]>
Commit | Line | Data |
---|---|---|
51580e79 | 1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 2 | * Copyright (c) 2016 Facebook |
fd978bf7 | 3 | * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io |
51580e79 AS |
4 | * |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of version 2 of the GNU General Public | |
7 | * License as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but | |
10 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | */ | |
838e9690 | 14 | #include <uapi/linux/btf.h> |
51580e79 AS |
15 | #include <linux/kernel.h> |
16 | #include <linux/types.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/bpf.h> | |
838e9690 | 19 | #include <linux/btf.h> |
58e2af8b | 20 | #include <linux/bpf_verifier.h> |
51580e79 AS |
21 | #include <linux/filter.h> |
22 | #include <net/netlink.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/vmalloc.h> | |
ebb676da | 25 | #include <linux/stringify.h> |
cc8b0b92 AS |
26 | #include <linux/bsearch.h> |
27 | #include <linux/sort.h> | |
c195651e | 28 | #include <linux/perf_event.h> |
d9762e84 | 29 | #include <linux/ctype.h> |
51580e79 | 30 | |
f4ac7e0b JK |
31 | #include "disasm.h" |
32 | ||
00176a34 JK |
33 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
34 | #define BPF_PROG_TYPE(_id, _name) \ | |
35 | [_id] = & _name ## _verifier_ops, | |
36 | #define BPF_MAP_TYPE(_id, _ops) | |
37 | #include <linux/bpf_types.h> | |
38 | #undef BPF_PROG_TYPE | |
39 | #undef BPF_MAP_TYPE | |
40 | }; | |
41 | ||
51580e79 AS |
42 | /* bpf_check() is a static code analyzer that walks eBPF program |
43 | * instruction by instruction and updates register/stack state. | |
44 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
45 | * | |
46 | * The first pass is depth-first-search to check that the program is a DAG. | |
47 | * It rejects the following programs: | |
48 | * - larger than BPF_MAXINSNS insns | |
49 | * - if loop is present (detected via back-edge) | |
50 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
51 | * - out of bounds or malformed jumps | |
52 | * The second pass is all possible path descent from the 1st insn. | |
53 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 54 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
55 | * insn is less then 4K, but there are too many branches that change stack/regs. |
56 | * Number of 'branches to be analyzed' is limited to 1k | |
57 | * | |
58 | * On entry to each instruction, each register has a type, and the instruction | |
59 | * changes the types of the registers depending on instruction semantics. | |
60 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
61 | * copied to R1. | |
62 | * | |
63 | * All registers are 64-bit. | |
64 | * R0 - return register | |
65 | * R1-R5 argument passing registers | |
66 | * R6-R9 callee saved registers | |
67 | * R10 - frame pointer read-only | |
68 | * | |
69 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
70 | * and has type PTR_TO_CTX. | |
71 | * | |
72 | * Verifier tracks arithmetic operations on pointers in case: | |
73 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
74 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
75 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
76 | * and 2nd arithmetic instruction is pattern matched to recognize | |
77 | * that it wants to construct a pointer to some element within stack. | |
78 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
79 | * (and -20 constant is saved for further stack bounds checking). | |
80 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
81 | * | |
f1174f77 | 82 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 83 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 84 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
85 | * |
86 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
87 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
88 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
89 | * |
90 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
91 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
92 | * | |
93 | * registers used to pass values to function calls are checked against | |
94 | * function argument constraints. | |
95 | * | |
96 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
97 | * It means that the register type passed to this function must be | |
98 | * PTR_TO_STACK and it will be used inside the function as | |
99 | * 'pointer to map element key' | |
100 | * | |
101 | * For example the argument constraints for bpf_map_lookup_elem(): | |
102 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
103 | * .arg1_type = ARG_CONST_MAP_PTR, | |
104 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
105 | * | |
106 | * ret_type says that this function returns 'pointer to map elem value or null' | |
107 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
108 | * 2nd argument should be a pointer to stack, which will be used inside | |
109 | * the helper function as a pointer to map element key. | |
110 | * | |
111 | * On the kernel side the helper function looks like: | |
112 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
113 | * { | |
114 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
115 | * void *key = (void *) (unsigned long) r2; | |
116 | * void *value; | |
117 | * | |
118 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
119 | * [key, key + map->key_size) bytes are valid and were initialized on | |
120 | * the stack of eBPF program. | |
121 | * } | |
122 | * | |
123 | * Corresponding eBPF program may look like: | |
124 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
125 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
126 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
127 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
128 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
129 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
130 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
131 | * | |
132 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
133 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
134 | * and were initialized prior to this call. | |
135 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
136 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
137 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
138 | * returns ether pointer to map value or NULL. | |
139 | * | |
140 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
141 | * insn, the register holding that pointer in the true branch changes state to | |
142 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
143 | * branch. See check_cond_jmp_op(). | |
144 | * | |
145 | * After the call R0 is set to return type of the function and registers R1-R5 | |
146 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
147 | * |
148 | * The following reference types represent a potential reference to a kernel | |
149 | * resource which, after first being allocated, must be checked and freed by | |
150 | * the BPF program: | |
151 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
152 | * | |
153 | * When the verifier sees a helper call return a reference type, it allocates a | |
154 | * pointer id for the reference and stores it in the current function state. | |
155 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
156 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
157 | * passes through a NULL-check conditional. For the branch wherein the state is | |
158 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
159 | * |
160 | * For each helper function that allocates a reference, such as | |
161 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
162 | * bpf_sk_release(). When a reference type passes into the release function, | |
163 | * the verifier also releases the reference. If any unchecked or unreleased | |
164 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
165 | */ |
166 | ||
17a52670 | 167 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 168 | struct bpf_verifier_stack_elem { |
17a52670 AS |
169 | /* verifer state is 'st' |
170 | * before processing instruction 'insn_idx' | |
171 | * and after processing instruction 'prev_insn_idx' | |
172 | */ | |
58e2af8b | 173 | struct bpf_verifier_state st; |
17a52670 AS |
174 | int insn_idx; |
175 | int prev_insn_idx; | |
58e2af8b | 176 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
177 | }; |
178 | ||
8e17c1b1 | 179 | #define BPF_COMPLEXITY_LIMIT_INSNS 131072 |
07016151 | 180 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 |
ceefbc96 | 181 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 182 | |
c93552c4 DB |
183 | #define BPF_MAP_PTR_UNPRIV 1UL |
184 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
185 | POISON_POINTER_DELTA)) | |
186 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
187 | ||
188 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) | |
189 | { | |
190 | return BPF_MAP_PTR(aux->map_state) == BPF_MAP_PTR_POISON; | |
191 | } | |
192 | ||
193 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
194 | { | |
195 | return aux->map_state & BPF_MAP_PTR_UNPRIV; | |
196 | } | |
197 | ||
198 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
199 | const struct bpf_map *map, bool unpriv) | |
200 | { | |
201 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
202 | unpriv |= bpf_map_ptr_unpriv(aux); | |
203 | aux->map_state = (unsigned long)map | | |
204 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
205 | } | |
fad73a1a | 206 | |
33ff9823 DB |
207 | struct bpf_call_arg_meta { |
208 | struct bpf_map *map_ptr; | |
435faee1 | 209 | bool raw_mode; |
36bbef52 | 210 | bool pkt_access; |
435faee1 DB |
211 | int regno; |
212 | int access_size; | |
849fa506 YS |
213 | s64 msize_smax_value; |
214 | u64 msize_umax_value; | |
fd978bf7 | 215 | int ptr_id; |
33ff9823 DB |
216 | }; |
217 | ||
cbd35700 AS |
218 | static DEFINE_MUTEX(bpf_verifier_lock); |
219 | ||
d9762e84 MKL |
220 | static const struct bpf_line_info * |
221 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
222 | { | |
223 | const struct bpf_line_info *linfo; | |
224 | const struct bpf_prog *prog; | |
225 | u32 i, nr_linfo; | |
226 | ||
227 | prog = env->prog; | |
228 | nr_linfo = prog->aux->nr_linfo; | |
229 | ||
230 | if (!nr_linfo || insn_off >= prog->len) | |
231 | return NULL; | |
232 | ||
233 | linfo = prog->aux->linfo; | |
234 | for (i = 1; i < nr_linfo; i++) | |
235 | if (insn_off < linfo[i].insn_off) | |
236 | break; | |
237 | ||
238 | return &linfo[i - 1]; | |
239 | } | |
240 | ||
77d2e05a MKL |
241 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
242 | va_list args) | |
cbd35700 | 243 | { |
a2a7d570 | 244 | unsigned int n; |
cbd35700 | 245 | |
a2a7d570 | 246 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
247 | |
248 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
249 | "verifier log line truncated - local buffer too short\n"); | |
250 | ||
251 | n = min(log->len_total - log->len_used - 1, n); | |
252 | log->kbuf[n] = '\0'; | |
253 | ||
254 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) | |
255 | log->len_used += n; | |
256 | else | |
257 | log->ubuf = NULL; | |
cbd35700 | 258 | } |
abe08840 JO |
259 | |
260 | /* log_level controls verbosity level of eBPF verifier. | |
261 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
262 | * so the user can figure out what's wrong with the program | |
430e68d1 | 263 | */ |
abe08840 JO |
264 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
265 | const char *fmt, ...) | |
266 | { | |
267 | va_list args; | |
268 | ||
77d2e05a MKL |
269 | if (!bpf_verifier_log_needed(&env->log)) |
270 | return; | |
271 | ||
abe08840 | 272 | va_start(args, fmt); |
77d2e05a | 273 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
274 | va_end(args); |
275 | } | |
276 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
277 | ||
278 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
279 | { | |
77d2e05a | 280 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
281 | va_list args; |
282 | ||
77d2e05a MKL |
283 | if (!bpf_verifier_log_needed(&env->log)) |
284 | return; | |
285 | ||
abe08840 | 286 | va_start(args, fmt); |
77d2e05a | 287 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
288 | va_end(args); |
289 | } | |
cbd35700 | 290 | |
d9762e84 MKL |
291 | static const char *ltrim(const char *s) |
292 | { | |
293 | while (isspace(*s)) | |
294 | s++; | |
295 | ||
296 | return s; | |
297 | } | |
298 | ||
299 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
300 | u32 insn_off, | |
301 | const char *prefix_fmt, ...) | |
302 | { | |
303 | const struct bpf_line_info *linfo; | |
304 | ||
305 | if (!bpf_verifier_log_needed(&env->log)) | |
306 | return; | |
307 | ||
308 | linfo = find_linfo(env, insn_off); | |
309 | if (!linfo || linfo == env->prev_linfo) | |
310 | return; | |
311 | ||
312 | if (prefix_fmt) { | |
313 | va_list args; | |
314 | ||
315 | va_start(args, prefix_fmt); | |
316 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
317 | va_end(args); | |
318 | } | |
319 | ||
320 | verbose(env, "%s\n", | |
321 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
322 | linfo->line_off))); | |
323 | ||
324 | env->prev_linfo = linfo; | |
325 | } | |
326 | ||
de8f3a83 DB |
327 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
328 | { | |
329 | return type == PTR_TO_PACKET || | |
330 | type == PTR_TO_PACKET_META; | |
331 | } | |
332 | ||
840b9615 JS |
333 | static bool reg_type_may_be_null(enum bpf_reg_type type) |
334 | { | |
fd978bf7 JS |
335 | return type == PTR_TO_MAP_VALUE_OR_NULL || |
336 | type == PTR_TO_SOCKET_OR_NULL; | |
337 | } | |
338 | ||
339 | static bool type_is_refcounted(enum bpf_reg_type type) | |
340 | { | |
341 | return type == PTR_TO_SOCKET; | |
342 | } | |
343 | ||
344 | static bool type_is_refcounted_or_null(enum bpf_reg_type type) | |
345 | { | |
346 | return type == PTR_TO_SOCKET || type == PTR_TO_SOCKET_OR_NULL; | |
347 | } | |
348 | ||
349 | static bool reg_is_refcounted(const struct bpf_reg_state *reg) | |
350 | { | |
351 | return type_is_refcounted(reg->type); | |
352 | } | |
353 | ||
354 | static bool reg_is_refcounted_or_null(const struct bpf_reg_state *reg) | |
355 | { | |
356 | return type_is_refcounted_or_null(reg->type); | |
357 | } | |
358 | ||
359 | static bool arg_type_is_refcounted(enum bpf_arg_type type) | |
360 | { | |
361 | return type == ARG_PTR_TO_SOCKET; | |
362 | } | |
363 | ||
364 | /* Determine whether the function releases some resources allocated by another | |
365 | * function call. The first reference type argument will be assumed to be | |
366 | * released by release_reference(). | |
367 | */ | |
368 | static bool is_release_function(enum bpf_func_id func_id) | |
369 | { | |
6acc9b43 | 370 | return func_id == BPF_FUNC_sk_release; |
840b9615 JS |
371 | } |
372 | ||
17a52670 AS |
373 | /* string representation of 'enum bpf_reg_type' */ |
374 | static const char * const reg_type_str[] = { | |
375 | [NOT_INIT] = "?", | |
f1174f77 | 376 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
377 | [PTR_TO_CTX] = "ctx", |
378 | [CONST_PTR_TO_MAP] = "map_ptr", | |
379 | [PTR_TO_MAP_VALUE] = "map_value", | |
380 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 381 | [PTR_TO_STACK] = "fp", |
969bf05e | 382 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 383 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 384 | [PTR_TO_PACKET_END] = "pkt_end", |
d58e468b | 385 | [PTR_TO_FLOW_KEYS] = "flow_keys", |
c64b7983 JS |
386 | [PTR_TO_SOCKET] = "sock", |
387 | [PTR_TO_SOCKET_OR_NULL] = "sock_or_null", | |
17a52670 AS |
388 | }; |
389 | ||
8efea21d EC |
390 | static char slot_type_char[] = { |
391 | [STACK_INVALID] = '?', | |
392 | [STACK_SPILL] = 'r', | |
393 | [STACK_MISC] = 'm', | |
394 | [STACK_ZERO] = '0', | |
395 | }; | |
396 | ||
4e92024a AS |
397 | static void print_liveness(struct bpf_verifier_env *env, |
398 | enum bpf_reg_liveness live) | |
399 | { | |
9242b5f5 | 400 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
401 | verbose(env, "_"); |
402 | if (live & REG_LIVE_READ) | |
403 | verbose(env, "r"); | |
404 | if (live & REG_LIVE_WRITTEN) | |
405 | verbose(env, "w"); | |
9242b5f5 AS |
406 | if (live & REG_LIVE_DONE) |
407 | verbose(env, "D"); | |
4e92024a AS |
408 | } |
409 | ||
f4d7e40a AS |
410 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
411 | const struct bpf_reg_state *reg) | |
412 | { | |
413 | struct bpf_verifier_state *cur = env->cur_state; | |
414 | ||
415 | return cur->frame[reg->frameno]; | |
416 | } | |
417 | ||
61bd5218 | 418 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 419 | const struct bpf_func_state *state) |
17a52670 | 420 | { |
f4d7e40a | 421 | const struct bpf_reg_state *reg; |
17a52670 AS |
422 | enum bpf_reg_type t; |
423 | int i; | |
424 | ||
f4d7e40a AS |
425 | if (state->frameno) |
426 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 427 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
428 | reg = &state->regs[i]; |
429 | t = reg->type; | |
17a52670 AS |
430 | if (t == NOT_INIT) |
431 | continue; | |
4e92024a AS |
432 | verbose(env, " R%d", i); |
433 | print_liveness(env, reg->live); | |
434 | verbose(env, "=%s", reg_type_str[t]); | |
f1174f77 EC |
435 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
436 | tnum_is_const(reg->var_off)) { | |
437 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 438 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f4d7e40a AS |
439 | if (t == PTR_TO_STACK) |
440 | verbose(env, ",call_%d", func(env, reg)->callsite); | |
f1174f77 | 441 | } else { |
61bd5218 | 442 | verbose(env, "(id=%d", reg->id); |
f1174f77 | 443 | if (t != SCALAR_VALUE) |
61bd5218 | 444 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 445 | if (type_is_pkt_pointer(t)) |
61bd5218 | 446 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
447 | else if (t == CONST_PTR_TO_MAP || |
448 | t == PTR_TO_MAP_VALUE || | |
449 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 450 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
451 | reg->map_ptr->key_size, |
452 | reg->map_ptr->value_size); | |
7d1238f2 EC |
453 | if (tnum_is_const(reg->var_off)) { |
454 | /* Typically an immediate SCALAR_VALUE, but | |
455 | * could be a pointer whose offset is too big | |
456 | * for reg->off | |
457 | */ | |
61bd5218 | 458 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
459 | } else { |
460 | if (reg->smin_value != reg->umin_value && | |
461 | reg->smin_value != S64_MIN) | |
61bd5218 | 462 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
463 | (long long)reg->smin_value); |
464 | if (reg->smax_value != reg->umax_value && | |
465 | reg->smax_value != S64_MAX) | |
61bd5218 | 466 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
467 | (long long)reg->smax_value); |
468 | if (reg->umin_value != 0) | |
61bd5218 | 469 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
470 | (unsigned long long)reg->umin_value); |
471 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 472 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
473 | (unsigned long long)reg->umax_value); |
474 | if (!tnum_is_unknown(reg->var_off)) { | |
475 | char tn_buf[48]; | |
f1174f77 | 476 | |
7d1238f2 | 477 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 478 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 479 | } |
f1174f77 | 480 | } |
61bd5218 | 481 | verbose(env, ")"); |
f1174f77 | 482 | } |
17a52670 | 483 | } |
638f5b90 | 484 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
485 | char types_buf[BPF_REG_SIZE + 1]; |
486 | bool valid = false; | |
487 | int j; | |
488 | ||
489 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
490 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
491 | valid = true; | |
492 | types_buf[j] = slot_type_char[ | |
493 | state->stack[i].slot_type[j]]; | |
494 | } | |
495 | types_buf[BPF_REG_SIZE] = 0; | |
496 | if (!valid) | |
497 | continue; | |
498 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
499 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
500 | if (state->stack[i].slot_type[0] == STACK_SPILL) | |
4e92024a | 501 | verbose(env, "=%s", |
638f5b90 | 502 | reg_type_str[state->stack[i].spilled_ptr.type]); |
8efea21d EC |
503 | else |
504 | verbose(env, "=%s", types_buf); | |
17a52670 | 505 | } |
fd978bf7 JS |
506 | if (state->acquired_refs && state->refs[0].id) { |
507 | verbose(env, " refs=%d", state->refs[0].id); | |
508 | for (i = 1; i < state->acquired_refs; i++) | |
509 | if (state->refs[i].id) | |
510 | verbose(env, ",%d", state->refs[i].id); | |
511 | } | |
61bd5218 | 512 | verbose(env, "\n"); |
17a52670 AS |
513 | } |
514 | ||
84dbf350 JS |
515 | #define COPY_STATE_FN(NAME, COUNT, FIELD, SIZE) \ |
516 | static int copy_##NAME##_state(struct bpf_func_state *dst, \ | |
517 | const struct bpf_func_state *src) \ | |
518 | { \ | |
519 | if (!src->FIELD) \ | |
520 | return 0; \ | |
521 | if (WARN_ON_ONCE(dst->COUNT < src->COUNT)) { \ | |
522 | /* internal bug, make state invalid to reject the program */ \ | |
523 | memset(dst, 0, sizeof(*dst)); \ | |
524 | return -EFAULT; \ | |
525 | } \ | |
526 | memcpy(dst->FIELD, src->FIELD, \ | |
527 | sizeof(*src->FIELD) * (src->COUNT / SIZE)); \ | |
528 | return 0; \ | |
638f5b90 | 529 | } |
fd978bf7 JS |
530 | /* copy_reference_state() */ |
531 | COPY_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
532 | /* copy_stack_state() */ |
533 | COPY_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
534 | #undef COPY_STATE_FN | |
535 | ||
536 | #define REALLOC_STATE_FN(NAME, COUNT, FIELD, SIZE) \ | |
537 | static int realloc_##NAME##_state(struct bpf_func_state *state, int size, \ | |
538 | bool copy_old) \ | |
539 | { \ | |
540 | u32 old_size = state->COUNT; \ | |
541 | struct bpf_##NAME##_state *new_##FIELD; \ | |
542 | int slot = size / SIZE; \ | |
543 | \ | |
544 | if (size <= old_size || !size) { \ | |
545 | if (copy_old) \ | |
546 | return 0; \ | |
547 | state->COUNT = slot * SIZE; \ | |
548 | if (!size && old_size) { \ | |
549 | kfree(state->FIELD); \ | |
550 | state->FIELD = NULL; \ | |
551 | } \ | |
552 | return 0; \ | |
553 | } \ | |
554 | new_##FIELD = kmalloc_array(slot, sizeof(struct bpf_##NAME##_state), \ | |
555 | GFP_KERNEL); \ | |
556 | if (!new_##FIELD) \ | |
557 | return -ENOMEM; \ | |
558 | if (copy_old) { \ | |
559 | if (state->FIELD) \ | |
560 | memcpy(new_##FIELD, state->FIELD, \ | |
561 | sizeof(*new_##FIELD) * (old_size / SIZE)); \ | |
562 | memset(new_##FIELD + old_size / SIZE, 0, \ | |
563 | sizeof(*new_##FIELD) * (size - old_size) / SIZE); \ | |
564 | } \ | |
565 | state->COUNT = slot * SIZE; \ | |
566 | kfree(state->FIELD); \ | |
567 | state->FIELD = new_##FIELD; \ | |
568 | return 0; \ | |
569 | } | |
fd978bf7 JS |
570 | /* realloc_reference_state() */ |
571 | REALLOC_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
572 | /* realloc_stack_state() */ |
573 | REALLOC_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
574 | #undef REALLOC_STATE_FN | |
638f5b90 AS |
575 | |
576 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
577 | * make it consume minimal amount of memory. check_stack_write() access from | |
f4d7e40a | 578 | * the program calls into realloc_func_state() to grow the stack size. |
84dbf350 JS |
579 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state |
580 | * which realloc_stack_state() copies over. It points to previous | |
581 | * bpf_verifier_state which is never reallocated. | |
638f5b90 | 582 | */ |
fd978bf7 JS |
583 | static int realloc_func_state(struct bpf_func_state *state, int stack_size, |
584 | int refs_size, bool copy_old) | |
638f5b90 | 585 | { |
fd978bf7 JS |
586 | int err = realloc_reference_state(state, refs_size, copy_old); |
587 | if (err) | |
588 | return err; | |
589 | return realloc_stack_state(state, stack_size, copy_old); | |
590 | } | |
591 | ||
592 | /* Acquire a pointer id from the env and update the state->refs to include | |
593 | * this new pointer reference. | |
594 | * On success, returns a valid pointer id to associate with the register | |
595 | * On failure, returns a negative errno. | |
638f5b90 | 596 | */ |
fd978bf7 | 597 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 598 | { |
fd978bf7 JS |
599 | struct bpf_func_state *state = cur_func(env); |
600 | int new_ofs = state->acquired_refs; | |
601 | int id, err; | |
602 | ||
603 | err = realloc_reference_state(state, state->acquired_refs + 1, true); | |
604 | if (err) | |
605 | return err; | |
606 | id = ++env->id_gen; | |
607 | state->refs[new_ofs].id = id; | |
608 | state->refs[new_ofs].insn_idx = insn_idx; | |
638f5b90 | 609 | |
fd978bf7 JS |
610 | return id; |
611 | } | |
612 | ||
613 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
614 | static int __release_reference_state(struct bpf_func_state *state, int ptr_id) | |
615 | { | |
616 | int i, last_idx; | |
617 | ||
618 | if (!ptr_id) | |
619 | return -EFAULT; | |
620 | ||
621 | last_idx = state->acquired_refs - 1; | |
622 | for (i = 0; i < state->acquired_refs; i++) { | |
623 | if (state->refs[i].id == ptr_id) { | |
624 | if (last_idx && i != last_idx) | |
625 | memcpy(&state->refs[i], &state->refs[last_idx], | |
626 | sizeof(*state->refs)); | |
627 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
628 | state->acquired_refs--; | |
638f5b90 | 629 | return 0; |
638f5b90 | 630 | } |
638f5b90 | 631 | } |
fd978bf7 JS |
632 | return -EFAULT; |
633 | } | |
634 | ||
635 | /* variation on the above for cases where we expect that there must be an | |
636 | * outstanding reference for the specified ptr_id. | |
637 | */ | |
638 | static int release_reference_state(struct bpf_verifier_env *env, int ptr_id) | |
639 | { | |
640 | struct bpf_func_state *state = cur_func(env); | |
641 | int err; | |
642 | ||
643 | err = __release_reference_state(state, ptr_id); | |
644 | if (WARN_ON_ONCE(err != 0)) | |
645 | verbose(env, "verifier internal error: can't release reference\n"); | |
646 | return err; | |
647 | } | |
648 | ||
649 | static int transfer_reference_state(struct bpf_func_state *dst, | |
650 | struct bpf_func_state *src) | |
651 | { | |
652 | int err = realloc_reference_state(dst, src->acquired_refs, false); | |
653 | if (err) | |
654 | return err; | |
655 | err = copy_reference_state(dst, src); | |
656 | if (err) | |
657 | return err; | |
638f5b90 AS |
658 | return 0; |
659 | } | |
660 | ||
f4d7e40a AS |
661 | static void free_func_state(struct bpf_func_state *state) |
662 | { | |
5896351e AS |
663 | if (!state) |
664 | return; | |
fd978bf7 | 665 | kfree(state->refs); |
f4d7e40a AS |
666 | kfree(state->stack); |
667 | kfree(state); | |
668 | } | |
669 | ||
1969db47 AS |
670 | static void free_verifier_state(struct bpf_verifier_state *state, |
671 | bool free_self) | |
638f5b90 | 672 | { |
f4d7e40a AS |
673 | int i; |
674 | ||
675 | for (i = 0; i <= state->curframe; i++) { | |
676 | free_func_state(state->frame[i]); | |
677 | state->frame[i] = NULL; | |
678 | } | |
1969db47 AS |
679 | if (free_self) |
680 | kfree(state); | |
638f5b90 AS |
681 | } |
682 | ||
683 | /* copy verifier state from src to dst growing dst stack space | |
684 | * when necessary to accommodate larger src stack | |
685 | */ | |
f4d7e40a AS |
686 | static int copy_func_state(struct bpf_func_state *dst, |
687 | const struct bpf_func_state *src) | |
638f5b90 AS |
688 | { |
689 | int err; | |
690 | ||
fd978bf7 JS |
691 | err = realloc_func_state(dst, src->allocated_stack, src->acquired_refs, |
692 | false); | |
693 | if (err) | |
694 | return err; | |
695 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); | |
696 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
697 | if (err) |
698 | return err; | |
638f5b90 AS |
699 | return copy_stack_state(dst, src); |
700 | } | |
701 | ||
f4d7e40a AS |
702 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
703 | const struct bpf_verifier_state *src) | |
704 | { | |
705 | struct bpf_func_state *dst; | |
706 | int i, err; | |
707 | ||
708 | /* if dst has more stack frames then src frame, free them */ | |
709 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
710 | free_func_state(dst_state->frame[i]); | |
711 | dst_state->frame[i] = NULL; | |
712 | } | |
979d63d5 | 713 | dst_state->speculative = src->speculative; |
f4d7e40a | 714 | dst_state->curframe = src->curframe; |
f4d7e40a AS |
715 | for (i = 0; i <= src->curframe; i++) { |
716 | dst = dst_state->frame[i]; | |
717 | if (!dst) { | |
718 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
719 | if (!dst) | |
720 | return -ENOMEM; | |
721 | dst_state->frame[i] = dst; | |
722 | } | |
723 | err = copy_func_state(dst, src->frame[i]); | |
724 | if (err) | |
725 | return err; | |
726 | } | |
727 | return 0; | |
728 | } | |
729 | ||
638f5b90 AS |
730 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
731 | int *insn_idx) | |
732 | { | |
733 | struct bpf_verifier_state *cur = env->cur_state; | |
734 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
735 | int err; | |
17a52670 AS |
736 | |
737 | if (env->head == NULL) | |
638f5b90 | 738 | return -ENOENT; |
17a52670 | 739 | |
638f5b90 AS |
740 | if (cur) { |
741 | err = copy_verifier_state(cur, &head->st); | |
742 | if (err) | |
743 | return err; | |
744 | } | |
745 | if (insn_idx) | |
746 | *insn_idx = head->insn_idx; | |
17a52670 | 747 | if (prev_insn_idx) |
638f5b90 AS |
748 | *prev_insn_idx = head->prev_insn_idx; |
749 | elem = head->next; | |
1969db47 | 750 | free_verifier_state(&head->st, false); |
638f5b90 | 751 | kfree(head); |
17a52670 AS |
752 | env->head = elem; |
753 | env->stack_size--; | |
638f5b90 | 754 | return 0; |
17a52670 AS |
755 | } |
756 | ||
58e2af8b | 757 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
758 | int insn_idx, int prev_insn_idx, |
759 | bool speculative) | |
17a52670 | 760 | { |
638f5b90 | 761 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 762 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 763 | int err; |
17a52670 | 764 | |
638f5b90 | 765 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
766 | if (!elem) |
767 | goto err; | |
768 | ||
17a52670 AS |
769 | elem->insn_idx = insn_idx; |
770 | elem->prev_insn_idx = prev_insn_idx; | |
771 | elem->next = env->head; | |
772 | env->head = elem; | |
773 | env->stack_size++; | |
1969db47 AS |
774 | err = copy_verifier_state(&elem->st, cur); |
775 | if (err) | |
776 | goto err; | |
979d63d5 | 777 | elem->st.speculative |= speculative; |
07016151 | 778 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
61bd5218 | 779 | verbose(env, "BPF program is too complex\n"); |
17a52670 AS |
780 | goto err; |
781 | } | |
782 | return &elem->st; | |
783 | err: | |
5896351e AS |
784 | free_verifier_state(env->cur_state, true); |
785 | env->cur_state = NULL; | |
17a52670 | 786 | /* pop all elements and return */ |
638f5b90 | 787 | while (!pop_stack(env, NULL, NULL)); |
17a52670 AS |
788 | return NULL; |
789 | } | |
790 | ||
791 | #define CALLER_SAVED_REGS 6 | |
792 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
793 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
794 | }; | |
795 | ||
f1174f77 EC |
796 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
797 | ||
b03c9f9f EC |
798 | /* Mark the unknown part of a register (variable offset or scalar value) as |
799 | * known to have the value @imm. | |
800 | */ | |
801 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
802 | { | |
a9c676bc AS |
803 | /* Clear id, off, and union(map_ptr, range) */ |
804 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
805 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
b03c9f9f EC |
806 | reg->var_off = tnum_const(imm); |
807 | reg->smin_value = (s64)imm; | |
808 | reg->smax_value = (s64)imm; | |
809 | reg->umin_value = imm; | |
810 | reg->umax_value = imm; | |
811 | } | |
812 | ||
f1174f77 EC |
813 | /* Mark the 'variable offset' part of a register as zero. This should be |
814 | * used only on registers holding a pointer type. | |
815 | */ | |
816 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 817 | { |
b03c9f9f | 818 | __mark_reg_known(reg, 0); |
f1174f77 | 819 | } |
a9789ef9 | 820 | |
cc2b14d5 AS |
821 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
822 | { | |
823 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
824 | reg->type = SCALAR_VALUE; |
825 | } | |
826 | ||
61bd5218 JK |
827 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
828 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
829 | { |
830 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 831 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
832 | /* Something bad happened, let's kill all regs */ |
833 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
834 | __mark_reg_not_init(regs + regno); | |
835 | return; | |
836 | } | |
837 | __mark_reg_known_zero(regs + regno); | |
838 | } | |
839 | ||
de8f3a83 DB |
840 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
841 | { | |
842 | return type_is_pkt_pointer(reg->type); | |
843 | } | |
844 | ||
845 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
846 | { | |
847 | return reg_is_pkt_pointer(reg) || | |
848 | reg->type == PTR_TO_PACKET_END; | |
849 | } | |
850 | ||
851 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
852 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
853 | enum bpf_reg_type which) | |
854 | { | |
855 | /* The register can already have a range from prior markings. | |
856 | * This is fine as long as it hasn't been advanced from its | |
857 | * origin. | |
858 | */ | |
859 | return reg->type == which && | |
860 | reg->id == 0 && | |
861 | reg->off == 0 && | |
862 | tnum_equals_const(reg->var_off, 0); | |
863 | } | |
864 | ||
b03c9f9f EC |
865 | /* Attempts to improve min/max values based on var_off information */ |
866 | static void __update_reg_bounds(struct bpf_reg_state *reg) | |
867 | { | |
868 | /* min signed is max(sign bit) | min(other bits) */ | |
869 | reg->smin_value = max_t(s64, reg->smin_value, | |
870 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
871 | /* max signed is min(sign bit) | max(other bits) */ | |
872 | reg->smax_value = min_t(s64, reg->smax_value, | |
873 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
874 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
875 | reg->umax_value = min(reg->umax_value, | |
876 | reg->var_off.value | reg->var_off.mask); | |
877 | } | |
878 | ||
879 | /* Uses signed min/max values to inform unsigned, and vice-versa */ | |
880 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) | |
881 | { | |
882 | /* Learn sign from signed bounds. | |
883 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
884 | * are the same, so combine. This works even in the negative case, e.g. | |
885 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
886 | */ | |
887 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
888 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
889 | reg->umin_value); | |
890 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
891 | reg->umax_value); | |
892 | return; | |
893 | } | |
894 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
895 | * boundary, so we must be careful. | |
896 | */ | |
897 | if ((s64)reg->umax_value >= 0) { | |
898 | /* Positive. We can't learn anything from the smin, but smax | |
899 | * is positive, hence safe. | |
900 | */ | |
901 | reg->smin_value = reg->umin_value; | |
902 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
903 | reg->umax_value); | |
904 | } else if ((s64)reg->umin_value < 0) { | |
905 | /* Negative. We can't learn anything from the smax, but smin | |
906 | * is negative, hence safe. | |
907 | */ | |
908 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
909 | reg->umin_value); | |
910 | reg->smax_value = reg->umax_value; | |
911 | } | |
912 | } | |
913 | ||
914 | /* Attempts to improve var_off based on unsigned min/max information */ | |
915 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
916 | { | |
917 | reg->var_off = tnum_intersect(reg->var_off, | |
918 | tnum_range(reg->umin_value, | |
919 | reg->umax_value)); | |
920 | } | |
921 | ||
922 | /* Reset the min/max bounds of a register */ | |
923 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
924 | { | |
925 | reg->smin_value = S64_MIN; | |
926 | reg->smax_value = S64_MAX; | |
927 | reg->umin_value = 0; | |
928 | reg->umax_value = U64_MAX; | |
929 | } | |
930 | ||
f1174f77 EC |
931 | /* Mark a register as having a completely unknown (scalar) value. */ |
932 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
933 | { | |
a9c676bc AS |
934 | /* |
935 | * Clear type, id, off, and union(map_ptr, range) and | |
936 | * padding between 'type' and union | |
937 | */ | |
938 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 939 | reg->type = SCALAR_VALUE; |
f1174f77 | 940 | reg->var_off = tnum_unknown; |
f4d7e40a | 941 | reg->frameno = 0; |
b03c9f9f | 942 | __mark_reg_unbounded(reg); |
f1174f77 EC |
943 | } |
944 | ||
61bd5218 JK |
945 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
946 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
947 | { |
948 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 949 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
950 | /* Something bad happened, let's kill all regs except FP */ |
951 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
952 | __mark_reg_not_init(regs + regno); |
953 | return; | |
954 | } | |
955 | __mark_reg_unknown(regs + regno); | |
956 | } | |
957 | ||
958 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
959 | { | |
960 | __mark_reg_unknown(reg); | |
961 | reg->type = NOT_INIT; | |
962 | } | |
963 | ||
61bd5218 JK |
964 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
965 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
966 | { |
967 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 968 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
969 | /* Something bad happened, let's kill all regs except FP */ |
970 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
971 | __mark_reg_not_init(regs + regno); |
972 | return; | |
973 | } | |
974 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
975 | } |
976 | ||
61bd5218 | 977 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 978 | struct bpf_func_state *state) |
17a52670 | 979 | { |
f4d7e40a | 980 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
981 | int i; |
982 | ||
dc503a8a | 983 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 984 | mark_reg_not_init(env, regs, i); |
dc503a8a | 985 | regs[i].live = REG_LIVE_NONE; |
679c782d | 986 | regs[i].parent = NULL; |
dc503a8a | 987 | } |
17a52670 AS |
988 | |
989 | /* frame pointer */ | |
f1174f77 | 990 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 991 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 992 | regs[BPF_REG_FP].frameno = state->frameno; |
17a52670 AS |
993 | |
994 | /* 1st arg to a function */ | |
995 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
61bd5218 | 996 | mark_reg_known_zero(env, regs, BPF_REG_1); |
6760bf2d DB |
997 | } |
998 | ||
f4d7e40a AS |
999 | #define BPF_MAIN_FUNC (-1) |
1000 | static void init_func_state(struct bpf_verifier_env *env, | |
1001 | struct bpf_func_state *state, | |
1002 | int callsite, int frameno, int subprogno) | |
1003 | { | |
1004 | state->callsite = callsite; | |
1005 | state->frameno = frameno; | |
1006 | state->subprogno = subprogno; | |
1007 | init_reg_state(env, state); | |
1008 | } | |
1009 | ||
17a52670 AS |
1010 | enum reg_arg_type { |
1011 | SRC_OP, /* register is used as source operand */ | |
1012 | DST_OP, /* register is used as destination operand */ | |
1013 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1014 | }; | |
1015 | ||
cc8b0b92 AS |
1016 | static int cmp_subprogs(const void *a, const void *b) |
1017 | { | |
9c8105bd JW |
1018 | return ((struct bpf_subprog_info *)a)->start - |
1019 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1020 | } |
1021 | ||
1022 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1023 | { | |
9c8105bd | 1024 | struct bpf_subprog_info *p; |
cc8b0b92 | 1025 | |
9c8105bd JW |
1026 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1027 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1028 | if (!p) |
1029 | return -ENOENT; | |
9c8105bd | 1030 | return p - env->subprog_info; |
cc8b0b92 AS |
1031 | |
1032 | } | |
1033 | ||
1034 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1035 | { | |
1036 | int insn_cnt = env->prog->len; | |
1037 | int ret; | |
1038 | ||
1039 | if (off >= insn_cnt || off < 0) { | |
1040 | verbose(env, "call to invalid destination\n"); | |
1041 | return -EINVAL; | |
1042 | } | |
1043 | ret = find_subprog(env, off); | |
1044 | if (ret >= 0) | |
1045 | return 0; | |
4cb3d99c | 1046 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1047 | verbose(env, "too many subprograms\n"); |
1048 | return -E2BIG; | |
1049 | } | |
9c8105bd JW |
1050 | env->subprog_info[env->subprog_cnt++].start = off; |
1051 | sort(env->subprog_info, env->subprog_cnt, | |
1052 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
cc8b0b92 AS |
1053 | return 0; |
1054 | } | |
1055 | ||
1056 | static int check_subprogs(struct bpf_verifier_env *env) | |
1057 | { | |
1058 | int i, ret, subprog_start, subprog_end, off, cur_subprog = 0; | |
9c8105bd | 1059 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 AS |
1060 | struct bpf_insn *insn = env->prog->insnsi; |
1061 | int insn_cnt = env->prog->len; | |
1062 | ||
f910cefa JW |
1063 | /* Add entry function. */ |
1064 | ret = add_subprog(env, 0); | |
1065 | if (ret < 0) | |
1066 | return ret; | |
1067 | ||
cc8b0b92 AS |
1068 | /* determine subprog starts. The end is one before the next starts */ |
1069 | for (i = 0; i < insn_cnt; i++) { | |
1070 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
1071 | continue; | |
1072 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
1073 | continue; | |
1074 | if (!env->allow_ptr_leaks) { | |
1075 | verbose(env, "function calls to other bpf functions are allowed for root only\n"); | |
1076 | return -EPERM; | |
1077 | } | |
cc8b0b92 AS |
1078 | ret = add_subprog(env, i + insn[i].imm + 1); |
1079 | if (ret < 0) | |
1080 | return ret; | |
1081 | } | |
1082 | ||
4cb3d99c JW |
1083 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
1084 | * logic. 'subprog_cnt' should not be increased. | |
1085 | */ | |
1086 | subprog[env->subprog_cnt].start = insn_cnt; | |
1087 | ||
cc8b0b92 AS |
1088 | if (env->log.level > 1) |
1089 | for (i = 0; i < env->subprog_cnt; i++) | |
9c8105bd | 1090 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 AS |
1091 | |
1092 | /* now check that all jumps are within the same subprog */ | |
4cb3d99c JW |
1093 | subprog_start = subprog[cur_subprog].start; |
1094 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
1095 | for (i = 0; i < insn_cnt; i++) { |
1096 | u8 code = insn[i].code; | |
1097 | ||
1098 | if (BPF_CLASS(code) != BPF_JMP) | |
1099 | goto next; | |
1100 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
1101 | goto next; | |
1102 | off = i + insn[i].off + 1; | |
1103 | if (off < subprog_start || off >= subprog_end) { | |
1104 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
1105 | return -EINVAL; | |
1106 | } | |
1107 | next: | |
1108 | if (i == subprog_end - 1) { | |
1109 | /* to avoid fall-through from one subprog into another | |
1110 | * the last insn of the subprog should be either exit | |
1111 | * or unconditional jump back | |
1112 | */ | |
1113 | if (code != (BPF_JMP | BPF_EXIT) && | |
1114 | code != (BPF_JMP | BPF_JA)) { | |
1115 | verbose(env, "last insn is not an exit or jmp\n"); | |
1116 | return -EINVAL; | |
1117 | } | |
1118 | subprog_start = subprog_end; | |
4cb3d99c JW |
1119 | cur_subprog++; |
1120 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 1121 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
1122 | } |
1123 | } | |
1124 | return 0; | |
1125 | } | |
1126 | ||
679c782d EC |
1127 | /* Parentage chain of this register (or stack slot) should take care of all |
1128 | * issues like callee-saved registers, stack slot allocation time, etc. | |
1129 | */ | |
f4d7e40a | 1130 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d EC |
1131 | const struct bpf_reg_state *state, |
1132 | struct bpf_reg_state *parent) | |
f4d7e40a AS |
1133 | { |
1134 | bool writes = parent == state->parent; /* Observe write marks */ | |
dc503a8a EC |
1135 | |
1136 | while (parent) { | |
1137 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 1138 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 1139 | break; |
9242b5f5 AS |
1140 | if (parent->live & REG_LIVE_DONE) { |
1141 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
1142 | reg_type_str[parent->type], | |
1143 | parent->var_off.value, parent->off); | |
1144 | return -EFAULT; | |
1145 | } | |
dc503a8a | 1146 | /* ... then we depend on parent's value */ |
679c782d | 1147 | parent->live |= REG_LIVE_READ; |
dc503a8a EC |
1148 | state = parent; |
1149 | parent = state->parent; | |
f4d7e40a | 1150 | writes = true; |
dc503a8a | 1151 | } |
f4d7e40a | 1152 | return 0; |
dc503a8a EC |
1153 | } |
1154 | ||
1155 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, | |
17a52670 AS |
1156 | enum reg_arg_type t) |
1157 | { | |
f4d7e40a AS |
1158 | struct bpf_verifier_state *vstate = env->cur_state; |
1159 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
1160 | struct bpf_reg_state *regs = state->regs; | |
dc503a8a | 1161 | |
17a52670 | 1162 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 1163 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
1164 | return -EINVAL; |
1165 | } | |
1166 | ||
1167 | if (t == SRC_OP) { | |
1168 | /* check whether register used as source operand can be read */ | |
1169 | if (regs[regno].type == NOT_INIT) { | |
61bd5218 | 1170 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
1171 | return -EACCES; |
1172 | } | |
679c782d EC |
1173 | /* We don't need to worry about FP liveness because it's read-only */ |
1174 | if (regno != BPF_REG_FP) | |
1175 | return mark_reg_read(env, ®s[regno], | |
1176 | regs[regno].parent); | |
17a52670 AS |
1177 | } else { |
1178 | /* check whether register used as dest operand can be written to */ | |
1179 | if (regno == BPF_REG_FP) { | |
61bd5218 | 1180 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
1181 | return -EACCES; |
1182 | } | |
dc503a8a | 1183 | regs[regno].live |= REG_LIVE_WRITTEN; |
17a52670 | 1184 | if (t == DST_OP) |
61bd5218 | 1185 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
1186 | } |
1187 | return 0; | |
1188 | } | |
1189 | ||
1be7f75d AS |
1190 | static bool is_spillable_regtype(enum bpf_reg_type type) |
1191 | { | |
1192 | switch (type) { | |
1193 | case PTR_TO_MAP_VALUE: | |
1194 | case PTR_TO_MAP_VALUE_OR_NULL: | |
1195 | case PTR_TO_STACK: | |
1196 | case PTR_TO_CTX: | |
969bf05e | 1197 | case PTR_TO_PACKET: |
de8f3a83 | 1198 | case PTR_TO_PACKET_META: |
969bf05e | 1199 | case PTR_TO_PACKET_END: |
d58e468b | 1200 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 1201 | case CONST_PTR_TO_MAP: |
c64b7983 JS |
1202 | case PTR_TO_SOCKET: |
1203 | case PTR_TO_SOCKET_OR_NULL: | |
1be7f75d AS |
1204 | return true; |
1205 | default: | |
1206 | return false; | |
1207 | } | |
1208 | } | |
1209 | ||
cc2b14d5 AS |
1210 | /* Does this register contain a constant zero? */ |
1211 | static bool register_is_null(struct bpf_reg_state *reg) | |
1212 | { | |
1213 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
1214 | } | |
1215 | ||
17a52670 AS |
1216 | /* check_stack_read/write functions track spill/fill of registers, |
1217 | * stack boundary and alignment are checked in check_mem_access() | |
1218 | */ | |
61bd5218 | 1219 | static int check_stack_write(struct bpf_verifier_env *env, |
f4d7e40a | 1220 | struct bpf_func_state *state, /* func where register points to */ |
af86ca4e | 1221 | int off, int size, int value_regno, int insn_idx) |
17a52670 | 1222 | { |
f4d7e40a | 1223 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 1224 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
f4d7e40a | 1225 | enum bpf_reg_type type; |
638f5b90 | 1226 | |
f4d7e40a | 1227 | err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE), |
fd978bf7 | 1228 | state->acquired_refs, true); |
638f5b90 AS |
1229 | if (err) |
1230 | return err; | |
9c399760 AS |
1231 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
1232 | * so it's aligned access and [off, off + size) are within stack limits | |
1233 | */ | |
638f5b90 AS |
1234 | if (!env->allow_ptr_leaks && |
1235 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
1236 | size != BPF_REG_SIZE) { | |
1237 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
1238 | return -EACCES; | |
1239 | } | |
17a52670 | 1240 | |
f4d7e40a | 1241 | cur = env->cur_state->frame[env->cur_state->curframe]; |
17a52670 | 1242 | if (value_regno >= 0 && |
f4d7e40a | 1243 | is_spillable_regtype((type = cur->regs[value_regno].type))) { |
17a52670 AS |
1244 | |
1245 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 1246 | if (size != BPF_REG_SIZE) { |
61bd5218 | 1247 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
1248 | return -EACCES; |
1249 | } | |
1250 | ||
f4d7e40a AS |
1251 | if (state != cur && type == PTR_TO_STACK) { |
1252 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); | |
1253 | return -EINVAL; | |
1254 | } | |
1255 | ||
17a52670 | 1256 | /* save register state */ |
f4d7e40a | 1257 | state->stack[spi].spilled_ptr = cur->regs[value_regno]; |
638f5b90 | 1258 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; |
17a52670 | 1259 | |
af86ca4e AS |
1260 | for (i = 0; i < BPF_REG_SIZE; i++) { |
1261 | if (state->stack[spi].slot_type[i] == STACK_MISC && | |
1262 | !env->allow_ptr_leaks) { | |
1263 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; | |
1264 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
1265 | ||
1266 | /* detected reuse of integer stack slot with a pointer | |
1267 | * which means either llvm is reusing stack slot or | |
1268 | * an attacker is trying to exploit CVE-2018-3639 | |
1269 | * (speculative store bypass) | |
1270 | * Have to sanitize that slot with preemptive | |
1271 | * store of zero. | |
1272 | */ | |
1273 | if (*poff && *poff != soff) { | |
1274 | /* disallow programs where single insn stores | |
1275 | * into two different stack slots, since verifier | |
1276 | * cannot sanitize them | |
1277 | */ | |
1278 | verbose(env, | |
1279 | "insn %d cannot access two stack slots fp%d and fp%d", | |
1280 | insn_idx, *poff, soff); | |
1281 | return -EINVAL; | |
1282 | } | |
1283 | *poff = soff; | |
1284 | } | |
638f5b90 | 1285 | state->stack[spi].slot_type[i] = STACK_SPILL; |
af86ca4e | 1286 | } |
9c399760 | 1287 | } else { |
cc2b14d5 AS |
1288 | u8 type = STACK_MISC; |
1289 | ||
679c782d EC |
1290 | /* regular write of data into stack destroys any spilled ptr */ |
1291 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d JW |
1292 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
1293 | if (state->stack[spi].slot_type[0] == STACK_SPILL) | |
1294 | for (i = 0; i < BPF_REG_SIZE; i++) | |
1295 | state->stack[spi].slot_type[i] = STACK_MISC; | |
9c399760 | 1296 | |
cc2b14d5 AS |
1297 | /* only mark the slot as written if all 8 bytes were written |
1298 | * otherwise read propagation may incorrectly stop too soon | |
1299 | * when stack slots are partially written. | |
1300 | * This heuristic means that read propagation will be | |
1301 | * conservative, since it will add reg_live_read marks | |
1302 | * to stack slots all the way to first state when programs | |
1303 | * writes+reads less than 8 bytes | |
1304 | */ | |
1305 | if (size == BPF_REG_SIZE) | |
1306 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1307 | ||
1308 | /* when we zero initialize stack slots mark them as such */ | |
1309 | if (value_regno >= 0 && | |
1310 | register_is_null(&cur->regs[value_regno])) | |
1311 | type = STACK_ZERO; | |
1312 | ||
0bae2d4d | 1313 | /* Mark slots affected by this stack write. */ |
9c399760 | 1314 | for (i = 0; i < size; i++) |
638f5b90 | 1315 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 1316 | type; |
17a52670 AS |
1317 | } |
1318 | return 0; | |
1319 | } | |
1320 | ||
61bd5218 | 1321 | static int check_stack_read(struct bpf_verifier_env *env, |
f4d7e40a AS |
1322 | struct bpf_func_state *reg_state /* func where register points to */, |
1323 | int off, int size, int value_regno) | |
17a52670 | 1324 | { |
f4d7e40a AS |
1325 | struct bpf_verifier_state *vstate = env->cur_state; |
1326 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 AS |
1327 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
1328 | u8 *stype; | |
17a52670 | 1329 | |
f4d7e40a | 1330 | if (reg_state->allocated_stack <= slot) { |
638f5b90 AS |
1331 | verbose(env, "invalid read from stack off %d+0 size %d\n", |
1332 | off, size); | |
1333 | return -EACCES; | |
1334 | } | |
f4d7e40a | 1335 | stype = reg_state->stack[spi].slot_type; |
17a52670 | 1336 | |
638f5b90 | 1337 | if (stype[0] == STACK_SPILL) { |
9c399760 | 1338 | if (size != BPF_REG_SIZE) { |
61bd5218 | 1339 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
1340 | return -EACCES; |
1341 | } | |
9c399760 | 1342 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 1343 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 1344 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
1345 | return -EACCES; |
1346 | } | |
1347 | } | |
1348 | ||
dc503a8a | 1349 | if (value_regno >= 0) { |
17a52670 | 1350 | /* restore register state from stack */ |
f4d7e40a | 1351 | state->regs[value_regno] = reg_state->stack[spi].spilled_ptr; |
2f18f62e AS |
1352 | /* mark reg as written since spilled pointer state likely |
1353 | * has its liveness marks cleared by is_state_visited() | |
1354 | * which resets stack/reg liveness for state transitions | |
1355 | */ | |
1356 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
dc503a8a | 1357 | } |
679c782d EC |
1358 | mark_reg_read(env, ®_state->stack[spi].spilled_ptr, |
1359 | reg_state->stack[spi].spilled_ptr.parent); | |
17a52670 AS |
1360 | return 0; |
1361 | } else { | |
cc2b14d5 AS |
1362 | int zeros = 0; |
1363 | ||
17a52670 | 1364 | for (i = 0; i < size; i++) { |
cc2b14d5 AS |
1365 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_MISC) |
1366 | continue; | |
1367 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_ZERO) { | |
1368 | zeros++; | |
1369 | continue; | |
17a52670 | 1370 | } |
cc2b14d5 AS |
1371 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
1372 | off, i, size); | |
1373 | return -EACCES; | |
1374 | } | |
679c782d EC |
1375 | mark_reg_read(env, ®_state->stack[spi].spilled_ptr, |
1376 | reg_state->stack[spi].spilled_ptr.parent); | |
cc2b14d5 AS |
1377 | if (value_regno >= 0) { |
1378 | if (zeros == size) { | |
1379 | /* any size read into register is zero extended, | |
1380 | * so the whole register == const_zero | |
1381 | */ | |
1382 | __mark_reg_const_zero(&state->regs[value_regno]); | |
1383 | } else { | |
1384 | /* have read misc data from the stack */ | |
1385 | mark_reg_unknown(env, state->regs, value_regno); | |
1386 | } | |
1387 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
17a52670 | 1388 | } |
17a52670 AS |
1389 | return 0; |
1390 | } | |
1391 | } | |
1392 | ||
e4298d25 DB |
1393 | static int check_stack_access(struct bpf_verifier_env *env, |
1394 | const struct bpf_reg_state *reg, | |
1395 | int off, int size) | |
1396 | { | |
1397 | /* Stack accesses must be at a fixed offset, so that we | |
1398 | * can determine what type of data were returned. See | |
1399 | * check_stack_read(). | |
1400 | */ | |
1401 | if (!tnum_is_const(reg->var_off)) { | |
1402 | char tn_buf[48]; | |
1403 | ||
1404 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1405 | verbose(env, "variable stack access var_off=%s off=%d size=%d", | |
1406 | tn_buf, off, size); | |
1407 | return -EACCES; | |
1408 | } | |
1409 | ||
1410 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
1411 | verbose(env, "invalid stack off=%d size=%d\n", off, size); | |
1412 | return -EACCES; | |
1413 | } | |
1414 | ||
1415 | return 0; | |
1416 | } | |
1417 | ||
17a52670 | 1418 | /* check read/write into map element returned by bpf_map_lookup_elem() */ |
f1174f77 | 1419 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1420 | int size, bool zero_size_allowed) |
17a52670 | 1421 | { |
638f5b90 AS |
1422 | struct bpf_reg_state *regs = cur_regs(env); |
1423 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 1424 | |
9fd29c08 YS |
1425 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1426 | off + size > map->value_size) { | |
61bd5218 | 1427 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
1428 | map->value_size, off, size); |
1429 | return -EACCES; | |
1430 | } | |
1431 | return 0; | |
1432 | } | |
1433 | ||
f1174f77 EC |
1434 | /* check read/write into a map element with possible variable offset */ |
1435 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 1436 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 1437 | { |
f4d7e40a AS |
1438 | struct bpf_verifier_state *vstate = env->cur_state; |
1439 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
1440 | struct bpf_reg_state *reg = &state->regs[regno]; |
1441 | int err; | |
1442 | ||
f1174f77 EC |
1443 | /* We may have adjusted the register to this map value, so we |
1444 | * need to try adding each of min_value and max_value to off | |
1445 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 1446 | */ |
61bd5218 JK |
1447 | if (env->log.level) |
1448 | print_verifier_state(env, state); | |
b7137c4e | 1449 | |
dbcfe5f7 GB |
1450 | /* The minimum value is only important with signed |
1451 | * comparisons where we can't assume the floor of a | |
1452 | * value is 0. If we are using signed variables for our | |
1453 | * index'es we need to make sure that whatever we use | |
1454 | * will have a set floor within our range. | |
1455 | */ | |
b7137c4e DB |
1456 | if (reg->smin_value < 0 && |
1457 | (reg->smin_value == S64_MIN || | |
1458 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
1459 | reg->smin_value + off < 0)) { | |
61bd5218 | 1460 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
1461 | regno); |
1462 | return -EACCES; | |
1463 | } | |
9fd29c08 YS |
1464 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
1465 | zero_size_allowed); | |
dbcfe5f7 | 1466 | if (err) { |
61bd5218 JK |
1467 | verbose(env, "R%d min value is outside of the array range\n", |
1468 | regno); | |
dbcfe5f7 GB |
1469 | return err; |
1470 | } | |
1471 | ||
b03c9f9f EC |
1472 | /* If we haven't set a max value then we need to bail since we can't be |
1473 | * sure we won't do bad things. | |
1474 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 1475 | */ |
b03c9f9f | 1476 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 1477 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
1478 | regno); |
1479 | return -EACCES; | |
1480 | } | |
9fd29c08 YS |
1481 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
1482 | zero_size_allowed); | |
f1174f77 | 1483 | if (err) |
61bd5218 JK |
1484 | verbose(env, "R%d max value is outside of the array range\n", |
1485 | regno); | |
f1174f77 | 1486 | return err; |
dbcfe5f7 GB |
1487 | } |
1488 | ||
969bf05e AS |
1489 | #define MAX_PACKET_OFF 0xffff |
1490 | ||
58e2af8b | 1491 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
1492 | const struct bpf_call_arg_meta *meta, |
1493 | enum bpf_access_type t) | |
4acf6c0b | 1494 | { |
36bbef52 | 1495 | switch (env->prog->type) { |
5d66fa7d | 1496 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
1497 | case BPF_PROG_TYPE_LWT_IN: |
1498 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 1499 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 1500 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 1501 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 1502 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
1503 | if (t == BPF_WRITE) |
1504 | return false; | |
7e57fbb2 | 1505 | /* fallthrough */ |
5d66fa7d DB |
1506 | |
1507 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
1508 | case BPF_PROG_TYPE_SCHED_CLS: |
1509 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 1510 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 1511 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 1512 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 1513 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
1514 | if (meta) |
1515 | return meta->pkt_access; | |
1516 | ||
1517 | env->seen_direct_write = true; | |
4acf6c0b BB |
1518 | return true; |
1519 | default: | |
1520 | return false; | |
1521 | } | |
1522 | } | |
1523 | ||
f1174f77 | 1524 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 1525 | int off, int size, bool zero_size_allowed) |
969bf05e | 1526 | { |
638f5b90 | 1527 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 1528 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 1529 | |
9fd29c08 YS |
1530 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1531 | (u64)off + size > reg->range) { | |
61bd5218 | 1532 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 1533 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
1534 | return -EACCES; |
1535 | } | |
1536 | return 0; | |
1537 | } | |
1538 | ||
f1174f77 | 1539 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1540 | int size, bool zero_size_allowed) |
f1174f77 | 1541 | { |
638f5b90 | 1542 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
1543 | struct bpf_reg_state *reg = ®s[regno]; |
1544 | int err; | |
1545 | ||
1546 | /* We may have added a variable offset to the packet pointer; but any | |
1547 | * reg->range we have comes after that. We are only checking the fixed | |
1548 | * offset. | |
1549 | */ | |
1550 | ||
1551 | /* We don't allow negative numbers, because we aren't tracking enough | |
1552 | * detail to prove they're safe. | |
1553 | */ | |
b03c9f9f | 1554 | if (reg->smin_value < 0) { |
61bd5218 | 1555 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
1556 | regno); |
1557 | return -EACCES; | |
1558 | } | |
9fd29c08 | 1559 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 1560 | if (err) { |
61bd5218 | 1561 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
1562 | return err; |
1563 | } | |
e647815a JW |
1564 | |
1565 | /* __check_packet_access has made sure "off + size - 1" is within u16. | |
1566 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, | |
1567 | * otherwise find_good_pkt_pointers would have refused to set range info | |
1568 | * that __check_packet_access would have rejected this pkt access. | |
1569 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. | |
1570 | */ | |
1571 | env->prog->aux->max_pkt_offset = | |
1572 | max_t(u32, env->prog->aux->max_pkt_offset, | |
1573 | off + reg->umax_value + size - 1); | |
1574 | ||
f1174f77 EC |
1575 | return err; |
1576 | } | |
1577 | ||
1578 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 1579 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 1580 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 1581 | { |
f96da094 DB |
1582 | struct bpf_insn_access_aux info = { |
1583 | .reg_type = *reg_type, | |
1584 | }; | |
31fd8581 | 1585 | |
4f9218aa | 1586 | if (env->ops->is_valid_access && |
5e43f899 | 1587 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
1588 | /* A non zero info.ctx_field_size indicates that this field is a |
1589 | * candidate for later verifier transformation to load the whole | |
1590 | * field and then apply a mask when accessed with a narrower | |
1591 | * access than actual ctx access size. A zero info.ctx_field_size | |
1592 | * will only allow for whole field access and rejects any other | |
1593 | * type of narrower access. | |
31fd8581 | 1594 | */ |
23994631 | 1595 | *reg_type = info.reg_type; |
31fd8581 | 1596 | |
4f9218aa | 1597 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
1598 | /* remember the offset of last byte accessed in ctx */ |
1599 | if (env->prog->aux->max_ctx_offset < off + size) | |
1600 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 1601 | return 0; |
32bbe007 | 1602 | } |
17a52670 | 1603 | |
61bd5218 | 1604 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
1605 | return -EACCES; |
1606 | } | |
1607 | ||
d58e468b PP |
1608 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
1609 | int size) | |
1610 | { | |
1611 | if (size < 0 || off < 0 || | |
1612 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
1613 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
1614 | off, size); | |
1615 | return -EACCES; | |
1616 | } | |
1617 | return 0; | |
1618 | } | |
1619 | ||
c64b7983 JS |
1620 | static int check_sock_access(struct bpf_verifier_env *env, u32 regno, int off, |
1621 | int size, enum bpf_access_type t) | |
1622 | { | |
1623 | struct bpf_reg_state *regs = cur_regs(env); | |
1624 | struct bpf_reg_state *reg = ®s[regno]; | |
1625 | struct bpf_insn_access_aux info; | |
1626 | ||
1627 | if (reg->smin_value < 0) { | |
1628 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
1629 | regno); | |
1630 | return -EACCES; | |
1631 | } | |
1632 | ||
1633 | if (!bpf_sock_is_valid_access(off, size, t, &info)) { | |
1634 | verbose(env, "invalid bpf_sock access off=%d size=%d\n", | |
1635 | off, size); | |
1636 | return -EACCES; | |
1637 | } | |
1638 | ||
1639 | return 0; | |
1640 | } | |
1641 | ||
4cabc5b1 DB |
1642 | static bool __is_pointer_value(bool allow_ptr_leaks, |
1643 | const struct bpf_reg_state *reg) | |
1be7f75d | 1644 | { |
4cabc5b1 | 1645 | if (allow_ptr_leaks) |
1be7f75d AS |
1646 | return false; |
1647 | ||
f1174f77 | 1648 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
1649 | } |
1650 | ||
2a159c6f DB |
1651 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) |
1652 | { | |
1653 | return cur_regs(env) + regno; | |
1654 | } | |
1655 | ||
4cabc5b1 DB |
1656 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1657 | { | |
2a159c6f | 1658 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
1659 | } |
1660 | ||
f37a8cb8 DB |
1661 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
1662 | { | |
2a159c6f | 1663 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 1664 | |
fd978bf7 JS |
1665 | return reg->type == PTR_TO_CTX || |
1666 | reg->type == PTR_TO_SOCKET; | |
f37a8cb8 DB |
1667 | } |
1668 | ||
ca369602 DB |
1669 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
1670 | { | |
2a159c6f | 1671 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
1672 | |
1673 | return type_is_pkt_pointer(reg->type); | |
1674 | } | |
1675 | ||
4b5defde DB |
1676 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
1677 | { | |
1678 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
1679 | ||
1680 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
1681 | return reg->type == PTR_TO_FLOW_KEYS; | |
1682 | } | |
1683 | ||
61bd5218 JK |
1684 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
1685 | const struct bpf_reg_state *reg, | |
d1174416 | 1686 | int off, int size, bool strict) |
969bf05e | 1687 | { |
f1174f77 | 1688 | struct tnum reg_off; |
e07b98d9 | 1689 | int ip_align; |
d1174416 DM |
1690 | |
1691 | /* Byte size accesses are always allowed. */ | |
1692 | if (!strict || size == 1) | |
1693 | return 0; | |
1694 | ||
e4eda884 DM |
1695 | /* For platforms that do not have a Kconfig enabling |
1696 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
1697 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
1698 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
1699 | * to this code only in strict mode where we want to emulate | |
1700 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
1701 | * unconditional IP align value of '2'. | |
e07b98d9 | 1702 | */ |
e4eda884 | 1703 | ip_align = 2; |
f1174f77 EC |
1704 | |
1705 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1706 | if (!tnum_is_aligned(reg_off, size)) { | |
1707 | char tn_buf[48]; | |
1708 | ||
1709 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1710 | verbose(env, |
1711 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1712 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1713 | return -EACCES; |
1714 | } | |
79adffcd | 1715 | |
969bf05e AS |
1716 | return 0; |
1717 | } | |
1718 | ||
61bd5218 JK |
1719 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1720 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1721 | const char *pointer_desc, |
1722 | int off, int size, bool strict) | |
79adffcd | 1723 | { |
f1174f77 EC |
1724 | struct tnum reg_off; |
1725 | ||
1726 | /* Byte size accesses are always allowed. */ | |
1727 | if (!strict || size == 1) | |
1728 | return 0; | |
1729 | ||
1730 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1731 | if (!tnum_is_aligned(reg_off, size)) { | |
1732 | char tn_buf[48]; | |
1733 | ||
1734 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 1735 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 1736 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
1737 | return -EACCES; |
1738 | } | |
1739 | ||
969bf05e AS |
1740 | return 0; |
1741 | } | |
1742 | ||
e07b98d9 | 1743 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
1744 | const struct bpf_reg_state *reg, int off, |
1745 | int size, bool strict_alignment_once) | |
79adffcd | 1746 | { |
ca369602 | 1747 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 1748 | const char *pointer_desc = ""; |
d1174416 | 1749 | |
79adffcd DB |
1750 | switch (reg->type) { |
1751 | case PTR_TO_PACKET: | |
de8f3a83 DB |
1752 | case PTR_TO_PACKET_META: |
1753 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
1754 | * right in front, treat it the very same way. | |
1755 | */ | |
61bd5218 | 1756 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
1757 | case PTR_TO_FLOW_KEYS: |
1758 | pointer_desc = "flow keys "; | |
1759 | break; | |
f1174f77 EC |
1760 | case PTR_TO_MAP_VALUE: |
1761 | pointer_desc = "value "; | |
1762 | break; | |
1763 | case PTR_TO_CTX: | |
1764 | pointer_desc = "context "; | |
1765 | break; | |
1766 | case PTR_TO_STACK: | |
1767 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
1768 | /* The stack spill tracking logic in check_stack_write() |
1769 | * and check_stack_read() relies on stack accesses being | |
1770 | * aligned. | |
1771 | */ | |
1772 | strict = true; | |
f1174f77 | 1773 | break; |
c64b7983 JS |
1774 | case PTR_TO_SOCKET: |
1775 | pointer_desc = "sock "; | |
1776 | break; | |
79adffcd | 1777 | default: |
f1174f77 | 1778 | break; |
79adffcd | 1779 | } |
61bd5218 JK |
1780 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
1781 | strict); | |
79adffcd DB |
1782 | } |
1783 | ||
f4d7e40a AS |
1784 | static int update_stack_depth(struct bpf_verifier_env *env, |
1785 | const struct bpf_func_state *func, | |
1786 | int off) | |
1787 | { | |
9c8105bd | 1788 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
1789 | |
1790 | if (stack >= -off) | |
1791 | return 0; | |
1792 | ||
1793 | /* update known max for given subprogram */ | |
9c8105bd | 1794 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
1795 | return 0; |
1796 | } | |
f4d7e40a | 1797 | |
70a87ffe AS |
1798 | /* starting from main bpf function walk all instructions of the function |
1799 | * and recursively walk all callees that given function can call. | |
1800 | * Ignore jump and exit insns. | |
1801 | * Since recursion is prevented by check_cfg() this algorithm | |
1802 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
1803 | */ | |
1804 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
1805 | { | |
9c8105bd JW |
1806 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
1807 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 1808 | struct bpf_insn *insn = env->prog->insnsi; |
70a87ffe AS |
1809 | int ret_insn[MAX_CALL_FRAMES]; |
1810 | int ret_prog[MAX_CALL_FRAMES]; | |
f4d7e40a | 1811 | |
70a87ffe AS |
1812 | process_func: |
1813 | /* round up to 32-bytes, since this is granularity | |
1814 | * of interpreter stack size | |
1815 | */ | |
9c8105bd | 1816 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 1817 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 1818 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 1819 | frame + 1, depth); |
f4d7e40a AS |
1820 | return -EACCES; |
1821 | } | |
70a87ffe | 1822 | continue_func: |
4cb3d99c | 1823 | subprog_end = subprog[idx + 1].start; |
70a87ffe AS |
1824 | for (; i < subprog_end; i++) { |
1825 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
1826 | continue; | |
1827 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
1828 | continue; | |
1829 | /* remember insn and function to return to */ | |
1830 | ret_insn[frame] = i + 1; | |
9c8105bd | 1831 | ret_prog[frame] = idx; |
70a87ffe AS |
1832 | |
1833 | /* find the callee */ | |
1834 | i = i + insn[i].imm + 1; | |
9c8105bd JW |
1835 | idx = find_subprog(env, i); |
1836 | if (idx < 0) { | |
70a87ffe AS |
1837 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
1838 | i); | |
1839 | return -EFAULT; | |
1840 | } | |
70a87ffe AS |
1841 | frame++; |
1842 | if (frame >= MAX_CALL_FRAMES) { | |
1843 | WARN_ONCE(1, "verifier bug. Call stack is too deep\n"); | |
1844 | return -EFAULT; | |
1845 | } | |
1846 | goto process_func; | |
1847 | } | |
1848 | /* end of for() loop means the last insn of the 'subprog' | |
1849 | * was reached. Doesn't matter whether it was JA or EXIT | |
1850 | */ | |
1851 | if (frame == 0) | |
1852 | return 0; | |
9c8105bd | 1853 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
1854 | frame--; |
1855 | i = ret_insn[frame]; | |
9c8105bd | 1856 | idx = ret_prog[frame]; |
70a87ffe | 1857 | goto continue_func; |
f4d7e40a AS |
1858 | } |
1859 | ||
19d28fbd | 1860 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
1861 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
1862 | const struct bpf_insn *insn, int idx) | |
1863 | { | |
1864 | int start = idx + insn->imm + 1, subprog; | |
1865 | ||
1866 | subprog = find_subprog(env, start); | |
1867 | if (subprog < 0) { | |
1868 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
1869 | start); | |
1870 | return -EFAULT; | |
1871 | } | |
9c8105bd | 1872 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 1873 | } |
19d28fbd | 1874 | #endif |
1ea47e01 | 1875 | |
58990d1f DB |
1876 | static int check_ctx_reg(struct bpf_verifier_env *env, |
1877 | const struct bpf_reg_state *reg, int regno) | |
1878 | { | |
1879 | /* Access to ctx or passing it to a helper is only allowed in | |
1880 | * its original, unmodified form. | |
1881 | */ | |
1882 | ||
1883 | if (reg->off) { | |
1884 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
1885 | regno, reg->off); | |
1886 | return -EACCES; | |
1887 | } | |
1888 | ||
1889 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
1890 | char tn_buf[48]; | |
1891 | ||
1892 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1893 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
1894 | return -EACCES; | |
1895 | } | |
1896 | ||
1897 | return 0; | |
1898 | } | |
1899 | ||
0c17d1d2 JH |
1900 | /* truncate register to smaller size (in bytes) |
1901 | * must be called with size < BPF_REG_SIZE | |
1902 | */ | |
1903 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
1904 | { | |
1905 | u64 mask; | |
1906 | ||
1907 | /* clear high bits in bit representation */ | |
1908 | reg->var_off = tnum_cast(reg->var_off, size); | |
1909 | ||
1910 | /* fix arithmetic bounds */ | |
1911 | mask = ((u64)1 << (size * 8)) - 1; | |
1912 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
1913 | reg->umin_value &= mask; | |
1914 | reg->umax_value &= mask; | |
1915 | } else { | |
1916 | reg->umin_value = 0; | |
1917 | reg->umax_value = mask; | |
1918 | } | |
1919 | reg->smin_value = reg->umin_value; | |
1920 | reg->smax_value = reg->umax_value; | |
1921 | } | |
1922 | ||
17a52670 AS |
1923 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
1924 | * if t==write, value_regno is a register which value is stored into memory | |
1925 | * if t==read, value_regno is a register which will receive the value from memory | |
1926 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
1927 | * if t==read && value_regno==-1, don't care what we read from memory | |
1928 | */ | |
ca369602 DB |
1929 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
1930 | int off, int bpf_size, enum bpf_access_type t, | |
1931 | int value_regno, bool strict_alignment_once) | |
17a52670 | 1932 | { |
638f5b90 AS |
1933 | struct bpf_reg_state *regs = cur_regs(env); |
1934 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 1935 | struct bpf_func_state *state; |
17a52670 AS |
1936 | int size, err = 0; |
1937 | ||
1938 | size = bpf_size_to_bytes(bpf_size); | |
1939 | if (size < 0) | |
1940 | return size; | |
1941 | ||
f1174f77 | 1942 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 1943 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
1944 | if (err) |
1945 | return err; | |
17a52670 | 1946 | |
f1174f77 EC |
1947 | /* for access checks, reg->off is just part of off */ |
1948 | off += reg->off; | |
1949 | ||
1950 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
1951 | if (t == BPF_WRITE && value_regno >= 0 && |
1952 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1953 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
1954 | return -EACCES; |
1955 | } | |
48461135 | 1956 | |
9fd29c08 | 1957 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 1958 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 1959 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 1960 | |
1a0dc1ac | 1961 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 1962 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 1963 | |
1be7f75d AS |
1964 | if (t == BPF_WRITE && value_regno >= 0 && |
1965 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 1966 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
1967 | return -EACCES; |
1968 | } | |
f1174f77 | 1969 | |
58990d1f DB |
1970 | err = check_ctx_reg(env, reg, regno); |
1971 | if (err < 0) | |
1972 | return err; | |
1973 | ||
31fd8581 | 1974 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 1975 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 1976 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
1977 | * PTR_TO_PACKET[_META,_END]. In the latter |
1978 | * case, we know the offset is zero. | |
f1174f77 EC |
1979 | */ |
1980 | if (reg_type == SCALAR_VALUE) | |
638f5b90 | 1981 | mark_reg_unknown(env, regs, value_regno); |
f1174f77 | 1982 | else |
638f5b90 | 1983 | mark_reg_known_zero(env, regs, |
61bd5218 | 1984 | value_regno); |
638f5b90 | 1985 | regs[value_regno].type = reg_type; |
969bf05e | 1986 | } |
17a52670 | 1987 | |
f1174f77 | 1988 | } else if (reg->type == PTR_TO_STACK) { |
f1174f77 | 1989 | off += reg->var_off.value; |
e4298d25 DB |
1990 | err = check_stack_access(env, reg, off, size); |
1991 | if (err) | |
1992 | return err; | |
8726679a | 1993 | |
f4d7e40a AS |
1994 | state = func(env, reg); |
1995 | err = update_stack_depth(env, state, off); | |
1996 | if (err) | |
1997 | return err; | |
8726679a | 1998 | |
638f5b90 | 1999 | if (t == BPF_WRITE) |
61bd5218 | 2000 | err = check_stack_write(env, state, off, size, |
af86ca4e | 2001 | value_regno, insn_idx); |
638f5b90 | 2002 | else |
61bd5218 JK |
2003 | err = check_stack_read(env, state, off, size, |
2004 | value_regno); | |
de8f3a83 | 2005 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 2006 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 2007 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
2008 | return -EACCES; |
2009 | } | |
4acf6c0b BB |
2010 | if (t == BPF_WRITE && value_regno >= 0 && |
2011 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
2012 | verbose(env, "R%d leaks addr into packet\n", |
2013 | value_regno); | |
4acf6c0b BB |
2014 | return -EACCES; |
2015 | } | |
9fd29c08 | 2016 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 2017 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 2018 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
2019 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
2020 | if (t == BPF_WRITE && value_regno >= 0 && | |
2021 | is_pointer_value(env, value_regno)) { | |
2022 | verbose(env, "R%d leaks addr into flow keys\n", | |
2023 | value_regno); | |
2024 | return -EACCES; | |
2025 | } | |
2026 | ||
2027 | err = check_flow_keys_access(env, off, size); | |
2028 | if (!err && t == BPF_READ && value_regno >= 0) | |
2029 | mark_reg_unknown(env, regs, value_regno); | |
c64b7983 JS |
2030 | } else if (reg->type == PTR_TO_SOCKET) { |
2031 | if (t == BPF_WRITE) { | |
2032 | verbose(env, "cannot write into socket\n"); | |
2033 | return -EACCES; | |
2034 | } | |
2035 | err = check_sock_access(env, regno, off, size, t); | |
2036 | if (!err && value_regno >= 0) | |
2037 | mark_reg_unknown(env, regs, value_regno); | |
17a52670 | 2038 | } else { |
61bd5218 JK |
2039 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
2040 | reg_type_str[reg->type]); | |
17a52670 AS |
2041 | return -EACCES; |
2042 | } | |
969bf05e | 2043 | |
f1174f77 | 2044 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 2045 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 2046 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 2047 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 2048 | } |
17a52670 AS |
2049 | return err; |
2050 | } | |
2051 | ||
31fd8581 | 2052 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 2053 | { |
17a52670 AS |
2054 | int err; |
2055 | ||
2056 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
2057 | insn->imm != 0) { | |
61bd5218 | 2058 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
2059 | return -EINVAL; |
2060 | } | |
2061 | ||
2062 | /* check src1 operand */ | |
dc503a8a | 2063 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2064 | if (err) |
2065 | return err; | |
2066 | ||
2067 | /* check src2 operand */ | |
dc503a8a | 2068 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2069 | if (err) |
2070 | return err; | |
2071 | ||
6bdf6abc | 2072 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 2073 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
2074 | return -EACCES; |
2075 | } | |
2076 | ||
ca369602 | 2077 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde DB |
2078 | is_pkt_reg(env, insn->dst_reg) || |
2079 | is_flow_key_reg(env, insn->dst_reg)) { | |
ca369602 | 2080 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", |
2a159c6f DB |
2081 | insn->dst_reg, |
2082 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
2083 | return -EACCES; |
2084 | } | |
2085 | ||
17a52670 | 2086 | /* check whether atomic_add can read the memory */ |
31fd8581 | 2087 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 2088 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
2089 | if (err) |
2090 | return err; | |
2091 | ||
2092 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 2093 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 2094 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
2095 | } |
2096 | ||
2097 | /* when register 'regno' is passed into function that will read 'access_size' | |
2098 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
2099 | * and all elements of stack are initialized. |
2100 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
2101 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 2102 | */ |
58e2af8b | 2103 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
2104 | int access_size, bool zero_size_allowed, |
2105 | struct bpf_call_arg_meta *meta) | |
17a52670 | 2106 | { |
2a159c6f | 2107 | struct bpf_reg_state *reg = reg_state(env, regno); |
f4d7e40a | 2108 | struct bpf_func_state *state = func(env, reg); |
638f5b90 | 2109 | int off, i, slot, spi; |
17a52670 | 2110 | |
914cb781 | 2111 | if (reg->type != PTR_TO_STACK) { |
f1174f77 | 2112 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 2113 | if (zero_size_allowed && access_size == 0 && |
914cb781 | 2114 | register_is_null(reg)) |
8e2fe1d9 DB |
2115 | return 0; |
2116 | ||
61bd5218 | 2117 | verbose(env, "R%d type=%s expected=%s\n", regno, |
914cb781 | 2118 | reg_type_str[reg->type], |
8e2fe1d9 | 2119 | reg_type_str[PTR_TO_STACK]); |
17a52670 | 2120 | return -EACCES; |
8e2fe1d9 | 2121 | } |
17a52670 | 2122 | |
f1174f77 | 2123 | /* Only allow fixed-offset stack reads */ |
914cb781 | 2124 | if (!tnum_is_const(reg->var_off)) { |
f1174f77 EC |
2125 | char tn_buf[48]; |
2126 | ||
914cb781 | 2127 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 2128 | verbose(env, "invalid variable stack read R%d var_off=%s\n", |
f1174f77 | 2129 | regno, tn_buf); |
ea25f914 | 2130 | return -EACCES; |
f1174f77 | 2131 | } |
914cb781 | 2132 | off = reg->off + reg->var_off.value; |
17a52670 | 2133 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || |
9fd29c08 | 2134 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { |
61bd5218 | 2135 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", |
17a52670 AS |
2136 | regno, off, access_size); |
2137 | return -EACCES; | |
2138 | } | |
2139 | ||
435faee1 DB |
2140 | if (meta && meta->raw_mode) { |
2141 | meta->access_size = access_size; | |
2142 | meta->regno = regno; | |
2143 | return 0; | |
2144 | } | |
2145 | ||
17a52670 | 2146 | for (i = 0; i < access_size; i++) { |
cc2b14d5 AS |
2147 | u8 *stype; |
2148 | ||
638f5b90 AS |
2149 | slot = -(off + i) - 1; |
2150 | spi = slot / BPF_REG_SIZE; | |
cc2b14d5 AS |
2151 | if (state->allocated_stack <= slot) |
2152 | goto err; | |
2153 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
2154 | if (*stype == STACK_MISC) | |
2155 | goto mark; | |
2156 | if (*stype == STACK_ZERO) { | |
2157 | /* helper can write anything into the stack */ | |
2158 | *stype = STACK_MISC; | |
2159 | goto mark; | |
17a52670 | 2160 | } |
cc2b14d5 AS |
2161 | err: |
2162 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", | |
2163 | off, i, access_size); | |
2164 | return -EACCES; | |
2165 | mark: | |
2166 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
2167 | * the whole slot to be marked as 'read' | |
2168 | */ | |
679c782d EC |
2169 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
2170 | state->stack[spi].spilled_ptr.parent); | |
17a52670 | 2171 | } |
f4d7e40a | 2172 | return update_stack_depth(env, state, off); |
17a52670 AS |
2173 | } |
2174 | ||
06c1c049 GB |
2175 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
2176 | int access_size, bool zero_size_allowed, | |
2177 | struct bpf_call_arg_meta *meta) | |
2178 | { | |
638f5b90 | 2179 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 2180 | |
f1174f77 | 2181 | switch (reg->type) { |
06c1c049 | 2182 | case PTR_TO_PACKET: |
de8f3a83 | 2183 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
2184 | return check_packet_access(env, regno, reg->off, access_size, |
2185 | zero_size_allowed); | |
06c1c049 | 2186 | case PTR_TO_MAP_VALUE: |
9fd29c08 YS |
2187 | return check_map_access(env, regno, reg->off, access_size, |
2188 | zero_size_allowed); | |
f1174f77 | 2189 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
2190 | return check_stack_boundary(env, regno, access_size, |
2191 | zero_size_allowed, meta); | |
2192 | } | |
2193 | } | |
2194 | ||
90133415 DB |
2195 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
2196 | { | |
2197 | return type == ARG_PTR_TO_MEM || | |
2198 | type == ARG_PTR_TO_MEM_OR_NULL || | |
2199 | type == ARG_PTR_TO_UNINIT_MEM; | |
2200 | } | |
2201 | ||
2202 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
2203 | { | |
2204 | return type == ARG_CONST_SIZE || | |
2205 | type == ARG_CONST_SIZE_OR_ZERO; | |
2206 | } | |
2207 | ||
58e2af8b | 2208 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
2209 | enum bpf_arg_type arg_type, |
2210 | struct bpf_call_arg_meta *meta) | |
17a52670 | 2211 | { |
638f5b90 | 2212 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 2213 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
2214 | int err = 0; |
2215 | ||
80f1d68c | 2216 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
2217 | return 0; |
2218 | ||
dc503a8a EC |
2219 | err = check_reg_arg(env, regno, SRC_OP); |
2220 | if (err) | |
2221 | return err; | |
17a52670 | 2222 | |
1be7f75d AS |
2223 | if (arg_type == ARG_ANYTHING) { |
2224 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
2225 | verbose(env, "R%d leaks addr into helper function\n", |
2226 | regno); | |
1be7f75d AS |
2227 | return -EACCES; |
2228 | } | |
80f1d68c | 2229 | return 0; |
1be7f75d | 2230 | } |
80f1d68c | 2231 | |
de8f3a83 | 2232 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 2233 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 2234 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
2235 | return -EACCES; |
2236 | } | |
2237 | ||
8e2fe1d9 | 2238 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
2ea864c5 MV |
2239 | arg_type == ARG_PTR_TO_MAP_VALUE || |
2240 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { | |
17a52670 | 2241 | expected_type = PTR_TO_STACK; |
d71962f3 | 2242 | if (!type_is_pkt_pointer(type) && type != PTR_TO_MAP_VALUE && |
de8f3a83 | 2243 | type != expected_type) |
6841de8b | 2244 | goto err_type; |
39f19ebb AS |
2245 | } else if (arg_type == ARG_CONST_SIZE || |
2246 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
2247 | expected_type = SCALAR_VALUE; |
2248 | if (type != expected_type) | |
6841de8b | 2249 | goto err_type; |
17a52670 AS |
2250 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
2251 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
2252 | if (type != expected_type) |
2253 | goto err_type; | |
608cd71a AS |
2254 | } else if (arg_type == ARG_PTR_TO_CTX) { |
2255 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
2256 | if (type != expected_type) |
2257 | goto err_type; | |
58990d1f DB |
2258 | err = check_ctx_reg(env, reg, regno); |
2259 | if (err < 0) | |
2260 | return err; | |
c64b7983 JS |
2261 | } else if (arg_type == ARG_PTR_TO_SOCKET) { |
2262 | expected_type = PTR_TO_SOCKET; | |
2263 | if (type != expected_type) | |
2264 | goto err_type; | |
fd978bf7 JS |
2265 | if (meta->ptr_id || !reg->id) { |
2266 | verbose(env, "verifier internal error: mismatched references meta=%d, reg=%d\n", | |
2267 | meta->ptr_id, reg->id); | |
2268 | return -EFAULT; | |
2269 | } | |
2270 | meta->ptr_id = reg->id; | |
90133415 | 2271 | } else if (arg_type_is_mem_ptr(arg_type)) { |
8e2fe1d9 DB |
2272 | expected_type = PTR_TO_STACK; |
2273 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 2274 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
2275 | * happens during stack boundary checking. |
2276 | */ | |
914cb781 | 2277 | if (register_is_null(reg) && |
db1ac496 | 2278 | arg_type == ARG_PTR_TO_MEM_OR_NULL) |
6841de8b | 2279 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
2280 | else if (!type_is_pkt_pointer(type) && |
2281 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 2282 | type != expected_type) |
6841de8b | 2283 | goto err_type; |
39f19ebb | 2284 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 | 2285 | } else { |
61bd5218 | 2286 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
2287 | return -EFAULT; |
2288 | } | |
2289 | ||
17a52670 AS |
2290 | if (arg_type == ARG_CONST_MAP_PTR) { |
2291 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 2292 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
2293 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
2294 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
2295 | * check that [key, key + map->key_size) are within | |
2296 | * stack limits and initialized | |
2297 | */ | |
33ff9823 | 2298 | if (!meta->map_ptr) { |
17a52670 AS |
2299 | /* in function declaration map_ptr must come before |
2300 | * map_key, so that it's verified and known before | |
2301 | * we have to check map_key here. Otherwise it means | |
2302 | * that kernel subsystem misconfigured verifier | |
2303 | */ | |
61bd5218 | 2304 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
2305 | return -EACCES; |
2306 | } | |
d71962f3 PC |
2307 | err = check_helper_mem_access(env, regno, |
2308 | meta->map_ptr->key_size, false, | |
2309 | NULL); | |
2ea864c5 MV |
2310 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE || |
2311 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { | |
17a52670 AS |
2312 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
2313 | * check [value, value + map->value_size) validity | |
2314 | */ | |
33ff9823 | 2315 | if (!meta->map_ptr) { |
17a52670 | 2316 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 2317 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
2318 | return -EACCES; |
2319 | } | |
2ea864c5 | 2320 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE); |
d71962f3 PC |
2321 | err = check_helper_mem_access(env, regno, |
2322 | meta->map_ptr->value_size, false, | |
2ea864c5 | 2323 | meta); |
90133415 | 2324 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 2325 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 2326 | |
849fa506 YS |
2327 | /* remember the mem_size which may be used later |
2328 | * to refine return values. | |
2329 | */ | |
2330 | meta->msize_smax_value = reg->smax_value; | |
2331 | meta->msize_umax_value = reg->umax_value; | |
2332 | ||
f1174f77 EC |
2333 | /* The register is SCALAR_VALUE; the access check |
2334 | * happens using its boundaries. | |
06c1c049 | 2335 | */ |
f1174f77 | 2336 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
2337 | /* For unprivileged variable accesses, disable raw |
2338 | * mode so that the program is required to | |
2339 | * initialize all the memory that the helper could | |
2340 | * just partially fill up. | |
2341 | */ | |
2342 | meta = NULL; | |
2343 | ||
b03c9f9f | 2344 | if (reg->smin_value < 0) { |
61bd5218 | 2345 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
2346 | regno); |
2347 | return -EACCES; | |
2348 | } | |
06c1c049 | 2349 | |
b03c9f9f | 2350 | if (reg->umin_value == 0) { |
f1174f77 EC |
2351 | err = check_helper_mem_access(env, regno - 1, 0, |
2352 | zero_size_allowed, | |
2353 | meta); | |
06c1c049 GB |
2354 | if (err) |
2355 | return err; | |
06c1c049 | 2356 | } |
f1174f77 | 2357 | |
b03c9f9f | 2358 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 2359 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
2360 | regno); |
2361 | return -EACCES; | |
2362 | } | |
2363 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 2364 | reg->umax_value, |
f1174f77 | 2365 | zero_size_allowed, meta); |
17a52670 AS |
2366 | } |
2367 | ||
2368 | return err; | |
6841de8b | 2369 | err_type: |
61bd5218 | 2370 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
2371 | reg_type_str[type], reg_type_str[expected_type]); |
2372 | return -EACCES; | |
17a52670 AS |
2373 | } |
2374 | ||
61bd5218 JK |
2375 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
2376 | struct bpf_map *map, int func_id) | |
35578d79 | 2377 | { |
35578d79 KX |
2378 | if (!map) |
2379 | return 0; | |
2380 | ||
6aff67c8 AS |
2381 | /* We need a two way check, first is from map perspective ... */ |
2382 | switch (map->map_type) { | |
2383 | case BPF_MAP_TYPE_PROG_ARRAY: | |
2384 | if (func_id != BPF_FUNC_tail_call) | |
2385 | goto error; | |
2386 | break; | |
2387 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
2388 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
2389 | func_id != BPF_FUNC_perf_event_output && |
2390 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
2391 | goto error; |
2392 | break; | |
2393 | case BPF_MAP_TYPE_STACK_TRACE: | |
2394 | if (func_id != BPF_FUNC_get_stackid) | |
2395 | goto error; | |
2396 | break; | |
4ed8ec52 | 2397 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 2398 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 2399 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
2400 | goto error; |
2401 | break; | |
cd339431 | 2402 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 2403 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
2404 | if (func_id != BPF_FUNC_get_local_storage) |
2405 | goto error; | |
2406 | break; | |
546ac1ff JF |
2407 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
2408 | * allow to be modified from bpf side. So do not allow lookup elements | |
2409 | * for now. | |
2410 | */ | |
2411 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 2412 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
2413 | goto error; |
2414 | break; | |
fbfc504a BT |
2415 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
2416 | * appear. | |
2417 | */ | |
6710e112 | 2418 | case BPF_MAP_TYPE_CPUMAP: |
fbfc504a | 2419 | case BPF_MAP_TYPE_XSKMAP: |
6710e112 JDB |
2420 | if (func_id != BPF_FUNC_redirect_map) |
2421 | goto error; | |
2422 | break; | |
56f668df | 2423 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 2424 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
2425 | if (func_id != BPF_FUNC_map_lookup_elem) |
2426 | goto error; | |
16a43625 | 2427 | break; |
174a79ff JF |
2428 | case BPF_MAP_TYPE_SOCKMAP: |
2429 | if (func_id != BPF_FUNC_sk_redirect_map && | |
2430 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb JF |
2431 | func_id != BPF_FUNC_map_delete_elem && |
2432 | func_id != BPF_FUNC_msg_redirect_map) | |
174a79ff JF |
2433 | goto error; |
2434 | break; | |
81110384 JF |
2435 | case BPF_MAP_TYPE_SOCKHASH: |
2436 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
2437 | func_id != BPF_FUNC_sock_hash_update && | |
2438 | func_id != BPF_FUNC_map_delete_elem && | |
2439 | func_id != BPF_FUNC_msg_redirect_hash) | |
2440 | goto error; | |
2441 | break; | |
2dbb9b9e MKL |
2442 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
2443 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
2444 | goto error; | |
2445 | break; | |
f1a2e44a MV |
2446 | case BPF_MAP_TYPE_QUEUE: |
2447 | case BPF_MAP_TYPE_STACK: | |
2448 | if (func_id != BPF_FUNC_map_peek_elem && | |
2449 | func_id != BPF_FUNC_map_pop_elem && | |
2450 | func_id != BPF_FUNC_map_push_elem) | |
2451 | goto error; | |
2452 | break; | |
6aff67c8 AS |
2453 | default: |
2454 | break; | |
2455 | } | |
2456 | ||
2457 | /* ... and second from the function itself. */ | |
2458 | switch (func_id) { | |
2459 | case BPF_FUNC_tail_call: | |
2460 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
2461 | goto error; | |
f910cefa | 2462 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
2463 | verbose(env, "tail_calls are not allowed in programs with bpf-to-bpf calls\n"); |
2464 | return -EINVAL; | |
2465 | } | |
6aff67c8 AS |
2466 | break; |
2467 | case BPF_FUNC_perf_event_read: | |
2468 | case BPF_FUNC_perf_event_output: | |
908432ca | 2469 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
2470 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
2471 | goto error; | |
2472 | break; | |
2473 | case BPF_FUNC_get_stackid: | |
2474 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
2475 | goto error; | |
2476 | break; | |
60d20f91 | 2477 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 2478 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
2479 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
2480 | goto error; | |
2481 | break; | |
97f91a7c | 2482 | case BPF_FUNC_redirect_map: |
9c270af3 | 2483 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
fbfc504a BT |
2484 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
2485 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
2486 | goto error; |
2487 | break; | |
174a79ff | 2488 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 2489 | case BPF_FUNC_msg_redirect_map: |
81110384 | 2490 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
2491 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
2492 | goto error; | |
2493 | break; | |
81110384 JF |
2494 | case BPF_FUNC_sk_redirect_hash: |
2495 | case BPF_FUNC_msg_redirect_hash: | |
2496 | case BPF_FUNC_sock_hash_update: | |
2497 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
2498 | goto error; |
2499 | break; | |
cd339431 | 2500 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
2501 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
2502 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
2503 | goto error; |
2504 | break; | |
2dbb9b9e MKL |
2505 | case BPF_FUNC_sk_select_reuseport: |
2506 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) | |
2507 | goto error; | |
2508 | break; | |
f1a2e44a MV |
2509 | case BPF_FUNC_map_peek_elem: |
2510 | case BPF_FUNC_map_pop_elem: | |
2511 | case BPF_FUNC_map_push_elem: | |
2512 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
2513 | map->map_type != BPF_MAP_TYPE_STACK) | |
2514 | goto error; | |
2515 | break; | |
6aff67c8 AS |
2516 | default: |
2517 | break; | |
35578d79 KX |
2518 | } |
2519 | ||
2520 | return 0; | |
6aff67c8 | 2521 | error: |
61bd5218 | 2522 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 2523 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 2524 | return -EINVAL; |
35578d79 KX |
2525 | } |
2526 | ||
90133415 | 2527 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
2528 | { |
2529 | int count = 0; | |
2530 | ||
39f19ebb | 2531 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2532 | count++; |
39f19ebb | 2533 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2534 | count++; |
39f19ebb | 2535 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2536 | count++; |
39f19ebb | 2537 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 2538 | count++; |
39f19ebb | 2539 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
2540 | count++; |
2541 | ||
90133415 DB |
2542 | /* We only support one arg being in raw mode at the moment, |
2543 | * which is sufficient for the helper functions we have | |
2544 | * right now. | |
2545 | */ | |
2546 | return count <= 1; | |
2547 | } | |
2548 | ||
2549 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
2550 | enum bpf_arg_type arg_next) | |
2551 | { | |
2552 | return (arg_type_is_mem_ptr(arg_curr) && | |
2553 | !arg_type_is_mem_size(arg_next)) || | |
2554 | (!arg_type_is_mem_ptr(arg_curr) && | |
2555 | arg_type_is_mem_size(arg_next)); | |
2556 | } | |
2557 | ||
2558 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
2559 | { | |
2560 | /* bpf_xxx(..., buf, len) call will access 'len' | |
2561 | * bytes from memory 'buf'. Both arg types need | |
2562 | * to be paired, so make sure there's no buggy | |
2563 | * helper function specification. | |
2564 | */ | |
2565 | if (arg_type_is_mem_size(fn->arg1_type) || | |
2566 | arg_type_is_mem_ptr(fn->arg5_type) || | |
2567 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
2568 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
2569 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
2570 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
2571 | return false; | |
2572 | ||
2573 | return true; | |
2574 | } | |
2575 | ||
fd978bf7 JS |
2576 | static bool check_refcount_ok(const struct bpf_func_proto *fn) |
2577 | { | |
2578 | int count = 0; | |
2579 | ||
2580 | if (arg_type_is_refcounted(fn->arg1_type)) | |
2581 | count++; | |
2582 | if (arg_type_is_refcounted(fn->arg2_type)) | |
2583 | count++; | |
2584 | if (arg_type_is_refcounted(fn->arg3_type)) | |
2585 | count++; | |
2586 | if (arg_type_is_refcounted(fn->arg4_type)) | |
2587 | count++; | |
2588 | if (arg_type_is_refcounted(fn->arg5_type)) | |
2589 | count++; | |
2590 | ||
2591 | /* We only support one arg being unreferenced at the moment, | |
2592 | * which is sufficient for the helper functions we have right now. | |
2593 | */ | |
2594 | return count <= 1; | |
2595 | } | |
2596 | ||
90133415 DB |
2597 | static int check_func_proto(const struct bpf_func_proto *fn) |
2598 | { | |
2599 | return check_raw_mode_ok(fn) && | |
fd978bf7 JS |
2600 | check_arg_pair_ok(fn) && |
2601 | check_refcount_ok(fn) ? 0 : -EINVAL; | |
435faee1 DB |
2602 | } |
2603 | ||
de8f3a83 DB |
2604 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
2605 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 2606 | */ |
f4d7e40a AS |
2607 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
2608 | struct bpf_func_state *state) | |
969bf05e | 2609 | { |
58e2af8b | 2610 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
2611 | int i; |
2612 | ||
2613 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 2614 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 2615 | mark_reg_unknown(env, regs, i); |
969bf05e | 2616 | |
f3709f69 JS |
2617 | bpf_for_each_spilled_reg(i, state, reg) { |
2618 | if (!reg) | |
969bf05e | 2619 | continue; |
de8f3a83 DB |
2620 | if (reg_is_pkt_pointer_any(reg)) |
2621 | __mark_reg_unknown(reg); | |
969bf05e AS |
2622 | } |
2623 | } | |
2624 | ||
f4d7e40a AS |
2625 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
2626 | { | |
2627 | struct bpf_verifier_state *vstate = env->cur_state; | |
2628 | int i; | |
2629 | ||
2630 | for (i = 0; i <= vstate->curframe; i++) | |
2631 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
2632 | } | |
2633 | ||
fd978bf7 JS |
2634 | static void release_reg_references(struct bpf_verifier_env *env, |
2635 | struct bpf_func_state *state, int id) | |
2636 | { | |
2637 | struct bpf_reg_state *regs = state->regs, *reg; | |
2638 | int i; | |
2639 | ||
2640 | for (i = 0; i < MAX_BPF_REG; i++) | |
2641 | if (regs[i].id == id) | |
2642 | mark_reg_unknown(env, regs, i); | |
2643 | ||
2644 | bpf_for_each_spilled_reg(i, state, reg) { | |
2645 | if (!reg) | |
2646 | continue; | |
2647 | if (reg_is_refcounted(reg) && reg->id == id) | |
2648 | __mark_reg_unknown(reg); | |
2649 | } | |
2650 | } | |
2651 | ||
2652 | /* The pointer with the specified id has released its reference to kernel | |
2653 | * resources. Identify all copies of the same pointer and clear the reference. | |
2654 | */ | |
2655 | static int release_reference(struct bpf_verifier_env *env, | |
2656 | struct bpf_call_arg_meta *meta) | |
2657 | { | |
2658 | struct bpf_verifier_state *vstate = env->cur_state; | |
2659 | int i; | |
2660 | ||
2661 | for (i = 0; i <= vstate->curframe; i++) | |
2662 | release_reg_references(env, vstate->frame[i], meta->ptr_id); | |
2663 | ||
2664 | return release_reference_state(env, meta->ptr_id); | |
2665 | } | |
2666 | ||
f4d7e40a AS |
2667 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
2668 | int *insn_idx) | |
2669 | { | |
2670 | struct bpf_verifier_state *state = env->cur_state; | |
2671 | struct bpf_func_state *caller, *callee; | |
fd978bf7 | 2672 | int i, err, subprog, target_insn; |
f4d7e40a | 2673 | |
aada9ce6 | 2674 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 2675 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 2676 | state->curframe + 2); |
f4d7e40a AS |
2677 | return -E2BIG; |
2678 | } | |
2679 | ||
2680 | target_insn = *insn_idx + insn->imm; | |
2681 | subprog = find_subprog(env, target_insn + 1); | |
2682 | if (subprog < 0) { | |
2683 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
2684 | target_insn + 1); | |
2685 | return -EFAULT; | |
2686 | } | |
2687 | ||
2688 | caller = state->frame[state->curframe]; | |
2689 | if (state->frame[state->curframe + 1]) { | |
2690 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
2691 | state->curframe + 1); | |
2692 | return -EFAULT; | |
2693 | } | |
2694 | ||
2695 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); | |
2696 | if (!callee) | |
2697 | return -ENOMEM; | |
2698 | state->frame[state->curframe + 1] = callee; | |
2699 | ||
2700 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
2701 | * into its own stack before reading from it. | |
2702 | * callee can read/write into caller's stack | |
2703 | */ | |
2704 | init_func_state(env, callee, | |
2705 | /* remember the callsite, it will be used by bpf_exit */ | |
2706 | *insn_idx /* callsite */, | |
2707 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 2708 | subprog /* subprog number within this prog */); |
f4d7e40a | 2709 | |
fd978bf7 JS |
2710 | /* Transfer references to the callee */ |
2711 | err = transfer_reference_state(callee, caller); | |
2712 | if (err) | |
2713 | return err; | |
2714 | ||
679c782d EC |
2715 | /* copy r1 - r5 args that callee can access. The copy includes parent |
2716 | * pointers, which connects us up to the liveness chain | |
2717 | */ | |
f4d7e40a AS |
2718 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) |
2719 | callee->regs[i] = caller->regs[i]; | |
2720 | ||
679c782d | 2721 | /* after the call registers r0 - r5 were scratched */ |
f4d7e40a AS |
2722 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
2723 | mark_reg_not_init(env, caller->regs, caller_saved[i]); | |
2724 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
2725 | } | |
2726 | ||
2727 | /* only increment it after check_reg_arg() finished */ | |
2728 | state->curframe++; | |
2729 | ||
2730 | /* and go analyze first insn of the callee */ | |
2731 | *insn_idx = target_insn; | |
2732 | ||
2733 | if (env->log.level) { | |
2734 | verbose(env, "caller:\n"); | |
2735 | print_verifier_state(env, caller); | |
2736 | verbose(env, "callee:\n"); | |
2737 | print_verifier_state(env, callee); | |
2738 | } | |
2739 | return 0; | |
2740 | } | |
2741 | ||
2742 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) | |
2743 | { | |
2744 | struct bpf_verifier_state *state = env->cur_state; | |
2745 | struct bpf_func_state *caller, *callee; | |
2746 | struct bpf_reg_state *r0; | |
fd978bf7 | 2747 | int err; |
f4d7e40a AS |
2748 | |
2749 | callee = state->frame[state->curframe]; | |
2750 | r0 = &callee->regs[BPF_REG_0]; | |
2751 | if (r0->type == PTR_TO_STACK) { | |
2752 | /* technically it's ok to return caller's stack pointer | |
2753 | * (or caller's caller's pointer) back to the caller, | |
2754 | * since these pointers are valid. Only current stack | |
2755 | * pointer will be invalid as soon as function exits, | |
2756 | * but let's be conservative | |
2757 | */ | |
2758 | verbose(env, "cannot return stack pointer to the caller\n"); | |
2759 | return -EINVAL; | |
2760 | } | |
2761 | ||
2762 | state->curframe--; | |
2763 | caller = state->frame[state->curframe]; | |
2764 | /* return to the caller whatever r0 had in the callee */ | |
2765 | caller->regs[BPF_REG_0] = *r0; | |
2766 | ||
fd978bf7 JS |
2767 | /* Transfer references to the caller */ |
2768 | err = transfer_reference_state(caller, callee); | |
2769 | if (err) | |
2770 | return err; | |
2771 | ||
f4d7e40a AS |
2772 | *insn_idx = callee->callsite + 1; |
2773 | if (env->log.level) { | |
2774 | verbose(env, "returning from callee:\n"); | |
2775 | print_verifier_state(env, callee); | |
2776 | verbose(env, "to caller at %d:\n", *insn_idx); | |
2777 | print_verifier_state(env, caller); | |
2778 | } | |
2779 | /* clear everything in the callee */ | |
2780 | free_func_state(callee); | |
2781 | state->frame[state->curframe + 1] = NULL; | |
2782 | return 0; | |
2783 | } | |
2784 | ||
849fa506 YS |
2785 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
2786 | int func_id, | |
2787 | struct bpf_call_arg_meta *meta) | |
2788 | { | |
2789 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
2790 | ||
2791 | if (ret_type != RET_INTEGER || | |
2792 | (func_id != BPF_FUNC_get_stack && | |
2793 | func_id != BPF_FUNC_probe_read_str)) | |
2794 | return; | |
2795 | ||
2796 | ret_reg->smax_value = meta->msize_smax_value; | |
2797 | ret_reg->umax_value = meta->msize_umax_value; | |
2798 | __reg_deduce_bounds(ret_reg); | |
2799 | __reg_bound_offset(ret_reg); | |
2800 | } | |
2801 | ||
c93552c4 DB |
2802 | static int |
2803 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
2804 | int func_id, int insn_idx) | |
2805 | { | |
2806 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
2807 | ||
2808 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
2809 | func_id != BPF_FUNC_map_lookup_elem && |
2810 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
2811 | func_id != BPF_FUNC_map_delete_elem && |
2812 | func_id != BPF_FUNC_map_push_elem && | |
2813 | func_id != BPF_FUNC_map_pop_elem && | |
2814 | func_id != BPF_FUNC_map_peek_elem) | |
c93552c4 | 2815 | return 0; |
09772d92 | 2816 | |
c93552c4 DB |
2817 | if (meta->map_ptr == NULL) { |
2818 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
2819 | return -EINVAL; | |
2820 | } | |
2821 | ||
2822 | if (!BPF_MAP_PTR(aux->map_state)) | |
2823 | bpf_map_ptr_store(aux, meta->map_ptr, | |
2824 | meta->map_ptr->unpriv_array); | |
2825 | else if (BPF_MAP_PTR(aux->map_state) != meta->map_ptr) | |
2826 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, | |
2827 | meta->map_ptr->unpriv_array); | |
2828 | return 0; | |
2829 | } | |
2830 | ||
fd978bf7 JS |
2831 | static int check_reference_leak(struct bpf_verifier_env *env) |
2832 | { | |
2833 | struct bpf_func_state *state = cur_func(env); | |
2834 | int i; | |
2835 | ||
2836 | for (i = 0; i < state->acquired_refs; i++) { | |
2837 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", | |
2838 | state->refs[i].id, state->refs[i].insn_idx); | |
2839 | } | |
2840 | return state->acquired_refs ? -EINVAL : 0; | |
2841 | } | |
2842 | ||
f4d7e40a | 2843 | static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 2844 | { |
17a52670 | 2845 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 2846 | struct bpf_reg_state *regs; |
33ff9823 | 2847 | struct bpf_call_arg_meta meta; |
969bf05e | 2848 | bool changes_data; |
17a52670 AS |
2849 | int i, err; |
2850 | ||
2851 | /* find function prototype */ | |
2852 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
2853 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
2854 | func_id); | |
17a52670 AS |
2855 | return -EINVAL; |
2856 | } | |
2857 | ||
00176a34 | 2858 | if (env->ops->get_func_proto) |
5e43f899 | 2859 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 2860 | if (!fn) { |
61bd5218 JK |
2861 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
2862 | func_id); | |
17a52670 AS |
2863 | return -EINVAL; |
2864 | } | |
2865 | ||
2866 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 2867 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 2868 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
2869 | return -EINVAL; |
2870 | } | |
2871 | ||
04514d13 | 2872 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 2873 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
2874 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
2875 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
2876 | func_id_name(func_id), func_id); | |
2877 | return -EINVAL; | |
2878 | } | |
969bf05e | 2879 | |
33ff9823 | 2880 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 2881 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 2882 | |
90133415 | 2883 | err = check_func_proto(fn); |
435faee1 | 2884 | if (err) { |
61bd5218 | 2885 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 2886 | func_id_name(func_id), func_id); |
435faee1 DB |
2887 | return err; |
2888 | } | |
2889 | ||
17a52670 | 2890 | /* check args */ |
33ff9823 | 2891 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
2892 | if (err) |
2893 | return err; | |
33ff9823 | 2894 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
2895 | if (err) |
2896 | return err; | |
33ff9823 | 2897 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
2898 | if (err) |
2899 | return err; | |
33ff9823 | 2900 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
2901 | if (err) |
2902 | return err; | |
33ff9823 | 2903 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
2904 | if (err) |
2905 | return err; | |
2906 | ||
c93552c4 DB |
2907 | err = record_func_map(env, &meta, func_id, insn_idx); |
2908 | if (err) | |
2909 | return err; | |
2910 | ||
435faee1 DB |
2911 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
2912 | * is inferred from register state. | |
2913 | */ | |
2914 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
2915 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
2916 | BPF_WRITE, -1, false); | |
435faee1 DB |
2917 | if (err) |
2918 | return err; | |
2919 | } | |
2920 | ||
fd978bf7 JS |
2921 | if (func_id == BPF_FUNC_tail_call) { |
2922 | err = check_reference_leak(env); | |
2923 | if (err) { | |
2924 | verbose(env, "tail_call would lead to reference leak\n"); | |
2925 | return err; | |
2926 | } | |
2927 | } else if (is_release_function(func_id)) { | |
2928 | err = release_reference(env, &meta); | |
2929 | if (err) | |
2930 | return err; | |
2931 | } | |
2932 | ||
638f5b90 | 2933 | regs = cur_regs(env); |
cd339431 RG |
2934 | |
2935 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
2936 | * this is required because get_local_storage() can't return an error. | |
2937 | */ | |
2938 | if (func_id == BPF_FUNC_get_local_storage && | |
2939 | !register_is_null(®s[BPF_REG_2])) { | |
2940 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
2941 | return -EINVAL; | |
2942 | } | |
2943 | ||
17a52670 | 2944 | /* reset caller saved regs */ |
dc503a8a | 2945 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 2946 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
2947 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
2948 | } | |
17a52670 | 2949 | |
dc503a8a | 2950 | /* update return register (already marked as written above) */ |
17a52670 | 2951 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 2952 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 2953 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
2954 | } else if (fn->ret_type == RET_VOID) { |
2955 | regs[BPF_REG_0].type = NOT_INIT; | |
3e6a4b3e RG |
2956 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL || |
2957 | fn->ret_type == RET_PTR_TO_MAP_VALUE) { | |
f1174f77 | 2958 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 2959 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
2960 | /* remember map_ptr, so that check_map_access() |
2961 | * can check 'value_size' boundary of memory access | |
2962 | * to map element returned from bpf_map_lookup_elem() | |
2963 | */ | |
33ff9823 | 2964 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
2965 | verbose(env, |
2966 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
2967 | return -EINVAL; |
2968 | } | |
33ff9823 | 2969 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
4d31f301 DB |
2970 | if (fn->ret_type == RET_PTR_TO_MAP_VALUE) { |
2971 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE; | |
2972 | } else { | |
2973 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
2974 | regs[BPF_REG_0].id = ++env->id_gen; | |
2975 | } | |
c64b7983 | 2976 | } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) { |
fd978bf7 JS |
2977 | int id = acquire_reference_state(env, insn_idx); |
2978 | if (id < 0) | |
2979 | return id; | |
c64b7983 JS |
2980 | mark_reg_known_zero(env, regs, BPF_REG_0); |
2981 | regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL; | |
fd978bf7 | 2982 | regs[BPF_REG_0].id = id; |
17a52670 | 2983 | } else { |
61bd5218 | 2984 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 2985 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
2986 | return -EINVAL; |
2987 | } | |
04fd61ab | 2988 | |
849fa506 YS |
2989 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
2990 | ||
61bd5218 | 2991 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
2992 | if (err) |
2993 | return err; | |
04fd61ab | 2994 | |
c195651e YS |
2995 | if (func_id == BPF_FUNC_get_stack && !env->prog->has_callchain_buf) { |
2996 | const char *err_str; | |
2997 | ||
2998 | #ifdef CONFIG_PERF_EVENTS | |
2999 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
3000 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
3001 | #else | |
3002 | err = -ENOTSUPP; | |
3003 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
3004 | #endif | |
3005 | if (err) { | |
3006 | verbose(env, err_str, func_id_name(func_id), func_id); | |
3007 | return err; | |
3008 | } | |
3009 | ||
3010 | env->prog->has_callchain_buf = true; | |
3011 | } | |
3012 | ||
969bf05e AS |
3013 | if (changes_data) |
3014 | clear_all_pkt_pointers(env); | |
3015 | return 0; | |
3016 | } | |
3017 | ||
b03c9f9f EC |
3018 | static bool signed_add_overflows(s64 a, s64 b) |
3019 | { | |
3020 | /* Do the add in u64, where overflow is well-defined */ | |
3021 | s64 res = (s64)((u64)a + (u64)b); | |
3022 | ||
3023 | if (b < 0) | |
3024 | return res > a; | |
3025 | return res < a; | |
3026 | } | |
3027 | ||
3028 | static bool signed_sub_overflows(s64 a, s64 b) | |
3029 | { | |
3030 | /* Do the sub in u64, where overflow is well-defined */ | |
3031 | s64 res = (s64)((u64)a - (u64)b); | |
3032 | ||
3033 | if (b < 0) | |
3034 | return res < a; | |
3035 | return res > a; | |
969bf05e AS |
3036 | } |
3037 | ||
bb7f0f98 AS |
3038 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
3039 | const struct bpf_reg_state *reg, | |
3040 | enum bpf_reg_type type) | |
3041 | { | |
3042 | bool known = tnum_is_const(reg->var_off); | |
3043 | s64 val = reg->var_off.value; | |
3044 | s64 smin = reg->smin_value; | |
3045 | ||
3046 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
3047 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
3048 | reg_type_str[type], val); | |
3049 | return false; | |
3050 | } | |
3051 | ||
3052 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
3053 | verbose(env, "%s pointer offset %d is not allowed\n", | |
3054 | reg_type_str[type], reg->off); | |
3055 | return false; | |
3056 | } | |
3057 | ||
3058 | if (smin == S64_MIN) { | |
3059 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
3060 | reg_type_str[type]); | |
3061 | return false; | |
3062 | } | |
3063 | ||
3064 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
3065 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
3066 | smin, reg_type_str[type]); | |
3067 | return false; | |
3068 | } | |
3069 | ||
3070 | return true; | |
3071 | } | |
3072 | ||
979d63d5 DB |
3073 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
3074 | { | |
3075 | return &env->insn_aux_data[env->insn_idx]; | |
3076 | } | |
3077 | ||
3078 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, | |
3079 | u32 *ptr_limit, u8 opcode, bool off_is_neg) | |
3080 | { | |
3081 | bool mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
3082 | (opcode == BPF_SUB && !off_is_neg); | |
3083 | u32 off; | |
3084 | ||
3085 | switch (ptr_reg->type) { | |
3086 | case PTR_TO_STACK: | |
3087 | off = ptr_reg->off + ptr_reg->var_off.value; | |
3088 | if (mask_to_left) | |
3089 | *ptr_limit = MAX_BPF_STACK + off; | |
3090 | else | |
3091 | *ptr_limit = -off; | |
3092 | return 0; | |
3093 | case PTR_TO_MAP_VALUE: | |
3094 | if (mask_to_left) { | |
3095 | *ptr_limit = ptr_reg->umax_value + ptr_reg->off; | |
3096 | } else { | |
3097 | off = ptr_reg->smin_value + ptr_reg->off; | |
3098 | *ptr_limit = ptr_reg->map_ptr->value_size - off; | |
3099 | } | |
3100 | return 0; | |
3101 | default: | |
3102 | return -EINVAL; | |
3103 | } | |
3104 | } | |
3105 | ||
d3bd7413 DB |
3106 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
3107 | const struct bpf_insn *insn) | |
3108 | { | |
3109 | return env->allow_ptr_leaks || BPF_SRC(insn->code) == BPF_K; | |
3110 | } | |
3111 | ||
3112 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
3113 | u32 alu_state, u32 alu_limit) | |
3114 | { | |
3115 | /* If we arrived here from different branches with different | |
3116 | * state or limits to sanitize, then this won't work. | |
3117 | */ | |
3118 | if (aux->alu_state && | |
3119 | (aux->alu_state != alu_state || | |
3120 | aux->alu_limit != alu_limit)) | |
3121 | return -EACCES; | |
3122 | ||
3123 | /* Corresponding fixup done in fixup_bpf_calls(). */ | |
3124 | aux->alu_state = alu_state; | |
3125 | aux->alu_limit = alu_limit; | |
3126 | return 0; | |
3127 | } | |
3128 | ||
3129 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
3130 | struct bpf_insn *insn) | |
3131 | { | |
3132 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
3133 | ||
3134 | if (can_skip_alu_sanitation(env, insn)) | |
3135 | return 0; | |
3136 | ||
3137 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
3138 | } | |
3139 | ||
979d63d5 DB |
3140 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
3141 | struct bpf_insn *insn, | |
3142 | const struct bpf_reg_state *ptr_reg, | |
3143 | struct bpf_reg_state *dst_reg, | |
3144 | bool off_is_neg) | |
3145 | { | |
3146 | struct bpf_verifier_state *vstate = env->cur_state; | |
3147 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
3148 | bool ptr_is_dst_reg = ptr_reg == dst_reg; | |
3149 | u8 opcode = BPF_OP(insn->code); | |
3150 | u32 alu_state, alu_limit; | |
3151 | struct bpf_reg_state tmp; | |
3152 | bool ret; | |
3153 | ||
d3bd7413 | 3154 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
3155 | return 0; |
3156 | ||
3157 | /* We already marked aux for masking from non-speculative | |
3158 | * paths, thus we got here in the first place. We only care | |
3159 | * to explore bad access from here. | |
3160 | */ | |
3161 | if (vstate->speculative) | |
3162 | goto do_sim; | |
3163 | ||
3164 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
3165 | alu_state |= ptr_is_dst_reg ? | |
3166 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
3167 | ||
3168 | if (retrieve_ptr_limit(ptr_reg, &alu_limit, opcode, off_is_neg)) | |
3169 | return 0; | |
d3bd7413 | 3170 | if (update_alu_sanitation_state(aux, alu_state, alu_limit)) |
979d63d5 | 3171 | return -EACCES; |
979d63d5 DB |
3172 | do_sim: |
3173 | /* Simulate and find potential out-of-bounds access under | |
3174 | * speculative execution from truncation as a result of | |
3175 | * masking when off was not within expected range. If off | |
3176 | * sits in dst, then we temporarily need to move ptr there | |
3177 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
3178 | * for cases where we use K-based arithmetic in one direction | |
3179 | * and truncated reg-based in the other in order to explore | |
3180 | * bad access. | |
3181 | */ | |
3182 | if (!ptr_is_dst_reg) { | |
3183 | tmp = *dst_reg; | |
3184 | *dst_reg = *ptr_reg; | |
3185 | } | |
3186 | ret = push_stack(env, env->insn_idx + 1, env->insn_idx, true); | |
3187 | if (!ptr_is_dst_reg) | |
3188 | *dst_reg = tmp; | |
3189 | return !ret ? -EFAULT : 0; | |
3190 | } | |
3191 | ||
f1174f77 | 3192 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
3193 | * Caller should also handle BPF_MOV case separately. |
3194 | * If we return -EACCES, caller may want to try again treating pointer as a | |
3195 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
3196 | */ | |
3197 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
3198 | struct bpf_insn *insn, | |
3199 | const struct bpf_reg_state *ptr_reg, | |
3200 | const struct bpf_reg_state *off_reg) | |
969bf05e | 3201 | { |
f4d7e40a AS |
3202 | struct bpf_verifier_state *vstate = env->cur_state; |
3203 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3204 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 3205 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
3206 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
3207 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
3208 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
3209 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
9d7eceed | 3210 | u32 dst = insn->dst_reg, src = insn->src_reg; |
969bf05e | 3211 | u8 opcode = BPF_OP(insn->code); |
979d63d5 | 3212 | int ret; |
969bf05e | 3213 | |
f1174f77 | 3214 | dst_reg = ®s[dst]; |
969bf05e | 3215 | |
6f16101e DB |
3216 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
3217 | smin_val > smax_val || umin_val > umax_val) { | |
3218 | /* Taint dst register if offset had invalid bounds derived from | |
3219 | * e.g. dead branches. | |
3220 | */ | |
3221 | __mark_reg_unknown(dst_reg); | |
3222 | return 0; | |
f1174f77 EC |
3223 | } |
3224 | ||
3225 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
3226 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
3227 | verbose(env, |
3228 | "R%d 32-bit pointer arithmetic prohibited\n", | |
3229 | dst); | |
f1174f77 | 3230 | return -EACCES; |
969bf05e AS |
3231 | } |
3232 | ||
aad2eeaf JS |
3233 | switch (ptr_reg->type) { |
3234 | case PTR_TO_MAP_VALUE_OR_NULL: | |
3235 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", | |
3236 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 3237 | return -EACCES; |
aad2eeaf JS |
3238 | case CONST_PTR_TO_MAP: |
3239 | case PTR_TO_PACKET_END: | |
c64b7983 JS |
3240 | case PTR_TO_SOCKET: |
3241 | case PTR_TO_SOCKET_OR_NULL: | |
aad2eeaf JS |
3242 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
3243 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 3244 | return -EACCES; |
9d7eceed DB |
3245 | case PTR_TO_MAP_VALUE: |
3246 | if (!env->allow_ptr_leaks && !known && (smin_val < 0) != (smax_val < 0)) { | |
3247 | verbose(env, "R%d has unknown scalar with mixed signed bounds, pointer arithmetic with it prohibited for !root\n", | |
3248 | off_reg == dst_reg ? dst : src); | |
3249 | return -EACCES; | |
3250 | } | |
3251 | /* fall-through */ | |
aad2eeaf JS |
3252 | default: |
3253 | break; | |
f1174f77 EC |
3254 | } |
3255 | ||
3256 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
3257 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 3258 | */ |
f1174f77 EC |
3259 | dst_reg->type = ptr_reg->type; |
3260 | dst_reg->id = ptr_reg->id; | |
969bf05e | 3261 | |
bb7f0f98 AS |
3262 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
3263 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
3264 | return -EINVAL; | |
3265 | ||
f1174f77 EC |
3266 | switch (opcode) { |
3267 | case BPF_ADD: | |
979d63d5 DB |
3268 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
3269 | if (ret < 0) { | |
3270 | verbose(env, "R%d tried to add from different maps or paths\n", dst); | |
3271 | return ret; | |
3272 | } | |
f1174f77 EC |
3273 | /* We can take a fixed offset as long as it doesn't overflow |
3274 | * the s32 'off' field | |
969bf05e | 3275 | */ |
b03c9f9f EC |
3276 | if (known && (ptr_reg->off + smin_val == |
3277 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 3278 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
3279 | dst_reg->smin_value = smin_ptr; |
3280 | dst_reg->smax_value = smax_ptr; | |
3281 | dst_reg->umin_value = umin_ptr; | |
3282 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 3283 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 3284 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 3285 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
3286 | break; |
3287 | } | |
f1174f77 EC |
3288 | /* A new variable offset is created. Note that off_reg->off |
3289 | * == 0, since it's a scalar. | |
3290 | * dst_reg gets the pointer type and since some positive | |
3291 | * integer value was added to the pointer, give it a new 'id' | |
3292 | * if it's a PTR_TO_PACKET. | |
3293 | * this creates a new 'base' pointer, off_reg (variable) gets | |
3294 | * added into the variable offset, and we copy the fixed offset | |
3295 | * from ptr_reg. | |
969bf05e | 3296 | */ |
b03c9f9f EC |
3297 | if (signed_add_overflows(smin_ptr, smin_val) || |
3298 | signed_add_overflows(smax_ptr, smax_val)) { | |
3299 | dst_reg->smin_value = S64_MIN; | |
3300 | dst_reg->smax_value = S64_MAX; | |
3301 | } else { | |
3302 | dst_reg->smin_value = smin_ptr + smin_val; | |
3303 | dst_reg->smax_value = smax_ptr + smax_val; | |
3304 | } | |
3305 | if (umin_ptr + umin_val < umin_ptr || | |
3306 | umax_ptr + umax_val < umax_ptr) { | |
3307 | dst_reg->umin_value = 0; | |
3308 | dst_reg->umax_value = U64_MAX; | |
3309 | } else { | |
3310 | dst_reg->umin_value = umin_ptr + umin_val; | |
3311 | dst_reg->umax_value = umax_ptr + umax_val; | |
3312 | } | |
f1174f77 EC |
3313 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
3314 | dst_reg->off = ptr_reg->off; | |
0962590e | 3315 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 3316 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
3317 | dst_reg->id = ++env->id_gen; |
3318 | /* something was added to pkt_ptr, set range to zero */ | |
0962590e | 3319 | dst_reg->raw = 0; |
f1174f77 EC |
3320 | } |
3321 | break; | |
3322 | case BPF_SUB: | |
979d63d5 DB |
3323 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
3324 | if (ret < 0) { | |
3325 | verbose(env, "R%d tried to sub from different maps or paths\n", dst); | |
3326 | return ret; | |
3327 | } | |
f1174f77 EC |
3328 | if (dst_reg == off_reg) { |
3329 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
3330 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
3331 | dst); | |
f1174f77 EC |
3332 | return -EACCES; |
3333 | } | |
3334 | /* We don't allow subtraction from FP, because (according to | |
3335 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
3336 | * be able to deal with it. | |
969bf05e | 3337 | */ |
f1174f77 | 3338 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
3339 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
3340 | dst); | |
f1174f77 EC |
3341 | return -EACCES; |
3342 | } | |
b03c9f9f EC |
3343 | if (known && (ptr_reg->off - smin_val == |
3344 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 3345 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
3346 | dst_reg->smin_value = smin_ptr; |
3347 | dst_reg->smax_value = smax_ptr; | |
3348 | dst_reg->umin_value = umin_ptr; | |
3349 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
3350 | dst_reg->var_off = ptr_reg->var_off; |
3351 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 3352 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 3353 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
3354 | break; |
3355 | } | |
f1174f77 EC |
3356 | /* A new variable offset is created. If the subtrahend is known |
3357 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 3358 | */ |
b03c9f9f EC |
3359 | if (signed_sub_overflows(smin_ptr, smax_val) || |
3360 | signed_sub_overflows(smax_ptr, smin_val)) { | |
3361 | /* Overflow possible, we know nothing */ | |
3362 | dst_reg->smin_value = S64_MIN; | |
3363 | dst_reg->smax_value = S64_MAX; | |
3364 | } else { | |
3365 | dst_reg->smin_value = smin_ptr - smax_val; | |
3366 | dst_reg->smax_value = smax_ptr - smin_val; | |
3367 | } | |
3368 | if (umin_ptr < umax_val) { | |
3369 | /* Overflow possible, we know nothing */ | |
3370 | dst_reg->umin_value = 0; | |
3371 | dst_reg->umax_value = U64_MAX; | |
3372 | } else { | |
3373 | /* Cannot overflow (as long as bounds are consistent) */ | |
3374 | dst_reg->umin_value = umin_ptr - umax_val; | |
3375 | dst_reg->umax_value = umax_ptr - umin_val; | |
3376 | } | |
f1174f77 EC |
3377 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
3378 | dst_reg->off = ptr_reg->off; | |
0962590e | 3379 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 3380 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
3381 | dst_reg->id = ++env->id_gen; |
3382 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 3383 | if (smin_val < 0) |
0962590e | 3384 | dst_reg->raw = 0; |
43188702 | 3385 | } |
f1174f77 EC |
3386 | break; |
3387 | case BPF_AND: | |
3388 | case BPF_OR: | |
3389 | case BPF_XOR: | |
82abbf8d AS |
3390 | /* bitwise ops on pointers are troublesome, prohibit. */ |
3391 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
3392 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
3393 | return -EACCES; |
3394 | default: | |
3395 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
3396 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
3397 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 3398 | return -EACCES; |
43188702 JF |
3399 | } |
3400 | ||
bb7f0f98 AS |
3401 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
3402 | return -EINVAL; | |
3403 | ||
b03c9f9f EC |
3404 | __update_reg_bounds(dst_reg); |
3405 | __reg_deduce_bounds(dst_reg); | |
3406 | __reg_bound_offset(dst_reg); | |
0d6303db DB |
3407 | |
3408 | /* For unprivileged we require that resulting offset must be in bounds | |
3409 | * in order to be able to sanitize access later on. | |
3410 | */ | |
e4298d25 DB |
3411 | if (!env->allow_ptr_leaks) { |
3412 | if (dst_reg->type == PTR_TO_MAP_VALUE && | |
3413 | check_map_access(env, dst, dst_reg->off, 1, false)) { | |
3414 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
3415 | "prohibited for !root\n", dst); | |
3416 | return -EACCES; | |
3417 | } else if (dst_reg->type == PTR_TO_STACK && | |
3418 | check_stack_access(env, dst_reg, dst_reg->off + | |
3419 | dst_reg->var_off.value, 1)) { | |
3420 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
3421 | "prohibited for !root\n", dst); | |
3422 | return -EACCES; | |
3423 | } | |
0d6303db DB |
3424 | } |
3425 | ||
43188702 JF |
3426 | return 0; |
3427 | } | |
3428 | ||
468f6eaf JH |
3429 | /* WARNING: This function does calculations on 64-bit values, but the actual |
3430 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
3431 | * need extra checks in the 32-bit case. | |
3432 | */ | |
f1174f77 EC |
3433 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
3434 | struct bpf_insn *insn, | |
3435 | struct bpf_reg_state *dst_reg, | |
3436 | struct bpf_reg_state src_reg) | |
969bf05e | 3437 | { |
638f5b90 | 3438 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 3439 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 3440 | bool src_known, dst_known; |
b03c9f9f EC |
3441 | s64 smin_val, smax_val; |
3442 | u64 umin_val, umax_val; | |
468f6eaf | 3443 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
d3bd7413 DB |
3444 | u32 dst = insn->dst_reg; |
3445 | int ret; | |
48461135 | 3446 | |
b799207e JH |
3447 | if (insn_bitness == 32) { |
3448 | /* Relevant for 32-bit RSH: Information can propagate towards | |
3449 | * LSB, so it isn't sufficient to only truncate the output to | |
3450 | * 32 bits. | |
3451 | */ | |
3452 | coerce_reg_to_size(dst_reg, 4); | |
3453 | coerce_reg_to_size(&src_reg, 4); | |
3454 | } | |
3455 | ||
b03c9f9f EC |
3456 | smin_val = src_reg.smin_value; |
3457 | smax_val = src_reg.smax_value; | |
3458 | umin_val = src_reg.umin_value; | |
3459 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
3460 | src_known = tnum_is_const(src_reg.var_off); |
3461 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 3462 | |
6f16101e DB |
3463 | if ((src_known && (smin_val != smax_val || umin_val != umax_val)) || |
3464 | smin_val > smax_val || umin_val > umax_val) { | |
3465 | /* Taint dst register if offset had invalid bounds derived from | |
3466 | * e.g. dead branches. | |
3467 | */ | |
3468 | __mark_reg_unknown(dst_reg); | |
3469 | return 0; | |
3470 | } | |
3471 | ||
bb7f0f98 AS |
3472 | if (!src_known && |
3473 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
3474 | __mark_reg_unknown(dst_reg); | |
3475 | return 0; | |
3476 | } | |
3477 | ||
48461135 JB |
3478 | switch (opcode) { |
3479 | case BPF_ADD: | |
d3bd7413 DB |
3480 | ret = sanitize_val_alu(env, insn); |
3481 | if (ret < 0) { | |
3482 | verbose(env, "R%d tried to add from different pointers or scalars\n", dst); | |
3483 | return ret; | |
3484 | } | |
b03c9f9f EC |
3485 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
3486 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
3487 | dst_reg->smin_value = S64_MIN; | |
3488 | dst_reg->smax_value = S64_MAX; | |
3489 | } else { | |
3490 | dst_reg->smin_value += smin_val; | |
3491 | dst_reg->smax_value += smax_val; | |
3492 | } | |
3493 | if (dst_reg->umin_value + umin_val < umin_val || | |
3494 | dst_reg->umax_value + umax_val < umax_val) { | |
3495 | dst_reg->umin_value = 0; | |
3496 | dst_reg->umax_value = U64_MAX; | |
3497 | } else { | |
3498 | dst_reg->umin_value += umin_val; | |
3499 | dst_reg->umax_value += umax_val; | |
3500 | } | |
f1174f77 | 3501 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
3502 | break; |
3503 | case BPF_SUB: | |
d3bd7413 DB |
3504 | ret = sanitize_val_alu(env, insn); |
3505 | if (ret < 0) { | |
3506 | verbose(env, "R%d tried to sub from different pointers or scalars\n", dst); | |
3507 | return ret; | |
3508 | } | |
b03c9f9f EC |
3509 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
3510 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
3511 | /* Overflow possible, we know nothing */ | |
3512 | dst_reg->smin_value = S64_MIN; | |
3513 | dst_reg->smax_value = S64_MAX; | |
3514 | } else { | |
3515 | dst_reg->smin_value -= smax_val; | |
3516 | dst_reg->smax_value -= smin_val; | |
3517 | } | |
3518 | if (dst_reg->umin_value < umax_val) { | |
3519 | /* Overflow possible, we know nothing */ | |
3520 | dst_reg->umin_value = 0; | |
3521 | dst_reg->umax_value = U64_MAX; | |
3522 | } else { | |
3523 | /* Cannot overflow (as long as bounds are consistent) */ | |
3524 | dst_reg->umin_value -= umax_val; | |
3525 | dst_reg->umax_value -= umin_val; | |
3526 | } | |
f1174f77 | 3527 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
3528 | break; |
3529 | case BPF_MUL: | |
b03c9f9f EC |
3530 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
3531 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 3532 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
3533 | __mark_reg_unbounded(dst_reg); |
3534 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
3535 | break; |
3536 | } | |
b03c9f9f EC |
3537 | /* Both values are positive, so we can work with unsigned and |
3538 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 3539 | */ |
b03c9f9f EC |
3540 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
3541 | /* Potential overflow, we know nothing */ | |
3542 | __mark_reg_unbounded(dst_reg); | |
3543 | /* (except what we can learn from the var_off) */ | |
3544 | __update_reg_bounds(dst_reg); | |
3545 | break; | |
3546 | } | |
3547 | dst_reg->umin_value *= umin_val; | |
3548 | dst_reg->umax_value *= umax_val; | |
3549 | if (dst_reg->umax_value > S64_MAX) { | |
3550 | /* Overflow possible, we know nothing */ | |
3551 | dst_reg->smin_value = S64_MIN; | |
3552 | dst_reg->smax_value = S64_MAX; | |
3553 | } else { | |
3554 | dst_reg->smin_value = dst_reg->umin_value; | |
3555 | dst_reg->smax_value = dst_reg->umax_value; | |
3556 | } | |
48461135 JB |
3557 | break; |
3558 | case BPF_AND: | |
f1174f77 | 3559 | if (src_known && dst_known) { |
b03c9f9f EC |
3560 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
3561 | src_reg.var_off.value); | |
f1174f77 EC |
3562 | break; |
3563 | } | |
b03c9f9f EC |
3564 | /* We get our minimum from the var_off, since that's inherently |
3565 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 3566 | */ |
f1174f77 | 3567 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
3568 | dst_reg->umin_value = dst_reg->var_off.value; |
3569 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
3570 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
3571 | /* Lose signed bounds when ANDing negative numbers, | |
3572 | * ain't nobody got time for that. | |
3573 | */ | |
3574 | dst_reg->smin_value = S64_MIN; | |
3575 | dst_reg->smax_value = S64_MAX; | |
3576 | } else { | |
3577 | /* ANDing two positives gives a positive, so safe to | |
3578 | * cast result into s64. | |
3579 | */ | |
3580 | dst_reg->smin_value = dst_reg->umin_value; | |
3581 | dst_reg->smax_value = dst_reg->umax_value; | |
3582 | } | |
3583 | /* We may learn something more from the var_off */ | |
3584 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
3585 | break; |
3586 | case BPF_OR: | |
3587 | if (src_known && dst_known) { | |
b03c9f9f EC |
3588 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
3589 | src_reg.var_off.value); | |
f1174f77 EC |
3590 | break; |
3591 | } | |
b03c9f9f EC |
3592 | /* We get our maximum from the var_off, and our minimum is the |
3593 | * maximum of the operands' minima | |
f1174f77 EC |
3594 | */ |
3595 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
3596 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
3597 | dst_reg->umax_value = dst_reg->var_off.value | | |
3598 | dst_reg->var_off.mask; | |
3599 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
3600 | /* Lose signed bounds when ORing negative numbers, | |
3601 | * ain't nobody got time for that. | |
3602 | */ | |
3603 | dst_reg->smin_value = S64_MIN; | |
3604 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 3605 | } else { |
b03c9f9f EC |
3606 | /* ORing two positives gives a positive, so safe to |
3607 | * cast result into s64. | |
3608 | */ | |
3609 | dst_reg->smin_value = dst_reg->umin_value; | |
3610 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 3611 | } |
b03c9f9f EC |
3612 | /* We may learn something more from the var_off */ |
3613 | __update_reg_bounds(dst_reg); | |
48461135 JB |
3614 | break; |
3615 | case BPF_LSH: | |
468f6eaf JH |
3616 | if (umax_val >= insn_bitness) { |
3617 | /* Shifts greater than 31 or 63 are undefined. | |
3618 | * This includes shifts by a negative number. | |
b03c9f9f | 3619 | */ |
61bd5218 | 3620 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
3621 | break; |
3622 | } | |
b03c9f9f EC |
3623 | /* We lose all sign bit information (except what we can pick |
3624 | * up from var_off) | |
48461135 | 3625 | */ |
b03c9f9f EC |
3626 | dst_reg->smin_value = S64_MIN; |
3627 | dst_reg->smax_value = S64_MAX; | |
3628 | /* If we might shift our top bit out, then we know nothing */ | |
3629 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
3630 | dst_reg->umin_value = 0; | |
3631 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 3632 | } else { |
b03c9f9f EC |
3633 | dst_reg->umin_value <<= umin_val; |
3634 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 3635 | } |
afbe1a5b | 3636 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
3637 | /* We may learn something more from the var_off */ |
3638 | __update_reg_bounds(dst_reg); | |
48461135 JB |
3639 | break; |
3640 | case BPF_RSH: | |
468f6eaf JH |
3641 | if (umax_val >= insn_bitness) { |
3642 | /* Shifts greater than 31 or 63 are undefined. | |
3643 | * This includes shifts by a negative number. | |
b03c9f9f | 3644 | */ |
61bd5218 | 3645 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
3646 | break; |
3647 | } | |
4374f256 EC |
3648 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
3649 | * be negative, then either: | |
3650 | * 1) src_reg might be zero, so the sign bit of the result is | |
3651 | * unknown, so we lose our signed bounds | |
3652 | * 2) it's known negative, thus the unsigned bounds capture the | |
3653 | * signed bounds | |
3654 | * 3) the signed bounds cross zero, so they tell us nothing | |
3655 | * about the result | |
3656 | * If the value in dst_reg is known nonnegative, then again the | |
3657 | * unsigned bounts capture the signed bounds. | |
3658 | * Thus, in all cases it suffices to blow away our signed bounds | |
3659 | * and rely on inferring new ones from the unsigned bounds and | |
3660 | * var_off of the result. | |
3661 | */ | |
3662 | dst_reg->smin_value = S64_MIN; | |
3663 | dst_reg->smax_value = S64_MAX; | |
afbe1a5b | 3664 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
3665 | dst_reg->umin_value >>= umax_val; |
3666 | dst_reg->umax_value >>= umin_val; | |
3667 | /* We may learn something more from the var_off */ | |
3668 | __update_reg_bounds(dst_reg); | |
48461135 | 3669 | break; |
9cbe1f5a YS |
3670 | case BPF_ARSH: |
3671 | if (umax_val >= insn_bitness) { | |
3672 | /* Shifts greater than 31 or 63 are undefined. | |
3673 | * This includes shifts by a negative number. | |
3674 | */ | |
3675 | mark_reg_unknown(env, regs, insn->dst_reg); | |
3676 | break; | |
3677 | } | |
3678 | ||
3679 | /* Upon reaching here, src_known is true and | |
3680 | * umax_val is equal to umin_val. | |
3681 | */ | |
3682 | dst_reg->smin_value >>= umin_val; | |
3683 | dst_reg->smax_value >>= umin_val; | |
3684 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val); | |
3685 | ||
3686 | /* blow away the dst_reg umin_value/umax_value and rely on | |
3687 | * dst_reg var_off to refine the result. | |
3688 | */ | |
3689 | dst_reg->umin_value = 0; | |
3690 | dst_reg->umax_value = U64_MAX; | |
3691 | __update_reg_bounds(dst_reg); | |
3692 | break; | |
48461135 | 3693 | default: |
61bd5218 | 3694 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
3695 | break; |
3696 | } | |
3697 | ||
468f6eaf JH |
3698 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
3699 | /* 32-bit ALU ops are (32,32)->32 */ | |
3700 | coerce_reg_to_size(dst_reg, 4); | |
468f6eaf JH |
3701 | } |
3702 | ||
b03c9f9f EC |
3703 | __reg_deduce_bounds(dst_reg); |
3704 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
3705 | return 0; |
3706 | } | |
3707 | ||
3708 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
3709 | * and var_off. | |
3710 | */ | |
3711 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
3712 | struct bpf_insn *insn) | |
3713 | { | |
f4d7e40a AS |
3714 | struct bpf_verifier_state *vstate = env->cur_state; |
3715 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3716 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
3717 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
3718 | u8 opcode = BPF_OP(insn->code); | |
f1174f77 EC |
3719 | |
3720 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
3721 | src_reg = NULL; |
3722 | if (dst_reg->type != SCALAR_VALUE) | |
3723 | ptr_reg = dst_reg; | |
3724 | if (BPF_SRC(insn->code) == BPF_X) { | |
3725 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
3726 | if (src_reg->type != SCALAR_VALUE) { |
3727 | if (dst_reg->type != SCALAR_VALUE) { | |
3728 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
3729 | * an arbitrary scalar. Disallow all math except |
3730 | * pointer subtraction | |
f1174f77 | 3731 | */ |
dd066823 | 3732 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
3733 | mark_reg_unknown(env, regs, insn->dst_reg); |
3734 | return 0; | |
f1174f77 | 3735 | } |
82abbf8d AS |
3736 | verbose(env, "R%d pointer %s pointer prohibited\n", |
3737 | insn->dst_reg, | |
3738 | bpf_alu_string[opcode >> 4]); | |
3739 | return -EACCES; | |
f1174f77 EC |
3740 | } else { |
3741 | /* scalar += pointer | |
3742 | * This is legal, but we have to reverse our | |
3743 | * src/dest handling in computing the range | |
3744 | */ | |
82abbf8d AS |
3745 | return adjust_ptr_min_max_vals(env, insn, |
3746 | src_reg, dst_reg); | |
f1174f77 EC |
3747 | } |
3748 | } else if (ptr_reg) { | |
3749 | /* pointer += scalar */ | |
82abbf8d AS |
3750 | return adjust_ptr_min_max_vals(env, insn, |
3751 | dst_reg, src_reg); | |
f1174f77 EC |
3752 | } |
3753 | } else { | |
3754 | /* Pretend the src is a reg with a known value, since we only | |
3755 | * need to be able to read from this state. | |
3756 | */ | |
3757 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 3758 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 3759 | src_reg = &off_reg; |
82abbf8d AS |
3760 | if (ptr_reg) /* pointer += K */ |
3761 | return adjust_ptr_min_max_vals(env, insn, | |
3762 | ptr_reg, src_reg); | |
f1174f77 EC |
3763 | } |
3764 | ||
3765 | /* Got here implies adding two SCALAR_VALUEs */ | |
3766 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 3767 | print_verifier_state(env, state); |
61bd5218 | 3768 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
3769 | return -EINVAL; |
3770 | } | |
3771 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 3772 | print_verifier_state(env, state); |
61bd5218 | 3773 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
3774 | return -EINVAL; |
3775 | } | |
3776 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
3777 | } |
3778 | ||
17a52670 | 3779 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 3780 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 3781 | { |
638f5b90 | 3782 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
3783 | u8 opcode = BPF_OP(insn->code); |
3784 | int err; | |
3785 | ||
3786 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
3787 | if (opcode == BPF_NEG) { | |
3788 | if (BPF_SRC(insn->code) != 0 || | |
3789 | insn->src_reg != BPF_REG_0 || | |
3790 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 3791 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
3792 | return -EINVAL; |
3793 | } | |
3794 | } else { | |
3795 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
3796 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
3797 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 3798 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
3799 | return -EINVAL; |
3800 | } | |
3801 | } | |
3802 | ||
3803 | /* check src operand */ | |
dc503a8a | 3804 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3805 | if (err) |
3806 | return err; | |
3807 | ||
1be7f75d | 3808 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 3809 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
3810 | insn->dst_reg); |
3811 | return -EACCES; | |
3812 | } | |
3813 | ||
17a52670 | 3814 | /* check dest operand */ |
dc503a8a | 3815 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
3816 | if (err) |
3817 | return err; | |
3818 | ||
3819 | } else if (opcode == BPF_MOV) { | |
3820 | ||
3821 | if (BPF_SRC(insn->code) == BPF_X) { | |
3822 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 3823 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
3824 | return -EINVAL; |
3825 | } | |
3826 | ||
3827 | /* check src operand */ | |
dc503a8a | 3828 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3829 | if (err) |
3830 | return err; | |
3831 | } else { | |
3832 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 3833 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
3834 | return -EINVAL; |
3835 | } | |
3836 | } | |
3837 | ||
fbeb1603 AF |
3838 | /* check dest operand, mark as required later */ |
3839 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
3840 | if (err) |
3841 | return err; | |
3842 | ||
3843 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
3844 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
3845 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
3846 | ||
17a52670 AS |
3847 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
3848 | /* case: R1 = R2 | |
3849 | * copy register state to dest reg | |
3850 | */ | |
e434b8cd JW |
3851 | *dst_reg = *src_reg; |
3852 | dst_reg->live |= REG_LIVE_WRITTEN; | |
17a52670 | 3853 | } else { |
f1174f77 | 3854 | /* R1 = (u32) R2 */ |
1be7f75d | 3855 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
3856 | verbose(env, |
3857 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
3858 | insn->src_reg); |
3859 | return -EACCES; | |
e434b8cd JW |
3860 | } else if (src_reg->type == SCALAR_VALUE) { |
3861 | *dst_reg = *src_reg; | |
3862 | dst_reg->live |= REG_LIVE_WRITTEN; | |
3863 | } else { | |
3864 | mark_reg_unknown(env, regs, | |
3865 | insn->dst_reg); | |
1be7f75d | 3866 | } |
e434b8cd | 3867 | coerce_reg_to_size(dst_reg, 4); |
17a52670 AS |
3868 | } |
3869 | } else { | |
3870 | /* case: R = imm | |
3871 | * remember the value we stored into this reg | |
3872 | */ | |
fbeb1603 AF |
3873 | /* clear any state __mark_reg_known doesn't set */ |
3874 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 3875 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
3876 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
3877 | __mark_reg_known(regs + insn->dst_reg, | |
3878 | insn->imm); | |
3879 | } else { | |
3880 | __mark_reg_known(regs + insn->dst_reg, | |
3881 | (u32)insn->imm); | |
3882 | } | |
17a52670 AS |
3883 | } |
3884 | ||
3885 | } else if (opcode > BPF_END) { | |
61bd5218 | 3886 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
3887 | return -EINVAL; |
3888 | ||
3889 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
3890 | ||
17a52670 AS |
3891 | if (BPF_SRC(insn->code) == BPF_X) { |
3892 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 3893 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
3894 | return -EINVAL; |
3895 | } | |
3896 | /* check src1 operand */ | |
dc503a8a | 3897 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3898 | if (err) |
3899 | return err; | |
3900 | } else { | |
3901 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 3902 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
3903 | return -EINVAL; |
3904 | } | |
3905 | } | |
3906 | ||
3907 | /* check src2 operand */ | |
dc503a8a | 3908 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3909 | if (err) |
3910 | return err; | |
3911 | ||
3912 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
3913 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 3914 | verbose(env, "div by zero\n"); |
17a52670 AS |
3915 | return -EINVAL; |
3916 | } | |
3917 | ||
229394e8 RV |
3918 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
3919 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
3920 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
3921 | ||
3922 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 3923 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
3924 | return -EINVAL; |
3925 | } | |
3926 | } | |
3927 | ||
1a0dc1ac | 3928 | /* check dest operand */ |
dc503a8a | 3929 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
3930 | if (err) |
3931 | return err; | |
3932 | ||
f1174f77 | 3933 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
3934 | } |
3935 | ||
3936 | return 0; | |
3937 | } | |
3938 | ||
f4d7e40a | 3939 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 3940 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 3941 | enum bpf_reg_type type, |
fb2a311a | 3942 | bool range_right_open) |
969bf05e | 3943 | { |
f4d7e40a | 3944 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
58e2af8b | 3945 | struct bpf_reg_state *regs = state->regs, *reg; |
fb2a311a | 3946 | u16 new_range; |
f4d7e40a | 3947 | int i, j; |
2d2be8ca | 3948 | |
fb2a311a DB |
3949 | if (dst_reg->off < 0 || |
3950 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
3951 | /* This doesn't give us any range */ |
3952 | return; | |
3953 | ||
b03c9f9f EC |
3954 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
3955 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
3956 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
3957 | * than pkt_end, but that's because it's also less than pkt. | |
3958 | */ | |
3959 | return; | |
3960 | ||
fb2a311a DB |
3961 | new_range = dst_reg->off; |
3962 | if (range_right_open) | |
3963 | new_range--; | |
3964 | ||
3965 | /* Examples for register markings: | |
2d2be8ca | 3966 | * |
fb2a311a | 3967 | * pkt_data in dst register: |
2d2be8ca DB |
3968 | * |
3969 | * r2 = r3; | |
3970 | * r2 += 8; | |
3971 | * if (r2 > pkt_end) goto <handle exception> | |
3972 | * <access okay> | |
3973 | * | |
b4e432f1 DB |
3974 | * r2 = r3; |
3975 | * r2 += 8; | |
3976 | * if (r2 < pkt_end) goto <access okay> | |
3977 | * <handle exception> | |
3978 | * | |
2d2be8ca DB |
3979 | * Where: |
3980 | * r2 == dst_reg, pkt_end == src_reg | |
3981 | * r2=pkt(id=n,off=8,r=0) | |
3982 | * r3=pkt(id=n,off=0,r=0) | |
3983 | * | |
fb2a311a | 3984 | * pkt_data in src register: |
2d2be8ca DB |
3985 | * |
3986 | * r2 = r3; | |
3987 | * r2 += 8; | |
3988 | * if (pkt_end >= r2) goto <access okay> | |
3989 | * <handle exception> | |
3990 | * | |
b4e432f1 DB |
3991 | * r2 = r3; |
3992 | * r2 += 8; | |
3993 | * if (pkt_end <= r2) goto <handle exception> | |
3994 | * <access okay> | |
3995 | * | |
2d2be8ca DB |
3996 | * Where: |
3997 | * pkt_end == dst_reg, r2 == src_reg | |
3998 | * r2=pkt(id=n,off=8,r=0) | |
3999 | * r3=pkt(id=n,off=0,r=0) | |
4000 | * | |
4001 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
4002 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
4003 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
4004 | * the check. | |
969bf05e | 4005 | */ |
2d2be8ca | 4006 | |
f1174f77 EC |
4007 | /* If our ids match, then we must have the same max_value. And we |
4008 | * don't care about the other reg's fixed offset, since if it's too big | |
4009 | * the range won't allow anything. | |
4010 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
4011 | */ | |
969bf05e | 4012 | for (i = 0; i < MAX_BPF_REG; i++) |
de8f3a83 | 4013 | if (regs[i].type == type && regs[i].id == dst_reg->id) |
b1977682 | 4014 | /* keep the maximum range already checked */ |
fb2a311a | 4015 | regs[i].range = max(regs[i].range, new_range); |
969bf05e | 4016 | |
f4d7e40a AS |
4017 | for (j = 0; j <= vstate->curframe; j++) { |
4018 | state = vstate->frame[j]; | |
f3709f69 JS |
4019 | bpf_for_each_spilled_reg(i, state, reg) { |
4020 | if (!reg) | |
f4d7e40a | 4021 | continue; |
f4d7e40a AS |
4022 | if (reg->type == type && reg->id == dst_reg->id) |
4023 | reg->range = max(reg->range, new_range); | |
4024 | } | |
969bf05e AS |
4025 | } |
4026 | } | |
4027 | ||
4f7b3e82 AS |
4028 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
4029 | * and return: | |
4030 | * 1 - branch will be taken and "goto target" will be executed | |
4031 | * 0 - branch will not be taken and fall-through to next insn | |
4032 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value range [0,10] | |
4033 | */ | |
4034 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
4035 | { | |
4036 | if (__is_pointer_value(false, reg)) | |
4037 | return -1; | |
4038 | ||
4039 | switch (opcode) { | |
4040 | case BPF_JEQ: | |
4041 | if (tnum_is_const(reg->var_off)) | |
4042 | return !!tnum_equals_const(reg->var_off, val); | |
4043 | break; | |
4044 | case BPF_JNE: | |
4045 | if (tnum_is_const(reg->var_off)) | |
4046 | return !tnum_equals_const(reg->var_off, val); | |
4047 | break; | |
960ea056 JK |
4048 | case BPF_JSET: |
4049 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
4050 | return 1; | |
4051 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
4052 | return 0; | |
4053 | break; | |
4f7b3e82 AS |
4054 | case BPF_JGT: |
4055 | if (reg->umin_value > val) | |
4056 | return 1; | |
4057 | else if (reg->umax_value <= val) | |
4058 | return 0; | |
4059 | break; | |
4060 | case BPF_JSGT: | |
4061 | if (reg->smin_value > (s64)val) | |
4062 | return 1; | |
4063 | else if (reg->smax_value < (s64)val) | |
4064 | return 0; | |
4065 | break; | |
4066 | case BPF_JLT: | |
4067 | if (reg->umax_value < val) | |
4068 | return 1; | |
4069 | else if (reg->umin_value >= val) | |
4070 | return 0; | |
4071 | break; | |
4072 | case BPF_JSLT: | |
4073 | if (reg->smax_value < (s64)val) | |
4074 | return 1; | |
4075 | else if (reg->smin_value >= (s64)val) | |
4076 | return 0; | |
4077 | break; | |
4078 | case BPF_JGE: | |
4079 | if (reg->umin_value >= val) | |
4080 | return 1; | |
4081 | else if (reg->umax_value < val) | |
4082 | return 0; | |
4083 | break; | |
4084 | case BPF_JSGE: | |
4085 | if (reg->smin_value >= (s64)val) | |
4086 | return 1; | |
4087 | else if (reg->smax_value < (s64)val) | |
4088 | return 0; | |
4089 | break; | |
4090 | case BPF_JLE: | |
4091 | if (reg->umax_value <= val) | |
4092 | return 1; | |
4093 | else if (reg->umin_value > val) | |
4094 | return 0; | |
4095 | break; | |
4096 | case BPF_JSLE: | |
4097 | if (reg->smax_value <= (s64)val) | |
4098 | return 1; | |
4099 | else if (reg->smin_value > (s64)val) | |
4100 | return 0; | |
4101 | break; | |
4102 | } | |
4103 | ||
4104 | return -1; | |
4105 | } | |
4106 | ||
48461135 JB |
4107 | /* Adjusts the register min/max values in the case that the dst_reg is the |
4108 | * variable register that we are working on, and src_reg is a constant or we're | |
4109 | * simply doing a BPF_K check. | |
f1174f77 | 4110 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
4111 | */ |
4112 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
4113 | struct bpf_reg_state *false_reg, u64 val, | |
4114 | u8 opcode) | |
4115 | { | |
f1174f77 EC |
4116 | /* If the dst_reg is a pointer, we can't learn anything about its |
4117 | * variable offset from the compare (unless src_reg were a pointer into | |
4118 | * the same object, but we don't bother with that. | |
4119 | * Since false_reg and true_reg have the same type by construction, we | |
4120 | * only need to check one of them for pointerness. | |
4121 | */ | |
4122 | if (__is_pointer_value(false, false_reg)) | |
4123 | return; | |
4cabc5b1 | 4124 | |
48461135 JB |
4125 | switch (opcode) { |
4126 | case BPF_JEQ: | |
4127 | /* If this is false then we know nothing Jon Snow, but if it is | |
4128 | * true then we know for sure. | |
4129 | */ | |
b03c9f9f | 4130 | __mark_reg_known(true_reg, val); |
48461135 JB |
4131 | break; |
4132 | case BPF_JNE: | |
4133 | /* If this is true we know nothing Jon Snow, but if it is false | |
4134 | * we know the value for sure; | |
4135 | */ | |
b03c9f9f | 4136 | __mark_reg_known(false_reg, val); |
48461135 | 4137 | break; |
960ea056 JK |
4138 | case BPF_JSET: |
4139 | false_reg->var_off = tnum_and(false_reg->var_off, | |
4140 | tnum_const(~val)); | |
4141 | if (is_power_of_2(val)) | |
4142 | true_reg->var_off = tnum_or(true_reg->var_off, | |
4143 | tnum_const(val)); | |
4144 | break; | |
48461135 | 4145 | case BPF_JGT: |
b03c9f9f EC |
4146 | false_reg->umax_value = min(false_reg->umax_value, val); |
4147 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
4148 | break; | |
48461135 | 4149 | case BPF_JSGT: |
b03c9f9f EC |
4150 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); |
4151 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
48461135 | 4152 | break; |
b4e432f1 DB |
4153 | case BPF_JLT: |
4154 | false_reg->umin_value = max(false_reg->umin_value, val); | |
4155 | true_reg->umax_value = min(true_reg->umax_value, val - 1); | |
4156 | break; | |
4157 | case BPF_JSLT: | |
4158 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
4159 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); | |
4160 | break; | |
48461135 | 4161 | case BPF_JGE: |
b03c9f9f EC |
4162 | false_reg->umax_value = min(false_reg->umax_value, val - 1); |
4163 | true_reg->umin_value = max(true_reg->umin_value, val); | |
4164 | break; | |
48461135 | 4165 | case BPF_JSGE: |
b03c9f9f EC |
4166 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); |
4167 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
48461135 | 4168 | break; |
b4e432f1 DB |
4169 | case BPF_JLE: |
4170 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
4171 | true_reg->umax_value = min(true_reg->umax_value, val); | |
4172 | break; | |
4173 | case BPF_JSLE: | |
4174 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
4175 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); | |
4176 | break; | |
48461135 JB |
4177 | default: |
4178 | break; | |
4179 | } | |
4180 | ||
b03c9f9f EC |
4181 | __reg_deduce_bounds(false_reg); |
4182 | __reg_deduce_bounds(true_reg); | |
4183 | /* We might have learned some bits from the bounds. */ | |
4184 | __reg_bound_offset(false_reg); | |
4185 | __reg_bound_offset(true_reg); | |
4186 | /* Intersecting with the old var_off might have improved our bounds | |
4187 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
4188 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
4189 | */ | |
4190 | __update_reg_bounds(false_reg); | |
4191 | __update_reg_bounds(true_reg); | |
48461135 JB |
4192 | } |
4193 | ||
f1174f77 EC |
4194 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
4195 | * the variable reg. | |
48461135 JB |
4196 | */ |
4197 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
4198 | struct bpf_reg_state *false_reg, u64 val, | |
4199 | u8 opcode) | |
4200 | { | |
f1174f77 EC |
4201 | if (__is_pointer_value(false, false_reg)) |
4202 | return; | |
4cabc5b1 | 4203 | |
48461135 JB |
4204 | switch (opcode) { |
4205 | case BPF_JEQ: | |
4206 | /* If this is false then we know nothing Jon Snow, but if it is | |
4207 | * true then we know for sure. | |
4208 | */ | |
b03c9f9f | 4209 | __mark_reg_known(true_reg, val); |
48461135 JB |
4210 | break; |
4211 | case BPF_JNE: | |
4212 | /* If this is true we know nothing Jon Snow, but if it is false | |
4213 | * we know the value for sure; | |
4214 | */ | |
b03c9f9f | 4215 | __mark_reg_known(false_reg, val); |
48461135 | 4216 | break; |
960ea056 JK |
4217 | case BPF_JSET: |
4218 | false_reg->var_off = tnum_and(false_reg->var_off, | |
4219 | tnum_const(~val)); | |
4220 | if (is_power_of_2(val)) | |
4221 | true_reg->var_off = tnum_or(true_reg->var_off, | |
4222 | tnum_const(val)); | |
4223 | break; | |
48461135 | 4224 | case BPF_JGT: |
b03c9f9f EC |
4225 | true_reg->umax_value = min(true_reg->umax_value, val - 1); |
4226 | false_reg->umin_value = max(false_reg->umin_value, val); | |
4227 | break; | |
48461135 | 4228 | case BPF_JSGT: |
b03c9f9f EC |
4229 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val - 1); |
4230 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val); | |
48461135 | 4231 | break; |
b4e432f1 DB |
4232 | case BPF_JLT: |
4233 | true_reg->umin_value = max(true_reg->umin_value, val + 1); | |
4234 | false_reg->umax_value = min(false_reg->umax_value, val); | |
4235 | break; | |
4236 | case BPF_JSLT: | |
4237 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val + 1); | |
4238 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val); | |
4239 | break; | |
48461135 | 4240 | case BPF_JGE: |
b03c9f9f EC |
4241 | true_reg->umax_value = min(true_reg->umax_value, val); |
4242 | false_reg->umin_value = max(false_reg->umin_value, val + 1); | |
4243 | break; | |
48461135 | 4244 | case BPF_JSGE: |
b03c9f9f EC |
4245 | true_reg->smax_value = min_t(s64, true_reg->smax_value, val); |
4246 | false_reg->smin_value = max_t(s64, false_reg->smin_value, val + 1); | |
48461135 | 4247 | break; |
b4e432f1 DB |
4248 | case BPF_JLE: |
4249 | true_reg->umin_value = max(true_reg->umin_value, val); | |
4250 | false_reg->umax_value = min(false_reg->umax_value, val - 1); | |
4251 | break; | |
4252 | case BPF_JSLE: | |
4253 | true_reg->smin_value = max_t(s64, true_reg->smin_value, val); | |
4254 | false_reg->smax_value = min_t(s64, false_reg->smax_value, val - 1); | |
4255 | break; | |
48461135 JB |
4256 | default: |
4257 | break; | |
4258 | } | |
4259 | ||
b03c9f9f EC |
4260 | __reg_deduce_bounds(false_reg); |
4261 | __reg_deduce_bounds(true_reg); | |
4262 | /* We might have learned some bits from the bounds. */ | |
4263 | __reg_bound_offset(false_reg); | |
4264 | __reg_bound_offset(true_reg); | |
4265 | /* Intersecting with the old var_off might have improved our bounds | |
4266 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
4267 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
4268 | */ | |
4269 | __update_reg_bounds(false_reg); | |
4270 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
4271 | } |
4272 | ||
4273 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
4274 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
4275 | struct bpf_reg_state *dst_reg) | |
4276 | { | |
b03c9f9f EC |
4277 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
4278 | dst_reg->umin_value); | |
4279 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
4280 | dst_reg->umax_value); | |
4281 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
4282 | dst_reg->smin_value); | |
4283 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
4284 | dst_reg->smax_value); | |
f1174f77 EC |
4285 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
4286 | dst_reg->var_off); | |
b03c9f9f EC |
4287 | /* We might have learned new bounds from the var_off. */ |
4288 | __update_reg_bounds(src_reg); | |
4289 | __update_reg_bounds(dst_reg); | |
4290 | /* We might have learned something about the sign bit. */ | |
4291 | __reg_deduce_bounds(src_reg); | |
4292 | __reg_deduce_bounds(dst_reg); | |
4293 | /* We might have learned some bits from the bounds. */ | |
4294 | __reg_bound_offset(src_reg); | |
4295 | __reg_bound_offset(dst_reg); | |
4296 | /* Intersecting with the old var_off might have improved our bounds | |
4297 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
4298 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
4299 | */ | |
4300 | __update_reg_bounds(src_reg); | |
4301 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
4302 | } |
4303 | ||
4304 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
4305 | struct bpf_reg_state *true_dst, | |
4306 | struct bpf_reg_state *false_src, | |
4307 | struct bpf_reg_state *false_dst, | |
4308 | u8 opcode) | |
4309 | { | |
4310 | switch (opcode) { | |
4311 | case BPF_JEQ: | |
4312 | __reg_combine_min_max(true_src, true_dst); | |
4313 | break; | |
4314 | case BPF_JNE: | |
4315 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 4316 | break; |
4cabc5b1 | 4317 | } |
48461135 JB |
4318 | } |
4319 | ||
fd978bf7 JS |
4320 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
4321 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 4322 | bool is_null) |
57a09bf0 | 4323 | { |
840b9615 | 4324 | if (reg_type_may_be_null(reg->type) && reg->id == id) { |
f1174f77 EC |
4325 | /* Old offset (both fixed and variable parts) should |
4326 | * have been known-zero, because we don't allow pointer | |
4327 | * arithmetic on pointers that might be NULL. | |
4328 | */ | |
b03c9f9f EC |
4329 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
4330 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 4331 | reg->off)) { |
b03c9f9f EC |
4332 | __mark_reg_known_zero(reg); |
4333 | reg->off = 0; | |
f1174f77 EC |
4334 | } |
4335 | if (is_null) { | |
4336 | reg->type = SCALAR_VALUE; | |
840b9615 JS |
4337 | } else if (reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
4338 | if (reg->map_ptr->inner_map_meta) { | |
4339 | reg->type = CONST_PTR_TO_MAP; | |
4340 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
4341 | } else { | |
4342 | reg->type = PTR_TO_MAP_VALUE; | |
4343 | } | |
c64b7983 JS |
4344 | } else if (reg->type == PTR_TO_SOCKET_OR_NULL) { |
4345 | reg->type = PTR_TO_SOCKET; | |
56f668df | 4346 | } |
fd978bf7 JS |
4347 | if (is_null || !reg_is_refcounted(reg)) { |
4348 | /* We don't need id from this point onwards anymore, | |
4349 | * thus we should better reset it, so that state | |
4350 | * pruning has chances to take effect. | |
4351 | */ | |
4352 | reg->id = 0; | |
56f668df | 4353 | } |
57a09bf0 TG |
4354 | } |
4355 | } | |
4356 | ||
4357 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
4358 | * be folded together at some point. | |
4359 | */ | |
840b9615 JS |
4360 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
4361 | bool is_null) | |
57a09bf0 | 4362 | { |
f4d7e40a | 4363 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
f3709f69 | 4364 | struct bpf_reg_state *reg, *regs = state->regs; |
a08dd0da | 4365 | u32 id = regs[regno].id; |
f4d7e40a | 4366 | int i, j; |
57a09bf0 | 4367 | |
fd978bf7 JS |
4368 | if (reg_is_refcounted_or_null(®s[regno]) && is_null) |
4369 | __release_reference_state(state, id); | |
4370 | ||
57a09bf0 | 4371 | for (i = 0; i < MAX_BPF_REG; i++) |
fd978bf7 | 4372 | mark_ptr_or_null_reg(state, ®s[i], id, is_null); |
57a09bf0 | 4373 | |
f4d7e40a AS |
4374 | for (j = 0; j <= vstate->curframe; j++) { |
4375 | state = vstate->frame[j]; | |
f3709f69 JS |
4376 | bpf_for_each_spilled_reg(i, state, reg) { |
4377 | if (!reg) | |
f4d7e40a | 4378 | continue; |
fd978bf7 | 4379 | mark_ptr_or_null_reg(state, reg, id, is_null); |
f4d7e40a | 4380 | } |
57a09bf0 TG |
4381 | } |
4382 | } | |
4383 | ||
5beca081 DB |
4384 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
4385 | struct bpf_reg_state *dst_reg, | |
4386 | struct bpf_reg_state *src_reg, | |
4387 | struct bpf_verifier_state *this_branch, | |
4388 | struct bpf_verifier_state *other_branch) | |
4389 | { | |
4390 | if (BPF_SRC(insn->code) != BPF_X) | |
4391 | return false; | |
4392 | ||
4393 | switch (BPF_OP(insn->code)) { | |
4394 | case BPF_JGT: | |
4395 | if ((dst_reg->type == PTR_TO_PACKET && | |
4396 | src_reg->type == PTR_TO_PACKET_END) || | |
4397 | (dst_reg->type == PTR_TO_PACKET_META && | |
4398 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
4399 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
4400 | find_good_pkt_pointers(this_branch, dst_reg, | |
4401 | dst_reg->type, false); | |
4402 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
4403 | src_reg->type == PTR_TO_PACKET) || | |
4404 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
4405 | src_reg->type == PTR_TO_PACKET_META)) { | |
4406 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
4407 | find_good_pkt_pointers(other_branch, src_reg, | |
4408 | src_reg->type, true); | |
4409 | } else { | |
4410 | return false; | |
4411 | } | |
4412 | break; | |
4413 | case BPF_JLT: | |
4414 | if ((dst_reg->type == PTR_TO_PACKET && | |
4415 | src_reg->type == PTR_TO_PACKET_END) || | |
4416 | (dst_reg->type == PTR_TO_PACKET_META && | |
4417 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
4418 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
4419 | find_good_pkt_pointers(other_branch, dst_reg, | |
4420 | dst_reg->type, true); | |
4421 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
4422 | src_reg->type == PTR_TO_PACKET) || | |
4423 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
4424 | src_reg->type == PTR_TO_PACKET_META)) { | |
4425 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
4426 | find_good_pkt_pointers(this_branch, src_reg, | |
4427 | src_reg->type, false); | |
4428 | } else { | |
4429 | return false; | |
4430 | } | |
4431 | break; | |
4432 | case BPF_JGE: | |
4433 | if ((dst_reg->type == PTR_TO_PACKET && | |
4434 | src_reg->type == PTR_TO_PACKET_END) || | |
4435 | (dst_reg->type == PTR_TO_PACKET_META && | |
4436 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
4437 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
4438 | find_good_pkt_pointers(this_branch, dst_reg, | |
4439 | dst_reg->type, true); | |
4440 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
4441 | src_reg->type == PTR_TO_PACKET) || | |
4442 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
4443 | src_reg->type == PTR_TO_PACKET_META)) { | |
4444 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
4445 | find_good_pkt_pointers(other_branch, src_reg, | |
4446 | src_reg->type, false); | |
4447 | } else { | |
4448 | return false; | |
4449 | } | |
4450 | break; | |
4451 | case BPF_JLE: | |
4452 | if ((dst_reg->type == PTR_TO_PACKET && | |
4453 | src_reg->type == PTR_TO_PACKET_END) || | |
4454 | (dst_reg->type == PTR_TO_PACKET_META && | |
4455 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
4456 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
4457 | find_good_pkt_pointers(other_branch, dst_reg, | |
4458 | dst_reg->type, false); | |
4459 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
4460 | src_reg->type == PTR_TO_PACKET) || | |
4461 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
4462 | src_reg->type == PTR_TO_PACKET_META)) { | |
4463 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
4464 | find_good_pkt_pointers(this_branch, src_reg, | |
4465 | src_reg->type, true); | |
4466 | } else { | |
4467 | return false; | |
4468 | } | |
4469 | break; | |
4470 | default: | |
4471 | return false; | |
4472 | } | |
4473 | ||
4474 | return true; | |
4475 | } | |
4476 | ||
58e2af8b | 4477 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
4478 | struct bpf_insn *insn, int *insn_idx) |
4479 | { | |
f4d7e40a AS |
4480 | struct bpf_verifier_state *this_branch = env->cur_state; |
4481 | struct bpf_verifier_state *other_branch; | |
4482 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
4483 | struct bpf_reg_state *dst_reg, *other_branch_regs; | |
17a52670 AS |
4484 | u8 opcode = BPF_OP(insn->code); |
4485 | int err; | |
4486 | ||
b4e432f1 | 4487 | if (opcode > BPF_JSLE) { |
61bd5218 | 4488 | verbose(env, "invalid BPF_JMP opcode %x\n", opcode); |
17a52670 AS |
4489 | return -EINVAL; |
4490 | } | |
4491 | ||
4492 | if (BPF_SRC(insn->code) == BPF_X) { | |
4493 | if (insn->imm != 0) { | |
61bd5218 | 4494 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
4495 | return -EINVAL; |
4496 | } | |
4497 | ||
4498 | /* check src1 operand */ | |
dc503a8a | 4499 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4500 | if (err) |
4501 | return err; | |
1be7f75d AS |
4502 | |
4503 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 4504 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
4505 | insn->src_reg); |
4506 | return -EACCES; | |
4507 | } | |
17a52670 AS |
4508 | } else { |
4509 | if (insn->src_reg != BPF_REG_0) { | |
61bd5218 | 4510 | verbose(env, "BPF_JMP uses reserved fields\n"); |
17a52670 AS |
4511 | return -EINVAL; |
4512 | } | |
4513 | } | |
4514 | ||
4515 | /* check src2 operand */ | |
dc503a8a | 4516 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4517 | if (err) |
4518 | return err; | |
4519 | ||
1a0dc1ac AS |
4520 | dst_reg = ®s[insn->dst_reg]; |
4521 | ||
4f7b3e82 AS |
4522 | if (BPF_SRC(insn->code) == BPF_K) { |
4523 | int pred = is_branch_taken(dst_reg, insn->imm, opcode); | |
4524 | ||
4525 | if (pred == 1) { | |
4526 | /* only follow the goto, ignore fall-through */ | |
17a52670 AS |
4527 | *insn_idx += insn->off; |
4528 | return 0; | |
4f7b3e82 AS |
4529 | } else if (pred == 0) { |
4530 | /* only follow fall-through branch, since | |
17a52670 AS |
4531 | * that's where the program will go |
4532 | */ | |
4533 | return 0; | |
4534 | } | |
4535 | } | |
4536 | ||
979d63d5 DB |
4537 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
4538 | false); | |
17a52670 AS |
4539 | if (!other_branch) |
4540 | return -EFAULT; | |
f4d7e40a | 4541 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 4542 | |
48461135 JB |
4543 | /* detect if we are comparing against a constant value so we can adjust |
4544 | * our min/max values for our dst register. | |
f1174f77 EC |
4545 | * this is only legit if both are scalars (or pointers to the same |
4546 | * object, I suppose, but we don't support that right now), because | |
4547 | * otherwise the different base pointers mean the offsets aren't | |
4548 | * comparable. | |
48461135 JB |
4549 | */ |
4550 | if (BPF_SRC(insn->code) == BPF_X) { | |
f1174f77 EC |
4551 | if (dst_reg->type == SCALAR_VALUE && |
4552 | regs[insn->src_reg].type == SCALAR_VALUE) { | |
4553 | if (tnum_is_const(regs[insn->src_reg].var_off)) | |
f4d7e40a | 4554 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
f1174f77 EC |
4555 | dst_reg, regs[insn->src_reg].var_off.value, |
4556 | opcode); | |
4557 | else if (tnum_is_const(dst_reg->var_off)) | |
f4d7e40a | 4558 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
f1174f77 EC |
4559 | ®s[insn->src_reg], |
4560 | dst_reg->var_off.value, opcode); | |
4561 | else if (opcode == BPF_JEQ || opcode == BPF_JNE) | |
4562 | /* Comparing for equality, we can combine knowledge */ | |
f4d7e40a AS |
4563 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
4564 | &other_branch_regs[insn->dst_reg], | |
f1174f77 EC |
4565 | ®s[insn->src_reg], |
4566 | ®s[insn->dst_reg], opcode); | |
4567 | } | |
4568 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 4569 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
48461135 JB |
4570 | dst_reg, insn->imm, opcode); |
4571 | } | |
4572 | ||
58e2af8b | 4573 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 4574 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac | 4575 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
840b9615 JS |
4576 | reg_type_may_be_null(dst_reg->type)) { |
4577 | /* Mark all identical registers in each branch as either | |
57a09bf0 TG |
4578 | * safe or unknown depending R == 0 or R != 0 conditional. |
4579 | */ | |
840b9615 JS |
4580 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
4581 | opcode == BPF_JNE); | |
4582 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
4583 | opcode == BPF_JEQ); | |
5beca081 DB |
4584 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
4585 | this_branch, other_branch) && | |
4586 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
4587 | verbose(env, "R%d pointer comparison prohibited\n", |
4588 | insn->dst_reg); | |
1be7f75d | 4589 | return -EACCES; |
17a52670 | 4590 | } |
61bd5218 | 4591 | if (env->log.level) |
f4d7e40a | 4592 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
4593 | return 0; |
4594 | } | |
4595 | ||
0246e64d AS |
4596 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
4597 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
4598 | { | |
4599 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
4600 | ||
4601 | return (struct bpf_map *) (unsigned long) imm64; | |
4602 | } | |
4603 | ||
17a52670 | 4604 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 4605 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 4606 | { |
638f5b90 | 4607 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
4608 | int err; |
4609 | ||
4610 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 4611 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
4612 | return -EINVAL; |
4613 | } | |
4614 | if (insn->off != 0) { | |
61bd5218 | 4615 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
4616 | return -EINVAL; |
4617 | } | |
4618 | ||
dc503a8a | 4619 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
4620 | if (err) |
4621 | return err; | |
4622 | ||
6b173873 | 4623 | if (insn->src_reg == 0) { |
6b173873 JK |
4624 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
4625 | ||
f1174f77 | 4626 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 4627 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 4628 | return 0; |
6b173873 | 4629 | } |
17a52670 AS |
4630 | |
4631 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
4632 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
4633 | ||
4634 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
4635 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
4636 | return 0; | |
4637 | } | |
4638 | ||
96be4325 DB |
4639 | static bool may_access_skb(enum bpf_prog_type type) |
4640 | { | |
4641 | switch (type) { | |
4642 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
4643 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 4644 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
4645 | return true; |
4646 | default: | |
4647 | return false; | |
4648 | } | |
4649 | } | |
4650 | ||
ddd872bc AS |
4651 | /* verify safety of LD_ABS|LD_IND instructions: |
4652 | * - they can only appear in the programs where ctx == skb | |
4653 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
4654 | * preserve R6-R9, and store return value into R0 | |
4655 | * | |
4656 | * Implicit input: | |
4657 | * ctx == skb == R6 == CTX | |
4658 | * | |
4659 | * Explicit input: | |
4660 | * SRC == any register | |
4661 | * IMM == 32-bit immediate | |
4662 | * | |
4663 | * Output: | |
4664 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
4665 | */ | |
58e2af8b | 4666 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 4667 | { |
638f5b90 | 4668 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 4669 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
4670 | int i, err; |
4671 | ||
24701ece | 4672 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 4673 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
4674 | return -EINVAL; |
4675 | } | |
4676 | ||
e0cea7ce DB |
4677 | if (!env->ops->gen_ld_abs) { |
4678 | verbose(env, "bpf verifier is misconfigured\n"); | |
4679 | return -EINVAL; | |
4680 | } | |
4681 | ||
f910cefa | 4682 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
4683 | /* when program has LD_ABS insn JITs and interpreter assume |
4684 | * that r1 == ctx == skb which is not the case for callees | |
4685 | * that can have arbitrary arguments. It's problematic | |
4686 | * for main prog as well since JITs would need to analyze | |
4687 | * all functions in order to make proper register save/restore | |
4688 | * decisions in the main prog. Hence disallow LD_ABS with calls | |
4689 | */ | |
4690 | verbose(env, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n"); | |
4691 | return -EINVAL; | |
4692 | } | |
4693 | ||
ddd872bc | 4694 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 4695 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 4696 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 4697 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
4698 | return -EINVAL; |
4699 | } | |
4700 | ||
4701 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 4702 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
4703 | if (err) |
4704 | return err; | |
4705 | ||
fd978bf7 JS |
4706 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
4707 | * gen_ld_abs() may terminate the program at runtime, leading to | |
4708 | * reference leak. | |
4709 | */ | |
4710 | err = check_reference_leak(env); | |
4711 | if (err) { | |
4712 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
4713 | return err; | |
4714 | } | |
4715 | ||
ddd872bc | 4716 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { |
61bd5218 JK |
4717 | verbose(env, |
4718 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
4719 | return -EINVAL; |
4720 | } | |
4721 | ||
4722 | if (mode == BPF_IND) { | |
4723 | /* check explicit source operand */ | |
dc503a8a | 4724 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
4725 | if (err) |
4726 | return err; | |
4727 | } | |
4728 | ||
4729 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 4730 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 4731 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
4732 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
4733 | } | |
ddd872bc AS |
4734 | |
4735 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
4736 | * the value fetched from the packet. |
4737 | * Already marked as written above. | |
ddd872bc | 4738 | */ |
61bd5218 | 4739 | mark_reg_unknown(env, regs, BPF_REG_0); |
ddd872bc AS |
4740 | return 0; |
4741 | } | |
4742 | ||
390ee7e2 AS |
4743 | static int check_return_code(struct bpf_verifier_env *env) |
4744 | { | |
4745 | struct bpf_reg_state *reg; | |
4746 | struct tnum range = tnum_range(0, 1); | |
4747 | ||
4748 | switch (env->prog->type) { | |
4749 | case BPF_PROG_TYPE_CGROUP_SKB: | |
4750 | case BPF_PROG_TYPE_CGROUP_SOCK: | |
4fbac77d | 4751 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
390ee7e2 | 4752 | case BPF_PROG_TYPE_SOCK_OPS: |
ebc614f6 | 4753 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
390ee7e2 AS |
4754 | break; |
4755 | default: | |
4756 | return 0; | |
4757 | } | |
4758 | ||
638f5b90 | 4759 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 4760 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 4761 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
4762 | reg_type_str[reg->type]); |
4763 | return -EINVAL; | |
4764 | } | |
4765 | ||
4766 | if (!tnum_in(range, reg->var_off)) { | |
61bd5218 | 4767 | verbose(env, "At program exit the register R0 "); |
390ee7e2 AS |
4768 | if (!tnum_is_unknown(reg->var_off)) { |
4769 | char tn_buf[48]; | |
4770 | ||
4771 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 4772 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 4773 | } else { |
61bd5218 | 4774 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 4775 | } |
61bd5218 | 4776 | verbose(env, " should have been 0 or 1\n"); |
390ee7e2 AS |
4777 | return -EINVAL; |
4778 | } | |
4779 | return 0; | |
4780 | } | |
4781 | ||
475fb78f AS |
4782 | /* non-recursive DFS pseudo code |
4783 | * 1 procedure DFS-iterative(G,v): | |
4784 | * 2 label v as discovered | |
4785 | * 3 let S be a stack | |
4786 | * 4 S.push(v) | |
4787 | * 5 while S is not empty | |
4788 | * 6 t <- S.pop() | |
4789 | * 7 if t is what we're looking for: | |
4790 | * 8 return t | |
4791 | * 9 for all edges e in G.adjacentEdges(t) do | |
4792 | * 10 if edge e is already labelled | |
4793 | * 11 continue with the next edge | |
4794 | * 12 w <- G.adjacentVertex(t,e) | |
4795 | * 13 if vertex w is not discovered and not explored | |
4796 | * 14 label e as tree-edge | |
4797 | * 15 label w as discovered | |
4798 | * 16 S.push(w) | |
4799 | * 17 continue at 5 | |
4800 | * 18 else if vertex w is discovered | |
4801 | * 19 label e as back-edge | |
4802 | * 20 else | |
4803 | * 21 // vertex w is explored | |
4804 | * 22 label e as forward- or cross-edge | |
4805 | * 23 label t as explored | |
4806 | * 24 S.pop() | |
4807 | * | |
4808 | * convention: | |
4809 | * 0x10 - discovered | |
4810 | * 0x11 - discovered and fall-through edge labelled | |
4811 | * 0x12 - discovered and fall-through and branch edges labelled | |
4812 | * 0x20 - explored | |
4813 | */ | |
4814 | ||
4815 | enum { | |
4816 | DISCOVERED = 0x10, | |
4817 | EXPLORED = 0x20, | |
4818 | FALLTHROUGH = 1, | |
4819 | BRANCH = 2, | |
4820 | }; | |
4821 | ||
58e2af8b | 4822 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 4823 | |
475fb78f AS |
4824 | static int *insn_stack; /* stack of insns to process */ |
4825 | static int cur_stack; /* current stack index */ | |
4826 | static int *insn_state; | |
4827 | ||
4828 | /* t, w, e - match pseudo-code above: | |
4829 | * t - index of current instruction | |
4830 | * w - next instruction | |
4831 | * e - edge | |
4832 | */ | |
58e2af8b | 4833 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
4834 | { |
4835 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
4836 | return 0; | |
4837 | ||
4838 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
4839 | return 0; | |
4840 | ||
4841 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 4842 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 4843 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
4844 | return -EINVAL; |
4845 | } | |
4846 | ||
f1bca824 AS |
4847 | if (e == BRANCH) |
4848 | /* mark branch target for state pruning */ | |
4849 | env->explored_states[w] = STATE_LIST_MARK; | |
4850 | ||
475fb78f AS |
4851 | if (insn_state[w] == 0) { |
4852 | /* tree-edge */ | |
4853 | insn_state[t] = DISCOVERED | e; | |
4854 | insn_state[w] = DISCOVERED; | |
4855 | if (cur_stack >= env->prog->len) | |
4856 | return -E2BIG; | |
4857 | insn_stack[cur_stack++] = w; | |
4858 | return 1; | |
4859 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
d9762e84 MKL |
4860 | verbose_linfo(env, t, "%d: ", t); |
4861 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 4862 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
4863 | return -EINVAL; |
4864 | } else if (insn_state[w] == EXPLORED) { | |
4865 | /* forward- or cross-edge */ | |
4866 | insn_state[t] = DISCOVERED | e; | |
4867 | } else { | |
61bd5218 | 4868 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
4869 | return -EFAULT; |
4870 | } | |
4871 | return 0; | |
4872 | } | |
4873 | ||
4874 | /* non-recursive depth-first-search to detect loops in BPF program | |
4875 | * loop == back-edge in directed graph | |
4876 | */ | |
58e2af8b | 4877 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
4878 | { |
4879 | struct bpf_insn *insns = env->prog->insnsi; | |
4880 | int insn_cnt = env->prog->len; | |
4881 | int ret = 0; | |
4882 | int i, t; | |
4883 | ||
4884 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
4885 | if (!insn_state) | |
4886 | return -ENOMEM; | |
4887 | ||
4888 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
4889 | if (!insn_stack) { | |
4890 | kfree(insn_state); | |
4891 | return -ENOMEM; | |
4892 | } | |
4893 | ||
4894 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
4895 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
4896 | cur_stack = 1; | |
4897 | ||
4898 | peek_stack: | |
4899 | if (cur_stack == 0) | |
4900 | goto check_state; | |
4901 | t = insn_stack[cur_stack - 1]; | |
4902 | ||
4903 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
4904 | u8 opcode = BPF_OP(insns[t].code); | |
4905 | ||
4906 | if (opcode == BPF_EXIT) { | |
4907 | goto mark_explored; | |
4908 | } else if (opcode == BPF_CALL) { | |
4909 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
4910 | if (ret == 1) | |
4911 | goto peek_stack; | |
4912 | else if (ret < 0) | |
4913 | goto err_free; | |
07016151 DB |
4914 | if (t + 1 < insn_cnt) |
4915 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
cc8b0b92 AS |
4916 | if (insns[t].src_reg == BPF_PSEUDO_CALL) { |
4917 | env->explored_states[t] = STATE_LIST_MARK; | |
4918 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env); | |
4919 | if (ret == 1) | |
4920 | goto peek_stack; | |
4921 | else if (ret < 0) | |
4922 | goto err_free; | |
4923 | } | |
475fb78f AS |
4924 | } else if (opcode == BPF_JA) { |
4925 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
4926 | ret = -EINVAL; | |
4927 | goto err_free; | |
4928 | } | |
4929 | /* unconditional jump with single edge */ | |
4930 | ret = push_insn(t, t + insns[t].off + 1, | |
4931 | FALLTHROUGH, env); | |
4932 | if (ret == 1) | |
4933 | goto peek_stack; | |
4934 | else if (ret < 0) | |
4935 | goto err_free; | |
f1bca824 AS |
4936 | /* tell verifier to check for equivalent states |
4937 | * after every call and jump | |
4938 | */ | |
c3de6317 AS |
4939 | if (t + 1 < insn_cnt) |
4940 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
4941 | } else { |
4942 | /* conditional jump with two edges */ | |
3c2ce60b | 4943 | env->explored_states[t] = STATE_LIST_MARK; |
475fb78f AS |
4944 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
4945 | if (ret == 1) | |
4946 | goto peek_stack; | |
4947 | else if (ret < 0) | |
4948 | goto err_free; | |
4949 | ||
4950 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
4951 | if (ret == 1) | |
4952 | goto peek_stack; | |
4953 | else if (ret < 0) | |
4954 | goto err_free; | |
4955 | } | |
4956 | } else { | |
4957 | /* all other non-branch instructions with single | |
4958 | * fall-through edge | |
4959 | */ | |
4960 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
4961 | if (ret == 1) | |
4962 | goto peek_stack; | |
4963 | else if (ret < 0) | |
4964 | goto err_free; | |
4965 | } | |
4966 | ||
4967 | mark_explored: | |
4968 | insn_state[t] = EXPLORED; | |
4969 | if (cur_stack-- <= 0) { | |
61bd5218 | 4970 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
4971 | ret = -EFAULT; |
4972 | goto err_free; | |
4973 | } | |
4974 | goto peek_stack; | |
4975 | ||
4976 | check_state: | |
4977 | for (i = 0; i < insn_cnt; i++) { | |
4978 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 4979 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
4980 | ret = -EINVAL; |
4981 | goto err_free; | |
4982 | } | |
4983 | } | |
4984 | ret = 0; /* cfg looks good */ | |
4985 | ||
4986 | err_free: | |
4987 | kfree(insn_state); | |
4988 | kfree(insn_stack); | |
4989 | return ret; | |
4990 | } | |
4991 | ||
838e9690 YS |
4992 | /* The minimum supported BTF func info size */ |
4993 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
4994 | #define MAX_FUNCINFO_REC_SIZE 252 | |
4995 | ||
c454a46b MKL |
4996 | static int check_btf_func(struct bpf_verifier_env *env, |
4997 | const union bpf_attr *attr, | |
4998 | union bpf_attr __user *uattr) | |
838e9690 | 4999 | { |
d0b2818e | 5000 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 5001 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 5002 | struct bpf_func_info *krecord; |
838e9690 | 5003 | const struct btf_type *type; |
c454a46b MKL |
5004 | struct bpf_prog *prog; |
5005 | const struct btf *btf; | |
838e9690 | 5006 | void __user *urecord; |
d0b2818e | 5007 | u32 prev_offset = 0; |
838e9690 YS |
5008 | int ret = 0; |
5009 | ||
5010 | nfuncs = attr->func_info_cnt; | |
5011 | if (!nfuncs) | |
5012 | return 0; | |
5013 | ||
5014 | if (nfuncs != env->subprog_cnt) { | |
5015 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
5016 | return -EINVAL; | |
5017 | } | |
5018 | ||
5019 | urec_size = attr->func_info_rec_size; | |
5020 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
5021 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
5022 | urec_size % sizeof(u32)) { | |
5023 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
5024 | return -EINVAL; | |
5025 | } | |
5026 | ||
c454a46b MKL |
5027 | prog = env->prog; |
5028 | btf = prog->aux->btf; | |
838e9690 YS |
5029 | |
5030 | urecord = u64_to_user_ptr(attr->func_info); | |
5031 | min_size = min_t(u32, krec_size, urec_size); | |
5032 | ||
ba64e7d8 | 5033 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
5034 | if (!krecord) |
5035 | return -ENOMEM; | |
ba64e7d8 | 5036 | |
838e9690 YS |
5037 | for (i = 0; i < nfuncs; i++) { |
5038 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
5039 | if (ret) { | |
5040 | if (ret == -E2BIG) { | |
5041 | verbose(env, "nonzero tailing record in func info"); | |
5042 | /* set the size kernel expects so loader can zero | |
5043 | * out the rest of the record. | |
5044 | */ | |
5045 | if (put_user(min_size, &uattr->func_info_rec_size)) | |
5046 | ret = -EFAULT; | |
5047 | } | |
c454a46b | 5048 | goto err_free; |
838e9690 YS |
5049 | } |
5050 | ||
ba64e7d8 | 5051 | if (copy_from_user(&krecord[i], urecord, min_size)) { |
838e9690 | 5052 | ret = -EFAULT; |
c454a46b | 5053 | goto err_free; |
838e9690 YS |
5054 | } |
5055 | ||
d30d42e0 | 5056 | /* check insn_off */ |
838e9690 | 5057 | if (i == 0) { |
d30d42e0 | 5058 | if (krecord[i].insn_off) { |
838e9690 | 5059 | verbose(env, |
d30d42e0 MKL |
5060 | "nonzero insn_off %u for the first func info record", |
5061 | krecord[i].insn_off); | |
838e9690 | 5062 | ret = -EINVAL; |
c454a46b | 5063 | goto err_free; |
838e9690 | 5064 | } |
d30d42e0 | 5065 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
5066 | verbose(env, |
5067 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 5068 | krecord[i].insn_off, prev_offset); |
838e9690 | 5069 | ret = -EINVAL; |
c454a46b | 5070 | goto err_free; |
838e9690 YS |
5071 | } |
5072 | ||
d30d42e0 | 5073 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 YS |
5074 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
5075 | ret = -EINVAL; | |
c454a46b | 5076 | goto err_free; |
838e9690 YS |
5077 | } |
5078 | ||
5079 | /* check type_id */ | |
ba64e7d8 | 5080 | type = btf_type_by_id(btf, krecord[i].type_id); |
838e9690 YS |
5081 | if (!type || BTF_INFO_KIND(type->info) != BTF_KIND_FUNC) { |
5082 | verbose(env, "invalid type id %d in func info", | |
ba64e7d8 | 5083 | krecord[i].type_id); |
838e9690 | 5084 | ret = -EINVAL; |
c454a46b | 5085 | goto err_free; |
838e9690 YS |
5086 | } |
5087 | ||
d30d42e0 | 5088 | prev_offset = krecord[i].insn_off; |
838e9690 YS |
5089 | urecord += urec_size; |
5090 | } | |
5091 | ||
ba64e7d8 YS |
5092 | prog->aux->func_info = krecord; |
5093 | prog->aux->func_info_cnt = nfuncs; | |
838e9690 YS |
5094 | return 0; |
5095 | ||
c454a46b | 5096 | err_free: |
ba64e7d8 | 5097 | kvfree(krecord); |
838e9690 YS |
5098 | return ret; |
5099 | } | |
5100 | ||
ba64e7d8 YS |
5101 | static void adjust_btf_func(struct bpf_verifier_env *env) |
5102 | { | |
5103 | int i; | |
5104 | ||
5105 | if (!env->prog->aux->func_info) | |
5106 | return; | |
5107 | ||
5108 | for (i = 0; i < env->subprog_cnt; i++) | |
d30d42e0 | 5109 | env->prog->aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
5110 | } |
5111 | ||
c454a46b MKL |
5112 | #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \ |
5113 | sizeof(((struct bpf_line_info *)(0))->line_col)) | |
5114 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE | |
5115 | ||
5116 | static int check_btf_line(struct bpf_verifier_env *env, | |
5117 | const union bpf_attr *attr, | |
5118 | union bpf_attr __user *uattr) | |
5119 | { | |
5120 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
5121 | struct bpf_subprog_info *sub; | |
5122 | struct bpf_line_info *linfo; | |
5123 | struct bpf_prog *prog; | |
5124 | const struct btf *btf; | |
5125 | void __user *ulinfo; | |
5126 | int err; | |
5127 | ||
5128 | nr_linfo = attr->line_info_cnt; | |
5129 | if (!nr_linfo) | |
5130 | return 0; | |
5131 | ||
5132 | rec_size = attr->line_info_rec_size; | |
5133 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
5134 | rec_size > MAX_LINEINFO_REC_SIZE || | |
5135 | rec_size & (sizeof(u32) - 1)) | |
5136 | return -EINVAL; | |
5137 | ||
5138 | /* Need to zero it in case the userspace may | |
5139 | * pass in a smaller bpf_line_info object. | |
5140 | */ | |
5141 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
5142 | GFP_KERNEL | __GFP_NOWARN); | |
5143 | if (!linfo) | |
5144 | return -ENOMEM; | |
5145 | ||
5146 | prog = env->prog; | |
5147 | btf = prog->aux->btf; | |
5148 | ||
5149 | s = 0; | |
5150 | sub = env->subprog_info; | |
5151 | ulinfo = u64_to_user_ptr(attr->line_info); | |
5152 | expected_size = sizeof(struct bpf_line_info); | |
5153 | ncopy = min_t(u32, expected_size, rec_size); | |
5154 | for (i = 0; i < nr_linfo; i++) { | |
5155 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
5156 | if (err) { | |
5157 | if (err == -E2BIG) { | |
5158 | verbose(env, "nonzero tailing record in line_info"); | |
5159 | if (put_user(expected_size, | |
5160 | &uattr->line_info_rec_size)) | |
5161 | err = -EFAULT; | |
5162 | } | |
5163 | goto err_free; | |
5164 | } | |
5165 | ||
5166 | if (copy_from_user(&linfo[i], ulinfo, ncopy)) { | |
5167 | err = -EFAULT; | |
5168 | goto err_free; | |
5169 | } | |
5170 | ||
5171 | /* | |
5172 | * Check insn_off to ensure | |
5173 | * 1) strictly increasing AND | |
5174 | * 2) bounded by prog->len | |
5175 | * | |
5176 | * The linfo[0].insn_off == 0 check logically falls into | |
5177 | * the later "missing bpf_line_info for func..." case | |
5178 | * because the first linfo[0].insn_off must be the | |
5179 | * first sub also and the first sub must have | |
5180 | * subprog_info[0].start == 0. | |
5181 | */ | |
5182 | if ((i && linfo[i].insn_off <= prev_offset) || | |
5183 | linfo[i].insn_off >= prog->len) { | |
5184 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
5185 | i, linfo[i].insn_off, prev_offset, | |
5186 | prog->len); | |
5187 | err = -EINVAL; | |
5188 | goto err_free; | |
5189 | } | |
5190 | ||
fdbaa0be MKL |
5191 | if (!prog->insnsi[linfo[i].insn_off].code) { |
5192 | verbose(env, | |
5193 | "Invalid insn code at line_info[%u].insn_off\n", | |
5194 | i); | |
5195 | err = -EINVAL; | |
5196 | goto err_free; | |
5197 | } | |
5198 | ||
23127b33 MKL |
5199 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
5200 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
5201 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
5202 | err = -EINVAL; | |
5203 | goto err_free; | |
5204 | } | |
5205 | ||
5206 | if (s != env->subprog_cnt) { | |
5207 | if (linfo[i].insn_off == sub[s].start) { | |
5208 | sub[s].linfo_idx = i; | |
5209 | s++; | |
5210 | } else if (sub[s].start < linfo[i].insn_off) { | |
5211 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
5212 | err = -EINVAL; | |
5213 | goto err_free; | |
5214 | } | |
5215 | } | |
5216 | ||
5217 | prev_offset = linfo[i].insn_off; | |
5218 | ulinfo += rec_size; | |
5219 | } | |
5220 | ||
5221 | if (s != env->subprog_cnt) { | |
5222 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
5223 | env->subprog_cnt - s, s); | |
5224 | err = -EINVAL; | |
5225 | goto err_free; | |
5226 | } | |
5227 | ||
5228 | prog->aux->linfo = linfo; | |
5229 | prog->aux->nr_linfo = nr_linfo; | |
5230 | ||
5231 | return 0; | |
5232 | ||
5233 | err_free: | |
5234 | kvfree(linfo); | |
5235 | return err; | |
5236 | } | |
5237 | ||
5238 | static int check_btf_info(struct bpf_verifier_env *env, | |
5239 | const union bpf_attr *attr, | |
5240 | union bpf_attr __user *uattr) | |
5241 | { | |
5242 | struct btf *btf; | |
5243 | int err; | |
5244 | ||
5245 | if (!attr->func_info_cnt && !attr->line_info_cnt) | |
5246 | return 0; | |
5247 | ||
5248 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
5249 | if (IS_ERR(btf)) | |
5250 | return PTR_ERR(btf); | |
5251 | env->prog->aux->btf = btf; | |
5252 | ||
5253 | err = check_btf_func(env, attr, uattr); | |
5254 | if (err) | |
5255 | return err; | |
5256 | ||
5257 | err = check_btf_line(env, attr, uattr); | |
5258 | if (err) | |
5259 | return err; | |
5260 | ||
5261 | return 0; | |
ba64e7d8 YS |
5262 | } |
5263 | ||
f1174f77 EC |
5264 | /* check %cur's range satisfies %old's */ |
5265 | static bool range_within(struct bpf_reg_state *old, | |
5266 | struct bpf_reg_state *cur) | |
5267 | { | |
b03c9f9f EC |
5268 | return old->umin_value <= cur->umin_value && |
5269 | old->umax_value >= cur->umax_value && | |
5270 | old->smin_value <= cur->smin_value && | |
5271 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
5272 | } |
5273 | ||
5274 | /* Maximum number of register states that can exist at once */ | |
5275 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
5276 | struct idpair { | |
5277 | u32 old; | |
5278 | u32 cur; | |
5279 | }; | |
5280 | ||
5281 | /* If in the old state two registers had the same id, then they need to have | |
5282 | * the same id in the new state as well. But that id could be different from | |
5283 | * the old state, so we need to track the mapping from old to new ids. | |
5284 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
5285 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
5286 | * regs with a different old id could still have new id 9, we don't care about | |
5287 | * that. | |
5288 | * So we look through our idmap to see if this old id has been seen before. If | |
5289 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 5290 | */ |
f1174f77 | 5291 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 5292 | { |
f1174f77 | 5293 | unsigned int i; |
969bf05e | 5294 | |
f1174f77 EC |
5295 | for (i = 0; i < ID_MAP_SIZE; i++) { |
5296 | if (!idmap[i].old) { | |
5297 | /* Reached an empty slot; haven't seen this id before */ | |
5298 | idmap[i].old = old_id; | |
5299 | idmap[i].cur = cur_id; | |
5300 | return true; | |
5301 | } | |
5302 | if (idmap[i].old == old_id) | |
5303 | return idmap[i].cur == cur_id; | |
5304 | } | |
5305 | /* We ran out of idmap slots, which should be impossible */ | |
5306 | WARN_ON_ONCE(1); | |
5307 | return false; | |
5308 | } | |
5309 | ||
9242b5f5 AS |
5310 | static void clean_func_state(struct bpf_verifier_env *env, |
5311 | struct bpf_func_state *st) | |
5312 | { | |
5313 | enum bpf_reg_liveness live; | |
5314 | int i, j; | |
5315 | ||
5316 | for (i = 0; i < BPF_REG_FP; i++) { | |
5317 | live = st->regs[i].live; | |
5318 | /* liveness must not touch this register anymore */ | |
5319 | st->regs[i].live |= REG_LIVE_DONE; | |
5320 | if (!(live & REG_LIVE_READ)) | |
5321 | /* since the register is unused, clear its state | |
5322 | * to make further comparison simpler | |
5323 | */ | |
5324 | __mark_reg_not_init(&st->regs[i]); | |
5325 | } | |
5326 | ||
5327 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
5328 | live = st->stack[i].spilled_ptr.live; | |
5329 | /* liveness must not touch this stack slot anymore */ | |
5330 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
5331 | if (!(live & REG_LIVE_READ)) { | |
5332 | __mark_reg_not_init(&st->stack[i].spilled_ptr); | |
5333 | for (j = 0; j < BPF_REG_SIZE; j++) | |
5334 | st->stack[i].slot_type[j] = STACK_INVALID; | |
5335 | } | |
5336 | } | |
5337 | } | |
5338 | ||
5339 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
5340 | struct bpf_verifier_state *st) | |
5341 | { | |
5342 | int i; | |
5343 | ||
5344 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
5345 | /* all regs in this state in all frames were already marked */ | |
5346 | return; | |
5347 | ||
5348 | for (i = 0; i <= st->curframe; i++) | |
5349 | clean_func_state(env, st->frame[i]); | |
5350 | } | |
5351 | ||
5352 | /* the parentage chains form a tree. | |
5353 | * the verifier states are added to state lists at given insn and | |
5354 | * pushed into state stack for future exploration. | |
5355 | * when the verifier reaches bpf_exit insn some of the verifer states | |
5356 | * stored in the state lists have their final liveness state already, | |
5357 | * but a lot of states will get revised from liveness point of view when | |
5358 | * the verifier explores other branches. | |
5359 | * Example: | |
5360 | * 1: r0 = 1 | |
5361 | * 2: if r1 == 100 goto pc+1 | |
5362 | * 3: r0 = 2 | |
5363 | * 4: exit | |
5364 | * when the verifier reaches exit insn the register r0 in the state list of | |
5365 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
5366 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
5367 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
5368 | * | |
5369 | * Since the verifier pushes the branch states as it sees them while exploring | |
5370 | * the program the condition of walking the branch instruction for the second | |
5371 | * time means that all states below this branch were already explored and | |
5372 | * their final liveness markes are already propagated. | |
5373 | * Hence when the verifier completes the search of state list in is_state_visited() | |
5374 | * we can call this clean_live_states() function to mark all liveness states | |
5375 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
5376 | * will not be used. | |
5377 | * This function also clears the registers and stack for states that !READ | |
5378 | * to simplify state merging. | |
5379 | * | |
5380 | * Important note here that walking the same branch instruction in the callee | |
5381 | * doesn't meant that the states are DONE. The verifier has to compare | |
5382 | * the callsites | |
5383 | */ | |
5384 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
5385 | struct bpf_verifier_state *cur) | |
5386 | { | |
5387 | struct bpf_verifier_state_list *sl; | |
5388 | int i; | |
5389 | ||
5390 | sl = env->explored_states[insn]; | |
5391 | if (!sl) | |
5392 | return; | |
5393 | ||
5394 | while (sl != STATE_LIST_MARK) { | |
5395 | if (sl->state.curframe != cur->curframe) | |
5396 | goto next; | |
5397 | for (i = 0; i <= cur->curframe; i++) | |
5398 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
5399 | goto next; | |
5400 | clean_verifier_state(env, &sl->state); | |
5401 | next: | |
5402 | sl = sl->next; | |
5403 | } | |
5404 | } | |
5405 | ||
f1174f77 | 5406 | /* Returns true if (rold safe implies rcur safe) */ |
1b688a19 EC |
5407 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
5408 | struct idpair *idmap) | |
f1174f77 | 5409 | { |
f4d7e40a AS |
5410 | bool equal; |
5411 | ||
dc503a8a EC |
5412 | if (!(rold->live & REG_LIVE_READ)) |
5413 | /* explored state didn't use this */ | |
5414 | return true; | |
5415 | ||
679c782d | 5416 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
5417 | |
5418 | if (rold->type == PTR_TO_STACK) | |
5419 | /* two stack pointers are equal only if they're pointing to | |
5420 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
5421 | */ | |
5422 | return equal && rold->frameno == rcur->frameno; | |
5423 | ||
5424 | if (equal) | |
969bf05e AS |
5425 | return true; |
5426 | ||
f1174f77 EC |
5427 | if (rold->type == NOT_INIT) |
5428 | /* explored state can't have used this */ | |
969bf05e | 5429 | return true; |
f1174f77 EC |
5430 | if (rcur->type == NOT_INIT) |
5431 | return false; | |
5432 | switch (rold->type) { | |
5433 | case SCALAR_VALUE: | |
5434 | if (rcur->type == SCALAR_VALUE) { | |
5435 | /* new val must satisfy old val knowledge */ | |
5436 | return range_within(rold, rcur) && | |
5437 | tnum_in(rold->var_off, rcur->var_off); | |
5438 | } else { | |
179d1c56 JH |
5439 | /* We're trying to use a pointer in place of a scalar. |
5440 | * Even if the scalar was unbounded, this could lead to | |
5441 | * pointer leaks because scalars are allowed to leak | |
5442 | * while pointers are not. We could make this safe in | |
5443 | * special cases if root is calling us, but it's | |
5444 | * probably not worth the hassle. | |
f1174f77 | 5445 | */ |
179d1c56 | 5446 | return false; |
f1174f77 EC |
5447 | } |
5448 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
5449 | /* If the new min/max/var_off satisfy the old ones and |
5450 | * everything else matches, we are OK. | |
5451 | * We don't care about the 'id' value, because nothing | |
5452 | * uses it for PTR_TO_MAP_VALUE (only for ..._OR_NULL) | |
5453 | */ | |
5454 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
5455 | range_within(rold, rcur) && | |
5456 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
5457 | case PTR_TO_MAP_VALUE_OR_NULL: |
5458 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
5459 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
5460 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
5461 | * checked, doing so could have affected others with the same | |
5462 | * id, and we can't check for that because we lost the id when | |
5463 | * we converted to a PTR_TO_MAP_VALUE. | |
5464 | */ | |
5465 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
5466 | return false; | |
5467 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
5468 | return false; | |
5469 | /* Check our ids match any regs they're supposed to */ | |
5470 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 5471 | case PTR_TO_PACKET_META: |
f1174f77 | 5472 | case PTR_TO_PACKET: |
de8f3a83 | 5473 | if (rcur->type != rold->type) |
f1174f77 EC |
5474 | return false; |
5475 | /* We must have at least as much range as the old ptr | |
5476 | * did, so that any accesses which were safe before are | |
5477 | * still safe. This is true even if old range < old off, | |
5478 | * since someone could have accessed through (ptr - k), or | |
5479 | * even done ptr -= k in a register, to get a safe access. | |
5480 | */ | |
5481 | if (rold->range > rcur->range) | |
5482 | return false; | |
5483 | /* If the offsets don't match, we can't trust our alignment; | |
5484 | * nor can we be sure that we won't fall out of range. | |
5485 | */ | |
5486 | if (rold->off != rcur->off) | |
5487 | return false; | |
5488 | /* id relations must be preserved */ | |
5489 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
5490 | return false; | |
5491 | /* new val must satisfy old val knowledge */ | |
5492 | return range_within(rold, rcur) && | |
5493 | tnum_in(rold->var_off, rcur->var_off); | |
5494 | case PTR_TO_CTX: | |
5495 | case CONST_PTR_TO_MAP: | |
f1174f77 | 5496 | case PTR_TO_PACKET_END: |
d58e468b | 5497 | case PTR_TO_FLOW_KEYS: |
c64b7983 JS |
5498 | case PTR_TO_SOCKET: |
5499 | case PTR_TO_SOCKET_OR_NULL: | |
f1174f77 EC |
5500 | /* Only valid matches are exact, which memcmp() above |
5501 | * would have accepted | |
5502 | */ | |
5503 | default: | |
5504 | /* Don't know what's going on, just say it's not safe */ | |
5505 | return false; | |
5506 | } | |
969bf05e | 5507 | |
f1174f77 EC |
5508 | /* Shouldn't get here; if we do, say it's not safe */ |
5509 | WARN_ON_ONCE(1); | |
969bf05e AS |
5510 | return false; |
5511 | } | |
5512 | ||
f4d7e40a AS |
5513 | static bool stacksafe(struct bpf_func_state *old, |
5514 | struct bpf_func_state *cur, | |
638f5b90 AS |
5515 | struct idpair *idmap) |
5516 | { | |
5517 | int i, spi; | |
5518 | ||
638f5b90 AS |
5519 | /* walk slots of the explored stack and ignore any additional |
5520 | * slots in the current stack, since explored(safe) state | |
5521 | * didn't use them | |
5522 | */ | |
5523 | for (i = 0; i < old->allocated_stack; i++) { | |
5524 | spi = i / BPF_REG_SIZE; | |
5525 | ||
b233920c AS |
5526 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
5527 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 5528 | /* explored state didn't use this */ |
fd05e57b | 5529 | continue; |
b233920c | 5530 | } |
cc2b14d5 | 5531 | |
638f5b90 AS |
5532 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
5533 | continue; | |
19e2dbb7 AS |
5534 | |
5535 | /* explored stack has more populated slots than current stack | |
5536 | * and these slots were used | |
5537 | */ | |
5538 | if (i >= cur->allocated_stack) | |
5539 | return false; | |
5540 | ||
cc2b14d5 AS |
5541 | /* if old state was safe with misc data in the stack |
5542 | * it will be safe with zero-initialized stack. | |
5543 | * The opposite is not true | |
5544 | */ | |
5545 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
5546 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
5547 | continue; | |
638f5b90 AS |
5548 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
5549 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
5550 | /* Ex: old explored (safe) state has STACK_SPILL in | |
5551 | * this stack slot, but current has has STACK_MISC -> | |
5552 | * this verifier states are not equivalent, | |
5553 | * return false to continue verification of this path | |
5554 | */ | |
5555 | return false; | |
5556 | if (i % BPF_REG_SIZE) | |
5557 | continue; | |
5558 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
5559 | continue; | |
5560 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
5561 | &cur->stack[spi].spilled_ptr, | |
5562 | idmap)) | |
5563 | /* when explored and current stack slot are both storing | |
5564 | * spilled registers, check that stored pointers types | |
5565 | * are the same as well. | |
5566 | * Ex: explored safe path could have stored | |
5567 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
5568 | * but current path has stored: | |
5569 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
5570 | * such verifier states are not equivalent. | |
5571 | * return false to continue verification of this path | |
5572 | */ | |
5573 | return false; | |
5574 | } | |
5575 | return true; | |
5576 | } | |
5577 | ||
fd978bf7 JS |
5578 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
5579 | { | |
5580 | if (old->acquired_refs != cur->acquired_refs) | |
5581 | return false; | |
5582 | return !memcmp(old->refs, cur->refs, | |
5583 | sizeof(*old->refs) * old->acquired_refs); | |
5584 | } | |
5585 | ||
f1bca824 AS |
5586 | /* compare two verifier states |
5587 | * | |
5588 | * all states stored in state_list are known to be valid, since | |
5589 | * verifier reached 'bpf_exit' instruction through them | |
5590 | * | |
5591 | * this function is called when verifier exploring different branches of | |
5592 | * execution popped from the state stack. If it sees an old state that has | |
5593 | * more strict register state and more strict stack state then this execution | |
5594 | * branch doesn't need to be explored further, since verifier already | |
5595 | * concluded that more strict state leads to valid finish. | |
5596 | * | |
5597 | * Therefore two states are equivalent if register state is more conservative | |
5598 | * and explored stack state is more conservative than the current one. | |
5599 | * Example: | |
5600 | * explored current | |
5601 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
5602 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
5603 | * | |
5604 | * In other words if current stack state (one being explored) has more | |
5605 | * valid slots than old one that already passed validation, it means | |
5606 | * the verifier can stop exploring and conclude that current state is valid too | |
5607 | * | |
5608 | * Similarly with registers. If explored state has register type as invalid | |
5609 | * whereas register type in current state is meaningful, it means that | |
5610 | * the current state will reach 'bpf_exit' instruction safely | |
5611 | */ | |
f4d7e40a AS |
5612 | static bool func_states_equal(struct bpf_func_state *old, |
5613 | struct bpf_func_state *cur) | |
f1bca824 | 5614 | { |
f1174f77 EC |
5615 | struct idpair *idmap; |
5616 | bool ret = false; | |
f1bca824 AS |
5617 | int i; |
5618 | ||
f1174f77 EC |
5619 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
5620 | /* If we failed to allocate the idmap, just say it's not safe */ | |
5621 | if (!idmap) | |
1a0dc1ac | 5622 | return false; |
f1174f77 EC |
5623 | |
5624 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 5625 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 5626 | goto out_free; |
f1bca824 AS |
5627 | } |
5628 | ||
638f5b90 AS |
5629 | if (!stacksafe(old, cur, idmap)) |
5630 | goto out_free; | |
fd978bf7 JS |
5631 | |
5632 | if (!refsafe(old, cur)) | |
5633 | goto out_free; | |
f1174f77 EC |
5634 | ret = true; |
5635 | out_free: | |
5636 | kfree(idmap); | |
5637 | return ret; | |
f1bca824 AS |
5638 | } |
5639 | ||
f4d7e40a AS |
5640 | static bool states_equal(struct bpf_verifier_env *env, |
5641 | struct bpf_verifier_state *old, | |
5642 | struct bpf_verifier_state *cur) | |
5643 | { | |
5644 | int i; | |
5645 | ||
5646 | if (old->curframe != cur->curframe) | |
5647 | return false; | |
5648 | ||
979d63d5 DB |
5649 | /* Verification state from speculative execution simulation |
5650 | * must never prune a non-speculative execution one. | |
5651 | */ | |
5652 | if (old->speculative && !cur->speculative) | |
5653 | return false; | |
5654 | ||
f4d7e40a AS |
5655 | /* for states to be equal callsites have to be the same |
5656 | * and all frame states need to be equivalent | |
5657 | */ | |
5658 | for (i = 0; i <= old->curframe; i++) { | |
5659 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
5660 | return false; | |
5661 | if (!func_states_equal(old->frame[i], cur->frame[i])) | |
5662 | return false; | |
5663 | } | |
5664 | return true; | |
5665 | } | |
5666 | ||
8e9cd9ce | 5667 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
5668 | * straight-line code between a state and its parent. When we arrive at an |
5669 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
5670 | * code, so read marks in the state must propagate to the parent regardless | |
5671 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 5672 | * in mark_reg_read() is for. |
8e9cd9ce | 5673 | */ |
f4d7e40a AS |
5674 | static int propagate_liveness(struct bpf_verifier_env *env, |
5675 | const struct bpf_verifier_state *vstate, | |
5676 | struct bpf_verifier_state *vparent) | |
dc503a8a | 5677 | { |
f4d7e40a AS |
5678 | int i, frame, err = 0; |
5679 | struct bpf_func_state *state, *parent; | |
dc503a8a | 5680 | |
f4d7e40a AS |
5681 | if (vparent->curframe != vstate->curframe) { |
5682 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
5683 | vparent->curframe, vstate->curframe); | |
5684 | return -EFAULT; | |
5685 | } | |
dc503a8a EC |
5686 | /* Propagate read liveness of registers... */ |
5687 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
5688 | /* We don't need to worry about FP liveness because it's read-only */ | |
5689 | for (i = 0; i < BPF_REG_FP; i++) { | |
f4d7e40a | 5690 | if (vparent->frame[vparent->curframe]->regs[i].live & REG_LIVE_READ) |
63f45f84 | 5691 | continue; |
f4d7e40a | 5692 | if (vstate->frame[vstate->curframe]->regs[i].live & REG_LIVE_READ) { |
679c782d EC |
5693 | err = mark_reg_read(env, &vstate->frame[vstate->curframe]->regs[i], |
5694 | &vparent->frame[vstate->curframe]->regs[i]); | |
f4d7e40a AS |
5695 | if (err) |
5696 | return err; | |
dc503a8a EC |
5697 | } |
5698 | } | |
f4d7e40a | 5699 | |
dc503a8a | 5700 | /* ... and stack slots */ |
f4d7e40a AS |
5701 | for (frame = 0; frame <= vstate->curframe; frame++) { |
5702 | state = vstate->frame[frame]; | |
5703 | parent = vparent->frame[frame]; | |
5704 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && | |
5705 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
f4d7e40a AS |
5706 | if (parent->stack[i].spilled_ptr.live & REG_LIVE_READ) |
5707 | continue; | |
5708 | if (state->stack[i].spilled_ptr.live & REG_LIVE_READ) | |
679c782d EC |
5709 | mark_reg_read(env, &state->stack[i].spilled_ptr, |
5710 | &parent->stack[i].spilled_ptr); | |
dc503a8a EC |
5711 | } |
5712 | } | |
f4d7e40a | 5713 | return err; |
dc503a8a EC |
5714 | } |
5715 | ||
58e2af8b | 5716 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 5717 | { |
58e2af8b JK |
5718 | struct bpf_verifier_state_list *new_sl; |
5719 | struct bpf_verifier_state_list *sl; | |
679c782d | 5720 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 5721 | int i, j, err, states_cnt = 0; |
f1bca824 AS |
5722 | |
5723 | sl = env->explored_states[insn_idx]; | |
5724 | if (!sl) | |
5725 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
5726 | * be doing state search here | |
5727 | */ | |
5728 | return 0; | |
5729 | ||
9242b5f5 AS |
5730 | clean_live_states(env, insn_idx, cur); |
5731 | ||
f1bca824 | 5732 | while (sl != STATE_LIST_MARK) { |
638f5b90 | 5733 | if (states_equal(env, &sl->state, cur)) { |
f1bca824 | 5734 | /* reached equivalent register/stack state, |
dc503a8a EC |
5735 | * prune the search. |
5736 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
5737 | * If we have any write marks in env->cur_state, they |
5738 | * will prevent corresponding reads in the continuation | |
5739 | * from reaching our parent (an explored_state). Our | |
5740 | * own state will get the read marks recorded, but | |
5741 | * they'll be immediately forgotten as we're pruning | |
5742 | * this state and will pop a new one. | |
f1bca824 | 5743 | */ |
f4d7e40a AS |
5744 | err = propagate_liveness(env, &sl->state, cur); |
5745 | if (err) | |
5746 | return err; | |
f1bca824 | 5747 | return 1; |
dc503a8a | 5748 | } |
f1bca824 | 5749 | sl = sl->next; |
ceefbc96 | 5750 | states_cnt++; |
f1bca824 AS |
5751 | } |
5752 | ||
ceefbc96 AS |
5753 | if (!env->allow_ptr_leaks && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
5754 | return 0; | |
5755 | ||
f1bca824 AS |
5756 | /* there were no equivalent states, remember current one. |
5757 | * technically the current state is not proven to be safe yet, | |
f4d7e40a AS |
5758 | * but it will either reach outer most bpf_exit (which means it's safe) |
5759 | * or it will be rejected. Since there are no loops, we won't be | |
5760 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) | |
5761 | * again on the way to bpf_exit | |
f1bca824 | 5762 | */ |
638f5b90 | 5763 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
5764 | if (!new_sl) |
5765 | return -ENOMEM; | |
5766 | ||
5767 | /* add new state to the head of linked list */ | |
679c782d EC |
5768 | new = &new_sl->state; |
5769 | err = copy_verifier_state(new, cur); | |
1969db47 | 5770 | if (err) { |
679c782d | 5771 | free_verifier_state(new, false); |
1969db47 AS |
5772 | kfree(new_sl); |
5773 | return err; | |
5774 | } | |
f1bca824 AS |
5775 | new_sl->next = env->explored_states[insn_idx]; |
5776 | env->explored_states[insn_idx] = new_sl; | |
7640ead9 JK |
5777 | /* connect new state to parentage chain. Current frame needs all |
5778 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
5779 | * to the stack implicitly by JITs) so in callers' frames connect just | |
5780 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
5781 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
5782 | * from callee with its full parentage chain, anyway. | |
5783 | */ | |
5784 | for (j = 0; j <= cur->curframe; j++) | |
5785 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
5786 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
8e9cd9ce EC |
5787 | /* clear write marks in current state: the writes we did are not writes |
5788 | * our child did, so they don't screen off its reads from us. | |
5789 | * (There are no read marks in current state, because reads always mark | |
5790 | * their parent and current state never has children yet. Only | |
5791 | * explored_states can get read marks.) | |
5792 | */ | |
dc503a8a | 5793 | for (i = 0; i < BPF_REG_FP; i++) |
f4d7e40a AS |
5794 | cur->frame[cur->curframe]->regs[i].live = REG_LIVE_NONE; |
5795 | ||
5796 | /* all stack frames are accessible from callee, clear them all */ | |
5797 | for (j = 0; j <= cur->curframe; j++) { | |
5798 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 5799 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 5800 | |
679c782d | 5801 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 5802 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
5803 | frame->stack[i].spilled_ptr.parent = |
5804 | &newframe->stack[i].spilled_ptr; | |
5805 | } | |
f4d7e40a | 5806 | } |
f1bca824 AS |
5807 | return 0; |
5808 | } | |
5809 | ||
c64b7983 JS |
5810 | /* Return true if it's OK to have the same insn return a different type. */ |
5811 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
5812 | { | |
5813 | switch (type) { | |
5814 | case PTR_TO_CTX: | |
5815 | case PTR_TO_SOCKET: | |
5816 | case PTR_TO_SOCKET_OR_NULL: | |
5817 | return false; | |
5818 | default: | |
5819 | return true; | |
5820 | } | |
5821 | } | |
5822 | ||
5823 | /* If an instruction was previously used with particular pointer types, then we | |
5824 | * need to be careful to avoid cases such as the below, where it may be ok | |
5825 | * for one branch accessing the pointer, but not ok for the other branch: | |
5826 | * | |
5827 | * R1 = sock_ptr | |
5828 | * goto X; | |
5829 | * ... | |
5830 | * R1 = some_other_valid_ptr; | |
5831 | * goto X; | |
5832 | * ... | |
5833 | * R2 = *(u32 *)(R1 + 0); | |
5834 | */ | |
5835 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
5836 | { | |
5837 | return src != prev && (!reg_type_mismatch_ok(src) || | |
5838 | !reg_type_mismatch_ok(prev)); | |
5839 | } | |
5840 | ||
58e2af8b | 5841 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 5842 | { |
638f5b90 | 5843 | struct bpf_verifier_state *state; |
17a52670 | 5844 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 5845 | struct bpf_reg_state *regs; |
f4d7e40a | 5846 | int insn_cnt = env->prog->len, i; |
17a52670 AS |
5847 | int insn_processed = 0; |
5848 | bool do_print_state = false; | |
5849 | ||
d9762e84 MKL |
5850 | env->prev_linfo = NULL; |
5851 | ||
638f5b90 AS |
5852 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
5853 | if (!state) | |
5854 | return -ENOMEM; | |
f4d7e40a | 5855 | state->curframe = 0; |
979d63d5 | 5856 | state->speculative = false; |
f4d7e40a AS |
5857 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); |
5858 | if (!state->frame[0]) { | |
5859 | kfree(state); | |
5860 | return -ENOMEM; | |
5861 | } | |
5862 | env->cur_state = state; | |
5863 | init_func_state(env, state->frame[0], | |
5864 | BPF_MAIN_FUNC /* callsite */, | |
5865 | 0 /* frameno */, | |
5866 | 0 /* subprogno, zero == main subprog */); | |
c08435ec | 5867 | |
17a52670 AS |
5868 | for (;;) { |
5869 | struct bpf_insn *insn; | |
5870 | u8 class; | |
5871 | int err; | |
5872 | ||
c08435ec | 5873 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 5874 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 5875 | env->insn_idx, insn_cnt); |
17a52670 AS |
5876 | return -EFAULT; |
5877 | } | |
5878 | ||
c08435ec | 5879 | insn = &insns[env->insn_idx]; |
17a52670 AS |
5880 | class = BPF_CLASS(insn->code); |
5881 | ||
07016151 | 5882 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
5883 | verbose(env, |
5884 | "BPF program is too large. Processed %d insn\n", | |
17a52670 AS |
5885 | insn_processed); |
5886 | return -E2BIG; | |
5887 | } | |
5888 | ||
c08435ec | 5889 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
5890 | if (err < 0) |
5891 | return err; | |
5892 | if (err == 1) { | |
5893 | /* found equivalent state, can prune the search */ | |
61bd5218 | 5894 | if (env->log.level) { |
f1bca824 | 5895 | if (do_print_state) |
979d63d5 DB |
5896 | verbose(env, "\nfrom %d to %d%s: safe\n", |
5897 | env->prev_insn_idx, env->insn_idx, | |
5898 | env->cur_state->speculative ? | |
5899 | " (speculative execution)" : ""); | |
f1bca824 | 5900 | else |
c08435ec | 5901 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
5902 | } |
5903 | goto process_bpf_exit; | |
5904 | } | |
5905 | ||
c3494801 AS |
5906 | if (signal_pending(current)) |
5907 | return -EAGAIN; | |
5908 | ||
3c2ce60b DB |
5909 | if (need_resched()) |
5910 | cond_resched(); | |
5911 | ||
61bd5218 JK |
5912 | if (env->log.level > 1 || (env->log.level && do_print_state)) { |
5913 | if (env->log.level > 1) | |
c08435ec | 5914 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 5915 | else |
979d63d5 DB |
5916 | verbose(env, "\nfrom %d to %d%s:", |
5917 | env->prev_insn_idx, env->insn_idx, | |
5918 | env->cur_state->speculative ? | |
5919 | " (speculative execution)" : ""); | |
f4d7e40a | 5920 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
5921 | do_print_state = false; |
5922 | } | |
5923 | ||
61bd5218 | 5924 | if (env->log.level) { |
7105e828 DB |
5925 | const struct bpf_insn_cbs cbs = { |
5926 | .cb_print = verbose, | |
abe08840 | 5927 | .private_data = env, |
7105e828 DB |
5928 | }; |
5929 | ||
c08435ec DB |
5930 | verbose_linfo(env, env->insn_idx, "; "); |
5931 | verbose(env, "%d: ", env->insn_idx); | |
abe08840 | 5932 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
5933 | } |
5934 | ||
cae1927c | 5935 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
5936 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
5937 | env->prev_insn_idx); | |
cae1927c JK |
5938 | if (err) |
5939 | return err; | |
5940 | } | |
13a27dfc | 5941 | |
638f5b90 | 5942 | regs = cur_regs(env); |
c08435ec | 5943 | env->insn_aux_data[env->insn_idx].seen = true; |
fd978bf7 | 5944 | |
17a52670 | 5945 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 5946 | err = check_alu_op(env, insn); |
17a52670 AS |
5947 | if (err) |
5948 | return err; | |
5949 | ||
5950 | } else if (class == BPF_LDX) { | |
3df126f3 | 5951 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
5952 | |
5953 | /* check for reserved fields is already done */ | |
5954 | ||
17a52670 | 5955 | /* check src operand */ |
dc503a8a | 5956 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
5957 | if (err) |
5958 | return err; | |
5959 | ||
dc503a8a | 5960 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
5961 | if (err) |
5962 | return err; | |
5963 | ||
725f9dcd AS |
5964 | src_reg_type = regs[insn->src_reg].type; |
5965 | ||
17a52670 AS |
5966 | /* check that memory (src_reg + off) is readable, |
5967 | * the state of dst_reg will be updated by this func | |
5968 | */ | |
c08435ec DB |
5969 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
5970 | insn->off, BPF_SIZE(insn->code), | |
5971 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
5972 | if (err) |
5973 | return err; | |
5974 | ||
c08435ec | 5975 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
5976 | |
5977 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
5978 | /* saw a valid insn |
5979 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 5980 | * save type to validate intersecting paths |
9bac3d6d | 5981 | */ |
3df126f3 | 5982 | *prev_src_type = src_reg_type; |
9bac3d6d | 5983 | |
c64b7983 | 5984 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
5985 | /* ABuser program is trying to use the same insn |
5986 | * dst_reg = *(u32*) (src_reg + off) | |
5987 | * with different pointer types: | |
5988 | * src_reg == ctx in one branch and | |
5989 | * src_reg == stack|map in some other branch. | |
5990 | * Reject it. | |
5991 | */ | |
61bd5218 | 5992 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
5993 | return -EINVAL; |
5994 | } | |
5995 | ||
17a52670 | 5996 | } else if (class == BPF_STX) { |
3df126f3 | 5997 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 5998 | |
17a52670 | 5999 | if (BPF_MODE(insn->code) == BPF_XADD) { |
c08435ec | 6000 | err = check_xadd(env, env->insn_idx, insn); |
17a52670 AS |
6001 | if (err) |
6002 | return err; | |
c08435ec | 6003 | env->insn_idx++; |
17a52670 AS |
6004 | continue; |
6005 | } | |
6006 | ||
17a52670 | 6007 | /* check src1 operand */ |
dc503a8a | 6008 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6009 | if (err) |
6010 | return err; | |
6011 | /* check src2 operand */ | |
dc503a8a | 6012 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6013 | if (err) |
6014 | return err; | |
6015 | ||
d691f9e8 AS |
6016 | dst_reg_type = regs[insn->dst_reg].type; |
6017 | ||
17a52670 | 6018 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
6019 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
6020 | insn->off, BPF_SIZE(insn->code), | |
6021 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
6022 | if (err) |
6023 | return err; | |
6024 | ||
c08435ec | 6025 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
6026 | |
6027 | if (*prev_dst_type == NOT_INIT) { | |
6028 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 6029 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 6030 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
6031 | return -EINVAL; |
6032 | } | |
6033 | ||
17a52670 AS |
6034 | } else if (class == BPF_ST) { |
6035 | if (BPF_MODE(insn->code) != BPF_MEM || | |
6036 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 6037 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
6038 | return -EINVAL; |
6039 | } | |
6040 | /* check src operand */ | |
dc503a8a | 6041 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6042 | if (err) |
6043 | return err; | |
6044 | ||
f37a8cb8 | 6045 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 6046 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f DB |
6047 | insn->dst_reg, |
6048 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
6049 | return -EACCES; |
6050 | } | |
6051 | ||
17a52670 | 6052 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
6053 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
6054 | insn->off, BPF_SIZE(insn->code), | |
6055 | BPF_WRITE, -1, false); | |
17a52670 AS |
6056 | if (err) |
6057 | return err; | |
6058 | ||
6059 | } else if (class == BPF_JMP) { | |
6060 | u8 opcode = BPF_OP(insn->code); | |
6061 | ||
6062 | if (opcode == BPF_CALL) { | |
6063 | if (BPF_SRC(insn->code) != BPF_K || | |
6064 | insn->off != 0 || | |
f4d7e40a AS |
6065 | (insn->src_reg != BPF_REG_0 && |
6066 | insn->src_reg != BPF_PSEUDO_CALL) || | |
17a52670 | 6067 | insn->dst_reg != BPF_REG_0) { |
61bd5218 | 6068 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
6069 | return -EINVAL; |
6070 | } | |
6071 | ||
f4d7e40a | 6072 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 6073 | err = check_func_call(env, insn, &env->insn_idx); |
f4d7e40a | 6074 | else |
c08435ec | 6075 | err = check_helper_call(env, insn->imm, env->insn_idx); |
17a52670 AS |
6076 | if (err) |
6077 | return err; | |
6078 | ||
6079 | } else if (opcode == BPF_JA) { | |
6080 | if (BPF_SRC(insn->code) != BPF_K || | |
6081 | insn->imm != 0 || | |
6082 | insn->src_reg != BPF_REG_0 || | |
6083 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 6084 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
6085 | return -EINVAL; |
6086 | } | |
6087 | ||
c08435ec | 6088 | env->insn_idx += insn->off + 1; |
17a52670 AS |
6089 | continue; |
6090 | ||
6091 | } else if (opcode == BPF_EXIT) { | |
6092 | if (BPF_SRC(insn->code) != BPF_K || | |
6093 | insn->imm != 0 || | |
6094 | insn->src_reg != BPF_REG_0 || | |
6095 | insn->dst_reg != BPF_REG_0) { | |
61bd5218 | 6096 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
6097 | return -EINVAL; |
6098 | } | |
6099 | ||
f4d7e40a AS |
6100 | if (state->curframe) { |
6101 | /* exit from nested function */ | |
c08435ec DB |
6102 | env->prev_insn_idx = env->insn_idx; |
6103 | err = prepare_func_exit(env, &env->insn_idx); | |
f4d7e40a AS |
6104 | if (err) |
6105 | return err; | |
6106 | do_print_state = true; | |
6107 | continue; | |
6108 | } | |
6109 | ||
fd978bf7 JS |
6110 | err = check_reference_leak(env); |
6111 | if (err) | |
6112 | return err; | |
6113 | ||
17a52670 AS |
6114 | /* eBPF calling convetion is such that R0 is used |
6115 | * to return the value from eBPF program. | |
6116 | * Make sure that it's readable at this time | |
6117 | * of bpf_exit, which means that program wrote | |
6118 | * something into it earlier | |
6119 | */ | |
dc503a8a | 6120 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
6121 | if (err) |
6122 | return err; | |
6123 | ||
1be7f75d | 6124 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 6125 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
6126 | return -EACCES; |
6127 | } | |
6128 | ||
390ee7e2 AS |
6129 | err = check_return_code(env); |
6130 | if (err) | |
6131 | return err; | |
f1bca824 | 6132 | process_bpf_exit: |
c08435ec DB |
6133 | err = pop_stack(env, &env->prev_insn_idx, |
6134 | &env->insn_idx); | |
638f5b90 AS |
6135 | if (err < 0) { |
6136 | if (err != -ENOENT) | |
6137 | return err; | |
17a52670 AS |
6138 | break; |
6139 | } else { | |
6140 | do_print_state = true; | |
6141 | continue; | |
6142 | } | |
6143 | } else { | |
c08435ec | 6144 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
6145 | if (err) |
6146 | return err; | |
6147 | } | |
6148 | } else if (class == BPF_LD) { | |
6149 | u8 mode = BPF_MODE(insn->code); | |
6150 | ||
6151 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
6152 | err = check_ld_abs(env, insn); |
6153 | if (err) | |
6154 | return err; | |
6155 | ||
17a52670 AS |
6156 | } else if (mode == BPF_IMM) { |
6157 | err = check_ld_imm(env, insn); | |
6158 | if (err) | |
6159 | return err; | |
6160 | ||
c08435ec DB |
6161 | env->insn_idx++; |
6162 | env->insn_aux_data[env->insn_idx].seen = true; | |
17a52670 | 6163 | } else { |
61bd5218 | 6164 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
6165 | return -EINVAL; |
6166 | } | |
6167 | } else { | |
61bd5218 | 6168 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
6169 | return -EINVAL; |
6170 | } | |
6171 | ||
c08435ec | 6172 | env->insn_idx++; |
17a52670 AS |
6173 | } |
6174 | ||
4bd95f4b DB |
6175 | verbose(env, "processed %d insns (limit %d), stack depth ", |
6176 | insn_processed, BPF_COMPLEXITY_LIMIT_INSNS); | |
f910cefa | 6177 | for (i = 0; i < env->subprog_cnt; i++) { |
9c8105bd | 6178 | u32 depth = env->subprog_info[i].stack_depth; |
f4d7e40a AS |
6179 | |
6180 | verbose(env, "%d", depth); | |
f910cefa | 6181 | if (i + 1 < env->subprog_cnt) |
f4d7e40a AS |
6182 | verbose(env, "+"); |
6183 | } | |
6184 | verbose(env, "\n"); | |
9c8105bd | 6185 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; |
17a52670 AS |
6186 | return 0; |
6187 | } | |
6188 | ||
56f668df MKL |
6189 | static int check_map_prealloc(struct bpf_map *map) |
6190 | { | |
6191 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
6192 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
6193 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
6194 | !(map->map_flags & BPF_F_NO_PREALLOC); |
6195 | } | |
6196 | ||
61bd5218 JK |
6197 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
6198 | struct bpf_map *map, | |
fdc15d38 AS |
6199 | struct bpf_prog *prog) |
6200 | ||
6201 | { | |
56f668df MKL |
6202 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
6203 | * preallocated hash maps, since doing memory allocation | |
6204 | * in overflow_handler can crash depending on where nmi got | |
6205 | * triggered. | |
6206 | */ | |
6207 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
6208 | if (!check_map_prealloc(map)) { | |
61bd5218 | 6209 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
6210 | return -EINVAL; |
6211 | } | |
6212 | if (map->inner_map_meta && | |
6213 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 6214 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
6215 | return -EINVAL; |
6216 | } | |
fdc15d38 | 6217 | } |
a3884572 JK |
6218 | |
6219 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && | |
09728266 | 6220 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
6221 | verbose(env, "offload device mismatch between prog and map\n"); |
6222 | return -EINVAL; | |
6223 | } | |
6224 | ||
fdc15d38 AS |
6225 | return 0; |
6226 | } | |
6227 | ||
b741f163 RG |
6228 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
6229 | { | |
6230 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
6231 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
6232 | } | |
6233 | ||
0246e64d AS |
6234 | /* look for pseudo eBPF instructions that access map FDs and |
6235 | * replace them with actual map pointers | |
6236 | */ | |
58e2af8b | 6237 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
6238 | { |
6239 | struct bpf_insn *insn = env->prog->insnsi; | |
6240 | int insn_cnt = env->prog->len; | |
fdc15d38 | 6241 | int i, j, err; |
0246e64d | 6242 | |
f1f7714e | 6243 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
6244 | if (err) |
6245 | return err; | |
6246 | ||
0246e64d | 6247 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 6248 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 6249 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 6250 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
6251 | return -EINVAL; |
6252 | } | |
6253 | ||
d691f9e8 AS |
6254 | if (BPF_CLASS(insn->code) == BPF_STX && |
6255 | ((BPF_MODE(insn->code) != BPF_MEM && | |
6256 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 6257 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
6258 | return -EINVAL; |
6259 | } | |
6260 | ||
0246e64d AS |
6261 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
6262 | struct bpf_map *map; | |
6263 | struct fd f; | |
6264 | ||
6265 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
6266 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
6267 | insn[1].off != 0) { | |
61bd5218 | 6268 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
6269 | return -EINVAL; |
6270 | } | |
6271 | ||
6272 | if (insn->src_reg == 0) | |
6273 | /* valid generic load 64-bit imm */ | |
6274 | goto next_insn; | |
6275 | ||
6276 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
61bd5218 JK |
6277 | verbose(env, |
6278 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
6279 | return -EINVAL; |
6280 | } | |
6281 | ||
6282 | f = fdget(insn->imm); | |
c2101297 | 6283 | map = __bpf_map_get(f); |
0246e64d | 6284 | if (IS_ERR(map)) { |
61bd5218 | 6285 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
0246e64d | 6286 | insn->imm); |
0246e64d AS |
6287 | return PTR_ERR(map); |
6288 | } | |
6289 | ||
61bd5218 | 6290 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
6291 | if (err) { |
6292 | fdput(f); | |
6293 | return err; | |
6294 | } | |
6295 | ||
0246e64d AS |
6296 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
6297 | insn[0].imm = (u32) (unsigned long) map; | |
6298 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
6299 | ||
6300 | /* check whether we recorded this map already */ | |
6301 | for (j = 0; j < env->used_map_cnt; j++) | |
6302 | if (env->used_maps[j] == map) { | |
6303 | fdput(f); | |
6304 | goto next_insn; | |
6305 | } | |
6306 | ||
6307 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
6308 | fdput(f); | |
6309 | return -E2BIG; | |
6310 | } | |
6311 | ||
0246e64d AS |
6312 | /* hold the map. If the program is rejected by verifier, |
6313 | * the map will be released by release_maps() or it | |
6314 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 6315 | * and all maps are released in free_used_maps() |
0246e64d | 6316 | */ |
92117d84 AS |
6317 | map = bpf_map_inc(map, false); |
6318 | if (IS_ERR(map)) { | |
6319 | fdput(f); | |
6320 | return PTR_ERR(map); | |
6321 | } | |
6322 | env->used_maps[env->used_map_cnt++] = map; | |
6323 | ||
b741f163 | 6324 | if (bpf_map_is_cgroup_storage(map) && |
de9cbbaa | 6325 | bpf_cgroup_storage_assign(env->prog, map)) { |
b741f163 | 6326 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
6327 | fdput(f); |
6328 | return -EBUSY; | |
6329 | } | |
6330 | ||
0246e64d AS |
6331 | fdput(f); |
6332 | next_insn: | |
6333 | insn++; | |
6334 | i++; | |
5e581dad DB |
6335 | continue; |
6336 | } | |
6337 | ||
6338 | /* Basic sanity check before we invest more work here. */ | |
6339 | if (!bpf_opcode_in_insntable(insn->code)) { | |
6340 | verbose(env, "unknown opcode %02x\n", insn->code); | |
6341 | return -EINVAL; | |
0246e64d AS |
6342 | } |
6343 | } | |
6344 | ||
6345 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
6346 | * 'struct bpf_map *' into a register instead of user map_fd. | |
6347 | * These pointers will be used later by verifier to validate map access. | |
6348 | */ | |
6349 | return 0; | |
6350 | } | |
6351 | ||
6352 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 6353 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 6354 | { |
8bad74f9 | 6355 | enum bpf_cgroup_storage_type stype; |
0246e64d AS |
6356 | int i; |
6357 | ||
8bad74f9 RG |
6358 | for_each_cgroup_storage_type(stype) { |
6359 | if (!env->prog->aux->cgroup_storage[stype]) | |
6360 | continue; | |
de9cbbaa | 6361 | bpf_cgroup_storage_release(env->prog, |
8bad74f9 RG |
6362 | env->prog->aux->cgroup_storage[stype]); |
6363 | } | |
de9cbbaa | 6364 | |
0246e64d AS |
6365 | for (i = 0; i < env->used_map_cnt; i++) |
6366 | bpf_map_put(env->used_maps[i]); | |
6367 | } | |
6368 | ||
6369 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 6370 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
6371 | { |
6372 | struct bpf_insn *insn = env->prog->insnsi; | |
6373 | int insn_cnt = env->prog->len; | |
6374 | int i; | |
6375 | ||
6376 | for (i = 0; i < insn_cnt; i++, insn++) | |
6377 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
6378 | insn->src_reg = 0; | |
6379 | } | |
6380 | ||
8041902d AS |
6381 | /* single env->prog->insni[off] instruction was replaced with the range |
6382 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
6383 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
6384 | */ | |
6385 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
6386 | u32 off, u32 cnt) | |
6387 | { | |
6388 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
c131187d | 6389 | int i; |
8041902d AS |
6390 | |
6391 | if (cnt == 1) | |
6392 | return 0; | |
fad953ce KC |
6393 | new_data = vzalloc(array_size(prog_len, |
6394 | sizeof(struct bpf_insn_aux_data))); | |
8041902d AS |
6395 | if (!new_data) |
6396 | return -ENOMEM; | |
6397 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
6398 | memcpy(new_data + off + cnt - 1, old_data + off, | |
6399 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
c131187d AS |
6400 | for (i = off; i < off + cnt - 1; i++) |
6401 | new_data[i].seen = true; | |
8041902d AS |
6402 | env->insn_aux_data = new_data; |
6403 | vfree(old_data); | |
6404 | return 0; | |
6405 | } | |
6406 | ||
cc8b0b92 AS |
6407 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
6408 | { | |
6409 | int i; | |
6410 | ||
6411 | if (len == 1) | |
6412 | return; | |
4cb3d99c JW |
6413 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
6414 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 6415 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 6416 | continue; |
9c8105bd | 6417 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
6418 | } |
6419 | } | |
6420 | ||
8041902d AS |
6421 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
6422 | const struct bpf_insn *patch, u32 len) | |
6423 | { | |
6424 | struct bpf_prog *new_prog; | |
6425 | ||
6426 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
6427 | if (!new_prog) | |
6428 | return NULL; | |
6429 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
6430 | return NULL; | |
cc8b0b92 | 6431 | adjust_subprog_starts(env, off, len); |
8041902d AS |
6432 | return new_prog; |
6433 | } | |
6434 | ||
52875a04 JK |
6435 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
6436 | u32 off, u32 cnt) | |
6437 | { | |
6438 | int i, j; | |
6439 | ||
6440 | /* find first prog starting at or after off (first to remove) */ | |
6441 | for (i = 0; i < env->subprog_cnt; i++) | |
6442 | if (env->subprog_info[i].start >= off) | |
6443 | break; | |
6444 | /* find first prog starting at or after off + cnt (first to stay) */ | |
6445 | for (j = i; j < env->subprog_cnt; j++) | |
6446 | if (env->subprog_info[j].start >= off + cnt) | |
6447 | break; | |
6448 | /* if j doesn't start exactly at off + cnt, we are just removing | |
6449 | * the front of previous prog | |
6450 | */ | |
6451 | if (env->subprog_info[j].start != off + cnt) | |
6452 | j--; | |
6453 | ||
6454 | if (j > i) { | |
6455 | struct bpf_prog_aux *aux = env->prog->aux; | |
6456 | int move; | |
6457 | ||
6458 | /* move fake 'exit' subprog as well */ | |
6459 | move = env->subprog_cnt + 1 - j; | |
6460 | ||
6461 | memmove(env->subprog_info + i, | |
6462 | env->subprog_info + j, | |
6463 | sizeof(*env->subprog_info) * move); | |
6464 | env->subprog_cnt -= j - i; | |
6465 | ||
6466 | /* remove func_info */ | |
6467 | if (aux->func_info) { | |
6468 | move = aux->func_info_cnt - j; | |
6469 | ||
6470 | memmove(aux->func_info + i, | |
6471 | aux->func_info + j, | |
6472 | sizeof(*aux->func_info) * move); | |
6473 | aux->func_info_cnt -= j - i; | |
6474 | /* func_info->insn_off is set after all code rewrites, | |
6475 | * in adjust_btf_func() - no need to adjust | |
6476 | */ | |
6477 | } | |
6478 | } else { | |
6479 | /* convert i from "first prog to remove" to "first to adjust" */ | |
6480 | if (env->subprog_info[i].start == off) | |
6481 | i++; | |
6482 | } | |
6483 | ||
6484 | /* update fake 'exit' subprog as well */ | |
6485 | for (; i <= env->subprog_cnt; i++) | |
6486 | env->subprog_info[i].start -= cnt; | |
6487 | ||
6488 | return 0; | |
6489 | } | |
6490 | ||
6491 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
6492 | u32 cnt) | |
6493 | { | |
6494 | struct bpf_prog *prog = env->prog; | |
6495 | u32 i, l_off, l_cnt, nr_linfo; | |
6496 | struct bpf_line_info *linfo; | |
6497 | ||
6498 | nr_linfo = prog->aux->nr_linfo; | |
6499 | if (!nr_linfo) | |
6500 | return 0; | |
6501 | ||
6502 | linfo = prog->aux->linfo; | |
6503 | ||
6504 | /* find first line info to remove, count lines to be removed */ | |
6505 | for (i = 0; i < nr_linfo; i++) | |
6506 | if (linfo[i].insn_off >= off) | |
6507 | break; | |
6508 | ||
6509 | l_off = i; | |
6510 | l_cnt = 0; | |
6511 | for (; i < nr_linfo; i++) | |
6512 | if (linfo[i].insn_off < off + cnt) | |
6513 | l_cnt++; | |
6514 | else | |
6515 | break; | |
6516 | ||
6517 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
6518 | * last removed linfo. prog is already modified, so prog->len == off | |
6519 | * means no live instructions after (tail of the program was removed). | |
6520 | */ | |
6521 | if (prog->len != off && l_cnt && | |
6522 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
6523 | l_cnt--; | |
6524 | linfo[--i].insn_off = off + cnt; | |
6525 | } | |
6526 | ||
6527 | /* remove the line info which refer to the removed instructions */ | |
6528 | if (l_cnt) { | |
6529 | memmove(linfo + l_off, linfo + i, | |
6530 | sizeof(*linfo) * (nr_linfo - i)); | |
6531 | ||
6532 | prog->aux->nr_linfo -= l_cnt; | |
6533 | nr_linfo = prog->aux->nr_linfo; | |
6534 | } | |
6535 | ||
6536 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
6537 | for (i = l_off; i < nr_linfo; i++) | |
6538 | linfo[i].insn_off -= cnt; | |
6539 | ||
6540 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
6541 | for (i = 0; i <= env->subprog_cnt; i++) | |
6542 | if (env->subprog_info[i].linfo_idx > l_off) { | |
6543 | /* program may have started in the removed region but | |
6544 | * may not be fully removed | |
6545 | */ | |
6546 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
6547 | env->subprog_info[i].linfo_idx -= l_cnt; | |
6548 | else | |
6549 | env->subprog_info[i].linfo_idx = l_off; | |
6550 | } | |
6551 | ||
6552 | return 0; | |
6553 | } | |
6554 | ||
6555 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
6556 | { | |
6557 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
6558 | unsigned int orig_prog_len = env->prog->len; | |
6559 | int err; | |
6560 | ||
6561 | err = bpf_remove_insns(env->prog, off, cnt); | |
6562 | if (err) | |
6563 | return err; | |
6564 | ||
6565 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
6566 | if (err) | |
6567 | return err; | |
6568 | ||
6569 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
6570 | if (err) | |
6571 | return err; | |
6572 | ||
6573 | memmove(aux_data + off, aux_data + off + cnt, | |
6574 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
6575 | ||
6576 | return 0; | |
6577 | } | |
6578 | ||
2a5418a1 DB |
6579 | /* The verifier does more data flow analysis than llvm and will not |
6580 | * explore branches that are dead at run time. Malicious programs can | |
6581 | * have dead code too. Therefore replace all dead at-run-time code | |
6582 | * with 'ja -1'. | |
6583 | * | |
6584 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
6585 | * program and through another bug we would manage to jump there, then | |
6586 | * we'd execute beyond program memory otherwise. Returning exception | |
6587 | * code also wouldn't work since we can have subprogs where the dead | |
6588 | * code could be located. | |
c131187d AS |
6589 | */ |
6590 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
6591 | { | |
6592 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 6593 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
6594 | struct bpf_insn *insn = env->prog->insnsi; |
6595 | const int insn_cnt = env->prog->len; | |
6596 | int i; | |
6597 | ||
6598 | for (i = 0; i < insn_cnt; i++) { | |
6599 | if (aux_data[i].seen) | |
6600 | continue; | |
2a5418a1 | 6601 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
6602 | } |
6603 | } | |
6604 | ||
e2ae4ca2 JK |
6605 | static bool insn_is_cond_jump(u8 code) |
6606 | { | |
6607 | u8 op; | |
6608 | ||
6609 | if (BPF_CLASS(code) != BPF_JMP) | |
6610 | return false; | |
6611 | ||
6612 | op = BPF_OP(code); | |
6613 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
6614 | } | |
6615 | ||
6616 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
6617 | { | |
6618 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
6619 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
6620 | struct bpf_insn *insn = env->prog->insnsi; | |
6621 | const int insn_cnt = env->prog->len; | |
6622 | int i; | |
6623 | ||
6624 | for (i = 0; i < insn_cnt; i++, insn++) { | |
6625 | if (!insn_is_cond_jump(insn->code)) | |
6626 | continue; | |
6627 | ||
6628 | if (!aux_data[i + 1].seen) | |
6629 | ja.off = insn->off; | |
6630 | else if (!aux_data[i + 1 + insn->off].seen) | |
6631 | ja.off = 0; | |
6632 | else | |
6633 | continue; | |
6634 | ||
6635 | memcpy(insn, &ja, sizeof(ja)); | |
6636 | } | |
6637 | } | |
6638 | ||
52875a04 JK |
6639 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
6640 | { | |
6641 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
6642 | int insn_cnt = env->prog->len; | |
6643 | int i, err; | |
6644 | ||
6645 | for (i = 0; i < insn_cnt; i++) { | |
6646 | int j; | |
6647 | ||
6648 | j = 0; | |
6649 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
6650 | j++; | |
6651 | if (!j) | |
6652 | continue; | |
6653 | ||
6654 | err = verifier_remove_insns(env, i, j); | |
6655 | if (err) | |
6656 | return err; | |
6657 | insn_cnt = env->prog->len; | |
6658 | } | |
6659 | ||
6660 | return 0; | |
6661 | } | |
6662 | ||
a1b14abc JK |
6663 | static int opt_remove_nops(struct bpf_verifier_env *env) |
6664 | { | |
6665 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
6666 | struct bpf_insn *insn = env->prog->insnsi; | |
6667 | int insn_cnt = env->prog->len; | |
6668 | int i, err; | |
6669 | ||
6670 | for (i = 0; i < insn_cnt; i++) { | |
6671 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
6672 | continue; | |
6673 | ||
6674 | err = verifier_remove_insns(env, i, 1); | |
6675 | if (err) | |
6676 | return err; | |
6677 | insn_cnt--; | |
6678 | i--; | |
6679 | } | |
6680 | ||
6681 | return 0; | |
6682 | } | |
6683 | ||
c64b7983 JS |
6684 | /* convert load instructions that access fields of a context type into a |
6685 | * sequence of instructions that access fields of the underlying structure: | |
6686 | * struct __sk_buff -> struct sk_buff | |
6687 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 6688 | */ |
58e2af8b | 6689 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 6690 | { |
00176a34 | 6691 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 6692 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 6693 | const int insn_cnt = env->prog->len; |
36bbef52 | 6694 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 6695 | u32 target_size, size_default, off; |
9bac3d6d | 6696 | struct bpf_prog *new_prog; |
d691f9e8 | 6697 | enum bpf_access_type type; |
f96da094 | 6698 | bool is_narrower_load; |
9bac3d6d | 6699 | |
b09928b9 DB |
6700 | if (ops->gen_prologue || env->seen_direct_write) { |
6701 | if (!ops->gen_prologue) { | |
6702 | verbose(env, "bpf verifier is misconfigured\n"); | |
6703 | return -EINVAL; | |
6704 | } | |
36bbef52 DB |
6705 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
6706 | env->prog); | |
6707 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 6708 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
6709 | return -EINVAL; |
6710 | } else if (cnt) { | |
8041902d | 6711 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
6712 | if (!new_prog) |
6713 | return -ENOMEM; | |
8041902d | 6714 | |
36bbef52 | 6715 | env->prog = new_prog; |
3df126f3 | 6716 | delta += cnt - 1; |
36bbef52 DB |
6717 | } |
6718 | } | |
6719 | ||
c64b7983 | 6720 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
6721 | return 0; |
6722 | ||
3df126f3 | 6723 | insn = env->prog->insnsi + delta; |
36bbef52 | 6724 | |
9bac3d6d | 6725 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 JS |
6726 | bpf_convert_ctx_access_t convert_ctx_access; |
6727 | ||
62c7989b DB |
6728 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
6729 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
6730 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 6731 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 6732 | type = BPF_READ; |
62c7989b DB |
6733 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
6734 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
6735 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 6736 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
6737 | type = BPF_WRITE; |
6738 | else | |
9bac3d6d AS |
6739 | continue; |
6740 | ||
af86ca4e AS |
6741 | if (type == BPF_WRITE && |
6742 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
6743 | struct bpf_insn patch[] = { | |
6744 | /* Sanitize suspicious stack slot with zero. | |
6745 | * There are no memory dependencies for this store, | |
6746 | * since it's only using frame pointer and immediate | |
6747 | * constant of zero | |
6748 | */ | |
6749 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
6750 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
6751 | 0), | |
6752 | /* the original STX instruction will immediately | |
6753 | * overwrite the same stack slot with appropriate value | |
6754 | */ | |
6755 | *insn, | |
6756 | }; | |
6757 | ||
6758 | cnt = ARRAY_SIZE(patch); | |
6759 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
6760 | if (!new_prog) | |
6761 | return -ENOMEM; | |
6762 | ||
6763 | delta += cnt - 1; | |
6764 | env->prog = new_prog; | |
6765 | insn = new_prog->insnsi + i + delta; | |
6766 | continue; | |
6767 | } | |
6768 | ||
c64b7983 JS |
6769 | switch (env->insn_aux_data[i + delta].ptr_type) { |
6770 | case PTR_TO_CTX: | |
6771 | if (!ops->convert_ctx_access) | |
6772 | continue; | |
6773 | convert_ctx_access = ops->convert_ctx_access; | |
6774 | break; | |
6775 | case PTR_TO_SOCKET: | |
6776 | convert_ctx_access = bpf_sock_convert_ctx_access; | |
6777 | break; | |
6778 | default: | |
9bac3d6d | 6779 | continue; |
c64b7983 | 6780 | } |
9bac3d6d | 6781 | |
31fd8581 | 6782 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 6783 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
6784 | |
6785 | /* If the read access is a narrower load of the field, | |
6786 | * convert to a 4/8-byte load, to minimum program type specific | |
6787 | * convert_ctx_access changes. If conversion is successful, | |
6788 | * we will apply proper mask to the result. | |
6789 | */ | |
f96da094 | 6790 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
6791 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
6792 | off = insn->off; | |
31fd8581 | 6793 | if (is_narrower_load) { |
f96da094 DB |
6794 | u8 size_code; |
6795 | ||
6796 | if (type == BPF_WRITE) { | |
61bd5218 | 6797 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
6798 | return -EINVAL; |
6799 | } | |
31fd8581 | 6800 | |
f96da094 | 6801 | size_code = BPF_H; |
31fd8581 YS |
6802 | if (ctx_field_size == 4) |
6803 | size_code = BPF_W; | |
6804 | else if (ctx_field_size == 8) | |
6805 | size_code = BPF_DW; | |
f96da094 | 6806 | |
bc23105c | 6807 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
6808 | insn->code = BPF_LDX | BPF_MEM | size_code; |
6809 | } | |
f96da094 DB |
6810 | |
6811 | target_size = 0; | |
c64b7983 JS |
6812 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
6813 | &target_size); | |
f96da094 DB |
6814 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
6815 | (ctx_field_size && !target_size)) { | |
61bd5218 | 6816 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
6817 | return -EINVAL; |
6818 | } | |
f96da094 DB |
6819 | |
6820 | if (is_narrower_load && size < target_size) { | |
46f53a65 AI |
6821 | u8 shift = (off & (size_default - 1)) * 8; |
6822 | ||
6823 | if (ctx_field_size <= 4) { | |
6824 | if (shift) | |
6825 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
6826 | insn->dst_reg, | |
6827 | shift); | |
31fd8581 | 6828 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 6829 | (1 << size * 8) - 1); |
46f53a65 AI |
6830 | } else { |
6831 | if (shift) | |
6832 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
6833 | insn->dst_reg, | |
6834 | shift); | |
31fd8581 | 6835 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 6836 | (1 << size * 8) - 1); |
46f53a65 | 6837 | } |
31fd8581 | 6838 | } |
9bac3d6d | 6839 | |
8041902d | 6840 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
6841 | if (!new_prog) |
6842 | return -ENOMEM; | |
6843 | ||
3df126f3 | 6844 | delta += cnt - 1; |
9bac3d6d AS |
6845 | |
6846 | /* keep walking new program and skip insns we just inserted */ | |
6847 | env->prog = new_prog; | |
3df126f3 | 6848 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
6849 | } |
6850 | ||
6851 | return 0; | |
6852 | } | |
6853 | ||
1c2a088a AS |
6854 | static int jit_subprogs(struct bpf_verifier_env *env) |
6855 | { | |
6856 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
6857 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
7105e828 | 6858 | struct bpf_insn *insn; |
1c2a088a | 6859 | void *old_bpf_func; |
c454a46b | 6860 | int err; |
1c2a088a | 6861 | |
f910cefa | 6862 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
6863 | return 0; |
6864 | ||
7105e828 | 6865 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
1c2a088a AS |
6866 | if (insn->code != (BPF_JMP | BPF_CALL) || |
6867 | insn->src_reg != BPF_PSEUDO_CALL) | |
6868 | continue; | |
c7a89784 DB |
6869 | /* Upon error here we cannot fall back to interpreter but |
6870 | * need a hard reject of the program. Thus -EFAULT is | |
6871 | * propagated in any case. | |
6872 | */ | |
1c2a088a AS |
6873 | subprog = find_subprog(env, i + insn->imm + 1); |
6874 | if (subprog < 0) { | |
6875 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
6876 | i + insn->imm + 1); | |
6877 | return -EFAULT; | |
6878 | } | |
6879 | /* temporarily remember subprog id inside insn instead of | |
6880 | * aux_data, since next loop will split up all insns into funcs | |
6881 | */ | |
f910cefa | 6882 | insn->off = subprog; |
1c2a088a AS |
6883 | /* remember original imm in case JIT fails and fallback |
6884 | * to interpreter will be needed | |
6885 | */ | |
6886 | env->insn_aux_data[i].call_imm = insn->imm; | |
6887 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
6888 | insn->imm = 1; | |
6889 | } | |
6890 | ||
c454a46b MKL |
6891 | err = bpf_prog_alloc_jited_linfo(prog); |
6892 | if (err) | |
6893 | goto out_undo_insn; | |
6894 | ||
6895 | err = -ENOMEM; | |
6396bb22 | 6896 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 6897 | if (!func) |
c7a89784 | 6898 | goto out_undo_insn; |
1c2a088a | 6899 | |
f910cefa | 6900 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 6901 | subprog_start = subprog_end; |
4cb3d99c | 6902 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
6903 | |
6904 | len = subprog_end - subprog_start; | |
6905 | func[i] = bpf_prog_alloc(bpf_prog_size(len), GFP_USER); | |
6906 | if (!func[i]) | |
6907 | goto out_free; | |
6908 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
6909 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 6910 | func[i]->type = prog->type; |
1c2a088a | 6911 | func[i]->len = len; |
4f74d809 DB |
6912 | if (bpf_prog_calc_tag(func[i])) |
6913 | goto out_free; | |
1c2a088a | 6914 | func[i]->is_func = 1; |
ba64e7d8 YS |
6915 | func[i]->aux->func_idx = i; |
6916 | /* the btf and func_info will be freed only at prog->aux */ | |
6917 | func[i]->aux->btf = prog->aux->btf; | |
6918 | func[i]->aux->func_info = prog->aux->func_info; | |
6919 | ||
1c2a088a AS |
6920 | /* Use bpf_prog_F_tag to indicate functions in stack traces. |
6921 | * Long term would need debug info to populate names | |
6922 | */ | |
6923 | func[i]->aux->name[0] = 'F'; | |
9c8105bd | 6924 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 6925 | func[i]->jit_requested = 1; |
c454a46b MKL |
6926 | func[i]->aux->linfo = prog->aux->linfo; |
6927 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
6928 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
6929 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
1c2a088a AS |
6930 | func[i] = bpf_int_jit_compile(func[i]); |
6931 | if (!func[i]->jited) { | |
6932 | err = -ENOTSUPP; | |
6933 | goto out_free; | |
6934 | } | |
6935 | cond_resched(); | |
6936 | } | |
6937 | /* at this point all bpf functions were successfully JITed | |
6938 | * now populate all bpf_calls with correct addresses and | |
6939 | * run last pass of JIT | |
6940 | */ | |
f910cefa | 6941 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
6942 | insn = func[i]->insnsi; |
6943 | for (j = 0; j < func[i]->len; j++, insn++) { | |
6944 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
6945 | insn->src_reg != BPF_PSEUDO_CALL) | |
6946 | continue; | |
6947 | subprog = insn->off; | |
1c2a088a AS |
6948 | insn->imm = (u64 (*)(u64, u64, u64, u64, u64)) |
6949 | func[subprog]->bpf_func - | |
6950 | __bpf_call_base; | |
6951 | } | |
2162fed4 SD |
6952 | |
6953 | /* we use the aux data to keep a list of the start addresses | |
6954 | * of the JITed images for each function in the program | |
6955 | * | |
6956 | * for some architectures, such as powerpc64, the imm field | |
6957 | * might not be large enough to hold the offset of the start | |
6958 | * address of the callee's JITed image from __bpf_call_base | |
6959 | * | |
6960 | * in such cases, we can lookup the start address of a callee | |
6961 | * by using its subprog id, available from the off field of | |
6962 | * the call instruction, as an index for this list | |
6963 | */ | |
6964 | func[i]->aux->func = func; | |
6965 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 6966 | } |
f910cefa | 6967 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
6968 | old_bpf_func = func[i]->bpf_func; |
6969 | tmp = bpf_int_jit_compile(func[i]); | |
6970 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
6971 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 6972 | err = -ENOTSUPP; |
1c2a088a AS |
6973 | goto out_free; |
6974 | } | |
6975 | cond_resched(); | |
6976 | } | |
6977 | ||
6978 | /* finally lock prog and jit images for all functions and | |
6979 | * populate kallsysm | |
6980 | */ | |
f910cefa | 6981 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
6982 | bpf_prog_lock_ro(func[i]); |
6983 | bpf_prog_kallsyms_add(func[i]); | |
6984 | } | |
7105e828 DB |
6985 | |
6986 | /* Last step: make now unused interpreter insns from main | |
6987 | * prog consistent for later dump requests, so they can | |
6988 | * later look the same as if they were interpreted only. | |
6989 | */ | |
6990 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
7105e828 DB |
6991 | if (insn->code != (BPF_JMP | BPF_CALL) || |
6992 | insn->src_reg != BPF_PSEUDO_CALL) | |
6993 | continue; | |
6994 | insn->off = env->insn_aux_data[i].call_imm; | |
6995 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 6996 | insn->imm = subprog; |
7105e828 DB |
6997 | } |
6998 | ||
1c2a088a AS |
6999 | prog->jited = 1; |
7000 | prog->bpf_func = func[0]->bpf_func; | |
7001 | prog->aux->func = func; | |
f910cefa | 7002 | prog->aux->func_cnt = env->subprog_cnt; |
c454a46b | 7003 | bpf_prog_free_unused_jited_linfo(prog); |
1c2a088a AS |
7004 | return 0; |
7005 | out_free: | |
f910cefa | 7006 | for (i = 0; i < env->subprog_cnt; i++) |
1c2a088a AS |
7007 | if (func[i]) |
7008 | bpf_jit_free(func[i]); | |
7009 | kfree(func); | |
c7a89784 | 7010 | out_undo_insn: |
1c2a088a AS |
7011 | /* cleanup main prog to be interpreted */ |
7012 | prog->jit_requested = 0; | |
7013 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
7014 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
7015 | insn->src_reg != BPF_PSEUDO_CALL) | |
7016 | continue; | |
7017 | insn->off = 0; | |
7018 | insn->imm = env->insn_aux_data[i].call_imm; | |
7019 | } | |
c454a46b | 7020 | bpf_prog_free_jited_linfo(prog); |
1c2a088a AS |
7021 | return err; |
7022 | } | |
7023 | ||
1ea47e01 AS |
7024 | static int fixup_call_args(struct bpf_verifier_env *env) |
7025 | { | |
19d28fbd | 7026 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
7027 | struct bpf_prog *prog = env->prog; |
7028 | struct bpf_insn *insn = prog->insnsi; | |
7029 | int i, depth; | |
19d28fbd | 7030 | #endif |
e4052d06 | 7031 | int err = 0; |
1ea47e01 | 7032 | |
e4052d06 QM |
7033 | if (env->prog->jit_requested && |
7034 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
7035 | err = jit_subprogs(env); |
7036 | if (err == 0) | |
1c2a088a | 7037 | return 0; |
c7a89784 DB |
7038 | if (err == -EFAULT) |
7039 | return err; | |
19d28fbd DM |
7040 | } |
7041 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1ea47e01 AS |
7042 | for (i = 0; i < prog->len; i++, insn++) { |
7043 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
7044 | insn->src_reg != BPF_PSEUDO_CALL) | |
7045 | continue; | |
7046 | depth = get_callee_stack_depth(env, insn, i); | |
7047 | if (depth < 0) | |
7048 | return depth; | |
7049 | bpf_patch_call_args(insn, depth); | |
7050 | } | |
19d28fbd DM |
7051 | err = 0; |
7052 | #endif | |
7053 | return err; | |
1ea47e01 AS |
7054 | } |
7055 | ||
79741b3b | 7056 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 7057 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
7058 | * |
7059 | * this function is called after eBPF program passed verification | |
7060 | */ | |
79741b3b | 7061 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 7062 | { |
79741b3b AS |
7063 | struct bpf_prog *prog = env->prog; |
7064 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 7065 | const struct bpf_func_proto *fn; |
79741b3b | 7066 | const int insn_cnt = prog->len; |
09772d92 | 7067 | const struct bpf_map_ops *ops; |
c93552c4 | 7068 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
7069 | struct bpf_insn insn_buf[16]; |
7070 | struct bpf_prog *new_prog; | |
7071 | struct bpf_map *map_ptr; | |
7072 | int i, cnt, delta = 0; | |
e245c5c6 | 7073 | |
79741b3b | 7074 | for (i = 0; i < insn_cnt; i++, insn++) { |
f6b1b3bf DB |
7075 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
7076 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
7077 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 7078 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf DB |
7079 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
7080 | struct bpf_insn mask_and_div[] = { | |
7081 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
7082 | /* Rx div 0 -> 0 */ | |
7083 | BPF_JMP_IMM(BPF_JNE, insn->src_reg, 0, 2), | |
7084 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), | |
7085 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
7086 | *insn, | |
7087 | }; | |
7088 | struct bpf_insn mask_and_mod[] = { | |
7089 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
7090 | /* Rx mod 0 -> Rx */ | |
7091 | BPF_JMP_IMM(BPF_JEQ, insn->src_reg, 0, 1), | |
7092 | *insn, | |
7093 | }; | |
7094 | struct bpf_insn *patchlet; | |
7095 | ||
7096 | if (insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
7097 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
7098 | patchlet = mask_and_div + (is64 ? 1 : 0); | |
7099 | cnt = ARRAY_SIZE(mask_and_div) - (is64 ? 1 : 0); | |
7100 | } else { | |
7101 | patchlet = mask_and_mod + (is64 ? 1 : 0); | |
7102 | cnt = ARRAY_SIZE(mask_and_mod) - (is64 ? 1 : 0); | |
7103 | } | |
7104 | ||
7105 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
7106 | if (!new_prog) |
7107 | return -ENOMEM; | |
7108 | ||
7109 | delta += cnt - 1; | |
7110 | env->prog = prog = new_prog; | |
7111 | insn = new_prog->insnsi + i + delta; | |
7112 | continue; | |
7113 | } | |
7114 | ||
e0cea7ce DB |
7115 | if (BPF_CLASS(insn->code) == BPF_LD && |
7116 | (BPF_MODE(insn->code) == BPF_ABS || | |
7117 | BPF_MODE(insn->code) == BPF_IND)) { | |
7118 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
7119 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
7120 | verbose(env, "bpf verifier is misconfigured\n"); | |
7121 | return -EINVAL; | |
7122 | } | |
7123 | ||
7124 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
7125 | if (!new_prog) | |
7126 | return -ENOMEM; | |
7127 | ||
7128 | delta += cnt - 1; | |
7129 | env->prog = prog = new_prog; | |
7130 | insn = new_prog->insnsi + i + delta; | |
7131 | continue; | |
7132 | } | |
7133 | ||
979d63d5 DB |
7134 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
7135 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
7136 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
7137 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
7138 | struct bpf_insn insn_buf[16]; | |
7139 | struct bpf_insn *patch = &insn_buf[0]; | |
7140 | bool issrc, isneg; | |
7141 | u32 off_reg; | |
7142 | ||
7143 | aux = &env->insn_aux_data[i + delta]; | |
7144 | if (!aux->alu_state) | |
7145 | continue; | |
7146 | ||
7147 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
7148 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
7149 | BPF_ALU_SANITIZE_SRC; | |
7150 | ||
7151 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
7152 | if (isneg) | |
7153 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
7154 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit - 1); | |
7155 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
7156 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
7157 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
7158 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
7159 | if (issrc) { | |
7160 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, | |
7161 | off_reg); | |
7162 | insn->src_reg = BPF_REG_AX; | |
7163 | } else { | |
7164 | *patch++ = BPF_ALU64_REG(BPF_AND, off_reg, | |
7165 | BPF_REG_AX); | |
7166 | } | |
7167 | if (isneg) | |
7168 | insn->code = insn->code == code_add ? | |
7169 | code_sub : code_add; | |
7170 | *patch++ = *insn; | |
7171 | if (issrc && isneg) | |
7172 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
7173 | cnt = patch - insn_buf; | |
7174 | ||
7175 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
7176 | if (!new_prog) | |
7177 | return -ENOMEM; | |
7178 | ||
7179 | delta += cnt - 1; | |
7180 | env->prog = prog = new_prog; | |
7181 | insn = new_prog->insnsi + i + delta; | |
7182 | continue; | |
7183 | } | |
7184 | ||
79741b3b AS |
7185 | if (insn->code != (BPF_JMP | BPF_CALL)) |
7186 | continue; | |
cc8b0b92 AS |
7187 | if (insn->src_reg == BPF_PSEUDO_CALL) |
7188 | continue; | |
e245c5c6 | 7189 | |
79741b3b AS |
7190 | if (insn->imm == BPF_FUNC_get_route_realm) |
7191 | prog->dst_needed = 1; | |
7192 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
7193 | bpf_user_rnd_init_once(); | |
9802d865 JB |
7194 | if (insn->imm == BPF_FUNC_override_return) |
7195 | prog->kprobe_override = 1; | |
79741b3b | 7196 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
7197 | /* If we tail call into other programs, we |
7198 | * cannot make any assumptions since they can | |
7199 | * be replaced dynamically during runtime in | |
7200 | * the program array. | |
7201 | */ | |
7202 | prog->cb_access = 1; | |
80a58d02 | 7203 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
e647815a | 7204 | env->prog->aux->max_pkt_offset = MAX_PACKET_OFF; |
7b9f6da1 | 7205 | |
79741b3b AS |
7206 | /* mark bpf_tail_call as different opcode to avoid |
7207 | * conditional branch in the interpeter for every normal | |
7208 | * call and to prevent accidental JITing by JIT compiler | |
7209 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 7210 | */ |
79741b3b | 7211 | insn->imm = 0; |
71189fa9 | 7212 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 7213 | |
c93552c4 DB |
7214 | aux = &env->insn_aux_data[i + delta]; |
7215 | if (!bpf_map_ptr_unpriv(aux)) | |
7216 | continue; | |
7217 | ||
b2157399 AS |
7218 | /* instead of changing every JIT dealing with tail_call |
7219 | * emit two extra insns: | |
7220 | * if (index >= max_entries) goto out; | |
7221 | * index &= array->index_mask; | |
7222 | * to avoid out-of-bounds cpu speculation | |
7223 | */ | |
c93552c4 | 7224 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 7225 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
7226 | return -EINVAL; |
7227 | } | |
c93552c4 DB |
7228 | |
7229 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
b2157399 AS |
7230 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
7231 | map_ptr->max_entries, 2); | |
7232 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
7233 | container_of(map_ptr, | |
7234 | struct bpf_array, | |
7235 | map)->index_mask); | |
7236 | insn_buf[2] = *insn; | |
7237 | cnt = 3; | |
7238 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
7239 | if (!new_prog) | |
7240 | return -ENOMEM; | |
7241 | ||
7242 | delta += cnt - 1; | |
7243 | env->prog = prog = new_prog; | |
7244 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
7245 | continue; |
7246 | } | |
e245c5c6 | 7247 | |
89c63074 | 7248 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
7249 | * and other inlining handlers are currently limited to 64 bit |
7250 | * only. | |
89c63074 | 7251 | */ |
60b58afc | 7252 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
7253 | (insn->imm == BPF_FUNC_map_lookup_elem || |
7254 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
7255 | insn->imm == BPF_FUNC_map_delete_elem || |
7256 | insn->imm == BPF_FUNC_map_push_elem || | |
7257 | insn->imm == BPF_FUNC_map_pop_elem || | |
7258 | insn->imm == BPF_FUNC_map_peek_elem)) { | |
c93552c4 DB |
7259 | aux = &env->insn_aux_data[i + delta]; |
7260 | if (bpf_map_ptr_poisoned(aux)) | |
7261 | goto patch_call_imm; | |
7262 | ||
7263 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
09772d92 DB |
7264 | ops = map_ptr->ops; |
7265 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
7266 | ops->map_gen_lookup) { | |
7267 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
7268 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
7269 | verbose(env, "bpf verifier is misconfigured\n"); | |
7270 | return -EINVAL; | |
7271 | } | |
81ed18ab | 7272 | |
09772d92 DB |
7273 | new_prog = bpf_patch_insn_data(env, i + delta, |
7274 | insn_buf, cnt); | |
7275 | if (!new_prog) | |
7276 | return -ENOMEM; | |
81ed18ab | 7277 | |
09772d92 DB |
7278 | delta += cnt - 1; |
7279 | env->prog = prog = new_prog; | |
7280 | insn = new_prog->insnsi + i + delta; | |
7281 | continue; | |
7282 | } | |
81ed18ab | 7283 | |
09772d92 DB |
7284 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
7285 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
7286 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
7287 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
7288 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
7289 | (int (*)(struct bpf_map *map, void *key, void *value, | |
7290 | u64 flags))NULL)); | |
84430d42 DB |
7291 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
7292 | (int (*)(struct bpf_map *map, void *value, | |
7293 | u64 flags))NULL)); | |
7294 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
7295 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
7296 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
7297 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
7298 | ||
09772d92 DB |
7299 | switch (insn->imm) { |
7300 | case BPF_FUNC_map_lookup_elem: | |
7301 | insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) - | |
7302 | __bpf_call_base; | |
7303 | continue; | |
7304 | case BPF_FUNC_map_update_elem: | |
7305 | insn->imm = BPF_CAST_CALL(ops->map_update_elem) - | |
7306 | __bpf_call_base; | |
7307 | continue; | |
7308 | case BPF_FUNC_map_delete_elem: | |
7309 | insn->imm = BPF_CAST_CALL(ops->map_delete_elem) - | |
7310 | __bpf_call_base; | |
7311 | continue; | |
84430d42 DB |
7312 | case BPF_FUNC_map_push_elem: |
7313 | insn->imm = BPF_CAST_CALL(ops->map_push_elem) - | |
7314 | __bpf_call_base; | |
7315 | continue; | |
7316 | case BPF_FUNC_map_pop_elem: | |
7317 | insn->imm = BPF_CAST_CALL(ops->map_pop_elem) - | |
7318 | __bpf_call_base; | |
7319 | continue; | |
7320 | case BPF_FUNC_map_peek_elem: | |
7321 | insn->imm = BPF_CAST_CALL(ops->map_peek_elem) - | |
7322 | __bpf_call_base; | |
7323 | continue; | |
09772d92 | 7324 | } |
81ed18ab | 7325 | |
09772d92 | 7326 | goto patch_call_imm; |
81ed18ab AS |
7327 | } |
7328 | ||
7329 | patch_call_imm: | |
5e43f899 | 7330 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
7331 | /* all functions that have prototype and verifier allowed |
7332 | * programs to call them, must be real in-kernel functions | |
7333 | */ | |
7334 | if (!fn->func) { | |
61bd5218 JK |
7335 | verbose(env, |
7336 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
7337 | func_id_name(insn->imm), insn->imm); |
7338 | return -EFAULT; | |
e245c5c6 | 7339 | } |
79741b3b | 7340 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 7341 | } |
e245c5c6 | 7342 | |
79741b3b AS |
7343 | return 0; |
7344 | } | |
e245c5c6 | 7345 | |
58e2af8b | 7346 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 7347 | { |
58e2af8b | 7348 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
7349 | int i; |
7350 | ||
7351 | if (!env->explored_states) | |
7352 | return; | |
7353 | ||
7354 | for (i = 0; i < env->prog->len; i++) { | |
7355 | sl = env->explored_states[i]; | |
7356 | ||
7357 | if (sl) | |
7358 | while (sl != STATE_LIST_MARK) { | |
7359 | sln = sl->next; | |
1969db47 | 7360 | free_verifier_state(&sl->state, false); |
f1bca824 AS |
7361 | kfree(sl); |
7362 | sl = sln; | |
7363 | } | |
7364 | } | |
7365 | ||
7366 | kfree(env->explored_states); | |
7367 | } | |
7368 | ||
838e9690 YS |
7369 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, |
7370 | union bpf_attr __user *uattr) | |
51580e79 | 7371 | { |
58e2af8b | 7372 | struct bpf_verifier_env *env; |
b9193c1b | 7373 | struct bpf_verifier_log *log; |
51580e79 | 7374 | int ret = -EINVAL; |
e2ae4ca2 | 7375 | bool is_priv; |
51580e79 | 7376 | |
eba0c929 AB |
7377 | /* no program is valid */ |
7378 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
7379 | return -EINVAL; | |
7380 | ||
58e2af8b | 7381 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
7382 | * allocate/free it every time bpf_check() is called |
7383 | */ | |
58e2af8b | 7384 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
7385 | if (!env) |
7386 | return -ENOMEM; | |
61bd5218 | 7387 | log = &env->log; |
cbd35700 | 7388 | |
fad953ce KC |
7389 | env->insn_aux_data = |
7390 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), | |
7391 | (*prog)->len)); | |
3df126f3 JK |
7392 | ret = -ENOMEM; |
7393 | if (!env->insn_aux_data) | |
7394 | goto err_free_env; | |
9bac3d6d | 7395 | env->prog = *prog; |
00176a34 | 7396 | env->ops = bpf_verifier_ops[env->prog->type]; |
0246e64d | 7397 | |
cbd35700 AS |
7398 | /* grab the mutex to protect few globals used by verifier */ |
7399 | mutex_lock(&bpf_verifier_lock); | |
7400 | ||
7401 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
7402 | /* user requested verbose verifier output | |
7403 | * and supplied buffer to store the verification trace | |
7404 | */ | |
e7bf8249 JK |
7405 | log->level = attr->log_level; |
7406 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
7407 | log->len_total = attr->log_size; | |
cbd35700 AS |
7408 | |
7409 | ret = -EINVAL; | |
e7bf8249 JK |
7410 | /* log attributes have to be sane */ |
7411 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 8 || | |
7412 | !log->level || !log->ubuf) | |
3df126f3 | 7413 | goto err_unlock; |
cbd35700 | 7414 | } |
1ad2f583 DB |
7415 | |
7416 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
7417 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 7418 | env->strict_alignment = true; |
e9ee9efc DM |
7419 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
7420 | env->strict_alignment = false; | |
cbd35700 | 7421 | |
e2ae4ca2 JK |
7422 | is_priv = capable(CAP_SYS_ADMIN); |
7423 | env->allow_ptr_leaks = is_priv; | |
7424 | ||
f4e3ec0d JK |
7425 | ret = replace_map_fd_with_map_ptr(env); |
7426 | if (ret < 0) | |
7427 | goto skip_full_check; | |
7428 | ||
cae1927c | 7429 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
a40a2632 | 7430 | ret = bpf_prog_offload_verifier_prep(env->prog); |
ab3f0063 | 7431 | if (ret) |
f4e3ec0d | 7432 | goto skip_full_check; |
ab3f0063 JK |
7433 | } |
7434 | ||
9bac3d6d | 7435 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 7436 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
7437 | GFP_USER); |
7438 | ret = -ENOMEM; | |
7439 | if (!env->explored_states) | |
7440 | goto skip_full_check; | |
7441 | ||
d9762e84 | 7442 | ret = check_subprogs(env); |
475fb78f AS |
7443 | if (ret < 0) |
7444 | goto skip_full_check; | |
7445 | ||
c454a46b | 7446 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
7447 | if (ret < 0) |
7448 | goto skip_full_check; | |
7449 | ||
d9762e84 MKL |
7450 | ret = check_cfg(env); |
7451 | if (ret < 0) | |
7452 | goto skip_full_check; | |
7453 | ||
17a52670 | 7454 | ret = do_check(env); |
8c01c4f8 CG |
7455 | if (env->cur_state) { |
7456 | free_verifier_state(env->cur_state, true); | |
7457 | env->cur_state = NULL; | |
7458 | } | |
cbd35700 | 7459 | |
c941ce9c QM |
7460 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
7461 | ret = bpf_prog_offload_finalize(env); | |
7462 | ||
0246e64d | 7463 | skip_full_check: |
638f5b90 | 7464 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 7465 | free_states(env); |
0246e64d | 7466 | |
c131187d | 7467 | if (ret == 0) |
9b38c405 | 7468 | ret = check_max_stack_depth(env); |
c131187d | 7469 | |
9b38c405 | 7470 | /* instruction rewrites happen after this point */ |
e2ae4ca2 JK |
7471 | if (is_priv) { |
7472 | if (ret == 0) | |
7473 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
7474 | if (ret == 0) |
7475 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
7476 | if (ret == 0) |
7477 | ret = opt_remove_nops(env); | |
52875a04 JK |
7478 | } else { |
7479 | if (ret == 0) | |
7480 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
7481 | } |
7482 | ||
9bac3d6d AS |
7483 | if (ret == 0) |
7484 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
7485 | ret = convert_ctx_accesses(env); | |
7486 | ||
e245c5c6 | 7487 | if (ret == 0) |
79741b3b | 7488 | ret = fixup_bpf_calls(env); |
e245c5c6 | 7489 | |
1ea47e01 AS |
7490 | if (ret == 0) |
7491 | ret = fixup_call_args(env); | |
7492 | ||
a2a7d570 | 7493 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 7494 | ret = -ENOSPC; |
a2a7d570 | 7495 | if (log->level && !log->ubuf) { |
cbd35700 | 7496 | ret = -EFAULT; |
a2a7d570 | 7497 | goto err_release_maps; |
cbd35700 AS |
7498 | } |
7499 | ||
0246e64d AS |
7500 | if (ret == 0 && env->used_map_cnt) { |
7501 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
7502 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
7503 | sizeof(env->used_maps[0]), | |
7504 | GFP_KERNEL); | |
0246e64d | 7505 | |
9bac3d6d | 7506 | if (!env->prog->aux->used_maps) { |
0246e64d | 7507 | ret = -ENOMEM; |
a2a7d570 | 7508 | goto err_release_maps; |
0246e64d AS |
7509 | } |
7510 | ||
9bac3d6d | 7511 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 7512 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 7513 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
7514 | |
7515 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
7516 | * bpf_ld_imm64 instructions | |
7517 | */ | |
7518 | convert_pseudo_ld_imm64(env); | |
7519 | } | |
cbd35700 | 7520 | |
ba64e7d8 YS |
7521 | if (ret == 0) |
7522 | adjust_btf_func(env); | |
7523 | ||
a2a7d570 | 7524 | err_release_maps: |
9bac3d6d | 7525 | if (!env->prog->aux->used_maps) |
0246e64d | 7526 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 7527 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
7528 | */ |
7529 | release_maps(env); | |
9bac3d6d | 7530 | *prog = env->prog; |
3df126f3 | 7531 | err_unlock: |
cbd35700 | 7532 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
7533 | vfree(env->insn_aux_data); |
7534 | err_free_env: | |
7535 | kfree(env); | |
51580e79 AS |
7536 | return ret; |
7537 | } |