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