else { | |
2664 | dst_reg->umin_value = umin_ptr + umin_val; | |
2665 | dst_reg->umax_value = umax_ptr + umax_val; | |
2666 | } | |
f1174f77 EC |
2667 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
2668 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 2669 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2670 | dst_reg->id = ++env->id_gen; |
2671 | /* something was added to pkt_ptr, set range to zero */ | |
2672 | dst_reg->range = 0; | |
2673 | } | |
2674 | break; | |
2675 | case BPF_SUB: | |
2676 | if (dst_reg == off_reg) { | |
2677 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
2678 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
2679 | dst); | |
f1174f77 EC |
2680 | return -EACCES; |
2681 | } | |
2682 | /* We don't allow subtraction from FP, because (according to | |
2683 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
2684 | * be able to deal with it. | |
969bf05e | 2685 | */ |
f1174f77 | 2686 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
2687 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
2688 | dst); | |
f1174f77 EC |
2689 | return -EACCES; |
2690 | } | |
b03c9f9f EC |
2691 | if (known && (ptr_reg->off - smin_val == |
2692 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 2693 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
2694 | dst_reg->smin_value = smin_ptr; |
2695 | dst_reg->smax_value = smax_ptr; | |
2696 | dst_reg->umin_value = umin_ptr; | |
2697 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
2698 | dst_reg->var_off = ptr_reg->var_off; |
2699 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 2700 | dst_reg->off = ptr_reg->off - smin_val; |
f1174f77 EC |
2701 | dst_reg->range = ptr_reg->range; |
2702 | break; | |
2703 | } | |
f1174f77 EC |
2704 | /* A new variable offset is created. If the subtrahend is known |
2705 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 2706 | */ |
b03c9f9f EC |
2707 | if (signed_sub_overflows(smin_ptr, smax_val) || |
2708 | signed_sub_overflows(smax_ptr, smin_val)) { | |
2709 | /* Overflow possible, we know nothing */ | |
2710 | dst_reg->smin_value = S64_MIN; | |
2711 | dst_reg->smax_value = S64_MAX; | |
2712 | } else { | |
2713 | dst_reg->smin_value = smin_ptr - smax_val; | |
2714 | dst_reg->smax_value = smax_ptr - smin_val; | |
2715 | } | |
2716 | if (umin_ptr < umax_val) { | |
2717 | /* Overflow possible, we know nothing */ | |
2718 | dst_reg->umin_value = 0; | |
2719 | dst_reg->umax_value = U64_MAX; | |
2720 | } else { | |
2721 | /* Cannot overflow (as long as bounds are consistent) */ | |
2722 | dst_reg->umin_value = umin_ptr - umax_val; | |
2723 | dst_reg->umax_value = umax_ptr - umin_val; | |
2724 | } | |
f1174f77 EC |
2725 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
2726 | dst_reg->off = ptr_reg->off; | |
de8f3a83 | 2727 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
2728 | dst_reg->id = ++env->id_gen; |
2729 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 2730 | if (smin_val < 0) |
f1174f77 | 2731 | dst_reg->range = 0; |
43188702 | 2732 | } |
f1174f77 EC |
2733 | break; |
2734 | case BPF_AND: | |
2735 | case BPF_OR: | |
2736 | case BPF_XOR: | |
82abbf8d AS |
2737 | /* bitwise ops on pointers are troublesome, prohibit. */ |
2738 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
2739 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
2740 | return -EACCES; |
2741 | default: | |
2742 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
2743 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
2744 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 2745 | return -EACCES; |
43188702 JF |
2746 | } |
2747 | ||
bb7f0f98 AS |
2748 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
2749 | return -EINVAL; | |
2750 | ||
b03c9f9f EC |
2751 | __update_reg_bounds(dst_reg); |
2752 | __reg_deduce_bounds(dst_reg); | |
2753 | __reg_bound_offset(dst_reg); | |
43188702 JF |
2754 | return 0; |
2755 | } | |
2756 | ||
468f6eaf JH |
2757 | /* WARNING: This function does calculations on 64-bit values, but the actual |
2758 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
2759 | * need extra checks in the 32-bit case. | |
2760 | */ | |
f1174f77 EC |
2761 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
2762 | struct bpf_insn *insn, | |
2763 | struct bpf_reg_state *dst_reg, | |
2764 | struct bpf_reg_state src_reg) | |
969bf05e | 2765 | { |
638f5b90 | 2766 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 2767 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 2768 | bool src_known, dst_known; |
b03c9f9f EC |
2769 | s64 smin_val, smax_val; |
2770 | u64 umin_val, umax_val; | |
468f6eaf | 2771 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
48461135 | 2772 | |
b03c9f9f EC |
2773 | smin_val = src_reg.smin_value; |
2774 | smax_val = src_reg.smax_value; | |
2775 | umin_val = src_reg.umin_value; | |
2776 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
2777 | src_known = tnum_is_const(src_reg.var_off); |
2778 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 2779 | |
6f16101e DB |
2780 | if ((src_known && (smin_val != smax_val || umin_val != umax_val)) || |
2781 | smin_val > smax_val || umin_val > umax_val) { | |
2782 | /* Taint dst register if offset had invalid bounds derived from | |
2783 | * e.g. dead branches. | |
2784 | */ | |
2785 | __mark_reg_unknown(dst_reg); | |
2786 | return 0; | |
2787 | } | |
2788 | ||
bb7f0f98 AS |
2789 | if (!src_known && |
2790 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
2791 | __mark_reg_unknown(dst_reg); | |
2792 | return 0; | |
2793 | } | |
2794 | ||
48461135 JB |
2795 | switch (opcode) { |
2796 | case BPF_ADD: | |
b03c9f9f EC |
2797 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
2798 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
2799 | dst_reg->smin_value = S64_MIN; | |
2800 | dst_reg->smax_value = S64_MAX; | |
2801 | } else { | |
2802 | dst_reg->smin_value += smin_val; | |
2803 | dst_reg->smax_value += smax_val; | |
2804 | } | |
2805 | if (dst_reg->umin_value + umin_val < umin_val || | |
2806 | dst_reg->umax_value + umax_val < umax_val) { | |
2807 | dst_reg->umin_value = 0; | |
2808 | dst_reg->umax_value = U64_MAX; | |
2809 | } else { | |
2810 | dst_reg->umin_value += umin_val; | |
2811 | dst_reg->umax_value += umax_val; | |
2812 | } | |
f1174f77 | 2813 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2814 | break; |
2815 | case BPF_SUB: | |
b03c9f9f EC |
2816 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
2817 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
2818 | /* Overflow possible, we know nothing */ | |
2819 | dst_reg->smin_value = S64_MIN; | |
2820 | dst_reg->smax_value = S64_MAX; | |
2821 | } else { | |
2822 | dst_reg->smin_value -= smax_val; | |
2823 | dst_reg->smax_value -= smin_val; | |
2824 | } | |
2825 | if (dst_reg->umin_value < umax_val) { | |
2826 | /* Overflow possible, we know nothing */ | |
2827 | dst_reg->umin_value = 0; | |
2828 | dst_reg->umax_value = U64_MAX; | |
2829 | } else { | |
2830 | /* Cannot overflow (as long as bounds are consistent) */ | |
2831 | dst_reg->umin_value -= umax_val; | |
2832 | dst_reg->umax_value -= umin_val; | |
2833 | } | |
f1174f77 | 2834 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
2835 | break; |
2836 | case BPF_MUL: | |
b03c9f9f EC |
2837 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
2838 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 2839 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
2840 | __mark_reg_unbounded(dst_reg); |
2841 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2842 | break; |
2843 | } | |
b03c9f9f EC |
2844 | /* Both values are positive, so we can work with unsigned and |
2845 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 2846 | */ |
b03c9f9f EC |
2847 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
2848 | /* Potential overflow, we know nothing */ | |
2849 | __mark_reg_unbounded(dst_reg); | |
2850 | /* (except what we can learn from the var_off) */ | |
2851 | __update_reg_bounds(dst_reg); | |
2852 | break; | |
2853 | } | |
2854 | dst_reg->umin_value *= umin_val; | |
2855 | dst_reg->umax_value *= umax_val; | |
2856 | if (dst_reg->umax_value > S64_MAX) { | |
2857 | /* Overflow possible, we know nothing */ | |
2858 | dst_reg->smin_value = S64_MIN; | |
2859 | dst_reg->smax_value = S64_MAX; | |
2860 | } else { | |
2861 | dst_reg->smin_value = dst_reg->umin_value; | |
2862 | dst_reg->smax_value = dst_reg->umax_value; | |
2863 | } | |
48461135 JB |
2864 | break; |
2865 | case BPF_AND: | |
f1174f77 | 2866 | if (src_known && dst_known) { |
b03c9f9f EC |
2867 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
2868 | src_reg.var_off.value); | |
f1174f77 EC |
2869 | break; |
2870 | } | |
b03c9f9f EC |
2871 | /* We get our minimum from the var_off, since that's inherently |
2872 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 2873 | */ |
f1174f77 | 2874 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
2875 | dst_reg->umin_value = dst_reg->var_off.value; |
2876 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
2877 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2878 | /* Lose signed bounds when ANDing negative numbers, | |
2879 | * ain't nobody got time for that. | |
2880 | */ | |
2881 | dst_reg->smin_value = S64_MIN; | |
2882 | dst_reg->smax_value = S64_MAX; | |
2883 | } else { | |
2884 | /* ANDing two positives gives a positive, so safe to | |
2885 | * cast result into s64. | |
2886 | */ | |
2887 | dst_reg->smin_value = dst_reg->umin_value; | |
2888 | dst_reg->smax_value = dst_reg->umax_value; | |
2889 | } | |
2890 | /* We may learn something more from the var_off */ | |
2891 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
2892 | break; |
2893 | case BPF_OR: | |
2894 | if (src_known && dst_known) { | |
b03c9f9f EC |
2895 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
2896 | src_reg.var_off.value); | |
f1174f77 EC |
2897 | break; |
2898 | } | |
b03c9f9f EC |
2899 | /* We get our maximum from the var_off, and our minimum is the |
2900 | * maximum of the operands' minima | |
f1174f77 EC |
2901 | */ |
2902 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
2903 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
2904 | dst_reg->umax_value = dst_reg->var_off.value | | |
2905 | dst_reg->var_off.mask; | |
2906 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
2907 | /* Lose signed bounds when ORing negative numbers, | |
2908 | * ain't nobody got time for that. | |
2909 | */ | |
2910 | dst_reg->smin_value = S64_MIN; | |
2911 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 2912 | } else { |
b03c9f9f EC |
2913 | /* ORing two positives gives a positive, so safe to |
2914 | * cast result into s64. | |
2915 | */ | |
2916 | dst_reg->smin_value = dst_reg->umin_value; | |
2917 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 2918 | } |
b03c9f9f EC |
2919 | /* We may learn something more from the var_off */ |
2920 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2921 | break; |
2922 | case BPF_LSH: | |
468f6eaf JH |
2923 | if (umax_val >= insn_bitness) { |
2924 | /* Shifts greater than 31 or 63 are undefined. | |
2925 | * This includes shifts by a negative number. | |
b03c9f9f | 2926 | */ |
61bd5218 | 2927 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2928 | break; |
2929 | } | |
b03c9f9f EC |
2930 | /* We lose all sign bit information (except what we can pick |
2931 | * up from var_off) | |
48461135 | 2932 | */ |
b03c9f9f EC |
2933 | dst_reg->smin_value = S64_MIN; |
2934 | dst_reg->smax_value = S64_MAX; | |
2935 | /* If we might shift our top bit out, then we know nothing */ | |
2936 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
2937 | dst_reg->umin_value = 0; | |
2938 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 2939 | } else { |
b03c9f9f EC |
2940 | dst_reg->umin_value <<= umin_val; |
2941 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 2942 | } |
afbe1a5b | 2943 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
2944 | /* We may learn something more from the var_off */ |
2945 | __update_reg_bounds(dst_reg); | |
48461135 JB |
2946 | break; |
2947 | case BPF_RSH: | |
468f6eaf JH |
2948 | if (umax_val >= insn_bitness) { |
2949 | /* Shifts greater than 31 or 63 are undefined. | |
2950 | * This includes shifts by a negative number. | |
b03c9f9f | 2951 | */ |
61bd5218 | 2952 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
2953 | break; |
2954 | } | |
4374f256 EC |
2955 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
2956 | * be negative, then either: | |
2957 | * 1) src_reg might be zero, so the sign bit of the result is | |
2958 | * unknown, so we lose our signed bounds | |
2959 | * 2) it's known negative, thus the unsigned bounds capture the | |
2960 | * signed bounds | |
2961 | * 3) the signed bounds cross zero, so they tell us nothing | |
2962 | * about the result | |
2963 | * If the value in dst_reg is known nonnegative, then again the | |
2964 | * unsigned bounts capture the signed bounds. | |
2965 | * Thus, in all cases it suffices to blow away our signed bounds | |
2966 | * and rely on inferring new ones from the unsigned bounds and | |
2967 | * var_off of the result. | |
2968 | */ | |
2969 | dst_reg->smin_value = S64_MIN; | |
2970 | dst_reg->smax_value = S64_MAX; | |
afbe1a5b | 2971 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
2972 | dst_reg->umin_value >>= umax_val; |
2973 | dst_reg->umax_value >>= umin_val; | |
2974 | /* We may learn something more from the var_off */ | |
2975 | __update_reg_bounds(dst_reg); | |
48461135 | 2976 | break; |
9cbe1f5a YS |
2977 | case BPF_ARSH: |
2978 | if (umax_val >= insn_bitness) { | |
2979 | /* Shifts greater than 31 or 63 are undefined. | |
2980 | * This includes shifts by a negative number. | |
2981 | */ | |
2982 | mark_reg_unknown(env, regs, insn->dst_reg); | |
2983 | break; | |
2984 | } | |
2985 | ||
2986 | /* Upon reaching here, src_known is true and | |
2987 | * umax_val is equal to umin_val. | |
2988 | */ | |
2989 | dst_reg->smin_value >>= umin_val; | |
2990 | dst_reg->smax_value >>= umin_val; | |
2991 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val); | |
2992 | ||
2993 | /* blow away the dst_reg umin_value/umax_value and rely on | |
2994 | * dst_reg var_off to refine the result. | |
2995 | */ | |
2996 | dst_reg->umin_value = 0; | |
2997 | dst_reg->umax_value = U64_MAX; | |
2998 | __update_reg_bounds(dst_reg); | |
2999 | break; | |
48461135 | 3000 | default: |
61bd5218 | 3001 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
3002 | break; |
3003 | } | |
3004 | ||
468f6eaf JH |
3005 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
3006 | /* 32-bit ALU ops are (32,32)->32 */ | |
3007 | coerce_reg_to_size(dst_reg, 4); | |
3008 | coerce_reg_to_size(&src_reg, 4); | |
3009 | } | |
3010 | ||
b03c9f9f EC |
3011 | __reg_deduce_bounds(dst_reg); |
3012 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
3013 | return 0; |
3014 | } | |
3015 | ||
3016 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
3017 | * and var_off. | |
3018 | */ | |
3019 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
3020 | struct bpf_insn *insn) | |
3021 | { | |
f4d7e40a AS |
3022 | struct bpf_verifier_state *vstate = env->cur_state; |
3023 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3024 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
3025 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
3026 | u8 opcode = BPF_OP(insn->code); | |
f1174f77 EC |
3027 | |
3028 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
3029 | src_reg = NULL; |
3030 | if (dst_reg->type != SCALAR_VALUE) | |
3031 | ptr_reg = dst_reg; | |
3032 | if (BPF_SRC(insn->code) == BPF_X) { | |
3033 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
3034 | if (src_reg->type != SCALAR_VALUE) { |
3035 | if (dst_reg->type != SCALAR_VALUE) { | |
3036 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
3037 | * an arbitrary scalar. Disallow all math except |
3038 | * pointer subtraction | |
f1174f77 | 3039 | */ |
82abbf8d AS |
3040 | if (opcode == BPF_SUB){ |
3041 | mark_reg_unknown(env, regs, insn->dst_reg); | |
3042 | return 0; | |
f1174f77 | 3043 | } |
82abbf8d AS |
3044 | verbose(env, "R%d pointer %s pointer prohibited\n", |
3045 | insn->dst_reg, | |
3046 | bpf_alu_string[opcode >> 4]); | |
3047 | return -EACCES; | |
f1174f77 EC |
3048 | } else { |
3049 | /* scalar += pointer | |
3050 | * This is legal, but we have to reverse our | |
3051 | * src/dest handling in computing the range | |
3052 | */ | |
82abbf8d AS |
3053 | return adjust_ptr_min_max_vals(env, insn, |
3054 | src_reg, dst_reg); | |
f1174f77 EC |
3055 | } |
3056 | } else if (ptr_reg) { | |
3057 | /* pointer += scalar */ | |
82abbf8d AS |
3058 | return adjust_ptr_min_max_vals(env, insn, |
3059 | dst_reg, src_reg); | |
f1174f77 EC |
3060 | } |
3061 | } else { | |
3062 | /* Pretend the src is a reg with a known value, since we only | |
3063 | * need to be able to read from this state. | |
3064 | */ | |
3065 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 3066 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 3067 | src_reg = &off_reg; |
82abbf8d AS |
3068 | if (ptr_reg) /* pointer += K */ |
3069 | return adjust_ptr_min_max_vals(env, insn, | |
3070 | ptr_reg, src_reg); | |
f1174f77 EC |
3071 | } |
3072 | ||
3073 | /* Got here implies adding two SCALAR_VALUEs */ | |
3074 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 3075 | print_verifier_state(env, state); |
61bd5218 | 3076 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
3077 | return -EINVAL; |
3078 | } | |
3079 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 3080 | print_verifier_state(env, state); |
61bd5218 | 3081 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
3082 | return -EINVAL; |
3083 | } | |
3084 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
3085 | } |
3086 | ||
17a52670 | 3087 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 3088 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3089 | { |
638f5b90 | 3090 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3091 | u8 opcode = BPF_OP(insn->code); |
3092 | int err; | |
3093 | ||
3094 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
3095 | if (opcode == BPF_NEG) { | |
3096 | if (BPF_SRC(insn->code) != 0 || | |
3097 | insn->src_reg != BPF_REG_0 || | |
3098 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 3099 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
3100 | return -EINVAL; |
3101 | } | |
3102 | } else { | |
3103 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
3104 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
3105 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 3106 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
3107 | return -EINVAL; |
3108 | } | |
3109 | } | |
3110 | ||
3111 | /* check src operand */ | |
dc503a8a | 3112 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3113 | if (err) |
3114 | return err; | |
3115 | ||
1be7f75d | 3116 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 3117 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
3118 | insn->dst_reg); |
3119 | return -EACCES; | |
3120 | } | |
3121 | ||
17a52670 | 3122 | /* check dest operand */ |
dc503a8a | 3123 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3124 | if (err) |
3125 | return err; | |
3126 | ||
3127 | } else if (opcode == BPF_MOV) { | |
3128 | ||
3129 | if (BPF_SRC(insn->code) == BPF_X) { | |
3130 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 3131 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
3132 | return -EINVAL; |
3133 | } | |
3134 | ||
3135 | /* check src operand */ | |
dc503a8a | 3136 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3137 | if (err) |
3138 | return err; | |
3139 | } else { | |
3140 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 3141 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
3142 | return -EINVAL; |
3143 | } | |
3144 | } | |
3145 | ||
3146 | /* check dest operand */ | |
dc503a8a | 3147 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3148 | if (err) |
3149 | return err; | |
3150 | ||
3151 | if (BPF_SRC(insn->code) == BPF_X) { | |
3152 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
3153 | /* case: R1 = R2 | |
3154 | * copy register state to dest reg | |
3155 | */ | |
3156 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
8fe2d6cc | 3157 | regs[insn->dst_reg].live |= REG_LIVE_WRITTEN; |
17a52670 | 3158 | } else { |
f1174f77 | 3159 | /* R1 = (u32) R2 */ |
1be7f75d | 3160 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
3161 | verbose(env, |
3162 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
3163 | insn->src_reg); |
3164 | return -EACCES; | |
3165 | } | |
61bd5218 | 3166 | mark_reg_unknown(env, regs, insn->dst_reg); |
0c17d1d2 | 3167 | coerce_reg_to_size(®s[insn->dst_reg], 4); |
17a52670 AS |
3168 | } |
3169 | } else { | |
3170 | /* case: R = imm | |
3171 | * remember the value we stored into this reg | |
3172 | */ | |
f1174f77 | 3173 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
3174 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
3175 | __mark_reg_known(regs + insn->dst_reg, | |
3176 | insn->imm); | |
3177 | } else { | |
3178 | __mark_reg_known(regs + insn->dst_reg, | |
3179 | (u32)insn->imm); | |
3180 | } | |
17a52670 AS |
3181 | } |
3182 | ||
3183 | } else if (opcode > BPF_END) { | |
61bd5218 | 3184 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
3185 | return -EINVAL; |
3186 | ||
3187 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
3188 | ||
17a52670 AS |
3189 | if (BPF_SRC(insn->code) == BPF_X) { |
3190 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 3191 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
3192 | return -EINVAL; |
3193 | } | |
3194 | /* check src1 operand */ | |
dc503a8a | 3195 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3196 | if (err) |
3197 | return err; | |
3198 | } else { | |
3199 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 3200 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
3201 | return -EINVAL; |
3202 | } | |
3203 | } | |
3204 | ||
3205 | /* check src2 operand */ | |
dc503a8a | 3206 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3207 | if (err) |
3208 | return err; | |
3209 | ||
3210 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
3211 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 3212 | verbose(env, "div by zero\n"); |
17a52670 AS |
3213 | return -EINVAL; |
3214 | } | |
3215 | ||
7891a87e DB |
3216 | if (opcode == BPF_ARSH && BPF_CLASS(insn->code) != BPF_ALU64) { |
3217 | verbose(env, "BPF_ARSH not supported for 32 bit ALU\n"); | |
3218 | return -EINVAL; | |
3219 | } | |
3220 | ||
229394e8 RV |
3221 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
3222 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
3223 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
3224 | ||
3225 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 3226 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
3227 | return -EINVAL; |
3228 | } | |
3229 | } | |
3230 | ||
1a0dc1ac | 3231 | /* check dest operand */ |
dc503a8a | 3232 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
3233 | if (err) |
3234 | return err; | |
3235 | ||
f1174f77 | 3236 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
3237 | } |
3238 | ||
3239 | return 0; | |
3240 | } | |
3241 | ||
f4d7e40a | 3242 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 3243 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 3244 | enum bpf_reg_type type, |
fb2a311a | 3245 | bool range_right_open) |
969bf05e | 3246 | { |
f4d7e40a | 3247 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
58e2af8b | 3248 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 3249 | u16 new_range; |
f4d7e40a | 3250 | int i, j; |
2d2be8ca | 3251 | |
fb2a311a DB |
3252 | if (dst_reg->off < 0 || |
3253 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
3254 | /* This doesn't give us any range */ |
3255 | return; | |
3256 | ||
b03c9f9f EC |
3257 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
3258 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
3259 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
3260 | * than pkt_end, but that's because it's also less than pkt. | |
3261 | */ | |
3262 | return; | |
3263 | ||
fb2a311a DB |
3264 | new_range = dst_reg->off; |
3265 | if (range_right_open) | |
3266 | new_range--; | |
3267 | ||
3268 | /* Examples for register markings: | |
2d2be8ca | 3269 | * |
fb2a311a | 3270 | * pkt_data in dst register: |
2d2be8ca DB |
3271 | * |
3272 | * r2 = r3; | |
3273 | * r2 += 8; | |
3274 | * if (r2 > pkt_end) goto <handle exception> | |
3275 | * <access okay> | |
3276 | * | |
b4e432f1 DB |
3277 | * r2 = r3; |
3278 | * r2 += 8; | |
3279 | * if (r2 < pkt_end) goto <access okay> | |
3280 | * <handle exception> | |
3281 | * | |
2d2be8ca DB |
3282 | * Where: |
3283 | * r2 == dst_reg, pkt_end == src_reg | |
3284 | * r2=pkt(id=n,off=8,r=0) | |
3285 | * r3=pkt(id=n,off=0,r=0) | |
3286 | * | |
fb2a311a | 3287 | * pkt_data in src register: |
2d2be8ca DB |
3288 | * |
3289 | * r2 = r3; | |
3290 | * r2 += 8; | |
3291 | * if (pkt_end >= r2) goto <access okay> | |
3292 | * <handle exception> | |
3293 | * | |
b4e432f1 DB |
3294 | * r2 = r3; |
3295 | * r2 += 8; | |
3296 | * if (pkt_end <= r2) goto <handle exception> | |
3297 | * <access okay> | |
3298 | * | |
2d2be8ca DB |
3299 | * Where: |
3300 | * pkt_end == dst_reg, r2 == src_reg | |
3301 | * r2=pkt(id=n,off=8,r=0) | |
3302 | * r3=pkt(id=n,off=0,r=0) | |
3303 | * | |
3304 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
3305 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
3306 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
3307 | * the check. | |
969bf05e | 3308 | */ |
2d2be8ca | 3309 | |
f1174f77 EC |
3310 | /* If our ids match, then we must have the same max_value. And we |
3311 | * don't care about the other reg's fixed offset, since if it's too big | |
3312 | * the range won't allow anything. | |
3313 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
3314 | */ | |
969bf05e | 3315 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 3316 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 3317 | /* keep the maximum range already checked */ |
fb2a311a | 3318 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 3319 | |
f4d7e40a AS |
3320 | for (j = 0; j <= vstate->curframe; j++) { |
3321 | state = vstate->frame[j]; | |
3322 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
3323 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
3324 | continue; | |
3325 | reg = &state->stack[i].spilled_ptr; | |
3326 | if (reg->type == type && reg->id == dst_reg->id) | |
3327 | reg->range = max(reg->range, new_range); | |
3328 | } | |
969bf05e AS |
3329 | } |
3330 | } | |
3331 | ||
48461135 JB |
3332 | /* Adjusts the register min/max values in the case that the dst_reg is the |
3333 | * variable register that we are working on, and src_reg is a constant or we're | |
3334 | * simply doing a BPF_K check. | |
f1174f77 | 3335 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
3336 | */ |
3337 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3338 | struct bpf_reg_state *false_reg, u64 val, | |
3339 | u8 opcode) | |
3340 | { | |
f1174f77 EC |
3341 | /* If the dst_reg is a pointer, we can't learn anything about its |
3342 | * variable offset from the compare (unless src_reg were a pointer into | |
3343 | * the same object, but we don't bother with that. | |
3344 | * Since false_reg and true_reg have the same type by construction, we | |
3345 | * only need to check one of them for pointerness. | |
3346 | */ | |
3347 | if (__is_pointer_value(false, false_reg)) | |
3348 | return; | |
4cabc5b1 | 3349 | |
48461135 JB |
3350 | switch (opcode) { |
3351 | case BPF_JEQ: | |
3352 | /* If this is false then we know nothing Jon Snow, but if it is | |
3353 | * true then we know for sure. | |
3354 | */ | |
b03c9f9f | 3355 | __mark_reg_known(true_reg, val); |
48461135 JB |
3356 | break; |
3357 | case BPF_JNE: | |
3358 | /* If this is true we know nothing Jon Snow, but if it is false | |
3359 | * we know the value for sure; | |
3360 | */ | |
b03c9f9f | 3361 | __mark_reg_known(false_reg, val); |
48461135 JB |
3362 | break; |
3363 | case BPF_JGT: | |
b03c9f9f EC |
3364 | false_reg->umax_value = min(false_reg->umax_value, val); |
3365 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
3366 | break; | |
48461135 | 3367 | case BPF_JSGT: |
b03c9f9f EC |
3368 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
3369 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 3370 | break; |
b4e432f1 DB |
3371 | case BPF_JLT: |
3372 | false_reg->umin_value = max(false_reg->umin_value, val); | |
3373 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
3374 | break; | |
3375 | case BPF_JSLT: | |
3376 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
3377 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
3378 | break; | |
48461135 | 3379 | case BPF_JGE: |
b03c9f9f EC |
3380 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
3381 | true_reg->umin_value = max(true_reg->umin_value, val); | |
3382 | break; | |
48461135 | 3383 | case BPF_JSGE: |
b03c9f9f EC |
3384 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
3385 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 3386 | break; |
b4e432f1 DB |
3387 | case BPF_JLE: |
3388 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
3389 | true_reg->umax_value = min(true_reg->umax_value, val); | |
3390 | break; | |
3391 | case BPF_JSLE: | |
3392 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
3393 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
3394 | break; | |
48461135 JB |
3395 | default: |
3396 | break; | |
3397 | } | |
3398 | ||
b03c9f9f EC |
3399 | __reg_deduce_bounds(false_reg); |
3400 | __reg_deduce_bounds(true_reg); | |
3401 | /* We might have learned some bits from the bounds. */ | |
3402 | __reg_bound_offset(false_reg); | |
3403 | __reg_bound_offset(true_reg); | |
3404 | /* Intersecting with the old var_off might have improved our bounds | |
3405 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3406 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3407 | */ | |
3408 | __update_reg_bounds(false_reg); | |
3409 | __update_reg_bounds(true_reg); | |
48461135 JB |
3410 | } |
3411 | ||
f1174f77 EC |
3412 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
3413 | * the variable reg. | |
48461135 JB |
3414 | */ |
3415 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3416 | struct bpf_reg_state *false_reg, u64 val, | |
3417 | u8 opcode) | |
3418 | { | |
f1174f77 EC |
3419 | if (__is_pointer_value(false, false_reg)) |
3420 | return; | |
4cabc5b1 | 3421 | |
48461135 JB |
3422 | switch (opcode) { |
3423 | case BPF_JEQ: | |
3424 | /* If this is false then we know nothing Jon Snow, but if it is | |
3425 | * true then we know for sure. | |
3426 | */ | |
b03c9f9f | 3427 | __mark_reg_known(true_reg, val); |
48461135 JB |
3428 | break; |
3429 | case BPF_JNE: | |
3430 | /* If this is true we know nothing Jon Snow, but if it is false | |
3431 | * we know the value for sure; | |
3432 | */ | |
b03c9f9f | 3433 | __mark_reg_known(false_reg, val); |
48461135 JB |
3434 | break; |
3435 | case BPF_JGT: | |
b03c9f9f EC |
3436 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
3437 | false_reg->umin_value = max(false_reg->umin_value, val); | |
3438 | break; | |
48461135 | 3439 | case BPF_JSGT: |
b03c9f9f EC |
3440 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
3441 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 3442 | break; |
b4e432f1 DB |
3443 | case BPF_JLT: |
3444 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
3445 | false_reg->umax_value = min(false_reg->umax_value, val); | |
3446 | break; | |
3447 | case BPF_JSLT: | |
3448 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
3449 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
3450 | break; | |
48461135 | 3451 | case BPF_JGE: |
b03c9f9f EC |
3452 | true_reg->umax_value = min(true_reg->umax_value, val); |
3453 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
3454 | break; | |
48461135 | 3455 | case BPF_JSGE: |
b03c9f9f EC |
3456 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
3457 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 3458 | break; |
b4e432f1 DB |
3459 | case BPF_JLE: |
3460 | true_reg->umin_value = max(true_reg->umin_value, val); | |
3461 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
3462 | break; | |
3463 | case BPF_JSLE: | |
3464 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
3465 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
3466 | break; | |
48461135 JB |
3467 | default: |
3468 | break; | |
3469 | } | |
3470 | ||
b03c9f9f EC |
3471 | __reg_deduce_bounds(false_reg); |
3472 | __reg_deduce_bounds(true_reg); | |
3473 | /* We might have learned some bits from the bounds. */ | |
3474 | __reg_bound_offset(false_reg); | |
3475 | __reg_bound_offset(true_reg); | |
3476 | /* Intersecting with the old var_off might have improved our bounds | |
3477 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3478 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3479 | */ | |
3480 | __update_reg_bounds(false_reg); | |
3481 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
3482 | } |
3483 | ||
3484 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
3485 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
3486 | struct bpf_reg_state *dst_reg) | |
3487 | { | |
b03c9f9f EC |
3488 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
3489 | dst_reg->umin_value); | |
3490 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
3491 | dst_reg->umax_value); | |
3492 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
3493 | dst_reg->smin_value); | |
3494 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
3495 | dst_reg->smax_value); | |
f1174f77 EC |
3496 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
3497 | dst_reg->var_off); | |
b03c9f9f EC |
3498 | /* We might have learned new bounds from the var_off. */ |
3499 | __update_reg_bounds(src_reg); | |
3500 | __update_reg_bounds(dst_reg); | |
3501 | /* We might have learned something about the sign bit. */ | |
3502 | __reg_deduce_bounds(src_reg); | |
3503 | __reg_deduce_bounds(dst_reg); | |
3504 | /* We might have learned some bits from the bounds. */ | |
3505 | __reg_bound_offset(src_reg); | |
3506 | __reg_bound_offset(dst_reg); | |
3507 | /* Intersecting with the old var_off might have improved our bounds | |
3508 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
3509 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
3510 | */ | |
3511 | __update_reg_bounds(src_reg); | |
3512 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
3513 | } |
3514 | ||
3515 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
3516 | struct bpf_reg_state *true_dst, | |
3517 | struct bpf_reg_state *false_src, | |
3518 | struct bpf_reg_state *false_dst, | |
3519 | u8 opcode) | |
3520 | { | |
3521 | switch (opcode) { | |
3522 | case BPF_JEQ: | |
3523 | __reg_combine_min_max(true_src, true_dst); | |
3524 | break; | |
3525 | case BPF_JNE: | |
3526 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 3527 | break; |
4cabc5b1 | 3528 | } |
48461135 JB |
3529 | } |
3530 | ||
57a09bf0 | 3531 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
f1174f77 | 3532 | bool is_null) |
57a09bf0 TG |
3533 | { |
3534 | struct bpf_reg_state *reg = ®s[regno]; | |
3535 | ||
3536 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
f1174f77 EC |
3537 | /* Old offset (both fixed and variable parts) should |
3538 | * have been known-zero, because we don't allow pointer | |
3539 | * arithmetic on pointers that might be NULL. | |
3540 | */ | |
b03c9f9f EC |
3541 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
3542 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 3543 | reg->off)) { |
b03c9f9f EC |
3544 | __mark_reg_known_zero(reg); |
3545 | reg->off = 0; | |
f1174f77 EC |
3546 | } |
3547 | if (is_null) { | |
3548 | reg->type = SCALAR_VALUE; | |
56f668df MKL |
3549 | } else if (reg->map_ptr->inner_map_meta) { |
3550 | reg->type = CONST_PTR_TO_MAP; | |
3551 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
3552 | } else { | |
f1174f77 | 3553 | reg->type = PTR_TO_MAP_VALUE; |
56f668df | 3554 | } |
a08dd0da DB |
3555 | /* We don't need id from this point onwards anymore, thus we |
3556 | * should better reset it, so that state pruning has chances | |
3557 | * to take effect. | |
3558 | */ | |
3559 | reg->id = 0; | |
57a09bf0 TG |
3560 | } |
3561 | } | |
3562 | ||
3563 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
3564 | * be folded together at some point. | |
3565 | */ | |
f4d7e40a | 3566 | static void mark_map_regs(struct bpf_verifier_state *vstate, u32 regno, |
f1174f77 | 3567 | bool is_null) |
57a09bf0 | 3568 | { |
f4d7e40a | 3569 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
57a09bf0 | 3570 | struct bpf_reg_state *regs = state->regs; |
a08dd0da | 3571 | u32 id = regs[regno].id; |
f4d7e40a | 3572 | int i, j; |
57a09bf0 TG |
3573 | |
3574 | for (i = 0; i < MAX_BPF_REG; i++) | |
f1174f77 | 3575 | mark_map_reg(regs, i, id, is_null); |
57a09bf0 | 3576 | |
f4d7e40a AS |
3577 | for (j = 0; j <= vstate->curframe; j++) { |
3578 | state = vstate->frame[j]; | |
3579 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
3580 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
3581 | continue; | |
3582 | mark_map_reg(&state->stack[i].spilled_ptr, 0, id, is_null); | |
3583 | } | |
57a09bf0 TG |
3584 | } |
3585 | } | |
3586 | ||
5beca081 DB |
3587 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
3588 | struct bpf_reg_state *dst_reg, | |
3589 | struct bpf_reg_state *src_reg, | |
3590 | struct bpf_verifier_state *this_branch, | |
3591 | struct bpf_verifier_state *other_branch) | |
3592 | { | |
3593 | if (BPF_SRC(insn->code) != BPF_X) | |
3594 | return false; | |
3595 | ||
3596 | switch (BPF_OP(insn->code)) { | |
3597 | case BPF_JGT: | |
3598 | if ((dst_reg->type == PTR_TO_PACKET && | |
3599 | src_reg->type == PTR_TO_PACKET_END) || | |
3600 | (dst_reg->type == PTR_TO_PACKET_META && | |
3601 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3602 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
3603 | find_good_pkt_pointers(this_branch, dst_reg, | |
3604 | dst_reg->type, false); | |
3605 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3606 | src_reg->type == PTR_TO_PACKET) || | |
3607 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3608 | src_reg->type == PTR_TO_PACKET_META)) { | |
3609 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
3610 | find_good_pkt_pointers(other_branch, src_reg, | |
3611 | src_reg->type, true); | |
3612 | } else { | |
3613 | return false; | |
3614 | } | |
3615 | break; | |
3616 | case BPF_JLT: | |
3617 | if ((dst_reg->type == PTR_TO_PACKET && | |
3618 | src_reg->type == PTR_TO_PACKET_END) || | |
3619 | (dst_reg->type == PTR_TO_PACKET_META && | |
3620 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3621 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
3622 | find_good_pkt_pointers(other_branch, dst_reg, | |
3623 | dst_reg->type, true); | |
3624 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3625 | src_reg->type == PTR_TO_PACKET) || | |
3626 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3627 | src_reg->type == PTR_TO_PACKET_META)) { | |
3628 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
3629 | find_good_pkt_pointers(this_branch, src_reg, | |
3630 | src_reg->type, false); | |
3631 | } else { | |
3632 | return false; | |
3633 | } | |
3634 | break; | |
3635 | case BPF_JGE: | |
3636 | if ((dst_reg->type == PTR_TO_PACKET && | |
3637 | src_reg->type == PTR_TO_PACKET_END) || | |
3638 | (dst_reg->type == PTR_TO_PACKET_META && | |
3639 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3640 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
3641 | find_good_pkt_pointers(this_branch, dst_reg, | |
3642 | dst_reg->type, true); | |
3643 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3644 | src_reg->type == PTR_TO_PACKET) || | |
3645 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3646 | src_reg->type == PTR_TO_PACKET_META)) { | |
3647 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
3648 | find_good_pkt_pointers(other_branch, src_reg, | |
3649 | src_reg->type, false); | |
3650 | } else { | |
3651 | return false; | |
3652 | } | |
3653 | break; | |
3654 | case BPF_JLE: | |
3655 | if ((dst_reg->type == PTR_TO_PACKET && | |
3656 | src_reg->type == PTR_TO_PACKET_END) || | |
3657 | (dst_reg->type == PTR_TO_PACKET_META && | |
3658 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
3659 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
3660 | find_good_pkt_pointers(other_branch, dst_reg, | |
3661 | dst_reg->type, false); | |
3662 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
3663 | src_reg->type == PTR_TO_PACKET) || | |
3664 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
3665 | src_reg->type == PTR_TO_PACKET_META)) { | |
3666 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
3667 | find_good_pkt_pointers(this_branch, src_reg, | |
3668 | src_reg->type, true); | |
3669 | } else { | |
3670 | return false; | |
3671 | } | |
3672 | break; | |
3673 | default: | |
3674 | return false; | |
3675 | } | |
3676 | ||
3677 | return true; | |
3678 | } | |
3679 | ||
58e2af8b | 3680 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
3681 | struct bpf_insn *insn, int *insn_idx) |
3682 | { | |
f4d7e40a AS |
3683 | struct bpf_verifier_state *this_branch = env->cur_state; |
3684 | struct bpf_verifier_state *other_branch; | |
3685 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
3686 | struct bpf_reg_state *dst_reg, *other_branch_regs; | |
17a52670 AS |
3687 | u8 opcode = BPF_OP(insn->code); |
3688 | int err; | |
3689 | ||
b4e432f1 | 3690 | if (opcode > BPF_JSLE) { |
61bd5218 | 3691 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
3692 | return -EINVAL; |
3693 | } | |
3694 | ||
3695 | if (BPF_SRC(insn->code) == BPF_X) { | |
3696 | if (insn->imm != 0) { | |
61bd5218 | 3697 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3698 | return -EINVAL; |
3699 | } | |
3700 | ||
3701 | /* check src1 operand */ | |
dc503a8a | 3702 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3703 | if (err) |
3704 | return err; | |
1be7f75d AS |
3705 | |
3706 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 3707 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
3708 | insn->src_reg); |
3709 | return -EACCES; | |
3710 | } | |
17a52670 AS |
3711 | } else { |
3712 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 3713 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
3714 | return -EINVAL; |
3715 | } | |
3716 | } | |
3717 | ||
3718 | /* check src2 operand */ | |
dc503a8a | 3719 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3720 | if (err) |
3721 | return err; | |
3722 | ||
1a0dc1ac AS |
3723 | dst_reg = ®s[insn->dst_reg]; |
3724 | ||
17a52670 AS |
3725 | /* detect if R == 0 where R was initialized to zero earlier */ |
3726 | if (BPF_SRC(insn->code) == BPF_K && | |
3727 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
f1174f77 | 3728 | dst_reg->type == SCALAR_VALUE && |
3bf15921 AS |
3729 | tnum_is_const(dst_reg->var_off)) { |
3730 | if ((opcode == BPF_JEQ && dst_reg->var_off.value == insn->imm) || | |
3731 | (opcode == BPF_JNE && dst_reg->var_off.value != insn->imm)) { | |
17a52670 AS |
3732 | /* if (imm == imm) goto pc+off; |
3733 | * only follow the goto, ignore fall-through | |
3734 | */ | |
3735 | *insn_idx += insn->off; | |
3736 | return 0; | |
3737 | } else { | |
3738 | /* if (imm != imm) goto pc+off; | |
3739 | * only follow fall-through branch, since | |
3740 | * that's where the program will go | |
3741 | */ | |
3742 | return 0; | |
3743 | } | |
3744 | } | |
3745 | ||
3746 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
3747 | if (!other_branch) | |
3748 | return -EFAULT; | |
f4d7e40a | 3749 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 3750 | |
48461135 JB |
3751 | /* detect if we are comparing against a constant value so we can adjust |
3752 | * our min/max values for our dst register. | |
f1174f77 EC |
3753 | * this is only legit if both are scalars (or pointers to the same |
3754 | * object, I suppose, but we don't support that right now), because | |
3755 | * otherwise the different base pointers mean the offsets aren't | |
3756 | * comparable. | |
48461135 JB |
3757 | */ |
3758 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
3759 | if (dst_reg->type == SCALAR_VALUE && |
3760 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
3761 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
f4d7e40a | 3762 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
f1174f77 EC |
3763 | dst_reg, regs[insn->src_reg].var_off.value, |
3764 | opcode); | |
3765 | else if (tnum_is_const(dst_reg->var_off)) | |
f4d7e40a | 3766 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
f1174f77 EC |
3767 | ®s[insn->src_reg], |
3768 | dst_reg->var_off.value, opcode); | |
3769 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
3770 | /* Comparing for equality, we can combine knowledge */ | |
f4d7e40a AS |
3771 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
3772 | &other_branch_regs[insn->dst_reg], | |
f1174f77 EC |
3773 | ®s[insn->src_reg], |
3774 | ®s[insn->dst_reg], opcode); | |
3775 | } | |
3776 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 3777 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
48461135 JB |
3778 | dst_reg, insn->imm, opcode); |
3779 | } | |
3780 | ||
58e2af8b | 3781 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 3782 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
3783 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
3784 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
3785 | /* Mark all identical map registers in each branch as either |
3786 | * safe or unknown depending R == 0 or R != 0 conditional. | |
3787 | */ | |
f1174f77 EC |
3788 | mark_map_regs(this_branch, insn->dst_reg, opcode == BPF_JNE); |
3789 | mark_map_regs(other_branch, insn->dst_reg, opcode == BPF_JEQ); | |
5beca081 DB |
3790 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
3791 | this_branch, other_branch) && | |
3792 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
3793 | verbose(env, "R%d pointer comparison prohibited\n", |
3794 | insn->dst_reg); | |
1be7f75d | 3795 | return -EACCES; |
17a52670 | 3796 | } |
61bd5218 | 3797 | if (env->log.level) |
f4d7e40a | 3798 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
3799 | return 0; |
3800 | } | |
3801 | ||
0246e64d AS |
3802 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
3803 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
3804 | { | |
3805 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
3806 | ||
3807 | return (struct bpf_map *) (unsigned long) imm64; | |
3808 | } | |
3809 | ||
17a52670 | 3810 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 3811 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3812 | { |
638f5b90 | 3813 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3814 | int err; |
3815 | ||
3816 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 3817 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
3818 | return -EINVAL; |
3819 | } | |
3820 | if (insn->off != 0) { | |
61bd5218 | 3821 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
3822 | return -EINVAL; |
3823 | } | |
3824 | ||
dc503a8a | 3825 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3826 | if (err) |
3827 | return err; | |
3828 | ||
6b173873 | 3829 | if (insn->src_reg == 0) { |
6b173873 JK |
3830 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
3831 | ||
f1174f77 | 3832 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 3833 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 3834 | return 0; |
6b173873 | 3835 | } |
17a52670 AS |
3836 | |
3837 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
3838 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
3839 | ||
3840 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
3841 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
3842 | return 0; | |
3843 | } | |
3844 | ||
96be4325 DB |
3845 | static bool may_access_skb(enum bpf_prog_type type) |
3846 | { | |
3847 | switch (type) { | |
3848 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
3849 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 3850 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
3851 | return true; |
3852 | default: | |
3853 | return false; | |
3854 | } | |
3855 | } | |
3856 | ||
ddd872bc AS |
3857 | /* verify safety of LD_ABS|LD_IND instructions: |
3858 | * - they can only appear in the programs where ctx == skb | |
3859 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
3860 | * preserve R6-R9, and store return value into R0 | |
3861 | * | |
3862 | * Implicit input: | |
3863 | * ctx == skb == R6 == CTX | |
3864 | * | |
3865 | * Explicit input: | |
3866 | * SRC == any register | |
3867 | * IMM == 32-bit immediate | |
3868 | * | |
3869 | * Output: | |
3870 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
3871 | */ | |
58e2af8b | 3872 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 3873 | { |
638f5b90 | 3874 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 3875 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
3876 | int i, err; |
3877 | ||
24701ece | 3878 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 3879 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
3880 | return -EINVAL; |
3881 | } | |
3882 | ||
f4d7e40a AS |
3883 | if (env->subprog_cnt) { |
3884 | /* when program has LD_ABS insn JITs and interpreter assume | |
3885 | * that r1 == ctx == skb which is not the case for callees | |
3886 | * that can have arbitrary arguments. It's problematic | |
3887 | * for main prog as well since JITs would need to analyze | |
3888 | * all functions in order to make proper register save/restore | |
3889 | * decisions in the main prog. Hence disallow LD_ABS with calls | |
3890 | */ | |
3891 | verbose(env, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n"); | |
3892 | return -EINVAL; | |
3893 | } | |
3894 | ||
ddd872bc | 3895 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 3896 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 3897 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 3898 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
3899 | return -EINVAL; |
3900 | } | |
3901 | ||
3902 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 3903 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
3904 | if (err) |
3905 | return err; | |
3906 | ||
3907 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
61bd5218 JK |
3908 | verbose(env, |
3909 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
3910 | return -EINVAL; |
3911 | } | |
3912 | ||
3913 | if (mode == BPF_IND) { | |
3914 | /* check explicit source operand */ | |
dc503a8a | 3915 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
3916 | if (err) |
3917 | return err; | |
3918 | } | |
3919 | ||
3920 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 3921 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3922 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3923 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3924 | } | |
ddd872bc AS |
3925 | |
3926 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
3927 | * the value fetched from the packet. |
3928 | * Already marked as written above. | |
ddd872bc | 3929 | */ |
61bd5218 | 3930 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
3931 | return 0; |
3932 | } | |
3933 | ||
390ee7e2 AS |
3934 | static int check_return_code(struct bpf_verifier_env *env) |
3935 | { | |
3936 | struct bpf_reg_state *reg; | |
3937 | struct tnum range = tnum_range(0, 1); | |
3938 | ||
3939 | switch (env->prog->type) { | |
3940 | case BPF_PROG_TYPE_CGROUP_SKB: | |
3941 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
4fbac77d | 3942 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
390ee7e2 | 3943 | case BPF_PROG_TYPE_SOCK_OPS: |
ebc614f6 | 3944 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
390ee7e2 AS |
3945 | break; |
3946 | default: | |
3947 | return 0; | |
3948 | } | |
3949 | ||
638f5b90 | 3950 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 3951 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 3952 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
3953 | reg_type_str[reg->type]); |
3954 | return -EINVAL; | |
3955 | } | |
3956 | ||
3957 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 3958 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
3959 | if (!tnum_is_unknown(reg->var_off)) { |
3960 | char tn_buf[48]; | |
3961 | ||
3962 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 3963 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 3964 | } else { |
61bd5218 | 3965 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 3966 | } |
61bd5218 | 3967 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
3968 | return -EINVAL; |
3969 | } | |
3970 | return 0; | |
3971 | } | |
3972 | ||
475fb78f AS |
3973 | /* non-recursive DFS pseudo code |
3974 | * 1 procedure DFS-iterative(G,v): | |
3975 | * 2 label v as discovered | |
3976 | * 3 let S be a stack | |
3977 | * 4 S.push(v) | |
3978 | * 5 while S is not empty | |
3979 | * 6 t <- S.pop() | |
3980 | * 7 if t is what we're looking for: | |
3981 | * 8 return t | |
3982 | * 9 for all edges e in G.adjacentEdges(t) do | |
3983 | * 10 if edge e is already labelled | |
3984 | * 11 continue with the next edge | |
3985 | * 12 w <- G.adjacentVertex(t,e) | |
3986 | * 13 if vertex w is not discovered and not explored | |
3987 | * 14 label e as tree-edge | |
3988 | * 15 label w as discovered | |
3989 | * 16 S.push(w) | |
3990 | * 17 continue at 5 | |
3991 | * 18 else if vertex w is discovered | |
3992 | * 19 label e as back-edge | |
3993 | * 20 else | |
3994 | * 21 // vertex w is explored | |
3995 | * 22 label e as forward- or cross-edge | |
3996 | * 23 label t as explored | |
3997 | * 24 S.pop() | |
3998 | * | |
3999 | * convention: | |
4000 | * 0x10 - discovered | |
4001 | * 0x11 - discovered and fall-through edge labelled | |
4002 | * 0x12 - discovered and fall-through and branch edges labelled | |
4003 | * 0x20 - explored | |
4004 | */ | |
4005 | ||
4006 | enum { | |
4007 | DISCOVERED = 0x10, | |
4008 | EXPLORED = 0x20, | |
4009 | FALLTHROUGH = 1, | |
4010 | BRANCH = 2, | |
4011 | }; | |
4012 | ||
58e2af8b | 4013 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 4014 | |
475fb78f AS |
4015 | static int *insn_stack; /* stack of insns to process */ |
4016 | static int cur_stack; /* current stack index */ | |
4017 | static int *insn_state; | |
4018 | ||
4019 | /* t, w, e - match pseudo-code above: | |
4020 | * t - index of current instruction | |
4021 | * w - next instruction | |
4022 | * e - edge | |
4023 | */ | |
58e2af8b | 4024 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
4025 | { |
4026 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
4027 | return 0; | |
4028 | ||
4029 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
4030 | return 0; | |
4031 | ||
4032 | if (w < 0 || w >= env->prog->len) { | |
61bd5218 | 4033 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
4034 | return -EINVAL; |
4035 | } | |
4036 | ||
f1bca824 AS |
4037 | if (e == BRANCH) |
4038 | /* mark branch target for state pruning */ | |
4039 | env->explored_states[w] = STATE_LIST_MARK; | |
4040 | ||
475fb78f AS |
4041 | if (insn_state[w] == 0) { |
4042 | /* tree-edge */ | |
4043 | insn_state[t] = DISCOVERED | e; | |
4044 | insn_state[w] = DISCOVERED; | |
4045 | if (cur_stack >= env->prog->len) | |
4046 | return -E2BIG; | |
4047 | insn_stack[cur_stack++] = w; | |
4048 | return 1; | |
4049 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
61bd5218 | 4050 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
4051 | return -EINVAL; |
4052 | } else if (insn_state[w] == EXPLORED) { | |
4053 | /* forward- or cross-edge */ | |
4054 | insn_state[t] = DISCOVERED | e; | |
4055 | } else { | |
61bd5218 | 4056 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
4057 | return -EFAULT; |
4058 | } | |
4059 | return 0; | |
4060 | } | |
4061 | ||
4062 | /* non-recursive depth-first-search to detect loops in BPF program | |
4063 | * loop == back-edge in directed graph | |
4064 | */ | |
58e2af8b | 4065 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
4066 | { |
4067 | struct bpf_insn *insns = env->prog->insnsi; | |
4068 | int insn_cnt = env->prog->len; | |
4069 | int ret = 0; | |
4070 | int i, t; | |
4071 | ||
cc8b0b92 AS |
4072 | ret = check_subprogs(env); |
4073 | if (ret < 0) | |
4074 | return ret; | |
4075 | ||
475fb78f AS |
4076 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
4077 | if (!insn_state) | |
4078 | return -ENOMEM; | |
4079 | ||
4080 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
4081 | if (!insn_stack) { | |
4082 | kfree(insn_state); | |
4083 | return -ENOMEM; | |
4084 | } | |
4085 | ||
4086 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
4087 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
4088 | cur_stack = 1; | |
4089 | ||
4090 | peek_stack: | |
4091 | if (cur_stack == 0) | |
4092 | goto check_state; | |
4093 | t = insn_stack[cur_stack - 1]; | |
4094 | ||
4095 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
4096 | u8 opcode = BPF_OP(insns[t].code); | |
4097 | ||
4098 | if (opcode == BPF_EXIT) { | |
4099 | goto mark_explored; | |
4100 | } else if (opcode == BPF_CALL) { | |
4101 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
4102 | if (ret == 1) | |
4103 | goto peek_stack; | |
4104 | else if (ret < 0) | |
4105 | goto err_free; | |
07016151 DB |
4106 | if (t + 1 < insn_cnt) |
4107 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
cc8b0b92 AS |
4108 | if (insns[t].src_reg == BPF_PSEUDO_CALL) { |
4109 | env->explored_states[t] = STATE_LIST_MARK; | |
4110 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env); | |
4111 | if (ret == 1) | |
4112 | goto peek_stack; | |
4113 | else if (ret < 0) | |
4114 | goto err_free; | |
4115 | } | |
475fb78f AS |
4116 | } else if (opcode == BPF_JA) { |
4117 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
4118 | ret = -EINVAL; | |
4119 | goto err_free; | |
4120 | } | |
4121 | /* unconditional jump with single edge */ | |
4122 | ret = push_insn(t, t + insns[t].off + 1, | |
4123 | FALLTHROUGH, env); | |
4124 | if (ret == 1) | |
4125 | goto peek_stack; | |
4126 | else if (ret < 0) | |
4127 | goto err_free; | |
f1bca824 AS |
4128 | /* tell verifier to check for equivalent states |
4129 | * after every call and jump | |
4130 | */ | |
c3de6317 AS |
4131 | if (t + 1 < insn_cnt) |
4132 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
4133 | } else { |
4134 | /* conditional jump with two edges */ | |
3c2ce60b | 4135 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
4136 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
4137 | if (ret == 1) | |
4138 | goto peek_stack; | |
4139 | else if (ret < 0) | |
4140 | goto err_free; | |
4141 | ||
4142 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
4143 | if (ret == 1) | |
4144 | goto peek_stack; | |
4145 | else if (ret < 0) | |
4146 | goto err_free; | |
4147 | } | |
4148 | } else { | |
4149 | /* all other non-branch instructions with single | |
4150 | * fall-through edge | |
4151 | */ | |
4152 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
4153 | if (ret == 1) | |
4154 | goto peek_stack; | |
4155 | else if (ret < 0) | |
4156 | goto err_free; | |
4157 | } | |
4158 | ||
4159 | mark_explored: | |
4160 | insn_state[t] = EXPLORED; | |
4161 | if (cur_stack-- <= 0) { | |
61bd5218 | 4162 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
4163 | ret = -EFAULT; |
4164 | goto err_free; | |
4165 | } | |
4166 | goto peek_stack; | |
4167 | ||
4168 | check_state: | |
4169 | for (i = 0; i < insn_cnt; i++) { | |
4170 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 4171 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
4172 | ret = -EINVAL; |
4173 | goto err_free; | |
4174 | } | |
4175 | } | |
4176 | ret = 0; /* cfg looks good */ | |
4177 | ||
4178 | err_free: | |
4179 | kfree(insn_state); | |
4180 | kfree(insn_stack); | |
4181 | return ret; | |
4182 | } | |
4183 | ||
f1174f77 EC |
4184 | /* check %cur's range satisfies %old's */ |
4185 | static bool range_within(struct bpf_reg_state *old, | |
4186 | struct bpf_reg_state *cur) | |
4187 | { | |
b03c9f9f EC |
4188 | return old->umin_value <= cur->umin_value && |
4189 | old->umax_value >= cur->umax_value && | |
4190 | old->smin_value <= cur->smin_value && | |
4191 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
4192 | } |
4193 | ||
4194 | /* Maximum number of register states that can exist at once */ | |
4195 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
4196 | struct idpair { | |
4197 | u32 old; | |
4198 | u32 cur; | |
4199 | }; | |
4200 | ||
4201 | /* If in the old state two registers had the same id, then they need to have | |
4202 | * the same id in the new state as well. But that id could be different from | |
4203 | * the old state, so we need to track the mapping from old to new ids. | |
4204 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
4205 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
4206 | * regs with a different old id could still have new id 9, we don't care about | |
4207 | * that. | |
4208 | * So we look through our idmap to see if this old id has been seen before. If | |
4209 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 4210 | */ |
f1174f77 | 4211 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 4212 | { |
f1174f77 | 4213 | unsigned int i; |
969bf05e | 4214 | |
f1174f77 EC |
4215 | for (i = 0; i < ID_MAP_SIZE; i++) { |
4216 | if (!idmap[i].old) { | |
4217 | /* Reached an empty slot; haven't seen this id before */ | |
4218 | idmap[i].old = old_id; | |
4219 | idmap[i].cur = cur_id; | |
4220 | return true; | |
4221 | } | |
4222 | if (idmap[i].old == old_id) | |
4223 | return idmap[i].cur == cur_id; | |
4224 | } | |
4225 | /* We ran out of idmap slots, which should be impossible */ | |
4226 | WARN_ON_ONCE(1); | |
4227 | return false; | |
4228 | } | |
4229 | ||
4230 | /* Returns true if (rold safe implies rcur safe) */ | |
1b688a19 EC |
4231 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
4232 | struct idpair *idmap) | |
f1174f77 | 4233 | { |
f4d7e40a AS |
4234 | bool equal; |
4235 | ||
dc503a8a EC |
4236 | if (!(rold->live & REG_LIVE_READ)) |
4237 | /* explored state didn't use this */ | |
4238 | return true; | |
4239 | ||
f4d7e40a AS |
4240 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, frameno)) == 0; |
4241 | ||
4242 | if (rold->type == PTR_TO_STACK) | |
4243 | /* two stack pointers are equal only if they're pointing to | |
4244 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
4245 | */ | |
4246 | return equal && rold->frameno == rcur->frameno; | |
4247 | ||
4248 | if (equal) | |
969bf05e AS |
4249 | return true; |
4250 | ||
f1174f77 EC |
4251 | if (rold->type == NOT_INIT) |
4252 | /* explored state can't have used this */ | |
969bf05e | 4253 | return true; |
f1174f77 EC |
4254 | if (rcur->type == NOT_INIT) |
4255 | return false; | |
4256 | switch (rold->type) { | |
4257 | case SCALAR_VALUE: | |
4258 | if (rcur->type == SCALAR_VALUE) { | |
4259 | /* new val must satisfy old val knowledge */ | |
4260 | return range_within(rold, rcur) && | |
4261 | tnum_in(rold->var_off, rcur->var_off); | |
4262 | } else { | |
179d1c56 JH |
4263 | /* We're trying to use a pointer in place of a scalar. |
4264 | * Even if the scalar was unbounded, this could lead to | |
4265 | * pointer leaks because scalars are allowed to leak | |
4266 | * while pointers are not. We could make this safe in | |
4267 | * special cases if root is calling us, but it's | |
4268 | * probably not worth the hassle. | |
f1174f77 | 4269 | */ |
179d1c56 | 4270 | return false; |
f1174f77 EC |
4271 | } |
4272 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
4273 | /* If the new min/max/var_off satisfy the old ones and |
4274 | * everything else matches, we are OK. | |
4275 | * We don't care about the 'id' value, because nothing | |
4276 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
4277 | */ | |
4278 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
4279 | range_within(rold, rcur) && | |
4280 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
4281 | case PTR_TO_MAP_VALUE_OR_NULL: |
4282 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
4283 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
4284 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
4285 | * checked, doing so could have affected others with the same | |
4286 | * id, and we can't check for that because we lost the id when | |
4287 | * we converted to a PTR_TO_MAP_VALUE. | |
4288 | */ | |
4289 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
4290 | return false; | |
4291 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
4292 | return false; | |
4293 | /* Check our ids match any regs they're supposed to */ | |
4294 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 4295 | case PTR_TO_PACKET_META: |
f1174f77 | 4296 | case PTR_TO_PACKET: |
de8f3a83 | 4297 | if (rcur->type != rold->type) |
f1174f77 EC |
4298 | return false; |
4299 | /* We must have at least as much range as the old ptr | |
4300 | * did, so that any accesses which were safe before are | |
4301 | * still safe. This is true even if old range < old off, | |
4302 | * since someone could have accessed through (ptr - k), or | |
4303 | * even done ptr -= k in a register, to get a safe access. | |
4304 | */ | |
4305 | if (rold->range > rcur->range) | |
4306 | return false; | |
4307 | /* If the offsets don't match, we can't trust our alignment; | |
4308 | * nor can we be sure that we won't fall out of range. | |
4309 | */ | |
4310 | if (rold->off != rcur->off) | |
4311 | return false; | |
4312 | /* id relations must be preserved */ | |
4313 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
4314 | return false; | |
4315 | /* new val must satisfy old val knowledge */ | |
4316 | return range_within(rold, rcur) && | |
4317 | tnum_in(rold->var_off, rcur->var_off); | |
4318 | case PTR_TO_CTX: | |
4319 | case CONST_PTR_TO_MAP: | |
f1174f77 EC |
4320 | case PTR_TO_PACKET_END: |
4321 | /* Only valid matches are exact, which memcmp() above | |
4322 | * would have accepted | |
4323 | */ | |
4324 | default: | |
4325 | /* Don't know what's going on, just say it's not safe */ | |
4326 | return false; | |
4327 | } | |
969bf05e | 4328 | |
f1174f77 EC |
4329 | /* Shouldn't get here; if we do, say it's not safe */ |
4330 | WARN_ON_ONCE(1); | |
969bf05e AS |
4331 | return false; |
4332 | } | |
4333 | ||
f4d7e40a AS |
4334 | static bool stacksafe(struct bpf_func_state *old, |
4335 | struct bpf_func_state *cur, | |
638f5b90 AS |
4336 | struct idpair *idmap) |
4337 | { | |
4338 | int i, spi; | |
4339 | ||
4340 | /* if explored stack has more populated slots than current stack | |
4341 | * such stacks are not equivalent | |
4342 | */ | |
4343 | if (old->allocated_stack > cur->allocated_stack) | |
4344 | return false; | |
4345 | ||
4346 | /* walk slots of the explored stack and ignore any additional | |
4347 | * slots in the current stack, since explored(safe) state | |
4348 | * didn't use them | |
4349 | */ | |
4350 | for (i = 0; i < old->allocated_stack; i++) { | |
4351 | spi = i / BPF_REG_SIZE; | |
4352 | ||
cc2b14d5 AS |
4353 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) |
4354 | /* explored state didn't use this */ | |
fd05e57b | 4355 | continue; |
cc2b14d5 | 4356 | |
638f5b90 AS |
4357 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
4358 | continue; | |
cc2b14d5 AS |
4359 | /* if old state was safe with misc data in the stack |
4360 | * it will be safe with zero-initialized stack. | |
4361 | * The opposite is not true | |
4362 | */ | |
4363 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
4364 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
4365 | continue; | |
638f5b90 AS |
4366 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
4367 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
4368 | /* Ex: old explored (safe) state has STACK_SPILL in | |
4369 | * this stack slot, but current has has STACK_MISC -> | |
4370 | * this verifier states are not equivalent, | |
4371 | * return false to continue verification of this path | |
4372 | */ | |
4373 | return false; | |
4374 | if (i % BPF_REG_SIZE) | |
4375 | continue; | |
4376 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
4377 | continue; | |
4378 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
4379 | &cur->stack[spi].spilled_ptr, | |
4380 | idmap)) | |
4381 | /* when explored and current stack slot are both storing | |
4382 | * spilled registers, check that stored pointers types | |
4383 | * are the same as well. | |
4384 | * Ex: explored safe path could have stored | |
4385 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
4386 | * but current path has stored: | |
4387 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
4388 | * such verifier states are not equivalent. | |
4389 | * return false to continue verification of this path | |
4390 | */ | |
4391 | return false; | |
4392 | } | |
4393 | return true; | |
4394 | } | |
4395 | ||
f1bca824 AS |
4396 | /* compare two verifier states |
4397 | * | |
4398 | * all states stored in state_list are known to be valid, since | |
4399 | * verifier reached 'bpf_exit' instruction through them | |
4400 | * | |
4401 | * this function is called when verifier exploring different branches of | |
4402 | * execution popped from the state stack. If it sees an old state that has | |
4403 | * more strict register state and more strict stack state then this execution | |
4404 | * branch doesn't need to be explored further, since verifier already | |
4405 | * concluded that more strict state leads to valid finish. | |
4406 | * | |
4407 | * Therefore two states are equivalent if register state is more conservative | |
4408 | * and explored stack state is more conservative than the current one. | |
4409 | * Example: | |
4410 | * explored current | |
4411 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
4412 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
4413 | * | |
4414 | * In other words if current stack state (one being explored) has more | |
4415 | * valid slots than old one that already passed validation, it means | |
4416 | * the verifier can stop exploring and conclude that current state is valid too | |
4417 | * | |
4418 | * Similarly with registers. If explored state has register type as invalid | |
4419 | * whereas register type in current state is meaningful, it means that | |
4420 | * the current state will reach 'bpf_exit' instruction safely | |
4421 | */ | |
f4d7e40a AS |
4422 | static bool func_states_equal(struct bpf_func_state *old, |
4423 | struct bpf_func_state *cur) | |
f1bca824 | 4424 | { |
f1174f77 EC |
4425 | struct idpair *idmap; |
4426 | bool ret = false; | |
f1bca824 AS |
4427 | int i; |
4428 | ||
f1174f77 EC |
4429 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
4430 | /* If we failed to allocate the idmap, just say it's not safe */ | |
4431 | if (!idmap) | |
1a0dc1ac | 4432 | return false; |
f1174f77 EC |
4433 | |
4434 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 4435 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 4436 | goto out_free; |
f1bca824 AS |
4437 | } |
4438 | ||
638f5b90 AS |
4439 | if (!stacksafe(old, cur, idmap)) |
4440 | goto out_free; | |
f1174f77 EC |
4441 | ret = true; |
4442 | out_free: | |
4443 | kfree(idmap); | |
4444 | return ret; | |
f1bca824 AS |
4445 | } |
4446 | ||
f4d7e40a AS |
4447 | static bool states_equal(struct bpf_verifier_env *env, |
4448 | struct bpf_verifier_state *old, | |
4449 | struct bpf_verifier_state *cur) | |
4450 | { | |
4451 | int i; | |
4452 | ||
4453 | if (old->curframe != cur->curframe) | |
4454 | return false; | |
4455 | ||
4456 | /* for states to be equal callsites have to be the same | |
4457 | * and all frame states need to be equivalent | |
4458 | */ | |
4459 | for (i = 0; i <= old->curframe; i++) { | |
4460 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
4461 | return false; | |
4462 | if (!func_states_equal(old->frame[i], cur->frame[i])) | |
4463 | return false; | |
4464 | } | |
4465 | return true; | |
4466 | } | |
4467 | ||
8e9cd9ce | 4468 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
4469 | * straight-line code between a state and its parent. When we arrive at an |
4470 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
4471 | * code, so read marks in the state must propagate to the parent regardless | |
4472 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
4473 | * in mark_reg_read() and mark_stack_slot_read() is for. | |
8e9cd9ce | 4474 | */ |
f4d7e40a AS |
4475 | static int propagate_liveness(struct bpf_verifier_env *env, |
4476 | const struct bpf_verifier_state *vstate, | |
4477 | struct bpf_verifier_state *vparent) | |
dc503a8a | 4478 | { |
f4d7e40a AS |
4479 | int i, frame, err = 0; |
4480 | struct bpf_func_state *state, *parent; | |
dc503a8a | 4481 | |
f4d7e40a AS |
4482 | if (vparent->curframe != vstate->curframe) { |
4483 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
4484 | vparent->curframe, vstate->curframe); | |
4485 | return -EFAULT; | |
4486 | } | |
dc503a8a EC |
4487 | /* Propagate read liveness of registers... */ |
4488 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
4489 | /* We don't need to worry about FP liveness because it's read-only */ | |
4490 | for (i = 0; i < BPF_REG_FP; i++) { | |
f4d7e40a | 4491 | if (vparent->frame[vparent->curframe]->regs[i].live & REG_LIVE_READ) |
63f45f84 | 4492 | continue; |
f4d7e40a AS |
4493 | if (vstate->frame[vstate->curframe]->regs[i].live & REG_LIVE_READ) { |
4494 | err = mark_reg_read(env, vstate, vparent, i); | |
4495 | if (err) | |
4496 | return err; | |
dc503a8a EC |
4497 | } |
4498 | } | |
f4d7e40a | 4499 | |
dc503a8a | 4500 | /* ... and stack slots */ |
f4d7e40a AS |
4501 | for (frame = 0; frame <= vstate->curframe; frame++) { |
4502 | state = vstate->frame[frame]; | |
4503 | parent = vparent->frame[frame]; | |
4504 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && | |
4505 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
f4d7e40a AS |
4506 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
4507 | continue; | |
4508 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) | |
4509 | mark_stack_slot_read(env, vstate, vparent, i, frame); | |
dc503a8a EC |
4510 | } |
4511 | } | |
f4d7e40a | 4512 | return err; |
dc503a8a EC |
4513 | } |
4514 | ||
58e2af8b | 4515 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 4516 | { |
58e2af8b JK |
4517 | struct bpf_verifier_state_list *new_sl; |
4518 | struct bpf_verifier_state_list *sl; | |
638f5b90 | 4519 | struct bpf_verifier_state *cur = env->cur_state; |
f4d7e40a | 4520 | int i, j, err; |
f1bca824 AS |
4521 | |
4522 | sl = env->explored_states[insn_idx]; | |
4523 | if (!sl) | |
4524 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
4525 | * be doing state search here | |
4526 | */ | |
4527 | return 0; | |
4528 | ||
4529 | while (sl != STATE_LIST_MARK) { | |
638f5b90 | 4530 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 4531 | /* reached equivalent register/stack state, |
dc503a8a EC |
4532 | * prune the search. |
4533 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
4534 | * If we have any write marks in env->cur_state, they |
4535 | * will prevent corresponding reads in the continuation | |
4536 | * from reaching our parent (an explored_state). Our | |
4537 | * own state will get the read marks recorded, but | |
4538 | * they'll be immediately forgotten as we're pruning | |
4539 | * this state and will pop a new one. | |
f1bca824 | 4540 | */ |
f4d7e40a AS |
4541 | err = propagate_liveness(env, &sl->state, cur); |
4542 | if (err) | |
4543 | return err; | |
f1bca824 | 4544 | return 1; |
dc503a8a | 4545 | } |
f1bca824 AS |
4546 | sl = sl->next; |
4547 | } | |
4548 | ||
4549 | /* there were no equivalent states, remember current one. | |
4550 | * technically the current state is not proven to be safe yet, | |
f4d7e40a AS |
4551 | * but it will either reach outer most bpf_exit (which means it's safe) |
4552 | * or it will be rejected. Since there are no loops, we won't be | |
4553 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) | |
4554 | * again on the way to bpf_exit | |
f1bca824 | 4555 | */ |
638f5b90 | 4556 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
4557 | if (!new_sl) |
4558 | return -ENOMEM; | |
4559 | ||
4560 | /* add new state to the head of linked list */ | |
1969db47 AS |
4561 | err = copy_verifier_state(&new_sl->state, cur); |
4562 | if (err) { | |
4563 | free_verifier_state(&new_sl->state, false); | |
4564 | kfree(new_sl); | |
4565 | return err; | |
4566 | } | |
f1bca824 AS |
4567 | new_sl->next = env->explored_states[insn_idx]; |
4568 | env->explored_states[insn_idx] = new_sl; | |
dc503a8a | 4569 | /* connect new state to parentage chain */ |
638f5b90 | 4570 | cur->parent = &new_sl->state; |
8e9cd9ce EC |
4571 | /* clear write marks in current state: the writes we did are not writes |
4572 | * our child did, so they don't screen off its reads from us. | |
4573 | * (There are no read marks in current state, because reads always mark | |
4574 | * their parent and current state never has children yet. Only | |
4575 | * explored_states can get read marks.) | |
4576 | */ | |
dc503a8a | 4577 | for (i = 0; i < BPF_REG_FP; i++) |
f4d7e40a AS |
4578 | cur->frame[cur->curframe]->regs[i].live = REG_LIVE_NONE; |
4579 | ||
4580 | /* all stack frames are accessible from callee, clear them all */ | |
4581 | for (j = 0; j <= cur->curframe; j++) { | |
4582 | struct bpf_func_state *frame = cur->frame[j]; | |
4583 | ||
4584 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) | |
cc2b14d5 | 4585 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
f4d7e40a | 4586 | } |
f1bca824 AS |
4587 | return 0; |
4588 | } | |
4589 | ||
58e2af8b | 4590 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 4591 | { |
638f5b90 | 4592 | struct bpf_verifier_state *state; |
17a52670 | 4593 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 4594 | struct bpf_reg_state *regs; |
f4d7e40a | 4595 | int insn_cnt = env->prog->len, i; |
17a52670 AS |
4596 | int insn_idx, prev_insn_idx = 0; |
4597 | int insn_processed = 0; | |
4598 | bool do_print_state = false; | |
4599 | ||
638f5b90 AS |
4600 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
4601 | if (!state) | |
4602 | return -ENOMEM; | |
f4d7e40a | 4603 | state->curframe = 0; |
dc503a8a | 4604 | state->parent = NULL; |
f4d7e40a AS |
4605 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); |
4606 | if (!state->frame[0]) { | |
4607 | kfree(state); | |
4608 | return -ENOMEM; | |
4609 | } | |
4610 | env->cur_state = state; | |
4611 | init_func_state(env, state->frame[0], | |
4612 | BPF_MAIN_FUNC /* callsite */, | |
4613 | 0 /* frameno */, | |
4614 | 0 /* subprogno, zero == main subprog */); | |
17a52670 AS |
4615 | insn_idx = 0; |
4616 | for (;;) { | |
4617 | struct bpf_insn *insn; | |
4618 | u8 class; | |
4619 | int err; | |
4620 | ||
4621 | if (insn_idx >= insn_cnt) { | |
61bd5218 | 4622 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
17a52670 AS |
4623 | insn_idx, insn_cnt); |
4624 | return -EFAULT; | |
4625 | } | |
4626 | ||
4627 | insn = &insns[insn_idx]; | |
4628 | class = BPF_CLASS(insn->code); | |
4629 | ||
07016151 | 4630 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
4631 | verbose(env, |
4632 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
4633 | insn_processed); |
4634 | return -E2BIG; | |
4635 | } | |
4636 | ||
f1bca824 AS |
4637 | err = is_state_visited(env, insn_idx); |
4638 | if (err < 0) | |
4639 | return err; | |
4640 | if (err == 1) { | |
4641 | /* found equivalent state, can prune the search */ | |
61bd5218 | 4642 | if (env->log.level) { |
f1bca824 | 4643 | if (do_print_state) |
61bd5218 | 4644 | verbose(env, "\nfrom %d to %d: safe\n", |
f1bca824 AS |
4645 | prev_insn_idx, insn_idx); |
4646 | else | |
61bd5218 | 4647 | verbose(env, "%d: safe\n", insn_idx); |
f1bca824 AS |
4648 | } |
4649 | goto process_bpf_exit; | |
4650 | } | |
4651 | ||
3c2ce60b DB |
4652 | if (need_resched()) |
4653 | cond_resched(); | |
4654 | ||
61bd5218 JK |
4655 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
4656 | if (env->log.level > 1) | |
4657 | verbose(env, "%d:", insn_idx); | |
c5fc9692 | 4658 | else |
61bd5218 | 4659 | verbose(env, "\nfrom %d to %d:", |
c5fc9692 | 4660 | prev_insn_idx, insn_idx); |
f4d7e40a | 4661 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
4662 | do_print_state = false; |
4663 | } | |
4664 | ||
61bd5218 | 4665 | if (env->log.level) { |
7105e828 DB |
4666 | const struct bpf_insn_cbs cbs = { |
4667 | .cb_print = verbose, | |
abe08840 | 4668 | .private_data = env, |
7105e828 DB |
4669 | }; |
4670 | ||
61bd5218 | 4671 | verbose(env, "%d: ", insn_idx); |
abe08840 | 4672 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
4673 | } |
4674 | ||
cae1927c JK |
4675 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
4676 | err = bpf_prog_offload_verify_insn(env, insn_idx, | |
4677 | prev_insn_idx); | |
4678 | if (err) | |
4679 | return err; | |
4680 | } | |
13a27dfc | 4681 | |
638f5b90 | 4682 | regs = cur_regs(env); |
c131187d | 4683 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 4684 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 4685 | err = check_alu_op(env, insn); |
17a52670 AS |
4686 | if (err) |
4687 | return err; | |
4688 | ||
4689 | } else if (class == BPF_LDX) { | |
3df126f3 | 4690 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
4691 | |
4692 | /* check for reserved fields is already done */ | |
4693 | ||
17a52670 | 4694 | /* check src operand */ |
dc503a8a | 4695 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4696 | if (err) |
4697 | return err; | |
4698 | ||
dc503a8a | 4699 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
4700 | if (err) |
4701 | return err; | |
4702 | ||
725f9dcd AS |
4703 | src_reg_type = regs[insn->src_reg].type; |
4704 | ||
17a52670 AS |
4705 | /* check that memory (src_reg + off) is readable, |
4706 | * the state of dst_reg will be updated by this func | |
4707 | */ | |
31fd8581 | 4708 | err = check_mem_access(env, insn_idx, insn->src_reg, insn->off, |
17a52670 | 4709 | BPF_SIZE(insn->code), BPF_READ, |
ca369602 | 4710 | insn->dst_reg, false); |
17a52670 AS |
4711 | if (err) |
4712 | return err; | |
4713 | ||
3df126f3 JK |
4714 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
4715 | ||
4716 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
4717 | /* saw a valid insn |
4718 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 4719 | * save type to validate intersecting paths |
9bac3d6d | 4720 | */ |
3df126f3 | 4721 | *prev_src_type = src_reg_type; |
9bac3d6d | 4722 | |
3df126f3 | 4723 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 4724 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 4725 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
4726 | /* ABuser program is trying to use the same insn |
4727 | * dst_reg = *(u32*) (src_reg + off) | |
4728 | * with different pointer types: | |
4729 | * src_reg == ctx in one branch and | |
4730 | * src_reg == stack|map in some other branch. | |
4731 | * Reject it. | |
4732 | */ | |
61bd5218 | 4733 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
4734 | return -EINVAL; |
4735 | } | |
4736 | ||
17a52670 | 4737 | } else if (class == BPF_STX) { |
3df126f3 | 4738 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 4739 | |
17a52670 | 4740 | if (BPF_MODE(insn->code) == BPF_XADD) { |
31fd8581 | 4741 | err = check_xadd(env, insn_idx, insn); |
17a52670 AS |
4742 | if (err) |
4743 | return err; | |
4744 | insn_idx++; | |
4745 | continue; | |
4746 | } | |
4747 | ||
17a52670 | 4748 | /* check src1 operand */ |
dc503a8a | 4749 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4750 | if (err) |
4751 | return err; | |
4752 | /* check src2 operand */ | |
dc503a8a | 4753 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4754 | if (err) |
4755 | return err; | |
4756 | ||
d691f9e8 AS |
4757 | dst_reg_type = regs[insn->dst_reg].type; |
4758 | ||
17a52670 | 4759 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 4760 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 | 4761 | BPF_SIZE(insn->code), BPF_WRITE, |
ca369602 | 4762 | insn->src_reg, false); |
17a52670 AS |
4763 | if (err) |
4764 | return err; | |
4765 | ||
3df126f3 JK |
4766 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
4767 | ||
4768 | if (*prev_dst_type == NOT_INIT) { | |
4769 | *prev_dst_type = dst_reg_type; | |
4770 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 4771 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 4772 | *prev_dst_type == PTR_TO_CTX)) { |
61bd5218 | 4773 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
4774 | return -EINVAL; |
4775 | } | |
4776 | ||
17a52670 AS |
4777 | } else if (class == BPF_ST) { |
4778 | if (BPF_MODE(insn->code) != BPF_MEM || | |
4779 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 4780 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
4781 | return -EINVAL; |
4782 | } | |
4783 | /* check src operand */ | |
dc503a8a | 4784 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4785 | if (err) |
4786 | return err; | |
4787 | ||
f37a8cb8 DB |
4788 | if (is_ctx_reg(env, insn->dst_reg)) { |
4789 | verbose(env, "BPF_ST stores into R%d context is not allowed\n", | |
4790 | insn->dst_reg); | |
4791 | return -EACCES; | |
4792 | } | |
4793 | ||
17a52670 | 4794 | /* check that memory (dst_reg + off) is writeable */ |
31fd8581 | 4795 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
17a52670 | 4796 | BPF_SIZE(insn->code), BPF_WRITE, |
ca369602 | 4797 | -1, false); |
17a52670 AS |
4798 | if (err) |
4799 | return err; | |
4800 | ||
4801 | } else if (class == BPF_JMP) { | |
4802 | u8 opcode = BPF_OP(insn->code); | |
4803 | ||
4804 | if (opcode == BPF_CALL) { | |
4805 | if (BPF_SRC(insn->code) != BPF_K || | |
4806 | insn->off != 0 || | |
f4d7e40a AS |
4807 | (insn->src_reg != BPF_REG_0 && |
4808 | insn->src_reg != BPF_PSEUDO_CALL) || | |
17a52670 | 4809 | insn->dst_reg != BPF_REG_0) { |
61bd5218 | 4810 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
4811 | return -EINVAL; |
4812 | } | |
4813 | ||
f4d7e40a AS |
4814 | if (insn->src_reg == BPF_PSEUDO_CALL) |
4815 | err = check_func_call(env, insn, &insn_idx); | |
4816 | else | |
4817 | err = check_helper_call(env, insn->imm, insn_idx); | |
17a52670 AS |
4818 | if (err) |
4819 | return err; | |
4820 | ||
4821 | } else if (opcode == BPF_JA) { | |
4822 | if (BPF_SRC(insn->code) != BPF_K || | |
4823 | insn->imm != 0 || | |
4824 | insn->src_reg != BPF_REG_0 || | |
4825 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4826 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
4827 | return -EINVAL; |
4828 | } | |
4829 | ||
4830 | insn_idx += insn->off + 1; | |
4831 | continue; | |
4832 | ||
4833 | } else if (opcode == BPF_EXIT) { | |
4834 | if (BPF_SRC(insn->code) != BPF_K || | |
4835 | insn->imm != 0 || | |
4836 | insn->src_reg != BPF_REG_0 || | |
4837 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 4838 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
4839 | return -EINVAL; |
4840 | } | |
4841 | ||
f4d7e40a AS |
4842 | if (state->curframe) { |
4843 | /* exit from nested function */ | |
4844 | prev_insn_idx = insn_idx; | |
4845 | err = prepare_func_exit(env, &insn_idx); | |
4846 | if (err) | |
4847 | return err; | |
4848 | do_print_state = true; | |
4849 | continue; | |
4850 | } | |
4851 | ||
17a52670 AS |
4852 | /* eBPF calling convetion is such that R0 is used |
4853 | * to return the value from eBPF program. | |
4854 | * Make sure that it's readable at this time | |
4855 | * of bpf_exit, which means that program wrote | |
4856 | * something into it earlier | |
4857 | */ | |
dc503a8a | 4858 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
4859 | if (err) |
4860 | return err; | |
4861 | ||
1be7f75d | 4862 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 4863 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
4864 | return -EACCES; |
4865 | } | |
4866 | ||
390ee7e2 AS |
4867 | err = check_return_code(env); |
4868 | if (err) | |
4869 | return err; | |
f1bca824 | 4870 | process_bpf_exit: |
638f5b90 AS |
4871 | err = pop_stack(env, &prev_insn_idx, &insn_idx); |
4872 | if (err < 0) { | |
4873 | if (err != -ENOENT) | |
4874 | return err; | |
17a52670 AS |
4875 | break; |
4876 | } else { | |
4877 | do_print_state = true; | |
4878 | continue; | |
4879 | } | |
4880 | } else { | |
4881 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
4882 | if (err) | |
4883 | return err; | |
4884 | } | |
4885 | } else if (class == BPF_LD) { | |
4886 | u8 mode = BPF_MODE(insn->code); | |
4887 | ||
4888 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
4889 | err = check_ld_abs(env, insn); |
4890 | if (err) | |
4891 | return err; | |
4892 | ||
17a52670 AS |
4893 | } else if (mode == BPF_IMM) { |
4894 | err = check_ld_imm(env, insn); | |
4895 | if (err) | |
4896 | return err; | |
4897 | ||
4898 | insn_idx++; | |
c131187d | 4899 | env->insn_aux_data[insn_idx].seen = true; |
17a52670 | 4900 | } else { |
61bd5218 | 4901 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
4902 | return -EINVAL; |
4903 | } | |
4904 | } else { | |
61bd5218 | 4905 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
4906 | return -EINVAL; |
4907 | } | |
4908 | ||
4909 | insn_idx++; | |
4910 | } | |
4911 | ||
4bd95f4b DB |
4912 | verbose(env, "processed %d insns (limit %d), stack depth ", |
4913 | insn_processed, BPF_COMPLEXITY_LIMIT_INSNS); | |
f4d7e40a AS |
4914 | for (i = 0; i < env->subprog_cnt + 1; i++) { |
4915 | u32 depth = env->subprog_stack_depth[i]; | |
4916 | ||
4917 | verbose(env, "%d", depth); | |
4918 | if (i + 1 < env->subprog_cnt + 1) | |
4919 | verbose(env, "+"); | |
4920 | } | |
4921 | verbose(env, "\n"); | |
4922 | env->prog->aux->stack_depth = env->subprog_stack_depth[0]; | |
17a52670 AS |
4923 | return 0; |
4924 | } | |
4925 | ||
56f668df MKL |
4926 | static int check_map_prealloc(struct bpf_map *map) |
4927 | { | |
4928 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
4929 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
4930 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
4931 | !(map->map_flags & BPF_F_NO_PREALLOC); |
4932 | } | |
4933 | ||
61bd5218 JK |
4934 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
4935 | struct bpf_map *map, | |
fdc15d38 AS |
4936 | struct bpf_prog *prog) |
4937 | ||
4938 | { | |
56f668df MKL |
4939 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
4940 | * preallocated hash maps, since doing memory allocation | |
4941 | * in overflow_handler can crash depending on where nmi got | |
4942 | * triggered. | |
4943 | */ | |
4944 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
4945 | if (!check_map_prealloc(map)) { | |
61bd5218 | 4946 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
4947 | return -EINVAL; |
4948 | } | |
4949 | if (map->inner_map_meta && | |
4950 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 4951 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
4952 | return -EINVAL; |
4953 | } | |
fdc15d38 | 4954 | } |
a3884572 JK |
4955 | |
4956 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && | |
4957 | !bpf_offload_dev_match(prog, map)) { | |
4958 | verbose(env, "offload device mismatch between prog and map\n"); | |
4959 | return -EINVAL; | |
4960 | } | |
4961 | ||
fdc15d38 AS |
4962 | return 0; |
4963 | } | |
4964 | ||
0246e64d AS |
4965 | /* look for pseudo eBPF instructions that access map FDs and |
4966 | * replace them with actual map pointers | |
4967 | */ | |
58e2af8b | 4968 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
4969 | { |
4970 | struct bpf_insn *insn = env->prog->insnsi; | |
4971 | int insn_cnt = env->prog->len; | |
fdc15d38 | 4972 | int i, j, err; |
0246e64d | 4973 | |
f1f7714e | 4974 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
4975 | if (err) |
4976 | return err; | |
4977 | ||
0246e64d | 4978 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 4979 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 4980 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 4981 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
4982 | return -EINVAL; |
4983 | } | |
4984 | ||
d691f9e8 AS |
4985 | if (BPF_CLASS(insn->code) == BPF_STX && |
4986 | ((BPF_MODE(insn->code) != BPF_MEM && | |
4987 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 4988 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
4989 | return -EINVAL; |
4990 | } | |
4991 | ||
0246e64d AS |
4992 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
4993 | struct bpf_map *map; | |
4994 | struct fd f; | |
4995 | ||
4996 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
4997 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
4998 | insn[1].off != 0) { | |
61bd5218 | 4999 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
5000 | return -EINVAL; |
5001 | } | |
5002 | ||
5003 | if (insn->src_reg == 0) | |
5004 | /* valid generic load 64-bit imm */ | |
5005 | goto next_insn; | |
5006 | ||
5007 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
5008 | verbose(env, |
5009 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
5010 | return -EINVAL; |
5011 | } | |
5012 | ||
5013 | f = fdget(insn->imm); | |
c2101297 | 5014 | map = __bpf_map_get(f); |
0246e64d | 5015 | if (IS_ERR(map)) { |
61bd5218 | 5016 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 5017 | insn->imm); |
0246e64d AS |
5018 | return PTR_ERR(map); |
5019 | } | |
5020 | ||
61bd5218 | 5021 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
5022 | if (err) { |
5023 | fdput(f); | |
5024 | return err; | |
5025 | } | |
5026 | ||
0246e64d AS |
5027 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
5028 | insn[0].imm = (u32) (unsigned long) map; | |
5029 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
5030 | ||
5031 | /* check whether we recorded this map already */ | |
5032 | for (j = 0; j < env->used_map_cnt; j++) | |
5033 | if (env->used_maps[j] == map) { | |
5034 | fdput(f); | |
5035 | goto next_insn; | |
5036 | } | |
5037 | ||
5038 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
5039 | fdput(f); | |
5040 | return -E2BIG; | |
5041 | } | |
5042 | ||
0246e64d AS |
5043 | /* hold the map. If the program is rejected by verifier, |
5044 | * the map will be released by release_maps() or it | |
5045 | * will be used by the valid program until it's unloaded | |
5046 | * and all maps are released in free_bpf_prog_info() | |
5047 | */ | |
92117d84 AS |
5048 | map = bpf_map_inc(map, false); |
5049 | if (IS_ERR(map)) { | |
5050 | fdput(f); | |
5051 | return PTR_ERR(map); | |
5052 | } | |
5053 | env->used_maps[env->used_map_cnt++] = map; | |
5054 | ||
0246e64d AS |
5055 | fdput(f); |
5056 | next_insn: | |
5057 | insn++; | |
5058 | i++; | |
5e581dad DB |
5059 | continue; |
5060 | } | |
5061 | ||
5062 | /* Basic sanity check before we invest more work here. */ | |
5063 | if (!bpf_opcode_in_insntable(insn->code)) { | |
5064 | verbose(env, "unknown opcode %02x\n", insn->code); | |
5065 | return -EINVAL; | |
0246e64d AS |
5066 | } |
5067 | } | |
5068 | ||
5069 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
5070 | * 'struct bpf_map *' into a register instead of user map_fd. | |
5071 | * These pointers will be used later by verifier to validate map access. | |
5072 | */ | |
5073 | return 0; | |
5074 | } | |
5075 | ||
5076 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 5077 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
5078 | { |
5079 | int i; | |
5080 | ||
5081 | for (i = 0; i < env->used_map_cnt; i++) | |
5082 | bpf_map_put(env->used_maps[i]); | |
5083 | } | |
5084 | ||
5085 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 5086 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
5087 | { |
5088 | struct bpf_insn *insn = env->prog->insnsi; | |
5089 | int insn_cnt = env->prog->len; | |
5090 | int i; | |
5091 | ||
5092 | for (i = 0; i < insn_cnt; i++, insn++) | |
5093 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
5094 | insn->src_reg = 0; | |
5095 | } | |
5096 | ||
8041902d AS |
5097 | /* single env->prog->insni[off] instruction was replaced with the range |
5098 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
5099 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
5100 | */ | |
5101 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
5102 | u32 off, u32 cnt) | |
5103 | { | |
5104 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
c131187d | 5105 | int i; |
8041902d AS |
5106 | |
5107 | if (cnt == 1) | |
5108 | return 0; | |
5109 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
5110 | if (!new_data) | |
5111 | return -ENOMEM; | |
5112 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
5113 | memcpy(new_data + off + cnt - 1, old_data + off, | |
5114 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
c131187d AS |
5115 | for (i = off; i < off + cnt - 1; i++) |
5116 | new_data[i].seen = true; | |
8041902d AS |
5117 | env->insn_aux_data = new_data; |
5118 | vfree(old_data); | |
5119 | return 0; | |
5120 | } | |
5121 | ||
cc8b0b92 AS |
5122 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
5123 | { | |
5124 | int i; | |
5125 | ||
5126 | if (len == 1) | |
5127 | return; | |
5128 | for (i = 0; i < env->subprog_cnt; i++) { | |
5129 | if (env->subprog_starts[i] < off) | |
5130 | continue; | |
5131 | env->subprog_starts[i] += len - 1; | |
5132 | } | |
5133 | } | |
5134 | ||
8041902d AS |
5135 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
5136 | const struct bpf_insn *patch, u32 len) | |
5137 | { | |
5138 | struct bpf_prog *new_prog; | |
5139 | ||
5140 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
5141 | if (!new_prog) | |
5142 | return NULL; | |
5143 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
5144 | return NULL; | |
cc8b0b92 | 5145 | adjust_subprog_starts(env, off, len); |
8041902d AS |
5146 | return new_prog; |
5147 | } | |
5148 | ||
2a5418a1 DB |
5149 | /* The verifier does more data flow analysis than llvm and will not |
5150 | * explore branches that are dead at run time. Malicious programs can | |
5151 | * have dead code too. Therefore replace all dead at-run-time code | |
5152 | * with 'ja -1'. | |
5153 | * | |
5154 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
5155 | * program and through another bug we would manage to jump there, then | |
5156 | * we'd execute beyond program memory otherwise. Returning exception | |
5157 | * code also wouldn't work since we can have subprogs where the dead | |
5158 | * code could be located. | |
c131187d AS |
5159 | */ |
5160 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
5161 | { | |
5162 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 5163 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
5164 | struct bpf_insn *insn = env->prog->insnsi; |
5165 | const int insn_cnt = env->prog->len; | |
5166 | int i; | |
5167 | ||
5168 | for (i = 0; i < insn_cnt; i++) { | |
5169 | if (aux_data[i].seen) | |
5170 | continue; | |
2a5418a1 | 5171 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
5172 | } |
5173 | } | |
5174 | ||
9bac3d6d AS |
5175 | /* convert load instructions that access fields of 'struct __sk_buff' |
5176 | * into sequence of instructions that access fields of 'struct sk_buff' | |
5177 | */ | |
58e2af8b | 5178 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 5179 | { |
00176a34 | 5180 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 5181 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 5182 | const int insn_cnt = env->prog->len; |
36bbef52 | 5183 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 5184 | struct bpf_prog *new_prog; |
d691f9e8 | 5185 | enum bpf_access_type type; |
f96da094 DB |
5186 | bool is_narrower_load; |
5187 | u32 target_size; | |
9bac3d6d | 5188 | |
36bbef52 DB |
5189 | if (ops->gen_prologue) { |
5190 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
5191 | env->prog); | |
5192 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 5193 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
5194 | return -EINVAL; |
5195 | } else if (cnt) { | |
8041902d | 5196 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
5197 | if (!new_prog) |
5198 | return -ENOMEM; | |
8041902d | 5199 | |
36bbef52 | 5200 | env->prog = new_prog; |
3df126f3 | 5201 | delta += cnt - 1; |
36bbef52 DB |
5202 | } |
5203 | } | |
5204 | ||
5205 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
5206 | return 0; |
5207 | ||
3df126f3 | 5208 | insn = env->prog->insnsi + delta; |
36bbef52 | 5209 | |
9bac3d6d | 5210 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
5211 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
5212 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
5213 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 5214 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 5215 | type = BPF_READ; |
62c7989b DB |
5216 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
5217 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
5218 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 5219 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
5220 | type = BPF_WRITE; |
5221 | else | |
9bac3d6d AS |
5222 | continue; |
5223 | ||
8041902d | 5224 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 5225 | continue; |
9bac3d6d | 5226 | |
31fd8581 | 5227 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 5228 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
5229 | |
5230 | /* If the read access is a narrower load of the field, | |
5231 | * convert to a 4/8-byte load, to minimum program type specific | |
5232 | * convert_ctx_access changes. If conversion is successful, | |
5233 | * we will apply proper mask to the result. | |
5234 | */ | |
f96da094 | 5235 | is_narrower_load = size < ctx_field_size; |
31fd8581 | 5236 | if (is_narrower_load) { |
f96da094 DB |
5237 | u32 off = insn->off; |
5238 | u8 size_code; | |
5239 | ||
5240 | if (type == BPF_WRITE) { | |
61bd5218 | 5241 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
5242 | return -EINVAL; |
5243 | } | |
31fd8581 | 5244 | |
f96da094 | 5245 | size_code = BPF_H; |
31fd8581 YS |
5246 | if (ctx_field_size == 4) |
5247 | size_code = BPF_W; | |
5248 | else if (ctx_field_size == 8) | |
5249 | size_code = BPF_DW; | |
f96da094 | 5250 | |
31fd8581 YS |
5251 | insn->off = off & ~(ctx_field_size - 1); |
5252 | insn->code = BPF_LDX | BPF_MEM | size_code; | |
5253 | } | |
f96da094 DB |
5254 | |
5255 | target_size = 0; | |
5256 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog, | |
5257 | &target_size); | |
5258 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || | |
5259 | (ctx_field_size && !target_size)) { | |
61bd5218 | 5260 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
5261 | return -EINVAL; |
5262 | } | |
f96da094 DB |
5263 | |
5264 | if (is_narrower_load && size < target_size) { | |
31fd8581 YS |
5265 | if (ctx_field_size <= 4) |
5266 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 5267 | (1 << size * 8) - 1); |
31fd8581 YS |
5268 | else |
5269 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, | |
f96da094 | 5270 | (1 << size * 8) - 1); |
31fd8581 | 5271 | } |
9bac3d6d | 5272 | |
8041902d | 5273 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
5274 | if (!new_prog) |
5275 | return -ENOMEM; | |
5276 | ||
3df126f3 | 5277 | delta += cnt - 1; |
9bac3d6d AS |
5278 | |
5279 | /* keep walking new program and skip insns we just inserted */ | |
5280 | env->prog = new_prog; | |
3df126f3 | 5281 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
5282 | } |
5283 | ||
5284 | return 0; | |
5285 | } | |
5286 | ||
1c2a088a AS |
5287 | static int jit_subprogs(struct bpf_verifier_env *env) |
5288 | { | |
5289 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
5290 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
7105e828 | 5291 | struct bpf_insn *insn; |
1c2a088a AS |
5292 | void *old_bpf_func; |
5293 | int err = -ENOMEM; | |
5294 | ||
5295 | if (env->subprog_cnt == 0) | |
5296 | return 0; | |
5297 | ||
7105e828 | 5298 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
1c2a088a AS |
5299 | if (insn->code != (BPF_JMP | BPF_CALL) || |
5300 | insn->src_reg != BPF_PSEUDO_CALL) | |
5301 | continue; | |
5302 | subprog = find_subprog(env, i + insn->imm + 1); | |
5303 | if (subprog < 0) { | |
5304 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
5305 | i + insn->imm + 1); | |
5306 | return -EFAULT; | |
5307 | } | |
5308 | /* temporarily remember subprog id inside insn instead of | |
5309 | * aux_data, since next loop will split up all insns into funcs | |
5310 | */ | |
5311 | insn->off = subprog + 1; | |
5312 | /* remember original imm in case JIT fails and fallback | |
5313 | * to interpreter will be needed | |
5314 | */ | |
5315 | env->insn_aux_data[i].call_imm = insn->imm; | |
5316 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
5317 | insn->imm = 1; | |
5318 | } | |
5319 | ||
5320 | func = kzalloc(sizeof(prog) * (env->subprog_cnt + 1), GFP_KERNEL); | |
5321 | if (!func) | |
5322 | return -ENOMEM; | |
5323 | ||
5324 | for (i = 0; i <= env->subprog_cnt; i++) { | |
5325 | subprog_start = subprog_end; | |
5326 | if (env->subprog_cnt == i) | |
5327 | subprog_end = prog->len; | |
5328 | else | |
5329 | subprog_end = env->subprog_starts[i]; | |
5330 | ||
5331 | len = subprog_end - subprog_start; | |
5332 | func[i] = bpf_prog_alloc(bpf_prog_size(len), GFP_USER); | |
5333 | if (!func[i]) | |
5334 | goto out_free; | |
5335 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
5336 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 5337 | func[i]->type = prog->type; |
1c2a088a | 5338 | func[i]->len = len; |
4f74d809 DB |
5339 | if (bpf_prog_calc_tag(func[i])) |
5340 | goto out_free; | |
1c2a088a AS |
5341 | func[i]->is_func = 1; |
5342 | /* Use bpf_prog_F_tag to indicate functions in stack traces. | |
5343 | * Long term would need debug info to populate names | |
5344 | */ | |
5345 | func[i]->aux->name[0] = 'F'; | |
5346 | func[i]->aux->stack_depth = env->subprog_stack_depth[i]; | |
5347 | func[i]->jit_requested = 1; | |
5348 | func[i] = bpf_int_jit_compile(func[i]); | |
5349 | if (!func[i]->jited) { | |
5350 | err = -ENOTSUPP; | |
5351 | goto out_free; | |
5352 | } | |
5353 | cond_resched(); | |
5354 | } | |
5355 | /* at this point all bpf functions were successfully JITed | |
5356 | * now populate all bpf_calls with correct addresses and | |
5357 | * run last pass of JIT | |
5358 | */ | |
5359 | for (i = 0; i <= env->subprog_cnt; i++) { | |
5360 | insn = func[i]->insnsi; | |
5361 | for (j = 0; j < func[i]->len; j++, insn++) { | |
5362 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
5363 | insn->src_reg != BPF_PSEUDO_CALL) | |
5364 | continue; | |
5365 | subprog = insn->off; | |
5366 | insn->off = 0; | |
5367 | insn->imm = (u64 (*)(u64, u64, u64, u64, u64)) | |
5368 | func[subprog]->bpf_func - | |
5369 | __bpf_call_base; | |
5370 | } | |
5371 | } | |
5372 | for (i = 0; i <= env->subprog_cnt; i++) { | |
5373 | old_bpf_func = func[i]->bpf_func; | |
5374 | tmp = bpf_int_jit_compile(func[i]); | |
5375 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
5376 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
5377 | err = -EFAULT; | |
5378 | goto out_free; | |
5379 | } | |
5380 | cond_resched(); | |
5381 | } | |
5382 | ||
5383 | /* finally lock prog and jit images for all functions and | |
5384 | * populate kallsysm | |
5385 | */ | |
5386 | for (i = 0; i <= env->subprog_cnt; i++) { | |
5387 | bpf_prog_lock_ro(func[i]); | |
5388 | bpf_prog_kallsyms_add(func[i]); | |
5389 | } | |
7105e828 DB |
5390 | |
5391 | /* Last step: make now unused interpreter insns from main | |
5392 | * prog consistent for later dump requests, so they can | |
5393 | * later look the same as if they were interpreted only. | |
5394 | */ | |
5395 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
5396 | unsigned long addr; | |
5397 | ||
5398 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
5399 | insn->src_reg != BPF_PSEUDO_CALL) | |
5400 | continue; | |
5401 | insn->off = env->insn_aux_data[i].call_imm; | |
5402 | subprog = find_subprog(env, i + insn->off + 1); | |
5403 | addr = (unsigned long)func[subprog + 1]->bpf_func; | |
5404 | addr &= PAGE_MASK; | |
5405 | insn->imm = (u64 (*)(u64, u64, u64, u64, u64)) | |
5406 | addr - __bpf_call_base; | |
5407 | } | |
5408 | ||
1c2a088a AS |
5409 | prog->jited = 1; |
5410 | prog->bpf_func = func[0]->bpf_func; | |
5411 | prog->aux->func = func; | |
5412 | prog->aux->func_cnt = env->subprog_cnt + 1; | |
5413 | return 0; | |
5414 | out_free: | |
5415 | for (i = 0; i <= env->subprog_cnt; i++) | |
5416 | if (func[i]) | |
5417 | bpf_jit_free(func[i]); | |
5418 | kfree(func); | |
5419 | /* cleanup main prog to be interpreted */ | |
5420 | prog->jit_requested = 0; | |
5421 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
5422 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
5423 | insn->src_reg != BPF_PSEUDO_CALL) | |
5424 | continue; | |
5425 | insn->off = 0; | |
5426 | insn->imm = env->insn_aux_data[i].call_imm; | |
5427 | } | |
5428 | return err; | |
5429 | } | |
5430 | ||
1ea47e01 AS |
5431 | static int fixup_call_args(struct bpf_verifier_env *env) |
5432 | { | |
19d28fbd | 5433 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
5434 | struct bpf_prog *prog = env->prog; |
5435 | struct bpf_insn *insn = prog->insnsi; | |
5436 | int i, depth; | |
19d28fbd DM |
5437 | #endif |
5438 | int err; | |
1ea47e01 | 5439 | |
19d28fbd DM |
5440 | err = 0; |
5441 | if (env->prog->jit_requested) { | |
5442 | err = jit_subprogs(env); | |
5443 | if (err == 0) | |
1c2a088a | 5444 | return 0; |
19d28fbd DM |
5445 | } |
5446 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1ea47e01 AS |
5447 | for (i = 0; i < prog->len; i++, insn++) { |
5448 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
5449 | insn->src_reg != BPF_PSEUDO_CALL) | |
5450 | continue; | |
5451 | depth = get_callee_stack_depth(env, insn, i); | |
5452 | if (depth < 0) | |
5453 | return depth; | |
5454 | bpf_patch_call_args(insn, depth); | |
5455 | } | |
19d28fbd DM |
5456 | err = 0; |
5457 | #endif | |
5458 | return err; | |
1ea47e01 AS |
5459 | } |
5460 | ||
79741b3b | 5461 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 5462 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
5463 | * |
5464 | * this function is called after eBPF program passed verification | |
5465 | */ | |
79741b3b | 5466 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 5467 | { |
79741b3b AS |
5468 | struct bpf_prog *prog = env->prog; |
5469 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 5470 | const struct bpf_func_proto *fn; |
79741b3b | 5471 | const int insn_cnt = prog->len; |
81ed18ab AS |
5472 | struct bpf_insn insn_buf[16]; |
5473 | struct bpf_prog *new_prog; | |
5474 | struct bpf_map *map_ptr; | |
5475 | int i, cnt, delta = 0; | |
e245c5c6 | 5476 | |
79741b3b | 5477 | for (i = 0; i < insn_cnt; i++, insn++) { |
f6b1b3bf DB |
5478 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
5479 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
5480 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 5481 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf DB |
5482 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
5483 | struct bpf_insn mask_and_div[] = { | |
5484 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
5485 | /* Rx div 0 -> 0 */ | |
5486 | BPF_JMP_IMM(BPF_JNE, insn->src_reg, 0, 2), | |
5487 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), | |
5488 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
5489 | *insn, | |
5490 | }; | |
5491 | struct bpf_insn mask_and_mod[] = { | |
5492 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
5493 | /* Rx mod 0 -> Rx */ | |
5494 | BPF_JMP_IMM(BPF_JEQ, insn->src_reg, 0, 1), | |
5495 | *insn, | |
5496 | }; | |
5497 | struct bpf_insn *patchlet; | |
5498 | ||
5499 | if (insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
5500 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
5501 | patchlet = mask_and_div + (is64 ? 1 : 0); | |
5502 | cnt = ARRAY_SIZE(mask_and_div) - (is64 ? 1 : 0); | |
5503 | } else { | |
5504 | patchlet = mask_and_mod + (is64 ? 1 : 0); | |
5505 | cnt = ARRAY_SIZE(mask_and_mod) - (is64 ? 1 : 0); | |
5506 | } | |
5507 | ||
5508 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
5509 | if (!new_prog) |
5510 | return -ENOMEM; | |
5511 | ||
5512 | delta += cnt - 1; | |
5513 | env->prog = prog = new_prog; | |
5514 | insn = new_prog->insnsi + i + delta; | |
5515 | continue; | |
5516 | } | |
5517 | ||
79741b3b AS |
5518 | if (insn->code != (BPF_JMP | BPF_CALL)) |
5519 | continue; | |
cc8b0b92 AS |
5520 | if (insn->src_reg == BPF_PSEUDO_CALL) |
5521 | continue; | |
e245c5c6 | 5522 | |
79741b3b AS |
5523 | if (insn->imm == BPF_FUNC_get_route_realm) |
5524 | prog->dst_needed = 1; | |
5525 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
5526 | bpf_user_rnd_init_once(); | |
9802d865 JB |
5527 | if (insn->imm == BPF_FUNC_override_return) |
5528 | prog->kprobe_override = 1; | |
79741b3b | 5529 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
5530 | /* If we tail call into other programs, we |
5531 | * cannot make any assumptions since they can | |
5532 | * be replaced dynamically during runtime in | |
5533 | * the program array. | |
5534 | */ | |
5535 | prog->cb_access = 1; | |
80a58d02 | 5536 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
7b9f6da1 | 5537 | |
79741b3b AS |
5538 | /* mark bpf_tail_call as different opcode to avoid |
5539 | * conditional branch in the interpeter for every normal | |
5540 | * call and to prevent accidental JITing by JIT compiler | |
5541 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 5542 | */ |
79741b3b | 5543 | insn->imm = 0; |
71189fa9 | 5544 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 AS |
5545 | |
5546 | /* instead of changing every JIT dealing with tail_call | |
5547 | * emit two extra insns: | |
5548 | * if (index >= max_entries) goto out; | |
5549 | * index &= array->index_mask; | |
5550 | * to avoid out-of-bounds cpu speculation | |
5551 | */ | |
5552 | map_ptr = env->insn_aux_data[i + delta].map_ptr; | |
5553 | if (map_ptr == BPF_MAP_PTR_POISON) { | |
40950343 | 5554 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
5555 | return -EINVAL; |
5556 | } | |
5557 | if (!map_ptr->unpriv_array) | |
5558 | continue; | |
5559 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, | |
5560 | map_ptr->max_entries, 2); | |
5561 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
5562 | container_of(map_ptr, | |
5563 | struct bpf_array, | |
5564 | map)->index_mask); | |
5565 | insn_buf[2] = *insn; | |
5566 | cnt = 3; | |
5567 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
5568 | if (!new_prog) | |
5569 | return -ENOMEM; | |
5570 | ||
5571 | delta += cnt - 1; | |
5572 | env->prog = prog = new_prog; | |
5573 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
5574 | continue; |
5575 | } | |
e245c5c6 | 5576 | |
89c63074 DB |
5577 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
5578 | * handlers are currently limited to 64 bit only. | |
5579 | */ | |
60b58afc | 5580 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
89c63074 | 5581 | insn->imm == BPF_FUNC_map_lookup_elem) { |
81ed18ab | 5582 | map_ptr = env->insn_aux_data[i + delta].map_ptr; |
fad73a1a MKL |
5583 | if (map_ptr == BPF_MAP_PTR_POISON || |
5584 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
5585 | goto patch_call_imm; |
5586 | ||
5587 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
5588 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 5589 | verbose(env, "bpf verifier is misconfigured\n"); |
81ed18ab AS |
5590 | return -EINVAL; |
5591 | } | |
5592 | ||
5593 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
5594 | cnt); | |
5595 | if (!new_prog) | |
5596 | return -ENOMEM; | |
5597 | ||
5598 | delta += cnt - 1; | |
5599 | ||
5600 | /* keep walking new program and skip insns we just inserted */ | |
5601 | env->prog = prog = new_prog; | |
5602 | insn = new_prog->insnsi + i + delta; | |
5603 | continue; | |
5604 | } | |
5605 | ||
109980b8 | 5606 | if (insn->imm == BPF_FUNC_redirect_map) { |
7c300131 DB |
5607 | /* Note, we cannot use prog directly as imm as subsequent |
5608 | * rewrites would still change the prog pointer. The only | |
5609 | * stable address we can use is aux, which also works with | |
5610 | * prog clones during blinding. | |
5611 | */ | |
5612 | u64 addr = (unsigned long)prog->aux; | |
109980b8 DB |
5613 | struct bpf_insn r4_ld[] = { |
5614 | BPF_LD_IMM64(BPF_REG_4, addr), | |
5615 | *insn, | |
5616 | }; | |
5617 | cnt = ARRAY_SIZE(r4_ld); | |
5618 | ||
5619 | new_prog = bpf_patch_insn_data(env, i + delta, r4_ld, cnt); | |
5620 | if (!new_prog) | |
5621 | return -ENOMEM; | |
5622 | ||
5623 | delta += cnt - 1; | |
5624 | env->prog = prog = new_prog; | |
5625 | insn = new_prog->insnsi + i + delta; | |
5626 | } | |
81ed18ab | 5627 | patch_call_imm: |
5e43f899 | 5628 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
5629 | /* all functions that have prototype and verifier allowed |
5630 | * programs to call them, must be real in-kernel functions | |
5631 | */ | |
5632 | if (!fn->func) { | |
61bd5218 JK |
5633 | verbose(env, |
5634 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
5635 | func_id_name(insn->imm), insn->imm); |
5636 | return -EFAULT; | |
e245c5c6 | 5637 | } |
79741b3b | 5638 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 5639 | } |
e245c5c6 | 5640 | |
79741b3b AS |
5641 | return 0; |
5642 | } | |
e245c5c6 | 5643 | |
58e2af8b | 5644 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 5645 | { |
58e2af8b | 5646 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
5647 | int i; |
5648 | ||
5649 | if (!env->explored_states) | |
5650 | return; | |
5651 | ||
5652 | for (i = 0; i < env->prog->len; i++) { | |
5653 | sl = env->explored_states[i]; | |
5654 | ||
5655 | if (sl) | |
5656 | while (sl != STATE_LIST_MARK) { | |
5657 | sln = sl->next; | |
1969db47 | 5658 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
5659 | kfree(sl); |
5660 | sl = sln; | |
5661 | } | |
5662 | } | |
5663 | ||
5664 | kfree(env->explored_states); | |
5665 | } | |
5666 | ||
9bac3d6d | 5667 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 5668 | { |
58e2af8b | 5669 | struct bpf_verifier_env *env; |
b9193c1b | 5670 | struct bpf_verifier_log *log; |
51580e79 AS |
5671 | int ret = -EINVAL; |
5672 | ||
eba0c929 AB |
5673 | /* no program is valid */ |
5674 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
5675 | return -EINVAL; | |
5676 | ||
58e2af8b | 5677 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
5678 | * allocate/free it every time bpf_check() is called |
5679 | */ | |
58e2af8b | 5680 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
5681 | if (!env) |
5682 | return -ENOMEM; | |
61bd5218 | 5683 | log = &env->log; |
cbd35700 | 5684 | |
3df126f3 JK |
5685 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
5686 | (*prog)->len); | |
5687 | ret = -ENOMEM; | |
5688 | if (!env->insn_aux_data) | |
5689 | goto err_free_env; | |
9bac3d6d | 5690 | env->prog = *prog; |
00176a34 | 5691 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 5692 | |
cbd35700 AS |
5693 | /* grab the mutex to protect few globals used by verifier */ |
5694 | mutex_lock(&bpf_verifier_lock); | |
5695 | ||
5696 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
5697 | /* user requested verbose verifier output | |
5698 | * and supplied buffer to store the verification trace | |
5699 | */ | |
e7bf8249 JK |
5700 | log->level = attr->log_level; |
5701 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
5702 | log->len_total = attr->log_size; | |
cbd35700 AS |
5703 | |
5704 | ret = -EINVAL; | |
e7bf8249 JK |
5705 | /* log attributes have to be sane */ |
5706 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
5707 | !log->level || !log->ubuf) | |
3df126f3 | 5708 | goto err_unlock; |
cbd35700 | 5709 | } |
1ad2f583 DB |
5710 | |
5711 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
5712 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 5713 | env->strict_alignment = true; |
cbd35700 | 5714 | |
cae1927c | 5715 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
ab3f0063 JK |
5716 | ret = bpf_prog_offload_verifier_prep(env); |
5717 | if (ret) | |
5718 | goto err_unlock; | |
5719 | } | |
5720 | ||
0246e64d AS |
5721 | ret = replace_map_fd_with_map_ptr(env); |
5722 | if (ret < 0) | |
5723 | goto skip_full_check; | |
5724 | ||
9bac3d6d | 5725 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 5726 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
5727 | GFP_USER); |
5728 | ret = -ENOMEM; | |
5729 | if (!env->explored_states) | |
5730 | goto skip_full_check; | |
5731 | ||
cc8b0b92 AS |
5732 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
5733 | ||
475fb78f AS |
5734 | ret = check_cfg(env); |
5735 | if (ret < 0) | |
5736 | goto skip_full_check; | |
5737 | ||
17a52670 | 5738 | ret = do_check(env); |
8c01c4f8 CG |
5739 | if (env->cur_state) { |
5740 | free_verifier_state(env->cur_state, true); | |
5741 | env->cur_state = NULL; | |
5742 | } | |
cbd35700 | 5743 | |
0246e64d | 5744 | skip_full_check: |
638f5b90 | 5745 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 5746 | free_states(env); |
0246e64d | 5747 | |
c131187d AS |
5748 | if (ret == 0) |
5749 | sanitize_dead_code(env); | |
5750 | ||
70a87ffe AS |
5751 | if (ret == 0) |
5752 | ret = check_max_stack_depth(env); | |
5753 | ||
9bac3d6d AS |
5754 | if (ret == 0) |
5755 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
5756 | ret = convert_ctx_accesses(env); | |
5757 | ||
e245c5c6 | 5758 | if (ret == 0) |
79741b3b | 5759 | ret = fixup_bpf_calls(env); |
e245c5c6 | 5760 | |
1ea47e01 AS |
5761 | if (ret == 0) |
5762 | ret = fixup_call_args(env); | |
5763 | ||
a2a7d570 | 5764 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 5765 | ret = -ENOSPC; |
a2a7d570 | 5766 | if (log->level && !log->ubuf) { |
cbd35700 | 5767 | ret = -EFAULT; |
a2a7d570 | 5768 | goto err_release_maps; |
cbd35700 AS |
5769 | } |
5770 | ||
0246e64d AS |
5771 | if (ret == 0 && env->used_map_cnt) { |
5772 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
5773 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
5774 | sizeof(env->used_maps[0]), | |
5775 | GFP_KERNEL); | |
0246e64d | 5776 | |
9bac3d6d | 5777 | if (!env->prog->aux->used_maps) { |
0246e64d | 5778 | ret = -ENOMEM; |
a2a7d570 | 5779 | goto err_release_maps; |
0246e64d AS |
5780 | } |
5781 | ||
9bac3d6d | 5782 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 5783 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 5784 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
5785 | |
5786 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
5787 | * bpf_ld_imm64 instructions | |
5788 | */ | |
5789 | convert_pseudo_ld_imm64(env); | |
5790 | } | |
cbd35700 | 5791 | |
a2a7d570 | 5792 | err_release_maps: |
9bac3d6d | 5793 | if (!env->prog->aux->used_maps) |
0246e64d AS |
5794 | /* if we didn't copy map pointers into bpf_prog_info, release |
5795 | * them now. Otherwise free_bpf_prog_info() will release them. | |
5796 | */ | |
5797 | release_maps(env); | |
9bac3d6d | 5798 | *prog = env->prog; |
3df126f3 | 5799 | err_unlock: |
cbd35700 | 5800 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
5801 | vfree(env->insn_aux_data); |
5802 | err_free_env: | |
5803 | kfree(env); | |
51580e79 AS |
5804 | return ret; |
5805 | } |