]>
Commit | Line | Data |
---|---|---|
5b497af4 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
51580e79 | 2 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 3 | * Copyright (c) 2016 Facebook |
fd978bf7 | 4 | * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io |
51580e79 | 5 | */ |
838e9690 | 6 | #include <uapi/linux/btf.h> |
51580e79 AS |
7 | #include <linux/kernel.h> |
8 | #include <linux/types.h> | |
9 | #include <linux/slab.h> | |
10 | #include <linux/bpf.h> | |
838e9690 | 11 | #include <linux/btf.h> |
58e2af8b | 12 | #include <linux/bpf_verifier.h> |
51580e79 AS |
13 | #include <linux/filter.h> |
14 | #include <net/netlink.h> | |
15 | #include <linux/file.h> | |
16 | #include <linux/vmalloc.h> | |
ebb676da | 17 | #include <linux/stringify.h> |
cc8b0b92 AS |
18 | #include <linux/bsearch.h> |
19 | #include <linux/sort.h> | |
c195651e | 20 | #include <linux/perf_event.h> |
d9762e84 | 21 | #include <linux/ctype.h> |
6ba43b76 | 22 | #include <linux/error-injection.h> |
9e4e01df | 23 | #include <linux/bpf_lsm.h> |
51580e79 | 24 | |
f4ac7e0b JK |
25 | #include "disasm.h" |
26 | ||
00176a34 | 27 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
91cc1a99 | 28 | #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \ |
00176a34 JK |
29 | [_id] = & _name ## _verifier_ops, |
30 | #define BPF_MAP_TYPE(_id, _ops) | |
f2e10bff | 31 | #define BPF_LINK_TYPE(_id, _name) |
00176a34 JK |
32 | #include <linux/bpf_types.h> |
33 | #undef BPF_PROG_TYPE | |
34 | #undef BPF_MAP_TYPE | |
f2e10bff | 35 | #undef BPF_LINK_TYPE |
00176a34 JK |
36 | }; |
37 | ||
51580e79 AS |
38 | /* bpf_check() is a static code analyzer that walks eBPF program |
39 | * instruction by instruction and updates register/stack state. | |
40 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
41 | * | |
42 | * The first pass is depth-first-search to check that the program is a DAG. | |
43 | * It rejects the following programs: | |
44 | * - larger than BPF_MAXINSNS insns | |
45 | * - if loop is present (detected via back-edge) | |
46 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
47 | * - out of bounds or malformed jumps | |
48 | * The second pass is all possible path descent from the 1st insn. | |
49 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 50 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
51 | * insn is less then 4K, but there are too many branches that change stack/regs. |
52 | * Number of 'branches to be analyzed' is limited to 1k | |
53 | * | |
54 | * On entry to each instruction, each register has a type, and the instruction | |
55 | * changes the types of the registers depending on instruction semantics. | |
56 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
57 | * copied to R1. | |
58 | * | |
59 | * All registers are 64-bit. | |
60 | * R0 - return register | |
61 | * R1-R5 argument passing registers | |
62 | * R6-R9 callee saved registers | |
63 | * R10 - frame pointer read-only | |
64 | * | |
65 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
66 | * and has type PTR_TO_CTX. | |
67 | * | |
68 | * Verifier tracks arithmetic operations on pointers in case: | |
69 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
70 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
71 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
72 | * and 2nd arithmetic instruction is pattern matched to recognize | |
73 | * that it wants to construct a pointer to some element within stack. | |
74 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
75 | * (and -20 constant is saved for further stack bounds checking). | |
76 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
77 | * | |
f1174f77 | 78 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 79 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 80 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
81 | * |
82 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
83 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
84 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
85 | * |
86 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
87 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
88 | * | |
89 | * registers used to pass values to function calls are checked against | |
90 | * function argument constraints. | |
91 | * | |
92 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
93 | * It means that the register type passed to this function must be | |
94 | * PTR_TO_STACK and it will be used inside the function as | |
95 | * 'pointer to map element key' | |
96 | * | |
97 | * For example the argument constraints for bpf_map_lookup_elem(): | |
98 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
99 | * .arg1_type = ARG_CONST_MAP_PTR, | |
100 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
101 | * | |
102 | * ret_type says that this function returns 'pointer to map elem value or null' | |
103 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
104 | * 2nd argument should be a pointer to stack, which will be used inside | |
105 | * the helper function as a pointer to map element key. | |
106 | * | |
107 | * On the kernel side the helper function looks like: | |
108 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
109 | * { | |
110 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
111 | * void *key = (void *) (unsigned long) r2; | |
112 | * void *value; | |
113 | * | |
114 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
115 | * [key, key + map->key_size) bytes are valid and were initialized on | |
116 | * the stack of eBPF program. | |
117 | * } | |
118 | * | |
119 | * Corresponding eBPF program may look like: | |
120 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
121 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
122 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
123 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
124 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
125 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
126 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
127 | * | |
128 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
129 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
130 | * and were initialized prior to this call. | |
131 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
132 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
133 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
134 | * returns ether pointer to map value or NULL. | |
135 | * | |
136 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
137 | * insn, the register holding that pointer in the true branch changes state to | |
138 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
139 | * branch. See check_cond_jmp_op(). | |
140 | * | |
141 | * After the call R0 is set to return type of the function and registers R1-R5 | |
142 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
143 | * |
144 | * The following reference types represent a potential reference to a kernel | |
145 | * resource which, after first being allocated, must be checked and freed by | |
146 | * the BPF program: | |
147 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
148 | * | |
149 | * When the verifier sees a helper call return a reference type, it allocates a | |
150 | * pointer id for the reference and stores it in the current function state. | |
151 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
152 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
153 | * passes through a NULL-check conditional. For the branch wherein the state is | |
154 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
155 | * |
156 | * For each helper function that allocates a reference, such as | |
157 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
158 | * bpf_sk_release(). When a reference type passes into the release function, | |
159 | * the verifier also releases the reference. If any unchecked or unreleased | |
160 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
161 | */ |
162 | ||
17a52670 | 163 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 164 | struct bpf_verifier_stack_elem { |
17a52670 AS |
165 | /* verifer state is 'st' |
166 | * before processing instruction 'insn_idx' | |
167 | * and after processing instruction 'prev_insn_idx' | |
168 | */ | |
58e2af8b | 169 | struct bpf_verifier_state st; |
17a52670 AS |
170 | int insn_idx; |
171 | int prev_insn_idx; | |
58e2af8b | 172 | struct bpf_verifier_stack_elem *next; |
6f8a57cc AN |
173 | /* length of verifier log at the time this state was pushed on stack */ |
174 | u32 log_pos; | |
cbd35700 AS |
175 | }; |
176 | ||
b285fcb7 | 177 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 178 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 179 | |
d2e4c1e6 DB |
180 | #define BPF_MAP_KEY_POISON (1ULL << 63) |
181 | #define BPF_MAP_KEY_SEEN (1ULL << 62) | |
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 | { | |
d2e4c1e6 | 190 | return BPF_MAP_PTR(aux->map_ptr_state) == BPF_MAP_PTR_POISON; |
c93552c4 DB |
191 | } |
192 | ||
193 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
194 | { | |
d2e4c1e6 | 195 | return aux->map_ptr_state & BPF_MAP_PTR_UNPRIV; |
c93552c4 DB |
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); | |
d2e4c1e6 DB |
203 | aux->map_ptr_state = (unsigned long)map | |
204 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
205 | } | |
206 | ||
207 | static bool bpf_map_key_poisoned(const struct bpf_insn_aux_data *aux) | |
208 | { | |
209 | return aux->map_key_state & BPF_MAP_KEY_POISON; | |
210 | } | |
211 | ||
212 | static bool bpf_map_key_unseen(const struct bpf_insn_aux_data *aux) | |
213 | { | |
214 | return !(aux->map_key_state & BPF_MAP_KEY_SEEN); | |
215 | } | |
216 | ||
217 | static u64 bpf_map_key_immediate(const struct bpf_insn_aux_data *aux) | |
218 | { | |
219 | return aux->map_key_state & ~(BPF_MAP_KEY_SEEN | BPF_MAP_KEY_POISON); | |
220 | } | |
221 | ||
222 | static void bpf_map_key_store(struct bpf_insn_aux_data *aux, u64 state) | |
223 | { | |
224 | bool poisoned = bpf_map_key_poisoned(aux); | |
225 | ||
226 | aux->map_key_state = state | BPF_MAP_KEY_SEEN | | |
227 | (poisoned ? BPF_MAP_KEY_POISON : 0ULL); | |
c93552c4 | 228 | } |
fad73a1a | 229 | |
33ff9823 DB |
230 | struct bpf_call_arg_meta { |
231 | struct bpf_map *map_ptr; | |
435faee1 | 232 | bool raw_mode; |
36bbef52 | 233 | bool pkt_access; |
435faee1 DB |
234 | int regno; |
235 | int access_size; | |
10060503 | 236 | u64 msize_max_value; |
1b986589 | 237 | int ref_obj_id; |
d83525ca | 238 | int func_id; |
a7658e1a | 239 | u32 btf_id; |
33ff9823 DB |
240 | }; |
241 | ||
8580ac94 AS |
242 | struct btf *btf_vmlinux; |
243 | ||
cbd35700 AS |
244 | static DEFINE_MUTEX(bpf_verifier_lock); |
245 | ||
d9762e84 MKL |
246 | static const struct bpf_line_info * |
247 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
248 | { | |
249 | const struct bpf_line_info *linfo; | |
250 | const struct bpf_prog *prog; | |
251 | u32 i, nr_linfo; | |
252 | ||
253 | prog = env->prog; | |
254 | nr_linfo = prog->aux->nr_linfo; | |
255 | ||
256 | if (!nr_linfo || insn_off >= prog->len) | |
257 | return NULL; | |
258 | ||
259 | linfo = prog->aux->linfo; | |
260 | for (i = 1; i < nr_linfo; i++) | |
261 | if (insn_off < linfo[i].insn_off) | |
262 | break; | |
263 | ||
264 | return &linfo[i - 1]; | |
265 | } | |
266 | ||
77d2e05a MKL |
267 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
268 | va_list args) | |
cbd35700 | 269 | { |
a2a7d570 | 270 | unsigned int n; |
cbd35700 | 271 | |
a2a7d570 | 272 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
273 | |
274 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
275 | "verifier log line truncated - local buffer too short\n"); | |
276 | ||
277 | n = min(log->len_total - log->len_used - 1, n); | |
278 | log->kbuf[n] = '\0'; | |
279 | ||
8580ac94 AS |
280 | if (log->level == BPF_LOG_KERNEL) { |
281 | pr_err("BPF:%s\n", log->kbuf); | |
282 | return; | |
283 | } | |
a2a7d570 JK |
284 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) |
285 | log->len_used += n; | |
286 | else | |
287 | log->ubuf = NULL; | |
cbd35700 | 288 | } |
abe08840 | 289 | |
6f8a57cc AN |
290 | static void bpf_vlog_reset(struct bpf_verifier_log *log, u32 new_pos) |
291 | { | |
292 | char zero = 0; | |
293 | ||
294 | if (!bpf_verifier_log_needed(log)) | |
295 | return; | |
296 | ||
297 | log->len_used = new_pos; | |
298 | if (put_user(zero, log->ubuf + new_pos)) | |
299 | log->ubuf = NULL; | |
300 | } | |
301 | ||
abe08840 JO |
302 | /* log_level controls verbosity level of eBPF verifier. |
303 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
304 | * so the user can figure out what's wrong with the program | |
430e68d1 | 305 | */ |
abe08840 JO |
306 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
307 | const char *fmt, ...) | |
308 | { | |
309 | va_list args; | |
310 | ||
77d2e05a MKL |
311 | if (!bpf_verifier_log_needed(&env->log)) |
312 | return; | |
313 | ||
abe08840 | 314 | va_start(args, fmt); |
77d2e05a | 315 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
316 | va_end(args); |
317 | } | |
318 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
319 | ||
320 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
321 | { | |
77d2e05a | 322 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
323 | va_list args; |
324 | ||
77d2e05a MKL |
325 | if (!bpf_verifier_log_needed(&env->log)) |
326 | return; | |
327 | ||
abe08840 | 328 | va_start(args, fmt); |
77d2e05a | 329 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
330 | va_end(args); |
331 | } | |
cbd35700 | 332 | |
9e15db66 AS |
333 | __printf(2, 3) void bpf_log(struct bpf_verifier_log *log, |
334 | const char *fmt, ...) | |
335 | { | |
336 | va_list args; | |
337 | ||
338 | if (!bpf_verifier_log_needed(log)) | |
339 | return; | |
340 | ||
341 | va_start(args, fmt); | |
342 | bpf_verifier_vlog(log, fmt, args); | |
343 | va_end(args); | |
344 | } | |
345 | ||
d9762e84 MKL |
346 | static const char *ltrim(const char *s) |
347 | { | |
348 | while (isspace(*s)) | |
349 | s++; | |
350 | ||
351 | return s; | |
352 | } | |
353 | ||
354 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
355 | u32 insn_off, | |
356 | const char *prefix_fmt, ...) | |
357 | { | |
358 | const struct bpf_line_info *linfo; | |
359 | ||
360 | if (!bpf_verifier_log_needed(&env->log)) | |
361 | return; | |
362 | ||
363 | linfo = find_linfo(env, insn_off); | |
364 | if (!linfo || linfo == env->prev_linfo) | |
365 | return; | |
366 | ||
367 | if (prefix_fmt) { | |
368 | va_list args; | |
369 | ||
370 | va_start(args, prefix_fmt); | |
371 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
372 | va_end(args); | |
373 | } | |
374 | ||
375 | verbose(env, "%s\n", | |
376 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
377 | linfo->line_off))); | |
378 | ||
379 | env->prev_linfo = linfo; | |
380 | } | |
381 | ||
de8f3a83 DB |
382 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
383 | { | |
384 | return type == PTR_TO_PACKET || | |
385 | type == PTR_TO_PACKET_META; | |
386 | } | |
387 | ||
46f8bc92 MKL |
388 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
389 | { | |
390 | return type == PTR_TO_SOCKET || | |
655a51e5 | 391 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
392 | type == PTR_TO_TCP_SOCK || |
393 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
394 | } |
395 | ||
840b9615 JS |
396 | static bool reg_type_may_be_null(enum bpf_reg_type type) |
397 | { | |
fd978bf7 | 398 | return type == PTR_TO_MAP_VALUE_OR_NULL || |
46f8bc92 | 399 | type == PTR_TO_SOCKET_OR_NULL || |
655a51e5 | 400 | type == PTR_TO_SOCK_COMMON_OR_NULL || |
b121b341 YS |
401 | type == PTR_TO_TCP_SOCK_OR_NULL || |
402 | type == PTR_TO_BTF_ID_OR_NULL; | |
fd978bf7 JS |
403 | } |
404 | ||
d83525ca AS |
405 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
406 | { | |
407 | return reg->type == PTR_TO_MAP_VALUE && | |
408 | map_value_has_spin_lock(reg->map_ptr); | |
409 | } | |
410 | ||
cba368c1 MKL |
411 | static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type) |
412 | { | |
413 | return type == PTR_TO_SOCKET || | |
414 | type == PTR_TO_SOCKET_OR_NULL || | |
415 | type == PTR_TO_TCP_SOCK || | |
416 | type == PTR_TO_TCP_SOCK_OR_NULL; | |
417 | } | |
418 | ||
1b986589 | 419 | static bool arg_type_may_be_refcounted(enum bpf_arg_type type) |
fd978bf7 | 420 | { |
1b986589 | 421 | return type == ARG_PTR_TO_SOCK_COMMON; |
fd978bf7 JS |
422 | } |
423 | ||
424 | /* Determine whether the function releases some resources allocated by another | |
425 | * function call. The first reference type argument will be assumed to be | |
426 | * released by release_reference(). | |
427 | */ | |
428 | static bool is_release_function(enum bpf_func_id func_id) | |
429 | { | |
6acc9b43 | 430 | return func_id == BPF_FUNC_sk_release; |
840b9615 JS |
431 | } |
432 | ||
64d85290 | 433 | static bool may_be_acquire_function(enum bpf_func_id func_id) |
46f8bc92 MKL |
434 | { |
435 | return func_id == BPF_FUNC_sk_lookup_tcp || | |
edbf8c01 | 436 | func_id == BPF_FUNC_sk_lookup_udp || |
64d85290 JS |
437 | func_id == BPF_FUNC_skc_lookup_tcp || |
438 | func_id == BPF_FUNC_map_lookup_elem; | |
439 | } | |
440 | ||
441 | static bool is_acquire_function(enum bpf_func_id func_id, | |
442 | const struct bpf_map *map) | |
443 | { | |
444 | enum bpf_map_type map_type = map ? map->map_type : BPF_MAP_TYPE_UNSPEC; | |
445 | ||
446 | if (func_id == BPF_FUNC_sk_lookup_tcp || | |
447 | func_id == BPF_FUNC_sk_lookup_udp || | |
448 | func_id == BPF_FUNC_skc_lookup_tcp) | |
449 | return true; | |
450 | ||
451 | if (func_id == BPF_FUNC_map_lookup_elem && | |
452 | (map_type == BPF_MAP_TYPE_SOCKMAP || | |
453 | map_type == BPF_MAP_TYPE_SOCKHASH)) | |
454 | return true; | |
455 | ||
456 | return false; | |
46f8bc92 MKL |
457 | } |
458 | ||
1b986589 MKL |
459 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
460 | { | |
461 | return func_id == BPF_FUNC_tcp_sock || | |
462 | func_id == BPF_FUNC_sk_fullsock; | |
463 | } | |
464 | ||
17a52670 AS |
465 | /* string representation of 'enum bpf_reg_type' */ |
466 | static const char * const reg_type_str[] = { | |
467 | [NOT_INIT] = "?", | |
f1174f77 | 468 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
469 | [PTR_TO_CTX] = "ctx", |
470 | [CONST_PTR_TO_MAP] = "map_ptr", | |
471 | [PTR_TO_MAP_VALUE] = "map_value", | |
472 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 473 | [PTR_TO_STACK] = "fp", |
969bf05e | 474 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 475 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 476 | [PTR_TO_PACKET_END] = "pkt_end", |
d58e468b | 477 | [PTR_TO_FLOW_KEYS] = "flow_keys", |
c64b7983 JS |
478 | [PTR_TO_SOCKET] = "sock", |
479 | [PTR_TO_SOCKET_OR_NULL] = "sock_or_null", | |
46f8bc92 MKL |
480 | [PTR_TO_SOCK_COMMON] = "sock_common", |
481 | [PTR_TO_SOCK_COMMON_OR_NULL] = "sock_common_or_null", | |
655a51e5 MKL |
482 | [PTR_TO_TCP_SOCK] = "tcp_sock", |
483 | [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", | |
9df1c28b | 484 | [PTR_TO_TP_BUFFER] = "tp_buffer", |
fada7fdc | 485 | [PTR_TO_XDP_SOCK] = "xdp_sock", |
9e15db66 | 486 | [PTR_TO_BTF_ID] = "ptr_", |
b121b341 | 487 | [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", |
17a52670 AS |
488 | }; |
489 | ||
8efea21d EC |
490 | static char slot_type_char[] = { |
491 | [STACK_INVALID] = '?', | |
492 | [STACK_SPILL] = 'r', | |
493 | [STACK_MISC] = 'm', | |
494 | [STACK_ZERO] = '0', | |
495 | }; | |
496 | ||
4e92024a AS |
497 | static void print_liveness(struct bpf_verifier_env *env, |
498 | enum bpf_reg_liveness live) | |
499 | { | |
9242b5f5 | 500 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
501 | verbose(env, "_"); |
502 | if (live & REG_LIVE_READ) | |
503 | verbose(env, "r"); | |
504 | if (live & REG_LIVE_WRITTEN) | |
505 | verbose(env, "w"); | |
9242b5f5 AS |
506 | if (live & REG_LIVE_DONE) |
507 | verbose(env, "D"); | |
4e92024a AS |
508 | } |
509 | ||
f4d7e40a AS |
510 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
511 | const struct bpf_reg_state *reg) | |
512 | { | |
513 | struct bpf_verifier_state *cur = env->cur_state; | |
514 | ||
515 | return cur->frame[reg->frameno]; | |
516 | } | |
517 | ||
9e15db66 AS |
518 | const char *kernel_type_name(u32 id) |
519 | { | |
520 | return btf_name_by_offset(btf_vmlinux, | |
521 | btf_type_by_id(btf_vmlinux, id)->name_off); | |
522 | } | |
523 | ||
61bd5218 | 524 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 525 | const struct bpf_func_state *state) |
17a52670 | 526 | { |
f4d7e40a | 527 | const struct bpf_reg_state *reg; |
17a52670 AS |
528 | enum bpf_reg_type t; |
529 | int i; | |
530 | ||
f4d7e40a AS |
531 | if (state->frameno) |
532 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 533 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
534 | reg = &state->regs[i]; |
535 | t = reg->type; | |
17a52670 AS |
536 | if (t == NOT_INIT) |
537 | continue; | |
4e92024a AS |
538 | verbose(env, " R%d", i); |
539 | print_liveness(env, reg->live); | |
540 | verbose(env, "=%s", reg_type_str[t]); | |
b5dc0163 AS |
541 | if (t == SCALAR_VALUE && reg->precise) |
542 | verbose(env, "P"); | |
f1174f77 EC |
543 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
544 | tnum_is_const(reg->var_off)) { | |
545 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 546 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f1174f77 | 547 | } else { |
b121b341 | 548 | if (t == PTR_TO_BTF_ID || t == PTR_TO_BTF_ID_OR_NULL) |
9e15db66 | 549 | verbose(env, "%s", kernel_type_name(reg->btf_id)); |
cba368c1 MKL |
550 | verbose(env, "(id=%d", reg->id); |
551 | if (reg_type_may_be_refcounted_or_null(t)) | |
552 | verbose(env, ",ref_obj_id=%d", reg->ref_obj_id); | |
f1174f77 | 553 | if (t != SCALAR_VALUE) |
61bd5218 | 554 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 555 | if (type_is_pkt_pointer(t)) |
61bd5218 | 556 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
557 | else if (t == CONST_PTR_TO_MAP || |
558 | t == PTR_TO_MAP_VALUE || | |
559 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 560 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
561 | reg->map_ptr->key_size, |
562 | reg->map_ptr->value_size); | |
7d1238f2 EC |
563 | if (tnum_is_const(reg->var_off)) { |
564 | /* Typically an immediate SCALAR_VALUE, but | |
565 | * could be a pointer whose offset is too big | |
566 | * for reg->off | |
567 | */ | |
61bd5218 | 568 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
569 | } else { |
570 | if (reg->smin_value != reg->umin_value && | |
571 | reg->smin_value != S64_MIN) | |
61bd5218 | 572 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
573 | (long long)reg->smin_value); |
574 | if (reg->smax_value != reg->umax_value && | |
575 | reg->smax_value != S64_MAX) | |
61bd5218 | 576 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
577 | (long long)reg->smax_value); |
578 | if (reg->umin_value != 0) | |
61bd5218 | 579 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
580 | (unsigned long long)reg->umin_value); |
581 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 582 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
583 | (unsigned long long)reg->umax_value); |
584 | if (!tnum_is_unknown(reg->var_off)) { | |
585 | char tn_buf[48]; | |
f1174f77 | 586 | |
7d1238f2 | 587 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 588 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 589 | } |
3f50f132 JF |
590 | if (reg->s32_min_value != reg->smin_value && |
591 | reg->s32_min_value != S32_MIN) | |
592 | verbose(env, ",s32_min_value=%d", | |
593 | (int)(reg->s32_min_value)); | |
594 | if (reg->s32_max_value != reg->smax_value && | |
595 | reg->s32_max_value != S32_MAX) | |
596 | verbose(env, ",s32_max_value=%d", | |
597 | (int)(reg->s32_max_value)); | |
598 | if (reg->u32_min_value != reg->umin_value && | |
599 | reg->u32_min_value != U32_MIN) | |
600 | verbose(env, ",u32_min_value=%d", | |
601 | (int)(reg->u32_min_value)); | |
602 | if (reg->u32_max_value != reg->umax_value && | |
603 | reg->u32_max_value != U32_MAX) | |
604 | verbose(env, ",u32_max_value=%d", | |
605 | (int)(reg->u32_max_value)); | |
f1174f77 | 606 | } |
61bd5218 | 607 | verbose(env, ")"); |
f1174f77 | 608 | } |
17a52670 | 609 | } |
638f5b90 | 610 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
611 | char types_buf[BPF_REG_SIZE + 1]; |
612 | bool valid = false; | |
613 | int j; | |
614 | ||
615 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
616 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
617 | valid = true; | |
618 | types_buf[j] = slot_type_char[ | |
619 | state->stack[i].slot_type[j]]; | |
620 | } | |
621 | types_buf[BPF_REG_SIZE] = 0; | |
622 | if (!valid) | |
623 | continue; | |
624 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
625 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
b5dc0163 AS |
626 | if (state->stack[i].slot_type[0] == STACK_SPILL) { |
627 | reg = &state->stack[i].spilled_ptr; | |
628 | t = reg->type; | |
629 | verbose(env, "=%s", reg_type_str[t]); | |
630 | if (t == SCALAR_VALUE && reg->precise) | |
631 | verbose(env, "P"); | |
632 | if (t == SCALAR_VALUE && tnum_is_const(reg->var_off)) | |
633 | verbose(env, "%lld", reg->var_off.value + reg->off); | |
634 | } else { | |
8efea21d | 635 | verbose(env, "=%s", types_buf); |
b5dc0163 | 636 | } |
17a52670 | 637 | } |
fd978bf7 JS |
638 | if (state->acquired_refs && state->refs[0].id) { |
639 | verbose(env, " refs=%d", state->refs[0].id); | |
640 | for (i = 1; i < state->acquired_refs; i++) | |
641 | if (state->refs[i].id) | |
642 | verbose(env, ",%d", state->refs[i].id); | |
643 | } | |
61bd5218 | 644 | verbose(env, "\n"); |
17a52670 AS |
645 | } |
646 | ||
84dbf350 JS |
647 | #define COPY_STATE_FN(NAME, COUNT, FIELD, SIZE) \ |
648 | static int copy_##NAME##_state(struct bpf_func_state *dst, \ | |
649 | const struct bpf_func_state *src) \ | |
650 | { \ | |
651 | if (!src->FIELD) \ | |
652 | return 0; \ | |
653 | if (WARN_ON_ONCE(dst->COUNT < src->COUNT)) { \ | |
654 | /* internal bug, make state invalid to reject the program */ \ | |
655 | memset(dst, 0, sizeof(*dst)); \ | |
656 | return -EFAULT; \ | |
657 | } \ | |
658 | memcpy(dst->FIELD, src->FIELD, \ | |
659 | sizeof(*src->FIELD) * (src->COUNT / SIZE)); \ | |
660 | return 0; \ | |
638f5b90 | 661 | } |
fd978bf7 JS |
662 | /* copy_reference_state() */ |
663 | COPY_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
664 | /* copy_stack_state() */ |
665 | COPY_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
666 | #undef COPY_STATE_FN | |
667 | ||
668 | #define REALLOC_STATE_FN(NAME, COUNT, FIELD, SIZE) \ | |
669 | static int realloc_##NAME##_state(struct bpf_func_state *state, int size, \ | |
670 | bool copy_old) \ | |
671 | { \ | |
672 | u32 old_size = state->COUNT; \ | |
673 | struct bpf_##NAME##_state *new_##FIELD; \ | |
674 | int slot = size / SIZE; \ | |
675 | \ | |
676 | if (size <= old_size || !size) { \ | |
677 | if (copy_old) \ | |
678 | return 0; \ | |
679 | state->COUNT = slot * SIZE; \ | |
680 | if (!size && old_size) { \ | |
681 | kfree(state->FIELD); \ | |
682 | state->FIELD = NULL; \ | |
683 | } \ | |
684 | return 0; \ | |
685 | } \ | |
686 | new_##FIELD = kmalloc_array(slot, sizeof(struct bpf_##NAME##_state), \ | |
687 | GFP_KERNEL); \ | |
688 | if (!new_##FIELD) \ | |
689 | return -ENOMEM; \ | |
690 | if (copy_old) { \ | |
691 | if (state->FIELD) \ | |
692 | memcpy(new_##FIELD, state->FIELD, \ | |
693 | sizeof(*new_##FIELD) * (old_size / SIZE)); \ | |
694 | memset(new_##FIELD + old_size / SIZE, 0, \ | |
695 | sizeof(*new_##FIELD) * (size - old_size) / SIZE); \ | |
696 | } \ | |
697 | state->COUNT = slot * SIZE; \ | |
698 | kfree(state->FIELD); \ | |
699 | state->FIELD = new_##FIELD; \ | |
700 | return 0; \ | |
701 | } | |
fd978bf7 JS |
702 | /* realloc_reference_state() */ |
703 | REALLOC_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
704 | /* realloc_stack_state() */ |
705 | REALLOC_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
706 | #undef REALLOC_STATE_FN | |
638f5b90 AS |
707 | |
708 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
709 | * make it consume minimal amount of memory. check_stack_write() access from | |
f4d7e40a | 710 | * the program calls into realloc_func_state() to grow the stack size. |
84dbf350 JS |
711 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state |
712 | * which realloc_stack_state() copies over. It points to previous | |
713 | * bpf_verifier_state which is never reallocated. | |
638f5b90 | 714 | */ |
fd978bf7 JS |
715 | static int realloc_func_state(struct bpf_func_state *state, int stack_size, |
716 | int refs_size, bool copy_old) | |
638f5b90 | 717 | { |
fd978bf7 JS |
718 | int err = realloc_reference_state(state, refs_size, copy_old); |
719 | if (err) | |
720 | return err; | |
721 | return realloc_stack_state(state, stack_size, copy_old); | |
722 | } | |
723 | ||
724 | /* Acquire a pointer id from the env and update the state->refs to include | |
725 | * this new pointer reference. | |
726 | * On success, returns a valid pointer id to associate with the register | |
727 | * On failure, returns a negative errno. | |
638f5b90 | 728 | */ |
fd978bf7 | 729 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 730 | { |
fd978bf7 JS |
731 | struct bpf_func_state *state = cur_func(env); |
732 | int new_ofs = state->acquired_refs; | |
733 | int id, err; | |
734 | ||
735 | err = realloc_reference_state(state, state->acquired_refs + 1, true); | |
736 | if (err) | |
737 | return err; | |
738 | id = ++env->id_gen; | |
739 | state->refs[new_ofs].id = id; | |
740 | state->refs[new_ofs].insn_idx = insn_idx; | |
638f5b90 | 741 | |
fd978bf7 JS |
742 | return id; |
743 | } | |
744 | ||
745 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 746 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
747 | { |
748 | int i, last_idx; | |
749 | ||
fd978bf7 JS |
750 | last_idx = state->acquired_refs - 1; |
751 | for (i = 0; i < state->acquired_refs; i++) { | |
752 | if (state->refs[i].id == ptr_id) { | |
753 | if (last_idx && i != last_idx) | |
754 | memcpy(&state->refs[i], &state->refs[last_idx], | |
755 | sizeof(*state->refs)); | |
756 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
757 | state->acquired_refs--; | |
638f5b90 | 758 | return 0; |
638f5b90 | 759 | } |
638f5b90 | 760 | } |
46f8bc92 | 761 | return -EINVAL; |
fd978bf7 JS |
762 | } |
763 | ||
764 | static int transfer_reference_state(struct bpf_func_state *dst, | |
765 | struct bpf_func_state *src) | |
766 | { | |
767 | int err = realloc_reference_state(dst, src->acquired_refs, false); | |
768 | if (err) | |
769 | return err; | |
770 | err = copy_reference_state(dst, src); | |
771 | if (err) | |
772 | return err; | |
638f5b90 AS |
773 | return 0; |
774 | } | |
775 | ||
f4d7e40a AS |
776 | static void free_func_state(struct bpf_func_state *state) |
777 | { | |
5896351e AS |
778 | if (!state) |
779 | return; | |
fd978bf7 | 780 | kfree(state->refs); |
f4d7e40a AS |
781 | kfree(state->stack); |
782 | kfree(state); | |
783 | } | |
784 | ||
b5dc0163 AS |
785 | static void clear_jmp_history(struct bpf_verifier_state *state) |
786 | { | |
787 | kfree(state->jmp_history); | |
788 | state->jmp_history = NULL; | |
789 | state->jmp_history_cnt = 0; | |
790 | } | |
791 | ||
1969db47 AS |
792 | static void free_verifier_state(struct bpf_verifier_state *state, |
793 | bool free_self) | |
638f5b90 | 794 | { |
f4d7e40a AS |
795 | int i; |
796 | ||
797 | for (i = 0; i <= state->curframe; i++) { | |
798 | free_func_state(state->frame[i]); | |
799 | state->frame[i] = NULL; | |
800 | } | |
b5dc0163 | 801 | clear_jmp_history(state); |
1969db47 AS |
802 | if (free_self) |
803 | kfree(state); | |
638f5b90 AS |
804 | } |
805 | ||
806 | /* copy verifier state from src to dst growing dst stack space | |
807 | * when necessary to accommodate larger src stack | |
808 | */ | |
f4d7e40a AS |
809 | static int copy_func_state(struct bpf_func_state *dst, |
810 | const struct bpf_func_state *src) | |
638f5b90 AS |
811 | { |
812 | int err; | |
813 | ||
fd978bf7 JS |
814 | err = realloc_func_state(dst, src->allocated_stack, src->acquired_refs, |
815 | false); | |
816 | if (err) | |
817 | return err; | |
818 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); | |
819 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
820 | if (err) |
821 | return err; | |
638f5b90 AS |
822 | return copy_stack_state(dst, src); |
823 | } | |
824 | ||
f4d7e40a AS |
825 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
826 | const struct bpf_verifier_state *src) | |
827 | { | |
828 | struct bpf_func_state *dst; | |
b5dc0163 | 829 | u32 jmp_sz = sizeof(struct bpf_idx_pair) * src->jmp_history_cnt; |
f4d7e40a AS |
830 | int i, err; |
831 | ||
b5dc0163 AS |
832 | if (dst_state->jmp_history_cnt < src->jmp_history_cnt) { |
833 | kfree(dst_state->jmp_history); | |
834 | dst_state->jmp_history = kmalloc(jmp_sz, GFP_USER); | |
835 | if (!dst_state->jmp_history) | |
836 | return -ENOMEM; | |
837 | } | |
838 | memcpy(dst_state->jmp_history, src->jmp_history, jmp_sz); | |
839 | dst_state->jmp_history_cnt = src->jmp_history_cnt; | |
840 | ||
f4d7e40a AS |
841 | /* if dst has more stack frames then src frame, free them */ |
842 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
843 | free_func_state(dst_state->frame[i]); | |
844 | dst_state->frame[i] = NULL; | |
845 | } | |
979d63d5 | 846 | dst_state->speculative = src->speculative; |
f4d7e40a | 847 | dst_state->curframe = src->curframe; |
d83525ca | 848 | dst_state->active_spin_lock = src->active_spin_lock; |
2589726d AS |
849 | dst_state->branches = src->branches; |
850 | dst_state->parent = src->parent; | |
b5dc0163 AS |
851 | dst_state->first_insn_idx = src->first_insn_idx; |
852 | dst_state->last_insn_idx = src->last_insn_idx; | |
f4d7e40a AS |
853 | for (i = 0; i <= src->curframe; i++) { |
854 | dst = dst_state->frame[i]; | |
855 | if (!dst) { | |
856 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
857 | if (!dst) | |
858 | return -ENOMEM; | |
859 | dst_state->frame[i] = dst; | |
860 | } | |
861 | err = copy_func_state(dst, src->frame[i]); | |
862 | if (err) | |
863 | return err; | |
864 | } | |
865 | return 0; | |
866 | } | |
867 | ||
2589726d AS |
868 | static void update_branch_counts(struct bpf_verifier_env *env, struct bpf_verifier_state *st) |
869 | { | |
870 | while (st) { | |
871 | u32 br = --st->branches; | |
872 | ||
873 | /* WARN_ON(br > 1) technically makes sense here, | |
874 | * but see comment in push_stack(), hence: | |
875 | */ | |
876 | WARN_ONCE((int)br < 0, | |
877 | "BUG update_branch_counts:branches_to_explore=%d\n", | |
878 | br); | |
879 | if (br) | |
880 | break; | |
881 | st = st->parent; | |
882 | } | |
883 | } | |
884 | ||
638f5b90 | 885 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
6f8a57cc | 886 | int *insn_idx, bool pop_log) |
638f5b90 AS |
887 | { |
888 | struct bpf_verifier_state *cur = env->cur_state; | |
889 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
890 | int err; | |
17a52670 AS |
891 | |
892 | if (env->head == NULL) | |
638f5b90 | 893 | return -ENOENT; |
17a52670 | 894 | |
638f5b90 AS |
895 | if (cur) { |
896 | err = copy_verifier_state(cur, &head->st); | |
897 | if (err) | |
898 | return err; | |
899 | } | |
6f8a57cc AN |
900 | if (pop_log) |
901 | bpf_vlog_reset(&env->log, head->log_pos); | |
638f5b90 AS |
902 | if (insn_idx) |
903 | *insn_idx = head->insn_idx; | |
17a52670 | 904 | if (prev_insn_idx) |
638f5b90 AS |
905 | *prev_insn_idx = head->prev_insn_idx; |
906 | elem = head->next; | |
1969db47 | 907 | free_verifier_state(&head->st, false); |
638f5b90 | 908 | kfree(head); |
17a52670 AS |
909 | env->head = elem; |
910 | env->stack_size--; | |
638f5b90 | 911 | return 0; |
17a52670 AS |
912 | } |
913 | ||
58e2af8b | 914 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
915 | int insn_idx, int prev_insn_idx, |
916 | bool speculative) | |
17a52670 | 917 | { |
638f5b90 | 918 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 919 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 920 | int err; |
17a52670 | 921 | |
638f5b90 | 922 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
923 | if (!elem) |
924 | goto err; | |
925 | ||
17a52670 AS |
926 | elem->insn_idx = insn_idx; |
927 | elem->prev_insn_idx = prev_insn_idx; | |
928 | elem->next = env->head; | |
6f8a57cc | 929 | elem->log_pos = env->log.len_used; |
17a52670 AS |
930 | env->head = elem; |
931 | env->stack_size++; | |
1969db47 AS |
932 | err = copy_verifier_state(&elem->st, cur); |
933 | if (err) | |
934 | goto err; | |
979d63d5 | 935 | elem->st.speculative |= speculative; |
b285fcb7 AS |
936 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
937 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
938 | env->stack_size); | |
17a52670 AS |
939 | goto err; |
940 | } | |
2589726d AS |
941 | if (elem->st.parent) { |
942 | ++elem->st.parent->branches; | |
943 | /* WARN_ON(branches > 2) technically makes sense here, | |
944 | * but | |
945 | * 1. speculative states will bump 'branches' for non-branch | |
946 | * instructions | |
947 | * 2. is_state_visited() heuristics may decide not to create | |
948 | * a new state for a sequence of branches and all such current | |
949 | * and cloned states will be pointing to a single parent state | |
950 | * which might have large 'branches' count. | |
951 | */ | |
952 | } | |
17a52670 AS |
953 | return &elem->st; |
954 | err: | |
5896351e AS |
955 | free_verifier_state(env->cur_state, true); |
956 | env->cur_state = NULL; | |
17a52670 | 957 | /* pop all elements and return */ |
6f8a57cc | 958 | while (!pop_stack(env, NULL, NULL, false)); |
17a52670 AS |
959 | return NULL; |
960 | } | |
961 | ||
962 | #define CALLER_SAVED_REGS 6 | |
963 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
964 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
965 | }; | |
966 | ||
f54c7898 DB |
967 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
968 | struct bpf_reg_state *reg); | |
f1174f77 | 969 | |
b03c9f9f EC |
970 | /* Mark the unknown part of a register (variable offset or scalar value) as |
971 | * known to have the value @imm. | |
972 | */ | |
973 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
974 | { | |
a9c676bc AS |
975 | /* Clear id, off, and union(map_ptr, range) */ |
976 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
977 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
b03c9f9f EC |
978 | reg->var_off = tnum_const(imm); |
979 | reg->smin_value = (s64)imm; | |
980 | reg->smax_value = (s64)imm; | |
981 | reg->umin_value = imm; | |
982 | reg->umax_value = imm; | |
3f50f132 JF |
983 | |
984 | reg->s32_min_value = (s32)imm; | |
985 | reg->s32_max_value = (s32)imm; | |
986 | reg->u32_min_value = (u32)imm; | |
987 | reg->u32_max_value = (u32)imm; | |
988 | } | |
989 | ||
990 | static void __mark_reg32_known(struct bpf_reg_state *reg, u64 imm) | |
991 | { | |
992 | reg->var_off = tnum_const_subreg(reg->var_off, imm); | |
993 | reg->s32_min_value = (s32)imm; | |
994 | reg->s32_max_value = (s32)imm; | |
995 | reg->u32_min_value = (u32)imm; | |
996 | reg->u32_max_value = (u32)imm; | |
b03c9f9f EC |
997 | } |
998 | ||
f1174f77 EC |
999 | /* Mark the 'variable offset' part of a register as zero. This should be |
1000 | * used only on registers holding a pointer type. | |
1001 | */ | |
1002 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 1003 | { |
b03c9f9f | 1004 | __mark_reg_known(reg, 0); |
f1174f77 | 1005 | } |
a9789ef9 | 1006 | |
cc2b14d5 AS |
1007 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
1008 | { | |
1009 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
1010 | reg->type = SCALAR_VALUE; |
1011 | } | |
1012 | ||
61bd5218 JK |
1013 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
1014 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1015 | { |
1016 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1017 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
1018 | /* Something bad happened, let's kill all regs */ |
1019 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
f54c7898 | 1020 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1021 | return; |
1022 | } | |
1023 | __mark_reg_known_zero(regs + regno); | |
1024 | } | |
1025 | ||
de8f3a83 DB |
1026 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
1027 | { | |
1028 | return type_is_pkt_pointer(reg->type); | |
1029 | } | |
1030 | ||
1031 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
1032 | { | |
1033 | return reg_is_pkt_pointer(reg) || | |
1034 | reg->type == PTR_TO_PACKET_END; | |
1035 | } | |
1036 | ||
1037 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
1038 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
1039 | enum bpf_reg_type which) | |
1040 | { | |
1041 | /* The register can already have a range from prior markings. | |
1042 | * This is fine as long as it hasn't been advanced from its | |
1043 | * origin. | |
1044 | */ | |
1045 | return reg->type == which && | |
1046 | reg->id == 0 && | |
1047 | reg->off == 0 && | |
1048 | tnum_equals_const(reg->var_off, 0); | |
1049 | } | |
1050 | ||
3f50f132 JF |
1051 | /* Reset the min/max bounds of a register */ |
1052 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
1053 | { | |
1054 | reg->smin_value = S64_MIN; | |
1055 | reg->smax_value = S64_MAX; | |
1056 | reg->umin_value = 0; | |
1057 | reg->umax_value = U64_MAX; | |
1058 | ||
1059 | reg->s32_min_value = S32_MIN; | |
1060 | reg->s32_max_value = S32_MAX; | |
1061 | reg->u32_min_value = 0; | |
1062 | reg->u32_max_value = U32_MAX; | |
1063 | } | |
1064 | ||
1065 | static void __mark_reg64_unbounded(struct bpf_reg_state *reg) | |
1066 | { | |
1067 | reg->smin_value = S64_MIN; | |
1068 | reg->smax_value = S64_MAX; | |
1069 | reg->umin_value = 0; | |
1070 | reg->umax_value = U64_MAX; | |
1071 | } | |
1072 | ||
1073 | static void __mark_reg32_unbounded(struct bpf_reg_state *reg) | |
1074 | { | |
1075 | reg->s32_min_value = S32_MIN; | |
1076 | reg->s32_max_value = S32_MAX; | |
1077 | reg->u32_min_value = 0; | |
1078 | reg->u32_max_value = U32_MAX; | |
1079 | } | |
1080 | ||
1081 | static void __update_reg32_bounds(struct bpf_reg_state *reg) | |
1082 | { | |
1083 | struct tnum var32_off = tnum_subreg(reg->var_off); | |
1084 | ||
1085 | /* min signed is max(sign bit) | min(other bits) */ | |
1086 | reg->s32_min_value = max_t(s32, reg->s32_min_value, | |
1087 | var32_off.value | (var32_off.mask & S32_MIN)); | |
1088 | /* max signed is min(sign bit) | max(other bits) */ | |
1089 | reg->s32_max_value = min_t(s32, reg->s32_max_value, | |
1090 | var32_off.value | (var32_off.mask & S32_MAX)); | |
1091 | reg->u32_min_value = max_t(u32, reg->u32_min_value, (u32)var32_off.value); | |
1092 | reg->u32_max_value = min(reg->u32_max_value, | |
1093 | (u32)(var32_off.value | var32_off.mask)); | |
1094 | } | |
1095 | ||
1096 | static void __update_reg64_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1097 | { |
1098 | /* min signed is max(sign bit) | min(other bits) */ | |
1099 | reg->smin_value = max_t(s64, reg->smin_value, | |
1100 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
1101 | /* max signed is min(sign bit) | max(other bits) */ | |
1102 | reg->smax_value = min_t(s64, reg->smax_value, | |
1103 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
1104 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
1105 | reg->umax_value = min(reg->umax_value, | |
1106 | reg->var_off.value | reg->var_off.mask); | |
1107 | } | |
1108 | ||
3f50f132 JF |
1109 | static void __update_reg_bounds(struct bpf_reg_state *reg) |
1110 | { | |
1111 | __update_reg32_bounds(reg); | |
1112 | __update_reg64_bounds(reg); | |
1113 | } | |
1114 | ||
b03c9f9f | 1115 | /* Uses signed min/max values to inform unsigned, and vice-versa */ |
3f50f132 JF |
1116 | static void __reg32_deduce_bounds(struct bpf_reg_state *reg) |
1117 | { | |
1118 | /* Learn sign from signed bounds. | |
1119 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1120 | * are the same, so combine. This works even in the negative case, e.g. | |
1121 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1122 | */ | |
1123 | if (reg->s32_min_value >= 0 || reg->s32_max_value < 0) { | |
1124 | reg->s32_min_value = reg->u32_min_value = | |
1125 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1126 | reg->s32_max_value = reg->u32_max_value = | |
1127 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1128 | return; | |
1129 | } | |
1130 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1131 | * boundary, so we must be careful. | |
1132 | */ | |
1133 | if ((s32)reg->u32_max_value >= 0) { | |
1134 | /* Positive. We can't learn anything from the smin, but smax | |
1135 | * is positive, hence safe. | |
1136 | */ | |
1137 | reg->s32_min_value = reg->u32_min_value; | |
1138 | reg->s32_max_value = reg->u32_max_value = | |
1139 | min_t(u32, reg->s32_max_value, reg->u32_max_value); | |
1140 | } else if ((s32)reg->u32_min_value < 0) { | |
1141 | /* Negative. We can't learn anything from the smax, but smin | |
1142 | * is negative, hence safe. | |
1143 | */ | |
1144 | reg->s32_min_value = reg->u32_min_value = | |
1145 | max_t(u32, reg->s32_min_value, reg->u32_min_value); | |
1146 | reg->s32_max_value = reg->u32_max_value; | |
1147 | } | |
1148 | } | |
1149 | ||
1150 | static void __reg64_deduce_bounds(struct bpf_reg_state *reg) | |
b03c9f9f EC |
1151 | { |
1152 | /* Learn sign from signed bounds. | |
1153 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
1154 | * are the same, so combine. This works even in the negative case, e.g. | |
1155 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
1156 | */ | |
1157 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
1158 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1159 | reg->umin_value); | |
1160 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1161 | reg->umax_value); | |
1162 | return; | |
1163 | } | |
1164 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
1165 | * boundary, so we must be careful. | |
1166 | */ | |
1167 | if ((s64)reg->umax_value >= 0) { | |
1168 | /* Positive. We can't learn anything from the smin, but smax | |
1169 | * is positive, hence safe. | |
1170 | */ | |
1171 | reg->smin_value = reg->umin_value; | |
1172 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
1173 | reg->umax_value); | |
1174 | } else if ((s64)reg->umin_value < 0) { | |
1175 | /* Negative. We can't learn anything from the smax, but smin | |
1176 | * is negative, hence safe. | |
1177 | */ | |
1178 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
1179 | reg->umin_value); | |
1180 | reg->smax_value = reg->umax_value; | |
1181 | } | |
1182 | } | |
1183 | ||
3f50f132 JF |
1184 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) |
1185 | { | |
1186 | __reg32_deduce_bounds(reg); | |
1187 | __reg64_deduce_bounds(reg); | |
1188 | } | |
1189 | ||
b03c9f9f EC |
1190 | /* Attempts to improve var_off based on unsigned min/max information */ |
1191 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
1192 | { | |
3f50f132 JF |
1193 | struct tnum var64_off = tnum_intersect(reg->var_off, |
1194 | tnum_range(reg->umin_value, | |
1195 | reg->umax_value)); | |
1196 | struct tnum var32_off = tnum_intersect(tnum_subreg(reg->var_off), | |
1197 | tnum_range(reg->u32_min_value, | |
1198 | reg->u32_max_value)); | |
1199 | ||
1200 | reg->var_off = tnum_or(tnum_clear_subreg(var64_off), var32_off); | |
b03c9f9f EC |
1201 | } |
1202 | ||
3f50f132 | 1203 | static void __reg_assign_32_into_64(struct bpf_reg_state *reg) |
b03c9f9f | 1204 | { |
3f50f132 JF |
1205 | reg->umin_value = reg->u32_min_value; |
1206 | reg->umax_value = reg->u32_max_value; | |
1207 | /* Attempt to pull 32-bit signed bounds into 64-bit bounds | |
1208 | * but must be positive otherwise set to worse case bounds | |
1209 | * and refine later from tnum. | |
1210 | */ | |
1211 | if (reg->s32_min_value > 0) | |
1212 | reg->smin_value = reg->s32_min_value; | |
1213 | else | |
1214 | reg->smin_value = 0; | |
1215 | if (reg->s32_max_value > 0) | |
1216 | reg->smax_value = reg->s32_max_value; | |
1217 | else | |
1218 | reg->smax_value = U32_MAX; | |
1219 | } | |
1220 | ||
1221 | static void __reg_combine_32_into_64(struct bpf_reg_state *reg) | |
1222 | { | |
1223 | /* special case when 64-bit register has upper 32-bit register | |
1224 | * zeroed. Typically happens after zext or <<32, >>32 sequence | |
1225 | * allowing us to use 32-bit bounds directly, | |
1226 | */ | |
1227 | if (tnum_equals_const(tnum_clear_subreg(reg->var_off), 0)) { | |
1228 | __reg_assign_32_into_64(reg); | |
1229 | } else { | |
1230 | /* Otherwise the best we can do is push lower 32bit known and | |
1231 | * unknown bits into register (var_off set from jmp logic) | |
1232 | * then learn as much as possible from the 64-bit tnum | |
1233 | * known and unknown bits. The previous smin/smax bounds are | |
1234 | * invalid here because of jmp32 compare so mark them unknown | |
1235 | * so they do not impact tnum bounds calculation. | |
1236 | */ | |
1237 | __mark_reg64_unbounded(reg); | |
1238 | __update_reg_bounds(reg); | |
1239 | } | |
1240 | ||
1241 | /* Intersecting with the old var_off might have improved our bounds | |
1242 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1243 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1244 | */ | |
1245 | __reg_deduce_bounds(reg); | |
1246 | __reg_bound_offset(reg); | |
1247 | __update_reg_bounds(reg); | |
1248 | } | |
1249 | ||
1250 | static bool __reg64_bound_s32(s64 a) | |
1251 | { | |
1252 | if (a > S32_MIN && a < S32_MAX) | |
1253 | return true; | |
1254 | return false; | |
1255 | } | |
1256 | ||
1257 | static bool __reg64_bound_u32(u64 a) | |
1258 | { | |
1259 | if (a > U32_MIN && a < U32_MAX) | |
1260 | return true; | |
1261 | return false; | |
1262 | } | |
1263 | ||
1264 | static void __reg_combine_64_into_32(struct bpf_reg_state *reg) | |
1265 | { | |
1266 | __mark_reg32_unbounded(reg); | |
1267 | ||
1268 | if (__reg64_bound_s32(reg->smin_value)) | |
1269 | reg->s32_min_value = (s32)reg->smin_value; | |
1270 | if (__reg64_bound_s32(reg->smax_value)) | |
1271 | reg->s32_max_value = (s32)reg->smax_value; | |
1272 | if (__reg64_bound_u32(reg->umin_value)) | |
1273 | reg->u32_min_value = (u32)reg->umin_value; | |
1274 | if (__reg64_bound_u32(reg->umax_value)) | |
1275 | reg->u32_max_value = (u32)reg->umax_value; | |
1276 | ||
1277 | /* Intersecting with the old var_off might have improved our bounds | |
1278 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
1279 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
1280 | */ | |
1281 | __reg_deduce_bounds(reg); | |
1282 | __reg_bound_offset(reg); | |
1283 | __update_reg_bounds(reg); | |
b03c9f9f EC |
1284 | } |
1285 | ||
f1174f77 | 1286 | /* Mark a register as having a completely unknown (scalar) value. */ |
f54c7898 DB |
1287 | static void __mark_reg_unknown(const struct bpf_verifier_env *env, |
1288 | struct bpf_reg_state *reg) | |
f1174f77 | 1289 | { |
a9c676bc AS |
1290 | /* |
1291 | * Clear type, id, off, and union(map_ptr, range) and | |
1292 | * padding between 'type' and union | |
1293 | */ | |
1294 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 1295 | reg->type = SCALAR_VALUE; |
f1174f77 | 1296 | reg->var_off = tnum_unknown; |
f4d7e40a | 1297 | reg->frameno = 0; |
2c78ee89 | 1298 | reg->precise = env->subprog_cnt > 1 || !env->bpf_capable; |
b03c9f9f | 1299 | __mark_reg_unbounded(reg); |
f1174f77 EC |
1300 | } |
1301 | ||
61bd5218 JK |
1302 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
1303 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1304 | { |
1305 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1306 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
1307 | /* Something bad happened, let's kill all regs except FP */ |
1308 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1309 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1310 | return; |
1311 | } | |
f54c7898 | 1312 | __mark_reg_unknown(env, regs + regno); |
f1174f77 EC |
1313 | } |
1314 | ||
f54c7898 DB |
1315 | static void __mark_reg_not_init(const struct bpf_verifier_env *env, |
1316 | struct bpf_reg_state *reg) | |
f1174f77 | 1317 | { |
f54c7898 | 1318 | __mark_reg_unknown(env, reg); |
f1174f77 EC |
1319 | reg->type = NOT_INIT; |
1320 | } | |
1321 | ||
61bd5218 JK |
1322 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
1323 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
1324 | { |
1325 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 1326 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
1327 | /* Something bad happened, let's kill all regs except FP */ |
1328 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f54c7898 | 1329 | __mark_reg_not_init(env, regs + regno); |
f1174f77 EC |
1330 | return; |
1331 | } | |
f54c7898 | 1332 | __mark_reg_not_init(env, regs + regno); |
a9789ef9 DB |
1333 | } |
1334 | ||
5327ed3d | 1335 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 1336 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 1337 | struct bpf_func_state *state) |
17a52670 | 1338 | { |
f4d7e40a | 1339 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
1340 | int i; |
1341 | ||
dc503a8a | 1342 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 1343 | mark_reg_not_init(env, regs, i); |
dc503a8a | 1344 | regs[i].live = REG_LIVE_NONE; |
679c782d | 1345 | regs[i].parent = NULL; |
5327ed3d | 1346 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 1347 | } |
17a52670 AS |
1348 | |
1349 | /* frame pointer */ | |
f1174f77 | 1350 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 1351 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 1352 | regs[BPF_REG_FP].frameno = state->frameno; |
6760bf2d DB |
1353 | } |
1354 | ||
f4d7e40a AS |
1355 | #define BPF_MAIN_FUNC (-1) |
1356 | static void init_func_state(struct bpf_verifier_env *env, | |
1357 | struct bpf_func_state *state, | |
1358 | int callsite, int frameno, int subprogno) | |
1359 | { | |
1360 | state->callsite = callsite; | |
1361 | state->frameno = frameno; | |
1362 | state->subprogno = subprogno; | |
1363 | init_reg_state(env, state); | |
1364 | } | |
1365 | ||
17a52670 AS |
1366 | enum reg_arg_type { |
1367 | SRC_OP, /* register is used as source operand */ | |
1368 | DST_OP, /* register is used as destination operand */ | |
1369 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1370 | }; | |
1371 | ||
cc8b0b92 AS |
1372 | static int cmp_subprogs(const void *a, const void *b) |
1373 | { | |
9c8105bd JW |
1374 | return ((struct bpf_subprog_info *)a)->start - |
1375 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1376 | } |
1377 | ||
1378 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1379 | { | |
9c8105bd | 1380 | struct bpf_subprog_info *p; |
cc8b0b92 | 1381 | |
9c8105bd JW |
1382 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1383 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1384 | if (!p) |
1385 | return -ENOENT; | |
9c8105bd | 1386 | return p - env->subprog_info; |
cc8b0b92 AS |
1387 | |
1388 | } | |
1389 | ||
1390 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1391 | { | |
1392 | int insn_cnt = env->prog->len; | |
1393 | int ret; | |
1394 | ||
1395 | if (off >= insn_cnt || off < 0) { | |
1396 | verbose(env, "call to invalid destination\n"); | |
1397 | return -EINVAL; | |
1398 | } | |
1399 | ret = find_subprog(env, off); | |
1400 | if (ret >= 0) | |
1401 | return 0; | |
4cb3d99c | 1402 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1403 | verbose(env, "too many subprograms\n"); |
1404 | return -E2BIG; | |
1405 | } | |
9c8105bd JW |
1406 | env->subprog_info[env->subprog_cnt++].start = off; |
1407 | sort(env->subprog_info, env->subprog_cnt, | |
1408 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
cc8b0b92 AS |
1409 | return 0; |
1410 | } | |
1411 | ||
1412 | static int check_subprogs(struct bpf_verifier_env *env) | |
1413 | { | |
1414 | int i, ret, subprog_start, subprog_end, off, cur_subprog = 0; | |
9c8105bd | 1415 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 AS |
1416 | struct bpf_insn *insn = env->prog->insnsi; |
1417 | int insn_cnt = env->prog->len; | |
1418 | ||
f910cefa JW |
1419 | /* Add entry function. */ |
1420 | ret = add_subprog(env, 0); | |
1421 | if (ret < 0) | |
1422 | return ret; | |
1423 | ||
cc8b0b92 AS |
1424 | /* determine subprog starts. The end is one before the next starts */ |
1425 | for (i = 0; i < insn_cnt; i++) { | |
1426 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
1427 | continue; | |
1428 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
1429 | continue; | |
2c78ee89 AS |
1430 | if (!env->bpf_capable) { |
1431 | verbose(env, | |
1432 | "function calls to other bpf functions are allowed for CAP_BPF and CAP_SYS_ADMIN\n"); | |
cc8b0b92 AS |
1433 | return -EPERM; |
1434 | } | |
cc8b0b92 AS |
1435 | ret = add_subprog(env, i + insn[i].imm + 1); |
1436 | if (ret < 0) | |
1437 | return ret; | |
1438 | } | |
1439 | ||
4cb3d99c JW |
1440 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
1441 | * logic. 'subprog_cnt' should not be increased. | |
1442 | */ | |
1443 | subprog[env->subprog_cnt].start = insn_cnt; | |
1444 | ||
06ee7115 | 1445 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 1446 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 1447 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 AS |
1448 | |
1449 | /* now check that all jumps are within the same subprog */ | |
4cb3d99c JW |
1450 | subprog_start = subprog[cur_subprog].start; |
1451 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
1452 | for (i = 0; i < insn_cnt; i++) { |
1453 | u8 code = insn[i].code; | |
1454 | ||
092ed096 | 1455 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
1456 | goto next; |
1457 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
1458 | goto next; | |
1459 | off = i + insn[i].off + 1; | |
1460 | if (off < subprog_start || off >= subprog_end) { | |
1461 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
1462 | return -EINVAL; | |
1463 | } | |
1464 | next: | |
1465 | if (i == subprog_end - 1) { | |
1466 | /* to avoid fall-through from one subprog into another | |
1467 | * the last insn of the subprog should be either exit | |
1468 | * or unconditional jump back | |
1469 | */ | |
1470 | if (code != (BPF_JMP | BPF_EXIT) && | |
1471 | code != (BPF_JMP | BPF_JA)) { | |
1472 | verbose(env, "last insn is not an exit or jmp\n"); | |
1473 | return -EINVAL; | |
1474 | } | |
1475 | subprog_start = subprog_end; | |
4cb3d99c JW |
1476 | cur_subprog++; |
1477 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 1478 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
1479 | } |
1480 | } | |
1481 | return 0; | |
1482 | } | |
1483 | ||
679c782d EC |
1484 | /* Parentage chain of this register (or stack slot) should take care of all |
1485 | * issues like callee-saved registers, stack slot allocation time, etc. | |
1486 | */ | |
f4d7e40a | 1487 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 1488 | const struct bpf_reg_state *state, |
5327ed3d | 1489 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
1490 | { |
1491 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 1492 | int cnt = 0; |
dc503a8a EC |
1493 | |
1494 | while (parent) { | |
1495 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 1496 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 1497 | break; |
9242b5f5 AS |
1498 | if (parent->live & REG_LIVE_DONE) { |
1499 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
1500 | reg_type_str[parent->type], | |
1501 | parent->var_off.value, parent->off); | |
1502 | return -EFAULT; | |
1503 | } | |
5327ed3d JW |
1504 | /* The first condition is more likely to be true than the |
1505 | * second, checked it first. | |
1506 | */ | |
1507 | if ((parent->live & REG_LIVE_READ) == flag || | |
1508 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
1509 | /* The parentage chain never changes and |
1510 | * this parent was already marked as LIVE_READ. | |
1511 | * There is no need to keep walking the chain again and | |
1512 | * keep re-marking all parents as LIVE_READ. | |
1513 | * This case happens when the same register is read | |
1514 | * multiple times without writes into it in-between. | |
5327ed3d JW |
1515 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
1516 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
1517 | */ |
1518 | break; | |
dc503a8a | 1519 | /* ... then we depend on parent's value */ |
5327ed3d JW |
1520 | parent->live |= flag; |
1521 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
1522 | if (flag == REG_LIVE_READ64) | |
1523 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
1524 | state = parent; |
1525 | parent = state->parent; | |
f4d7e40a | 1526 | writes = true; |
06ee7115 | 1527 | cnt++; |
dc503a8a | 1528 | } |
06ee7115 AS |
1529 | |
1530 | if (env->longest_mark_read_walk < cnt) | |
1531 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 1532 | return 0; |
dc503a8a EC |
1533 | } |
1534 | ||
5327ed3d JW |
1535 | /* This function is supposed to be used by the following 32-bit optimization |
1536 | * code only. It returns TRUE if the source or destination register operates | |
1537 | * on 64-bit, otherwise return FALSE. | |
1538 | */ | |
1539 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
1540 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
1541 | { | |
1542 | u8 code, class, op; | |
1543 | ||
1544 | code = insn->code; | |
1545 | class = BPF_CLASS(code); | |
1546 | op = BPF_OP(code); | |
1547 | if (class == BPF_JMP) { | |
1548 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
1549 | * conservatively. | |
1550 | */ | |
1551 | if (op == BPF_EXIT) | |
1552 | return true; | |
1553 | if (op == BPF_CALL) { | |
1554 | /* BPF to BPF call will reach here because of marking | |
1555 | * caller saved clobber with DST_OP_NO_MARK for which we | |
1556 | * don't care the register def because they are anyway | |
1557 | * marked as NOT_INIT already. | |
1558 | */ | |
1559 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
1560 | return false; | |
1561 | /* Helper call will reach here because of arg type | |
1562 | * check, conservatively return TRUE. | |
1563 | */ | |
1564 | if (t == SRC_OP) | |
1565 | return true; | |
1566 | ||
1567 | return false; | |
1568 | } | |
1569 | } | |
1570 | ||
1571 | if (class == BPF_ALU64 || class == BPF_JMP || | |
1572 | /* BPF_END always use BPF_ALU class. */ | |
1573 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) | |
1574 | return true; | |
1575 | ||
1576 | if (class == BPF_ALU || class == BPF_JMP32) | |
1577 | return false; | |
1578 | ||
1579 | if (class == BPF_LDX) { | |
1580 | if (t != SRC_OP) | |
1581 | return BPF_SIZE(code) == BPF_DW; | |
1582 | /* LDX source must be ptr. */ | |
1583 | return true; | |
1584 | } | |
1585 | ||
1586 | if (class == BPF_STX) { | |
1587 | if (reg->type != SCALAR_VALUE) | |
1588 | return true; | |
1589 | return BPF_SIZE(code) == BPF_DW; | |
1590 | } | |
1591 | ||
1592 | if (class == BPF_LD) { | |
1593 | u8 mode = BPF_MODE(code); | |
1594 | ||
1595 | /* LD_IMM64 */ | |
1596 | if (mode == BPF_IMM) | |
1597 | return true; | |
1598 | ||
1599 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
1600 | if (t != SRC_OP) | |
1601 | return false; | |
1602 | ||
1603 | /* Implicit ctx ptr. */ | |
1604 | if (regno == BPF_REG_6) | |
1605 | return true; | |
1606 | ||
1607 | /* Explicit source could be any width. */ | |
1608 | return true; | |
1609 | } | |
1610 | ||
1611 | if (class == BPF_ST) | |
1612 | /* The only source register for BPF_ST is a ptr. */ | |
1613 | return true; | |
1614 | ||
1615 | /* Conservatively return true at default. */ | |
1616 | return true; | |
1617 | } | |
1618 | ||
b325fbca JW |
1619 | /* Return TRUE if INSN doesn't have explicit value define. */ |
1620 | static bool insn_no_def(struct bpf_insn *insn) | |
1621 | { | |
1622 | u8 class = BPF_CLASS(insn->code); | |
1623 | ||
1624 | return (class == BPF_JMP || class == BPF_JMP32 || | |
1625 | class == BPF_STX || class == BPF_ST); | |
1626 | } | |
1627 | ||
1628 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
1629 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
1630 | { | |
1631 | if (insn_no_def(insn)) | |
1632 | return false; | |
1633 | ||
1634 | return !is_reg64(env, insn, insn->dst_reg, NULL, DST_OP); | |
1635 | } | |
1636 | ||
5327ed3d JW |
1637 | static void mark_insn_zext(struct bpf_verifier_env *env, |
1638 | struct bpf_reg_state *reg) | |
1639 | { | |
1640 | s32 def_idx = reg->subreg_def; | |
1641 | ||
1642 | if (def_idx == DEF_NOT_SUBREG) | |
1643 | return; | |
1644 | ||
1645 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
1646 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
1647 | reg->subreg_def = DEF_NOT_SUBREG; | |
1648 | } | |
1649 | ||
dc503a8a | 1650 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
1651 | enum reg_arg_type t) |
1652 | { | |
f4d7e40a AS |
1653 | struct bpf_verifier_state *vstate = env->cur_state; |
1654 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 1655 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 1656 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 1657 | bool rw64; |
dc503a8a | 1658 | |
17a52670 | 1659 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 1660 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
1661 | return -EINVAL; |
1662 | } | |
1663 | ||
c342dc10 | 1664 | reg = ®s[regno]; |
5327ed3d | 1665 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
1666 | if (t == SRC_OP) { |
1667 | /* check whether register used as source operand can be read */ | |
c342dc10 | 1668 | if (reg->type == NOT_INIT) { |
61bd5218 | 1669 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
1670 | return -EACCES; |
1671 | } | |
679c782d | 1672 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
1673 | if (regno == BPF_REG_FP) |
1674 | return 0; | |
1675 | ||
5327ed3d JW |
1676 | if (rw64) |
1677 | mark_insn_zext(env, reg); | |
1678 | ||
1679 | return mark_reg_read(env, reg, reg->parent, | |
1680 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
1681 | } else { |
1682 | /* check whether register used as dest operand can be written to */ | |
1683 | if (regno == BPF_REG_FP) { | |
61bd5218 | 1684 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
1685 | return -EACCES; |
1686 | } | |
c342dc10 | 1687 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 1688 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 1689 | if (t == DST_OP) |
61bd5218 | 1690 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
1691 | } |
1692 | return 0; | |
1693 | } | |
1694 | ||
b5dc0163 AS |
1695 | /* for any branch, call, exit record the history of jmps in the given state */ |
1696 | static int push_jmp_history(struct bpf_verifier_env *env, | |
1697 | struct bpf_verifier_state *cur) | |
1698 | { | |
1699 | u32 cnt = cur->jmp_history_cnt; | |
1700 | struct bpf_idx_pair *p; | |
1701 | ||
1702 | cnt++; | |
1703 | p = krealloc(cur->jmp_history, cnt * sizeof(*p), GFP_USER); | |
1704 | if (!p) | |
1705 | return -ENOMEM; | |
1706 | p[cnt - 1].idx = env->insn_idx; | |
1707 | p[cnt - 1].prev_idx = env->prev_insn_idx; | |
1708 | cur->jmp_history = p; | |
1709 | cur->jmp_history_cnt = cnt; | |
1710 | return 0; | |
1711 | } | |
1712 | ||
1713 | /* Backtrack one insn at a time. If idx is not at the top of recorded | |
1714 | * history then previous instruction came from straight line execution. | |
1715 | */ | |
1716 | static int get_prev_insn_idx(struct bpf_verifier_state *st, int i, | |
1717 | u32 *history) | |
1718 | { | |
1719 | u32 cnt = *history; | |
1720 | ||
1721 | if (cnt && st->jmp_history[cnt - 1].idx == i) { | |
1722 | i = st->jmp_history[cnt - 1].prev_idx; | |
1723 | (*history)--; | |
1724 | } else { | |
1725 | i--; | |
1726 | } | |
1727 | return i; | |
1728 | } | |
1729 | ||
1730 | /* For given verifier state backtrack_insn() is called from the last insn to | |
1731 | * the first insn. Its purpose is to compute a bitmask of registers and | |
1732 | * stack slots that needs precision in the parent verifier state. | |
1733 | */ | |
1734 | static int backtrack_insn(struct bpf_verifier_env *env, int idx, | |
1735 | u32 *reg_mask, u64 *stack_mask) | |
1736 | { | |
1737 | const struct bpf_insn_cbs cbs = { | |
1738 | .cb_print = verbose, | |
1739 | .private_data = env, | |
1740 | }; | |
1741 | struct bpf_insn *insn = env->prog->insnsi + idx; | |
1742 | u8 class = BPF_CLASS(insn->code); | |
1743 | u8 opcode = BPF_OP(insn->code); | |
1744 | u8 mode = BPF_MODE(insn->code); | |
1745 | u32 dreg = 1u << insn->dst_reg; | |
1746 | u32 sreg = 1u << insn->src_reg; | |
1747 | u32 spi; | |
1748 | ||
1749 | if (insn->code == 0) | |
1750 | return 0; | |
1751 | if (env->log.level & BPF_LOG_LEVEL) { | |
1752 | verbose(env, "regs=%x stack=%llx before ", *reg_mask, *stack_mask); | |
1753 | verbose(env, "%d: ", idx); | |
1754 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); | |
1755 | } | |
1756 | ||
1757 | if (class == BPF_ALU || class == BPF_ALU64) { | |
1758 | if (!(*reg_mask & dreg)) | |
1759 | return 0; | |
1760 | if (opcode == BPF_MOV) { | |
1761 | if (BPF_SRC(insn->code) == BPF_X) { | |
1762 | /* dreg = sreg | |
1763 | * dreg needs precision after this insn | |
1764 | * sreg needs precision before this insn | |
1765 | */ | |
1766 | *reg_mask &= ~dreg; | |
1767 | *reg_mask |= sreg; | |
1768 | } else { | |
1769 | /* dreg = K | |
1770 | * dreg needs precision after this insn. | |
1771 | * Corresponding register is already marked | |
1772 | * as precise=true in this verifier state. | |
1773 | * No further markings in parent are necessary | |
1774 | */ | |
1775 | *reg_mask &= ~dreg; | |
1776 | } | |
1777 | } else { | |
1778 | if (BPF_SRC(insn->code) == BPF_X) { | |
1779 | /* dreg += sreg | |
1780 | * both dreg and sreg need precision | |
1781 | * before this insn | |
1782 | */ | |
1783 | *reg_mask |= sreg; | |
1784 | } /* else dreg += K | |
1785 | * dreg still needs precision before this insn | |
1786 | */ | |
1787 | } | |
1788 | } else if (class == BPF_LDX) { | |
1789 | if (!(*reg_mask & dreg)) | |
1790 | return 0; | |
1791 | *reg_mask &= ~dreg; | |
1792 | ||
1793 | /* scalars can only be spilled into stack w/o losing precision. | |
1794 | * Load from any other memory can be zero extended. | |
1795 | * The desire to keep that precision is already indicated | |
1796 | * by 'precise' mark in corresponding register of this state. | |
1797 | * No further tracking necessary. | |
1798 | */ | |
1799 | if (insn->src_reg != BPF_REG_FP) | |
1800 | return 0; | |
1801 | if (BPF_SIZE(insn->code) != BPF_DW) | |
1802 | return 0; | |
1803 | ||
1804 | /* dreg = *(u64 *)[fp - off] was a fill from the stack. | |
1805 | * that [fp - off] slot contains scalar that needs to be | |
1806 | * tracked with precision | |
1807 | */ | |
1808 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
1809 | if (spi >= 64) { | |
1810 | verbose(env, "BUG spi %d\n", spi); | |
1811 | WARN_ONCE(1, "verifier backtracking bug"); | |
1812 | return -EFAULT; | |
1813 | } | |
1814 | *stack_mask |= 1ull << spi; | |
b3b50f05 | 1815 | } else if (class == BPF_STX || class == BPF_ST) { |
b5dc0163 | 1816 | if (*reg_mask & dreg) |
b3b50f05 | 1817 | /* stx & st shouldn't be using _scalar_ dst_reg |
b5dc0163 AS |
1818 | * to access memory. It means backtracking |
1819 | * encountered a case of pointer subtraction. | |
1820 | */ | |
1821 | return -ENOTSUPP; | |
1822 | /* scalars can only be spilled into stack */ | |
1823 | if (insn->dst_reg != BPF_REG_FP) | |
1824 | return 0; | |
1825 | if (BPF_SIZE(insn->code) != BPF_DW) | |
1826 | return 0; | |
1827 | spi = (-insn->off - 1) / BPF_REG_SIZE; | |
1828 | if (spi >= 64) { | |
1829 | verbose(env, "BUG spi %d\n", spi); | |
1830 | WARN_ONCE(1, "verifier backtracking bug"); | |
1831 | return -EFAULT; | |
1832 | } | |
1833 | if (!(*stack_mask & (1ull << spi))) | |
1834 | return 0; | |
1835 | *stack_mask &= ~(1ull << spi); | |
b3b50f05 AN |
1836 | if (class == BPF_STX) |
1837 | *reg_mask |= sreg; | |
b5dc0163 AS |
1838 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
1839 | if (opcode == BPF_CALL) { | |
1840 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
1841 | return -ENOTSUPP; | |
1842 | /* regular helper call sets R0 */ | |
1843 | *reg_mask &= ~1; | |
1844 | if (*reg_mask & 0x3f) { | |
1845 | /* if backtracing was looking for registers R1-R5 | |
1846 | * they should have been found already. | |
1847 | */ | |
1848 | verbose(env, "BUG regs %x\n", *reg_mask); | |
1849 | WARN_ONCE(1, "verifier backtracking bug"); | |
1850 | return -EFAULT; | |
1851 | } | |
1852 | } else if (opcode == BPF_EXIT) { | |
1853 | return -ENOTSUPP; | |
1854 | } | |
1855 | } else if (class == BPF_LD) { | |
1856 | if (!(*reg_mask & dreg)) | |
1857 | return 0; | |
1858 | *reg_mask &= ~dreg; | |
1859 | /* It's ld_imm64 or ld_abs or ld_ind. | |
1860 | * For ld_imm64 no further tracking of precision | |
1861 | * into parent is necessary | |
1862 | */ | |
1863 | if (mode == BPF_IND || mode == BPF_ABS) | |
1864 | /* to be analyzed */ | |
1865 | return -ENOTSUPP; | |
b5dc0163 AS |
1866 | } |
1867 | return 0; | |
1868 | } | |
1869 | ||
1870 | /* the scalar precision tracking algorithm: | |
1871 | * . at the start all registers have precise=false. | |
1872 | * . scalar ranges are tracked as normal through alu and jmp insns. | |
1873 | * . once precise value of the scalar register is used in: | |
1874 | * . ptr + scalar alu | |
1875 | * . if (scalar cond K|scalar) | |
1876 | * . helper_call(.., scalar, ...) where ARG_CONST is expected | |
1877 | * backtrack through the verifier states and mark all registers and | |
1878 | * stack slots with spilled constants that these scalar regisers | |
1879 | * should be precise. | |
1880 | * . during state pruning two registers (or spilled stack slots) | |
1881 | * are equivalent if both are not precise. | |
1882 | * | |
1883 | * Note the verifier cannot simply walk register parentage chain, | |
1884 | * since many different registers and stack slots could have been | |
1885 | * used to compute single precise scalar. | |
1886 | * | |
1887 | * The approach of starting with precise=true for all registers and then | |
1888 | * backtrack to mark a register as not precise when the verifier detects | |
1889 | * that program doesn't care about specific value (e.g., when helper | |
1890 | * takes register as ARG_ANYTHING parameter) is not safe. | |
1891 | * | |
1892 | * It's ok to walk single parentage chain of the verifier states. | |
1893 | * It's possible that this backtracking will go all the way till 1st insn. | |
1894 | * All other branches will be explored for needing precision later. | |
1895 | * | |
1896 | * The backtracking needs to deal with cases like: | |
1897 | * R8=map_value(id=0,off=0,ks=4,vs=1952,imm=0) R9_w=map_value(id=0,off=40,ks=4,vs=1952,imm=0) | |
1898 | * r9 -= r8 | |
1899 | * r5 = r9 | |
1900 | * if r5 > 0x79f goto pc+7 | |
1901 | * R5_w=inv(id=0,umax_value=1951,var_off=(0x0; 0x7ff)) | |
1902 | * r5 += 1 | |
1903 | * ... | |
1904 | * call bpf_perf_event_output#25 | |
1905 | * where .arg5_type = ARG_CONST_SIZE_OR_ZERO | |
1906 | * | |
1907 | * and this case: | |
1908 | * r6 = 1 | |
1909 | * call foo // uses callee's r6 inside to compute r0 | |
1910 | * r0 += r6 | |
1911 | * if r0 == 0 goto | |
1912 | * | |
1913 | * to track above reg_mask/stack_mask needs to be independent for each frame. | |
1914 | * | |
1915 | * Also if parent's curframe > frame where backtracking started, | |
1916 | * the verifier need to mark registers in both frames, otherwise callees | |
1917 | * may incorrectly prune callers. This is similar to | |
1918 | * commit 7640ead93924 ("bpf: verifier: make sure callees don't prune with caller differences") | |
1919 | * | |
1920 | * For now backtracking falls back into conservative marking. | |
1921 | */ | |
1922 | static void mark_all_scalars_precise(struct bpf_verifier_env *env, | |
1923 | struct bpf_verifier_state *st) | |
1924 | { | |
1925 | struct bpf_func_state *func; | |
1926 | struct bpf_reg_state *reg; | |
1927 | int i, j; | |
1928 | ||
1929 | /* big hammer: mark all scalars precise in this path. | |
1930 | * pop_stack may still get !precise scalars. | |
1931 | */ | |
1932 | for (; st; st = st->parent) | |
1933 | for (i = 0; i <= st->curframe; i++) { | |
1934 | func = st->frame[i]; | |
1935 | for (j = 0; j < BPF_REG_FP; j++) { | |
1936 | reg = &func->regs[j]; | |
1937 | if (reg->type != SCALAR_VALUE) | |
1938 | continue; | |
1939 | reg->precise = true; | |
1940 | } | |
1941 | for (j = 0; j < func->allocated_stack / BPF_REG_SIZE; j++) { | |
1942 | if (func->stack[j].slot_type[0] != STACK_SPILL) | |
1943 | continue; | |
1944 | reg = &func->stack[j].spilled_ptr; | |
1945 | if (reg->type != SCALAR_VALUE) | |
1946 | continue; | |
1947 | reg->precise = true; | |
1948 | } | |
1949 | } | |
1950 | } | |
1951 | ||
a3ce685d AS |
1952 | static int __mark_chain_precision(struct bpf_verifier_env *env, int regno, |
1953 | int spi) | |
b5dc0163 AS |
1954 | { |
1955 | struct bpf_verifier_state *st = env->cur_state; | |
1956 | int first_idx = st->first_insn_idx; | |
1957 | int last_idx = env->insn_idx; | |
1958 | struct bpf_func_state *func; | |
1959 | struct bpf_reg_state *reg; | |
a3ce685d AS |
1960 | u32 reg_mask = regno >= 0 ? 1u << regno : 0; |
1961 | u64 stack_mask = spi >= 0 ? 1ull << spi : 0; | |
b5dc0163 | 1962 | bool skip_first = true; |
a3ce685d | 1963 | bool new_marks = false; |
b5dc0163 AS |
1964 | int i, err; |
1965 | ||
2c78ee89 | 1966 | if (!env->bpf_capable) |
b5dc0163 AS |
1967 | return 0; |
1968 | ||
1969 | func = st->frame[st->curframe]; | |
a3ce685d AS |
1970 | if (regno >= 0) { |
1971 | reg = &func->regs[regno]; | |
1972 | if (reg->type != SCALAR_VALUE) { | |
1973 | WARN_ONCE(1, "backtracing misuse"); | |
1974 | return -EFAULT; | |
1975 | } | |
1976 | if (!reg->precise) | |
1977 | new_marks = true; | |
1978 | else | |
1979 | reg_mask = 0; | |
1980 | reg->precise = true; | |
b5dc0163 | 1981 | } |
b5dc0163 | 1982 | |
a3ce685d AS |
1983 | while (spi >= 0) { |
1984 | if (func->stack[spi].slot_type[0] != STACK_SPILL) { | |
1985 | stack_mask = 0; | |
1986 | break; | |
1987 | } | |
1988 | reg = &func->stack[spi].spilled_ptr; | |
1989 | if (reg->type != SCALAR_VALUE) { | |
1990 | stack_mask = 0; | |
1991 | break; | |
1992 | } | |
1993 | if (!reg->precise) | |
1994 | new_marks = true; | |
1995 | else | |
1996 | stack_mask = 0; | |
1997 | reg->precise = true; | |
1998 | break; | |
1999 | } | |
2000 | ||
2001 | if (!new_marks) | |
2002 | return 0; | |
2003 | if (!reg_mask && !stack_mask) | |
2004 | return 0; | |
b5dc0163 AS |
2005 | for (;;) { |
2006 | DECLARE_BITMAP(mask, 64); | |
b5dc0163 AS |
2007 | u32 history = st->jmp_history_cnt; |
2008 | ||
2009 | if (env->log.level & BPF_LOG_LEVEL) | |
2010 | verbose(env, "last_idx %d first_idx %d\n", last_idx, first_idx); | |
2011 | for (i = last_idx;;) { | |
2012 | if (skip_first) { | |
2013 | err = 0; | |
2014 | skip_first = false; | |
2015 | } else { | |
2016 | err = backtrack_insn(env, i, ®_mask, &stack_mask); | |
2017 | } | |
2018 | if (err == -ENOTSUPP) { | |
2019 | mark_all_scalars_precise(env, st); | |
2020 | return 0; | |
2021 | } else if (err) { | |
2022 | return err; | |
2023 | } | |
2024 | if (!reg_mask && !stack_mask) | |
2025 | /* Found assignment(s) into tracked register in this state. | |
2026 | * Since this state is already marked, just return. | |
2027 | * Nothing to be tracked further in the parent state. | |
2028 | */ | |
2029 | return 0; | |
2030 | if (i == first_idx) | |
2031 | break; | |
2032 | i = get_prev_insn_idx(st, i, &history); | |
2033 | if (i >= env->prog->len) { | |
2034 | /* This can happen if backtracking reached insn 0 | |
2035 | * and there are still reg_mask or stack_mask | |
2036 | * to backtrack. | |
2037 | * It means the backtracking missed the spot where | |
2038 | * particular register was initialized with a constant. | |
2039 | */ | |
2040 | verbose(env, "BUG backtracking idx %d\n", i); | |
2041 | WARN_ONCE(1, "verifier backtracking bug"); | |
2042 | return -EFAULT; | |
2043 | } | |
2044 | } | |
2045 | st = st->parent; | |
2046 | if (!st) | |
2047 | break; | |
2048 | ||
a3ce685d | 2049 | new_marks = false; |
b5dc0163 AS |
2050 | func = st->frame[st->curframe]; |
2051 | bitmap_from_u64(mask, reg_mask); | |
2052 | for_each_set_bit(i, mask, 32) { | |
2053 | reg = &func->regs[i]; | |
a3ce685d AS |
2054 | if (reg->type != SCALAR_VALUE) { |
2055 | reg_mask &= ~(1u << i); | |
b5dc0163 | 2056 | continue; |
a3ce685d | 2057 | } |
b5dc0163 AS |
2058 | if (!reg->precise) |
2059 | new_marks = true; | |
2060 | reg->precise = true; | |
2061 | } | |
2062 | ||
2063 | bitmap_from_u64(mask, stack_mask); | |
2064 | for_each_set_bit(i, mask, 64) { | |
2065 | if (i >= func->allocated_stack / BPF_REG_SIZE) { | |
2339cd6c AS |
2066 | /* the sequence of instructions: |
2067 | * 2: (bf) r3 = r10 | |
2068 | * 3: (7b) *(u64 *)(r3 -8) = r0 | |
2069 | * 4: (79) r4 = *(u64 *)(r10 -8) | |
2070 | * doesn't contain jmps. It's backtracked | |
2071 | * as a single block. | |
2072 | * During backtracking insn 3 is not recognized as | |
2073 | * stack access, so at the end of backtracking | |
2074 | * stack slot fp-8 is still marked in stack_mask. | |
2075 | * However the parent state may not have accessed | |
2076 | * fp-8 and it's "unallocated" stack space. | |
2077 | * In such case fallback to conservative. | |
b5dc0163 | 2078 | */ |
2339cd6c AS |
2079 | mark_all_scalars_precise(env, st); |
2080 | return 0; | |
b5dc0163 AS |
2081 | } |
2082 | ||
a3ce685d AS |
2083 | if (func->stack[i].slot_type[0] != STACK_SPILL) { |
2084 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2085 | continue; |
a3ce685d | 2086 | } |
b5dc0163 | 2087 | reg = &func->stack[i].spilled_ptr; |
a3ce685d AS |
2088 | if (reg->type != SCALAR_VALUE) { |
2089 | stack_mask &= ~(1ull << i); | |
b5dc0163 | 2090 | continue; |
a3ce685d | 2091 | } |
b5dc0163 AS |
2092 | if (!reg->precise) |
2093 | new_marks = true; | |
2094 | reg->precise = true; | |
2095 | } | |
2096 | if (env->log.level & BPF_LOG_LEVEL) { | |
2097 | print_verifier_state(env, func); | |
2098 | verbose(env, "parent %s regs=%x stack=%llx marks\n", | |
2099 | new_marks ? "didn't have" : "already had", | |
2100 | reg_mask, stack_mask); | |
2101 | } | |
2102 | ||
a3ce685d AS |
2103 | if (!reg_mask && !stack_mask) |
2104 | break; | |
b5dc0163 AS |
2105 | if (!new_marks) |
2106 | break; | |
2107 | ||
2108 | last_idx = st->last_insn_idx; | |
2109 | first_idx = st->first_insn_idx; | |
2110 | } | |
2111 | return 0; | |
2112 | } | |
2113 | ||
a3ce685d AS |
2114 | static int mark_chain_precision(struct bpf_verifier_env *env, int regno) |
2115 | { | |
2116 | return __mark_chain_precision(env, regno, -1); | |
2117 | } | |
2118 | ||
2119 | static int mark_chain_precision_stack(struct bpf_verifier_env *env, int spi) | |
2120 | { | |
2121 | return __mark_chain_precision(env, -1, spi); | |
2122 | } | |
b5dc0163 | 2123 | |
1be7f75d AS |
2124 | static bool is_spillable_regtype(enum bpf_reg_type type) |
2125 | { | |
2126 | switch (type) { | |
2127 | case PTR_TO_MAP_VALUE: | |
2128 | case PTR_TO_MAP_VALUE_OR_NULL: | |
2129 | case PTR_TO_STACK: | |
2130 | case PTR_TO_CTX: | |
969bf05e | 2131 | case PTR_TO_PACKET: |
de8f3a83 | 2132 | case PTR_TO_PACKET_META: |
969bf05e | 2133 | case PTR_TO_PACKET_END: |
d58e468b | 2134 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 2135 | case CONST_PTR_TO_MAP: |
c64b7983 JS |
2136 | case PTR_TO_SOCKET: |
2137 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
2138 | case PTR_TO_SOCK_COMMON: |
2139 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
2140 | case PTR_TO_TCP_SOCK: |
2141 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 2142 | case PTR_TO_XDP_SOCK: |
65726b5b | 2143 | case PTR_TO_BTF_ID: |
b121b341 | 2144 | case PTR_TO_BTF_ID_OR_NULL: |
1be7f75d AS |
2145 | return true; |
2146 | default: | |
2147 | return false; | |
2148 | } | |
2149 | } | |
2150 | ||
cc2b14d5 AS |
2151 | /* Does this register contain a constant zero? */ |
2152 | static bool register_is_null(struct bpf_reg_state *reg) | |
2153 | { | |
2154 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
2155 | } | |
2156 | ||
f7cf25b2 AS |
2157 | static bool register_is_const(struct bpf_reg_state *reg) |
2158 | { | |
2159 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
2160 | } | |
2161 | ||
6e7e63cb JH |
2162 | static bool __is_pointer_value(bool allow_ptr_leaks, |
2163 | const struct bpf_reg_state *reg) | |
2164 | { | |
2165 | if (allow_ptr_leaks) | |
2166 | return false; | |
2167 | ||
2168 | return reg->type != SCALAR_VALUE; | |
2169 | } | |
2170 | ||
f7cf25b2 AS |
2171 | static void save_register_state(struct bpf_func_state *state, |
2172 | int spi, struct bpf_reg_state *reg) | |
2173 | { | |
2174 | int i; | |
2175 | ||
2176 | state->stack[spi].spilled_ptr = *reg; | |
2177 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2178 | ||
2179 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2180 | state->stack[spi].slot_type[i] = STACK_SPILL; | |
2181 | } | |
2182 | ||
17a52670 AS |
2183 | /* check_stack_read/write functions track spill/fill of registers, |
2184 | * stack boundary and alignment are checked in check_mem_access() | |
2185 | */ | |
61bd5218 | 2186 | static int check_stack_write(struct bpf_verifier_env *env, |
f4d7e40a | 2187 | struct bpf_func_state *state, /* func where register points to */ |
af86ca4e | 2188 | int off, int size, int value_regno, int insn_idx) |
17a52670 | 2189 | { |
f4d7e40a | 2190 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 2191 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
b5dc0163 | 2192 | u32 dst_reg = env->prog->insnsi[insn_idx].dst_reg; |
f7cf25b2 | 2193 | struct bpf_reg_state *reg = NULL; |
638f5b90 | 2194 | |
f4d7e40a | 2195 | err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE), |
fd978bf7 | 2196 | state->acquired_refs, true); |
638f5b90 AS |
2197 | if (err) |
2198 | return err; | |
9c399760 AS |
2199 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
2200 | * so it's aligned access and [off, off + size) are within stack limits | |
2201 | */ | |
638f5b90 AS |
2202 | if (!env->allow_ptr_leaks && |
2203 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
2204 | size != BPF_REG_SIZE) { | |
2205 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
2206 | return -EACCES; | |
2207 | } | |
17a52670 | 2208 | |
f4d7e40a | 2209 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
2210 | if (value_regno >= 0) |
2211 | reg = &cur->regs[value_regno]; | |
17a52670 | 2212 | |
f7cf25b2 | 2213 | if (reg && size == BPF_REG_SIZE && register_is_const(reg) && |
2c78ee89 | 2214 | !register_is_null(reg) && env->bpf_capable) { |
b5dc0163 AS |
2215 | if (dst_reg != BPF_REG_FP) { |
2216 | /* The backtracking logic can only recognize explicit | |
2217 | * stack slot address like [fp - 8]. Other spill of | |
2218 | * scalar via different register has to be conervative. | |
2219 | * Backtrack from here and mark all registers as precise | |
2220 | * that contributed into 'reg' being a constant. | |
2221 | */ | |
2222 | err = mark_chain_precision(env, value_regno); | |
2223 | if (err) | |
2224 | return err; | |
2225 | } | |
f7cf25b2 AS |
2226 | save_register_state(state, spi, reg); |
2227 | } else if (reg && is_spillable_regtype(reg->type)) { | |
17a52670 | 2228 | /* register containing pointer is being spilled into stack */ |
9c399760 | 2229 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 2230 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 2231 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
2232 | return -EACCES; |
2233 | } | |
2234 | ||
f7cf25b2 | 2235 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
2236 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
2237 | return -EINVAL; | |
2238 | } | |
2239 | ||
2c78ee89 | 2240 | if (!env->bypass_spec_v4) { |
f7cf25b2 | 2241 | bool sanitize = false; |
17a52670 | 2242 | |
f7cf25b2 AS |
2243 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
2244 | register_is_const(&state->stack[spi].spilled_ptr)) | |
2245 | sanitize = true; | |
2246 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2247 | if (state->stack[spi].slot_type[i] == STACK_MISC) { | |
2248 | sanitize = true; | |
2249 | break; | |
2250 | } | |
2251 | if (sanitize) { | |
af86ca4e AS |
2252 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; |
2253 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
2254 | ||
2255 | /* detected reuse of integer stack slot with a pointer | |
2256 | * which means either llvm is reusing stack slot or | |
2257 | * an attacker is trying to exploit CVE-2018-3639 | |
2258 | * (speculative store bypass) | |
2259 | * Have to sanitize that slot with preemptive | |
2260 | * store of zero. | |
2261 | */ | |
2262 | if (*poff && *poff != soff) { | |
2263 | /* disallow programs where single insn stores | |
2264 | * into two different stack slots, since verifier | |
2265 | * cannot sanitize them | |
2266 | */ | |
2267 | verbose(env, | |
2268 | "insn %d cannot access two stack slots fp%d and fp%d", | |
2269 | insn_idx, *poff, soff); | |
2270 | return -EINVAL; | |
2271 | } | |
2272 | *poff = soff; | |
2273 | } | |
af86ca4e | 2274 | } |
f7cf25b2 | 2275 | save_register_state(state, spi, reg); |
9c399760 | 2276 | } else { |
cc2b14d5 AS |
2277 | u8 type = STACK_MISC; |
2278 | ||
679c782d EC |
2279 | /* regular write of data into stack destroys any spilled ptr */ |
2280 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d JW |
2281 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
2282 | if (state->stack[spi].slot_type[0] == STACK_SPILL) | |
2283 | for (i = 0; i < BPF_REG_SIZE; i++) | |
2284 | state->stack[spi].slot_type[i] = STACK_MISC; | |
9c399760 | 2285 | |
cc2b14d5 AS |
2286 | /* only mark the slot as written if all 8 bytes were written |
2287 | * otherwise read propagation may incorrectly stop too soon | |
2288 | * when stack slots are partially written. | |
2289 | * This heuristic means that read propagation will be | |
2290 | * conservative, since it will add reg_live_read marks | |
2291 | * to stack slots all the way to first state when programs | |
2292 | * writes+reads less than 8 bytes | |
2293 | */ | |
2294 | if (size == BPF_REG_SIZE) | |
2295 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
2296 | ||
2297 | /* when we zero initialize stack slots mark them as such */ | |
b5dc0163 AS |
2298 | if (reg && register_is_null(reg)) { |
2299 | /* backtracking doesn't work for STACK_ZERO yet. */ | |
2300 | err = mark_chain_precision(env, value_regno); | |
2301 | if (err) | |
2302 | return err; | |
cc2b14d5 | 2303 | type = STACK_ZERO; |
b5dc0163 | 2304 | } |
cc2b14d5 | 2305 | |
0bae2d4d | 2306 | /* Mark slots affected by this stack write. */ |
9c399760 | 2307 | for (i = 0; i < size; i++) |
638f5b90 | 2308 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 2309 | type; |
17a52670 AS |
2310 | } |
2311 | return 0; | |
2312 | } | |
2313 | ||
61bd5218 | 2314 | static int check_stack_read(struct bpf_verifier_env *env, |
f4d7e40a AS |
2315 | struct bpf_func_state *reg_state /* func where register points to */, |
2316 | int off, int size, int value_regno) | |
17a52670 | 2317 | { |
f4d7e40a AS |
2318 | struct bpf_verifier_state *vstate = env->cur_state; |
2319 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 2320 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 2321 | struct bpf_reg_state *reg; |
638f5b90 | 2322 | u8 *stype; |
17a52670 | 2323 | |
f4d7e40a | 2324 | if (reg_state->allocated_stack <= slot) { |
638f5b90 AS |
2325 | verbose(env, "invalid read from stack off %d+0 size %d\n", |
2326 | off, size); | |
2327 | return -EACCES; | |
2328 | } | |
f4d7e40a | 2329 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 2330 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 2331 | |
638f5b90 | 2332 | if (stype[0] == STACK_SPILL) { |
9c399760 | 2333 | if (size != BPF_REG_SIZE) { |
f7cf25b2 AS |
2334 | if (reg->type != SCALAR_VALUE) { |
2335 | verbose_linfo(env, env->insn_idx, "; "); | |
2336 | verbose(env, "invalid size of register fill\n"); | |
2337 | return -EACCES; | |
2338 | } | |
2339 | if (value_regno >= 0) { | |
2340 | mark_reg_unknown(env, state->regs, value_regno); | |
2341 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
2342 | } | |
2343 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
2344 | return 0; | |
17a52670 | 2345 | } |
9c399760 | 2346 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 2347 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 2348 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
2349 | return -EACCES; |
2350 | } | |
2351 | } | |
2352 | ||
dc503a8a | 2353 | if (value_regno >= 0) { |
17a52670 | 2354 | /* restore register state from stack */ |
f7cf25b2 | 2355 | state->regs[value_regno] = *reg; |
2f18f62e AS |
2356 | /* mark reg as written since spilled pointer state likely |
2357 | * has its liveness marks cleared by is_state_visited() | |
2358 | * which resets stack/reg liveness for state transitions | |
2359 | */ | |
2360 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
6e7e63cb JH |
2361 | } else if (__is_pointer_value(env->allow_ptr_leaks, reg)) { |
2362 | /* If value_regno==-1, the caller is asking us whether | |
2363 | * it is acceptable to use this value as a SCALAR_VALUE | |
2364 | * (e.g. for XADD). | |
2365 | * We must not allow unprivileged callers to do that | |
2366 | * with spilled pointers. | |
2367 | */ | |
2368 | verbose(env, "leaking pointer from stack off %d\n", | |
2369 | off); | |
2370 | return -EACCES; | |
dc503a8a | 2371 | } |
f7cf25b2 | 2372 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 | 2373 | } else { |
cc2b14d5 AS |
2374 | int zeros = 0; |
2375 | ||
17a52670 | 2376 | for (i = 0; i < size; i++) { |
cc2b14d5 AS |
2377 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_MISC) |
2378 | continue; | |
2379 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_ZERO) { | |
2380 | zeros++; | |
2381 | continue; | |
17a52670 | 2382 | } |
cc2b14d5 AS |
2383 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
2384 | off, i, size); | |
2385 | return -EACCES; | |
2386 | } | |
f7cf25b2 | 2387 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
cc2b14d5 AS |
2388 | if (value_regno >= 0) { |
2389 | if (zeros == size) { | |
2390 | /* any size read into register is zero extended, | |
2391 | * so the whole register == const_zero | |
2392 | */ | |
2393 | __mark_reg_const_zero(&state->regs[value_regno]); | |
b5dc0163 AS |
2394 | /* backtracking doesn't support STACK_ZERO yet, |
2395 | * so mark it precise here, so that later | |
2396 | * backtracking can stop here. | |
2397 | * Backtracking may not need this if this register | |
2398 | * doesn't participate in pointer adjustment. | |
2399 | * Forward propagation of precise flag is not | |
2400 | * necessary either. This mark is only to stop | |
2401 | * backtracking. Any register that contributed | |
2402 | * to const 0 was marked precise before spill. | |
2403 | */ | |
2404 | state->regs[value_regno].precise = true; | |
cc2b14d5 AS |
2405 | } else { |
2406 | /* have read misc data from the stack */ | |
2407 | mark_reg_unknown(env, state->regs, value_regno); | |
2408 | } | |
2409 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
17a52670 | 2410 | } |
17a52670 | 2411 | } |
f7cf25b2 | 2412 | return 0; |
17a52670 AS |
2413 | } |
2414 | ||
e4298d25 DB |
2415 | static int check_stack_access(struct bpf_verifier_env *env, |
2416 | const struct bpf_reg_state *reg, | |
2417 | int off, int size) | |
2418 | { | |
2419 | /* Stack accesses must be at a fixed offset, so that we | |
2420 | * can determine what type of data were returned. See | |
2421 | * check_stack_read(). | |
2422 | */ | |
2423 | if (!tnum_is_const(reg->var_off)) { | |
2424 | char tn_buf[48]; | |
2425 | ||
2426 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1fbd20f8 | 2427 | verbose(env, "variable stack access var_off=%s off=%d size=%d\n", |
e4298d25 DB |
2428 | tn_buf, off, size); |
2429 | return -EACCES; | |
2430 | } | |
2431 | ||
2432 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
2433 | verbose(env, "invalid stack off=%d size=%d\n", off, size); | |
2434 | return -EACCES; | |
2435 | } | |
2436 | ||
2437 | return 0; | |
2438 | } | |
2439 | ||
591fe988 DB |
2440 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
2441 | int off, int size, enum bpf_access_type type) | |
2442 | { | |
2443 | struct bpf_reg_state *regs = cur_regs(env); | |
2444 | struct bpf_map *map = regs[regno].map_ptr; | |
2445 | u32 cap = bpf_map_flags_to_cap(map); | |
2446 | ||
2447 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
2448 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
2449 | map->value_size, off, size); | |
2450 | return -EACCES; | |
2451 | } | |
2452 | ||
2453 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
2454 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
2455 | map->value_size, off, size); | |
2456 | return -EACCES; | |
2457 | } | |
2458 | ||
2459 | return 0; | |
2460 | } | |
2461 | ||
17a52670 | 2462 | /* check read/write into map element returned by bpf_map_lookup_elem() */ |
f1174f77 | 2463 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 2464 | int size, bool zero_size_allowed) |
17a52670 | 2465 | { |
638f5b90 AS |
2466 | struct bpf_reg_state *regs = cur_regs(env); |
2467 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 2468 | |
9fd29c08 YS |
2469 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
2470 | off + size > map->value_size) { | |
61bd5218 | 2471 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
2472 | map->value_size, off, size); |
2473 | return -EACCES; | |
2474 | } | |
2475 | return 0; | |
2476 | } | |
2477 | ||
f1174f77 EC |
2478 | /* check read/write into a map element with possible variable offset */ |
2479 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 2480 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 2481 | { |
f4d7e40a AS |
2482 | struct bpf_verifier_state *vstate = env->cur_state; |
2483 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
2484 | struct bpf_reg_state *reg = &state->regs[regno]; |
2485 | int err; | |
2486 | ||
f1174f77 EC |
2487 | /* We may have adjusted the register to this map value, so we |
2488 | * need to try adding each of min_value and max_value to off | |
2489 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 2490 | */ |
06ee7115 | 2491 | if (env->log.level & BPF_LOG_LEVEL) |
61bd5218 | 2492 | print_verifier_state(env, state); |
b7137c4e | 2493 | |
dbcfe5f7 GB |
2494 | /* The minimum value is only important with signed |
2495 | * comparisons where we can't assume the floor of a | |
2496 | * value is 0. If we are using signed variables for our | |
2497 | * index'es we need to make sure that whatever we use | |
2498 | * will have a set floor within our range. | |
2499 | */ | |
b7137c4e DB |
2500 | if (reg->smin_value < 0 && |
2501 | (reg->smin_value == S64_MIN || | |
2502 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
2503 | reg->smin_value + off < 0)) { | |
61bd5218 | 2504 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
2505 | regno); |
2506 | return -EACCES; | |
2507 | } | |
9fd29c08 YS |
2508 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
2509 | zero_size_allowed); | |
dbcfe5f7 | 2510 | if (err) { |
61bd5218 JK |
2511 | verbose(env, "R%d min value is outside of the array range\n", |
2512 | regno); | |
dbcfe5f7 GB |
2513 | return err; |
2514 | } | |
2515 | ||
b03c9f9f EC |
2516 | /* If we haven't set a max value then we need to bail since we can't be |
2517 | * sure we won't do bad things. | |
2518 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 2519 | */ |
b03c9f9f | 2520 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 2521 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
2522 | regno); |
2523 | return -EACCES; | |
2524 | } | |
9fd29c08 YS |
2525 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
2526 | zero_size_allowed); | |
f1174f77 | 2527 | if (err) |
61bd5218 JK |
2528 | verbose(env, "R%d max value is outside of the array range\n", |
2529 | regno); | |
d83525ca AS |
2530 | |
2531 | if (map_value_has_spin_lock(reg->map_ptr)) { | |
2532 | u32 lock = reg->map_ptr->spin_lock_off; | |
2533 | ||
2534 | /* if any part of struct bpf_spin_lock can be touched by | |
2535 | * load/store reject this program. | |
2536 | * To check that [x1, x2) overlaps with [y1, y2) | |
2537 | * it is sufficient to check x1 < y2 && y1 < x2. | |
2538 | */ | |
2539 | if (reg->smin_value + off < lock + sizeof(struct bpf_spin_lock) && | |
2540 | lock < reg->umax_value + off + size) { | |
2541 | verbose(env, "bpf_spin_lock cannot be accessed directly by load/store\n"); | |
2542 | return -EACCES; | |
2543 | } | |
2544 | } | |
f1174f77 | 2545 | return err; |
dbcfe5f7 GB |
2546 | } |
2547 | ||
969bf05e AS |
2548 | #define MAX_PACKET_OFF 0xffff |
2549 | ||
58e2af8b | 2550 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
2551 | const struct bpf_call_arg_meta *meta, |
2552 | enum bpf_access_type t) | |
4acf6c0b | 2553 | { |
36bbef52 | 2554 | switch (env->prog->type) { |
5d66fa7d | 2555 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
2556 | case BPF_PROG_TYPE_LWT_IN: |
2557 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 2558 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 2559 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 2560 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 2561 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
2562 | if (t == BPF_WRITE) |
2563 | return false; | |
7e57fbb2 | 2564 | /* fallthrough */ |
5d66fa7d DB |
2565 | |
2566 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
2567 | case BPF_PROG_TYPE_SCHED_CLS: |
2568 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 2569 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 2570 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 2571 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 2572 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
2573 | if (meta) |
2574 | return meta->pkt_access; | |
2575 | ||
2576 | env->seen_direct_write = true; | |
4acf6c0b | 2577 | return true; |
0d01da6a SF |
2578 | |
2579 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: | |
2580 | if (t == BPF_WRITE) | |
2581 | env->seen_direct_write = true; | |
2582 | ||
2583 | return true; | |
2584 | ||
4acf6c0b BB |
2585 | default: |
2586 | return false; | |
2587 | } | |
2588 | } | |
2589 | ||
f1174f77 | 2590 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 2591 | int off, int size, bool zero_size_allowed) |
969bf05e | 2592 | { |
638f5b90 | 2593 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 2594 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 2595 | |
9fd29c08 YS |
2596 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
2597 | (u64)off + size > reg->range) { | |
61bd5218 | 2598 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 2599 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
2600 | return -EACCES; |
2601 | } | |
2602 | return 0; | |
2603 | } | |
2604 | ||
f1174f77 | 2605 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 2606 | int size, bool zero_size_allowed) |
f1174f77 | 2607 | { |
638f5b90 | 2608 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
2609 | struct bpf_reg_state *reg = ®s[regno]; |
2610 | int err; | |
2611 | ||
2612 | /* We may have added a variable offset to the packet pointer; but any | |
2613 | * reg->range we have comes after that. We are only checking the fixed | |
2614 | * offset. | |
2615 | */ | |
2616 | ||
2617 | /* We don't allow negative numbers, because we aren't tracking enough | |
2618 | * detail to prove they're safe. | |
2619 | */ | |
b03c9f9f | 2620 | if (reg->smin_value < 0) { |
61bd5218 | 2621 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
2622 | regno); |
2623 | return -EACCES; | |
2624 | } | |
9fd29c08 | 2625 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 2626 | if (err) { |
61bd5218 | 2627 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
2628 | return err; |
2629 | } | |
e647815a JW |
2630 | |
2631 | /* __check_packet_access has made sure "off + size - 1" is within u16. | |
2632 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, | |
2633 | * otherwise find_good_pkt_pointers would have refused to set range info | |
2634 | * that __check_packet_access would have rejected this pkt access. | |
2635 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. | |
2636 | */ | |
2637 | env->prog->aux->max_pkt_offset = | |
2638 | max_t(u32, env->prog->aux->max_pkt_offset, | |
2639 | off + reg->umax_value + size - 1); | |
2640 | ||
f1174f77 EC |
2641 | return err; |
2642 | } | |
2643 | ||
2644 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 2645 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
9e15db66 AS |
2646 | enum bpf_access_type t, enum bpf_reg_type *reg_type, |
2647 | u32 *btf_id) | |
17a52670 | 2648 | { |
f96da094 DB |
2649 | struct bpf_insn_access_aux info = { |
2650 | .reg_type = *reg_type, | |
9e15db66 | 2651 | .log = &env->log, |
f96da094 | 2652 | }; |
31fd8581 | 2653 | |
4f9218aa | 2654 | if (env->ops->is_valid_access && |
5e43f899 | 2655 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
2656 | /* A non zero info.ctx_field_size indicates that this field is a |
2657 | * candidate for later verifier transformation to load the whole | |
2658 | * field and then apply a mask when accessed with a narrower | |
2659 | * access than actual ctx access size. A zero info.ctx_field_size | |
2660 | * will only allow for whole field access and rejects any other | |
2661 | * type of narrower access. | |
31fd8581 | 2662 | */ |
23994631 | 2663 | *reg_type = info.reg_type; |
31fd8581 | 2664 | |
b121b341 | 2665 | if (*reg_type == PTR_TO_BTF_ID || *reg_type == PTR_TO_BTF_ID_OR_NULL) |
9e15db66 AS |
2666 | *btf_id = info.btf_id; |
2667 | else | |
2668 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; | |
32bbe007 AS |
2669 | /* remember the offset of last byte accessed in ctx */ |
2670 | if (env->prog->aux->max_ctx_offset < off + size) | |
2671 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 2672 | return 0; |
32bbe007 | 2673 | } |
17a52670 | 2674 | |
61bd5218 | 2675 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
2676 | return -EACCES; |
2677 | } | |
2678 | ||
d58e468b PP |
2679 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
2680 | int size) | |
2681 | { | |
2682 | if (size < 0 || off < 0 || | |
2683 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
2684 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
2685 | off, size); | |
2686 | return -EACCES; | |
2687 | } | |
2688 | return 0; | |
2689 | } | |
2690 | ||
5f456649 MKL |
2691 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
2692 | u32 regno, int off, int size, | |
2693 | enum bpf_access_type t) | |
c64b7983 JS |
2694 | { |
2695 | struct bpf_reg_state *regs = cur_regs(env); | |
2696 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 2697 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 2698 | bool valid; |
c64b7983 JS |
2699 | |
2700 | if (reg->smin_value < 0) { | |
2701 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
2702 | regno); | |
2703 | return -EACCES; | |
2704 | } | |
2705 | ||
46f8bc92 MKL |
2706 | switch (reg->type) { |
2707 | case PTR_TO_SOCK_COMMON: | |
2708 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
2709 | break; | |
2710 | case PTR_TO_SOCKET: | |
2711 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
2712 | break; | |
655a51e5 MKL |
2713 | case PTR_TO_TCP_SOCK: |
2714 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
2715 | break; | |
fada7fdc JL |
2716 | case PTR_TO_XDP_SOCK: |
2717 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
2718 | break; | |
46f8bc92 MKL |
2719 | default: |
2720 | valid = false; | |
c64b7983 JS |
2721 | } |
2722 | ||
5f456649 | 2723 | |
46f8bc92 MKL |
2724 | if (valid) { |
2725 | env->insn_aux_data[insn_idx].ctx_field_size = | |
2726 | info.ctx_field_size; | |
2727 | return 0; | |
2728 | } | |
2729 | ||
2730 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
2731 | regno, reg_type_str[reg->type], off, size); | |
2732 | ||
2733 | return -EACCES; | |
c64b7983 JS |
2734 | } |
2735 | ||
2a159c6f DB |
2736 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) |
2737 | { | |
2738 | return cur_regs(env) + regno; | |
2739 | } | |
2740 | ||
4cabc5b1 DB |
2741 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
2742 | { | |
2a159c6f | 2743 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
2744 | } |
2745 | ||
f37a8cb8 DB |
2746 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
2747 | { | |
2a159c6f | 2748 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 2749 | |
46f8bc92 MKL |
2750 | return reg->type == PTR_TO_CTX; |
2751 | } | |
2752 | ||
2753 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
2754 | { | |
2755 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
2756 | ||
2757 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
2758 | } |
2759 | ||
ca369602 DB |
2760 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
2761 | { | |
2a159c6f | 2762 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
2763 | |
2764 | return type_is_pkt_pointer(reg->type); | |
2765 | } | |
2766 | ||
4b5defde DB |
2767 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
2768 | { | |
2769 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
2770 | ||
2771 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
2772 | return reg->type == PTR_TO_FLOW_KEYS; | |
2773 | } | |
2774 | ||
61bd5218 JK |
2775 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
2776 | const struct bpf_reg_state *reg, | |
d1174416 | 2777 | int off, int size, bool strict) |
969bf05e | 2778 | { |
f1174f77 | 2779 | struct tnum reg_off; |
e07b98d9 | 2780 | int ip_align; |
d1174416 DM |
2781 | |
2782 | /* Byte size accesses are always allowed. */ | |
2783 | if (!strict || size == 1) | |
2784 | return 0; | |
2785 | ||
e4eda884 DM |
2786 | /* For platforms that do not have a Kconfig enabling |
2787 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
2788 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
2789 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
2790 | * to this code only in strict mode where we want to emulate | |
2791 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
2792 | * unconditional IP align value of '2'. | |
e07b98d9 | 2793 | */ |
e4eda884 | 2794 | ip_align = 2; |
f1174f77 EC |
2795 | |
2796 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
2797 | if (!tnum_is_aligned(reg_off, size)) { | |
2798 | char tn_buf[48]; | |
2799 | ||
2800 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
2801 | verbose(env, |
2802 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 2803 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
2804 | return -EACCES; |
2805 | } | |
79adffcd | 2806 | |
969bf05e AS |
2807 | return 0; |
2808 | } | |
2809 | ||
61bd5218 JK |
2810 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
2811 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
2812 | const char *pointer_desc, |
2813 | int off, int size, bool strict) | |
79adffcd | 2814 | { |
f1174f77 EC |
2815 | struct tnum reg_off; |
2816 | ||
2817 | /* Byte size accesses are always allowed. */ | |
2818 | if (!strict || size == 1) | |
2819 | return 0; | |
2820 | ||
2821 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
2822 | if (!tnum_is_aligned(reg_off, size)) { | |
2823 | char tn_buf[48]; | |
2824 | ||
2825 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 2826 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 2827 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
2828 | return -EACCES; |
2829 | } | |
2830 | ||
969bf05e AS |
2831 | return 0; |
2832 | } | |
2833 | ||
e07b98d9 | 2834 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
2835 | const struct bpf_reg_state *reg, int off, |
2836 | int size, bool strict_alignment_once) | |
79adffcd | 2837 | { |
ca369602 | 2838 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 2839 | const char *pointer_desc = ""; |
d1174416 | 2840 | |
79adffcd DB |
2841 | switch (reg->type) { |
2842 | case PTR_TO_PACKET: | |
de8f3a83 DB |
2843 | case PTR_TO_PACKET_META: |
2844 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
2845 | * right in front, treat it the very same way. | |
2846 | */ | |
61bd5218 | 2847 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
2848 | case PTR_TO_FLOW_KEYS: |
2849 | pointer_desc = "flow keys "; | |
2850 | break; | |
f1174f77 EC |
2851 | case PTR_TO_MAP_VALUE: |
2852 | pointer_desc = "value "; | |
2853 | break; | |
2854 | case PTR_TO_CTX: | |
2855 | pointer_desc = "context "; | |
2856 | break; | |
2857 | case PTR_TO_STACK: | |
2858 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
2859 | /* The stack spill tracking logic in check_stack_write() |
2860 | * and check_stack_read() relies on stack accesses being | |
2861 | * aligned. | |
2862 | */ | |
2863 | strict = true; | |
f1174f77 | 2864 | break; |
c64b7983 JS |
2865 | case PTR_TO_SOCKET: |
2866 | pointer_desc = "sock "; | |
2867 | break; | |
46f8bc92 MKL |
2868 | case PTR_TO_SOCK_COMMON: |
2869 | pointer_desc = "sock_common "; | |
2870 | break; | |
655a51e5 MKL |
2871 | case PTR_TO_TCP_SOCK: |
2872 | pointer_desc = "tcp_sock "; | |
2873 | break; | |
fada7fdc JL |
2874 | case PTR_TO_XDP_SOCK: |
2875 | pointer_desc = "xdp_sock "; | |
2876 | break; | |
79adffcd | 2877 | default: |
f1174f77 | 2878 | break; |
79adffcd | 2879 | } |
61bd5218 JK |
2880 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
2881 | strict); | |
79adffcd DB |
2882 | } |
2883 | ||
f4d7e40a AS |
2884 | static int update_stack_depth(struct bpf_verifier_env *env, |
2885 | const struct bpf_func_state *func, | |
2886 | int off) | |
2887 | { | |
9c8105bd | 2888 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
2889 | |
2890 | if (stack >= -off) | |
2891 | return 0; | |
2892 | ||
2893 | /* update known max for given subprogram */ | |
9c8105bd | 2894 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
2895 | return 0; |
2896 | } | |
f4d7e40a | 2897 | |
70a87ffe AS |
2898 | /* starting from main bpf function walk all instructions of the function |
2899 | * and recursively walk all callees that given function can call. | |
2900 | * Ignore jump and exit insns. | |
2901 | * Since recursion is prevented by check_cfg() this algorithm | |
2902 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
2903 | */ | |
2904 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
2905 | { | |
9c8105bd JW |
2906 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
2907 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 2908 | struct bpf_insn *insn = env->prog->insnsi; |
70a87ffe AS |
2909 | int ret_insn[MAX_CALL_FRAMES]; |
2910 | int ret_prog[MAX_CALL_FRAMES]; | |
f4d7e40a | 2911 | |
70a87ffe AS |
2912 | process_func: |
2913 | /* round up to 32-bytes, since this is granularity | |
2914 | * of interpreter stack size | |
2915 | */ | |
9c8105bd | 2916 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 2917 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 2918 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 2919 | frame + 1, depth); |
f4d7e40a AS |
2920 | return -EACCES; |
2921 | } | |
70a87ffe | 2922 | continue_func: |
4cb3d99c | 2923 | subprog_end = subprog[idx + 1].start; |
70a87ffe AS |
2924 | for (; i < subprog_end; i++) { |
2925 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
2926 | continue; | |
2927 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
2928 | continue; | |
2929 | /* remember insn and function to return to */ | |
2930 | ret_insn[frame] = i + 1; | |
9c8105bd | 2931 | ret_prog[frame] = idx; |
70a87ffe AS |
2932 | |
2933 | /* find the callee */ | |
2934 | i = i + insn[i].imm + 1; | |
9c8105bd JW |
2935 | idx = find_subprog(env, i); |
2936 | if (idx < 0) { | |
70a87ffe AS |
2937 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
2938 | i); | |
2939 | return -EFAULT; | |
2940 | } | |
70a87ffe AS |
2941 | frame++; |
2942 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
2943 | verbose(env, "the call stack of %d frames is too deep !\n", |
2944 | frame); | |
2945 | return -E2BIG; | |
70a87ffe AS |
2946 | } |
2947 | goto process_func; | |
2948 | } | |
2949 | /* end of for() loop means the last insn of the 'subprog' | |
2950 | * was reached. Doesn't matter whether it was JA or EXIT | |
2951 | */ | |
2952 | if (frame == 0) | |
2953 | return 0; | |
9c8105bd | 2954 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
2955 | frame--; |
2956 | i = ret_insn[frame]; | |
9c8105bd | 2957 | idx = ret_prog[frame]; |
70a87ffe | 2958 | goto continue_func; |
f4d7e40a AS |
2959 | } |
2960 | ||
19d28fbd | 2961 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
2962 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
2963 | const struct bpf_insn *insn, int idx) | |
2964 | { | |
2965 | int start = idx + insn->imm + 1, subprog; | |
2966 | ||
2967 | subprog = find_subprog(env, start); | |
2968 | if (subprog < 0) { | |
2969 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
2970 | start); | |
2971 | return -EFAULT; | |
2972 | } | |
9c8105bd | 2973 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 2974 | } |
19d28fbd | 2975 | #endif |
1ea47e01 | 2976 | |
51c39bb1 AS |
2977 | int check_ctx_reg(struct bpf_verifier_env *env, |
2978 | const struct bpf_reg_state *reg, int regno) | |
58990d1f DB |
2979 | { |
2980 | /* Access to ctx or passing it to a helper is only allowed in | |
2981 | * its original, unmodified form. | |
2982 | */ | |
2983 | ||
2984 | if (reg->off) { | |
2985 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
2986 | regno, reg->off); | |
2987 | return -EACCES; | |
2988 | } | |
2989 | ||
2990 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
2991 | char tn_buf[48]; | |
2992 | ||
2993 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
2994 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
2995 | return -EACCES; | |
2996 | } | |
2997 | ||
2998 | return 0; | |
2999 | } | |
3000 | ||
9df1c28b MM |
3001 | static int check_tp_buffer_access(struct bpf_verifier_env *env, |
3002 | const struct bpf_reg_state *reg, | |
3003 | int regno, int off, int size) | |
3004 | { | |
3005 | if (off < 0) { | |
3006 | verbose(env, | |
3007 | "R%d invalid tracepoint buffer access: off=%d, size=%d", | |
3008 | regno, off, size); | |
3009 | return -EACCES; | |
3010 | } | |
3011 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3012 | char tn_buf[48]; | |
3013 | ||
3014 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3015 | verbose(env, | |
3016 | "R%d invalid variable buffer offset: off=%d, var_off=%s", | |
3017 | regno, off, tn_buf); | |
3018 | return -EACCES; | |
3019 | } | |
3020 | if (off + size > env->prog->aux->max_tp_access) | |
3021 | env->prog->aux->max_tp_access = off + size; | |
3022 | ||
3023 | return 0; | |
3024 | } | |
3025 | ||
3f50f132 JF |
3026 | /* BPF architecture zero extends alu32 ops into 64-bit registesr */ |
3027 | static void zext_32_to_64(struct bpf_reg_state *reg) | |
3028 | { | |
3029 | reg->var_off = tnum_subreg(reg->var_off); | |
3030 | __reg_assign_32_into_64(reg); | |
3031 | } | |
9df1c28b | 3032 | |
0c17d1d2 JH |
3033 | /* truncate register to smaller size (in bytes) |
3034 | * must be called with size < BPF_REG_SIZE | |
3035 | */ | |
3036 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
3037 | { | |
3038 | u64 mask; | |
3039 | ||
3040 | /* clear high bits in bit representation */ | |
3041 | reg->var_off = tnum_cast(reg->var_off, size); | |
3042 | ||
3043 | /* fix arithmetic bounds */ | |
3044 | mask = ((u64)1 << (size * 8)) - 1; | |
3045 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
3046 | reg->umin_value &= mask; | |
3047 | reg->umax_value &= mask; | |
3048 | } else { | |
3049 | reg->umin_value = 0; | |
3050 | reg->umax_value = mask; | |
3051 | } | |
3052 | reg->smin_value = reg->umin_value; | |
3053 | reg->smax_value = reg->umax_value; | |
3f50f132 JF |
3054 | |
3055 | /* If size is smaller than 32bit register the 32bit register | |
3056 | * values are also truncated so we push 64-bit bounds into | |
3057 | * 32-bit bounds. Above were truncated < 32-bits already. | |
3058 | */ | |
3059 | if (size >= 4) | |
3060 | return; | |
3061 | __reg_combine_64_into_32(reg); | |
0c17d1d2 JH |
3062 | } |
3063 | ||
a23740ec AN |
3064 | static bool bpf_map_is_rdonly(const struct bpf_map *map) |
3065 | { | |
3066 | return (map->map_flags & BPF_F_RDONLY_PROG) && map->frozen; | |
3067 | } | |
3068 | ||
3069 | static int bpf_map_direct_read(struct bpf_map *map, int off, int size, u64 *val) | |
3070 | { | |
3071 | void *ptr; | |
3072 | u64 addr; | |
3073 | int err; | |
3074 | ||
3075 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
3076 | if (err) | |
3077 | return err; | |
2dedd7d2 | 3078 | ptr = (void *)(long)addr + off; |
a23740ec AN |
3079 | |
3080 | switch (size) { | |
3081 | case sizeof(u8): | |
3082 | *val = (u64)*(u8 *)ptr; | |
3083 | break; | |
3084 | case sizeof(u16): | |
3085 | *val = (u64)*(u16 *)ptr; | |
3086 | break; | |
3087 | case sizeof(u32): | |
3088 | *val = (u64)*(u32 *)ptr; | |
3089 | break; | |
3090 | case sizeof(u64): | |
3091 | *val = *(u64 *)ptr; | |
3092 | break; | |
3093 | default: | |
3094 | return -EINVAL; | |
3095 | } | |
3096 | return 0; | |
3097 | } | |
3098 | ||
9e15db66 AS |
3099 | static int check_ptr_to_btf_access(struct bpf_verifier_env *env, |
3100 | struct bpf_reg_state *regs, | |
3101 | int regno, int off, int size, | |
3102 | enum bpf_access_type atype, | |
3103 | int value_regno) | |
3104 | { | |
3105 | struct bpf_reg_state *reg = regs + regno; | |
3106 | const struct btf_type *t = btf_type_by_id(btf_vmlinux, reg->btf_id); | |
3107 | const char *tname = btf_name_by_offset(btf_vmlinux, t->name_off); | |
3108 | u32 btf_id; | |
3109 | int ret; | |
3110 | ||
9e15db66 AS |
3111 | if (off < 0) { |
3112 | verbose(env, | |
3113 | "R%d is ptr_%s invalid negative access: off=%d\n", | |
3114 | regno, tname, off); | |
3115 | return -EACCES; | |
3116 | } | |
3117 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
3118 | char tn_buf[48]; | |
3119 | ||
3120 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3121 | verbose(env, | |
3122 | "R%d is ptr_%s invalid variable offset: off=%d, var_off=%s\n", | |
3123 | regno, tname, off, tn_buf); | |
3124 | return -EACCES; | |
3125 | } | |
3126 | ||
27ae7997 MKL |
3127 | if (env->ops->btf_struct_access) { |
3128 | ret = env->ops->btf_struct_access(&env->log, t, off, size, | |
3129 | atype, &btf_id); | |
3130 | } else { | |
3131 | if (atype != BPF_READ) { | |
3132 | verbose(env, "only read is supported\n"); | |
3133 | return -EACCES; | |
3134 | } | |
3135 | ||
3136 | ret = btf_struct_access(&env->log, t, off, size, atype, | |
3137 | &btf_id); | |
3138 | } | |
3139 | ||
9e15db66 AS |
3140 | if (ret < 0) |
3141 | return ret; | |
3142 | ||
8ff3571f | 3143 | if (atype == BPF_READ && value_regno >= 0) { |
27ae7997 MKL |
3144 | if (ret == SCALAR_VALUE) { |
3145 | mark_reg_unknown(env, regs, value_regno); | |
3146 | return 0; | |
3147 | } | |
3148 | mark_reg_known_zero(env, regs, value_regno); | |
3149 | regs[value_regno].type = PTR_TO_BTF_ID; | |
3150 | regs[value_regno].btf_id = btf_id; | |
9e15db66 | 3151 | } |
27ae7997 | 3152 | |
9e15db66 AS |
3153 | return 0; |
3154 | } | |
3155 | ||
17a52670 AS |
3156 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
3157 | * if t==write, value_regno is a register which value is stored into memory | |
3158 | * if t==read, value_regno is a register which will receive the value from memory | |
3159 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
3160 | * if t==read && value_regno==-1, don't care what we read from memory | |
3161 | */ | |
ca369602 DB |
3162 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
3163 | int off, int bpf_size, enum bpf_access_type t, | |
3164 | int value_regno, bool strict_alignment_once) | |
17a52670 | 3165 | { |
638f5b90 AS |
3166 | struct bpf_reg_state *regs = cur_regs(env); |
3167 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 3168 | struct bpf_func_state *state; |
17a52670 AS |
3169 | int size, err = 0; |
3170 | ||
3171 | size = bpf_size_to_bytes(bpf_size); | |
3172 | if (size < 0) | |
3173 | return size; | |
3174 | ||
f1174f77 | 3175 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 3176 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
3177 | if (err) |
3178 | return err; | |
17a52670 | 3179 | |
f1174f77 EC |
3180 | /* for access checks, reg->off is just part of off */ |
3181 | off += reg->off; | |
3182 | ||
3183 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
3184 | if (t == BPF_WRITE && value_regno >= 0 && |
3185 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 3186 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
3187 | return -EACCES; |
3188 | } | |
591fe988 DB |
3189 | err = check_map_access_type(env, regno, off, size, t); |
3190 | if (err) | |
3191 | return err; | |
9fd29c08 | 3192 | err = check_map_access(env, regno, off, size, false); |
a23740ec AN |
3193 | if (!err && t == BPF_READ && value_regno >= 0) { |
3194 | struct bpf_map *map = reg->map_ptr; | |
3195 | ||
3196 | /* if map is read-only, track its contents as scalars */ | |
3197 | if (tnum_is_const(reg->var_off) && | |
3198 | bpf_map_is_rdonly(map) && | |
3199 | map->ops->map_direct_value_addr) { | |
3200 | int map_off = off + reg->var_off.value; | |
3201 | u64 val = 0; | |
3202 | ||
3203 | err = bpf_map_direct_read(map, map_off, size, | |
3204 | &val); | |
3205 | if (err) | |
3206 | return err; | |
3207 | ||
3208 | regs[value_regno].type = SCALAR_VALUE; | |
3209 | __mark_reg_known(®s[value_regno], val); | |
3210 | } else { | |
3211 | mark_reg_unknown(env, regs, value_regno); | |
3212 | } | |
3213 | } | |
1a0dc1ac | 3214 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 3215 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
9e15db66 | 3216 | u32 btf_id = 0; |
19de99f7 | 3217 | |
1be7f75d AS |
3218 | if (t == BPF_WRITE && value_regno >= 0 && |
3219 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 3220 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
3221 | return -EACCES; |
3222 | } | |
f1174f77 | 3223 | |
58990d1f DB |
3224 | err = check_ctx_reg(env, reg, regno); |
3225 | if (err < 0) | |
3226 | return err; | |
3227 | ||
9e15db66 AS |
3228 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type, &btf_id); |
3229 | if (err) | |
3230 | verbose_linfo(env, insn_idx, "; "); | |
969bf05e | 3231 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 3232 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
3233 | * PTR_TO_PACKET[_META,_END]. In the latter |
3234 | * case, we know the offset is zero. | |
f1174f77 | 3235 | */ |
46f8bc92 | 3236 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 3237 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 3238 | } else { |
638f5b90 | 3239 | mark_reg_known_zero(env, regs, |
61bd5218 | 3240 | value_regno); |
46f8bc92 MKL |
3241 | if (reg_type_may_be_null(reg_type)) |
3242 | regs[value_regno].id = ++env->id_gen; | |
5327ed3d JW |
3243 | /* A load of ctx field could have different |
3244 | * actual load size with the one encoded in the | |
3245 | * insn. When the dst is PTR, it is for sure not | |
3246 | * a sub-register. | |
3247 | */ | |
3248 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
b121b341 YS |
3249 | if (reg_type == PTR_TO_BTF_ID || |
3250 | reg_type == PTR_TO_BTF_ID_OR_NULL) | |
9e15db66 | 3251 | regs[value_regno].btf_id = btf_id; |
46f8bc92 | 3252 | } |
638f5b90 | 3253 | regs[value_regno].type = reg_type; |
969bf05e | 3254 | } |
17a52670 | 3255 | |
f1174f77 | 3256 | } else if (reg->type == PTR_TO_STACK) { |
f1174f77 | 3257 | off += reg->var_off.value; |
e4298d25 DB |
3258 | err = check_stack_access(env, reg, off, size); |
3259 | if (err) | |
3260 | return err; | |
8726679a | 3261 | |
f4d7e40a AS |
3262 | state = func(env, reg); |
3263 | err = update_stack_depth(env, state, off); | |
3264 | if (err) | |
3265 | return err; | |
8726679a | 3266 | |
638f5b90 | 3267 | if (t == BPF_WRITE) |
61bd5218 | 3268 | err = check_stack_write(env, state, off, size, |
af86ca4e | 3269 | value_regno, insn_idx); |
638f5b90 | 3270 | else |
61bd5218 JK |
3271 | err = check_stack_read(env, state, off, size, |
3272 | value_regno); | |
de8f3a83 | 3273 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 3274 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 3275 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
3276 | return -EACCES; |
3277 | } | |
4acf6c0b BB |
3278 | if (t == BPF_WRITE && value_regno >= 0 && |
3279 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
3280 | verbose(env, "R%d leaks addr into packet\n", |
3281 | value_regno); | |
4acf6c0b BB |
3282 | return -EACCES; |
3283 | } | |
9fd29c08 | 3284 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 3285 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 3286 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
3287 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
3288 | if (t == BPF_WRITE && value_regno >= 0 && | |
3289 | is_pointer_value(env, value_regno)) { | |
3290 | verbose(env, "R%d leaks addr into flow keys\n", | |
3291 | value_regno); | |
3292 | return -EACCES; | |
3293 | } | |
3294 | ||
3295 | err = check_flow_keys_access(env, off, size); | |
3296 | if (!err && t == BPF_READ && value_regno >= 0) | |
3297 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 3298 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 3299 | if (t == BPF_WRITE) { |
46f8bc92 MKL |
3300 | verbose(env, "R%d cannot write into %s\n", |
3301 | regno, reg_type_str[reg->type]); | |
c64b7983 JS |
3302 | return -EACCES; |
3303 | } | |
5f456649 | 3304 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
3305 | if (!err && value_regno >= 0) |
3306 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
3307 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
3308 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
3309 | if (!err && t == BPF_READ && value_regno >= 0) | |
3310 | mark_reg_unknown(env, regs, value_regno); | |
9e15db66 AS |
3311 | } else if (reg->type == PTR_TO_BTF_ID) { |
3312 | err = check_ptr_to_btf_access(env, regs, regno, off, size, t, | |
3313 | value_regno); | |
17a52670 | 3314 | } else { |
61bd5218 JK |
3315 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
3316 | reg_type_str[reg->type]); | |
17a52670 AS |
3317 | return -EACCES; |
3318 | } | |
969bf05e | 3319 | |
f1174f77 | 3320 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 3321 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 3322 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 3323 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 3324 | } |
17a52670 AS |
3325 | return err; |
3326 | } | |
3327 | ||
31fd8581 | 3328 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 3329 | { |
17a52670 AS |
3330 | int err; |
3331 | ||
3332 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
3333 | insn->imm != 0) { | |
61bd5218 | 3334 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
3335 | return -EINVAL; |
3336 | } | |
3337 | ||
3338 | /* check src1 operand */ | |
dc503a8a | 3339 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
3340 | if (err) |
3341 | return err; | |
3342 | ||
3343 | /* check src2 operand */ | |
dc503a8a | 3344 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
3345 | if (err) |
3346 | return err; | |
3347 | ||
6bdf6abc | 3348 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 3349 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
3350 | return -EACCES; |
3351 | } | |
3352 | ||
ca369602 | 3353 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 3354 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
3355 | is_flow_key_reg(env, insn->dst_reg) || |
3356 | is_sk_reg(env, insn->dst_reg)) { | |
ca369602 | 3357 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", |
2a159c6f DB |
3358 | insn->dst_reg, |
3359 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
3360 | return -EACCES; |
3361 | } | |
3362 | ||
17a52670 | 3363 | /* check whether atomic_add can read the memory */ |
31fd8581 | 3364 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 3365 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
3366 | if (err) |
3367 | return err; | |
3368 | ||
3369 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 3370 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 3371 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
3372 | } |
3373 | ||
2011fccf AI |
3374 | static int __check_stack_boundary(struct bpf_verifier_env *env, u32 regno, |
3375 | int off, int access_size, | |
3376 | bool zero_size_allowed) | |
3377 | { | |
3378 | struct bpf_reg_state *reg = reg_state(env, regno); | |
3379 | ||
3380 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || | |
3381 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { | |
3382 | if (tnum_is_const(reg->var_off)) { | |
3383 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", | |
3384 | regno, off, access_size); | |
3385 | } else { | |
3386 | char tn_buf[48]; | |
3387 | ||
3388 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3389 | verbose(env, "invalid stack type R%d var_off=%s access_size=%d\n", | |
3390 | regno, tn_buf, access_size); | |
3391 | } | |
3392 | return -EACCES; | |
3393 | } | |
3394 | return 0; | |
3395 | } | |
3396 | ||
17a52670 AS |
3397 | /* when register 'regno' is passed into function that will read 'access_size' |
3398 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
3399 | * and all elements of stack are initialized. |
3400 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
3401 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 3402 | */ |
58e2af8b | 3403 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
3404 | int access_size, bool zero_size_allowed, |
3405 | struct bpf_call_arg_meta *meta) | |
17a52670 | 3406 | { |
2a159c6f | 3407 | struct bpf_reg_state *reg = reg_state(env, regno); |
f4d7e40a | 3408 | struct bpf_func_state *state = func(env, reg); |
f7cf25b2 | 3409 | int err, min_off, max_off, i, j, slot, spi; |
17a52670 | 3410 | |
914cb781 | 3411 | if (reg->type != PTR_TO_STACK) { |
f1174f77 | 3412 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 3413 | if (zero_size_allowed && access_size == 0 && |
914cb781 | 3414 | register_is_null(reg)) |
8e2fe1d9 DB |
3415 | return 0; |
3416 | ||
61bd5218 | 3417 | verbose(env, "R%d type=%s expected=%s\n", regno, |
914cb781 | 3418 | reg_type_str[reg->type], |
8e2fe1d9 | 3419 | reg_type_str[PTR_TO_STACK]); |
17a52670 | 3420 | return -EACCES; |
8e2fe1d9 | 3421 | } |
17a52670 | 3422 | |
2011fccf AI |
3423 | if (tnum_is_const(reg->var_off)) { |
3424 | min_off = max_off = reg->var_off.value + reg->off; | |
3425 | err = __check_stack_boundary(env, regno, min_off, access_size, | |
3426 | zero_size_allowed); | |
3427 | if (err) | |
3428 | return err; | |
3429 | } else { | |
088ec26d AI |
3430 | /* Variable offset is prohibited for unprivileged mode for |
3431 | * simplicity since it requires corresponding support in | |
3432 | * Spectre masking for stack ALU. | |
3433 | * See also retrieve_ptr_limit(). | |
3434 | */ | |
2c78ee89 | 3435 | if (!env->bypass_spec_v1) { |
088ec26d | 3436 | char tn_buf[48]; |
f1174f77 | 3437 | |
088ec26d AI |
3438 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
3439 | verbose(env, "R%d indirect variable offset stack access prohibited for !root, var_off=%s\n", | |
3440 | regno, tn_buf); | |
3441 | return -EACCES; | |
3442 | } | |
f2bcd05e AI |
3443 | /* Only initialized buffer on stack is allowed to be accessed |
3444 | * with variable offset. With uninitialized buffer it's hard to | |
3445 | * guarantee that whole memory is marked as initialized on | |
3446 | * helper return since specific bounds are unknown what may | |
3447 | * cause uninitialized stack leaking. | |
3448 | */ | |
3449 | if (meta && meta->raw_mode) | |
3450 | meta = NULL; | |
3451 | ||
107c26a7 AI |
3452 | if (reg->smax_value >= BPF_MAX_VAR_OFF || |
3453 | reg->smax_value <= -BPF_MAX_VAR_OFF) { | |
3454 | verbose(env, "R%d unbounded indirect variable offset stack access\n", | |
3455 | regno); | |
3456 | return -EACCES; | |
3457 | } | |
2011fccf | 3458 | min_off = reg->smin_value + reg->off; |
107c26a7 | 3459 | max_off = reg->smax_value + reg->off; |
2011fccf AI |
3460 | err = __check_stack_boundary(env, regno, min_off, access_size, |
3461 | zero_size_allowed); | |
107c26a7 AI |
3462 | if (err) { |
3463 | verbose(env, "R%d min value is outside of stack bound\n", | |
3464 | regno); | |
2011fccf | 3465 | return err; |
107c26a7 | 3466 | } |
2011fccf AI |
3467 | err = __check_stack_boundary(env, regno, max_off, access_size, |
3468 | zero_size_allowed); | |
107c26a7 AI |
3469 | if (err) { |
3470 | verbose(env, "R%d max value is outside of stack bound\n", | |
3471 | regno); | |
2011fccf | 3472 | return err; |
107c26a7 | 3473 | } |
17a52670 AS |
3474 | } |
3475 | ||
435faee1 DB |
3476 | if (meta && meta->raw_mode) { |
3477 | meta->access_size = access_size; | |
3478 | meta->regno = regno; | |
3479 | return 0; | |
3480 | } | |
3481 | ||
2011fccf | 3482 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
3483 | u8 *stype; |
3484 | ||
2011fccf | 3485 | slot = -i - 1; |
638f5b90 | 3486 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
3487 | if (state->allocated_stack <= slot) |
3488 | goto err; | |
3489 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
3490 | if (*stype == STACK_MISC) | |
3491 | goto mark; | |
3492 | if (*stype == STACK_ZERO) { | |
3493 | /* helper can write anything into the stack */ | |
3494 | *stype = STACK_MISC; | |
3495 | goto mark; | |
17a52670 | 3496 | } |
1d68f22b YS |
3497 | |
3498 | if (state->stack[spi].slot_type[0] == STACK_SPILL && | |
3499 | state->stack[spi].spilled_ptr.type == PTR_TO_BTF_ID) | |
3500 | goto mark; | |
3501 | ||
f7cf25b2 AS |
3502 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
3503 | state->stack[spi].spilled_ptr.type == SCALAR_VALUE) { | |
f54c7898 | 3504 | __mark_reg_unknown(env, &state->stack[spi].spilled_ptr); |
f7cf25b2 AS |
3505 | for (j = 0; j < BPF_REG_SIZE; j++) |
3506 | state->stack[spi].slot_type[j] = STACK_MISC; | |
3507 | goto mark; | |
3508 | } | |
3509 | ||
cc2b14d5 | 3510 | err: |
2011fccf AI |
3511 | if (tnum_is_const(reg->var_off)) { |
3512 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", | |
3513 | min_off, i - min_off, access_size); | |
3514 | } else { | |
3515 | char tn_buf[48]; | |
3516 | ||
3517 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
3518 | verbose(env, "invalid indirect read from stack var_off %s+%d size %d\n", | |
3519 | tn_buf, i - min_off, access_size); | |
3520 | } | |
cc2b14d5 AS |
3521 | return -EACCES; |
3522 | mark: | |
3523 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
3524 | * the whole slot to be marked as 'read' | |
3525 | */ | |
679c782d | 3526 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
3527 | state->stack[spi].spilled_ptr.parent, |
3528 | REG_LIVE_READ64); | |
17a52670 | 3529 | } |
2011fccf | 3530 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
3531 | } |
3532 | ||
06c1c049 GB |
3533 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
3534 | int access_size, bool zero_size_allowed, | |
3535 | struct bpf_call_arg_meta *meta) | |
3536 | { | |
638f5b90 | 3537 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 3538 | |
f1174f77 | 3539 | switch (reg->type) { |
06c1c049 | 3540 | case PTR_TO_PACKET: |
de8f3a83 | 3541 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
3542 | return check_packet_access(env, regno, reg->off, access_size, |
3543 | zero_size_allowed); | |
06c1c049 | 3544 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
3545 | if (check_map_access_type(env, regno, reg->off, access_size, |
3546 | meta && meta->raw_mode ? BPF_WRITE : | |
3547 | BPF_READ)) | |
3548 | return -EACCES; | |
9fd29c08 YS |
3549 | return check_map_access(env, regno, reg->off, access_size, |
3550 | zero_size_allowed); | |
f1174f77 | 3551 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
3552 | return check_stack_boundary(env, regno, access_size, |
3553 | zero_size_allowed, meta); | |
3554 | } | |
3555 | } | |
3556 | ||
d83525ca AS |
3557 | /* Implementation details: |
3558 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL | |
3559 | * Two bpf_map_lookups (even with the same key) will have different reg->id. | |
3560 | * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after | |
3561 | * value_or_null->value transition, since the verifier only cares about | |
3562 | * the range of access to valid map value pointer and doesn't care about actual | |
3563 | * address of the map element. | |
3564 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps | |
3565 | * reg->id > 0 after value_or_null->value transition. By doing so | |
3566 | * two bpf_map_lookups will be considered two different pointers that | |
3567 | * point to different bpf_spin_locks. | |
3568 | * The verifier allows taking only one bpf_spin_lock at a time to avoid | |
3569 | * dead-locks. | |
3570 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
3571 | * reg_is_refcounted() logic. The verifier needs to remember only | |
3572 | * one spin_lock instead of array of acquired_refs. | |
3573 | * cur_state->active_spin_lock remembers which map value element got locked | |
3574 | * and clears it after bpf_spin_unlock. | |
3575 | */ | |
3576 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
3577 | bool is_lock) | |
3578 | { | |
3579 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
3580 | struct bpf_verifier_state *cur = env->cur_state; | |
3581 | bool is_const = tnum_is_const(reg->var_off); | |
3582 | struct bpf_map *map = reg->map_ptr; | |
3583 | u64 val = reg->var_off.value; | |
3584 | ||
3585 | if (reg->type != PTR_TO_MAP_VALUE) { | |
3586 | verbose(env, "R%d is not a pointer to map_value\n", regno); | |
3587 | return -EINVAL; | |
3588 | } | |
3589 | if (!is_const) { | |
3590 | verbose(env, | |
3591 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
3592 | regno); | |
3593 | return -EINVAL; | |
3594 | } | |
3595 | if (!map->btf) { | |
3596 | verbose(env, | |
3597 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
3598 | map->name); | |
3599 | return -EINVAL; | |
3600 | } | |
3601 | if (!map_value_has_spin_lock(map)) { | |
3602 | if (map->spin_lock_off == -E2BIG) | |
3603 | verbose(env, | |
3604 | "map '%s' has more than one 'struct bpf_spin_lock'\n", | |
3605 | map->name); | |
3606 | else if (map->spin_lock_off == -ENOENT) | |
3607 | verbose(env, | |
3608 | "map '%s' doesn't have 'struct bpf_spin_lock'\n", | |
3609 | map->name); | |
3610 | else | |
3611 | verbose(env, | |
3612 | "map '%s' is not a struct type or bpf_spin_lock is mangled\n", | |
3613 | map->name); | |
3614 | return -EINVAL; | |
3615 | } | |
3616 | if (map->spin_lock_off != val + reg->off) { | |
3617 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock'\n", | |
3618 | val + reg->off); | |
3619 | return -EINVAL; | |
3620 | } | |
3621 | if (is_lock) { | |
3622 | if (cur->active_spin_lock) { | |
3623 | verbose(env, | |
3624 | "Locking two bpf_spin_locks are not allowed\n"); | |
3625 | return -EINVAL; | |
3626 | } | |
3627 | cur->active_spin_lock = reg->id; | |
3628 | } else { | |
3629 | if (!cur->active_spin_lock) { | |
3630 | verbose(env, "bpf_spin_unlock without taking a lock\n"); | |
3631 | return -EINVAL; | |
3632 | } | |
3633 | if (cur->active_spin_lock != reg->id) { | |
3634 | verbose(env, "bpf_spin_unlock of different lock\n"); | |
3635 | return -EINVAL; | |
3636 | } | |
3637 | cur->active_spin_lock = 0; | |
3638 | } | |
3639 | return 0; | |
3640 | } | |
3641 | ||
90133415 DB |
3642 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
3643 | { | |
3644 | return type == ARG_PTR_TO_MEM || | |
3645 | type == ARG_PTR_TO_MEM_OR_NULL || | |
3646 | type == ARG_PTR_TO_UNINIT_MEM; | |
3647 | } | |
3648 | ||
3649 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
3650 | { | |
3651 | return type == ARG_CONST_SIZE || | |
3652 | type == ARG_CONST_SIZE_OR_ZERO; | |
3653 | } | |
3654 | ||
57c3bb72 AI |
3655 | static bool arg_type_is_int_ptr(enum bpf_arg_type type) |
3656 | { | |
3657 | return type == ARG_PTR_TO_INT || | |
3658 | type == ARG_PTR_TO_LONG; | |
3659 | } | |
3660 | ||
3661 | static int int_ptr_type_to_size(enum bpf_arg_type type) | |
3662 | { | |
3663 | if (type == ARG_PTR_TO_INT) | |
3664 | return sizeof(u32); | |
3665 | else if (type == ARG_PTR_TO_LONG) | |
3666 | return sizeof(u64); | |
3667 | ||
3668 | return -EINVAL; | |
3669 | } | |
3670 | ||
58e2af8b | 3671 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
3672 | enum bpf_arg_type arg_type, |
3673 | struct bpf_call_arg_meta *meta) | |
17a52670 | 3674 | { |
638f5b90 | 3675 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 3676 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
3677 | int err = 0; |
3678 | ||
80f1d68c | 3679 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
3680 | return 0; |
3681 | ||
dc503a8a EC |
3682 | err = check_reg_arg(env, regno, SRC_OP); |
3683 | if (err) | |
3684 | return err; | |
17a52670 | 3685 | |
1be7f75d AS |
3686 | if (arg_type == ARG_ANYTHING) { |
3687 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
3688 | verbose(env, "R%d leaks addr into helper function\n", |
3689 | regno); | |
1be7f75d AS |
3690 | return -EACCES; |
3691 | } | |
80f1d68c | 3692 | return 0; |
1be7f75d | 3693 | } |
80f1d68c | 3694 | |
de8f3a83 | 3695 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 3696 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 3697 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
3698 | return -EACCES; |
3699 | } | |
3700 | ||
8e2fe1d9 | 3701 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
2ea864c5 | 3702 | arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
3703 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE || |
3704 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) { | |
17a52670 | 3705 | expected_type = PTR_TO_STACK; |
6ac99e8f MKL |
3706 | if (register_is_null(reg) && |
3707 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) | |
3708 | /* final test in check_stack_boundary() */; | |
3709 | else if (!type_is_pkt_pointer(type) && | |
3710 | type != PTR_TO_MAP_VALUE && | |
3711 | type != expected_type) | |
6841de8b | 3712 | goto err_type; |
39f19ebb AS |
3713 | } else if (arg_type == ARG_CONST_SIZE || |
3714 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
3715 | expected_type = SCALAR_VALUE; |
3716 | if (type != expected_type) | |
6841de8b | 3717 | goto err_type; |
17a52670 AS |
3718 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
3719 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
3720 | if (type != expected_type) |
3721 | goto err_type; | |
f318903c DB |
3722 | } else if (arg_type == ARG_PTR_TO_CTX || |
3723 | arg_type == ARG_PTR_TO_CTX_OR_NULL) { | |
608cd71a | 3724 | expected_type = PTR_TO_CTX; |
f318903c DB |
3725 | if (!(register_is_null(reg) && |
3726 | arg_type == ARG_PTR_TO_CTX_OR_NULL)) { | |
3727 | if (type != expected_type) | |
3728 | goto err_type; | |
3729 | err = check_ctx_reg(env, reg, regno); | |
3730 | if (err < 0) | |
3731 | return err; | |
3732 | } | |
46f8bc92 MKL |
3733 | } else if (arg_type == ARG_PTR_TO_SOCK_COMMON) { |
3734 | expected_type = PTR_TO_SOCK_COMMON; | |
3735 | /* Any sk pointer can be ARG_PTR_TO_SOCK_COMMON */ | |
3736 | if (!type_is_sk_pointer(type)) | |
3737 | goto err_type; | |
1b986589 MKL |
3738 | if (reg->ref_obj_id) { |
3739 | if (meta->ref_obj_id) { | |
3740 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
3741 | regno, reg->ref_obj_id, | |
3742 | meta->ref_obj_id); | |
3743 | return -EFAULT; | |
3744 | } | |
3745 | meta->ref_obj_id = reg->ref_obj_id; | |
fd978bf7 | 3746 | } |
6ac99e8f MKL |
3747 | } else if (arg_type == ARG_PTR_TO_SOCKET) { |
3748 | expected_type = PTR_TO_SOCKET; | |
3749 | if (type != expected_type) | |
3750 | goto err_type; | |
a7658e1a AS |
3751 | } else if (arg_type == ARG_PTR_TO_BTF_ID) { |
3752 | expected_type = PTR_TO_BTF_ID; | |
3753 | if (type != expected_type) | |
3754 | goto err_type; | |
3755 | if (reg->btf_id != meta->btf_id) { | |
3756 | verbose(env, "Helper has type %s got %s in R%d\n", | |
3757 | kernel_type_name(meta->btf_id), | |
3758 | kernel_type_name(reg->btf_id), regno); | |
3759 | ||
3760 | return -EACCES; | |
3761 | } | |
3762 | if (!tnum_is_const(reg->var_off) || reg->var_off.value || reg->off) { | |
3763 | verbose(env, "R%d is a pointer to in-kernel struct with non-zero offset\n", | |
3764 | regno); | |
3765 | return -EACCES; | |
3766 | } | |
d83525ca AS |
3767 | } else if (arg_type == ARG_PTR_TO_SPIN_LOCK) { |
3768 | if (meta->func_id == BPF_FUNC_spin_lock) { | |
3769 | if (process_spin_lock(env, regno, true)) | |
3770 | return -EACCES; | |
3771 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { | |
3772 | if (process_spin_lock(env, regno, false)) | |
3773 | return -EACCES; | |
3774 | } else { | |
3775 | verbose(env, "verifier internal error\n"); | |
3776 | return -EFAULT; | |
3777 | } | |
90133415 | 3778 | } else if (arg_type_is_mem_ptr(arg_type)) { |
8e2fe1d9 DB |
3779 | expected_type = PTR_TO_STACK; |
3780 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 3781 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
3782 | * happens during stack boundary checking. |
3783 | */ | |
914cb781 | 3784 | if (register_is_null(reg) && |
db1ac496 | 3785 | arg_type == ARG_PTR_TO_MEM_OR_NULL) |
6841de8b | 3786 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
3787 | else if (!type_is_pkt_pointer(type) && |
3788 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 3789 | type != expected_type) |
6841de8b | 3790 | goto err_type; |
39f19ebb | 3791 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
57c3bb72 AI |
3792 | } else if (arg_type_is_int_ptr(arg_type)) { |
3793 | expected_type = PTR_TO_STACK; | |
3794 | if (!type_is_pkt_pointer(type) && | |
3795 | type != PTR_TO_MAP_VALUE && | |
3796 | type != expected_type) | |
3797 | goto err_type; | |
17a52670 | 3798 | } else { |
61bd5218 | 3799 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
3800 | return -EFAULT; |
3801 | } | |
3802 | ||
17a52670 AS |
3803 | if (arg_type == ARG_CONST_MAP_PTR) { |
3804 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 3805 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
3806 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
3807 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
3808 | * check that [key, key + map->key_size) are within | |
3809 | * stack limits and initialized | |
3810 | */ | |
33ff9823 | 3811 | if (!meta->map_ptr) { |
17a52670 AS |
3812 | /* in function declaration map_ptr must come before |
3813 | * map_key, so that it's verified and known before | |
3814 | * we have to check map_key here. Otherwise it means | |
3815 | * that kernel subsystem misconfigured verifier | |
3816 | */ | |
61bd5218 | 3817 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
3818 | return -EACCES; |
3819 | } | |
d71962f3 PC |
3820 | err = check_helper_mem_access(env, regno, |
3821 | meta->map_ptr->key_size, false, | |
3822 | NULL); | |
2ea864c5 | 3823 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
3824 | (arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL && |
3825 | !register_is_null(reg)) || | |
2ea864c5 | 3826 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { |
17a52670 AS |
3827 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
3828 | * check [value, value + map->value_size) validity | |
3829 | */ | |
33ff9823 | 3830 | if (!meta->map_ptr) { |
17a52670 | 3831 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 3832 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
3833 | return -EACCES; |
3834 | } | |
2ea864c5 | 3835 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE); |
d71962f3 PC |
3836 | err = check_helper_mem_access(env, regno, |
3837 | meta->map_ptr->value_size, false, | |
2ea864c5 | 3838 | meta); |
90133415 | 3839 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 3840 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 3841 | |
10060503 JF |
3842 | /* This is used to refine r0 return value bounds for helpers |
3843 | * that enforce this value as an upper bound on return values. | |
3844 | * See do_refine_retval_range() for helpers that can refine | |
3845 | * the return value. C type of helper is u32 so we pull register | |
3846 | * bound from umax_value however, if negative verifier errors | |
3847 | * out. Only upper bounds can be learned because retval is an | |
3848 | * int type and negative retvals are allowed. | |
849fa506 | 3849 | */ |
10060503 | 3850 | meta->msize_max_value = reg->umax_value; |
849fa506 | 3851 | |
f1174f77 EC |
3852 | /* The register is SCALAR_VALUE; the access check |
3853 | * happens using its boundaries. | |
06c1c049 | 3854 | */ |
f1174f77 | 3855 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
3856 | /* For unprivileged variable accesses, disable raw |
3857 | * mode so that the program is required to | |
3858 | * initialize all the memory that the helper could | |
3859 | * just partially fill up. | |
3860 | */ | |
3861 | meta = NULL; | |
3862 | ||
b03c9f9f | 3863 | if (reg->smin_value < 0) { |
61bd5218 | 3864 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
3865 | regno); |
3866 | return -EACCES; | |
3867 | } | |
06c1c049 | 3868 | |
b03c9f9f | 3869 | if (reg->umin_value == 0) { |
f1174f77 EC |
3870 | err = check_helper_mem_access(env, regno - 1, 0, |
3871 | zero_size_allowed, | |
3872 | meta); | |
06c1c049 GB |
3873 | if (err) |
3874 | return err; | |
06c1c049 | 3875 | } |
f1174f77 | 3876 | |
b03c9f9f | 3877 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 3878 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
3879 | regno); |
3880 | return -EACCES; | |
3881 | } | |
3882 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 3883 | reg->umax_value, |
f1174f77 | 3884 | zero_size_allowed, meta); |
b5dc0163 AS |
3885 | if (!err) |
3886 | err = mark_chain_precision(env, regno); | |
57c3bb72 AI |
3887 | } else if (arg_type_is_int_ptr(arg_type)) { |
3888 | int size = int_ptr_type_to_size(arg_type); | |
3889 | ||
3890 | err = check_helper_mem_access(env, regno, size, false, meta); | |
3891 | if (err) | |
3892 | return err; | |
3893 | err = check_ptr_alignment(env, reg, 0, size, true); | |
17a52670 AS |
3894 | } |
3895 | ||
3896 | return err; | |
6841de8b | 3897 | err_type: |
61bd5218 | 3898 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
3899 | reg_type_str[type], reg_type_str[expected_type]); |
3900 | return -EACCES; | |
17a52670 AS |
3901 | } |
3902 | ||
61bd5218 JK |
3903 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
3904 | struct bpf_map *map, int func_id) | |
35578d79 | 3905 | { |
35578d79 KX |
3906 | if (!map) |
3907 | return 0; | |
3908 | ||
6aff67c8 AS |
3909 | /* We need a two way check, first is from map perspective ... */ |
3910 | switch (map->map_type) { | |
3911 | case BPF_MAP_TYPE_PROG_ARRAY: | |
3912 | if (func_id != BPF_FUNC_tail_call) | |
3913 | goto error; | |
3914 | break; | |
3915 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
3916 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca | 3917 | func_id != BPF_FUNC_perf_event_output && |
a7658e1a | 3918 | func_id != BPF_FUNC_skb_output && |
d831ee84 EC |
3919 | func_id != BPF_FUNC_perf_event_read_value && |
3920 | func_id != BPF_FUNC_xdp_output) | |
6aff67c8 AS |
3921 | goto error; |
3922 | break; | |
3923 | case BPF_MAP_TYPE_STACK_TRACE: | |
3924 | if (func_id != BPF_FUNC_get_stackid) | |
3925 | goto error; | |
3926 | break; | |
4ed8ec52 | 3927 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 3928 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 3929 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
3930 | goto error; |
3931 | break; | |
cd339431 | 3932 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 3933 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
3934 | if (func_id != BPF_FUNC_get_local_storage) |
3935 | goto error; | |
3936 | break; | |
546ac1ff | 3937 | case BPF_MAP_TYPE_DEVMAP: |
6f9d451a | 3938 | case BPF_MAP_TYPE_DEVMAP_HASH: |
0cdbb4b0 THJ |
3939 | if (func_id != BPF_FUNC_redirect_map && |
3940 | func_id != BPF_FUNC_map_lookup_elem) | |
546ac1ff JF |
3941 | goto error; |
3942 | break; | |
fbfc504a BT |
3943 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
3944 | * appear. | |
3945 | */ | |
6710e112 JDB |
3946 | case BPF_MAP_TYPE_CPUMAP: |
3947 | if (func_id != BPF_FUNC_redirect_map) | |
3948 | goto error; | |
3949 | break; | |
fada7fdc JL |
3950 | case BPF_MAP_TYPE_XSKMAP: |
3951 | if (func_id != BPF_FUNC_redirect_map && | |
3952 | func_id != BPF_FUNC_map_lookup_elem) | |
3953 | goto error; | |
3954 | break; | |
56f668df | 3955 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 3956 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
3957 | if (func_id != BPF_FUNC_map_lookup_elem) |
3958 | goto error; | |
16a43625 | 3959 | break; |
174a79ff JF |
3960 | case BPF_MAP_TYPE_SOCKMAP: |
3961 | if (func_id != BPF_FUNC_sk_redirect_map && | |
3962 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb | 3963 | func_id != BPF_FUNC_map_delete_elem && |
9fed9000 | 3964 | func_id != BPF_FUNC_msg_redirect_map && |
64d85290 JS |
3965 | func_id != BPF_FUNC_sk_select_reuseport && |
3966 | func_id != BPF_FUNC_map_lookup_elem) | |
174a79ff JF |
3967 | goto error; |
3968 | break; | |
81110384 JF |
3969 | case BPF_MAP_TYPE_SOCKHASH: |
3970 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
3971 | func_id != BPF_FUNC_sock_hash_update && | |
3972 | func_id != BPF_FUNC_map_delete_elem && | |
9fed9000 | 3973 | func_id != BPF_FUNC_msg_redirect_hash && |
64d85290 JS |
3974 | func_id != BPF_FUNC_sk_select_reuseport && |
3975 | func_id != BPF_FUNC_map_lookup_elem) | |
81110384 JF |
3976 | goto error; |
3977 | break; | |
2dbb9b9e MKL |
3978 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
3979 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
3980 | goto error; | |
3981 | break; | |
f1a2e44a MV |
3982 | case BPF_MAP_TYPE_QUEUE: |
3983 | case BPF_MAP_TYPE_STACK: | |
3984 | if (func_id != BPF_FUNC_map_peek_elem && | |
3985 | func_id != BPF_FUNC_map_pop_elem && | |
3986 | func_id != BPF_FUNC_map_push_elem) | |
3987 | goto error; | |
3988 | break; | |
6ac99e8f MKL |
3989 | case BPF_MAP_TYPE_SK_STORAGE: |
3990 | if (func_id != BPF_FUNC_sk_storage_get && | |
3991 | func_id != BPF_FUNC_sk_storage_delete) | |
3992 | goto error; | |
3993 | break; | |
6aff67c8 AS |
3994 | default: |
3995 | break; | |
3996 | } | |
3997 | ||
3998 | /* ... and second from the function itself. */ | |
3999 | switch (func_id) { | |
4000 | case BPF_FUNC_tail_call: | |
4001 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
4002 | goto error; | |
f910cefa | 4003 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
4004 | verbose(env, "tail_calls are not allowed in programs with bpf-to-bpf calls\n"); |
4005 | return -EINVAL; | |
4006 | } | |
6aff67c8 AS |
4007 | break; |
4008 | case BPF_FUNC_perf_event_read: | |
4009 | case BPF_FUNC_perf_event_output: | |
908432ca | 4010 | case BPF_FUNC_perf_event_read_value: |
a7658e1a | 4011 | case BPF_FUNC_skb_output: |
d831ee84 | 4012 | case BPF_FUNC_xdp_output: |
6aff67c8 AS |
4013 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
4014 | goto error; | |
4015 | break; | |
4016 | case BPF_FUNC_get_stackid: | |
4017 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
4018 | goto error; | |
4019 | break; | |
60d20f91 | 4020 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 4021 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
4022 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
4023 | goto error; | |
4024 | break; | |
97f91a7c | 4025 | case BPF_FUNC_redirect_map: |
9c270af3 | 4026 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
6f9d451a | 4027 | map->map_type != BPF_MAP_TYPE_DEVMAP_HASH && |
fbfc504a BT |
4028 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
4029 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
4030 | goto error; |
4031 | break; | |
174a79ff | 4032 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 4033 | case BPF_FUNC_msg_redirect_map: |
81110384 | 4034 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
4035 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
4036 | goto error; | |
4037 | break; | |
81110384 JF |
4038 | case BPF_FUNC_sk_redirect_hash: |
4039 | case BPF_FUNC_msg_redirect_hash: | |
4040 | case BPF_FUNC_sock_hash_update: | |
4041 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
4042 | goto error; |
4043 | break; | |
cd339431 | 4044 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
4045 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
4046 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
4047 | goto error; |
4048 | break; | |
2dbb9b9e | 4049 | case BPF_FUNC_sk_select_reuseport: |
9fed9000 JS |
4050 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY && |
4051 | map->map_type != BPF_MAP_TYPE_SOCKMAP && | |
4052 | map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
2dbb9b9e MKL |
4053 | goto error; |
4054 | break; | |
f1a2e44a MV |
4055 | case BPF_FUNC_map_peek_elem: |
4056 | case BPF_FUNC_map_pop_elem: | |
4057 | case BPF_FUNC_map_push_elem: | |
4058 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
4059 | map->map_type != BPF_MAP_TYPE_STACK) | |
4060 | goto error; | |
4061 | break; | |
6ac99e8f MKL |
4062 | case BPF_FUNC_sk_storage_get: |
4063 | case BPF_FUNC_sk_storage_delete: | |
4064 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
4065 | goto error; | |
4066 | break; | |
6aff67c8 AS |
4067 | default: |
4068 | break; | |
35578d79 KX |
4069 | } |
4070 | ||
4071 | return 0; | |
6aff67c8 | 4072 | error: |
61bd5218 | 4073 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 4074 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 4075 | return -EINVAL; |
35578d79 KX |
4076 | } |
4077 | ||
90133415 | 4078 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
4079 | { |
4080 | int count = 0; | |
4081 | ||
39f19ebb | 4082 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 4083 | count++; |
39f19ebb | 4084 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 4085 | count++; |
39f19ebb | 4086 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 4087 | count++; |
39f19ebb | 4088 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 4089 | count++; |
39f19ebb | 4090 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
4091 | count++; |
4092 | ||
90133415 DB |
4093 | /* We only support one arg being in raw mode at the moment, |
4094 | * which is sufficient for the helper functions we have | |
4095 | * right now. | |
4096 | */ | |
4097 | return count <= 1; | |
4098 | } | |
4099 | ||
4100 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
4101 | enum bpf_arg_type arg_next) | |
4102 | { | |
4103 | return (arg_type_is_mem_ptr(arg_curr) && | |
4104 | !arg_type_is_mem_size(arg_next)) || | |
4105 | (!arg_type_is_mem_ptr(arg_curr) && | |
4106 | arg_type_is_mem_size(arg_next)); | |
4107 | } | |
4108 | ||
4109 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
4110 | { | |
4111 | /* bpf_xxx(..., buf, len) call will access 'len' | |
4112 | * bytes from memory 'buf'. Both arg types need | |
4113 | * to be paired, so make sure there's no buggy | |
4114 | * helper function specification. | |
4115 | */ | |
4116 | if (arg_type_is_mem_size(fn->arg1_type) || | |
4117 | arg_type_is_mem_ptr(fn->arg5_type) || | |
4118 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
4119 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
4120 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
4121 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
4122 | return false; | |
4123 | ||
4124 | return true; | |
4125 | } | |
4126 | ||
1b986589 | 4127 | static bool check_refcount_ok(const struct bpf_func_proto *fn, int func_id) |
fd978bf7 JS |
4128 | { |
4129 | int count = 0; | |
4130 | ||
1b986589 | 4131 | if (arg_type_may_be_refcounted(fn->arg1_type)) |
fd978bf7 | 4132 | count++; |
1b986589 | 4133 | if (arg_type_may_be_refcounted(fn->arg2_type)) |
fd978bf7 | 4134 | count++; |
1b986589 | 4135 | if (arg_type_may_be_refcounted(fn->arg3_type)) |
fd978bf7 | 4136 | count++; |
1b986589 | 4137 | if (arg_type_may_be_refcounted(fn->arg4_type)) |
fd978bf7 | 4138 | count++; |
1b986589 | 4139 | if (arg_type_may_be_refcounted(fn->arg5_type)) |
fd978bf7 JS |
4140 | count++; |
4141 | ||
1b986589 MKL |
4142 | /* A reference acquiring function cannot acquire |
4143 | * another refcounted ptr. | |
4144 | */ | |
64d85290 | 4145 | if (may_be_acquire_function(func_id) && count) |
1b986589 MKL |
4146 | return false; |
4147 | ||
fd978bf7 JS |
4148 | /* We only support one arg being unreferenced at the moment, |
4149 | * which is sufficient for the helper functions we have right now. | |
4150 | */ | |
4151 | return count <= 1; | |
4152 | } | |
4153 | ||
1b986589 | 4154 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
4155 | { |
4156 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 4157 | check_arg_pair_ok(fn) && |
1b986589 | 4158 | check_refcount_ok(fn, func_id) ? 0 : -EINVAL; |
435faee1 DB |
4159 | } |
4160 | ||
de8f3a83 DB |
4161 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
4162 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 4163 | */ |
f4d7e40a AS |
4164 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
4165 | struct bpf_func_state *state) | |
969bf05e | 4166 | { |
58e2af8b | 4167 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
4168 | int i; |
4169 | ||
4170 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 4171 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 4172 | mark_reg_unknown(env, regs, i); |
969bf05e | 4173 | |
f3709f69 JS |
4174 | bpf_for_each_spilled_reg(i, state, reg) { |
4175 | if (!reg) | |
969bf05e | 4176 | continue; |
de8f3a83 | 4177 | if (reg_is_pkt_pointer_any(reg)) |
f54c7898 | 4178 | __mark_reg_unknown(env, reg); |
969bf05e AS |
4179 | } |
4180 | } | |
4181 | ||
f4d7e40a AS |
4182 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
4183 | { | |
4184 | struct bpf_verifier_state *vstate = env->cur_state; | |
4185 | int i; | |
4186 | ||
4187 | for (i = 0; i <= vstate->curframe; i++) | |
4188 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
4189 | } | |
4190 | ||
fd978bf7 | 4191 | static void release_reg_references(struct bpf_verifier_env *env, |
1b986589 MKL |
4192 | struct bpf_func_state *state, |
4193 | int ref_obj_id) | |
fd978bf7 JS |
4194 | { |
4195 | struct bpf_reg_state *regs = state->regs, *reg; | |
4196 | int i; | |
4197 | ||
4198 | for (i = 0; i < MAX_BPF_REG; i++) | |
1b986589 | 4199 | if (regs[i].ref_obj_id == ref_obj_id) |
fd978bf7 JS |
4200 | mark_reg_unknown(env, regs, i); |
4201 | ||
4202 | bpf_for_each_spilled_reg(i, state, reg) { | |
4203 | if (!reg) | |
4204 | continue; | |
1b986589 | 4205 | if (reg->ref_obj_id == ref_obj_id) |
f54c7898 | 4206 | __mark_reg_unknown(env, reg); |
fd978bf7 JS |
4207 | } |
4208 | } | |
4209 | ||
4210 | /* The pointer with the specified id has released its reference to kernel | |
4211 | * resources. Identify all copies of the same pointer and clear the reference. | |
4212 | */ | |
4213 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 4214 | int ref_obj_id) |
fd978bf7 JS |
4215 | { |
4216 | struct bpf_verifier_state *vstate = env->cur_state; | |
1b986589 | 4217 | int err; |
fd978bf7 JS |
4218 | int i; |
4219 | ||
1b986589 MKL |
4220 | err = release_reference_state(cur_func(env), ref_obj_id); |
4221 | if (err) | |
4222 | return err; | |
4223 | ||
fd978bf7 | 4224 | for (i = 0; i <= vstate->curframe; i++) |
1b986589 | 4225 | release_reg_references(env, vstate->frame[i], ref_obj_id); |
fd978bf7 | 4226 | |
1b986589 | 4227 | return 0; |
fd978bf7 JS |
4228 | } |
4229 | ||
51c39bb1 AS |
4230 | static void clear_caller_saved_regs(struct bpf_verifier_env *env, |
4231 | struct bpf_reg_state *regs) | |
4232 | { | |
4233 | int i; | |
4234 | ||
4235 | /* after the call registers r0 - r5 were scratched */ | |
4236 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
4237 | mark_reg_not_init(env, regs, caller_saved[i]); | |
4238 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
4239 | } | |
4240 | } | |
4241 | ||
f4d7e40a AS |
4242 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
4243 | int *insn_idx) | |
4244 | { | |
4245 | struct bpf_verifier_state *state = env->cur_state; | |
51c39bb1 | 4246 | struct bpf_func_info_aux *func_info_aux; |
f4d7e40a | 4247 | struct bpf_func_state *caller, *callee; |
fd978bf7 | 4248 | int i, err, subprog, target_insn; |
51c39bb1 | 4249 | bool is_global = false; |
f4d7e40a | 4250 | |
aada9ce6 | 4251 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 4252 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 4253 | state->curframe + 2); |
f4d7e40a AS |
4254 | return -E2BIG; |
4255 | } | |
4256 | ||
4257 | target_insn = *insn_idx + insn->imm; | |
4258 | subprog = find_subprog(env, target_insn + 1); | |
4259 | if (subprog < 0) { | |
4260 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
4261 | target_insn + 1); | |
4262 | return -EFAULT; | |
4263 | } | |
4264 | ||
4265 | caller = state->frame[state->curframe]; | |
4266 | if (state->frame[state->curframe + 1]) { | |
4267 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
4268 | state->curframe + 1); | |
4269 | return -EFAULT; | |
4270 | } | |
4271 | ||
51c39bb1 AS |
4272 | func_info_aux = env->prog->aux->func_info_aux; |
4273 | if (func_info_aux) | |
4274 | is_global = func_info_aux[subprog].linkage == BTF_FUNC_GLOBAL; | |
4275 | err = btf_check_func_arg_match(env, subprog, caller->regs); | |
4276 | if (err == -EFAULT) | |
4277 | return err; | |
4278 | if (is_global) { | |
4279 | if (err) { | |
4280 | verbose(env, "Caller passes invalid args into func#%d\n", | |
4281 | subprog); | |
4282 | return err; | |
4283 | } else { | |
4284 | if (env->log.level & BPF_LOG_LEVEL) | |
4285 | verbose(env, | |
4286 | "Func#%d is global and valid. Skipping.\n", | |
4287 | subprog); | |
4288 | clear_caller_saved_regs(env, caller->regs); | |
4289 | ||
4290 | /* All global functions return SCALAR_VALUE */ | |
4291 | mark_reg_unknown(env, caller->regs, BPF_REG_0); | |
4292 | ||
4293 | /* continue with next insn after call */ | |
4294 | return 0; | |
4295 | } | |
4296 | } | |
4297 | ||
f4d7e40a AS |
4298 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); |
4299 | if (!callee) | |
4300 | return -ENOMEM; | |
4301 | state->frame[state->curframe + 1] = callee; | |
4302 | ||
4303 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
4304 | * into its own stack before reading from it. | |
4305 | * callee can read/write into caller's stack | |
4306 | */ | |
4307 | init_func_state(env, callee, | |
4308 | /* remember the callsite, it will be used by bpf_exit */ | |
4309 | *insn_idx /* callsite */, | |
4310 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 4311 | subprog /* subprog number within this prog */); |
f4d7e40a | 4312 | |
fd978bf7 JS |
4313 | /* Transfer references to the callee */ |
4314 | err = transfer_reference_state(callee, caller); | |
4315 | if (err) | |
4316 | return err; | |
4317 | ||
679c782d EC |
4318 | /* copy r1 - r5 args that callee can access. The copy includes parent |
4319 | * pointers, which connects us up to the liveness chain | |
4320 | */ | |
f4d7e40a AS |
4321 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) |
4322 | callee->regs[i] = caller->regs[i]; | |
4323 | ||
51c39bb1 | 4324 | clear_caller_saved_regs(env, caller->regs); |
f4d7e40a AS |
4325 | |
4326 | /* only increment it after check_reg_arg() finished */ | |
4327 | state->curframe++; | |
4328 | ||
4329 | /* and go analyze first insn of the callee */ | |
4330 | *insn_idx = target_insn; | |
4331 | ||
06ee7115 | 4332 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
4333 | verbose(env, "caller:\n"); |
4334 | print_verifier_state(env, caller); | |
4335 | verbose(env, "callee:\n"); | |
4336 | print_verifier_state(env, callee); | |
4337 | } | |
4338 | return 0; | |
4339 | } | |
4340 | ||
4341 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) | |
4342 | { | |
4343 | struct bpf_verifier_state *state = env->cur_state; | |
4344 | struct bpf_func_state *caller, *callee; | |
4345 | struct bpf_reg_state *r0; | |
fd978bf7 | 4346 | int err; |
f4d7e40a AS |
4347 | |
4348 | callee = state->frame[state->curframe]; | |
4349 | r0 = &callee->regs[BPF_REG_0]; | |
4350 | if (r0->type == PTR_TO_STACK) { | |
4351 | /* technically it's ok to return caller's stack pointer | |
4352 | * (or caller's caller's pointer) back to the caller, | |
4353 | * since these pointers are valid. Only current stack | |
4354 | * pointer will be invalid as soon as function exits, | |
4355 | * but let's be conservative | |
4356 | */ | |
4357 | verbose(env, "cannot return stack pointer to the caller\n"); | |
4358 | return -EINVAL; | |
4359 | } | |
4360 | ||
4361 | state->curframe--; | |
4362 | caller = state->frame[state->curframe]; | |
4363 | /* return to the caller whatever r0 had in the callee */ | |
4364 | caller->regs[BPF_REG_0] = *r0; | |
4365 | ||
fd978bf7 JS |
4366 | /* Transfer references to the caller */ |
4367 | err = transfer_reference_state(caller, callee); | |
4368 | if (err) | |
4369 | return err; | |
4370 | ||
f4d7e40a | 4371 | *insn_idx = callee->callsite + 1; |
06ee7115 | 4372 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
4373 | verbose(env, "returning from callee:\n"); |
4374 | print_verifier_state(env, callee); | |
4375 | verbose(env, "to caller at %d:\n", *insn_idx); | |
4376 | print_verifier_state(env, caller); | |
4377 | } | |
4378 | /* clear everything in the callee */ | |
4379 | free_func_state(callee); | |
4380 | state->frame[state->curframe + 1] = NULL; | |
4381 | return 0; | |
4382 | } | |
4383 | ||
849fa506 YS |
4384 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
4385 | int func_id, | |
4386 | struct bpf_call_arg_meta *meta) | |
4387 | { | |
4388 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
4389 | ||
4390 | if (ret_type != RET_INTEGER || | |
4391 | (func_id != BPF_FUNC_get_stack && | |
47cc0ed5 DB |
4392 | func_id != BPF_FUNC_probe_read_str && |
4393 | func_id != BPF_FUNC_probe_read_kernel_str && | |
4394 | func_id != BPF_FUNC_probe_read_user_str)) | |
849fa506 YS |
4395 | return; |
4396 | ||
10060503 | 4397 | ret_reg->smax_value = meta->msize_max_value; |
fa123ac0 | 4398 | ret_reg->s32_max_value = meta->msize_max_value; |
849fa506 YS |
4399 | __reg_deduce_bounds(ret_reg); |
4400 | __reg_bound_offset(ret_reg); | |
10060503 | 4401 | __update_reg_bounds(ret_reg); |
849fa506 YS |
4402 | } |
4403 | ||
c93552c4 DB |
4404 | static int |
4405 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
4406 | int func_id, int insn_idx) | |
4407 | { | |
4408 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 4409 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
4410 | |
4411 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
4412 | func_id != BPF_FUNC_map_lookup_elem && |
4413 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
4414 | func_id != BPF_FUNC_map_delete_elem && |
4415 | func_id != BPF_FUNC_map_push_elem && | |
4416 | func_id != BPF_FUNC_map_pop_elem && | |
4417 | func_id != BPF_FUNC_map_peek_elem) | |
c93552c4 | 4418 | return 0; |
09772d92 | 4419 | |
591fe988 | 4420 | if (map == NULL) { |
c93552c4 DB |
4421 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
4422 | return -EINVAL; | |
4423 | } | |
4424 | ||
591fe988 DB |
4425 | /* In case of read-only, some additional restrictions |
4426 | * need to be applied in order to prevent altering the | |
4427 | * state of the map from program side. | |
4428 | */ | |
4429 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
4430 | (func_id == BPF_FUNC_map_delete_elem || | |
4431 | func_id == BPF_FUNC_map_update_elem || | |
4432 | func_id == BPF_FUNC_map_push_elem || | |
4433 | func_id == BPF_FUNC_map_pop_elem)) { | |
4434 | verbose(env, "write into map forbidden\n"); | |
4435 | return -EACCES; | |
4436 | } | |
4437 | ||
d2e4c1e6 | 4438 | if (!BPF_MAP_PTR(aux->map_ptr_state)) |
c93552c4 | 4439 | bpf_map_ptr_store(aux, meta->map_ptr, |
2c78ee89 | 4440 | !meta->map_ptr->bypass_spec_v1); |
d2e4c1e6 | 4441 | else if (BPF_MAP_PTR(aux->map_ptr_state) != meta->map_ptr) |
c93552c4 | 4442 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, |
2c78ee89 | 4443 | !meta->map_ptr->bypass_spec_v1); |
c93552c4 DB |
4444 | return 0; |
4445 | } | |
4446 | ||
d2e4c1e6 DB |
4447 | static int |
4448 | record_func_key(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
4449 | int func_id, int insn_idx) | |
4450 | { | |
4451 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
4452 | struct bpf_reg_state *regs = cur_regs(env), *reg; | |
4453 | struct bpf_map *map = meta->map_ptr; | |
4454 | struct tnum range; | |
4455 | u64 val; | |
cc52d914 | 4456 | int err; |
d2e4c1e6 DB |
4457 | |
4458 | if (func_id != BPF_FUNC_tail_call) | |
4459 | return 0; | |
4460 | if (!map || map->map_type != BPF_MAP_TYPE_PROG_ARRAY) { | |
4461 | verbose(env, "kernel subsystem misconfigured verifier\n"); | |
4462 | return -EINVAL; | |
4463 | } | |
4464 | ||
4465 | range = tnum_range(0, map->max_entries - 1); | |
4466 | reg = ®s[BPF_REG_3]; | |
4467 | ||
4468 | if (!register_is_const(reg) || !tnum_in(range, reg->var_off)) { | |
4469 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
4470 | return 0; | |
4471 | } | |
4472 | ||
cc52d914 DB |
4473 | err = mark_chain_precision(env, BPF_REG_3); |
4474 | if (err) | |
4475 | return err; | |
4476 | ||
d2e4c1e6 DB |
4477 | val = reg->var_off.value; |
4478 | if (bpf_map_key_unseen(aux)) | |
4479 | bpf_map_key_store(aux, val); | |
4480 | else if (!bpf_map_key_poisoned(aux) && | |
4481 | bpf_map_key_immediate(aux) != val) | |
4482 | bpf_map_key_store(aux, BPF_MAP_KEY_POISON); | |
4483 | return 0; | |
4484 | } | |
4485 | ||
fd978bf7 JS |
4486 | static int check_reference_leak(struct bpf_verifier_env *env) |
4487 | { | |
4488 | struct bpf_func_state *state = cur_func(env); | |
4489 | int i; | |
4490 | ||
4491 | for (i = 0; i < state->acquired_refs; i++) { | |
4492 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", | |
4493 | state->refs[i].id, state->refs[i].insn_idx); | |
4494 | } | |
4495 | return state->acquired_refs ? -EINVAL : 0; | |
4496 | } | |
4497 | ||
f4d7e40a | 4498 | static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 4499 | { |
17a52670 | 4500 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 4501 | struct bpf_reg_state *regs; |
33ff9823 | 4502 | struct bpf_call_arg_meta meta; |
969bf05e | 4503 | bool changes_data; |
17a52670 AS |
4504 | int i, err; |
4505 | ||
4506 | /* find function prototype */ | |
4507 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
4508 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
4509 | func_id); | |
17a52670 AS |
4510 | return -EINVAL; |
4511 | } | |
4512 | ||
00176a34 | 4513 | if (env->ops->get_func_proto) |
5e43f899 | 4514 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 4515 | if (!fn) { |
61bd5218 JK |
4516 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
4517 | func_id); | |
17a52670 AS |
4518 | return -EINVAL; |
4519 | } | |
4520 | ||
4521 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 4522 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 4523 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
4524 | return -EINVAL; |
4525 | } | |
4526 | ||
04514d13 | 4527 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 4528 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
4529 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
4530 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
4531 | func_id_name(func_id), func_id); | |
4532 | return -EINVAL; | |
4533 | } | |
969bf05e | 4534 | |
33ff9823 | 4535 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 4536 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 4537 | |
1b986589 | 4538 | err = check_func_proto(fn, func_id); |
435faee1 | 4539 | if (err) { |
61bd5218 | 4540 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 4541 | func_id_name(func_id), func_id); |
435faee1 DB |
4542 | return err; |
4543 | } | |
4544 | ||
d83525ca | 4545 | meta.func_id = func_id; |
17a52670 | 4546 | /* check args */ |
a7658e1a | 4547 | for (i = 0; i < 5; i++) { |
9cc31b3a AS |
4548 | err = btf_resolve_helper_id(&env->log, fn, i); |
4549 | if (err > 0) | |
4550 | meta.btf_id = err; | |
a7658e1a AS |
4551 | err = check_func_arg(env, BPF_REG_1 + i, fn->arg_type[i], &meta); |
4552 | if (err) | |
4553 | return err; | |
4554 | } | |
17a52670 | 4555 | |
c93552c4 DB |
4556 | err = record_func_map(env, &meta, func_id, insn_idx); |
4557 | if (err) | |
4558 | return err; | |
4559 | ||
d2e4c1e6 DB |
4560 | err = record_func_key(env, &meta, func_id, insn_idx); |
4561 | if (err) | |
4562 | return err; | |
4563 | ||
435faee1 DB |
4564 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
4565 | * is inferred from register state. | |
4566 | */ | |
4567 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
4568 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
4569 | BPF_WRITE, -1, false); | |
435faee1 DB |
4570 | if (err) |
4571 | return err; | |
4572 | } | |
4573 | ||
fd978bf7 JS |
4574 | if (func_id == BPF_FUNC_tail_call) { |
4575 | err = check_reference_leak(env); | |
4576 | if (err) { | |
4577 | verbose(env, "tail_call would lead to reference leak\n"); | |
4578 | return err; | |
4579 | } | |
4580 | } else if (is_release_function(func_id)) { | |
1b986589 | 4581 | err = release_reference(env, meta.ref_obj_id); |
46f8bc92 MKL |
4582 | if (err) { |
4583 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
4584 | func_id_name(func_id), func_id); | |
fd978bf7 | 4585 | return err; |
46f8bc92 | 4586 | } |
fd978bf7 JS |
4587 | } |
4588 | ||
638f5b90 | 4589 | regs = cur_regs(env); |
cd339431 RG |
4590 | |
4591 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
4592 | * this is required because get_local_storage() can't return an error. | |
4593 | */ | |
4594 | if (func_id == BPF_FUNC_get_local_storage && | |
4595 | !register_is_null(®s[BPF_REG_2])) { | |
4596 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
4597 | return -EINVAL; | |
4598 | } | |
4599 | ||
17a52670 | 4600 | /* reset caller saved regs */ |
dc503a8a | 4601 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 4602 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
4603 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
4604 | } | |
17a52670 | 4605 | |
5327ed3d JW |
4606 | /* helper call returns 64-bit value. */ |
4607 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
4608 | ||
dc503a8a | 4609 | /* update return register (already marked as written above) */ |
17a52670 | 4610 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 4611 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 4612 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
4613 | } else if (fn->ret_type == RET_VOID) { |
4614 | regs[BPF_REG_0].type = NOT_INIT; | |
3e6a4b3e RG |
4615 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL || |
4616 | fn->ret_type == RET_PTR_TO_MAP_VALUE) { | |
f1174f77 | 4617 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 4618 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
4619 | /* remember map_ptr, so that check_map_access() |
4620 | * can check 'value_size' boundary of memory access | |
4621 | * to map element returned from bpf_map_lookup_elem() | |
4622 | */ | |
33ff9823 | 4623 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
4624 | verbose(env, |
4625 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
4626 | return -EINVAL; |
4627 | } | |
33ff9823 | 4628 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
4d31f301 DB |
4629 | if (fn->ret_type == RET_PTR_TO_MAP_VALUE) { |
4630 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE; | |
e16d2f1a AS |
4631 | if (map_value_has_spin_lock(meta.map_ptr)) |
4632 | regs[BPF_REG_0].id = ++env->id_gen; | |
4d31f301 DB |
4633 | } else { |
4634 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
4635 | regs[BPF_REG_0].id = ++env->id_gen; | |
4636 | } | |
c64b7983 JS |
4637 | } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) { |
4638 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
4639 | regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL; | |
0f3adc28 | 4640 | regs[BPF_REG_0].id = ++env->id_gen; |
85a51f8c LB |
4641 | } else if (fn->ret_type == RET_PTR_TO_SOCK_COMMON_OR_NULL) { |
4642 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
4643 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON_OR_NULL; | |
4644 | regs[BPF_REG_0].id = ++env->id_gen; | |
655a51e5 MKL |
4645 | } else if (fn->ret_type == RET_PTR_TO_TCP_SOCK_OR_NULL) { |
4646 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
4647 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK_OR_NULL; | |
4648 | regs[BPF_REG_0].id = ++env->id_gen; | |
17a52670 | 4649 | } else { |
61bd5218 | 4650 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 4651 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
4652 | return -EINVAL; |
4653 | } | |
04fd61ab | 4654 | |
0f3adc28 | 4655 | if (is_ptr_cast_function(func_id)) { |
1b986589 MKL |
4656 | /* For release_reference() */ |
4657 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
64d85290 | 4658 | } else if (is_acquire_function(func_id, meta.map_ptr)) { |
0f3adc28 LB |
4659 | int id = acquire_reference_state(env, insn_idx); |
4660 | ||
4661 | if (id < 0) | |
4662 | return id; | |
4663 | /* For mark_ptr_or_null_reg() */ | |
4664 | regs[BPF_REG_0].id = id; | |
4665 | /* For release_reference() */ | |
4666 | regs[BPF_REG_0].ref_obj_id = id; | |
4667 | } | |
1b986589 | 4668 | |
849fa506 YS |
4669 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
4670 | ||
61bd5218 | 4671 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
4672 | if (err) |
4673 | return err; | |
04fd61ab | 4674 | |
c195651e YS |
4675 | if (func_id == BPF_FUNC_get_stack && !env->prog->has_callchain_buf) { |
4676 | const char *err_str; | |
4677 | ||
4678 | #ifdef CONFIG_PERF_EVENTS | |
4679 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
4680 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
4681 | #else | |
4682 | err = -ENOTSUPP; | |
4683 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
4684 | #endif | |
4685 | if (err) { | |
4686 | verbose(env, err_str, func_id_name(func_id), func_id); | |
4687 | return err; | |
4688 | } | |
4689 | ||
4690 | env->prog->has_callchain_buf = true; | |
4691 | } | |
4692 | ||
969bf05e AS |
4693 | if (changes_data) |
4694 | clear_all_pkt_pointers(env); | |
4695 | return 0; | |
4696 | } | |
4697 | ||
b03c9f9f EC |
4698 | static bool signed_add_overflows(s64 a, s64 b) |
4699 | { | |
4700 | /* Do the add in u64, where overflow is well-defined */ | |
4701 | s64 res = (s64)((u64)a + (u64)b); | |
4702 | ||
4703 | if (b < 0) | |
4704 | return res > a; | |
4705 | return res < a; | |
4706 | } | |
4707 | ||
3f50f132 JF |
4708 | static bool signed_add32_overflows(s64 a, s64 b) |
4709 | { | |
4710 | /* Do the add in u32, where overflow is well-defined */ | |
4711 | s32 res = (s32)((u32)a + (u32)b); | |
4712 | ||
4713 | if (b < 0) | |
4714 | return res > a; | |
4715 | return res < a; | |
4716 | } | |
4717 | ||
4718 | static bool signed_sub_overflows(s32 a, s32 b) | |
b03c9f9f EC |
4719 | { |
4720 | /* Do the sub in u64, where overflow is well-defined */ | |
4721 | s64 res = (s64)((u64)a - (u64)b); | |
4722 | ||
4723 | if (b < 0) | |
4724 | return res < a; | |
4725 | return res > a; | |
969bf05e AS |
4726 | } |
4727 | ||
3f50f132 JF |
4728 | static bool signed_sub32_overflows(s32 a, s32 b) |
4729 | { | |
4730 | /* Do the sub in u64, where overflow is well-defined */ | |
4731 | s32 res = (s32)((u32)a - (u32)b); | |
4732 | ||
4733 | if (b < 0) | |
4734 | return res < a; | |
4735 | return res > a; | |
4736 | } | |
4737 | ||
bb7f0f98 AS |
4738 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
4739 | const struct bpf_reg_state *reg, | |
4740 | enum bpf_reg_type type) | |
4741 | { | |
4742 | bool known = tnum_is_const(reg->var_off); | |
4743 | s64 val = reg->var_off.value; | |
4744 | s64 smin = reg->smin_value; | |
4745 | ||
4746 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
4747 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
4748 | reg_type_str[type], val); | |
4749 | return false; | |
4750 | } | |
4751 | ||
4752 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
4753 | verbose(env, "%s pointer offset %d is not allowed\n", | |
4754 | reg_type_str[type], reg->off); | |
4755 | return false; | |
4756 | } | |
4757 | ||
4758 | if (smin == S64_MIN) { | |
4759 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
4760 | reg_type_str[type]); | |
4761 | return false; | |
4762 | } | |
4763 | ||
4764 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
4765 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
4766 | smin, reg_type_str[type]); | |
4767 | return false; | |
4768 | } | |
4769 | ||
4770 | return true; | |
4771 | } | |
4772 | ||
979d63d5 DB |
4773 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
4774 | { | |
4775 | return &env->insn_aux_data[env->insn_idx]; | |
4776 | } | |
4777 | ||
4778 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, | |
4779 | u32 *ptr_limit, u8 opcode, bool off_is_neg) | |
4780 | { | |
4781 | bool mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
4782 | (opcode == BPF_SUB && !off_is_neg); | |
4783 | u32 off; | |
4784 | ||
4785 | switch (ptr_reg->type) { | |
4786 | case PTR_TO_STACK: | |
088ec26d AI |
4787 | /* Indirect variable offset stack access is prohibited in |
4788 | * unprivileged mode so it's not handled here. | |
4789 | */ | |
979d63d5 DB |
4790 | off = ptr_reg->off + ptr_reg->var_off.value; |
4791 | if (mask_to_left) | |
4792 | *ptr_limit = MAX_BPF_STACK + off; | |
4793 | else | |
4794 | *ptr_limit = -off; | |
4795 | return 0; | |
4796 | case PTR_TO_MAP_VALUE: | |
4797 | if (mask_to_left) { | |
4798 | *ptr_limit = ptr_reg->umax_value + ptr_reg->off; | |
4799 | } else { | |
4800 | off = ptr_reg->smin_value + ptr_reg->off; | |
4801 | *ptr_limit = ptr_reg->map_ptr->value_size - off; | |
4802 | } | |
4803 | return 0; | |
4804 | default: | |
4805 | return -EINVAL; | |
4806 | } | |
4807 | } | |
4808 | ||
d3bd7413 DB |
4809 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
4810 | const struct bpf_insn *insn) | |
4811 | { | |
2c78ee89 | 4812 | return env->bypass_spec_v1 || BPF_SRC(insn->code) == BPF_K; |
d3bd7413 DB |
4813 | } |
4814 | ||
4815 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
4816 | u32 alu_state, u32 alu_limit) | |
4817 | { | |
4818 | /* If we arrived here from different branches with different | |
4819 | * state or limits to sanitize, then this won't work. | |
4820 | */ | |
4821 | if (aux->alu_state && | |
4822 | (aux->alu_state != alu_state || | |
4823 | aux->alu_limit != alu_limit)) | |
4824 | return -EACCES; | |
4825 | ||
4826 | /* Corresponding fixup done in fixup_bpf_calls(). */ | |
4827 | aux->alu_state = alu_state; | |
4828 | aux->alu_limit = alu_limit; | |
4829 | return 0; | |
4830 | } | |
4831 | ||
4832 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
4833 | struct bpf_insn *insn) | |
4834 | { | |
4835 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
4836 | ||
4837 | if (can_skip_alu_sanitation(env, insn)) | |
4838 | return 0; | |
4839 | ||
4840 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
4841 | } | |
4842 | ||
979d63d5 DB |
4843 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
4844 | struct bpf_insn *insn, | |
4845 | const struct bpf_reg_state *ptr_reg, | |
4846 | struct bpf_reg_state *dst_reg, | |
4847 | bool off_is_neg) | |
4848 | { | |
4849 | struct bpf_verifier_state *vstate = env->cur_state; | |
4850 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
4851 | bool ptr_is_dst_reg = ptr_reg == dst_reg; | |
4852 | u8 opcode = BPF_OP(insn->code); | |
4853 | u32 alu_state, alu_limit; | |
4854 | struct bpf_reg_state tmp; | |
4855 | bool ret; | |
4856 | ||
d3bd7413 | 4857 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
4858 | return 0; |
4859 | ||
4860 | /* We already marked aux for masking from non-speculative | |
4861 | * paths, thus we got here in the first place. We only care | |
4862 | * to explore bad access from here. | |
4863 | */ | |
4864 | if (vstate->speculative) | |
4865 | goto do_sim; | |
4866 | ||
4867 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
4868 | alu_state |= ptr_is_dst_reg ? | |
4869 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
4870 | ||
4871 | if (retrieve_ptr_limit(ptr_reg, &alu_limit, opcode, off_is_neg)) | |
4872 | return 0; | |
d3bd7413 | 4873 | if (update_alu_sanitation_state(aux, alu_state, alu_limit)) |
979d63d5 | 4874 | return -EACCES; |
979d63d5 DB |
4875 | do_sim: |
4876 | /* Simulate and find potential out-of-bounds access under | |
4877 | * speculative execution from truncation as a result of | |
4878 | * masking when off was not within expected range. If off | |
4879 | * sits in dst, then we temporarily need to move ptr there | |
4880 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
4881 | * for cases where we use K-based arithmetic in one direction | |
4882 | * and truncated reg-based in the other in order to explore | |
4883 | * bad access. | |
4884 | */ | |
4885 | if (!ptr_is_dst_reg) { | |
4886 | tmp = *dst_reg; | |
4887 | *dst_reg = *ptr_reg; | |
4888 | } | |
4889 | ret = push_stack(env, env->insn_idx + 1, env->insn_idx, true); | |
0803278b | 4890 | if (!ptr_is_dst_reg && ret) |
979d63d5 DB |
4891 | *dst_reg = tmp; |
4892 | return !ret ? -EFAULT : 0; | |
4893 | } | |
4894 | ||
f1174f77 | 4895 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
4896 | * Caller should also handle BPF_MOV case separately. |
4897 | * If we return -EACCES, caller may want to try again treating pointer as a | |
4898 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
4899 | */ | |
4900 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
4901 | struct bpf_insn *insn, | |
4902 | const struct bpf_reg_state *ptr_reg, | |
4903 | const struct bpf_reg_state *off_reg) | |
969bf05e | 4904 | { |
f4d7e40a AS |
4905 | struct bpf_verifier_state *vstate = env->cur_state; |
4906 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
4907 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 4908 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
4909 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
4910 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
4911 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
4912 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
9d7eceed | 4913 | u32 dst = insn->dst_reg, src = insn->src_reg; |
969bf05e | 4914 | u8 opcode = BPF_OP(insn->code); |
979d63d5 | 4915 | int ret; |
969bf05e | 4916 | |
f1174f77 | 4917 | dst_reg = ®s[dst]; |
969bf05e | 4918 | |
6f16101e DB |
4919 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
4920 | smin_val > smax_val || umin_val > umax_val) { | |
4921 | /* Taint dst register if offset had invalid bounds derived from | |
4922 | * e.g. dead branches. | |
4923 | */ | |
f54c7898 | 4924 | __mark_reg_unknown(env, dst_reg); |
6f16101e | 4925 | return 0; |
f1174f77 EC |
4926 | } |
4927 | ||
4928 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
4929 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
4930 | verbose(env, |
4931 | "R%d 32-bit pointer arithmetic prohibited\n", | |
4932 | dst); | |
f1174f77 | 4933 | return -EACCES; |
969bf05e AS |
4934 | } |
4935 | ||
aad2eeaf JS |
4936 | switch (ptr_reg->type) { |
4937 | case PTR_TO_MAP_VALUE_OR_NULL: | |
4938 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", | |
4939 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 4940 | return -EACCES; |
aad2eeaf JS |
4941 | case CONST_PTR_TO_MAP: |
4942 | case PTR_TO_PACKET_END: | |
c64b7983 JS |
4943 | case PTR_TO_SOCKET: |
4944 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
4945 | case PTR_TO_SOCK_COMMON: |
4946 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
4947 | case PTR_TO_TCP_SOCK: |
4948 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 4949 | case PTR_TO_XDP_SOCK: |
aad2eeaf JS |
4950 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
4951 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 4952 | return -EACCES; |
9d7eceed DB |
4953 | case PTR_TO_MAP_VALUE: |
4954 | if (!env->allow_ptr_leaks && !known && (smin_val < 0) != (smax_val < 0)) { | |
4955 | verbose(env, "R%d has unknown scalar with mixed signed bounds, pointer arithmetic with it prohibited for !root\n", | |
4956 | off_reg == dst_reg ? dst : src); | |
4957 | return -EACCES; | |
4958 | } | |
4959 | /* fall-through */ | |
aad2eeaf JS |
4960 | default: |
4961 | break; | |
f1174f77 EC |
4962 | } |
4963 | ||
4964 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
4965 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 4966 | */ |
f1174f77 EC |
4967 | dst_reg->type = ptr_reg->type; |
4968 | dst_reg->id = ptr_reg->id; | |
969bf05e | 4969 | |
bb7f0f98 AS |
4970 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
4971 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
4972 | return -EINVAL; | |
4973 | ||
3f50f132 JF |
4974 | /* pointer types do not carry 32-bit bounds at the moment. */ |
4975 | __mark_reg32_unbounded(dst_reg); | |
4976 | ||
f1174f77 EC |
4977 | switch (opcode) { |
4978 | case BPF_ADD: | |
979d63d5 DB |
4979 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
4980 | if (ret < 0) { | |
4981 | verbose(env, "R%d tried to add from different maps or paths\n", dst); | |
4982 | return ret; | |
4983 | } | |
f1174f77 EC |
4984 | /* We can take a fixed offset as long as it doesn't overflow |
4985 | * the s32 'off' field | |
969bf05e | 4986 | */ |
b03c9f9f EC |
4987 | if (known && (ptr_reg->off + smin_val == |
4988 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 4989 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
4990 | dst_reg->smin_value = smin_ptr; |
4991 | dst_reg->smax_value = smax_ptr; | |
4992 | dst_reg->umin_value = umin_ptr; | |
4993 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 4994 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 4995 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 4996 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
4997 | break; |
4998 | } | |
f1174f77 EC |
4999 | /* A new variable offset is created. Note that off_reg->off |
5000 | * == 0, since it's a scalar. | |
5001 | * dst_reg gets the pointer type and since some positive | |
5002 | * integer value was added to the pointer, give it a new 'id' | |
5003 | * if it's a PTR_TO_PACKET. | |
5004 | * this creates a new 'base' pointer, off_reg (variable) gets | |
5005 | * added into the variable offset, and we copy the fixed offset | |
5006 | * from ptr_reg. | |
969bf05e | 5007 | */ |
b03c9f9f EC |
5008 | if (signed_add_overflows(smin_ptr, smin_val) || |
5009 | signed_add_overflows(smax_ptr, smax_val)) { | |
5010 | dst_reg->smin_value = S64_MIN; | |
5011 | dst_reg->smax_value = S64_MAX; | |
5012 | } else { | |
5013 | dst_reg->smin_value = smin_ptr + smin_val; | |
5014 | dst_reg->smax_value = smax_ptr + smax_val; | |
5015 | } | |
5016 | if (umin_ptr + umin_val < umin_ptr || | |
5017 | umax_ptr + umax_val < umax_ptr) { | |
5018 | dst_reg->umin_value = 0; | |
5019 | dst_reg->umax_value = U64_MAX; | |
5020 | } else { | |
5021 | dst_reg->umin_value = umin_ptr + umin_val; | |
5022 | dst_reg->umax_value = umax_ptr + umax_val; | |
5023 | } | |
f1174f77 EC |
5024 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
5025 | dst_reg->off = ptr_reg->off; | |
0962590e | 5026 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 5027 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
5028 | dst_reg->id = ++env->id_gen; |
5029 | /* something was added to pkt_ptr, set range to zero */ | |
0962590e | 5030 | dst_reg->raw = 0; |
f1174f77 EC |
5031 | } |
5032 | break; | |
5033 | case BPF_SUB: | |
979d63d5 DB |
5034 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
5035 | if (ret < 0) { | |
5036 | verbose(env, "R%d tried to sub from different maps or paths\n", dst); | |
5037 | return ret; | |
5038 | } | |
f1174f77 EC |
5039 | if (dst_reg == off_reg) { |
5040 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
5041 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
5042 | dst); | |
f1174f77 EC |
5043 | return -EACCES; |
5044 | } | |
5045 | /* We don't allow subtraction from FP, because (according to | |
5046 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
5047 | * be able to deal with it. | |
969bf05e | 5048 | */ |
f1174f77 | 5049 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
5050 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
5051 | dst); | |
f1174f77 EC |
5052 | return -EACCES; |
5053 | } | |
b03c9f9f EC |
5054 | if (known && (ptr_reg->off - smin_val == |
5055 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 5056 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
5057 | dst_reg->smin_value = smin_ptr; |
5058 | dst_reg->smax_value = smax_ptr; | |
5059 | dst_reg->umin_value = umin_ptr; | |
5060 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
5061 | dst_reg->var_off = ptr_reg->var_off; |
5062 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 5063 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 5064 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
5065 | break; |
5066 | } | |
f1174f77 EC |
5067 | /* A new variable offset is created. If the subtrahend is known |
5068 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 5069 | */ |
b03c9f9f EC |
5070 | if (signed_sub_overflows(smin_ptr, smax_val) || |
5071 | signed_sub_overflows(smax_ptr, smin_val)) { | |
5072 | /* Overflow possible, we know nothing */ | |
5073 | dst_reg->smin_value = S64_MIN; | |
5074 | dst_reg->smax_value = S64_MAX; | |
5075 | } else { | |
5076 | dst_reg->smin_value = smin_ptr - smax_val; | |
5077 | dst_reg->smax_value = smax_ptr - smin_val; | |
5078 | } | |
5079 | if (umin_ptr < umax_val) { | |
5080 | /* Overflow possible, we know nothing */ | |
5081 | dst_reg->umin_value = 0; | |
5082 | dst_reg->umax_value = U64_MAX; | |
5083 | } else { | |
5084 | /* Cannot overflow (as long as bounds are consistent) */ | |
5085 | dst_reg->umin_value = umin_ptr - umax_val; | |
5086 | dst_reg->umax_value = umax_ptr - umin_val; | |
5087 | } | |
f1174f77 EC |
5088 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
5089 | dst_reg->off = ptr_reg->off; | |
0962590e | 5090 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 5091 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
5092 | dst_reg->id = ++env->id_gen; |
5093 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 5094 | if (smin_val < 0) |
0962590e | 5095 | dst_reg->raw = 0; |
43188702 | 5096 | } |
f1174f77 EC |
5097 | break; |
5098 | case BPF_AND: | |
5099 | case BPF_OR: | |
5100 | case BPF_XOR: | |
82abbf8d AS |
5101 | /* bitwise ops on pointers are troublesome, prohibit. */ |
5102 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
5103 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
5104 | return -EACCES; |
5105 | default: | |
5106 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
5107 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
5108 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 5109 | return -EACCES; |
43188702 JF |
5110 | } |
5111 | ||
bb7f0f98 AS |
5112 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
5113 | return -EINVAL; | |
5114 | ||
b03c9f9f EC |
5115 | __update_reg_bounds(dst_reg); |
5116 | __reg_deduce_bounds(dst_reg); | |
5117 | __reg_bound_offset(dst_reg); | |
0d6303db DB |
5118 | |
5119 | /* For unprivileged we require that resulting offset must be in bounds | |
5120 | * in order to be able to sanitize access later on. | |
5121 | */ | |
2c78ee89 | 5122 | if (!env->bypass_spec_v1) { |
e4298d25 DB |
5123 | if (dst_reg->type == PTR_TO_MAP_VALUE && |
5124 | check_map_access(env, dst, dst_reg->off, 1, false)) { | |
5125 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
5126 | "prohibited for !root\n", dst); | |
5127 | return -EACCES; | |
5128 | } else if (dst_reg->type == PTR_TO_STACK && | |
5129 | check_stack_access(env, dst_reg, dst_reg->off + | |
5130 | dst_reg->var_off.value, 1)) { | |
5131 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
5132 | "prohibited for !root\n", dst); | |
5133 | return -EACCES; | |
5134 | } | |
0d6303db DB |
5135 | } |
5136 | ||
43188702 JF |
5137 | return 0; |
5138 | } | |
5139 | ||
3f50f132 JF |
5140 | static void scalar32_min_max_add(struct bpf_reg_state *dst_reg, |
5141 | struct bpf_reg_state *src_reg) | |
5142 | { | |
5143 | s32 smin_val = src_reg->s32_min_value; | |
5144 | s32 smax_val = src_reg->s32_max_value; | |
5145 | u32 umin_val = src_reg->u32_min_value; | |
5146 | u32 umax_val = src_reg->u32_max_value; | |
5147 | ||
5148 | if (signed_add32_overflows(dst_reg->s32_min_value, smin_val) || | |
5149 | signed_add32_overflows(dst_reg->s32_max_value, smax_val)) { | |
5150 | dst_reg->s32_min_value = S32_MIN; | |
5151 | dst_reg->s32_max_value = S32_MAX; | |
5152 | } else { | |
5153 | dst_reg->s32_min_value += smin_val; | |
5154 | dst_reg->s32_max_value += smax_val; | |
5155 | } | |
5156 | if (dst_reg->u32_min_value + umin_val < umin_val || | |
5157 | dst_reg->u32_max_value + umax_val < umax_val) { | |
5158 | dst_reg->u32_min_value = 0; | |
5159 | dst_reg->u32_max_value = U32_MAX; | |
5160 | } else { | |
5161 | dst_reg->u32_min_value += umin_val; | |
5162 | dst_reg->u32_max_value += umax_val; | |
5163 | } | |
5164 | } | |
5165 | ||
07cd2631 JF |
5166 | static void scalar_min_max_add(struct bpf_reg_state *dst_reg, |
5167 | struct bpf_reg_state *src_reg) | |
5168 | { | |
5169 | s64 smin_val = src_reg->smin_value; | |
5170 | s64 smax_val = src_reg->smax_value; | |
5171 | u64 umin_val = src_reg->umin_value; | |
5172 | u64 umax_val = src_reg->umax_value; | |
5173 | ||
5174 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || | |
5175 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
5176 | dst_reg->smin_value = S64_MIN; | |
5177 | dst_reg->smax_value = S64_MAX; | |
5178 | } else { | |
5179 | dst_reg->smin_value += smin_val; | |
5180 | dst_reg->smax_value += smax_val; | |
5181 | } | |
5182 | if (dst_reg->umin_value + umin_val < umin_val || | |
5183 | dst_reg->umax_value + umax_val < umax_val) { | |
5184 | dst_reg->umin_value = 0; | |
5185 | dst_reg->umax_value = U64_MAX; | |
5186 | } else { | |
5187 | dst_reg->umin_value += umin_val; | |
5188 | dst_reg->umax_value += umax_val; | |
5189 | } | |
3f50f132 JF |
5190 | } |
5191 | ||
5192 | static void scalar32_min_max_sub(struct bpf_reg_state *dst_reg, | |
5193 | struct bpf_reg_state *src_reg) | |
5194 | { | |
5195 | s32 smin_val = src_reg->s32_min_value; | |
5196 | s32 smax_val = src_reg->s32_max_value; | |
5197 | u32 umin_val = src_reg->u32_min_value; | |
5198 | u32 umax_val = src_reg->u32_max_value; | |
5199 | ||
5200 | if (signed_sub32_overflows(dst_reg->s32_min_value, smax_val) || | |
5201 | signed_sub32_overflows(dst_reg->s32_max_value, smin_val)) { | |
5202 | /* Overflow possible, we know nothing */ | |
5203 | dst_reg->s32_min_value = S32_MIN; | |
5204 | dst_reg->s32_max_value = S32_MAX; | |
5205 | } else { | |
5206 | dst_reg->s32_min_value -= smax_val; | |
5207 | dst_reg->s32_max_value -= smin_val; | |
5208 | } | |
5209 | if (dst_reg->u32_min_value < umax_val) { | |
5210 | /* Overflow possible, we know nothing */ | |
5211 | dst_reg->u32_min_value = 0; | |
5212 | dst_reg->u32_max_value = U32_MAX; | |
5213 | } else { | |
5214 | /* Cannot overflow (as long as bounds are consistent) */ | |
5215 | dst_reg->u32_min_value -= umax_val; | |
5216 | dst_reg->u32_max_value -= umin_val; | |
5217 | } | |
07cd2631 JF |
5218 | } |
5219 | ||
5220 | static void scalar_min_max_sub(struct bpf_reg_state *dst_reg, | |
5221 | struct bpf_reg_state *src_reg) | |
5222 | { | |
5223 | s64 smin_val = src_reg->smin_value; | |
5224 | s64 smax_val = src_reg->smax_value; | |
5225 | u64 umin_val = src_reg->umin_value; | |
5226 | u64 umax_val = src_reg->umax_value; | |
5227 | ||
5228 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || | |
5229 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
5230 | /* Overflow possible, we know nothing */ | |
5231 | dst_reg->smin_value = S64_MIN; | |
5232 | dst_reg->smax_value = S64_MAX; | |
5233 | } else { | |
5234 | dst_reg->smin_value -= smax_val; | |
5235 | dst_reg->smax_value -= smin_val; | |
5236 | } | |
5237 | if (dst_reg->umin_value < umax_val) { | |
5238 | /* Overflow possible, we know nothing */ | |
5239 | dst_reg->umin_value = 0; | |
5240 | dst_reg->umax_value = U64_MAX; | |
5241 | } else { | |
5242 | /* Cannot overflow (as long as bounds are consistent) */ | |
5243 | dst_reg->umin_value -= umax_val; | |
5244 | dst_reg->umax_value -= umin_val; | |
5245 | } | |
3f50f132 JF |
5246 | } |
5247 | ||
5248 | static void scalar32_min_max_mul(struct bpf_reg_state *dst_reg, | |
5249 | struct bpf_reg_state *src_reg) | |
5250 | { | |
5251 | s32 smin_val = src_reg->s32_min_value; | |
5252 | u32 umin_val = src_reg->u32_min_value; | |
5253 | u32 umax_val = src_reg->u32_max_value; | |
5254 | ||
5255 | if (smin_val < 0 || dst_reg->s32_min_value < 0) { | |
5256 | /* Ain't nobody got time to multiply that sign */ | |
5257 | __mark_reg32_unbounded(dst_reg); | |
5258 | return; | |
5259 | } | |
5260 | /* Both values are positive, so we can work with unsigned and | |
5261 | * copy the result to signed (unless it exceeds S32_MAX). | |
5262 | */ | |
5263 | if (umax_val > U16_MAX || dst_reg->u32_max_value > U16_MAX) { | |
5264 | /* Potential overflow, we know nothing */ | |
5265 | __mark_reg32_unbounded(dst_reg); | |
5266 | return; | |
5267 | } | |
5268 | dst_reg->u32_min_value *= umin_val; | |
5269 | dst_reg->u32_max_value *= umax_val; | |
5270 | if (dst_reg->u32_max_value > S32_MAX) { | |
5271 | /* Overflow possible, we know nothing */ | |
5272 | dst_reg->s32_min_value = S32_MIN; | |
5273 | dst_reg->s32_max_value = S32_MAX; | |
5274 | } else { | |
5275 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
5276 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
5277 | } | |
07cd2631 JF |
5278 | } |
5279 | ||
5280 | static void scalar_min_max_mul(struct bpf_reg_state *dst_reg, | |
5281 | struct bpf_reg_state *src_reg) | |
5282 | { | |
5283 | s64 smin_val = src_reg->smin_value; | |
5284 | u64 umin_val = src_reg->umin_value; | |
5285 | u64 umax_val = src_reg->umax_value; | |
5286 | ||
07cd2631 JF |
5287 | if (smin_val < 0 || dst_reg->smin_value < 0) { |
5288 | /* Ain't nobody got time to multiply that sign */ | |
3f50f132 | 5289 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
5290 | return; |
5291 | } | |
5292 | /* Both values are positive, so we can work with unsigned and | |
5293 | * copy the result to signed (unless it exceeds S64_MAX). | |
5294 | */ | |
5295 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { | |
5296 | /* Potential overflow, we know nothing */ | |
3f50f132 | 5297 | __mark_reg64_unbounded(dst_reg); |
07cd2631 JF |
5298 | return; |
5299 | } | |
5300 | dst_reg->umin_value *= umin_val; | |
5301 | dst_reg->umax_value *= umax_val; | |
5302 | if (dst_reg->umax_value > S64_MAX) { | |
5303 | /* Overflow possible, we know nothing */ | |
5304 | dst_reg->smin_value = S64_MIN; | |
5305 | dst_reg->smax_value = S64_MAX; | |
5306 | } else { | |
5307 | dst_reg->smin_value = dst_reg->umin_value; | |
5308 | dst_reg->smax_value = dst_reg->umax_value; | |
5309 | } | |
5310 | } | |
5311 | ||
3f50f132 JF |
5312 | static void scalar32_min_max_and(struct bpf_reg_state *dst_reg, |
5313 | struct bpf_reg_state *src_reg) | |
5314 | { | |
5315 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
5316 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
5317 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5318 | s32 smin_val = src_reg->s32_min_value; | |
5319 | u32 umax_val = src_reg->u32_max_value; | |
5320 | ||
5321 | /* Assuming scalar64_min_max_and will be called so its safe | |
5322 | * to skip updating register for known 32-bit case. | |
5323 | */ | |
5324 | if (src_known && dst_known) | |
5325 | return; | |
5326 | ||
5327 | /* We get our minimum from the var_off, since that's inherently | |
5328 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
5329 | */ | |
5330 | dst_reg->u32_min_value = var32_off.value; | |
5331 | dst_reg->u32_max_value = min(dst_reg->u32_max_value, umax_val); | |
5332 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
5333 | /* Lose signed bounds when ANDing negative numbers, | |
5334 | * ain't nobody got time for that. | |
5335 | */ | |
5336 | dst_reg->s32_min_value = S32_MIN; | |
5337 | dst_reg->s32_max_value = S32_MAX; | |
5338 | } else { | |
5339 | /* ANDing two positives gives a positive, so safe to | |
5340 | * cast result into s64. | |
5341 | */ | |
5342 | dst_reg->s32_min_value = dst_reg->u32_min_value; | |
5343 | dst_reg->s32_max_value = dst_reg->u32_max_value; | |
5344 | } | |
5345 | ||
5346 | } | |
5347 | ||
07cd2631 JF |
5348 | static void scalar_min_max_and(struct bpf_reg_state *dst_reg, |
5349 | struct bpf_reg_state *src_reg) | |
5350 | { | |
3f50f132 JF |
5351 | bool src_known = tnum_is_const(src_reg->var_off); |
5352 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
5353 | s64 smin_val = src_reg->smin_value; |
5354 | u64 umax_val = src_reg->umax_value; | |
5355 | ||
3f50f132 JF |
5356 | if (src_known && dst_known) { |
5357 | __mark_reg_known(dst_reg, dst_reg->var_off.value & | |
5358 | src_reg->var_off.value); | |
5359 | return; | |
5360 | } | |
5361 | ||
07cd2631 JF |
5362 | /* We get our minimum from the var_off, since that's inherently |
5363 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
5364 | */ | |
07cd2631 JF |
5365 | dst_reg->umin_value = dst_reg->var_off.value; |
5366 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
5367 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
5368 | /* Lose signed bounds when ANDing negative numbers, | |
5369 | * ain't nobody got time for that. | |
5370 | */ | |
5371 | dst_reg->smin_value = S64_MIN; | |
5372 | dst_reg->smax_value = S64_MAX; | |
5373 | } else { | |
5374 | /* ANDing two positives gives a positive, so safe to | |
5375 | * cast result into s64. | |
5376 | */ | |
5377 | dst_reg->smin_value = dst_reg->umin_value; | |
5378 | dst_reg->smax_value = dst_reg->umax_value; | |
5379 | } | |
5380 | /* We may learn something more from the var_off */ | |
5381 | __update_reg_bounds(dst_reg); | |
5382 | } | |
5383 | ||
3f50f132 JF |
5384 | static void scalar32_min_max_or(struct bpf_reg_state *dst_reg, |
5385 | struct bpf_reg_state *src_reg) | |
5386 | { | |
5387 | bool src_known = tnum_subreg_is_const(src_reg->var_off); | |
5388 | bool dst_known = tnum_subreg_is_const(dst_reg->var_off); | |
5389 | struct tnum var32_off = tnum_subreg(dst_reg->var_off); | |
5390 | s32 smin_val = src_reg->smin_value; | |
5391 | u32 umin_val = src_reg->umin_value; | |
5392 | ||
5393 | /* Assuming scalar64_min_max_or will be called so it is safe | |
5394 | * to skip updating register for known case. | |
5395 | */ | |
5396 | if (src_known && dst_known) | |
5397 | return; | |
5398 | ||
5399 | /* We get our maximum from the var_off, and our minimum is the | |
5400 | * maximum of the operands' minima | |
5401 | */ | |
5402 | dst_reg->u32_min_value = max(dst_reg->u32_min_value, umin_val); | |
5403 | dst_reg->u32_max_value = var32_off.value | var32_off.mask; | |
5404 | if (dst_reg->s32_min_value < 0 || smin_val < 0) { | |
5405 | /* Lose signed bounds when ORing negative numbers, | |
5406 | * ain't nobody got time for that. | |
5407 | */ | |
5408 | dst_reg->s32_min_value = S32_MIN; | |
5409 | dst_reg->s32_max_value = S32_MAX; | |
5410 | } else { | |
5411 | /* ORing two positives gives a positive, so safe to | |
5412 | * cast result into s64. | |
5413 | */ | |
5414 | dst_reg->s32_min_value = dst_reg->umin_value; | |
5415 | dst_reg->s32_max_value = dst_reg->umax_value; | |
5416 | } | |
5417 | } | |
5418 | ||
07cd2631 JF |
5419 | static void scalar_min_max_or(struct bpf_reg_state *dst_reg, |
5420 | struct bpf_reg_state *src_reg) | |
5421 | { | |
3f50f132 JF |
5422 | bool src_known = tnum_is_const(src_reg->var_off); |
5423 | bool dst_known = tnum_is_const(dst_reg->var_off); | |
07cd2631 JF |
5424 | s64 smin_val = src_reg->smin_value; |
5425 | u64 umin_val = src_reg->umin_value; | |
5426 | ||
3f50f132 JF |
5427 | if (src_known && dst_known) { |
5428 | __mark_reg_known(dst_reg, dst_reg->var_off.value | | |
5429 | src_reg->var_off.value); | |
5430 | return; | |
5431 | } | |
5432 | ||
07cd2631 JF |
5433 | /* We get our maximum from the var_off, and our minimum is the |
5434 | * maximum of the operands' minima | |
5435 | */ | |
07cd2631 JF |
5436 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
5437 | dst_reg->umax_value = dst_reg->var_off.value | dst_reg->var_off.mask; | |
5438 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
5439 | /* Lose signed bounds when ORing negative numbers, | |
5440 | * ain't nobody got time for that. | |
5441 | */ | |
5442 | dst_reg->smin_value = S64_MIN; | |
5443 | dst_reg->smax_value = S64_MAX; | |
5444 | } else { | |
5445 | /* ORing two positives gives a positive, so safe to | |
5446 | * cast result into s64. | |
5447 | */ | |
5448 | dst_reg->smin_value = dst_reg->umin_value; | |
5449 | dst_reg->smax_value = dst_reg->umax_value; | |
5450 | } | |
5451 | /* We may learn something more from the var_off */ | |
5452 | __update_reg_bounds(dst_reg); | |
5453 | } | |
5454 | ||
3f50f132 JF |
5455 | static void __scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, |
5456 | u64 umin_val, u64 umax_val) | |
07cd2631 | 5457 | { |
07cd2631 JF |
5458 | /* We lose all sign bit information (except what we can pick |
5459 | * up from var_off) | |
5460 | */ | |
3f50f132 JF |
5461 | dst_reg->s32_min_value = S32_MIN; |
5462 | dst_reg->s32_max_value = S32_MAX; | |
5463 | /* If we might shift our top bit out, then we know nothing */ | |
5464 | if (umax_val > 31 || dst_reg->u32_max_value > 1ULL << (31 - umax_val)) { | |
5465 | dst_reg->u32_min_value = 0; | |
5466 | dst_reg->u32_max_value = U32_MAX; | |
5467 | } else { | |
5468 | dst_reg->u32_min_value <<= umin_val; | |
5469 | dst_reg->u32_max_value <<= umax_val; | |
5470 | } | |
5471 | } | |
5472 | ||
5473 | static void scalar32_min_max_lsh(struct bpf_reg_state *dst_reg, | |
5474 | struct bpf_reg_state *src_reg) | |
5475 | { | |
5476 | u32 umax_val = src_reg->u32_max_value; | |
5477 | u32 umin_val = src_reg->u32_min_value; | |
5478 | /* u32 alu operation will zext upper bits */ | |
5479 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
5480 | ||
5481 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
5482 | dst_reg->var_off = tnum_subreg(tnum_lshift(subreg, umin_val)); | |
5483 | /* Not required but being careful mark reg64 bounds as unknown so | |
5484 | * that we are forced to pick them up from tnum and zext later and | |
5485 | * if some path skips this step we are still safe. | |
5486 | */ | |
5487 | __mark_reg64_unbounded(dst_reg); | |
5488 | __update_reg32_bounds(dst_reg); | |
5489 | } | |
5490 | ||
5491 | static void __scalar64_min_max_lsh(struct bpf_reg_state *dst_reg, | |
5492 | u64 umin_val, u64 umax_val) | |
5493 | { | |
5494 | /* Special case <<32 because it is a common compiler pattern to sign | |
5495 | * extend subreg by doing <<32 s>>32. In this case if 32bit bounds are | |
5496 | * positive we know this shift will also be positive so we can track | |
5497 | * bounds correctly. Otherwise we lose all sign bit information except | |
5498 | * what we can pick up from var_off. Perhaps we can generalize this | |
5499 | * later to shifts of any length. | |
5500 | */ | |
5501 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_max_value >= 0) | |
5502 | dst_reg->smax_value = (s64)dst_reg->s32_max_value << 32; | |
5503 | else | |
5504 | dst_reg->smax_value = S64_MAX; | |
5505 | ||
5506 | if (umin_val == 32 && umax_val == 32 && dst_reg->s32_min_value >= 0) | |
5507 | dst_reg->smin_value = (s64)dst_reg->s32_min_value << 32; | |
5508 | else | |
5509 | dst_reg->smin_value = S64_MIN; | |
5510 | ||
07cd2631 JF |
5511 | /* If we might shift our top bit out, then we know nothing */ |
5512 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
5513 | dst_reg->umin_value = 0; | |
5514 | dst_reg->umax_value = U64_MAX; | |
5515 | } else { | |
5516 | dst_reg->umin_value <<= umin_val; | |
5517 | dst_reg->umax_value <<= umax_val; | |
5518 | } | |
3f50f132 JF |
5519 | } |
5520 | ||
5521 | static void scalar_min_max_lsh(struct bpf_reg_state *dst_reg, | |
5522 | struct bpf_reg_state *src_reg) | |
5523 | { | |
5524 | u64 umax_val = src_reg->umax_value; | |
5525 | u64 umin_val = src_reg->umin_value; | |
5526 | ||
5527 | /* scalar64 calc uses 32bit unshifted bounds so must be called first */ | |
5528 | __scalar64_min_max_lsh(dst_reg, umin_val, umax_val); | |
5529 | __scalar32_min_max_lsh(dst_reg, umin_val, umax_val); | |
5530 | ||
07cd2631 JF |
5531 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
5532 | /* We may learn something more from the var_off */ | |
5533 | __update_reg_bounds(dst_reg); | |
5534 | } | |
5535 | ||
3f50f132 JF |
5536 | static void scalar32_min_max_rsh(struct bpf_reg_state *dst_reg, |
5537 | struct bpf_reg_state *src_reg) | |
5538 | { | |
5539 | struct tnum subreg = tnum_subreg(dst_reg->var_off); | |
5540 | u32 umax_val = src_reg->u32_max_value; | |
5541 | u32 umin_val = src_reg->u32_min_value; | |
5542 | ||
5543 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
5544 | * be negative, then either: | |
5545 | * 1) src_reg might be zero, so the sign bit of the result is | |
5546 | * unknown, so we lose our signed bounds | |
5547 | * 2) it's known negative, thus the unsigned bounds capture the | |
5548 | * signed bounds | |
5549 | * 3) the signed bounds cross zero, so they tell us nothing | |
5550 | * about the result | |
5551 | * If the value in dst_reg is known nonnegative, then again the | |
5552 | * unsigned bounts capture the signed bounds. | |
5553 | * Thus, in all cases it suffices to blow away our signed bounds | |
5554 | * and rely on inferring new ones from the unsigned bounds and | |
5555 | * var_off of the result. | |
5556 | */ | |
5557 | dst_reg->s32_min_value = S32_MIN; | |
5558 | dst_reg->s32_max_value = S32_MAX; | |
5559 | ||
5560 | dst_reg->var_off = tnum_rshift(subreg, umin_val); | |
5561 | dst_reg->u32_min_value >>= umax_val; | |
5562 | dst_reg->u32_max_value >>= umin_val; | |
5563 | ||
5564 | __mark_reg64_unbounded(dst_reg); | |
5565 | __update_reg32_bounds(dst_reg); | |
5566 | } | |
5567 | ||
07cd2631 JF |
5568 | static void scalar_min_max_rsh(struct bpf_reg_state *dst_reg, |
5569 | struct bpf_reg_state *src_reg) | |
5570 | { | |
5571 | u64 umax_val = src_reg->umax_value; | |
5572 | u64 umin_val = src_reg->umin_value; | |
5573 | ||
5574 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might | |
5575 | * be negative, then either: | |
5576 | * 1) src_reg might be zero, so the sign bit of the result is | |
5577 | * unknown, so we lose our signed bounds | |
5578 | * 2) it's known negative, thus the unsigned bounds capture the | |
5579 | * signed bounds | |
5580 | * 3) the signed bounds cross zero, so they tell us nothing | |
5581 | * about the result | |
5582 | * If the value in dst_reg is known nonnegative, then again the | |
5583 | * unsigned bounts capture the signed bounds. | |
5584 | * Thus, in all cases it suffices to blow away our signed bounds | |
5585 | * and rely on inferring new ones from the unsigned bounds and | |
5586 | * var_off of the result. | |
5587 | */ | |
5588 | dst_reg->smin_value = S64_MIN; | |
5589 | dst_reg->smax_value = S64_MAX; | |
5590 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); | |
5591 | dst_reg->umin_value >>= umax_val; | |
5592 | dst_reg->umax_value >>= umin_val; | |
3f50f132 JF |
5593 | |
5594 | /* Its not easy to operate on alu32 bounds here because it depends | |
5595 | * on bits being shifted in. Take easy way out and mark unbounded | |
5596 | * so we can recalculate later from tnum. | |
5597 | */ | |
5598 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
5599 | __update_reg_bounds(dst_reg); |
5600 | } | |
5601 | ||
3f50f132 JF |
5602 | static void scalar32_min_max_arsh(struct bpf_reg_state *dst_reg, |
5603 | struct bpf_reg_state *src_reg) | |
07cd2631 | 5604 | { |
3f50f132 | 5605 | u64 umin_val = src_reg->u32_min_value; |
07cd2631 JF |
5606 | |
5607 | /* Upon reaching here, src_known is true and | |
5608 | * umax_val is equal to umin_val. | |
5609 | */ | |
3f50f132 JF |
5610 | dst_reg->s32_min_value = (u32)(((s32)dst_reg->s32_min_value) >> umin_val); |
5611 | dst_reg->s32_max_value = (u32)(((s32)dst_reg->s32_max_value) >> umin_val); | |
07cd2631 | 5612 | |
3f50f132 JF |
5613 | dst_reg->var_off = tnum_arshift(tnum_subreg(dst_reg->var_off), umin_val, 32); |
5614 | ||
5615 | /* blow away the dst_reg umin_value/umax_value and rely on | |
5616 | * dst_reg var_off to refine the result. | |
5617 | */ | |
5618 | dst_reg->u32_min_value = 0; | |
5619 | dst_reg->u32_max_value = U32_MAX; | |
5620 | ||
5621 | __mark_reg64_unbounded(dst_reg); | |
5622 | __update_reg32_bounds(dst_reg); | |
5623 | } | |
5624 | ||
5625 | static void scalar_min_max_arsh(struct bpf_reg_state *dst_reg, | |
5626 | struct bpf_reg_state *src_reg) | |
5627 | { | |
5628 | u64 umin_val = src_reg->umin_value; | |
5629 | ||
5630 | /* Upon reaching here, src_known is true and umax_val is equal | |
5631 | * to umin_val. | |
5632 | */ | |
5633 | dst_reg->smin_value >>= umin_val; | |
5634 | dst_reg->smax_value >>= umin_val; | |
5635 | ||
5636 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val, 64); | |
07cd2631 JF |
5637 | |
5638 | /* blow away the dst_reg umin_value/umax_value and rely on | |
5639 | * dst_reg var_off to refine the result. | |
5640 | */ | |
5641 | dst_reg->umin_value = 0; | |
5642 | dst_reg->umax_value = U64_MAX; | |
3f50f132 JF |
5643 | |
5644 | /* Its not easy to operate on alu32 bounds here because it depends | |
5645 | * on bits being shifted in from upper 32-bits. Take easy way out | |
5646 | * and mark unbounded so we can recalculate later from tnum. | |
5647 | */ | |
5648 | __mark_reg32_unbounded(dst_reg); | |
07cd2631 JF |
5649 | __update_reg_bounds(dst_reg); |
5650 | } | |
5651 | ||
468f6eaf JH |
5652 | /* WARNING: This function does calculations on 64-bit values, but the actual |
5653 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
5654 | * need extra checks in the 32-bit case. | |
5655 | */ | |
f1174f77 EC |
5656 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
5657 | struct bpf_insn *insn, | |
5658 | struct bpf_reg_state *dst_reg, | |
5659 | struct bpf_reg_state src_reg) | |
969bf05e | 5660 | { |
638f5b90 | 5661 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 5662 | u8 opcode = BPF_OP(insn->code); |
b0b3fb67 | 5663 | bool src_known; |
b03c9f9f EC |
5664 | s64 smin_val, smax_val; |
5665 | u64 umin_val, umax_val; | |
3f50f132 JF |
5666 | s32 s32_min_val, s32_max_val; |
5667 | u32 u32_min_val, u32_max_val; | |
468f6eaf | 5668 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
d3bd7413 DB |
5669 | u32 dst = insn->dst_reg; |
5670 | int ret; | |
3f50f132 | 5671 | bool alu32 = (BPF_CLASS(insn->code) != BPF_ALU64); |
b799207e | 5672 | |
b03c9f9f EC |
5673 | smin_val = src_reg.smin_value; |
5674 | smax_val = src_reg.smax_value; | |
5675 | umin_val = src_reg.umin_value; | |
5676 | umax_val = src_reg.umax_value; | |
f23cc643 | 5677 | |
3f50f132 JF |
5678 | s32_min_val = src_reg.s32_min_value; |
5679 | s32_max_val = src_reg.s32_max_value; | |
5680 | u32_min_val = src_reg.u32_min_value; | |
5681 | u32_max_val = src_reg.u32_max_value; | |
5682 | ||
5683 | if (alu32) { | |
5684 | src_known = tnum_subreg_is_const(src_reg.var_off); | |
3f50f132 JF |
5685 | if ((src_known && |
5686 | (s32_min_val != s32_max_val || u32_min_val != u32_max_val)) || | |
5687 | s32_min_val > s32_max_val || u32_min_val > u32_max_val) { | |
5688 | /* Taint dst register if offset had invalid bounds | |
5689 | * derived from e.g. dead branches. | |
5690 | */ | |
5691 | __mark_reg_unknown(env, dst_reg); | |
5692 | return 0; | |
5693 | } | |
5694 | } else { | |
5695 | src_known = tnum_is_const(src_reg.var_off); | |
3f50f132 JF |
5696 | if ((src_known && |
5697 | (smin_val != smax_val || umin_val != umax_val)) || | |
5698 | smin_val > smax_val || umin_val > umax_val) { | |
5699 | /* Taint dst register if offset had invalid bounds | |
5700 | * derived from e.g. dead branches. | |
5701 | */ | |
5702 | __mark_reg_unknown(env, dst_reg); | |
5703 | return 0; | |
5704 | } | |
6f16101e DB |
5705 | } |
5706 | ||
bb7f0f98 AS |
5707 | if (!src_known && |
5708 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
f54c7898 | 5709 | __mark_reg_unknown(env, dst_reg); |
bb7f0f98 AS |
5710 | return 0; |
5711 | } | |
5712 | ||
3f50f132 JF |
5713 | /* Calculate sign/unsigned bounds and tnum for alu32 and alu64 bit ops. |
5714 | * There are two classes of instructions: The first class we track both | |
5715 | * alu32 and alu64 sign/unsigned bounds independently this provides the | |
5716 | * greatest amount of precision when alu operations are mixed with jmp32 | |
5717 | * operations. These operations are BPF_ADD, BPF_SUB, BPF_MUL, BPF_ADD, | |
5718 | * and BPF_OR. This is possible because these ops have fairly easy to | |
5719 | * understand and calculate behavior in both 32-bit and 64-bit alu ops. | |
5720 | * See alu32 verifier tests for examples. The second class of | |
5721 | * operations, BPF_LSH, BPF_RSH, and BPF_ARSH, however are not so easy | |
5722 | * with regards to tracking sign/unsigned bounds because the bits may | |
5723 | * cross subreg boundaries in the alu64 case. When this happens we mark | |
5724 | * the reg unbounded in the subreg bound space and use the resulting | |
5725 | * tnum to calculate an approximation of the sign/unsigned bounds. | |
5726 | */ | |
48461135 JB |
5727 | switch (opcode) { |
5728 | case BPF_ADD: | |
d3bd7413 DB |
5729 | ret = sanitize_val_alu(env, insn); |
5730 | if (ret < 0) { | |
5731 | verbose(env, "R%d tried to add from different pointers or scalars\n", dst); | |
5732 | return ret; | |
5733 | } | |
3f50f132 | 5734 | scalar32_min_max_add(dst_reg, &src_reg); |
07cd2631 | 5735 | scalar_min_max_add(dst_reg, &src_reg); |
3f50f132 | 5736 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
5737 | break; |
5738 | case BPF_SUB: | |
d3bd7413 DB |
5739 | ret = sanitize_val_alu(env, insn); |
5740 | if (ret < 0) { | |
5741 | verbose(env, "R%d tried to sub from different pointers or scalars\n", dst); | |
5742 | return ret; | |
5743 | } | |
3f50f132 | 5744 | scalar32_min_max_sub(dst_reg, &src_reg); |
07cd2631 | 5745 | scalar_min_max_sub(dst_reg, &src_reg); |
3f50f132 | 5746 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
5747 | break; |
5748 | case BPF_MUL: | |
3f50f132 JF |
5749 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
5750 | scalar32_min_max_mul(dst_reg, &src_reg); | |
07cd2631 | 5751 | scalar_min_max_mul(dst_reg, &src_reg); |
48461135 JB |
5752 | break; |
5753 | case BPF_AND: | |
3f50f132 JF |
5754 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
5755 | scalar32_min_max_and(dst_reg, &src_reg); | |
07cd2631 | 5756 | scalar_min_max_and(dst_reg, &src_reg); |
f1174f77 EC |
5757 | break; |
5758 | case BPF_OR: | |
3f50f132 JF |
5759 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); |
5760 | scalar32_min_max_or(dst_reg, &src_reg); | |
07cd2631 | 5761 | scalar_min_max_or(dst_reg, &src_reg); |
48461135 JB |
5762 | break; |
5763 | case BPF_LSH: | |
468f6eaf JH |
5764 | if (umax_val >= insn_bitness) { |
5765 | /* Shifts greater than 31 or 63 are undefined. | |
5766 | * This includes shifts by a negative number. | |
b03c9f9f | 5767 | */ |
61bd5218 | 5768 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
5769 | break; |
5770 | } | |
3f50f132 JF |
5771 | if (alu32) |
5772 | scalar32_min_max_lsh(dst_reg, &src_reg); | |
5773 | else | |
5774 | scalar_min_max_lsh(dst_reg, &src_reg); | |
48461135 JB |
5775 | break; |
5776 | case BPF_RSH: | |
468f6eaf JH |
5777 | if (umax_val >= insn_bitness) { |
5778 | /* Shifts greater than 31 or 63 are undefined. | |
5779 | * This includes shifts by a negative number. | |
b03c9f9f | 5780 | */ |
61bd5218 | 5781 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
5782 | break; |
5783 | } | |
3f50f132 JF |
5784 | if (alu32) |
5785 | scalar32_min_max_rsh(dst_reg, &src_reg); | |
5786 | else | |
5787 | scalar_min_max_rsh(dst_reg, &src_reg); | |
48461135 | 5788 | break; |
9cbe1f5a YS |
5789 | case BPF_ARSH: |
5790 | if (umax_val >= insn_bitness) { | |
5791 | /* Shifts greater than 31 or 63 are undefined. | |
5792 | * This includes shifts by a negative number. | |
5793 | */ | |
5794 | mark_reg_unknown(env, regs, insn->dst_reg); | |
5795 | break; | |
5796 | } | |
3f50f132 JF |
5797 | if (alu32) |
5798 | scalar32_min_max_arsh(dst_reg, &src_reg); | |
5799 | else | |
5800 | scalar_min_max_arsh(dst_reg, &src_reg); | |
9cbe1f5a | 5801 | break; |
48461135 | 5802 | default: |
61bd5218 | 5803 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
5804 | break; |
5805 | } | |
5806 | ||
3f50f132 JF |
5807 | /* ALU32 ops are zero extended into 64bit register */ |
5808 | if (alu32) | |
5809 | zext_32_to_64(dst_reg); | |
468f6eaf | 5810 | |
294f2fc6 | 5811 | __update_reg_bounds(dst_reg); |
b03c9f9f EC |
5812 | __reg_deduce_bounds(dst_reg); |
5813 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
5814 | return 0; |
5815 | } | |
5816 | ||
5817 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
5818 | * and var_off. | |
5819 | */ | |
5820 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
5821 | struct bpf_insn *insn) | |
5822 | { | |
f4d7e40a AS |
5823 | struct bpf_verifier_state *vstate = env->cur_state; |
5824 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5825 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
5826 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
5827 | u8 opcode = BPF_OP(insn->code); | |
b5dc0163 | 5828 | int err; |
f1174f77 EC |
5829 | |
5830 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
5831 | src_reg = NULL; |
5832 | if (dst_reg->type != SCALAR_VALUE) | |
5833 | ptr_reg = dst_reg; | |
5834 | if (BPF_SRC(insn->code) == BPF_X) { | |
5835 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
5836 | if (src_reg->type != SCALAR_VALUE) { |
5837 | if (dst_reg->type != SCALAR_VALUE) { | |
5838 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
5839 | * an arbitrary scalar. Disallow all math except |
5840 | * pointer subtraction | |
f1174f77 | 5841 | */ |
dd066823 | 5842 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
5843 | mark_reg_unknown(env, regs, insn->dst_reg); |
5844 | return 0; | |
f1174f77 | 5845 | } |
82abbf8d AS |
5846 | verbose(env, "R%d pointer %s pointer prohibited\n", |
5847 | insn->dst_reg, | |
5848 | bpf_alu_string[opcode >> 4]); | |
5849 | return -EACCES; | |
f1174f77 EC |
5850 | } else { |
5851 | /* scalar += pointer | |
5852 | * This is legal, but we have to reverse our | |
5853 | * src/dest handling in computing the range | |
5854 | */ | |
b5dc0163 AS |
5855 | err = mark_chain_precision(env, insn->dst_reg); |
5856 | if (err) | |
5857 | return err; | |
82abbf8d AS |
5858 | return adjust_ptr_min_max_vals(env, insn, |
5859 | src_reg, dst_reg); | |
f1174f77 EC |
5860 | } |
5861 | } else if (ptr_reg) { | |
5862 | /* pointer += scalar */ | |
b5dc0163 AS |
5863 | err = mark_chain_precision(env, insn->src_reg); |
5864 | if (err) | |
5865 | return err; | |
82abbf8d AS |
5866 | return adjust_ptr_min_max_vals(env, insn, |
5867 | dst_reg, src_reg); | |
f1174f77 EC |
5868 | } |
5869 | } else { | |
5870 | /* Pretend the src is a reg with a known value, since we only | |
5871 | * need to be able to read from this state. | |
5872 | */ | |
5873 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 5874 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 5875 | src_reg = &off_reg; |
82abbf8d AS |
5876 | if (ptr_reg) /* pointer += K */ |
5877 | return adjust_ptr_min_max_vals(env, insn, | |
5878 | ptr_reg, src_reg); | |
f1174f77 EC |
5879 | } |
5880 | ||
5881 | /* Got here implies adding two SCALAR_VALUEs */ | |
5882 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 5883 | print_verifier_state(env, state); |
61bd5218 | 5884 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
5885 | return -EINVAL; |
5886 | } | |
5887 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 5888 | print_verifier_state(env, state); |
61bd5218 | 5889 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
5890 | return -EINVAL; |
5891 | } | |
5892 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
5893 | } |
5894 | ||
17a52670 | 5895 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 5896 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 5897 | { |
638f5b90 | 5898 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
5899 | u8 opcode = BPF_OP(insn->code); |
5900 | int err; | |
5901 | ||
5902 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
5903 | if (opcode == BPF_NEG) { | |
5904 | if (BPF_SRC(insn->code) != 0 || | |
5905 | insn->src_reg != BPF_REG_0 || | |
5906 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 5907 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
5908 | return -EINVAL; |
5909 | } | |
5910 | } else { | |
5911 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
5912 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
5913 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 5914 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
5915 | return -EINVAL; |
5916 | } | |
5917 | } | |
5918 | ||
5919 | /* check src operand */ | |
dc503a8a | 5920 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
5921 | if (err) |
5922 | return err; | |
5923 | ||
1be7f75d | 5924 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 5925 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
5926 | insn->dst_reg); |
5927 | return -EACCES; | |
5928 | } | |
5929 | ||
17a52670 | 5930 | /* check dest operand */ |
dc503a8a | 5931 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
5932 | if (err) |
5933 | return err; | |
5934 | ||
5935 | } else if (opcode == BPF_MOV) { | |
5936 | ||
5937 | if (BPF_SRC(insn->code) == BPF_X) { | |
5938 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 5939 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
5940 | return -EINVAL; |
5941 | } | |
5942 | ||
5943 | /* check src operand */ | |
dc503a8a | 5944 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
5945 | if (err) |
5946 | return err; | |
5947 | } else { | |
5948 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 5949 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
5950 | return -EINVAL; |
5951 | } | |
5952 | } | |
5953 | ||
fbeb1603 AF |
5954 | /* check dest operand, mark as required later */ |
5955 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
5956 | if (err) |
5957 | return err; | |
5958 | ||
5959 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
5960 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
5961 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
5962 | ||
17a52670 AS |
5963 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
5964 | /* case: R1 = R2 | |
5965 | * copy register state to dest reg | |
5966 | */ | |
e434b8cd JW |
5967 | *dst_reg = *src_reg; |
5968 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 5969 | dst_reg->subreg_def = DEF_NOT_SUBREG; |
17a52670 | 5970 | } else { |
f1174f77 | 5971 | /* R1 = (u32) R2 */ |
1be7f75d | 5972 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
5973 | verbose(env, |
5974 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
5975 | insn->src_reg); |
5976 | return -EACCES; | |
e434b8cd JW |
5977 | } else if (src_reg->type == SCALAR_VALUE) { |
5978 | *dst_reg = *src_reg; | |
5979 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 5980 | dst_reg->subreg_def = env->insn_idx + 1; |
e434b8cd JW |
5981 | } else { |
5982 | mark_reg_unknown(env, regs, | |
5983 | insn->dst_reg); | |
1be7f75d | 5984 | } |
3f50f132 | 5985 | zext_32_to_64(dst_reg); |
17a52670 AS |
5986 | } |
5987 | } else { | |
5988 | /* case: R = imm | |
5989 | * remember the value we stored into this reg | |
5990 | */ | |
fbeb1603 AF |
5991 | /* clear any state __mark_reg_known doesn't set */ |
5992 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 5993 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
5994 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
5995 | __mark_reg_known(regs + insn->dst_reg, | |
5996 | insn->imm); | |
5997 | } else { | |
5998 | __mark_reg_known(regs + insn->dst_reg, | |
5999 | (u32)insn->imm); | |
6000 | } | |
17a52670 AS |
6001 | } |
6002 | ||
6003 | } else if (opcode > BPF_END) { | |
61bd5218 | 6004 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
6005 | return -EINVAL; |
6006 | ||
6007 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
6008 | ||
17a52670 AS |
6009 | if (BPF_SRC(insn->code) == BPF_X) { |
6010 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 6011 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
6012 | return -EINVAL; |
6013 | } | |
6014 | /* check src1 operand */ | |
dc503a8a | 6015 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6016 | if (err) |
6017 | return err; | |
6018 | } else { | |
6019 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 6020 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
6021 | return -EINVAL; |
6022 | } | |
6023 | } | |
6024 | ||
6025 | /* check src2 operand */ | |
dc503a8a | 6026 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6027 | if (err) |
6028 | return err; | |
6029 | ||
6030 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
6031 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 6032 | verbose(env, "div by zero\n"); |
17a52670 AS |
6033 | return -EINVAL; |
6034 | } | |
6035 | ||
229394e8 RV |
6036 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
6037 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
6038 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
6039 | ||
6040 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 6041 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
6042 | return -EINVAL; |
6043 | } | |
6044 | } | |
6045 | ||
1a0dc1ac | 6046 | /* check dest operand */ |
dc503a8a | 6047 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
6048 | if (err) |
6049 | return err; | |
6050 | ||
f1174f77 | 6051 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
6052 | } |
6053 | ||
6054 | return 0; | |
6055 | } | |
6056 | ||
c6a9efa1 PC |
6057 | static void __find_good_pkt_pointers(struct bpf_func_state *state, |
6058 | struct bpf_reg_state *dst_reg, | |
6059 | enum bpf_reg_type type, u16 new_range) | |
6060 | { | |
6061 | struct bpf_reg_state *reg; | |
6062 | int i; | |
6063 | ||
6064 | for (i = 0; i < MAX_BPF_REG; i++) { | |
6065 | reg = &state->regs[i]; | |
6066 | if (reg->type == type && reg->id == dst_reg->id) | |
6067 | /* keep the maximum range already checked */ | |
6068 | reg->range = max(reg->range, new_range); | |
6069 | } | |
6070 | ||
6071 | bpf_for_each_spilled_reg(i, state, reg) { | |
6072 | if (!reg) | |
6073 | continue; | |
6074 | if (reg->type == type && reg->id == dst_reg->id) | |
6075 | reg->range = max(reg->range, new_range); | |
6076 | } | |
6077 | } | |
6078 | ||
f4d7e40a | 6079 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 6080 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 6081 | enum bpf_reg_type type, |
fb2a311a | 6082 | bool range_right_open) |
969bf05e | 6083 | { |
fb2a311a | 6084 | u16 new_range; |
c6a9efa1 | 6085 | int i; |
2d2be8ca | 6086 | |
fb2a311a DB |
6087 | if (dst_reg->off < 0 || |
6088 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
6089 | /* This doesn't give us any range */ |
6090 | return; | |
6091 | ||
b03c9f9f EC |
6092 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
6093 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
6094 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
6095 | * than pkt_end, but that's because it's also less than pkt. | |
6096 | */ | |
6097 | return; | |
6098 | ||
fb2a311a DB |
6099 | new_range = dst_reg->off; |
6100 | if (range_right_open) | |
6101 | new_range--; | |
6102 | ||
6103 | /* Examples for register markings: | |
2d2be8ca | 6104 | * |
fb2a311a | 6105 | * pkt_data in dst register: |
2d2be8ca DB |
6106 | * |
6107 | * r2 = r3; | |
6108 | * r2 += 8; | |
6109 | * if (r2 > pkt_end) goto <handle exception> | |
6110 | * <access okay> | |
6111 | * | |
b4e432f1 DB |
6112 | * r2 = r3; |
6113 | * r2 += 8; | |
6114 | * if (r2 < pkt_end) goto <access okay> | |
6115 | * <handle exception> | |
6116 | * | |
2d2be8ca DB |
6117 | * Where: |
6118 | * r2 == dst_reg, pkt_end == src_reg | |
6119 | * r2=pkt(id=n,off=8,r=0) | |
6120 | * r3=pkt(id=n,off=0,r=0) | |
6121 | * | |
fb2a311a | 6122 | * pkt_data in src register: |
2d2be8ca DB |
6123 | * |
6124 | * r2 = r3; | |
6125 | * r2 += 8; | |
6126 | * if (pkt_end >= r2) goto <access okay> | |
6127 | * <handle exception> | |
6128 | * | |
b4e432f1 DB |
6129 | * r2 = r3; |
6130 | * r2 += 8; | |
6131 | * if (pkt_end <= r2) goto <handle exception> | |
6132 | * <access okay> | |
6133 | * | |
2d2be8ca DB |
6134 | * Where: |
6135 | * pkt_end == dst_reg, r2 == src_reg | |
6136 | * r2=pkt(id=n,off=8,r=0) | |
6137 | * r3=pkt(id=n,off=0,r=0) | |
6138 | * | |
6139 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
6140 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
6141 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
6142 | * the check. | |
969bf05e | 6143 | */ |
2d2be8ca | 6144 | |
f1174f77 EC |
6145 | /* If our ids match, then we must have the same max_value. And we |
6146 | * don't care about the other reg's fixed offset, since if it's too big | |
6147 | * the range won't allow anything. | |
6148 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
6149 | */ | |
c6a9efa1 PC |
6150 | for (i = 0; i <= vstate->curframe; i++) |
6151 | __find_good_pkt_pointers(vstate->frame[i], dst_reg, type, | |
6152 | new_range); | |
969bf05e AS |
6153 | } |
6154 | ||
3f50f132 | 6155 | static int is_branch32_taken(struct bpf_reg_state *reg, u32 val, u8 opcode) |
4f7b3e82 | 6156 | { |
3f50f132 JF |
6157 | struct tnum subreg = tnum_subreg(reg->var_off); |
6158 | s32 sval = (s32)val; | |
a72dafaf | 6159 | |
3f50f132 JF |
6160 | switch (opcode) { |
6161 | case BPF_JEQ: | |
6162 | if (tnum_is_const(subreg)) | |
6163 | return !!tnum_equals_const(subreg, val); | |
6164 | break; | |
6165 | case BPF_JNE: | |
6166 | if (tnum_is_const(subreg)) | |
6167 | return !tnum_equals_const(subreg, val); | |
6168 | break; | |
6169 | case BPF_JSET: | |
6170 | if ((~subreg.mask & subreg.value) & val) | |
6171 | return 1; | |
6172 | if (!((subreg.mask | subreg.value) & val)) | |
6173 | return 0; | |
6174 | break; | |
6175 | case BPF_JGT: | |
6176 | if (reg->u32_min_value > val) | |
6177 | return 1; | |
6178 | else if (reg->u32_max_value <= val) | |
6179 | return 0; | |
6180 | break; | |
6181 | case BPF_JSGT: | |
6182 | if (reg->s32_min_value > sval) | |
6183 | return 1; | |
6184 | else if (reg->s32_max_value < sval) | |
6185 | return 0; | |
6186 | break; | |
6187 | case BPF_JLT: | |
6188 | if (reg->u32_max_value < val) | |
6189 | return 1; | |
6190 | else if (reg->u32_min_value >= val) | |
6191 | return 0; | |
6192 | break; | |
6193 | case BPF_JSLT: | |
6194 | if (reg->s32_max_value < sval) | |
6195 | return 1; | |
6196 | else if (reg->s32_min_value >= sval) | |
6197 | return 0; | |
6198 | break; | |
6199 | case BPF_JGE: | |
6200 | if (reg->u32_min_value >= val) | |
6201 | return 1; | |
6202 | else if (reg->u32_max_value < val) | |
6203 | return 0; | |
6204 | break; | |
6205 | case BPF_JSGE: | |
6206 | if (reg->s32_min_value >= sval) | |
6207 | return 1; | |
6208 | else if (reg->s32_max_value < sval) | |
6209 | return 0; | |
6210 | break; | |
6211 | case BPF_JLE: | |
6212 | if (reg->u32_max_value <= val) | |
6213 | return 1; | |
6214 | else if (reg->u32_min_value > val) | |
6215 | return 0; | |
6216 | break; | |
6217 | case BPF_JSLE: | |
6218 | if (reg->s32_max_value <= sval) | |
6219 | return 1; | |
6220 | else if (reg->s32_min_value > sval) | |
6221 | return 0; | |
6222 | break; | |
6223 | } | |
4f7b3e82 | 6224 | |
3f50f132 JF |
6225 | return -1; |
6226 | } | |
092ed096 | 6227 | |
3f50f132 JF |
6228 | |
6229 | static int is_branch64_taken(struct bpf_reg_state *reg, u64 val, u8 opcode) | |
6230 | { | |
6231 | s64 sval = (s64)val; | |
a72dafaf | 6232 | |
4f7b3e82 AS |
6233 | switch (opcode) { |
6234 | case BPF_JEQ: | |
6235 | if (tnum_is_const(reg->var_off)) | |
6236 | return !!tnum_equals_const(reg->var_off, val); | |
6237 | break; | |
6238 | case BPF_JNE: | |
6239 | if (tnum_is_const(reg->var_off)) | |
6240 | return !tnum_equals_const(reg->var_off, val); | |
6241 | break; | |
960ea056 JK |
6242 | case BPF_JSET: |
6243 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
6244 | return 1; | |
6245 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
6246 | return 0; | |
6247 | break; | |
4f7b3e82 AS |
6248 | case BPF_JGT: |
6249 | if (reg->umin_value > val) | |
6250 | return 1; | |
6251 | else if (reg->umax_value <= val) | |
6252 | return 0; | |
6253 | break; | |
6254 | case BPF_JSGT: | |
a72dafaf | 6255 | if (reg->smin_value > sval) |
4f7b3e82 | 6256 | return 1; |
a72dafaf | 6257 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
6258 | return 0; |
6259 | break; | |
6260 | case BPF_JLT: | |
6261 | if (reg->umax_value < val) | |
6262 | return 1; | |
6263 | else if (reg->umin_value >= val) | |
6264 | return 0; | |
6265 | break; | |
6266 | case BPF_JSLT: | |
a72dafaf | 6267 | if (reg->smax_value < sval) |
4f7b3e82 | 6268 | return 1; |
a72dafaf | 6269 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
6270 | return 0; |
6271 | break; | |
6272 | case BPF_JGE: | |
6273 | if (reg->umin_value >= val) | |
6274 | return 1; | |
6275 | else if (reg->umax_value < val) | |
6276 | return 0; | |
6277 | break; | |
6278 | case BPF_JSGE: | |
a72dafaf | 6279 | if (reg->smin_value >= sval) |
4f7b3e82 | 6280 | return 1; |
a72dafaf | 6281 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
6282 | return 0; |
6283 | break; | |
6284 | case BPF_JLE: | |
6285 | if (reg->umax_value <= val) | |
6286 | return 1; | |
6287 | else if (reg->umin_value > val) | |
6288 | return 0; | |
6289 | break; | |
6290 | case BPF_JSLE: | |
a72dafaf | 6291 | if (reg->smax_value <= sval) |
4f7b3e82 | 6292 | return 1; |
a72dafaf | 6293 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
6294 | return 0; |
6295 | break; | |
6296 | } | |
6297 | ||
6298 | return -1; | |
6299 | } | |
6300 | ||
3f50f132 JF |
6301 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
6302 | * and return: | |
6303 | * 1 - branch will be taken and "goto target" will be executed | |
6304 | * 0 - branch will not be taken and fall-through to next insn | |
6305 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value | |
6306 | * range [0,10] | |
604dca5e | 6307 | */ |
3f50f132 JF |
6308 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
6309 | bool is_jmp32) | |
604dca5e | 6310 | { |
3f50f132 JF |
6311 | if (__is_pointer_value(false, reg)) |
6312 | return -1; | |
604dca5e | 6313 | |
3f50f132 JF |
6314 | if (is_jmp32) |
6315 | return is_branch32_taken(reg, val, opcode); | |
6316 | return is_branch64_taken(reg, val, opcode); | |
604dca5e JH |
6317 | } |
6318 | ||
48461135 JB |
6319 | /* Adjusts the register min/max values in the case that the dst_reg is the |
6320 | * variable register that we are working on, and src_reg is a constant or we're | |
6321 | * simply doing a BPF_K check. | |
f1174f77 | 6322 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
6323 | */ |
6324 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
6325 | struct bpf_reg_state *false_reg, |
6326 | u64 val, u32 val32, | |
092ed096 | 6327 | u8 opcode, bool is_jmp32) |
48461135 | 6328 | { |
3f50f132 JF |
6329 | struct tnum false_32off = tnum_subreg(false_reg->var_off); |
6330 | struct tnum false_64off = false_reg->var_off; | |
6331 | struct tnum true_32off = tnum_subreg(true_reg->var_off); | |
6332 | struct tnum true_64off = true_reg->var_off; | |
6333 | s64 sval = (s64)val; | |
6334 | s32 sval32 = (s32)val32; | |
a72dafaf | 6335 | |
f1174f77 EC |
6336 | /* If the dst_reg is a pointer, we can't learn anything about its |
6337 | * variable offset from the compare (unless src_reg were a pointer into | |
6338 | * the same object, but we don't bother with that. | |
6339 | * Since false_reg and true_reg have the same type by construction, we | |
6340 | * only need to check one of them for pointerness. | |
6341 | */ | |
6342 | if (__is_pointer_value(false, false_reg)) | |
6343 | return; | |
4cabc5b1 | 6344 | |
48461135 JB |
6345 | switch (opcode) { |
6346 | case BPF_JEQ: | |
48461135 | 6347 | case BPF_JNE: |
a72dafaf JW |
6348 | { |
6349 | struct bpf_reg_state *reg = | |
6350 | opcode == BPF_JEQ ? true_reg : false_reg; | |
6351 | ||
6352 | /* For BPF_JEQ, if this is false we know nothing Jon Snow, but | |
6353 | * if it is true we know the value for sure. Likewise for | |
6354 | * BPF_JNE. | |
48461135 | 6355 | */ |
3f50f132 JF |
6356 | if (is_jmp32) |
6357 | __mark_reg32_known(reg, val32); | |
6358 | else | |
092ed096 | 6359 | __mark_reg_known(reg, val); |
48461135 | 6360 | break; |
a72dafaf | 6361 | } |
960ea056 | 6362 | case BPF_JSET: |
3f50f132 JF |
6363 | if (is_jmp32) { |
6364 | false_32off = tnum_and(false_32off, tnum_const(~val32)); | |
6365 | if (is_power_of_2(val32)) | |
6366 | true_32off = tnum_or(true_32off, | |
6367 | tnum_const(val32)); | |
6368 | } else { | |
6369 | false_64off = tnum_and(false_64off, tnum_const(~val)); | |
6370 | if (is_power_of_2(val)) | |
6371 | true_64off = tnum_or(true_64off, | |
6372 | tnum_const(val)); | |
6373 | } | |
960ea056 | 6374 | break; |
48461135 | 6375 | case BPF_JGE: |
a72dafaf JW |
6376 | case BPF_JGT: |
6377 | { | |
3f50f132 JF |
6378 | if (is_jmp32) { |
6379 | u32 false_umax = opcode == BPF_JGT ? val32 : val32 - 1; | |
6380 | u32 true_umin = opcode == BPF_JGT ? val32 + 1 : val32; | |
6381 | ||
6382 | false_reg->u32_max_value = min(false_reg->u32_max_value, | |
6383 | false_umax); | |
6384 | true_reg->u32_min_value = max(true_reg->u32_min_value, | |
6385 | true_umin); | |
6386 | } else { | |
6387 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
6388 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
6389 | ||
6390 | false_reg->umax_value = min(false_reg->umax_value, false_umax); | |
6391 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
6392 | } | |
b03c9f9f | 6393 | break; |
a72dafaf | 6394 | } |
48461135 | 6395 | case BPF_JSGE: |
a72dafaf JW |
6396 | case BPF_JSGT: |
6397 | { | |
3f50f132 JF |
6398 | if (is_jmp32) { |
6399 | s32 false_smax = opcode == BPF_JSGT ? sval32 : sval32 - 1; | |
6400 | s32 true_smin = opcode == BPF_JSGT ? sval32 + 1 : sval32; | |
a72dafaf | 6401 | |
3f50f132 JF |
6402 | false_reg->s32_max_value = min(false_reg->s32_max_value, false_smax); |
6403 | true_reg->s32_min_value = max(true_reg->s32_min_value, true_smin); | |
6404 | } else { | |
6405 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
6406 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
6407 | ||
6408 | false_reg->smax_value = min(false_reg->smax_value, false_smax); | |
6409 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
6410 | } | |
48461135 | 6411 | break; |
a72dafaf | 6412 | } |
b4e432f1 | 6413 | case BPF_JLE: |
a72dafaf JW |
6414 | case BPF_JLT: |
6415 | { | |
3f50f132 JF |
6416 | if (is_jmp32) { |
6417 | u32 false_umin = opcode == BPF_JLT ? val32 : val32 + 1; | |
6418 | u32 true_umax = opcode == BPF_JLT ? val32 - 1 : val32; | |
6419 | ||
6420 | false_reg->u32_min_value = max(false_reg->u32_min_value, | |
6421 | false_umin); | |
6422 | true_reg->u32_max_value = min(true_reg->u32_max_value, | |
6423 | true_umax); | |
6424 | } else { | |
6425 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
6426 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
6427 | ||
6428 | false_reg->umin_value = max(false_reg->umin_value, false_umin); | |
6429 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
6430 | } | |
b4e432f1 | 6431 | break; |
a72dafaf | 6432 | } |
b4e432f1 | 6433 | case BPF_JSLE: |
a72dafaf JW |
6434 | case BPF_JSLT: |
6435 | { | |
3f50f132 JF |
6436 | if (is_jmp32) { |
6437 | s32 false_smin = opcode == BPF_JSLT ? sval32 : sval32 + 1; | |
6438 | s32 true_smax = opcode == BPF_JSLT ? sval32 - 1 : sval32; | |
a72dafaf | 6439 | |
3f50f132 JF |
6440 | false_reg->s32_min_value = max(false_reg->s32_min_value, false_smin); |
6441 | true_reg->s32_max_value = min(true_reg->s32_max_value, true_smax); | |
6442 | } else { | |
6443 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
6444 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
6445 | ||
6446 | false_reg->smin_value = max(false_reg->smin_value, false_smin); | |
6447 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
6448 | } | |
b4e432f1 | 6449 | break; |
a72dafaf | 6450 | } |
48461135 | 6451 | default: |
0fc31b10 | 6452 | return; |
48461135 JB |
6453 | } |
6454 | ||
3f50f132 JF |
6455 | if (is_jmp32) { |
6456 | false_reg->var_off = tnum_or(tnum_clear_subreg(false_64off), | |
6457 | tnum_subreg(false_32off)); | |
6458 | true_reg->var_off = tnum_or(tnum_clear_subreg(true_64off), | |
6459 | tnum_subreg(true_32off)); | |
6460 | __reg_combine_32_into_64(false_reg); | |
6461 | __reg_combine_32_into_64(true_reg); | |
6462 | } else { | |
6463 | false_reg->var_off = false_64off; | |
6464 | true_reg->var_off = true_64off; | |
6465 | __reg_combine_64_into_32(false_reg); | |
6466 | __reg_combine_64_into_32(true_reg); | |
6467 | } | |
48461135 JB |
6468 | } |
6469 | ||
f1174f77 EC |
6470 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
6471 | * the variable reg. | |
48461135 JB |
6472 | */ |
6473 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
3f50f132 JF |
6474 | struct bpf_reg_state *false_reg, |
6475 | u64 val, u32 val32, | |
092ed096 | 6476 | u8 opcode, bool is_jmp32) |
48461135 | 6477 | { |
0fc31b10 JH |
6478 | /* How can we transform "a <op> b" into "b <op> a"? */ |
6479 | static const u8 opcode_flip[16] = { | |
6480 | /* these stay the same */ | |
6481 | [BPF_JEQ >> 4] = BPF_JEQ, | |
6482 | [BPF_JNE >> 4] = BPF_JNE, | |
6483 | [BPF_JSET >> 4] = BPF_JSET, | |
6484 | /* these swap "lesser" and "greater" (L and G in the opcodes) */ | |
6485 | [BPF_JGE >> 4] = BPF_JLE, | |
6486 | [BPF_JGT >> 4] = BPF_JLT, | |
6487 | [BPF_JLE >> 4] = BPF_JGE, | |
6488 | [BPF_JLT >> 4] = BPF_JGT, | |
6489 | [BPF_JSGE >> 4] = BPF_JSLE, | |
6490 | [BPF_JSGT >> 4] = BPF_JSLT, | |
6491 | [BPF_JSLE >> 4] = BPF_JSGE, | |
6492 | [BPF_JSLT >> 4] = BPF_JSGT | |
6493 | }; | |
6494 | opcode = opcode_flip[opcode >> 4]; | |
6495 | /* This uses zero as "not present in table"; luckily the zero opcode, | |
6496 | * BPF_JA, can't get here. | |
b03c9f9f | 6497 | */ |
0fc31b10 | 6498 | if (opcode) |
3f50f132 | 6499 | reg_set_min_max(true_reg, false_reg, val, val32, opcode, is_jmp32); |
f1174f77 EC |
6500 | } |
6501 | ||
6502 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
6503 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
6504 | struct bpf_reg_state *dst_reg) | |
6505 | { | |
b03c9f9f EC |
6506 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
6507 | dst_reg->umin_value); | |
6508 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
6509 | dst_reg->umax_value); | |
6510 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
6511 | dst_reg->smin_value); | |
6512 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
6513 | dst_reg->smax_value); | |
f1174f77 EC |
6514 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
6515 | dst_reg->var_off); | |
b03c9f9f EC |
6516 | /* We might have learned new bounds from the var_off. */ |
6517 | __update_reg_bounds(src_reg); | |
6518 | __update_reg_bounds(dst_reg); | |
6519 | /* We might have learned something about the sign bit. */ | |
6520 | __reg_deduce_bounds(src_reg); | |
6521 | __reg_deduce_bounds(dst_reg); | |
6522 | /* We might have learned some bits from the bounds. */ | |
6523 | __reg_bound_offset(src_reg); | |
6524 | __reg_bound_offset(dst_reg); | |
6525 | /* Intersecting with the old var_off might have improved our bounds | |
6526 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
6527 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
6528 | */ | |
6529 | __update_reg_bounds(src_reg); | |
6530 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
6531 | } |
6532 | ||
6533 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
6534 | struct bpf_reg_state *true_dst, | |
6535 | struct bpf_reg_state *false_src, | |
6536 | struct bpf_reg_state *false_dst, | |
6537 | u8 opcode) | |
6538 | { | |
6539 | switch (opcode) { | |
6540 | case BPF_JEQ: | |
6541 | __reg_combine_min_max(true_src, true_dst); | |
6542 | break; | |
6543 | case BPF_JNE: | |
6544 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 6545 | break; |
4cabc5b1 | 6546 | } |
48461135 JB |
6547 | } |
6548 | ||
fd978bf7 JS |
6549 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
6550 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 6551 | bool is_null) |
57a09bf0 | 6552 | { |
840b9615 | 6553 | if (reg_type_may_be_null(reg->type) && reg->id == id) { |
f1174f77 EC |
6554 | /* Old offset (both fixed and variable parts) should |
6555 | * have been known-zero, because we don't allow pointer | |
6556 | * arithmetic on pointers that might be NULL. | |
6557 | */ | |
b03c9f9f EC |
6558 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
6559 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 6560 | reg->off)) { |
b03c9f9f EC |
6561 | __mark_reg_known_zero(reg); |
6562 | reg->off = 0; | |
f1174f77 EC |
6563 | } |
6564 | if (is_null) { | |
6565 | reg->type = SCALAR_VALUE; | |
840b9615 | 6566 | } else if (reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
64d85290 JS |
6567 | const struct bpf_map *map = reg->map_ptr; |
6568 | ||
6569 | if (map->inner_map_meta) { | |
840b9615 | 6570 | reg->type = CONST_PTR_TO_MAP; |
64d85290 JS |
6571 | reg->map_ptr = map->inner_map_meta; |
6572 | } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { | |
fada7fdc | 6573 | reg->type = PTR_TO_XDP_SOCK; |
64d85290 JS |
6574 | } else if (map->map_type == BPF_MAP_TYPE_SOCKMAP || |
6575 | map->map_type == BPF_MAP_TYPE_SOCKHASH) { | |
6576 | reg->type = PTR_TO_SOCKET; | |
840b9615 JS |
6577 | } else { |
6578 | reg->type = PTR_TO_MAP_VALUE; | |
6579 | } | |
c64b7983 JS |
6580 | } else if (reg->type == PTR_TO_SOCKET_OR_NULL) { |
6581 | reg->type = PTR_TO_SOCKET; | |
46f8bc92 MKL |
6582 | } else if (reg->type == PTR_TO_SOCK_COMMON_OR_NULL) { |
6583 | reg->type = PTR_TO_SOCK_COMMON; | |
655a51e5 MKL |
6584 | } else if (reg->type == PTR_TO_TCP_SOCK_OR_NULL) { |
6585 | reg->type = PTR_TO_TCP_SOCK; | |
b121b341 YS |
6586 | } else if (reg->type == PTR_TO_BTF_ID_OR_NULL) { |
6587 | reg->type = PTR_TO_BTF_ID; | |
56f668df | 6588 | } |
1b986589 MKL |
6589 | if (is_null) { |
6590 | /* We don't need id and ref_obj_id from this point | |
6591 | * onwards anymore, thus we should better reset it, | |
6592 | * so that state pruning has chances to take effect. | |
6593 | */ | |
6594 | reg->id = 0; | |
6595 | reg->ref_obj_id = 0; | |
6596 | } else if (!reg_may_point_to_spin_lock(reg)) { | |
6597 | /* For not-NULL ptr, reg->ref_obj_id will be reset | |
6598 | * in release_reg_references(). | |
6599 | * | |
6600 | * reg->id is still used by spin_lock ptr. Other | |
6601 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
6602 | */ |
6603 | reg->id = 0; | |
56f668df | 6604 | } |
57a09bf0 TG |
6605 | } |
6606 | } | |
6607 | ||
c6a9efa1 PC |
6608 | static void __mark_ptr_or_null_regs(struct bpf_func_state *state, u32 id, |
6609 | bool is_null) | |
6610 | { | |
6611 | struct bpf_reg_state *reg; | |
6612 | int i; | |
6613 | ||
6614 | for (i = 0; i < MAX_BPF_REG; i++) | |
6615 | mark_ptr_or_null_reg(state, &state->regs[i], id, is_null); | |
6616 | ||
6617 | bpf_for_each_spilled_reg(i, state, reg) { | |
6618 | if (!reg) | |
6619 | continue; | |
6620 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
6621 | } | |
6622 | } | |
6623 | ||
57a09bf0 TG |
6624 | /* The logic is similar to find_good_pkt_pointers(), both could eventually |
6625 | * be folded together at some point. | |
6626 | */ | |
840b9615 JS |
6627 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
6628 | bool is_null) | |
57a09bf0 | 6629 | { |
f4d7e40a | 6630 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
c6a9efa1 | 6631 | struct bpf_reg_state *regs = state->regs; |
1b986589 | 6632 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 6633 | u32 id = regs[regno].id; |
c6a9efa1 | 6634 | int i; |
57a09bf0 | 6635 | |
1b986589 MKL |
6636 | if (ref_obj_id && ref_obj_id == id && is_null) |
6637 | /* regs[regno] is in the " == NULL" branch. | |
6638 | * No one could have freed the reference state before | |
6639 | * doing the NULL check. | |
6640 | */ | |
6641 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 6642 | |
c6a9efa1 PC |
6643 | for (i = 0; i <= vstate->curframe; i++) |
6644 | __mark_ptr_or_null_regs(vstate->frame[i], id, is_null); | |
57a09bf0 TG |
6645 | } |
6646 | ||
5beca081 DB |
6647 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
6648 | struct bpf_reg_state *dst_reg, | |
6649 | struct bpf_reg_state *src_reg, | |
6650 | struct bpf_verifier_state *this_branch, | |
6651 | struct bpf_verifier_state *other_branch) | |
6652 | { | |
6653 | if (BPF_SRC(insn->code) != BPF_X) | |
6654 | return false; | |
6655 | ||
092ed096 JW |
6656 | /* Pointers are always 64-bit. */ |
6657 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
6658 | return false; | |
6659 | ||
5beca081 DB |
6660 | switch (BPF_OP(insn->code)) { |
6661 | case BPF_JGT: | |
6662 | if ((dst_reg->type == PTR_TO_PACKET && | |
6663 | src_reg->type == PTR_TO_PACKET_END) || | |
6664 | (dst_reg->type == PTR_TO_PACKET_META && | |
6665 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
6666 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
6667 | find_good_pkt_pointers(this_branch, dst_reg, | |
6668 | dst_reg->type, false); | |
6669 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
6670 | src_reg->type == PTR_TO_PACKET) || | |
6671 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
6672 | src_reg->type == PTR_TO_PACKET_META)) { | |
6673 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
6674 | find_good_pkt_pointers(other_branch, src_reg, | |
6675 | src_reg->type, true); | |
6676 | } else { | |
6677 | return false; | |
6678 | } | |
6679 | break; | |
6680 | case BPF_JLT: | |
6681 | if ((dst_reg->type == PTR_TO_PACKET && | |
6682 | src_reg->type == PTR_TO_PACKET_END) || | |
6683 | (dst_reg->type == PTR_TO_PACKET_META && | |
6684 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
6685 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
6686 | find_good_pkt_pointers(other_branch, dst_reg, | |
6687 | dst_reg->type, true); | |
6688 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
6689 | src_reg->type == PTR_TO_PACKET) || | |
6690 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
6691 | src_reg->type == PTR_TO_PACKET_META)) { | |
6692 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
6693 | find_good_pkt_pointers(this_branch, src_reg, | |
6694 | src_reg->type, false); | |
6695 | } else { | |
6696 | return false; | |
6697 | } | |
6698 | break; | |
6699 | case BPF_JGE: | |
6700 | if ((dst_reg->type == PTR_TO_PACKET && | |
6701 | src_reg->type == PTR_TO_PACKET_END) || | |
6702 | (dst_reg->type == PTR_TO_PACKET_META && | |
6703 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
6704 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
6705 | find_good_pkt_pointers(this_branch, dst_reg, | |
6706 | dst_reg->type, true); | |
6707 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
6708 | src_reg->type == PTR_TO_PACKET) || | |
6709 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
6710 | src_reg->type == PTR_TO_PACKET_META)) { | |
6711 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
6712 | find_good_pkt_pointers(other_branch, src_reg, | |
6713 | src_reg->type, false); | |
6714 | } else { | |
6715 | return false; | |
6716 | } | |
6717 | break; | |
6718 | case BPF_JLE: | |
6719 | if ((dst_reg->type == PTR_TO_PACKET && | |
6720 | src_reg->type == PTR_TO_PACKET_END) || | |
6721 | (dst_reg->type == PTR_TO_PACKET_META && | |
6722 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
6723 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
6724 | find_good_pkt_pointers(other_branch, dst_reg, | |
6725 | dst_reg->type, false); | |
6726 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
6727 | src_reg->type == PTR_TO_PACKET) || | |
6728 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
6729 | src_reg->type == PTR_TO_PACKET_META)) { | |
6730 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
6731 | find_good_pkt_pointers(this_branch, src_reg, | |
6732 | src_reg->type, true); | |
6733 | } else { | |
6734 | return false; | |
6735 | } | |
6736 | break; | |
6737 | default: | |
6738 | return false; | |
6739 | } | |
6740 | ||
6741 | return true; | |
6742 | } | |
6743 | ||
58e2af8b | 6744 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
6745 | struct bpf_insn *insn, int *insn_idx) |
6746 | { | |
f4d7e40a AS |
6747 | struct bpf_verifier_state *this_branch = env->cur_state; |
6748 | struct bpf_verifier_state *other_branch; | |
6749 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 6750 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
17a52670 | 6751 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 6752 | bool is_jmp32; |
fb8d251e | 6753 | int pred = -1; |
17a52670 AS |
6754 | int err; |
6755 | ||
092ed096 JW |
6756 | /* Only conditional jumps are expected to reach here. */ |
6757 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
6758 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
6759 | return -EINVAL; |
6760 | } | |
6761 | ||
6762 | if (BPF_SRC(insn->code) == BPF_X) { | |
6763 | if (insn->imm != 0) { | |
092ed096 | 6764 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
6765 | return -EINVAL; |
6766 | } | |
6767 | ||
6768 | /* check src1 operand */ | |
dc503a8a | 6769 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6770 | if (err) |
6771 | return err; | |
1be7f75d AS |
6772 | |
6773 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 6774 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
6775 | insn->src_reg); |
6776 | return -EACCES; | |
6777 | } | |
fb8d251e | 6778 | src_reg = ®s[insn->src_reg]; |
17a52670 AS |
6779 | } else { |
6780 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 6781 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
6782 | return -EINVAL; |
6783 | } | |
6784 | } | |
6785 | ||
6786 | /* check src2 operand */ | |
dc503a8a | 6787 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6788 | if (err) |
6789 | return err; | |
6790 | ||
1a0dc1ac | 6791 | dst_reg = ®s[insn->dst_reg]; |
092ed096 | 6792 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 6793 | |
3f50f132 JF |
6794 | if (BPF_SRC(insn->code) == BPF_K) { |
6795 | pred = is_branch_taken(dst_reg, insn->imm, opcode, is_jmp32); | |
6796 | } else if (src_reg->type == SCALAR_VALUE && | |
6797 | is_jmp32 && tnum_is_const(tnum_subreg(src_reg->var_off))) { | |
6798 | pred = is_branch_taken(dst_reg, | |
6799 | tnum_subreg(src_reg->var_off).value, | |
6800 | opcode, | |
6801 | is_jmp32); | |
6802 | } else if (src_reg->type == SCALAR_VALUE && | |
6803 | !is_jmp32 && tnum_is_const(src_reg->var_off)) { | |
6804 | pred = is_branch_taken(dst_reg, | |
6805 | src_reg->var_off.value, | |
6806 | opcode, | |
6807 | is_jmp32); | |
6808 | } | |
6809 | ||
b5dc0163 AS |
6810 | if (pred >= 0) { |
6811 | err = mark_chain_precision(env, insn->dst_reg); | |
6812 | if (BPF_SRC(insn->code) == BPF_X && !err) | |
6813 | err = mark_chain_precision(env, insn->src_reg); | |
6814 | if (err) | |
6815 | return err; | |
6816 | } | |
fb8d251e AS |
6817 | if (pred == 1) { |
6818 | /* only follow the goto, ignore fall-through */ | |
6819 | *insn_idx += insn->off; | |
6820 | return 0; | |
6821 | } else if (pred == 0) { | |
6822 | /* only follow fall-through branch, since | |
6823 | * that's where the program will go | |
6824 | */ | |
6825 | return 0; | |
17a52670 AS |
6826 | } |
6827 | ||
979d63d5 DB |
6828 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
6829 | false); | |
17a52670 AS |
6830 | if (!other_branch) |
6831 | return -EFAULT; | |
f4d7e40a | 6832 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 6833 | |
48461135 JB |
6834 | /* detect if we are comparing against a constant value so we can adjust |
6835 | * our min/max values for our dst register. | |
f1174f77 EC |
6836 | * this is only legit if both are scalars (or pointers to the same |
6837 | * object, I suppose, but we don't support that right now), because | |
6838 | * otherwise the different base pointers mean the offsets aren't | |
6839 | * comparable. | |
48461135 JB |
6840 | */ |
6841 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 | 6842 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
092ed096 | 6843 | |
f1174f77 | 6844 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
6845 | src_reg->type == SCALAR_VALUE) { |
6846 | if (tnum_is_const(src_reg->var_off) || | |
3f50f132 JF |
6847 | (is_jmp32 && |
6848 | tnum_is_const(tnum_subreg(src_reg->var_off)))) | |
f4d7e40a | 6849 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 6850 | dst_reg, |
3f50f132 JF |
6851 | src_reg->var_off.value, |
6852 | tnum_subreg(src_reg->var_off).value, | |
092ed096 JW |
6853 | opcode, is_jmp32); |
6854 | else if (tnum_is_const(dst_reg->var_off) || | |
3f50f132 JF |
6855 | (is_jmp32 && |
6856 | tnum_is_const(tnum_subreg(dst_reg->var_off)))) | |
f4d7e40a | 6857 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 | 6858 | src_reg, |
3f50f132 JF |
6859 | dst_reg->var_off.value, |
6860 | tnum_subreg(dst_reg->var_off).value, | |
092ed096 JW |
6861 | opcode, is_jmp32); |
6862 | else if (!is_jmp32 && | |
6863 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 6864 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
6865 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
6866 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 6867 | src_reg, dst_reg, opcode); |
f1174f77 EC |
6868 | } |
6869 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 6870 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
3f50f132 JF |
6871 | dst_reg, insn->imm, (u32)insn->imm, |
6872 | opcode, is_jmp32); | |
48461135 JB |
6873 | } |
6874 | ||
092ed096 JW |
6875 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
6876 | * NOTE: these optimizations below are related with pointer comparison | |
6877 | * which will never be JMP32. | |
6878 | */ | |
6879 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 6880 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
840b9615 JS |
6881 | reg_type_may_be_null(dst_reg->type)) { |
6882 | /* Mark all identical registers in each branch as either | |
57a09bf0 TG |
6883 | * safe or unknown depending R == 0 or R != 0 conditional. |
6884 | */ | |
840b9615 JS |
6885 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
6886 | opcode == BPF_JNE); | |
6887 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
6888 | opcode == BPF_JEQ); | |
5beca081 DB |
6889 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
6890 | this_branch, other_branch) && | |
6891 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
6892 | verbose(env, "R%d pointer comparison prohibited\n", |
6893 | insn->dst_reg); | |
1be7f75d | 6894 | return -EACCES; |
17a52670 | 6895 | } |
06ee7115 | 6896 | if (env->log.level & BPF_LOG_LEVEL) |
f4d7e40a | 6897 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
6898 | return 0; |
6899 | } | |
6900 | ||
17a52670 | 6901 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 6902 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 6903 | { |
d8eca5bb | 6904 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 6905 | struct bpf_reg_state *regs = cur_regs(env); |
d8eca5bb | 6906 | struct bpf_map *map; |
17a52670 AS |
6907 | int err; |
6908 | ||
6909 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 6910 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
6911 | return -EINVAL; |
6912 | } | |
6913 | if (insn->off != 0) { | |
61bd5218 | 6914 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
6915 | return -EINVAL; |
6916 | } | |
6917 | ||
dc503a8a | 6918 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
6919 | if (err) |
6920 | return err; | |
6921 | ||
6b173873 | 6922 | if (insn->src_reg == 0) { |
6b173873 JK |
6923 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
6924 | ||
f1174f77 | 6925 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 6926 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 6927 | return 0; |
6b173873 | 6928 | } |
17a52670 | 6929 | |
d8eca5bb DB |
6930 | map = env->used_maps[aux->map_index]; |
6931 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
6932 | regs[insn->dst_reg].map_ptr = map; | |
6933 | ||
6934 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE) { | |
6935 | regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; | |
6936 | regs[insn->dst_reg].off = aux->map_off; | |
6937 | if (map_value_has_spin_lock(map)) | |
6938 | regs[insn->dst_reg].id = ++env->id_gen; | |
6939 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD) { | |
6940 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
6941 | } else { | |
6942 | verbose(env, "bpf verifier is misconfigured\n"); | |
6943 | return -EINVAL; | |
6944 | } | |
17a52670 | 6945 | |
17a52670 AS |
6946 | return 0; |
6947 | } | |
6948 | ||
96be4325 DB |
6949 | static bool may_access_skb(enum bpf_prog_type type) |
6950 | { | |
6951 | switch (type) { | |
6952 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
6953 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 6954 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
6955 | return true; |
6956 | default: | |
6957 | return false; | |
6958 | } | |
6959 | } | |
6960 | ||
ddd872bc AS |
6961 | /* verify safety of LD_ABS|LD_IND instructions: |
6962 | * - they can only appear in the programs where ctx == skb | |
6963 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
6964 | * preserve R6-R9, and store return value into R0 | |
6965 | * | |
6966 | * Implicit input: | |
6967 | * ctx == skb == R6 == CTX | |
6968 | * | |
6969 | * Explicit input: | |
6970 | * SRC == any register | |
6971 | * IMM == 32-bit immediate | |
6972 | * | |
6973 | * Output: | |
6974 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
6975 | */ | |
58e2af8b | 6976 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 6977 | { |
638f5b90 | 6978 | struct bpf_reg_state *regs = cur_regs(env); |
6d4f151a | 6979 | static const int ctx_reg = BPF_REG_6; |
ddd872bc | 6980 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
6981 | int i, err; |
6982 | ||
24701ece | 6983 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 6984 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
6985 | return -EINVAL; |
6986 | } | |
6987 | ||
e0cea7ce DB |
6988 | if (!env->ops->gen_ld_abs) { |
6989 | verbose(env, "bpf verifier is misconfigured\n"); | |
6990 | return -EINVAL; | |
6991 | } | |
6992 | ||
f910cefa | 6993 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
6994 | /* when program has LD_ABS insn JITs and interpreter assume |
6995 | * that r1 == ctx == skb which is not the case for callees | |
6996 | * that can have arbitrary arguments. It's problematic | |
6997 | * for main prog as well since JITs would need to analyze | |
6998 | * all functions in order to make proper register save/restore | |
6999 | * decisions in the main prog. Hence disallow LD_ABS with calls | |
7000 | */ | |
7001 | verbose(env, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n"); | |
7002 | return -EINVAL; | |
7003 | } | |
7004 | ||
ddd872bc | 7005 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 7006 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 7007 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 7008 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
7009 | return -EINVAL; |
7010 | } | |
7011 | ||
7012 | /* check whether implicit source operand (register R6) is readable */ | |
6d4f151a | 7013 | err = check_reg_arg(env, ctx_reg, SRC_OP); |
ddd872bc AS |
7014 | if (err) |
7015 | return err; | |
7016 | ||
fd978bf7 JS |
7017 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
7018 | * gen_ld_abs() may terminate the program at runtime, leading to | |
7019 | * reference leak. | |
7020 | */ | |
7021 | err = check_reference_leak(env); | |
7022 | if (err) { | |
7023 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
7024 | return err; | |
7025 | } | |
7026 | ||
d83525ca AS |
7027 | if (env->cur_state->active_spin_lock) { |
7028 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); | |
7029 | return -EINVAL; | |
7030 | } | |
7031 | ||
6d4f151a | 7032 | if (regs[ctx_reg].type != PTR_TO_CTX) { |
61bd5218 JK |
7033 | verbose(env, |
7034 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
7035 | return -EINVAL; |
7036 | } | |
7037 | ||
7038 | if (mode == BPF_IND) { | |
7039 | /* check explicit source operand */ | |
dc503a8a | 7040 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
7041 | if (err) |
7042 | return err; | |
7043 | } | |
7044 | ||
6d4f151a DB |
7045 | err = check_ctx_reg(env, ®s[ctx_reg], ctx_reg); |
7046 | if (err < 0) | |
7047 | return err; | |
7048 | ||
ddd872bc | 7049 | /* reset caller saved regs to unreadable */ |
dc503a8a | 7050 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 7051 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
7052 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
7053 | } | |
ddd872bc AS |
7054 | |
7055 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
7056 | * the value fetched from the packet. |
7057 | * Already marked as written above. | |
ddd872bc | 7058 | */ |
61bd5218 | 7059 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
7060 | /* ld_abs load up to 32-bit skb data. */ |
7061 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
7062 | return 0; |
7063 | } | |
7064 | ||
390ee7e2 AS |
7065 | static int check_return_code(struct bpf_verifier_env *env) |
7066 | { | |
5cf1e914 | 7067 | struct tnum enforce_attach_type_range = tnum_unknown; |
27ae7997 | 7068 | const struct bpf_prog *prog = env->prog; |
390ee7e2 AS |
7069 | struct bpf_reg_state *reg; |
7070 | struct tnum range = tnum_range(0, 1); | |
27ae7997 MKL |
7071 | int err; |
7072 | ||
9e4e01df KS |
7073 | /* LSM and struct_ops func-ptr's return type could be "void" */ |
7074 | if ((env->prog->type == BPF_PROG_TYPE_STRUCT_OPS || | |
7075 | env->prog->type == BPF_PROG_TYPE_LSM) && | |
27ae7997 MKL |
7076 | !prog->aux->attach_func_proto->type) |
7077 | return 0; | |
7078 | ||
7079 | /* eBPF calling convetion is such that R0 is used | |
7080 | * to return the value from eBPF program. | |
7081 | * Make sure that it's readable at this time | |
7082 | * of bpf_exit, which means that program wrote | |
7083 | * something into it earlier | |
7084 | */ | |
7085 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); | |
7086 | if (err) | |
7087 | return err; | |
7088 | ||
7089 | if (is_pointer_value(env, BPF_REG_0)) { | |
7090 | verbose(env, "R0 leaks addr as return value\n"); | |
7091 | return -EACCES; | |
7092 | } | |
390ee7e2 AS |
7093 | |
7094 | switch (env->prog->type) { | |
983695fa DB |
7095 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
7096 | if (env->prog->expected_attach_type == BPF_CGROUP_UDP4_RECVMSG || | |
1b66d253 DB |
7097 | env->prog->expected_attach_type == BPF_CGROUP_UDP6_RECVMSG || |
7098 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETPEERNAME || | |
7099 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETPEERNAME || | |
7100 | env->prog->expected_attach_type == BPF_CGROUP_INET4_GETSOCKNAME || | |
7101 | env->prog->expected_attach_type == BPF_CGROUP_INET6_GETSOCKNAME) | |
983695fa | 7102 | range = tnum_range(1, 1); |
ed4ed404 | 7103 | break; |
390ee7e2 | 7104 | case BPF_PROG_TYPE_CGROUP_SKB: |
5cf1e914 | 7105 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
7106 | range = tnum_range(0, 3); | |
7107 | enforce_attach_type_range = tnum_range(2, 3); | |
7108 | } | |
ed4ed404 | 7109 | break; |
390ee7e2 AS |
7110 | case BPF_PROG_TYPE_CGROUP_SOCK: |
7111 | case BPF_PROG_TYPE_SOCK_OPS: | |
ebc614f6 | 7112 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 7113 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
0d01da6a | 7114 | case BPF_PROG_TYPE_CGROUP_SOCKOPT: |
390ee7e2 | 7115 | break; |
15ab09bd AS |
7116 | case BPF_PROG_TYPE_RAW_TRACEPOINT: |
7117 | if (!env->prog->aux->attach_btf_id) | |
7118 | return 0; | |
7119 | range = tnum_const(0); | |
7120 | break; | |
15d83c4d | 7121 | case BPF_PROG_TYPE_TRACING: |
e92888c7 YS |
7122 | switch (env->prog->expected_attach_type) { |
7123 | case BPF_TRACE_FENTRY: | |
7124 | case BPF_TRACE_FEXIT: | |
7125 | range = tnum_const(0); | |
7126 | break; | |
7127 | case BPF_TRACE_RAW_TP: | |
7128 | case BPF_MODIFY_RETURN: | |
15d83c4d | 7129 | return 0; |
2ec0616e DB |
7130 | case BPF_TRACE_ITER: |
7131 | break; | |
e92888c7 YS |
7132 | default: |
7133 | return -ENOTSUPP; | |
7134 | } | |
15d83c4d | 7135 | break; |
e92888c7 YS |
7136 | case BPF_PROG_TYPE_EXT: |
7137 | /* freplace program can return anything as its return value | |
7138 | * depends on the to-be-replaced kernel func or bpf program. | |
7139 | */ | |
390ee7e2 AS |
7140 | default: |
7141 | return 0; | |
7142 | } | |
7143 | ||
638f5b90 | 7144 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 7145 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 7146 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
7147 | reg_type_str[reg->type]); |
7148 | return -EINVAL; | |
7149 | } | |
7150 | ||
7151 | if (!tnum_in(range, reg->var_off)) { | |
5cf1e914 | 7152 | char tn_buf[48]; |
7153 | ||
61bd5218 | 7154 | verbose(env, "At program exit the register R0 "); |
390ee7e2 | 7155 | if (!tnum_is_unknown(reg->var_off)) { |
390ee7e2 | 7156 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 7157 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 7158 | } else { |
61bd5218 | 7159 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 7160 | } |
5cf1e914 | 7161 | tnum_strn(tn_buf, sizeof(tn_buf), range); |
983695fa | 7162 | verbose(env, " should have been in %s\n", tn_buf); |
390ee7e2 AS |
7163 | return -EINVAL; |
7164 | } | |
5cf1e914 | 7165 | |
7166 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
7167 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
7168 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
7169 | return 0; |
7170 | } | |
7171 | ||
475fb78f AS |
7172 | /* non-recursive DFS pseudo code |
7173 | * 1 procedure DFS-iterative(G,v): | |
7174 | * 2 label v as discovered | |
7175 | * 3 let S be a stack | |
7176 | * 4 S.push(v) | |
7177 | * 5 while S is not empty | |
7178 | * 6 t <- S.pop() | |
7179 | * 7 if t is what we're looking for: | |
7180 | * 8 return t | |
7181 | * 9 for all edges e in G.adjacentEdges(t) do | |
7182 | * 10 if edge e is already labelled | |
7183 | * 11 continue with the next edge | |
7184 | * 12 w <- G.adjacentVertex(t,e) | |
7185 | * 13 if vertex w is not discovered and not explored | |
7186 | * 14 label e as tree-edge | |
7187 | * 15 label w as discovered | |
7188 | * 16 S.push(w) | |
7189 | * 17 continue at 5 | |
7190 | * 18 else if vertex w is discovered | |
7191 | * 19 label e as back-edge | |
7192 | * 20 else | |
7193 | * 21 // vertex w is explored | |
7194 | * 22 label e as forward- or cross-edge | |
7195 | * 23 label t as explored | |
7196 | * 24 S.pop() | |
7197 | * | |
7198 | * convention: | |
7199 | * 0x10 - discovered | |
7200 | * 0x11 - discovered and fall-through edge labelled | |
7201 | * 0x12 - discovered and fall-through and branch edges labelled | |
7202 | * 0x20 - explored | |
7203 | */ | |
7204 | ||
7205 | enum { | |
7206 | DISCOVERED = 0x10, | |
7207 | EXPLORED = 0x20, | |
7208 | FALLTHROUGH = 1, | |
7209 | BRANCH = 2, | |
7210 | }; | |
7211 | ||
dc2a4ebc AS |
7212 | static u32 state_htab_size(struct bpf_verifier_env *env) |
7213 | { | |
7214 | return env->prog->len; | |
7215 | } | |
7216 | ||
5d839021 AS |
7217 | static struct bpf_verifier_state_list **explored_state( |
7218 | struct bpf_verifier_env *env, | |
7219 | int idx) | |
7220 | { | |
dc2a4ebc AS |
7221 | struct bpf_verifier_state *cur = env->cur_state; |
7222 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
7223 | ||
7224 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
7225 | } |
7226 | ||
7227 | static void init_explored_state(struct bpf_verifier_env *env, int idx) | |
7228 | { | |
a8f500af | 7229 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 7230 | } |
f1bca824 | 7231 | |
475fb78f AS |
7232 | /* t, w, e - match pseudo-code above: |
7233 | * t - index of current instruction | |
7234 | * w - next instruction | |
7235 | * e - edge | |
7236 | */ | |
2589726d AS |
7237 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env, |
7238 | bool loop_ok) | |
475fb78f | 7239 | { |
7df737e9 AS |
7240 | int *insn_stack = env->cfg.insn_stack; |
7241 | int *insn_state = env->cfg.insn_state; | |
7242 | ||
475fb78f AS |
7243 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
7244 | return 0; | |
7245 | ||
7246 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
7247 | return 0; | |
7248 | ||
7249 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 7250 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 7251 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
7252 | return -EINVAL; |
7253 | } | |
7254 | ||
f1bca824 AS |
7255 | if (e == BRANCH) |
7256 | /* mark branch target for state pruning */ | |
5d839021 | 7257 | init_explored_state(env, w); |
f1bca824 | 7258 | |
475fb78f AS |
7259 | if (insn_state[w] == 0) { |
7260 | /* tree-edge */ | |
7261 | insn_state[t] = DISCOVERED | e; | |
7262 | insn_state[w] = DISCOVERED; | |
7df737e9 | 7263 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 7264 | return -E2BIG; |
7df737e9 | 7265 | insn_stack[env->cfg.cur_stack++] = w; |
475fb78f AS |
7266 | return 1; |
7267 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
2c78ee89 | 7268 | if (loop_ok && env->bpf_capable) |
2589726d | 7269 | return 0; |
d9762e84 MKL |
7270 | verbose_linfo(env, t, "%d: ", t); |
7271 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 7272 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
7273 | return -EINVAL; |
7274 | } else if (insn_state[w] == EXPLORED) { | |
7275 | /* forward- or cross-edge */ | |
7276 | insn_state[t] = DISCOVERED | e; | |
7277 | } else { | |
61bd5218 | 7278 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
7279 | return -EFAULT; |
7280 | } | |
7281 | return 0; | |
7282 | } | |
7283 | ||
7284 | /* non-recursive depth-first-search to detect loops in BPF program | |
7285 | * loop == back-edge in directed graph | |
7286 | */ | |
58e2af8b | 7287 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
7288 | { |
7289 | struct bpf_insn *insns = env->prog->insnsi; | |
7290 | int insn_cnt = env->prog->len; | |
7df737e9 | 7291 | int *insn_stack, *insn_state; |
475fb78f AS |
7292 | int ret = 0; |
7293 | int i, t; | |
7294 | ||
7df737e9 | 7295 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
7296 | if (!insn_state) |
7297 | return -ENOMEM; | |
7298 | ||
7df737e9 | 7299 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 7300 | if (!insn_stack) { |
71dde681 | 7301 | kvfree(insn_state); |
475fb78f AS |
7302 | return -ENOMEM; |
7303 | } | |
7304 | ||
7305 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
7306 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 7307 | env->cfg.cur_stack = 1; |
475fb78f AS |
7308 | |
7309 | peek_stack: | |
7df737e9 | 7310 | if (env->cfg.cur_stack == 0) |
475fb78f | 7311 | goto check_state; |
7df737e9 | 7312 | t = insn_stack[env->cfg.cur_stack - 1]; |
475fb78f | 7313 | |
092ed096 JW |
7314 | if (BPF_CLASS(insns[t].code) == BPF_JMP || |
7315 | BPF_CLASS(insns[t].code) == BPF_JMP32) { | |
475fb78f AS |
7316 | u8 opcode = BPF_OP(insns[t].code); |
7317 | ||
7318 | if (opcode == BPF_EXIT) { | |
7319 | goto mark_explored; | |
7320 | } else if (opcode == BPF_CALL) { | |
2589726d | 7321 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); |
475fb78f AS |
7322 | if (ret == 1) |
7323 | goto peek_stack; | |
7324 | else if (ret < 0) | |
7325 | goto err_free; | |
07016151 | 7326 | if (t + 1 < insn_cnt) |
5d839021 | 7327 | init_explored_state(env, t + 1); |
cc8b0b92 | 7328 | if (insns[t].src_reg == BPF_PSEUDO_CALL) { |
5d839021 | 7329 | init_explored_state(env, t); |
2589726d AS |
7330 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, |
7331 | env, false); | |
cc8b0b92 AS |
7332 | if (ret == 1) |
7333 | goto peek_stack; | |
7334 | else if (ret < 0) | |
7335 | goto err_free; | |
7336 | } | |
475fb78f AS |
7337 | } else if (opcode == BPF_JA) { |
7338 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
7339 | ret = -EINVAL; | |
7340 | goto err_free; | |
7341 | } | |
7342 | /* unconditional jump with single edge */ | |
7343 | ret = push_insn(t, t + insns[t].off + 1, | |
2589726d | 7344 | FALLTHROUGH, env, true); |
475fb78f AS |
7345 | if (ret == 1) |
7346 | goto peek_stack; | |
7347 | else if (ret < 0) | |
7348 | goto err_free; | |
b5dc0163 AS |
7349 | /* unconditional jmp is not a good pruning point, |
7350 | * but it's marked, since backtracking needs | |
7351 | * to record jmp history in is_state_visited(). | |
7352 | */ | |
7353 | init_explored_state(env, t + insns[t].off + 1); | |
f1bca824 AS |
7354 | /* tell verifier to check for equivalent states |
7355 | * after every call and jump | |
7356 | */ | |
c3de6317 | 7357 | if (t + 1 < insn_cnt) |
5d839021 | 7358 | init_explored_state(env, t + 1); |
475fb78f AS |
7359 | } else { |
7360 | /* conditional jump with two edges */ | |
5d839021 | 7361 | init_explored_state(env, t); |
2589726d | 7362 | ret = push_insn(t, t + 1, FALLTHROUGH, env, true); |
475fb78f AS |
7363 | if (ret == 1) |
7364 | goto peek_stack; | |
7365 | else if (ret < 0) | |
7366 | goto err_free; | |
7367 | ||
2589726d | 7368 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env, true); |
475fb78f AS |
7369 | if (ret == 1) |
7370 | goto peek_stack; | |
7371 | else if (ret < 0) | |
7372 | goto err_free; | |
7373 | } | |
7374 | } else { | |
7375 | /* all other non-branch instructions with single | |
7376 | * fall-through edge | |
7377 | */ | |
2589726d | 7378 | ret = push_insn(t, t + 1, FALLTHROUGH, env, false); |
475fb78f AS |
7379 | if (ret == 1) |
7380 | goto peek_stack; | |
7381 | else if (ret < 0) | |
7382 | goto err_free; | |
7383 | } | |
7384 | ||
7385 | mark_explored: | |
7386 | insn_state[t] = EXPLORED; | |
7df737e9 | 7387 | if (env->cfg.cur_stack-- <= 0) { |
61bd5218 | 7388 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
7389 | ret = -EFAULT; |
7390 | goto err_free; | |
7391 | } | |
7392 | goto peek_stack; | |
7393 | ||
7394 | check_state: | |
7395 | for (i = 0; i < insn_cnt; i++) { | |
7396 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 7397 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
7398 | ret = -EINVAL; |
7399 | goto err_free; | |
7400 | } | |
7401 | } | |
7402 | ret = 0; /* cfg looks good */ | |
7403 | ||
7404 | err_free: | |
71dde681 AS |
7405 | kvfree(insn_state); |
7406 | kvfree(insn_stack); | |
7df737e9 | 7407 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
7408 | return ret; |
7409 | } | |
7410 | ||
838e9690 YS |
7411 | /* The minimum supported BTF func info size */ |
7412 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
7413 | #define MAX_FUNCINFO_REC_SIZE 252 | |
7414 | ||
c454a46b MKL |
7415 | static int check_btf_func(struct bpf_verifier_env *env, |
7416 | const union bpf_attr *attr, | |
7417 | union bpf_attr __user *uattr) | |
838e9690 | 7418 | { |
d0b2818e | 7419 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 7420 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 7421 | struct bpf_func_info *krecord; |
8c1b6e69 | 7422 | struct bpf_func_info_aux *info_aux = NULL; |
838e9690 | 7423 | const struct btf_type *type; |
c454a46b MKL |
7424 | struct bpf_prog *prog; |
7425 | const struct btf *btf; | |
838e9690 | 7426 | void __user *urecord; |
d0b2818e | 7427 | u32 prev_offset = 0; |
838e9690 YS |
7428 | int ret = 0; |
7429 | ||
7430 | nfuncs = attr->func_info_cnt; | |
7431 | if (!nfuncs) | |
7432 | return 0; | |
7433 | ||
7434 | if (nfuncs != env->subprog_cnt) { | |
7435 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
7436 | return -EINVAL; | |
7437 | } | |
7438 | ||
7439 | urec_size = attr->func_info_rec_size; | |
7440 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
7441 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
7442 | urec_size % sizeof(u32)) { | |
7443 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
7444 | return -EINVAL; | |
7445 | } | |
7446 | ||
c454a46b MKL |
7447 | prog = env->prog; |
7448 | btf = prog->aux->btf; | |
838e9690 YS |
7449 | |
7450 | urecord = u64_to_user_ptr(attr->func_info); | |
7451 | min_size = min_t(u32, krec_size, urec_size); | |
7452 | ||
ba64e7d8 | 7453 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
7454 | if (!krecord) |
7455 | return -ENOMEM; | |
8c1b6e69 AS |
7456 | info_aux = kcalloc(nfuncs, sizeof(*info_aux), GFP_KERNEL | __GFP_NOWARN); |
7457 | if (!info_aux) | |
7458 | goto err_free; | |
ba64e7d8 | 7459 | |
838e9690 YS |
7460 | for (i = 0; i < nfuncs; i++) { |
7461 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
7462 | if (ret) { | |
7463 | if (ret == -E2BIG) { | |
7464 | verbose(env, "nonzero tailing record in func info"); | |
7465 | /* set the size kernel expects so loader can zero | |
7466 | * out the rest of the record. | |
7467 | */ | |
7468 | if (put_user(min_size, &uattr->func_info_rec_size)) | |
7469 | ret = -EFAULT; | |
7470 | } | |
c454a46b | 7471 | goto err_free; |
838e9690 YS |
7472 | } |
7473 | ||
ba64e7d8 | 7474 | if (copy_from_user(&krecord[i], urecord, min_size)) { |
838e9690 | 7475 | ret = -EFAULT; |
c454a46b | 7476 | goto err_free; |
838e9690 YS |
7477 | } |
7478 | ||
d30d42e0 | 7479 | /* check insn_off */ |
838e9690 | 7480 | if (i == 0) { |
d30d42e0 | 7481 | if (krecord[i].insn_off) { |
838e9690 | 7482 | verbose(env, |
d30d42e0 MKL |
7483 | "nonzero insn_off %u for the first func info record", |
7484 | krecord[i].insn_off); | |
838e9690 | 7485 | ret = -EINVAL; |
c454a46b | 7486 | goto err_free; |
838e9690 | 7487 | } |
d30d42e0 | 7488 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
7489 | verbose(env, |
7490 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 7491 | krecord[i].insn_off, prev_offset); |
838e9690 | 7492 | ret = -EINVAL; |
c454a46b | 7493 | goto err_free; |
838e9690 YS |
7494 | } |
7495 | ||
d30d42e0 | 7496 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 YS |
7497 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
7498 | ret = -EINVAL; | |
c454a46b | 7499 | goto err_free; |
838e9690 YS |
7500 | } |
7501 | ||
7502 | /* check type_id */ | |
ba64e7d8 | 7503 | type = btf_type_by_id(btf, krecord[i].type_id); |
51c39bb1 | 7504 | if (!type || !btf_type_is_func(type)) { |
838e9690 | 7505 | verbose(env, "invalid type id %d in func info", |
ba64e7d8 | 7506 | krecord[i].type_id); |
838e9690 | 7507 | ret = -EINVAL; |
c454a46b | 7508 | goto err_free; |
838e9690 | 7509 | } |
51c39bb1 | 7510 | info_aux[i].linkage = BTF_INFO_VLEN(type->info); |
d30d42e0 | 7511 | prev_offset = krecord[i].insn_off; |
838e9690 YS |
7512 | urecord += urec_size; |
7513 | } | |
7514 | ||
ba64e7d8 YS |
7515 | prog->aux->func_info = krecord; |
7516 | prog->aux->func_info_cnt = nfuncs; | |
8c1b6e69 | 7517 | prog->aux->func_info_aux = info_aux; |
838e9690 YS |
7518 | return 0; |
7519 | ||
c454a46b | 7520 | err_free: |
ba64e7d8 | 7521 | kvfree(krecord); |
8c1b6e69 | 7522 | kfree(info_aux); |
838e9690 YS |
7523 | return ret; |
7524 | } | |
7525 | ||
ba64e7d8 YS |
7526 | static void adjust_btf_func(struct bpf_verifier_env *env) |
7527 | { | |
8c1b6e69 | 7528 | struct bpf_prog_aux *aux = env->prog->aux; |
ba64e7d8 YS |
7529 | int i; |
7530 | ||
8c1b6e69 | 7531 | if (!aux->func_info) |
ba64e7d8 YS |
7532 | return; |
7533 | ||
7534 | for (i = 0; i < env->subprog_cnt; i++) | |
8c1b6e69 | 7535 | aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
7536 | } |
7537 | ||
c454a46b MKL |
7538 | #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \ |
7539 | sizeof(((struct bpf_line_info *)(0))->line_col)) | |
7540 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE | |
7541 | ||
7542 | static int check_btf_line(struct bpf_verifier_env *env, | |
7543 | const union bpf_attr *attr, | |
7544 | union bpf_attr __user *uattr) | |
7545 | { | |
7546 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
7547 | struct bpf_subprog_info *sub; | |
7548 | struct bpf_line_info *linfo; | |
7549 | struct bpf_prog *prog; | |
7550 | const struct btf *btf; | |
7551 | void __user *ulinfo; | |
7552 | int err; | |
7553 | ||
7554 | nr_linfo = attr->line_info_cnt; | |
7555 | if (!nr_linfo) | |
7556 | return 0; | |
7557 | ||
7558 | rec_size = attr->line_info_rec_size; | |
7559 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
7560 | rec_size > MAX_LINEINFO_REC_SIZE || | |
7561 | rec_size & (sizeof(u32) - 1)) | |
7562 | return -EINVAL; | |
7563 | ||
7564 | /* Need to zero it in case the userspace may | |
7565 | * pass in a smaller bpf_line_info object. | |
7566 | */ | |
7567 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
7568 | GFP_KERNEL | __GFP_NOWARN); | |
7569 | if (!linfo) | |
7570 | return -ENOMEM; | |
7571 | ||
7572 | prog = env->prog; | |
7573 | btf = prog->aux->btf; | |
7574 | ||
7575 | s = 0; | |
7576 | sub = env->subprog_info; | |
7577 | ulinfo = u64_to_user_ptr(attr->line_info); | |
7578 | expected_size = sizeof(struct bpf_line_info); | |
7579 | ncopy = min_t(u32, expected_size, rec_size); | |
7580 | for (i = 0; i < nr_linfo; i++) { | |
7581 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
7582 | if (err) { | |
7583 | if (err == -E2BIG) { | |
7584 | verbose(env, "nonzero tailing record in line_info"); | |
7585 | if (put_user(expected_size, | |
7586 | &uattr->line_info_rec_size)) | |
7587 | err = -EFAULT; | |
7588 | } | |
7589 | goto err_free; | |
7590 | } | |
7591 | ||
7592 | if (copy_from_user(&linfo[i], ulinfo, ncopy)) { | |
7593 | err = -EFAULT; | |
7594 | goto err_free; | |
7595 | } | |
7596 | ||
7597 | /* | |
7598 | * Check insn_off to ensure | |
7599 | * 1) strictly increasing AND | |
7600 | * 2) bounded by prog->len | |
7601 | * | |
7602 | * The linfo[0].insn_off == 0 check logically falls into | |
7603 | * the later "missing bpf_line_info for func..." case | |
7604 | * because the first linfo[0].insn_off must be the | |
7605 | * first sub also and the first sub must have | |
7606 | * subprog_info[0].start == 0. | |
7607 | */ | |
7608 | if ((i && linfo[i].insn_off <= prev_offset) || | |
7609 | linfo[i].insn_off >= prog->len) { | |
7610 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
7611 | i, linfo[i].insn_off, prev_offset, | |
7612 | prog->len); | |
7613 | err = -EINVAL; | |
7614 | goto err_free; | |
7615 | } | |
7616 | ||
fdbaa0be MKL |
7617 | if (!prog->insnsi[linfo[i].insn_off].code) { |
7618 | verbose(env, | |
7619 | "Invalid insn code at line_info[%u].insn_off\n", | |
7620 | i); | |
7621 | err = -EINVAL; | |
7622 | goto err_free; | |
7623 | } | |
7624 | ||
23127b33 MKL |
7625 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
7626 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
7627 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
7628 | err = -EINVAL; | |
7629 | goto err_free; | |
7630 | } | |
7631 | ||
7632 | if (s != env->subprog_cnt) { | |
7633 | if (linfo[i].insn_off == sub[s].start) { | |
7634 | sub[s].linfo_idx = i; | |
7635 | s++; | |
7636 | } else if (sub[s].start < linfo[i].insn_off) { | |
7637 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
7638 | err = -EINVAL; | |
7639 | goto err_free; | |
7640 | } | |
7641 | } | |
7642 | ||
7643 | prev_offset = linfo[i].insn_off; | |
7644 | ulinfo += rec_size; | |
7645 | } | |
7646 | ||
7647 | if (s != env->subprog_cnt) { | |
7648 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
7649 | env->subprog_cnt - s, s); | |
7650 | err = -EINVAL; | |
7651 | goto err_free; | |
7652 | } | |
7653 | ||
7654 | prog->aux->linfo = linfo; | |
7655 | prog->aux->nr_linfo = nr_linfo; | |
7656 | ||
7657 | return 0; | |
7658 | ||
7659 | err_free: | |
7660 | kvfree(linfo); | |
7661 | return err; | |
7662 | } | |
7663 | ||
7664 | static int check_btf_info(struct bpf_verifier_env *env, | |
7665 | const union bpf_attr *attr, | |
7666 | union bpf_attr __user *uattr) | |
7667 | { | |
7668 | struct btf *btf; | |
7669 | int err; | |
7670 | ||
7671 | if (!attr->func_info_cnt && !attr->line_info_cnt) | |
7672 | return 0; | |
7673 | ||
7674 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
7675 | if (IS_ERR(btf)) | |
7676 | return PTR_ERR(btf); | |
7677 | env->prog->aux->btf = btf; | |
7678 | ||
7679 | err = check_btf_func(env, attr, uattr); | |
7680 | if (err) | |
7681 | return err; | |
7682 | ||
7683 | err = check_btf_line(env, attr, uattr); | |
7684 | if (err) | |
7685 | return err; | |
7686 | ||
7687 | return 0; | |
ba64e7d8 YS |
7688 | } |
7689 | ||
f1174f77 EC |
7690 | /* check %cur's range satisfies %old's */ |
7691 | static bool range_within(struct bpf_reg_state *old, | |
7692 | struct bpf_reg_state *cur) | |
7693 | { | |
b03c9f9f EC |
7694 | return old->umin_value <= cur->umin_value && |
7695 | old->umax_value >= cur->umax_value && | |
7696 | old->smin_value <= cur->smin_value && | |
7697 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
7698 | } |
7699 | ||
7700 | /* Maximum number of register states that can exist at once */ | |
7701 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
7702 | struct idpair { | |
7703 | u32 old; | |
7704 | u32 cur; | |
7705 | }; | |
7706 | ||
7707 | /* If in the old state two registers had the same id, then they need to have | |
7708 | * the same id in the new state as well. But that id could be different from | |
7709 | * the old state, so we need to track the mapping from old to new ids. | |
7710 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
7711 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
7712 | * regs with a different old id could still have new id 9, we don't care about | |
7713 | * that. | |
7714 | * So we look through our idmap to see if this old id has been seen before. If | |
7715 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 7716 | */ |
f1174f77 | 7717 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 7718 | { |
f1174f77 | 7719 | unsigned int i; |
969bf05e | 7720 | |
f1174f77 EC |
7721 | for (i = 0; i < ID_MAP_SIZE; i++) { |
7722 | if (!idmap[i].old) { | |
7723 | /* Reached an empty slot; haven't seen this id before */ | |
7724 | idmap[i].old = old_id; | |
7725 | idmap[i].cur = cur_id; | |
7726 | return true; | |
7727 | } | |
7728 | if (idmap[i].old == old_id) | |
7729 | return idmap[i].cur == cur_id; | |
7730 | } | |
7731 | /* We ran out of idmap slots, which should be impossible */ | |
7732 | WARN_ON_ONCE(1); | |
7733 | return false; | |
7734 | } | |
7735 | ||
9242b5f5 AS |
7736 | static void clean_func_state(struct bpf_verifier_env *env, |
7737 | struct bpf_func_state *st) | |
7738 | { | |
7739 | enum bpf_reg_liveness live; | |
7740 | int i, j; | |
7741 | ||
7742 | for (i = 0; i < BPF_REG_FP; i++) { | |
7743 | live = st->regs[i].live; | |
7744 | /* liveness must not touch this register anymore */ | |
7745 | st->regs[i].live |= REG_LIVE_DONE; | |
7746 | if (!(live & REG_LIVE_READ)) | |
7747 | /* since the register is unused, clear its state | |
7748 | * to make further comparison simpler | |
7749 | */ | |
f54c7898 | 7750 | __mark_reg_not_init(env, &st->regs[i]); |
9242b5f5 AS |
7751 | } |
7752 | ||
7753 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
7754 | live = st->stack[i].spilled_ptr.live; | |
7755 | /* liveness must not touch this stack slot anymore */ | |
7756 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
7757 | if (!(live & REG_LIVE_READ)) { | |
f54c7898 | 7758 | __mark_reg_not_init(env, &st->stack[i].spilled_ptr); |
9242b5f5 AS |
7759 | for (j = 0; j < BPF_REG_SIZE; j++) |
7760 | st->stack[i].slot_type[j] = STACK_INVALID; | |
7761 | } | |
7762 | } | |
7763 | } | |
7764 | ||
7765 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
7766 | struct bpf_verifier_state *st) | |
7767 | { | |
7768 | int i; | |
7769 | ||
7770 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
7771 | /* all regs in this state in all frames were already marked */ | |
7772 | return; | |
7773 | ||
7774 | for (i = 0; i <= st->curframe; i++) | |
7775 | clean_func_state(env, st->frame[i]); | |
7776 | } | |
7777 | ||
7778 | /* the parentage chains form a tree. | |
7779 | * the verifier states are added to state lists at given insn and | |
7780 | * pushed into state stack for future exploration. | |
7781 | * when the verifier reaches bpf_exit insn some of the verifer states | |
7782 | * stored in the state lists have their final liveness state already, | |
7783 | * but a lot of states will get revised from liveness point of view when | |
7784 | * the verifier explores other branches. | |
7785 | * Example: | |
7786 | * 1: r0 = 1 | |
7787 | * 2: if r1 == 100 goto pc+1 | |
7788 | * 3: r0 = 2 | |
7789 | * 4: exit | |
7790 | * when the verifier reaches exit insn the register r0 in the state list of | |
7791 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
7792 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
7793 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
7794 | * | |
7795 | * Since the verifier pushes the branch states as it sees them while exploring | |
7796 | * the program the condition of walking the branch instruction for the second | |
7797 | * time means that all states below this branch were already explored and | |
7798 | * their final liveness markes are already propagated. | |
7799 | * Hence when the verifier completes the search of state list in is_state_visited() | |
7800 | * we can call this clean_live_states() function to mark all liveness states | |
7801 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
7802 | * will not be used. | |
7803 | * This function also clears the registers and stack for states that !READ | |
7804 | * to simplify state merging. | |
7805 | * | |
7806 | * Important note here that walking the same branch instruction in the callee | |
7807 | * doesn't meant that the states are DONE. The verifier has to compare | |
7808 | * the callsites | |
7809 | */ | |
7810 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
7811 | struct bpf_verifier_state *cur) | |
7812 | { | |
7813 | struct bpf_verifier_state_list *sl; | |
7814 | int i; | |
7815 | ||
5d839021 | 7816 | sl = *explored_state(env, insn); |
a8f500af | 7817 | while (sl) { |
2589726d AS |
7818 | if (sl->state.branches) |
7819 | goto next; | |
dc2a4ebc AS |
7820 | if (sl->state.insn_idx != insn || |
7821 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
7822 | goto next; |
7823 | for (i = 0; i <= cur->curframe; i++) | |
7824 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
7825 | goto next; | |
7826 | clean_verifier_state(env, &sl->state); | |
7827 | next: | |
7828 | sl = sl->next; | |
7829 | } | |
7830 | } | |
7831 | ||
f1174f77 | 7832 | /* Returns true if (rold safe implies rcur safe) */ |
1b688a19 EC |
7833 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
7834 | struct idpair *idmap) | |
f1174f77 | 7835 | { |
f4d7e40a AS |
7836 | bool equal; |
7837 | ||
dc503a8a EC |
7838 | if (!(rold->live & REG_LIVE_READ)) |
7839 | /* explored state didn't use this */ | |
7840 | return true; | |
7841 | ||
679c782d | 7842 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
7843 | |
7844 | if (rold->type == PTR_TO_STACK) | |
7845 | /* two stack pointers are equal only if they're pointing to | |
7846 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
7847 | */ | |
7848 | return equal && rold->frameno == rcur->frameno; | |
7849 | ||
7850 | if (equal) | |
969bf05e AS |
7851 | return true; |
7852 | ||
f1174f77 EC |
7853 | if (rold->type == NOT_INIT) |
7854 | /* explored state can't have used this */ | |
969bf05e | 7855 | return true; |
f1174f77 EC |
7856 | if (rcur->type == NOT_INIT) |
7857 | return false; | |
7858 | switch (rold->type) { | |
7859 | case SCALAR_VALUE: | |
7860 | if (rcur->type == SCALAR_VALUE) { | |
b5dc0163 AS |
7861 | if (!rold->precise && !rcur->precise) |
7862 | return true; | |
f1174f77 EC |
7863 | /* new val must satisfy old val knowledge */ |
7864 | return range_within(rold, rcur) && | |
7865 | tnum_in(rold->var_off, rcur->var_off); | |
7866 | } else { | |
179d1c56 JH |
7867 | /* We're trying to use a pointer in place of a scalar. |
7868 | * Even if the scalar was unbounded, this could lead to | |
7869 | * pointer leaks because scalars are allowed to leak | |
7870 | * while pointers are not. We could make this safe in | |
7871 | * special cases if root is calling us, but it's | |
7872 | * probably not worth the hassle. | |
f1174f77 | 7873 | */ |
179d1c56 | 7874 | return false; |
f1174f77 EC |
7875 | } |
7876 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
7877 | /* If the new min/max/var_off satisfy the old ones and |
7878 | * everything else matches, we are OK. | |
d83525ca AS |
7879 | * 'id' is not compared, since it's only used for maps with |
7880 | * bpf_spin_lock inside map element and in such cases if | |
7881 | * the rest of the prog is valid for one map element then | |
7882 | * it's valid for all map elements regardless of the key | |
7883 | * used in bpf_map_lookup() | |
1b688a19 EC |
7884 | */ |
7885 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
7886 | range_within(rold, rcur) && | |
7887 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
7888 | case PTR_TO_MAP_VALUE_OR_NULL: |
7889 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
7890 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
7891 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
7892 | * checked, doing so could have affected others with the same | |
7893 | * id, and we can't check for that because we lost the id when | |
7894 | * we converted to a PTR_TO_MAP_VALUE. | |
7895 | */ | |
7896 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
7897 | return false; | |
7898 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
7899 | return false; | |
7900 | /* Check our ids match any regs they're supposed to */ | |
7901 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 7902 | case PTR_TO_PACKET_META: |
f1174f77 | 7903 | case PTR_TO_PACKET: |
de8f3a83 | 7904 | if (rcur->type != rold->type) |
f1174f77 EC |
7905 | return false; |
7906 | /* We must have at least as much range as the old ptr | |
7907 | * did, so that any accesses which were safe before are | |
7908 | * still safe. This is true even if old range < old off, | |
7909 | * since someone could have accessed through (ptr - k), or | |
7910 | * even done ptr -= k in a register, to get a safe access. | |
7911 | */ | |
7912 | if (rold->range > rcur->range) | |
7913 | return false; | |
7914 | /* If the offsets don't match, we can't trust our alignment; | |
7915 | * nor can we be sure that we won't fall out of range. | |
7916 | */ | |
7917 | if (rold->off != rcur->off) | |
7918 | return false; | |
7919 | /* id relations must be preserved */ | |
7920 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
7921 | return false; | |
7922 | /* new val must satisfy old val knowledge */ | |
7923 | return range_within(rold, rcur) && | |
7924 | tnum_in(rold->var_off, rcur->var_off); | |
7925 | case PTR_TO_CTX: | |
7926 | case CONST_PTR_TO_MAP: | |
f1174f77 | 7927 | case PTR_TO_PACKET_END: |
d58e468b | 7928 | case PTR_TO_FLOW_KEYS: |
c64b7983 JS |
7929 | case PTR_TO_SOCKET: |
7930 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
7931 | case PTR_TO_SOCK_COMMON: |
7932 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
7933 | case PTR_TO_TCP_SOCK: |
7934 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 7935 | case PTR_TO_XDP_SOCK: |
f1174f77 EC |
7936 | /* Only valid matches are exact, which memcmp() above |
7937 | * would have accepted | |
7938 | */ | |
7939 | default: | |
7940 | /* Don't know what's going on, just say it's not safe */ | |
7941 | return false; | |
7942 | } | |
969bf05e | 7943 | |
f1174f77 EC |
7944 | /* Shouldn't get here; if we do, say it's not safe */ |
7945 | WARN_ON_ONCE(1); | |
969bf05e AS |
7946 | return false; |
7947 | } | |
7948 | ||
f4d7e40a AS |
7949 | static bool stacksafe(struct bpf_func_state *old, |
7950 | struct bpf_func_state *cur, | |
638f5b90 AS |
7951 | struct idpair *idmap) |
7952 | { | |
7953 | int i, spi; | |
7954 | ||
638f5b90 AS |
7955 | /* walk slots of the explored stack and ignore any additional |
7956 | * slots in the current stack, since explored(safe) state | |
7957 | * didn't use them | |
7958 | */ | |
7959 | for (i = 0; i < old->allocated_stack; i++) { | |
7960 | spi = i / BPF_REG_SIZE; | |
7961 | ||
b233920c AS |
7962 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
7963 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 7964 | /* explored state didn't use this */ |
fd05e57b | 7965 | continue; |
b233920c | 7966 | } |
cc2b14d5 | 7967 | |
638f5b90 AS |
7968 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
7969 | continue; | |
19e2dbb7 AS |
7970 | |
7971 | /* explored stack has more populated slots than current stack | |
7972 | * and these slots were used | |
7973 | */ | |
7974 | if (i >= cur->allocated_stack) | |
7975 | return false; | |
7976 | ||
cc2b14d5 AS |
7977 | /* if old state was safe with misc data in the stack |
7978 | * it will be safe with zero-initialized stack. | |
7979 | * The opposite is not true | |
7980 | */ | |
7981 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
7982 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
7983 | continue; | |
638f5b90 AS |
7984 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
7985 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
7986 | /* Ex: old explored (safe) state has STACK_SPILL in | |
7987 | * this stack slot, but current has has STACK_MISC -> | |
7988 | * this verifier states are not equivalent, | |
7989 | * return false to continue verification of this path | |
7990 | */ | |
7991 | return false; | |
7992 | if (i % BPF_REG_SIZE) | |
7993 | continue; | |
7994 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
7995 | continue; | |
7996 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
7997 | &cur->stack[spi].spilled_ptr, | |
7998 | idmap)) | |
7999 | /* when explored and current stack slot are both storing | |
8000 | * spilled registers, check that stored pointers types | |
8001 | * are the same as well. | |
8002 | * Ex: explored safe path could have stored | |
8003 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
8004 | * but current path has stored: | |
8005 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
8006 | * such verifier states are not equivalent. | |
8007 | * return false to continue verification of this path | |
8008 | */ | |
8009 | return false; | |
8010 | } | |
8011 | return true; | |
8012 | } | |
8013 | ||
fd978bf7 JS |
8014 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
8015 | { | |
8016 | if (old->acquired_refs != cur->acquired_refs) | |
8017 | return false; | |
8018 | return !memcmp(old->refs, cur->refs, | |
8019 | sizeof(*old->refs) * old->acquired_refs); | |
8020 | } | |
8021 | ||
f1bca824 AS |
8022 | /* compare two verifier states |
8023 | * | |
8024 | * all states stored in state_list are known to be valid, since | |
8025 | * verifier reached 'bpf_exit' instruction through them | |
8026 | * | |
8027 | * this function is called when verifier exploring different branches of | |
8028 | * execution popped from the state stack. If it sees an old state that has | |
8029 | * more strict register state and more strict stack state then this execution | |
8030 | * branch doesn't need to be explored further, since verifier already | |
8031 | * concluded that more strict state leads to valid finish. | |
8032 | * | |
8033 | * Therefore two states are equivalent if register state is more conservative | |
8034 | * and explored stack state is more conservative than the current one. | |
8035 | * Example: | |
8036 | * explored current | |
8037 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
8038 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
8039 | * | |
8040 | * In other words if current stack state (one being explored) has more | |
8041 | * valid slots than old one that already passed validation, it means | |
8042 | * the verifier can stop exploring and conclude that current state is valid too | |
8043 | * | |
8044 | * Similarly with registers. If explored state has register type as invalid | |
8045 | * whereas register type in current state is meaningful, it means that | |
8046 | * the current state will reach 'bpf_exit' instruction safely | |
8047 | */ | |
f4d7e40a AS |
8048 | static bool func_states_equal(struct bpf_func_state *old, |
8049 | struct bpf_func_state *cur) | |
f1bca824 | 8050 | { |
f1174f77 EC |
8051 | struct idpair *idmap; |
8052 | bool ret = false; | |
f1bca824 AS |
8053 | int i; |
8054 | ||
f1174f77 EC |
8055 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
8056 | /* If we failed to allocate the idmap, just say it's not safe */ | |
8057 | if (!idmap) | |
1a0dc1ac | 8058 | return false; |
f1174f77 EC |
8059 | |
8060 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 8061 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 8062 | goto out_free; |
f1bca824 AS |
8063 | } |
8064 | ||
638f5b90 AS |
8065 | if (!stacksafe(old, cur, idmap)) |
8066 | goto out_free; | |
fd978bf7 JS |
8067 | |
8068 | if (!refsafe(old, cur)) | |
8069 | goto out_free; | |
f1174f77 EC |
8070 | ret = true; |
8071 | out_free: | |
8072 | kfree(idmap); | |
8073 | return ret; | |
f1bca824 AS |
8074 | } |
8075 | ||
f4d7e40a AS |
8076 | static bool states_equal(struct bpf_verifier_env *env, |
8077 | struct bpf_verifier_state *old, | |
8078 | struct bpf_verifier_state *cur) | |
8079 | { | |
8080 | int i; | |
8081 | ||
8082 | if (old->curframe != cur->curframe) | |
8083 | return false; | |
8084 | ||
979d63d5 DB |
8085 | /* Verification state from speculative execution simulation |
8086 | * must never prune a non-speculative execution one. | |
8087 | */ | |
8088 | if (old->speculative && !cur->speculative) | |
8089 | return false; | |
8090 | ||
d83525ca AS |
8091 | if (old->active_spin_lock != cur->active_spin_lock) |
8092 | return false; | |
8093 | ||
f4d7e40a AS |
8094 | /* for states to be equal callsites have to be the same |
8095 | * and all frame states need to be equivalent | |
8096 | */ | |
8097 | for (i = 0; i <= old->curframe; i++) { | |
8098 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
8099 | return false; | |
8100 | if (!func_states_equal(old->frame[i], cur->frame[i])) | |
8101 | return false; | |
8102 | } | |
8103 | return true; | |
8104 | } | |
8105 | ||
5327ed3d JW |
8106 | /* Return 0 if no propagation happened. Return negative error code if error |
8107 | * happened. Otherwise, return the propagated bit. | |
8108 | */ | |
55e7f3b5 JW |
8109 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
8110 | struct bpf_reg_state *reg, | |
8111 | struct bpf_reg_state *parent_reg) | |
8112 | { | |
5327ed3d JW |
8113 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
8114 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
8115 | int err; |
8116 | ||
5327ed3d JW |
8117 | /* When comes here, read flags of PARENT_REG or REG could be any of |
8118 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
8119 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
8120 | */ | |
8121 | if (parent_flag == REG_LIVE_READ64 || | |
8122 | /* Or if there is no read flag from REG. */ | |
8123 | !flag || | |
8124 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
8125 | parent_flag == flag) | |
55e7f3b5 JW |
8126 | return 0; |
8127 | ||
5327ed3d | 8128 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
8129 | if (err) |
8130 | return err; | |
8131 | ||
5327ed3d | 8132 | return flag; |
55e7f3b5 JW |
8133 | } |
8134 | ||
8e9cd9ce | 8135 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
8136 | * straight-line code between a state and its parent. When we arrive at an |
8137 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
8138 | * code, so read marks in the state must propagate to the parent regardless | |
8139 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 8140 | * in mark_reg_read() is for. |
8e9cd9ce | 8141 | */ |
f4d7e40a AS |
8142 | static int propagate_liveness(struct bpf_verifier_env *env, |
8143 | const struct bpf_verifier_state *vstate, | |
8144 | struct bpf_verifier_state *vparent) | |
dc503a8a | 8145 | { |
3f8cafa4 | 8146 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 8147 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 8148 | int i, frame, err = 0; |
dc503a8a | 8149 | |
f4d7e40a AS |
8150 | if (vparent->curframe != vstate->curframe) { |
8151 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
8152 | vparent->curframe, vstate->curframe); | |
8153 | return -EFAULT; | |
8154 | } | |
dc503a8a EC |
8155 | /* Propagate read liveness of registers... */ |
8156 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 8157 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
8158 | parent = vparent->frame[frame]; |
8159 | state = vstate->frame[frame]; | |
8160 | parent_reg = parent->regs; | |
8161 | state_reg = state->regs; | |
83d16312 JK |
8162 | /* We don't need to worry about FP liveness, it's read-only */ |
8163 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
8164 | err = propagate_liveness_reg(env, &state_reg[i], |
8165 | &parent_reg[i]); | |
5327ed3d | 8166 | if (err < 0) |
3f8cafa4 | 8167 | return err; |
5327ed3d JW |
8168 | if (err == REG_LIVE_READ64) |
8169 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 8170 | } |
f4d7e40a | 8171 | |
1b04aee7 | 8172 | /* Propagate stack slots. */ |
f4d7e40a AS |
8173 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
8174 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
8175 | parent_reg = &parent->stack[i].spilled_ptr; |
8176 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
8177 | err = propagate_liveness_reg(env, state_reg, |
8178 | parent_reg); | |
5327ed3d | 8179 | if (err < 0) |
3f8cafa4 | 8180 | return err; |
dc503a8a EC |
8181 | } |
8182 | } | |
5327ed3d | 8183 | return 0; |
dc503a8a EC |
8184 | } |
8185 | ||
a3ce685d AS |
8186 | /* find precise scalars in the previous equivalent state and |
8187 | * propagate them into the current state | |
8188 | */ | |
8189 | static int propagate_precision(struct bpf_verifier_env *env, | |
8190 | const struct bpf_verifier_state *old) | |
8191 | { | |
8192 | struct bpf_reg_state *state_reg; | |
8193 | struct bpf_func_state *state; | |
8194 | int i, err = 0; | |
8195 | ||
8196 | state = old->frame[old->curframe]; | |
8197 | state_reg = state->regs; | |
8198 | for (i = 0; i < BPF_REG_FP; i++, state_reg++) { | |
8199 | if (state_reg->type != SCALAR_VALUE || | |
8200 | !state_reg->precise) | |
8201 | continue; | |
8202 | if (env->log.level & BPF_LOG_LEVEL2) | |
8203 | verbose(env, "propagating r%d\n", i); | |
8204 | err = mark_chain_precision(env, i); | |
8205 | if (err < 0) | |
8206 | return err; | |
8207 | } | |
8208 | ||
8209 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { | |
8210 | if (state->stack[i].slot_type[0] != STACK_SPILL) | |
8211 | continue; | |
8212 | state_reg = &state->stack[i].spilled_ptr; | |
8213 | if (state_reg->type != SCALAR_VALUE || | |
8214 | !state_reg->precise) | |
8215 | continue; | |
8216 | if (env->log.level & BPF_LOG_LEVEL2) | |
8217 | verbose(env, "propagating fp%d\n", | |
8218 | (-i - 1) * BPF_REG_SIZE); | |
8219 | err = mark_chain_precision_stack(env, i); | |
8220 | if (err < 0) | |
8221 | return err; | |
8222 | } | |
8223 | return 0; | |
8224 | } | |
8225 | ||
2589726d AS |
8226 | static bool states_maybe_looping(struct bpf_verifier_state *old, |
8227 | struct bpf_verifier_state *cur) | |
8228 | { | |
8229 | struct bpf_func_state *fold, *fcur; | |
8230 | int i, fr = cur->curframe; | |
8231 | ||
8232 | if (old->curframe != fr) | |
8233 | return false; | |
8234 | ||
8235 | fold = old->frame[fr]; | |
8236 | fcur = cur->frame[fr]; | |
8237 | for (i = 0; i < MAX_BPF_REG; i++) | |
8238 | if (memcmp(&fold->regs[i], &fcur->regs[i], | |
8239 | offsetof(struct bpf_reg_state, parent))) | |
8240 | return false; | |
8241 | return true; | |
8242 | } | |
8243 | ||
8244 | ||
58e2af8b | 8245 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 8246 | { |
58e2af8b | 8247 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 8248 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 8249 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 8250 | int i, j, err, states_cnt = 0; |
10d274e8 | 8251 | bool add_new_state = env->test_state_freq ? true : false; |
f1bca824 | 8252 | |
b5dc0163 | 8253 | cur->last_insn_idx = env->prev_insn_idx; |
a8f500af | 8254 | if (!env->insn_aux_data[insn_idx].prune_point) |
f1bca824 AS |
8255 | /* this 'insn_idx' instruction wasn't marked, so we will not |
8256 | * be doing state search here | |
8257 | */ | |
8258 | return 0; | |
8259 | ||
2589726d AS |
8260 | /* bpf progs typically have pruning point every 4 instructions |
8261 | * http://vger.kernel.org/bpfconf2019.html#session-1 | |
8262 | * Do not add new state for future pruning if the verifier hasn't seen | |
8263 | * at least 2 jumps and at least 8 instructions. | |
8264 | * This heuristics helps decrease 'total_states' and 'peak_states' metric. | |
8265 | * In tests that amounts to up to 50% reduction into total verifier | |
8266 | * memory consumption and 20% verifier time speedup. | |
8267 | */ | |
8268 | if (env->jmps_processed - env->prev_jmps_processed >= 2 && | |
8269 | env->insn_processed - env->prev_insn_processed >= 8) | |
8270 | add_new_state = true; | |
8271 | ||
a8f500af AS |
8272 | pprev = explored_state(env, insn_idx); |
8273 | sl = *pprev; | |
8274 | ||
9242b5f5 AS |
8275 | clean_live_states(env, insn_idx, cur); |
8276 | ||
a8f500af | 8277 | while (sl) { |
dc2a4ebc AS |
8278 | states_cnt++; |
8279 | if (sl->state.insn_idx != insn_idx) | |
8280 | goto next; | |
2589726d AS |
8281 | if (sl->state.branches) { |
8282 | if (states_maybe_looping(&sl->state, cur) && | |
8283 | states_equal(env, &sl->state, cur)) { | |
8284 | verbose_linfo(env, insn_idx, "; "); | |
8285 | verbose(env, "infinite loop detected at insn %d\n", insn_idx); | |
8286 | return -EINVAL; | |
8287 | } | |
8288 | /* if the verifier is processing a loop, avoid adding new state | |
8289 | * too often, since different loop iterations have distinct | |
8290 | * states and may not help future pruning. | |
8291 | * This threshold shouldn't be too low to make sure that | |
8292 | * a loop with large bound will be rejected quickly. | |
8293 | * The most abusive loop will be: | |
8294 | * r1 += 1 | |
8295 | * if r1 < 1000000 goto pc-2 | |
8296 | * 1M insn_procssed limit / 100 == 10k peak states. | |
8297 | * This threshold shouldn't be too high either, since states | |
8298 | * at the end of the loop are likely to be useful in pruning. | |
8299 | */ | |
8300 | if (env->jmps_processed - env->prev_jmps_processed < 20 && | |
8301 | env->insn_processed - env->prev_insn_processed < 100) | |
8302 | add_new_state = false; | |
8303 | goto miss; | |
8304 | } | |
638f5b90 | 8305 | if (states_equal(env, &sl->state, cur)) { |
9f4686c4 | 8306 | sl->hit_cnt++; |
f1bca824 | 8307 | /* reached equivalent register/stack state, |
dc503a8a EC |
8308 | * prune the search. |
8309 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
8310 | * If we have any write marks in env->cur_state, they |
8311 | * will prevent corresponding reads in the continuation | |
8312 | * from reaching our parent (an explored_state). Our | |
8313 | * own state will get the read marks recorded, but | |
8314 | * they'll be immediately forgotten as we're pruning | |
8315 | * this state and will pop a new one. | |
f1bca824 | 8316 | */ |
f4d7e40a | 8317 | err = propagate_liveness(env, &sl->state, cur); |
a3ce685d AS |
8318 | |
8319 | /* if previous state reached the exit with precision and | |
8320 | * current state is equivalent to it (except precsion marks) | |
8321 | * the precision needs to be propagated back in | |
8322 | * the current state. | |
8323 | */ | |
8324 | err = err ? : push_jmp_history(env, cur); | |
8325 | err = err ? : propagate_precision(env, &sl->state); | |
f4d7e40a AS |
8326 | if (err) |
8327 | return err; | |
f1bca824 | 8328 | return 1; |
dc503a8a | 8329 | } |
2589726d AS |
8330 | miss: |
8331 | /* when new state is not going to be added do not increase miss count. | |
8332 | * Otherwise several loop iterations will remove the state | |
8333 | * recorded earlier. The goal of these heuristics is to have | |
8334 | * states from some iterations of the loop (some in the beginning | |
8335 | * and some at the end) to help pruning. | |
8336 | */ | |
8337 | if (add_new_state) | |
8338 | sl->miss_cnt++; | |
9f4686c4 AS |
8339 | /* heuristic to determine whether this state is beneficial |
8340 | * to keep checking from state equivalence point of view. | |
8341 | * Higher numbers increase max_states_per_insn and verification time, | |
8342 | * but do not meaningfully decrease insn_processed. | |
8343 | */ | |
8344 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
8345 | /* the state is unlikely to be useful. Remove it to | |
8346 | * speed up verification | |
8347 | */ | |
8348 | *pprev = sl->next; | |
8349 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
2589726d AS |
8350 | u32 br = sl->state.branches; |
8351 | ||
8352 | WARN_ONCE(br, | |
8353 | "BUG live_done but branches_to_explore %d\n", | |
8354 | br); | |
9f4686c4 AS |
8355 | free_verifier_state(&sl->state, false); |
8356 | kfree(sl); | |
8357 | env->peak_states--; | |
8358 | } else { | |
8359 | /* cannot free this state, since parentage chain may | |
8360 | * walk it later. Add it for free_list instead to | |
8361 | * be freed at the end of verification | |
8362 | */ | |
8363 | sl->next = env->free_list; | |
8364 | env->free_list = sl; | |
8365 | } | |
8366 | sl = *pprev; | |
8367 | continue; | |
8368 | } | |
dc2a4ebc | 8369 | next: |
9f4686c4 AS |
8370 | pprev = &sl->next; |
8371 | sl = *pprev; | |
f1bca824 AS |
8372 | } |
8373 | ||
06ee7115 AS |
8374 | if (env->max_states_per_insn < states_cnt) |
8375 | env->max_states_per_insn = states_cnt; | |
8376 | ||
2c78ee89 | 8377 | if (!env->bpf_capable && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
b5dc0163 | 8378 | return push_jmp_history(env, cur); |
ceefbc96 | 8379 | |
2589726d | 8380 | if (!add_new_state) |
b5dc0163 | 8381 | return push_jmp_history(env, cur); |
ceefbc96 | 8382 | |
2589726d AS |
8383 | /* There were no equivalent states, remember the current one. |
8384 | * Technically the current state is not proven to be safe yet, | |
f4d7e40a | 8385 | * but it will either reach outer most bpf_exit (which means it's safe) |
2589726d | 8386 | * or it will be rejected. When there are no loops the verifier won't be |
f4d7e40a | 8387 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) |
2589726d AS |
8388 | * again on the way to bpf_exit. |
8389 | * When looping the sl->state.branches will be > 0 and this state | |
8390 | * will not be considered for equivalence until branches == 0. | |
f1bca824 | 8391 | */ |
638f5b90 | 8392 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
8393 | if (!new_sl) |
8394 | return -ENOMEM; | |
06ee7115 AS |
8395 | env->total_states++; |
8396 | env->peak_states++; | |
2589726d AS |
8397 | env->prev_jmps_processed = env->jmps_processed; |
8398 | env->prev_insn_processed = env->insn_processed; | |
f1bca824 AS |
8399 | |
8400 | /* add new state to the head of linked list */ | |
679c782d EC |
8401 | new = &new_sl->state; |
8402 | err = copy_verifier_state(new, cur); | |
1969db47 | 8403 | if (err) { |
679c782d | 8404 | free_verifier_state(new, false); |
1969db47 AS |
8405 | kfree(new_sl); |
8406 | return err; | |
8407 | } | |
dc2a4ebc | 8408 | new->insn_idx = insn_idx; |
2589726d AS |
8409 | WARN_ONCE(new->branches != 1, |
8410 | "BUG is_state_visited:branches_to_explore=%d insn %d\n", new->branches, insn_idx); | |
b5dc0163 | 8411 | |
2589726d | 8412 | cur->parent = new; |
b5dc0163 AS |
8413 | cur->first_insn_idx = insn_idx; |
8414 | clear_jmp_history(cur); | |
5d839021 AS |
8415 | new_sl->next = *explored_state(env, insn_idx); |
8416 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
8417 | /* connect new state to parentage chain. Current frame needs all |
8418 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
8419 | * to the stack implicitly by JITs) so in callers' frames connect just | |
8420 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
8421 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
8422 | * from callee with its full parentage chain, anyway. | |
8423 | */ | |
8e9cd9ce EC |
8424 | /* clear write marks in current state: the writes we did are not writes |
8425 | * our child did, so they don't screen off its reads from us. | |
8426 | * (There are no read marks in current state, because reads always mark | |
8427 | * their parent and current state never has children yet. Only | |
8428 | * explored_states can get read marks.) | |
8429 | */ | |
eea1c227 AS |
8430 | for (j = 0; j <= cur->curframe; j++) { |
8431 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
8432 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
8433 | for (i = 0; i < BPF_REG_FP; i++) | |
8434 | cur->frame[j]->regs[i].live = REG_LIVE_NONE; | |
8435 | } | |
f4d7e40a AS |
8436 | |
8437 | /* all stack frames are accessible from callee, clear them all */ | |
8438 | for (j = 0; j <= cur->curframe; j++) { | |
8439 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 8440 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 8441 | |
679c782d | 8442 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 8443 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
8444 | frame->stack[i].spilled_ptr.parent = |
8445 | &newframe->stack[i].spilled_ptr; | |
8446 | } | |
f4d7e40a | 8447 | } |
f1bca824 AS |
8448 | return 0; |
8449 | } | |
8450 | ||
c64b7983 JS |
8451 | /* Return true if it's OK to have the same insn return a different type. */ |
8452 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
8453 | { | |
8454 | switch (type) { | |
8455 | case PTR_TO_CTX: | |
8456 | case PTR_TO_SOCKET: | |
8457 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
8458 | case PTR_TO_SOCK_COMMON: |
8459 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
8460 | case PTR_TO_TCP_SOCK: |
8461 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 8462 | case PTR_TO_XDP_SOCK: |
2a02759e | 8463 | case PTR_TO_BTF_ID: |
b121b341 | 8464 | case PTR_TO_BTF_ID_OR_NULL: |
c64b7983 JS |
8465 | return false; |
8466 | default: | |
8467 | return true; | |
8468 | } | |
8469 | } | |
8470 | ||
8471 | /* If an instruction was previously used with particular pointer types, then we | |
8472 | * need to be careful to avoid cases such as the below, where it may be ok | |
8473 | * for one branch accessing the pointer, but not ok for the other branch: | |
8474 | * | |
8475 | * R1 = sock_ptr | |
8476 | * goto X; | |
8477 | * ... | |
8478 | * R1 = some_other_valid_ptr; | |
8479 | * goto X; | |
8480 | * ... | |
8481 | * R2 = *(u32 *)(R1 + 0); | |
8482 | */ | |
8483 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
8484 | { | |
8485 | return src != prev && (!reg_type_mismatch_ok(src) || | |
8486 | !reg_type_mismatch_ok(prev)); | |
8487 | } | |
8488 | ||
58e2af8b | 8489 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 8490 | { |
6f8a57cc | 8491 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 | 8492 | struct bpf_verifier_state *state = env->cur_state; |
17a52670 | 8493 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 8494 | struct bpf_reg_state *regs; |
06ee7115 | 8495 | int insn_cnt = env->prog->len; |
17a52670 | 8496 | bool do_print_state = false; |
b5dc0163 | 8497 | int prev_insn_idx = -1; |
17a52670 | 8498 | |
17a52670 AS |
8499 | for (;;) { |
8500 | struct bpf_insn *insn; | |
8501 | u8 class; | |
8502 | int err; | |
8503 | ||
b5dc0163 | 8504 | env->prev_insn_idx = prev_insn_idx; |
c08435ec | 8505 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 8506 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 8507 | env->insn_idx, insn_cnt); |
17a52670 AS |
8508 | return -EFAULT; |
8509 | } | |
8510 | ||
c08435ec | 8511 | insn = &insns[env->insn_idx]; |
17a52670 AS |
8512 | class = BPF_CLASS(insn->code); |
8513 | ||
06ee7115 | 8514 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
8515 | verbose(env, |
8516 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 8517 | env->insn_processed); |
17a52670 AS |
8518 | return -E2BIG; |
8519 | } | |
8520 | ||
c08435ec | 8521 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
8522 | if (err < 0) |
8523 | return err; | |
8524 | if (err == 1) { | |
8525 | /* found equivalent state, can prune the search */ | |
06ee7115 | 8526 | if (env->log.level & BPF_LOG_LEVEL) { |
f1bca824 | 8527 | if (do_print_state) |
979d63d5 DB |
8528 | verbose(env, "\nfrom %d to %d%s: safe\n", |
8529 | env->prev_insn_idx, env->insn_idx, | |
8530 | env->cur_state->speculative ? | |
8531 | " (speculative execution)" : ""); | |
f1bca824 | 8532 | else |
c08435ec | 8533 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
8534 | } |
8535 | goto process_bpf_exit; | |
8536 | } | |
8537 | ||
c3494801 AS |
8538 | if (signal_pending(current)) |
8539 | return -EAGAIN; | |
8540 | ||
3c2ce60b DB |
8541 | if (need_resched()) |
8542 | cond_resched(); | |
8543 | ||
06ee7115 AS |
8544 | if (env->log.level & BPF_LOG_LEVEL2 || |
8545 | (env->log.level & BPF_LOG_LEVEL && do_print_state)) { | |
8546 | if (env->log.level & BPF_LOG_LEVEL2) | |
c08435ec | 8547 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 8548 | else |
979d63d5 DB |
8549 | verbose(env, "\nfrom %d to %d%s:", |
8550 | env->prev_insn_idx, env->insn_idx, | |
8551 | env->cur_state->speculative ? | |
8552 | " (speculative execution)" : ""); | |
f4d7e40a | 8553 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
8554 | do_print_state = false; |
8555 | } | |
8556 | ||
06ee7115 | 8557 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 DB |
8558 | const struct bpf_insn_cbs cbs = { |
8559 | .cb_print = verbose, | |
abe08840 | 8560 | .private_data = env, |
7105e828 DB |
8561 | }; |
8562 | ||
c08435ec DB |
8563 | verbose_linfo(env, env->insn_idx, "; "); |
8564 | verbose(env, "%d: ", env->insn_idx); | |
abe08840 | 8565 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
8566 | } |
8567 | ||
cae1927c | 8568 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
8569 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
8570 | env->prev_insn_idx); | |
cae1927c JK |
8571 | if (err) |
8572 | return err; | |
8573 | } | |
13a27dfc | 8574 | |
638f5b90 | 8575 | regs = cur_regs(env); |
51c39bb1 | 8576 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; |
b5dc0163 | 8577 | prev_insn_idx = env->insn_idx; |
fd978bf7 | 8578 | |
17a52670 | 8579 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 8580 | err = check_alu_op(env, insn); |
17a52670 AS |
8581 | if (err) |
8582 | return err; | |
8583 | ||
8584 | } else if (class == BPF_LDX) { | |
3df126f3 | 8585 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
8586 | |
8587 | /* check for reserved fields is already done */ | |
8588 | ||
17a52670 | 8589 | /* check src operand */ |
dc503a8a | 8590 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
8591 | if (err) |
8592 | return err; | |
8593 | ||
dc503a8a | 8594 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
8595 | if (err) |
8596 | return err; | |
8597 | ||
725f9dcd AS |
8598 | src_reg_type = regs[insn->src_reg].type; |
8599 | ||
17a52670 AS |
8600 | /* check that memory (src_reg + off) is readable, |
8601 | * the state of dst_reg will be updated by this func | |
8602 | */ | |
c08435ec DB |
8603 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
8604 | insn->off, BPF_SIZE(insn->code), | |
8605 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
8606 | if (err) |
8607 | return err; | |
8608 | ||
c08435ec | 8609 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
8610 | |
8611 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
8612 | /* saw a valid insn |
8613 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 8614 | * save type to validate intersecting paths |
9bac3d6d | 8615 | */ |
3df126f3 | 8616 | *prev_src_type = src_reg_type; |
9bac3d6d | 8617 | |
c64b7983 | 8618 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
8619 | /* ABuser program is trying to use the same insn |
8620 | * dst_reg = *(u32*) (src_reg + off) | |
8621 | * with different pointer types: | |
8622 | * src_reg == ctx in one branch and | |
8623 | * src_reg == stack|map in some other branch. | |
8624 | * Reject it. | |
8625 | */ | |
61bd5218 | 8626 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
8627 | return -EINVAL; |
8628 | } | |
8629 | ||
17a52670 | 8630 | } else if (class == BPF_STX) { |
3df126f3 | 8631 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 8632 | |
17a52670 | 8633 | if (BPF_MODE(insn->code) == BPF_XADD) { |
c08435ec | 8634 | err = check_xadd(env, env->insn_idx, insn); |
17a52670 AS |
8635 | if (err) |
8636 | return err; | |
c08435ec | 8637 | env->insn_idx++; |
17a52670 AS |
8638 | continue; |
8639 | } | |
8640 | ||
17a52670 | 8641 | /* check src1 operand */ |
dc503a8a | 8642 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
8643 | if (err) |
8644 | return err; | |
8645 | /* check src2 operand */ | |
dc503a8a | 8646 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
8647 | if (err) |
8648 | return err; | |
8649 | ||
d691f9e8 AS |
8650 | dst_reg_type = regs[insn->dst_reg].type; |
8651 | ||
17a52670 | 8652 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
8653 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
8654 | insn->off, BPF_SIZE(insn->code), | |
8655 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
8656 | if (err) |
8657 | return err; | |
8658 | ||
c08435ec | 8659 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
8660 | |
8661 | if (*prev_dst_type == NOT_INIT) { | |
8662 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 8663 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 8664 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
8665 | return -EINVAL; |
8666 | } | |
8667 | ||
17a52670 AS |
8668 | } else if (class == BPF_ST) { |
8669 | if (BPF_MODE(insn->code) != BPF_MEM || | |
8670 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 8671 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
8672 | return -EINVAL; |
8673 | } | |
8674 | /* check src operand */ | |
dc503a8a | 8675 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
8676 | if (err) |
8677 | return err; | |
8678 | ||
f37a8cb8 | 8679 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 8680 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f DB |
8681 | insn->dst_reg, |
8682 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
8683 | return -EACCES; |
8684 | } | |
8685 | ||
17a52670 | 8686 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
8687 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
8688 | insn->off, BPF_SIZE(insn->code), | |
8689 | BPF_WRITE, -1, false); | |
17a52670 AS |
8690 | if (err) |
8691 | return err; | |
8692 | ||
092ed096 | 8693 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
8694 | u8 opcode = BPF_OP(insn->code); |
8695 | ||
2589726d | 8696 | env->jmps_processed++; |
17a52670 AS |
8697 | if (opcode == BPF_CALL) { |
8698 | if (BPF_SRC(insn->code) != BPF_K || | |
8699 | insn->off != 0 || | |
f4d7e40a AS |
8700 | (insn->src_reg != BPF_REG_0 && |
8701 | insn->src_reg != BPF_PSEUDO_CALL) || | |
092ed096 JW |
8702 | insn->dst_reg != BPF_REG_0 || |
8703 | class == BPF_JMP32) { | |
61bd5218 | 8704 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
8705 | return -EINVAL; |
8706 | } | |
8707 | ||
d83525ca AS |
8708 | if (env->cur_state->active_spin_lock && |
8709 | (insn->src_reg == BPF_PSEUDO_CALL || | |
8710 | insn->imm != BPF_FUNC_spin_unlock)) { | |
8711 | verbose(env, "function calls are not allowed while holding a lock\n"); | |
8712 | return -EINVAL; | |
8713 | } | |
f4d7e40a | 8714 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 8715 | err = check_func_call(env, insn, &env->insn_idx); |
f4d7e40a | 8716 | else |
c08435ec | 8717 | err = check_helper_call(env, insn->imm, env->insn_idx); |
17a52670 AS |
8718 | if (err) |
8719 | return err; | |
8720 | ||
8721 | } else if (opcode == BPF_JA) { | |
8722 | if (BPF_SRC(insn->code) != BPF_K || | |
8723 | insn->imm != 0 || | |
8724 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
8725 | insn->dst_reg != BPF_REG_0 || |
8726 | class == BPF_JMP32) { | |
61bd5218 | 8727 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
8728 | return -EINVAL; |
8729 | } | |
8730 | ||
c08435ec | 8731 | env->insn_idx += insn->off + 1; |
17a52670 AS |
8732 | continue; |
8733 | ||
8734 | } else if (opcode == BPF_EXIT) { | |
8735 | if (BPF_SRC(insn->code) != BPF_K || | |
8736 | insn->imm != 0 || | |
8737 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
8738 | insn->dst_reg != BPF_REG_0 || |
8739 | class == BPF_JMP32) { | |
61bd5218 | 8740 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
8741 | return -EINVAL; |
8742 | } | |
8743 | ||
d83525ca AS |
8744 | if (env->cur_state->active_spin_lock) { |
8745 | verbose(env, "bpf_spin_unlock is missing\n"); | |
8746 | return -EINVAL; | |
8747 | } | |
8748 | ||
f4d7e40a AS |
8749 | if (state->curframe) { |
8750 | /* exit from nested function */ | |
c08435ec | 8751 | err = prepare_func_exit(env, &env->insn_idx); |
f4d7e40a AS |
8752 | if (err) |
8753 | return err; | |
8754 | do_print_state = true; | |
8755 | continue; | |
8756 | } | |
8757 | ||
fd978bf7 JS |
8758 | err = check_reference_leak(env); |
8759 | if (err) | |
8760 | return err; | |
8761 | ||
390ee7e2 AS |
8762 | err = check_return_code(env); |
8763 | if (err) | |
8764 | return err; | |
f1bca824 | 8765 | process_bpf_exit: |
2589726d | 8766 | update_branch_counts(env, env->cur_state); |
b5dc0163 | 8767 | err = pop_stack(env, &prev_insn_idx, |
6f8a57cc | 8768 | &env->insn_idx, pop_log); |
638f5b90 AS |
8769 | if (err < 0) { |
8770 | if (err != -ENOENT) | |
8771 | return err; | |
17a52670 AS |
8772 | break; |
8773 | } else { | |
8774 | do_print_state = true; | |
8775 | continue; | |
8776 | } | |
8777 | } else { | |
c08435ec | 8778 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
8779 | if (err) |
8780 | return err; | |
8781 | } | |
8782 | } else if (class == BPF_LD) { | |
8783 | u8 mode = BPF_MODE(insn->code); | |
8784 | ||
8785 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
8786 | err = check_ld_abs(env, insn); |
8787 | if (err) | |
8788 | return err; | |
8789 | ||
17a52670 AS |
8790 | } else if (mode == BPF_IMM) { |
8791 | err = check_ld_imm(env, insn); | |
8792 | if (err) | |
8793 | return err; | |
8794 | ||
c08435ec | 8795 | env->insn_idx++; |
51c39bb1 | 8796 | env->insn_aux_data[env->insn_idx].seen = env->pass_cnt; |
17a52670 | 8797 | } else { |
61bd5218 | 8798 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
8799 | return -EINVAL; |
8800 | } | |
8801 | } else { | |
61bd5218 | 8802 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
8803 | return -EINVAL; |
8804 | } | |
8805 | ||
c08435ec | 8806 | env->insn_idx++; |
17a52670 AS |
8807 | } |
8808 | ||
8809 | return 0; | |
8810 | } | |
8811 | ||
56f668df MKL |
8812 | static int check_map_prealloc(struct bpf_map *map) |
8813 | { | |
8814 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
8815 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
8816 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
8817 | !(map->map_flags & BPF_F_NO_PREALLOC); |
8818 | } | |
8819 | ||
d83525ca AS |
8820 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
8821 | { | |
8822 | switch (type) { | |
8823 | case BPF_PROG_TYPE_KPROBE: | |
8824 | case BPF_PROG_TYPE_TRACEPOINT: | |
8825 | case BPF_PROG_TYPE_PERF_EVENT: | |
8826 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
8827 | return true; | |
8828 | default: | |
8829 | return false; | |
8830 | } | |
8831 | } | |
8832 | ||
94dacdbd TG |
8833 | static bool is_preallocated_map(struct bpf_map *map) |
8834 | { | |
8835 | if (!check_map_prealloc(map)) | |
8836 | return false; | |
8837 | if (map->inner_map_meta && !check_map_prealloc(map->inner_map_meta)) | |
8838 | return false; | |
8839 | return true; | |
8840 | } | |
8841 | ||
61bd5218 JK |
8842 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
8843 | struct bpf_map *map, | |
fdc15d38 AS |
8844 | struct bpf_prog *prog) |
8845 | ||
8846 | { | |
94dacdbd TG |
8847 | /* |
8848 | * Validate that trace type programs use preallocated hash maps. | |
8849 | * | |
8850 | * For programs attached to PERF events this is mandatory as the | |
8851 | * perf NMI can hit any arbitrary code sequence. | |
8852 | * | |
8853 | * All other trace types using preallocated hash maps are unsafe as | |
8854 | * well because tracepoint or kprobes can be inside locked regions | |
8855 | * of the memory allocator or at a place where a recursion into the | |
8856 | * memory allocator would see inconsistent state. | |
8857 | * | |
2ed905c5 TG |
8858 | * On RT enabled kernels run-time allocation of all trace type |
8859 | * programs is strictly prohibited due to lock type constraints. On | |
8860 | * !RT kernels it is allowed for backwards compatibility reasons for | |
8861 | * now, but warnings are emitted so developers are made aware of | |
8862 | * the unsafety and can fix their programs before this is enforced. | |
56f668df | 8863 | */ |
94dacdbd TG |
8864 | if (is_tracing_prog_type(prog->type) && !is_preallocated_map(map)) { |
8865 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
61bd5218 | 8866 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
8867 | return -EINVAL; |
8868 | } | |
2ed905c5 TG |
8869 | if (IS_ENABLED(CONFIG_PREEMPT_RT)) { |
8870 | verbose(env, "trace type programs can only use preallocated hash map\n"); | |
8871 | return -EINVAL; | |
8872 | } | |
94dacdbd TG |
8873 | WARN_ONCE(1, "trace type BPF program uses run-time allocation\n"); |
8874 | verbose(env, "trace type programs with run-time allocated hash maps are unsafe. Switch to preallocated hash maps.\n"); | |
fdc15d38 | 8875 | } |
a3884572 | 8876 | |
d83525ca AS |
8877 | if ((is_tracing_prog_type(prog->type) || |
8878 | prog->type == BPF_PROG_TYPE_SOCKET_FILTER) && | |
8879 | map_value_has_spin_lock(map)) { | |
8880 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
8881 | return -EINVAL; | |
8882 | } | |
8883 | ||
a3884572 | 8884 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && |
09728266 | 8885 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
8886 | verbose(env, "offload device mismatch between prog and map\n"); |
8887 | return -EINVAL; | |
8888 | } | |
8889 | ||
85d33df3 MKL |
8890 | if (map->map_type == BPF_MAP_TYPE_STRUCT_OPS) { |
8891 | verbose(env, "bpf_struct_ops map cannot be used in prog\n"); | |
8892 | return -EINVAL; | |
8893 | } | |
8894 | ||
fdc15d38 AS |
8895 | return 0; |
8896 | } | |
8897 | ||
b741f163 RG |
8898 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
8899 | { | |
8900 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
8901 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
8902 | } | |
8903 | ||
0246e64d AS |
8904 | /* look for pseudo eBPF instructions that access map FDs and |
8905 | * replace them with actual map pointers | |
8906 | */ | |
58e2af8b | 8907 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
8908 | { |
8909 | struct bpf_insn *insn = env->prog->insnsi; | |
8910 | int insn_cnt = env->prog->len; | |
fdc15d38 | 8911 | int i, j, err; |
0246e64d | 8912 | |
f1f7714e | 8913 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
8914 | if (err) |
8915 | return err; | |
8916 | ||
0246e64d | 8917 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 8918 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 8919 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 8920 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
8921 | return -EINVAL; |
8922 | } | |
8923 | ||
d691f9e8 AS |
8924 | if (BPF_CLASS(insn->code) == BPF_STX && |
8925 | ((BPF_MODE(insn->code) != BPF_MEM && | |
8926 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 8927 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
8928 | return -EINVAL; |
8929 | } | |
8930 | ||
0246e64d | 8931 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 8932 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
8933 | struct bpf_map *map; |
8934 | struct fd f; | |
d8eca5bb | 8935 | u64 addr; |
0246e64d AS |
8936 | |
8937 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
8938 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
8939 | insn[1].off != 0) { | |
61bd5218 | 8940 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
8941 | return -EINVAL; |
8942 | } | |
8943 | ||
d8eca5bb | 8944 | if (insn[0].src_reg == 0) |
0246e64d AS |
8945 | /* valid generic load 64-bit imm */ |
8946 | goto next_insn; | |
8947 | ||
d8eca5bb DB |
8948 | /* In final convert_pseudo_ld_imm64() step, this is |
8949 | * converted into regular 64-bit imm load insn. | |
8950 | */ | |
8951 | if ((insn[0].src_reg != BPF_PSEUDO_MAP_FD && | |
8952 | insn[0].src_reg != BPF_PSEUDO_MAP_VALUE) || | |
8953 | (insn[0].src_reg == BPF_PSEUDO_MAP_FD && | |
8954 | insn[1].imm != 0)) { | |
8955 | verbose(env, | |
8956 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
8957 | return -EINVAL; |
8958 | } | |
8959 | ||
20182390 | 8960 | f = fdget(insn[0].imm); |
c2101297 | 8961 | map = __bpf_map_get(f); |
0246e64d | 8962 | if (IS_ERR(map)) { |
61bd5218 | 8963 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 8964 | insn[0].imm); |
0246e64d AS |
8965 | return PTR_ERR(map); |
8966 | } | |
8967 | ||
61bd5218 | 8968 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
8969 | if (err) { |
8970 | fdput(f); | |
8971 | return err; | |
8972 | } | |
8973 | ||
d8eca5bb DB |
8974 | aux = &env->insn_aux_data[i]; |
8975 | if (insn->src_reg == BPF_PSEUDO_MAP_FD) { | |
8976 | addr = (unsigned long)map; | |
8977 | } else { | |
8978 | u32 off = insn[1].imm; | |
8979 | ||
8980 | if (off >= BPF_MAX_VAR_OFF) { | |
8981 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
8982 | fdput(f); | |
8983 | return -EINVAL; | |
8984 | } | |
8985 | ||
8986 | if (!map->ops->map_direct_value_addr) { | |
8987 | verbose(env, "no direct value access support for this map type\n"); | |
8988 | fdput(f); | |
8989 | return -EINVAL; | |
8990 | } | |
8991 | ||
8992 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
8993 | if (err) { | |
8994 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
8995 | map->value_size, off); | |
8996 | fdput(f); | |
8997 | return err; | |
8998 | } | |
8999 | ||
9000 | aux->map_off = off; | |
9001 | addr += off; | |
9002 | } | |
9003 | ||
9004 | insn[0].imm = (u32)addr; | |
9005 | insn[1].imm = addr >> 32; | |
0246e64d AS |
9006 | |
9007 | /* check whether we recorded this map already */ | |
d8eca5bb | 9008 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 9009 | if (env->used_maps[j] == map) { |
d8eca5bb | 9010 | aux->map_index = j; |
0246e64d AS |
9011 | fdput(f); |
9012 | goto next_insn; | |
9013 | } | |
d8eca5bb | 9014 | } |
0246e64d AS |
9015 | |
9016 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
9017 | fdput(f); | |
9018 | return -E2BIG; | |
9019 | } | |
9020 | ||
0246e64d AS |
9021 | /* hold the map. If the program is rejected by verifier, |
9022 | * the map will be released by release_maps() or it | |
9023 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 9024 | * and all maps are released in free_used_maps() |
0246e64d | 9025 | */ |
1e0bd5a0 | 9026 | bpf_map_inc(map); |
d8eca5bb DB |
9027 | |
9028 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
9029 | env->used_maps[env->used_map_cnt++] = map; |
9030 | ||
b741f163 | 9031 | if (bpf_map_is_cgroup_storage(map) && |
e4730423 | 9032 | bpf_cgroup_storage_assign(env->prog->aux, map)) { |
b741f163 | 9033 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
9034 | fdput(f); |
9035 | return -EBUSY; | |
9036 | } | |
9037 | ||
0246e64d AS |
9038 | fdput(f); |
9039 | next_insn: | |
9040 | insn++; | |
9041 | i++; | |
5e581dad DB |
9042 | continue; |
9043 | } | |
9044 | ||
9045 | /* Basic sanity check before we invest more work here. */ | |
9046 | if (!bpf_opcode_in_insntable(insn->code)) { | |
9047 | verbose(env, "unknown opcode %02x\n", insn->code); | |
9048 | return -EINVAL; | |
0246e64d AS |
9049 | } |
9050 | } | |
9051 | ||
9052 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
9053 | * 'struct bpf_map *' into a register instead of user map_fd. | |
9054 | * These pointers will be used later by verifier to validate map access. | |
9055 | */ | |
9056 | return 0; | |
9057 | } | |
9058 | ||
9059 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 9060 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 9061 | { |
a2ea0746 DB |
9062 | __bpf_free_used_maps(env->prog->aux, env->used_maps, |
9063 | env->used_map_cnt); | |
0246e64d AS |
9064 | } |
9065 | ||
9066 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 9067 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
9068 | { |
9069 | struct bpf_insn *insn = env->prog->insnsi; | |
9070 | int insn_cnt = env->prog->len; | |
9071 | int i; | |
9072 | ||
9073 | for (i = 0; i < insn_cnt; i++, insn++) | |
9074 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
9075 | insn->src_reg = 0; | |
9076 | } | |
9077 | ||
8041902d AS |
9078 | /* single env->prog->insni[off] instruction was replaced with the range |
9079 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
9080 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
9081 | */ | |
b325fbca JW |
9082 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, |
9083 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d AS |
9084 | { |
9085 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
b325fbca JW |
9086 | struct bpf_insn *insn = new_prog->insnsi; |
9087 | u32 prog_len; | |
c131187d | 9088 | int i; |
8041902d | 9089 | |
b325fbca JW |
9090 | /* aux info at OFF always needs adjustment, no matter fast path |
9091 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
9092 | * original insn at old prog. | |
9093 | */ | |
9094 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
9095 | ||
8041902d AS |
9096 | if (cnt == 1) |
9097 | return 0; | |
b325fbca | 9098 | prog_len = new_prog->len; |
fad953ce KC |
9099 | new_data = vzalloc(array_size(prog_len, |
9100 | sizeof(struct bpf_insn_aux_data))); | |
8041902d AS |
9101 | if (!new_data) |
9102 | return -ENOMEM; | |
9103 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
9104 | memcpy(new_data + off + cnt - 1, old_data + off, | |
9105 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 9106 | for (i = off; i < off + cnt - 1; i++) { |
51c39bb1 | 9107 | new_data[i].seen = env->pass_cnt; |
b325fbca JW |
9108 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
9109 | } | |
8041902d AS |
9110 | env->insn_aux_data = new_data; |
9111 | vfree(old_data); | |
9112 | return 0; | |
9113 | } | |
9114 | ||
cc8b0b92 AS |
9115 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
9116 | { | |
9117 | int i; | |
9118 | ||
9119 | if (len == 1) | |
9120 | return; | |
4cb3d99c JW |
9121 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
9122 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 9123 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 9124 | continue; |
9c8105bd | 9125 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
9126 | } |
9127 | } | |
9128 | ||
8041902d AS |
9129 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
9130 | const struct bpf_insn *patch, u32 len) | |
9131 | { | |
9132 | struct bpf_prog *new_prog; | |
9133 | ||
9134 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
9135 | if (IS_ERR(new_prog)) { |
9136 | if (PTR_ERR(new_prog) == -ERANGE) | |
9137 | verbose(env, | |
9138 | "insn %d cannot be patched due to 16-bit range\n", | |
9139 | env->insn_aux_data[off].orig_idx); | |
8041902d | 9140 | return NULL; |
4f73379e | 9141 | } |
b325fbca | 9142 | if (adjust_insn_aux_data(env, new_prog, off, len)) |
8041902d | 9143 | return NULL; |
cc8b0b92 | 9144 | adjust_subprog_starts(env, off, len); |
8041902d AS |
9145 | return new_prog; |
9146 | } | |
9147 | ||
52875a04 JK |
9148 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
9149 | u32 off, u32 cnt) | |
9150 | { | |
9151 | int i, j; | |
9152 | ||
9153 | /* find first prog starting at or after off (first to remove) */ | |
9154 | for (i = 0; i < env->subprog_cnt; i++) | |
9155 | if (env->subprog_info[i].start >= off) | |
9156 | break; | |
9157 | /* find first prog starting at or after off + cnt (first to stay) */ | |
9158 | for (j = i; j < env->subprog_cnt; j++) | |
9159 | if (env->subprog_info[j].start >= off + cnt) | |
9160 | break; | |
9161 | /* if j doesn't start exactly at off + cnt, we are just removing | |
9162 | * the front of previous prog | |
9163 | */ | |
9164 | if (env->subprog_info[j].start != off + cnt) | |
9165 | j--; | |
9166 | ||
9167 | if (j > i) { | |
9168 | struct bpf_prog_aux *aux = env->prog->aux; | |
9169 | int move; | |
9170 | ||
9171 | /* move fake 'exit' subprog as well */ | |
9172 | move = env->subprog_cnt + 1 - j; | |
9173 | ||
9174 | memmove(env->subprog_info + i, | |
9175 | env->subprog_info + j, | |
9176 | sizeof(*env->subprog_info) * move); | |
9177 | env->subprog_cnt -= j - i; | |
9178 | ||
9179 | /* remove func_info */ | |
9180 | if (aux->func_info) { | |
9181 | move = aux->func_info_cnt - j; | |
9182 | ||
9183 | memmove(aux->func_info + i, | |
9184 | aux->func_info + j, | |
9185 | sizeof(*aux->func_info) * move); | |
9186 | aux->func_info_cnt -= j - i; | |
9187 | /* func_info->insn_off is set after all code rewrites, | |
9188 | * in adjust_btf_func() - no need to adjust | |
9189 | */ | |
9190 | } | |
9191 | } else { | |
9192 | /* convert i from "first prog to remove" to "first to adjust" */ | |
9193 | if (env->subprog_info[i].start == off) | |
9194 | i++; | |
9195 | } | |
9196 | ||
9197 | /* update fake 'exit' subprog as well */ | |
9198 | for (; i <= env->subprog_cnt; i++) | |
9199 | env->subprog_info[i].start -= cnt; | |
9200 | ||
9201 | return 0; | |
9202 | } | |
9203 | ||
9204 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
9205 | u32 cnt) | |
9206 | { | |
9207 | struct bpf_prog *prog = env->prog; | |
9208 | u32 i, l_off, l_cnt, nr_linfo; | |
9209 | struct bpf_line_info *linfo; | |
9210 | ||
9211 | nr_linfo = prog->aux->nr_linfo; | |
9212 | if (!nr_linfo) | |
9213 | return 0; | |
9214 | ||
9215 | linfo = prog->aux->linfo; | |
9216 | ||
9217 | /* find first line info to remove, count lines to be removed */ | |
9218 | for (i = 0; i < nr_linfo; i++) | |
9219 | if (linfo[i].insn_off >= off) | |
9220 | break; | |
9221 | ||
9222 | l_off = i; | |
9223 | l_cnt = 0; | |
9224 | for (; i < nr_linfo; i++) | |
9225 | if (linfo[i].insn_off < off + cnt) | |
9226 | l_cnt++; | |
9227 | else | |
9228 | break; | |
9229 | ||
9230 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
9231 | * last removed linfo. prog is already modified, so prog->len == off | |
9232 | * means no live instructions after (tail of the program was removed). | |
9233 | */ | |
9234 | if (prog->len != off && l_cnt && | |
9235 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
9236 | l_cnt--; | |
9237 | linfo[--i].insn_off = off + cnt; | |
9238 | } | |
9239 | ||
9240 | /* remove the line info which refer to the removed instructions */ | |
9241 | if (l_cnt) { | |
9242 | memmove(linfo + l_off, linfo + i, | |
9243 | sizeof(*linfo) * (nr_linfo - i)); | |
9244 | ||
9245 | prog->aux->nr_linfo -= l_cnt; | |
9246 | nr_linfo = prog->aux->nr_linfo; | |
9247 | } | |
9248 | ||
9249 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
9250 | for (i = l_off; i < nr_linfo; i++) | |
9251 | linfo[i].insn_off -= cnt; | |
9252 | ||
9253 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
9254 | for (i = 0; i <= env->subprog_cnt; i++) | |
9255 | if (env->subprog_info[i].linfo_idx > l_off) { | |
9256 | /* program may have started in the removed region but | |
9257 | * may not be fully removed | |
9258 | */ | |
9259 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
9260 | env->subprog_info[i].linfo_idx -= l_cnt; | |
9261 | else | |
9262 | env->subprog_info[i].linfo_idx = l_off; | |
9263 | } | |
9264 | ||
9265 | return 0; | |
9266 | } | |
9267 | ||
9268 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
9269 | { | |
9270 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
9271 | unsigned int orig_prog_len = env->prog->len; | |
9272 | int err; | |
9273 | ||
08ca90af JK |
9274 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9275 | bpf_prog_offload_remove_insns(env, off, cnt); | |
9276 | ||
52875a04 JK |
9277 | err = bpf_remove_insns(env->prog, off, cnt); |
9278 | if (err) | |
9279 | return err; | |
9280 | ||
9281 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
9282 | if (err) | |
9283 | return err; | |
9284 | ||
9285 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
9286 | if (err) | |
9287 | return err; | |
9288 | ||
9289 | memmove(aux_data + off, aux_data + off + cnt, | |
9290 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
9291 | ||
9292 | return 0; | |
9293 | } | |
9294 | ||
2a5418a1 DB |
9295 | /* The verifier does more data flow analysis than llvm and will not |
9296 | * explore branches that are dead at run time. Malicious programs can | |
9297 | * have dead code too. Therefore replace all dead at-run-time code | |
9298 | * with 'ja -1'. | |
9299 | * | |
9300 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
9301 | * program and through another bug we would manage to jump there, then | |
9302 | * we'd execute beyond program memory otherwise. Returning exception | |
9303 | * code also wouldn't work since we can have subprogs where the dead | |
9304 | * code could be located. | |
c131187d AS |
9305 | */ |
9306 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
9307 | { | |
9308 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 9309 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
9310 | struct bpf_insn *insn = env->prog->insnsi; |
9311 | const int insn_cnt = env->prog->len; | |
9312 | int i; | |
9313 | ||
9314 | for (i = 0; i < insn_cnt; i++) { | |
9315 | if (aux_data[i].seen) | |
9316 | continue; | |
2a5418a1 | 9317 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
9318 | } |
9319 | } | |
9320 | ||
e2ae4ca2 JK |
9321 | static bool insn_is_cond_jump(u8 code) |
9322 | { | |
9323 | u8 op; | |
9324 | ||
092ed096 JW |
9325 | if (BPF_CLASS(code) == BPF_JMP32) |
9326 | return true; | |
9327 | ||
e2ae4ca2 JK |
9328 | if (BPF_CLASS(code) != BPF_JMP) |
9329 | return false; | |
9330 | ||
9331 | op = BPF_OP(code); | |
9332 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
9333 | } | |
9334 | ||
9335 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
9336 | { | |
9337 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
9338 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
9339 | struct bpf_insn *insn = env->prog->insnsi; | |
9340 | const int insn_cnt = env->prog->len; | |
9341 | int i; | |
9342 | ||
9343 | for (i = 0; i < insn_cnt; i++, insn++) { | |
9344 | if (!insn_is_cond_jump(insn->code)) | |
9345 | continue; | |
9346 | ||
9347 | if (!aux_data[i + 1].seen) | |
9348 | ja.off = insn->off; | |
9349 | else if (!aux_data[i + 1 + insn->off].seen) | |
9350 | ja.off = 0; | |
9351 | else | |
9352 | continue; | |
9353 | ||
08ca90af JK |
9354 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9355 | bpf_prog_offload_replace_insn(env, i, &ja); | |
9356 | ||
e2ae4ca2 JK |
9357 | memcpy(insn, &ja, sizeof(ja)); |
9358 | } | |
9359 | } | |
9360 | ||
52875a04 JK |
9361 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
9362 | { | |
9363 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
9364 | int insn_cnt = env->prog->len; | |
9365 | int i, err; | |
9366 | ||
9367 | for (i = 0; i < insn_cnt; i++) { | |
9368 | int j; | |
9369 | ||
9370 | j = 0; | |
9371 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
9372 | j++; | |
9373 | if (!j) | |
9374 | continue; | |
9375 | ||
9376 | err = verifier_remove_insns(env, i, j); | |
9377 | if (err) | |
9378 | return err; | |
9379 | insn_cnt = env->prog->len; | |
9380 | } | |
9381 | ||
9382 | return 0; | |
9383 | } | |
9384 | ||
a1b14abc JK |
9385 | static int opt_remove_nops(struct bpf_verifier_env *env) |
9386 | { | |
9387 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
9388 | struct bpf_insn *insn = env->prog->insnsi; | |
9389 | int insn_cnt = env->prog->len; | |
9390 | int i, err; | |
9391 | ||
9392 | for (i = 0; i < insn_cnt; i++) { | |
9393 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
9394 | continue; | |
9395 | ||
9396 | err = verifier_remove_insns(env, i, 1); | |
9397 | if (err) | |
9398 | return err; | |
9399 | insn_cnt--; | |
9400 | i--; | |
9401 | } | |
9402 | ||
9403 | return 0; | |
9404 | } | |
9405 | ||
d6c2308c JW |
9406 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
9407 | const union bpf_attr *attr) | |
a4b1d3c1 | 9408 | { |
d6c2308c | 9409 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 9410 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 9411 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 9412 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 9413 | struct bpf_prog *new_prog; |
d6c2308c | 9414 | bool rnd_hi32; |
a4b1d3c1 | 9415 | |
d6c2308c | 9416 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 9417 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
9418 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
9419 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
9420 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
9421 | for (i = 0; i < len; i++) { |
9422 | int adj_idx = i + delta; | |
9423 | struct bpf_insn insn; | |
9424 | ||
d6c2308c JW |
9425 | insn = insns[adj_idx]; |
9426 | if (!aux[adj_idx].zext_dst) { | |
9427 | u8 code, class; | |
9428 | u32 imm_rnd; | |
9429 | ||
9430 | if (!rnd_hi32) | |
9431 | continue; | |
9432 | ||
9433 | code = insn.code; | |
9434 | class = BPF_CLASS(code); | |
9435 | if (insn_no_def(&insn)) | |
9436 | continue; | |
9437 | ||
9438 | /* NOTE: arg "reg" (the fourth one) is only used for | |
9439 | * BPF_STX which has been ruled out in above | |
9440 | * check, it is safe to pass NULL here. | |
9441 | */ | |
9442 | if (is_reg64(env, &insn, insn.dst_reg, NULL, DST_OP)) { | |
9443 | if (class == BPF_LD && | |
9444 | BPF_MODE(code) == BPF_IMM) | |
9445 | i++; | |
9446 | continue; | |
9447 | } | |
9448 | ||
9449 | /* ctx load could be transformed into wider load. */ | |
9450 | if (class == BPF_LDX && | |
9451 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
9452 | continue; | |
9453 | ||
9454 | imm_rnd = get_random_int(); | |
9455 | rnd_hi32_patch[0] = insn; | |
9456 | rnd_hi32_patch[1].imm = imm_rnd; | |
9457 | rnd_hi32_patch[3].dst_reg = insn.dst_reg; | |
9458 | patch = rnd_hi32_patch; | |
9459 | patch_len = 4; | |
9460 | goto apply_patch_buffer; | |
9461 | } | |
9462 | ||
9463 | if (!bpf_jit_needs_zext()) | |
a4b1d3c1 JW |
9464 | continue; |
9465 | ||
a4b1d3c1 JW |
9466 | zext_patch[0] = insn; |
9467 | zext_patch[1].dst_reg = insn.dst_reg; | |
9468 | zext_patch[1].src_reg = insn.dst_reg; | |
d6c2308c JW |
9469 | patch = zext_patch; |
9470 | patch_len = 2; | |
9471 | apply_patch_buffer: | |
9472 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
9473 | if (!new_prog) |
9474 | return -ENOMEM; | |
9475 | env->prog = new_prog; | |
9476 | insns = new_prog->insnsi; | |
9477 | aux = env->insn_aux_data; | |
d6c2308c | 9478 | delta += patch_len - 1; |
a4b1d3c1 JW |
9479 | } |
9480 | ||
9481 | return 0; | |
9482 | } | |
9483 | ||
c64b7983 JS |
9484 | /* convert load instructions that access fields of a context type into a |
9485 | * sequence of instructions that access fields of the underlying structure: | |
9486 | * struct __sk_buff -> struct sk_buff | |
9487 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 9488 | */ |
58e2af8b | 9489 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 9490 | { |
00176a34 | 9491 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 9492 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 9493 | const int insn_cnt = env->prog->len; |
36bbef52 | 9494 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 9495 | u32 target_size, size_default, off; |
9bac3d6d | 9496 | struct bpf_prog *new_prog; |
d691f9e8 | 9497 | enum bpf_access_type type; |
f96da094 | 9498 | bool is_narrower_load; |
9bac3d6d | 9499 | |
b09928b9 DB |
9500 | if (ops->gen_prologue || env->seen_direct_write) { |
9501 | if (!ops->gen_prologue) { | |
9502 | verbose(env, "bpf verifier is misconfigured\n"); | |
9503 | return -EINVAL; | |
9504 | } | |
36bbef52 DB |
9505 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
9506 | env->prog); | |
9507 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 9508 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
9509 | return -EINVAL; |
9510 | } else if (cnt) { | |
8041902d | 9511 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
9512 | if (!new_prog) |
9513 | return -ENOMEM; | |
8041902d | 9514 | |
36bbef52 | 9515 | env->prog = new_prog; |
3df126f3 | 9516 | delta += cnt - 1; |
36bbef52 DB |
9517 | } |
9518 | } | |
9519 | ||
c64b7983 | 9520 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
9521 | return 0; |
9522 | ||
3df126f3 | 9523 | insn = env->prog->insnsi + delta; |
36bbef52 | 9524 | |
9bac3d6d | 9525 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 JS |
9526 | bpf_convert_ctx_access_t convert_ctx_access; |
9527 | ||
62c7989b DB |
9528 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
9529 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
9530 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 9531 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 9532 | type = BPF_READ; |
62c7989b DB |
9533 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
9534 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
9535 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 9536 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
9537 | type = BPF_WRITE; |
9538 | else | |
9bac3d6d AS |
9539 | continue; |
9540 | ||
af86ca4e AS |
9541 | if (type == BPF_WRITE && |
9542 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
9543 | struct bpf_insn patch[] = { | |
9544 | /* Sanitize suspicious stack slot with zero. | |
9545 | * There are no memory dependencies for this store, | |
9546 | * since it's only using frame pointer and immediate | |
9547 | * constant of zero | |
9548 | */ | |
9549 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
9550 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
9551 | 0), | |
9552 | /* the original STX instruction will immediately | |
9553 | * overwrite the same stack slot with appropriate value | |
9554 | */ | |
9555 | *insn, | |
9556 | }; | |
9557 | ||
9558 | cnt = ARRAY_SIZE(patch); | |
9559 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
9560 | if (!new_prog) | |
9561 | return -ENOMEM; | |
9562 | ||
9563 | delta += cnt - 1; | |
9564 | env->prog = new_prog; | |
9565 | insn = new_prog->insnsi + i + delta; | |
9566 | continue; | |
9567 | } | |
9568 | ||
c64b7983 JS |
9569 | switch (env->insn_aux_data[i + delta].ptr_type) { |
9570 | case PTR_TO_CTX: | |
9571 | if (!ops->convert_ctx_access) | |
9572 | continue; | |
9573 | convert_ctx_access = ops->convert_ctx_access; | |
9574 | break; | |
9575 | case PTR_TO_SOCKET: | |
46f8bc92 | 9576 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
9577 | convert_ctx_access = bpf_sock_convert_ctx_access; |
9578 | break; | |
655a51e5 MKL |
9579 | case PTR_TO_TCP_SOCK: |
9580 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
9581 | break; | |
fada7fdc JL |
9582 | case PTR_TO_XDP_SOCK: |
9583 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
9584 | break; | |
2a02759e | 9585 | case PTR_TO_BTF_ID: |
27ae7997 MKL |
9586 | if (type == BPF_READ) { |
9587 | insn->code = BPF_LDX | BPF_PROBE_MEM | | |
9588 | BPF_SIZE((insn)->code); | |
9589 | env->prog->aux->num_exentries++; | |
9590 | } else if (env->prog->type != BPF_PROG_TYPE_STRUCT_OPS) { | |
2a02759e AS |
9591 | verbose(env, "Writes through BTF pointers are not allowed\n"); |
9592 | return -EINVAL; | |
9593 | } | |
2a02759e | 9594 | continue; |
c64b7983 | 9595 | default: |
9bac3d6d | 9596 | continue; |
c64b7983 | 9597 | } |
9bac3d6d | 9598 | |
31fd8581 | 9599 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 9600 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
9601 | |
9602 | /* If the read access is a narrower load of the field, | |
9603 | * convert to a 4/8-byte load, to minimum program type specific | |
9604 | * convert_ctx_access changes. If conversion is successful, | |
9605 | * we will apply proper mask to the result. | |
9606 | */ | |
f96da094 | 9607 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
9608 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
9609 | off = insn->off; | |
31fd8581 | 9610 | if (is_narrower_load) { |
f96da094 DB |
9611 | u8 size_code; |
9612 | ||
9613 | if (type == BPF_WRITE) { | |
61bd5218 | 9614 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
9615 | return -EINVAL; |
9616 | } | |
31fd8581 | 9617 | |
f96da094 | 9618 | size_code = BPF_H; |
31fd8581 YS |
9619 | if (ctx_field_size == 4) |
9620 | size_code = BPF_W; | |
9621 | else if (ctx_field_size == 8) | |
9622 | size_code = BPF_DW; | |
f96da094 | 9623 | |
bc23105c | 9624 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
9625 | insn->code = BPF_LDX | BPF_MEM | size_code; |
9626 | } | |
f96da094 DB |
9627 | |
9628 | target_size = 0; | |
c64b7983 JS |
9629 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
9630 | &target_size); | |
f96da094 DB |
9631 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
9632 | (ctx_field_size && !target_size)) { | |
61bd5218 | 9633 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
9634 | return -EINVAL; |
9635 | } | |
f96da094 DB |
9636 | |
9637 | if (is_narrower_load && size < target_size) { | |
d895a0f1 IL |
9638 | u8 shift = bpf_ctx_narrow_access_offset( |
9639 | off, size, size_default) * 8; | |
46f53a65 AI |
9640 | if (ctx_field_size <= 4) { |
9641 | if (shift) | |
9642 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
9643 | insn->dst_reg, | |
9644 | shift); | |
31fd8581 | 9645 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 9646 | (1 << size * 8) - 1); |
46f53a65 AI |
9647 | } else { |
9648 | if (shift) | |
9649 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
9650 | insn->dst_reg, | |
9651 | shift); | |
31fd8581 | 9652 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 9653 | (1ULL << size * 8) - 1); |
46f53a65 | 9654 | } |
31fd8581 | 9655 | } |
9bac3d6d | 9656 | |
8041902d | 9657 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
9658 | if (!new_prog) |
9659 | return -ENOMEM; | |
9660 | ||
3df126f3 | 9661 | delta += cnt - 1; |
9bac3d6d AS |
9662 | |
9663 | /* keep walking new program and skip insns we just inserted */ | |
9664 | env->prog = new_prog; | |
3df126f3 | 9665 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
9666 | } |
9667 | ||
9668 | return 0; | |
9669 | } | |
9670 | ||
1c2a088a AS |
9671 | static int jit_subprogs(struct bpf_verifier_env *env) |
9672 | { | |
9673 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
9674 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
7105e828 | 9675 | struct bpf_insn *insn; |
1c2a088a | 9676 | void *old_bpf_func; |
c454a46b | 9677 | int err; |
1c2a088a | 9678 | |
f910cefa | 9679 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
9680 | return 0; |
9681 | ||
7105e828 | 9682 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
1c2a088a AS |
9683 | if (insn->code != (BPF_JMP | BPF_CALL) || |
9684 | insn->src_reg != BPF_PSEUDO_CALL) | |
9685 | continue; | |
c7a89784 DB |
9686 | /* Upon error here we cannot fall back to interpreter but |
9687 | * need a hard reject of the program. Thus -EFAULT is | |
9688 | * propagated in any case. | |
9689 | */ | |
1c2a088a AS |
9690 | subprog = find_subprog(env, i + insn->imm + 1); |
9691 | if (subprog < 0) { | |
9692 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
9693 | i + insn->imm + 1); | |
9694 | return -EFAULT; | |
9695 | } | |
9696 | /* temporarily remember subprog id inside insn instead of | |
9697 | * aux_data, since next loop will split up all insns into funcs | |
9698 | */ | |
f910cefa | 9699 | insn->off = subprog; |
1c2a088a AS |
9700 | /* remember original imm in case JIT fails and fallback |
9701 | * to interpreter will be needed | |
9702 | */ | |
9703 | env->insn_aux_data[i].call_imm = insn->imm; | |
9704 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
9705 | insn->imm = 1; | |
9706 | } | |
9707 | ||
c454a46b MKL |
9708 | err = bpf_prog_alloc_jited_linfo(prog); |
9709 | if (err) | |
9710 | goto out_undo_insn; | |
9711 | ||
9712 | err = -ENOMEM; | |
6396bb22 | 9713 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 9714 | if (!func) |
c7a89784 | 9715 | goto out_undo_insn; |
1c2a088a | 9716 | |
f910cefa | 9717 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 9718 | subprog_start = subprog_end; |
4cb3d99c | 9719 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
9720 | |
9721 | len = subprog_end - subprog_start; | |
492ecee8 AS |
9722 | /* BPF_PROG_RUN doesn't call subprogs directly, |
9723 | * hence main prog stats include the runtime of subprogs. | |
9724 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
9725 | * func[i]->aux->stats will never be accessed and stays NULL | |
9726 | */ | |
9727 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
9728 | if (!func[i]) |
9729 | goto out_free; | |
9730 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
9731 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 9732 | func[i]->type = prog->type; |
1c2a088a | 9733 | func[i]->len = len; |
4f74d809 DB |
9734 | if (bpf_prog_calc_tag(func[i])) |
9735 | goto out_free; | |
1c2a088a | 9736 | func[i]->is_func = 1; |
ba64e7d8 YS |
9737 | func[i]->aux->func_idx = i; |
9738 | /* the btf and func_info will be freed only at prog->aux */ | |
9739 | func[i]->aux->btf = prog->aux->btf; | |
9740 | func[i]->aux->func_info = prog->aux->func_info; | |
9741 | ||
1c2a088a AS |
9742 | /* Use bpf_prog_F_tag to indicate functions in stack traces. |
9743 | * Long term would need debug info to populate names | |
9744 | */ | |
9745 | func[i]->aux->name[0] = 'F'; | |
9c8105bd | 9746 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 9747 | func[i]->jit_requested = 1; |
c454a46b MKL |
9748 | func[i]->aux->linfo = prog->aux->linfo; |
9749 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
9750 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
9751 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
1c2a088a AS |
9752 | func[i] = bpf_int_jit_compile(func[i]); |
9753 | if (!func[i]->jited) { | |
9754 | err = -ENOTSUPP; | |
9755 | goto out_free; | |
9756 | } | |
9757 | cond_resched(); | |
9758 | } | |
9759 | /* at this point all bpf functions were successfully JITed | |
9760 | * now populate all bpf_calls with correct addresses and | |
9761 | * run last pass of JIT | |
9762 | */ | |
f910cefa | 9763 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
9764 | insn = func[i]->insnsi; |
9765 | for (j = 0; j < func[i]->len; j++, insn++) { | |
9766 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
9767 | insn->src_reg != BPF_PSEUDO_CALL) | |
9768 | continue; | |
9769 | subprog = insn->off; | |
0d306c31 PB |
9770 | insn->imm = BPF_CAST_CALL(func[subprog]->bpf_func) - |
9771 | __bpf_call_base; | |
1c2a088a | 9772 | } |
2162fed4 SD |
9773 | |
9774 | /* we use the aux data to keep a list of the start addresses | |
9775 | * of the JITed images for each function in the program | |
9776 | * | |
9777 | * for some architectures, such as powerpc64, the imm field | |
9778 | * might not be large enough to hold the offset of the start | |
9779 | * address of the callee's JITed image from __bpf_call_base | |
9780 | * | |
9781 | * in such cases, we can lookup the start address of a callee | |
9782 | * by using its subprog id, available from the off field of | |
9783 | * the call instruction, as an index for this list | |
9784 | */ | |
9785 | func[i]->aux->func = func; | |
9786 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 9787 | } |
f910cefa | 9788 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
9789 | old_bpf_func = func[i]->bpf_func; |
9790 | tmp = bpf_int_jit_compile(func[i]); | |
9791 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
9792 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 9793 | err = -ENOTSUPP; |
1c2a088a AS |
9794 | goto out_free; |
9795 | } | |
9796 | cond_resched(); | |
9797 | } | |
9798 | ||
9799 | /* finally lock prog and jit images for all functions and | |
9800 | * populate kallsysm | |
9801 | */ | |
f910cefa | 9802 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
9803 | bpf_prog_lock_ro(func[i]); |
9804 | bpf_prog_kallsyms_add(func[i]); | |
9805 | } | |
7105e828 DB |
9806 | |
9807 | /* Last step: make now unused interpreter insns from main | |
9808 | * prog consistent for later dump requests, so they can | |
9809 | * later look the same as if they were interpreted only. | |
9810 | */ | |
9811 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
7105e828 DB |
9812 | if (insn->code != (BPF_JMP | BPF_CALL) || |
9813 | insn->src_reg != BPF_PSEUDO_CALL) | |
9814 | continue; | |
9815 | insn->off = env->insn_aux_data[i].call_imm; | |
9816 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 9817 | insn->imm = subprog; |
7105e828 DB |
9818 | } |
9819 | ||
1c2a088a AS |
9820 | prog->jited = 1; |
9821 | prog->bpf_func = func[0]->bpf_func; | |
9822 | prog->aux->func = func; | |
f910cefa | 9823 | prog->aux->func_cnt = env->subprog_cnt; |
c454a46b | 9824 | bpf_prog_free_unused_jited_linfo(prog); |
1c2a088a AS |
9825 | return 0; |
9826 | out_free: | |
f910cefa | 9827 | for (i = 0; i < env->subprog_cnt; i++) |
1c2a088a AS |
9828 | if (func[i]) |
9829 | bpf_jit_free(func[i]); | |
9830 | kfree(func); | |
c7a89784 | 9831 | out_undo_insn: |
1c2a088a AS |
9832 | /* cleanup main prog to be interpreted */ |
9833 | prog->jit_requested = 0; | |
9834 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
9835 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
9836 | insn->src_reg != BPF_PSEUDO_CALL) | |
9837 | continue; | |
9838 | insn->off = 0; | |
9839 | insn->imm = env->insn_aux_data[i].call_imm; | |
9840 | } | |
c454a46b | 9841 | bpf_prog_free_jited_linfo(prog); |
1c2a088a AS |
9842 | return err; |
9843 | } | |
9844 | ||
1ea47e01 AS |
9845 | static int fixup_call_args(struct bpf_verifier_env *env) |
9846 | { | |
19d28fbd | 9847 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
9848 | struct bpf_prog *prog = env->prog; |
9849 | struct bpf_insn *insn = prog->insnsi; | |
9850 | int i, depth; | |
19d28fbd | 9851 | #endif |
e4052d06 | 9852 | int err = 0; |
1ea47e01 | 9853 | |
e4052d06 QM |
9854 | if (env->prog->jit_requested && |
9855 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
9856 | err = jit_subprogs(env); |
9857 | if (err == 0) | |
1c2a088a | 9858 | return 0; |
c7a89784 DB |
9859 | if (err == -EFAULT) |
9860 | return err; | |
19d28fbd DM |
9861 | } |
9862 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1ea47e01 AS |
9863 | for (i = 0; i < prog->len; i++, insn++) { |
9864 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
9865 | insn->src_reg != BPF_PSEUDO_CALL) | |
9866 | continue; | |
9867 | depth = get_callee_stack_depth(env, insn, i); | |
9868 | if (depth < 0) | |
9869 | return depth; | |
9870 | bpf_patch_call_args(insn, depth); | |
9871 | } | |
19d28fbd DM |
9872 | err = 0; |
9873 | #endif | |
9874 | return err; | |
1ea47e01 AS |
9875 | } |
9876 | ||
79741b3b | 9877 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 9878 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
9879 | * |
9880 | * this function is called after eBPF program passed verification | |
9881 | */ | |
79741b3b | 9882 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 9883 | { |
79741b3b | 9884 | struct bpf_prog *prog = env->prog; |
d2e4c1e6 | 9885 | bool expect_blinding = bpf_jit_blinding_enabled(prog); |
79741b3b | 9886 | struct bpf_insn *insn = prog->insnsi; |
e245c5c6 | 9887 | const struct bpf_func_proto *fn; |
79741b3b | 9888 | const int insn_cnt = prog->len; |
09772d92 | 9889 | const struct bpf_map_ops *ops; |
c93552c4 | 9890 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
9891 | struct bpf_insn insn_buf[16]; |
9892 | struct bpf_prog *new_prog; | |
9893 | struct bpf_map *map_ptr; | |
d2e4c1e6 | 9894 | int i, ret, cnt, delta = 0; |
e245c5c6 | 9895 | |
79741b3b | 9896 | for (i = 0; i < insn_cnt; i++, insn++) { |
f6b1b3bf DB |
9897 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
9898 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
9899 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 9900 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf DB |
9901 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
9902 | struct bpf_insn mask_and_div[] = { | |
9903 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
9904 | /* Rx div 0 -> 0 */ | |
9905 | BPF_JMP_IMM(BPF_JNE, insn->src_reg, 0, 2), | |
9906 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), | |
9907 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
9908 | *insn, | |
9909 | }; | |
9910 | struct bpf_insn mask_and_mod[] = { | |
9911 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
9912 | /* Rx mod 0 -> Rx */ | |
9913 | BPF_JMP_IMM(BPF_JEQ, insn->src_reg, 0, 1), | |
9914 | *insn, | |
9915 | }; | |
9916 | struct bpf_insn *patchlet; | |
9917 | ||
9918 | if (insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
9919 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
9920 | patchlet = mask_and_div + (is64 ? 1 : 0); | |
9921 | cnt = ARRAY_SIZE(mask_and_div) - (is64 ? 1 : 0); | |
9922 | } else { | |
9923 | patchlet = mask_and_mod + (is64 ? 1 : 0); | |
9924 | cnt = ARRAY_SIZE(mask_and_mod) - (is64 ? 1 : 0); | |
9925 | } | |
9926 | ||
9927 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
9928 | if (!new_prog) |
9929 | return -ENOMEM; | |
9930 | ||
9931 | delta += cnt - 1; | |
9932 | env->prog = prog = new_prog; | |
9933 | insn = new_prog->insnsi + i + delta; | |
9934 | continue; | |
9935 | } | |
9936 | ||
e0cea7ce DB |
9937 | if (BPF_CLASS(insn->code) == BPF_LD && |
9938 | (BPF_MODE(insn->code) == BPF_ABS || | |
9939 | BPF_MODE(insn->code) == BPF_IND)) { | |
9940 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
9941 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
9942 | verbose(env, "bpf verifier is misconfigured\n"); | |
9943 | return -EINVAL; | |
9944 | } | |
9945 | ||
9946 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
9947 | if (!new_prog) | |
9948 | return -ENOMEM; | |
9949 | ||
9950 | delta += cnt - 1; | |
9951 | env->prog = prog = new_prog; | |
9952 | insn = new_prog->insnsi + i + delta; | |
9953 | continue; | |
9954 | } | |
9955 | ||
979d63d5 DB |
9956 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
9957 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
9958 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
9959 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
9960 | struct bpf_insn insn_buf[16]; | |
9961 | struct bpf_insn *patch = &insn_buf[0]; | |
9962 | bool issrc, isneg; | |
9963 | u32 off_reg; | |
9964 | ||
9965 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
9966 | if (!aux->alu_state || |
9967 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
9968 | continue; |
9969 | ||
9970 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
9971 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
9972 | BPF_ALU_SANITIZE_SRC; | |
9973 | ||
9974 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
9975 | if (isneg) | |
9976 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
9977 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit - 1); | |
9978 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
9979 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
9980 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
9981 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
9982 | if (issrc) { | |
9983 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, | |
9984 | off_reg); | |
9985 | insn->src_reg = BPF_REG_AX; | |
9986 | } else { | |
9987 | *patch++ = BPF_ALU64_REG(BPF_AND, off_reg, | |
9988 | BPF_REG_AX); | |
9989 | } | |
9990 | if (isneg) | |
9991 | insn->code = insn->code == code_add ? | |
9992 | code_sub : code_add; | |
9993 | *patch++ = *insn; | |
9994 | if (issrc && isneg) | |
9995 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
9996 | cnt = patch - insn_buf; | |
9997 | ||
9998 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
9999 | if (!new_prog) | |
10000 | return -ENOMEM; | |
10001 | ||
10002 | delta += cnt - 1; | |
10003 | env->prog = prog = new_prog; | |
10004 | insn = new_prog->insnsi + i + delta; | |
10005 | continue; | |
10006 | } | |
10007 | ||
79741b3b AS |
10008 | if (insn->code != (BPF_JMP | BPF_CALL)) |
10009 | continue; | |
cc8b0b92 AS |
10010 | if (insn->src_reg == BPF_PSEUDO_CALL) |
10011 | continue; | |
e245c5c6 | 10012 | |
79741b3b AS |
10013 | if (insn->imm == BPF_FUNC_get_route_realm) |
10014 | prog->dst_needed = 1; | |
10015 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
10016 | bpf_user_rnd_init_once(); | |
9802d865 JB |
10017 | if (insn->imm == BPF_FUNC_override_return) |
10018 | prog->kprobe_override = 1; | |
79741b3b | 10019 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
10020 | /* If we tail call into other programs, we |
10021 | * cannot make any assumptions since they can | |
10022 | * be replaced dynamically during runtime in | |
10023 | * the program array. | |
10024 | */ | |
10025 | prog->cb_access = 1; | |
80a58d02 | 10026 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
e647815a | 10027 | env->prog->aux->max_pkt_offset = MAX_PACKET_OFF; |
7b9f6da1 | 10028 | |
79741b3b AS |
10029 | /* mark bpf_tail_call as different opcode to avoid |
10030 | * conditional branch in the interpeter for every normal | |
10031 | * call and to prevent accidental JITing by JIT compiler | |
10032 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 10033 | */ |
79741b3b | 10034 | insn->imm = 0; |
71189fa9 | 10035 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 10036 | |
c93552c4 | 10037 | aux = &env->insn_aux_data[i + delta]; |
2c78ee89 | 10038 | if (env->bpf_capable && !expect_blinding && |
cc52d914 | 10039 | prog->jit_requested && |
d2e4c1e6 DB |
10040 | !bpf_map_key_poisoned(aux) && |
10041 | !bpf_map_ptr_poisoned(aux) && | |
10042 | !bpf_map_ptr_unpriv(aux)) { | |
10043 | struct bpf_jit_poke_descriptor desc = { | |
10044 | .reason = BPF_POKE_REASON_TAIL_CALL, | |
10045 | .tail_call.map = BPF_MAP_PTR(aux->map_ptr_state), | |
10046 | .tail_call.key = bpf_map_key_immediate(aux), | |
10047 | }; | |
10048 | ||
10049 | ret = bpf_jit_add_poke_descriptor(prog, &desc); | |
10050 | if (ret < 0) { | |
10051 | verbose(env, "adding tail call poke descriptor failed\n"); | |
10052 | return ret; | |
10053 | } | |
10054 | ||
10055 | insn->imm = ret + 1; | |
10056 | continue; | |
10057 | } | |
10058 | ||
c93552c4 DB |
10059 | if (!bpf_map_ptr_unpriv(aux)) |
10060 | continue; | |
10061 | ||
b2157399 AS |
10062 | /* instead of changing every JIT dealing with tail_call |
10063 | * emit two extra insns: | |
10064 | * if (index >= max_entries) goto out; | |
10065 | * index &= array->index_mask; | |
10066 | * to avoid out-of-bounds cpu speculation | |
10067 | */ | |
c93552c4 | 10068 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 10069 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
10070 | return -EINVAL; |
10071 | } | |
c93552c4 | 10072 | |
d2e4c1e6 | 10073 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
b2157399 AS |
10074 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
10075 | map_ptr->max_entries, 2); | |
10076 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
10077 | container_of(map_ptr, | |
10078 | struct bpf_array, | |
10079 | map)->index_mask); | |
10080 | insn_buf[2] = *insn; | |
10081 | cnt = 3; | |
10082 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
10083 | if (!new_prog) | |
10084 | return -ENOMEM; | |
10085 | ||
10086 | delta += cnt - 1; | |
10087 | env->prog = prog = new_prog; | |
10088 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
10089 | continue; |
10090 | } | |
e245c5c6 | 10091 | |
89c63074 | 10092 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
10093 | * and other inlining handlers are currently limited to 64 bit |
10094 | * only. | |
89c63074 | 10095 | */ |
60b58afc | 10096 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
10097 | (insn->imm == BPF_FUNC_map_lookup_elem || |
10098 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
10099 | insn->imm == BPF_FUNC_map_delete_elem || |
10100 | insn->imm == BPF_FUNC_map_push_elem || | |
10101 | insn->imm == BPF_FUNC_map_pop_elem || | |
10102 | insn->imm == BPF_FUNC_map_peek_elem)) { | |
c93552c4 DB |
10103 | aux = &env->insn_aux_data[i + delta]; |
10104 | if (bpf_map_ptr_poisoned(aux)) | |
10105 | goto patch_call_imm; | |
10106 | ||
d2e4c1e6 | 10107 | map_ptr = BPF_MAP_PTR(aux->map_ptr_state); |
09772d92 DB |
10108 | ops = map_ptr->ops; |
10109 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
10110 | ops->map_gen_lookup) { | |
10111 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
10112 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
10113 | verbose(env, "bpf verifier is misconfigured\n"); | |
10114 | return -EINVAL; | |
10115 | } | |
81ed18ab | 10116 | |
09772d92 DB |
10117 | new_prog = bpf_patch_insn_data(env, i + delta, |
10118 | insn_buf, cnt); | |
10119 | if (!new_prog) | |
10120 | return -ENOMEM; | |
81ed18ab | 10121 | |
09772d92 DB |
10122 | delta += cnt - 1; |
10123 | env->prog = prog = new_prog; | |
10124 | insn = new_prog->insnsi + i + delta; | |
10125 | continue; | |
10126 | } | |
81ed18ab | 10127 | |
09772d92 DB |
10128 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
10129 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
10130 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
10131 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
10132 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
10133 | (int (*)(struct bpf_map *map, void *key, void *value, | |
10134 | u64 flags))NULL)); | |
84430d42 DB |
10135 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
10136 | (int (*)(struct bpf_map *map, void *value, | |
10137 | u64 flags))NULL)); | |
10138 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
10139 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
10140 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
10141 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
10142 | ||
09772d92 DB |
10143 | switch (insn->imm) { |
10144 | case BPF_FUNC_map_lookup_elem: | |
10145 | insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) - | |
10146 | __bpf_call_base; | |
10147 | continue; | |
10148 | case BPF_FUNC_map_update_elem: | |
10149 | insn->imm = BPF_CAST_CALL(ops->map_update_elem) - | |
10150 | __bpf_call_base; | |
10151 | continue; | |
10152 | case BPF_FUNC_map_delete_elem: | |
10153 | insn->imm = BPF_CAST_CALL(ops->map_delete_elem) - | |
10154 | __bpf_call_base; | |
10155 | continue; | |
84430d42 DB |
10156 | case BPF_FUNC_map_push_elem: |
10157 | insn->imm = BPF_CAST_CALL(ops->map_push_elem) - | |
10158 | __bpf_call_base; | |
10159 | continue; | |
10160 | case BPF_FUNC_map_pop_elem: | |
10161 | insn->imm = BPF_CAST_CALL(ops->map_pop_elem) - | |
10162 | __bpf_call_base; | |
10163 | continue; | |
10164 | case BPF_FUNC_map_peek_elem: | |
10165 | insn->imm = BPF_CAST_CALL(ops->map_peek_elem) - | |
10166 | __bpf_call_base; | |
10167 | continue; | |
09772d92 | 10168 | } |
81ed18ab | 10169 | |
09772d92 | 10170 | goto patch_call_imm; |
81ed18ab AS |
10171 | } |
10172 | ||
5576b991 MKL |
10173 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
10174 | insn->imm == BPF_FUNC_jiffies64) { | |
10175 | struct bpf_insn ld_jiffies_addr[2] = { | |
10176 | BPF_LD_IMM64(BPF_REG_0, | |
10177 | (unsigned long)&jiffies), | |
10178 | }; | |
10179 | ||
10180 | insn_buf[0] = ld_jiffies_addr[0]; | |
10181 | insn_buf[1] = ld_jiffies_addr[1]; | |
10182 | insn_buf[2] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, | |
10183 | BPF_REG_0, 0); | |
10184 | cnt = 3; | |
10185 | ||
10186 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
10187 | cnt); | |
10188 | if (!new_prog) | |
10189 | return -ENOMEM; | |
10190 | ||
10191 | delta += cnt - 1; | |
10192 | env->prog = prog = new_prog; | |
10193 | insn = new_prog->insnsi + i + delta; | |
10194 | continue; | |
10195 | } | |
10196 | ||
81ed18ab | 10197 | patch_call_imm: |
5e43f899 | 10198 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
10199 | /* all functions that have prototype and verifier allowed |
10200 | * programs to call them, must be real in-kernel functions | |
10201 | */ | |
10202 | if (!fn->func) { | |
61bd5218 JK |
10203 | verbose(env, |
10204 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
10205 | func_id_name(insn->imm), insn->imm); |
10206 | return -EFAULT; | |
e245c5c6 | 10207 | } |
79741b3b | 10208 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 10209 | } |
e245c5c6 | 10210 | |
d2e4c1e6 DB |
10211 | /* Since poke tab is now finalized, publish aux to tracker. */ |
10212 | for (i = 0; i < prog->aux->size_poke_tab; i++) { | |
10213 | map_ptr = prog->aux->poke_tab[i].tail_call.map; | |
10214 | if (!map_ptr->ops->map_poke_track || | |
10215 | !map_ptr->ops->map_poke_untrack || | |
10216 | !map_ptr->ops->map_poke_run) { | |
10217 | verbose(env, "bpf verifier is misconfigured\n"); | |
10218 | return -EINVAL; | |
10219 | } | |
10220 | ||
10221 | ret = map_ptr->ops->map_poke_track(map_ptr, prog->aux); | |
10222 | if (ret < 0) { | |
10223 | verbose(env, "tracking tail call prog failed\n"); | |
10224 | return ret; | |
10225 | } | |
10226 | } | |
10227 | ||
79741b3b AS |
10228 | return 0; |
10229 | } | |
e245c5c6 | 10230 | |
58e2af8b | 10231 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 10232 | { |
58e2af8b | 10233 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
10234 | int i; |
10235 | ||
9f4686c4 AS |
10236 | sl = env->free_list; |
10237 | while (sl) { | |
10238 | sln = sl->next; | |
10239 | free_verifier_state(&sl->state, false); | |
10240 | kfree(sl); | |
10241 | sl = sln; | |
10242 | } | |
51c39bb1 | 10243 | env->free_list = NULL; |
9f4686c4 | 10244 | |
f1bca824 AS |
10245 | if (!env->explored_states) |
10246 | return; | |
10247 | ||
dc2a4ebc | 10248 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
10249 | sl = env->explored_states[i]; |
10250 | ||
a8f500af AS |
10251 | while (sl) { |
10252 | sln = sl->next; | |
10253 | free_verifier_state(&sl->state, false); | |
10254 | kfree(sl); | |
10255 | sl = sln; | |
10256 | } | |
51c39bb1 | 10257 | env->explored_states[i] = NULL; |
f1bca824 | 10258 | } |
51c39bb1 | 10259 | } |
f1bca824 | 10260 | |
51c39bb1 AS |
10261 | /* The verifier is using insn_aux_data[] to store temporary data during |
10262 | * verification and to store information for passes that run after the | |
10263 | * verification like dead code sanitization. do_check_common() for subprogram N | |
10264 | * may analyze many other subprograms. sanitize_insn_aux_data() clears all | |
10265 | * temporary data after do_check_common() finds that subprogram N cannot be | |
10266 | * verified independently. pass_cnt counts the number of times | |
10267 | * do_check_common() was run and insn->aux->seen tells the pass number | |
10268 | * insn_aux_data was touched. These variables are compared to clear temporary | |
10269 | * data from failed pass. For testing and experiments do_check_common() can be | |
10270 | * run multiple times even when prior attempt to verify is unsuccessful. | |
10271 | */ | |
10272 | static void sanitize_insn_aux_data(struct bpf_verifier_env *env) | |
10273 | { | |
10274 | struct bpf_insn *insn = env->prog->insnsi; | |
10275 | struct bpf_insn_aux_data *aux; | |
10276 | int i, class; | |
10277 | ||
10278 | for (i = 0; i < env->prog->len; i++) { | |
10279 | class = BPF_CLASS(insn[i].code); | |
10280 | if (class != BPF_LDX && class != BPF_STX) | |
10281 | continue; | |
10282 | aux = &env->insn_aux_data[i]; | |
10283 | if (aux->seen != env->pass_cnt) | |
10284 | continue; | |
10285 | memset(aux, 0, offsetof(typeof(*aux), orig_idx)); | |
10286 | } | |
f1bca824 AS |
10287 | } |
10288 | ||
51c39bb1 AS |
10289 | static int do_check_common(struct bpf_verifier_env *env, int subprog) |
10290 | { | |
6f8a57cc | 10291 | bool pop_log = !(env->log.level & BPF_LOG_LEVEL2); |
51c39bb1 AS |
10292 | struct bpf_verifier_state *state; |
10293 | struct bpf_reg_state *regs; | |
10294 | int ret, i; | |
10295 | ||
10296 | env->prev_linfo = NULL; | |
10297 | env->pass_cnt++; | |
10298 | ||
10299 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); | |
10300 | if (!state) | |
10301 | return -ENOMEM; | |
10302 | state->curframe = 0; | |
10303 | state->speculative = false; | |
10304 | state->branches = 1; | |
10305 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); | |
10306 | if (!state->frame[0]) { | |
10307 | kfree(state); | |
10308 | return -ENOMEM; | |
10309 | } | |
10310 | env->cur_state = state; | |
10311 | init_func_state(env, state->frame[0], | |
10312 | BPF_MAIN_FUNC /* callsite */, | |
10313 | 0 /* frameno */, | |
10314 | subprog); | |
10315 | ||
10316 | regs = state->frame[state->curframe]->regs; | |
be8704ff | 10317 | if (subprog || env->prog->type == BPF_PROG_TYPE_EXT) { |
51c39bb1 AS |
10318 | ret = btf_prepare_func_args(env, subprog, regs); |
10319 | if (ret) | |
10320 | goto out; | |
10321 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) { | |
10322 | if (regs[i].type == PTR_TO_CTX) | |
10323 | mark_reg_known_zero(env, regs, i); | |
10324 | else if (regs[i].type == SCALAR_VALUE) | |
10325 | mark_reg_unknown(env, regs, i); | |
10326 | } | |
10327 | } else { | |
10328 | /* 1st arg to a function */ | |
10329 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
10330 | mark_reg_known_zero(env, regs, BPF_REG_1); | |
10331 | ret = btf_check_func_arg_match(env, subprog, regs); | |
10332 | if (ret == -EFAULT) | |
10333 | /* unlikely verifier bug. abort. | |
10334 | * ret == 0 and ret < 0 are sadly acceptable for | |
10335 | * main() function due to backward compatibility. | |
10336 | * Like socket filter program may be written as: | |
10337 | * int bpf_prog(struct pt_regs *ctx) | |
10338 | * and never dereference that ctx in the program. | |
10339 | * 'struct pt_regs' is a type mismatch for socket | |
10340 | * filter that should be using 'struct __sk_buff'. | |
10341 | */ | |
10342 | goto out; | |
10343 | } | |
10344 | ||
10345 | ret = do_check(env); | |
10346 | out: | |
f59bbfc2 AS |
10347 | /* check for NULL is necessary, since cur_state can be freed inside |
10348 | * do_check() under memory pressure. | |
10349 | */ | |
10350 | if (env->cur_state) { | |
10351 | free_verifier_state(env->cur_state, true); | |
10352 | env->cur_state = NULL; | |
10353 | } | |
6f8a57cc AN |
10354 | while (!pop_stack(env, NULL, NULL, false)); |
10355 | if (!ret && pop_log) | |
10356 | bpf_vlog_reset(&env->log, 0); | |
51c39bb1 AS |
10357 | free_states(env); |
10358 | if (ret) | |
10359 | /* clean aux data in case subprog was rejected */ | |
10360 | sanitize_insn_aux_data(env); | |
10361 | return ret; | |
10362 | } | |
10363 | ||
10364 | /* Verify all global functions in a BPF program one by one based on their BTF. | |
10365 | * All global functions must pass verification. Otherwise the whole program is rejected. | |
10366 | * Consider: | |
10367 | * int bar(int); | |
10368 | * int foo(int f) | |
10369 | * { | |
10370 | * return bar(f); | |
10371 | * } | |
10372 | * int bar(int b) | |
10373 | * { | |
10374 | * ... | |
10375 | * } | |
10376 | * foo() will be verified first for R1=any_scalar_value. During verification it | |
10377 | * will be assumed that bar() already verified successfully and call to bar() | |
10378 | * from foo() will be checked for type match only. Later bar() will be verified | |
10379 | * independently to check that it's safe for R1=any_scalar_value. | |
10380 | */ | |
10381 | static int do_check_subprogs(struct bpf_verifier_env *env) | |
10382 | { | |
10383 | struct bpf_prog_aux *aux = env->prog->aux; | |
10384 | int i, ret; | |
10385 | ||
10386 | if (!aux->func_info) | |
10387 | return 0; | |
10388 | ||
10389 | for (i = 1; i < env->subprog_cnt; i++) { | |
10390 | if (aux->func_info_aux[i].linkage != BTF_FUNC_GLOBAL) | |
10391 | continue; | |
10392 | env->insn_idx = env->subprog_info[i].start; | |
10393 | WARN_ON_ONCE(env->insn_idx == 0); | |
10394 | ret = do_check_common(env, i); | |
10395 | if (ret) { | |
10396 | return ret; | |
10397 | } else if (env->log.level & BPF_LOG_LEVEL) { | |
10398 | verbose(env, | |
10399 | "Func#%d is safe for any args that match its prototype\n", | |
10400 | i); | |
10401 | } | |
10402 | } | |
10403 | return 0; | |
10404 | } | |
10405 | ||
10406 | static int do_check_main(struct bpf_verifier_env *env) | |
10407 | { | |
10408 | int ret; | |
10409 | ||
10410 | env->insn_idx = 0; | |
10411 | ret = do_check_common(env, 0); | |
10412 | if (!ret) | |
10413 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; | |
10414 | return ret; | |
10415 | } | |
10416 | ||
10417 | ||
06ee7115 AS |
10418 | static void print_verification_stats(struct bpf_verifier_env *env) |
10419 | { | |
10420 | int i; | |
10421 | ||
10422 | if (env->log.level & BPF_LOG_STATS) { | |
10423 | verbose(env, "verification time %lld usec\n", | |
10424 | div_u64(env->verification_time, 1000)); | |
10425 | verbose(env, "stack depth "); | |
10426 | for (i = 0; i < env->subprog_cnt; i++) { | |
10427 | u32 depth = env->subprog_info[i].stack_depth; | |
10428 | ||
10429 | verbose(env, "%d", depth); | |
10430 | if (i + 1 < env->subprog_cnt) | |
10431 | verbose(env, "+"); | |
10432 | } | |
10433 | verbose(env, "\n"); | |
10434 | } | |
10435 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
10436 | "total_states %d peak_states %d mark_read %d\n", | |
10437 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
10438 | env->max_states_per_insn, env->total_states, | |
10439 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
10440 | } |
10441 | ||
27ae7997 MKL |
10442 | static int check_struct_ops_btf_id(struct bpf_verifier_env *env) |
10443 | { | |
10444 | const struct btf_type *t, *func_proto; | |
10445 | const struct bpf_struct_ops *st_ops; | |
10446 | const struct btf_member *member; | |
10447 | struct bpf_prog *prog = env->prog; | |
10448 | u32 btf_id, member_idx; | |
10449 | const char *mname; | |
10450 | ||
10451 | btf_id = prog->aux->attach_btf_id; | |
10452 | st_ops = bpf_struct_ops_find(btf_id); | |
10453 | if (!st_ops) { | |
10454 | verbose(env, "attach_btf_id %u is not a supported struct\n", | |
10455 | btf_id); | |
10456 | return -ENOTSUPP; | |
10457 | } | |
10458 | ||
10459 | t = st_ops->type; | |
10460 | member_idx = prog->expected_attach_type; | |
10461 | if (member_idx >= btf_type_vlen(t)) { | |
10462 | verbose(env, "attach to invalid member idx %u of struct %s\n", | |
10463 | member_idx, st_ops->name); | |
10464 | return -EINVAL; | |
10465 | } | |
10466 | ||
10467 | member = &btf_type_member(t)[member_idx]; | |
10468 | mname = btf_name_by_offset(btf_vmlinux, member->name_off); | |
10469 | func_proto = btf_type_resolve_func_ptr(btf_vmlinux, member->type, | |
10470 | NULL); | |
10471 | if (!func_proto) { | |
10472 | verbose(env, "attach to invalid member %s(@idx %u) of struct %s\n", | |
10473 | mname, member_idx, st_ops->name); | |
10474 | return -EINVAL; | |
10475 | } | |
10476 | ||
10477 | if (st_ops->check_member) { | |
10478 | int err = st_ops->check_member(t, member); | |
10479 | ||
10480 | if (err) { | |
10481 | verbose(env, "attach to unsupported member %s of struct %s\n", | |
10482 | mname, st_ops->name); | |
10483 | return err; | |
10484 | } | |
10485 | } | |
10486 | ||
10487 | prog->aux->attach_func_proto = func_proto; | |
10488 | prog->aux->attach_func_name = mname; | |
10489 | env->ops = st_ops->verifier_ops; | |
10490 | ||
10491 | return 0; | |
10492 | } | |
6ba43b76 KS |
10493 | #define SECURITY_PREFIX "security_" |
10494 | ||
10495 | static int check_attach_modify_return(struct bpf_verifier_env *env) | |
10496 | { | |
10497 | struct bpf_prog *prog = env->prog; | |
10498 | unsigned long addr = (unsigned long) prog->aux->trampoline->func.addr; | |
10499 | ||
6ba43b76 KS |
10500 | /* This is expected to be cleaned up in the future with the KRSI effort |
10501 | * introducing the LSM_HOOK macro for cleaning up lsm_hooks.h. | |
10502 | */ | |
69191754 KS |
10503 | if (within_error_injection_list(addr) || |
10504 | !strncmp(SECURITY_PREFIX, prog->aux->attach_func_name, | |
10505 | sizeof(SECURITY_PREFIX) - 1)) | |
6ba43b76 | 10506 | return 0; |
6ba43b76 KS |
10507 | |
10508 | verbose(env, "fmod_ret attach_btf_id %u (%s) is not modifiable\n", | |
10509 | prog->aux->attach_btf_id, prog->aux->attach_func_name); | |
10510 | ||
10511 | return -EINVAL; | |
10512 | } | |
27ae7997 | 10513 | |
38207291 MKL |
10514 | static int check_attach_btf_id(struct bpf_verifier_env *env) |
10515 | { | |
10516 | struct bpf_prog *prog = env->prog; | |
be8704ff | 10517 | bool prog_extension = prog->type == BPF_PROG_TYPE_EXT; |
5b92a28a | 10518 | struct bpf_prog *tgt_prog = prog->aux->linked_prog; |
38207291 | 10519 | u32 btf_id = prog->aux->attach_btf_id; |
f1b9509c | 10520 | const char prefix[] = "btf_trace_"; |
15d83c4d | 10521 | struct btf_func_model fmodel; |
5b92a28a | 10522 | int ret = 0, subprog = -1, i; |
fec56f58 | 10523 | struct bpf_trampoline *tr; |
38207291 | 10524 | const struct btf_type *t; |
5b92a28a | 10525 | bool conservative = true; |
38207291 | 10526 | const char *tname; |
5b92a28a | 10527 | struct btf *btf; |
fec56f58 | 10528 | long addr; |
5b92a28a | 10529 | u64 key; |
38207291 | 10530 | |
27ae7997 MKL |
10531 | if (prog->type == BPF_PROG_TYPE_STRUCT_OPS) |
10532 | return check_struct_ops_btf_id(env); | |
10533 | ||
9e4e01df KS |
10534 | if (prog->type != BPF_PROG_TYPE_TRACING && |
10535 | prog->type != BPF_PROG_TYPE_LSM && | |
10536 | !prog_extension) | |
f1b9509c | 10537 | return 0; |
38207291 | 10538 | |
f1b9509c AS |
10539 | if (!btf_id) { |
10540 | verbose(env, "Tracing programs must provide btf_id\n"); | |
10541 | return -EINVAL; | |
10542 | } | |
5b92a28a AS |
10543 | btf = bpf_prog_get_target_btf(prog); |
10544 | if (!btf) { | |
10545 | verbose(env, | |
10546 | "FENTRY/FEXIT program can only be attached to another program annotated with BTF\n"); | |
10547 | return -EINVAL; | |
10548 | } | |
10549 | t = btf_type_by_id(btf, btf_id); | |
f1b9509c AS |
10550 | if (!t) { |
10551 | verbose(env, "attach_btf_id %u is invalid\n", btf_id); | |
10552 | return -EINVAL; | |
10553 | } | |
5b92a28a | 10554 | tname = btf_name_by_offset(btf, t->name_off); |
f1b9509c AS |
10555 | if (!tname) { |
10556 | verbose(env, "attach_btf_id %u doesn't have a name\n", btf_id); | |
10557 | return -EINVAL; | |
10558 | } | |
5b92a28a AS |
10559 | if (tgt_prog) { |
10560 | struct bpf_prog_aux *aux = tgt_prog->aux; | |
10561 | ||
10562 | for (i = 0; i < aux->func_info_cnt; i++) | |
10563 | if (aux->func_info[i].type_id == btf_id) { | |
10564 | subprog = i; | |
10565 | break; | |
10566 | } | |
10567 | if (subprog == -1) { | |
10568 | verbose(env, "Subprog %s doesn't exist\n", tname); | |
10569 | return -EINVAL; | |
10570 | } | |
10571 | conservative = aux->func_info_aux[subprog].unreliable; | |
be8704ff AS |
10572 | if (prog_extension) { |
10573 | if (conservative) { | |
10574 | verbose(env, | |
10575 | "Cannot replace static functions\n"); | |
10576 | return -EINVAL; | |
10577 | } | |
10578 | if (!prog->jit_requested) { | |
10579 | verbose(env, | |
10580 | "Extension programs should be JITed\n"); | |
10581 | return -EINVAL; | |
10582 | } | |
10583 | env->ops = bpf_verifier_ops[tgt_prog->type]; | |
03f87c0b | 10584 | prog->expected_attach_type = tgt_prog->expected_attach_type; |
be8704ff AS |
10585 | } |
10586 | if (!tgt_prog->jited) { | |
10587 | verbose(env, "Can attach to only JITed progs\n"); | |
10588 | return -EINVAL; | |
10589 | } | |
10590 | if (tgt_prog->type == prog->type) { | |
10591 | /* Cannot fentry/fexit another fentry/fexit program. | |
10592 | * Cannot attach program extension to another extension. | |
10593 | * It's ok to attach fentry/fexit to extension program. | |
10594 | */ | |
10595 | verbose(env, "Cannot recursively attach\n"); | |
10596 | return -EINVAL; | |
10597 | } | |
10598 | if (tgt_prog->type == BPF_PROG_TYPE_TRACING && | |
10599 | prog_extension && | |
10600 | (tgt_prog->expected_attach_type == BPF_TRACE_FENTRY || | |
10601 | tgt_prog->expected_attach_type == BPF_TRACE_FEXIT)) { | |
10602 | /* Program extensions can extend all program types | |
10603 | * except fentry/fexit. The reason is the following. | |
10604 | * The fentry/fexit programs are used for performance | |
10605 | * analysis, stats and can be attached to any program | |
10606 | * type except themselves. When extension program is | |
10607 | * replacing XDP function it is necessary to allow | |
10608 | * performance analysis of all functions. Both original | |
10609 | * XDP program and its program extension. Hence | |
10610 | * attaching fentry/fexit to BPF_PROG_TYPE_EXT is | |
10611 | * allowed. If extending of fentry/fexit was allowed it | |
10612 | * would be possible to create long call chain | |
10613 | * fentry->extension->fentry->extension beyond | |
10614 | * reasonable stack size. Hence extending fentry is not | |
10615 | * allowed. | |
10616 | */ | |
10617 | verbose(env, "Cannot extend fentry/fexit\n"); | |
10618 | return -EINVAL; | |
10619 | } | |
5b92a28a AS |
10620 | key = ((u64)aux->id) << 32 | btf_id; |
10621 | } else { | |
be8704ff AS |
10622 | if (prog_extension) { |
10623 | verbose(env, "Cannot replace kernel functions\n"); | |
10624 | return -EINVAL; | |
10625 | } | |
5b92a28a AS |
10626 | key = btf_id; |
10627 | } | |
f1b9509c AS |
10628 | |
10629 | switch (prog->expected_attach_type) { | |
10630 | case BPF_TRACE_RAW_TP: | |
5b92a28a AS |
10631 | if (tgt_prog) { |
10632 | verbose(env, | |
10633 | "Only FENTRY/FEXIT progs are attachable to another BPF prog\n"); | |
10634 | return -EINVAL; | |
10635 | } | |
38207291 MKL |
10636 | if (!btf_type_is_typedef(t)) { |
10637 | verbose(env, "attach_btf_id %u is not a typedef\n", | |
10638 | btf_id); | |
10639 | return -EINVAL; | |
10640 | } | |
f1b9509c | 10641 | if (strncmp(prefix, tname, sizeof(prefix) - 1)) { |
38207291 MKL |
10642 | verbose(env, "attach_btf_id %u points to wrong type name %s\n", |
10643 | btf_id, tname); | |
10644 | return -EINVAL; | |
10645 | } | |
10646 | tname += sizeof(prefix) - 1; | |
5b92a28a | 10647 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
10648 | if (!btf_type_is_ptr(t)) |
10649 | /* should never happen in valid vmlinux build */ | |
10650 | return -EINVAL; | |
5b92a28a | 10651 | t = btf_type_by_id(btf, t->type); |
38207291 MKL |
10652 | if (!btf_type_is_func_proto(t)) |
10653 | /* should never happen in valid vmlinux build */ | |
10654 | return -EINVAL; | |
10655 | ||
10656 | /* remember two read only pointers that are valid for | |
10657 | * the life time of the kernel | |
10658 | */ | |
10659 | prog->aux->attach_func_name = tname; | |
10660 | prog->aux->attach_func_proto = t; | |
10661 | prog->aux->attach_btf_trace = true; | |
f1b9509c | 10662 | return 0; |
15d83c4d YS |
10663 | case BPF_TRACE_ITER: |
10664 | if (!btf_type_is_func(t)) { | |
10665 | verbose(env, "attach_btf_id %u is not a function\n", | |
10666 | btf_id); | |
10667 | return -EINVAL; | |
10668 | } | |
10669 | t = btf_type_by_id(btf, t->type); | |
10670 | if (!btf_type_is_func_proto(t)) | |
10671 | return -EINVAL; | |
10672 | prog->aux->attach_func_name = tname; | |
10673 | prog->aux->attach_func_proto = t; | |
10674 | if (!bpf_iter_prog_supported(prog)) | |
10675 | return -EINVAL; | |
10676 | ret = btf_distill_func_proto(&env->log, btf, t, | |
10677 | tname, &fmodel); | |
10678 | return ret; | |
be8704ff AS |
10679 | default: |
10680 | if (!prog_extension) | |
10681 | return -EINVAL; | |
10682 | /* fallthrough */ | |
ae240823 | 10683 | case BPF_MODIFY_RETURN: |
9e4e01df | 10684 | case BPF_LSM_MAC: |
fec56f58 AS |
10685 | case BPF_TRACE_FENTRY: |
10686 | case BPF_TRACE_FEXIT: | |
9e4e01df KS |
10687 | prog->aux->attach_func_name = tname; |
10688 | if (prog->type == BPF_PROG_TYPE_LSM) { | |
10689 | ret = bpf_lsm_verify_prog(&env->log, prog); | |
10690 | if (ret < 0) | |
10691 | return ret; | |
10692 | } | |
10693 | ||
fec56f58 AS |
10694 | if (!btf_type_is_func(t)) { |
10695 | verbose(env, "attach_btf_id %u is not a function\n", | |
10696 | btf_id); | |
10697 | return -EINVAL; | |
10698 | } | |
be8704ff AS |
10699 | if (prog_extension && |
10700 | btf_check_type_match(env, prog, btf, t)) | |
10701 | return -EINVAL; | |
5b92a28a | 10702 | t = btf_type_by_id(btf, t->type); |
fec56f58 AS |
10703 | if (!btf_type_is_func_proto(t)) |
10704 | return -EINVAL; | |
5b92a28a | 10705 | tr = bpf_trampoline_lookup(key); |
fec56f58 AS |
10706 | if (!tr) |
10707 | return -ENOMEM; | |
5b92a28a | 10708 | /* t is either vmlinux type or another program's type */ |
fec56f58 AS |
10709 | prog->aux->attach_func_proto = t; |
10710 | mutex_lock(&tr->mutex); | |
10711 | if (tr->func.addr) { | |
10712 | prog->aux->trampoline = tr; | |
10713 | goto out; | |
10714 | } | |
5b92a28a AS |
10715 | if (tgt_prog && conservative) { |
10716 | prog->aux->attach_func_proto = NULL; | |
10717 | t = NULL; | |
10718 | } | |
10719 | ret = btf_distill_func_proto(&env->log, btf, t, | |
fec56f58 AS |
10720 | tname, &tr->func.model); |
10721 | if (ret < 0) | |
10722 | goto out; | |
5b92a28a | 10723 | if (tgt_prog) { |
e9eeec58 YS |
10724 | if (subprog == 0) |
10725 | addr = (long) tgt_prog->bpf_func; | |
10726 | else | |
10727 | addr = (long) tgt_prog->aux->func[subprog]->bpf_func; | |
5b92a28a AS |
10728 | } else { |
10729 | addr = kallsyms_lookup_name(tname); | |
10730 | if (!addr) { | |
10731 | verbose(env, | |
10732 | "The address of function %s cannot be found\n", | |
10733 | tname); | |
10734 | ret = -ENOENT; | |
10735 | goto out; | |
10736 | } | |
fec56f58 AS |
10737 | } |
10738 | tr->func.addr = (void *)addr; | |
10739 | prog->aux->trampoline = tr; | |
6ba43b76 KS |
10740 | |
10741 | if (prog->expected_attach_type == BPF_MODIFY_RETURN) | |
10742 | ret = check_attach_modify_return(env); | |
fec56f58 AS |
10743 | out: |
10744 | mutex_unlock(&tr->mutex); | |
10745 | if (ret) | |
10746 | bpf_trampoline_put(tr); | |
10747 | return ret; | |
38207291 | 10748 | } |
38207291 MKL |
10749 | } |
10750 | ||
838e9690 YS |
10751 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, |
10752 | union bpf_attr __user *uattr) | |
51580e79 | 10753 | { |
06ee7115 | 10754 | u64 start_time = ktime_get_ns(); |
58e2af8b | 10755 | struct bpf_verifier_env *env; |
b9193c1b | 10756 | struct bpf_verifier_log *log; |
9e4c24e7 | 10757 | int i, len, ret = -EINVAL; |
e2ae4ca2 | 10758 | bool is_priv; |
51580e79 | 10759 | |
eba0c929 AB |
10760 | /* no program is valid */ |
10761 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
10762 | return -EINVAL; | |
10763 | ||
58e2af8b | 10764 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
10765 | * allocate/free it every time bpf_check() is called |
10766 | */ | |
58e2af8b | 10767 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
10768 | if (!env) |
10769 | return -ENOMEM; | |
61bd5218 | 10770 | log = &env->log; |
cbd35700 | 10771 | |
9e4c24e7 | 10772 | len = (*prog)->len; |
fad953ce | 10773 | env->insn_aux_data = |
9e4c24e7 | 10774 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
10775 | ret = -ENOMEM; |
10776 | if (!env->insn_aux_data) | |
10777 | goto err_free_env; | |
9e4c24e7 JK |
10778 | for (i = 0; i < len; i++) |
10779 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 10780 | env->prog = *prog; |
00176a34 | 10781 | env->ops = bpf_verifier_ops[env->prog->type]; |
2c78ee89 | 10782 | is_priv = bpf_capable(); |
0246e64d | 10783 | |
8580ac94 AS |
10784 | if (!btf_vmlinux && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) { |
10785 | mutex_lock(&bpf_verifier_lock); | |
10786 | if (!btf_vmlinux) | |
10787 | btf_vmlinux = btf_parse_vmlinux(); | |
10788 | mutex_unlock(&bpf_verifier_lock); | |
10789 | } | |
10790 | ||
cbd35700 | 10791 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
10792 | if (!is_priv) |
10793 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 AS |
10794 | |
10795 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
10796 | /* user requested verbose verifier output | |
10797 | * and supplied buffer to store the verification trace | |
10798 | */ | |
e7bf8249 JK |
10799 | log->level = attr->log_level; |
10800 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
10801 | log->len_total = attr->log_size; | |
cbd35700 AS |
10802 | |
10803 | ret = -EINVAL; | |
e7bf8249 | 10804 | /* log attributes have to be sane */ |
7a9f5c65 | 10805 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 || |
06ee7115 | 10806 | !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK) |
3df126f3 | 10807 | goto err_unlock; |
cbd35700 | 10808 | } |
1ad2f583 | 10809 | |
8580ac94 AS |
10810 | if (IS_ERR(btf_vmlinux)) { |
10811 | /* Either gcc or pahole or kernel are broken. */ | |
10812 | verbose(env, "in-kernel BTF is malformed\n"); | |
10813 | ret = PTR_ERR(btf_vmlinux); | |
38207291 | 10814 | goto skip_full_check; |
8580ac94 AS |
10815 | } |
10816 | ||
1ad2f583 DB |
10817 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); |
10818 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 10819 | env->strict_alignment = true; |
e9ee9efc DM |
10820 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
10821 | env->strict_alignment = false; | |
cbd35700 | 10822 | |
2c78ee89 AS |
10823 | env->allow_ptr_leaks = bpf_allow_ptr_leaks(); |
10824 | env->bypass_spec_v1 = bpf_bypass_spec_v1(); | |
10825 | env->bypass_spec_v4 = bpf_bypass_spec_v4(); | |
10826 | env->bpf_capable = bpf_capable(); | |
e2ae4ca2 | 10827 | |
10d274e8 AS |
10828 | if (is_priv) |
10829 | env->test_state_freq = attr->prog_flags & BPF_F_TEST_STATE_FREQ; | |
10830 | ||
f4e3ec0d JK |
10831 | ret = replace_map_fd_with_map_ptr(env); |
10832 | if (ret < 0) | |
10833 | goto skip_full_check; | |
10834 | ||
cae1927c | 10835 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
a40a2632 | 10836 | ret = bpf_prog_offload_verifier_prep(env->prog); |
ab3f0063 | 10837 | if (ret) |
f4e3ec0d | 10838 | goto skip_full_check; |
ab3f0063 JK |
10839 | } |
10840 | ||
dc2a4ebc | 10841 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 10842 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
10843 | GFP_USER); |
10844 | ret = -ENOMEM; | |
10845 | if (!env->explored_states) | |
10846 | goto skip_full_check; | |
10847 | ||
d9762e84 | 10848 | ret = check_subprogs(env); |
475fb78f AS |
10849 | if (ret < 0) |
10850 | goto skip_full_check; | |
10851 | ||
c454a46b | 10852 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
10853 | if (ret < 0) |
10854 | goto skip_full_check; | |
10855 | ||
be8704ff AS |
10856 | ret = check_attach_btf_id(env); |
10857 | if (ret) | |
10858 | goto skip_full_check; | |
10859 | ||
d9762e84 MKL |
10860 | ret = check_cfg(env); |
10861 | if (ret < 0) | |
10862 | goto skip_full_check; | |
10863 | ||
51c39bb1 AS |
10864 | ret = do_check_subprogs(env); |
10865 | ret = ret ?: do_check_main(env); | |
cbd35700 | 10866 | |
c941ce9c QM |
10867 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
10868 | ret = bpf_prog_offload_finalize(env); | |
10869 | ||
0246e64d | 10870 | skip_full_check: |
51c39bb1 | 10871 | kvfree(env->explored_states); |
0246e64d | 10872 | |
c131187d | 10873 | if (ret == 0) |
9b38c405 | 10874 | ret = check_max_stack_depth(env); |
c131187d | 10875 | |
9b38c405 | 10876 | /* instruction rewrites happen after this point */ |
e2ae4ca2 JK |
10877 | if (is_priv) { |
10878 | if (ret == 0) | |
10879 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
10880 | if (ret == 0) |
10881 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
10882 | if (ret == 0) |
10883 | ret = opt_remove_nops(env); | |
52875a04 JK |
10884 | } else { |
10885 | if (ret == 0) | |
10886 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
10887 | } |
10888 | ||
9bac3d6d AS |
10889 | if (ret == 0) |
10890 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
10891 | ret = convert_ctx_accesses(env); | |
10892 | ||
e245c5c6 | 10893 | if (ret == 0) |
79741b3b | 10894 | ret = fixup_bpf_calls(env); |
e245c5c6 | 10895 | |
a4b1d3c1 JW |
10896 | /* do 32-bit optimization after insn patching has done so those patched |
10897 | * insns could be handled correctly. | |
10898 | */ | |
d6c2308c JW |
10899 | if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) { |
10900 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); | |
10901 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
10902 | : false; | |
a4b1d3c1 JW |
10903 | } |
10904 | ||
1ea47e01 AS |
10905 | if (ret == 0) |
10906 | ret = fixup_call_args(env); | |
10907 | ||
06ee7115 AS |
10908 | env->verification_time = ktime_get_ns() - start_time; |
10909 | print_verification_stats(env); | |
10910 | ||
a2a7d570 | 10911 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 10912 | ret = -ENOSPC; |
a2a7d570 | 10913 | if (log->level && !log->ubuf) { |
cbd35700 | 10914 | ret = -EFAULT; |
a2a7d570 | 10915 | goto err_release_maps; |
cbd35700 AS |
10916 | } |
10917 | ||
0246e64d AS |
10918 | if (ret == 0 && env->used_map_cnt) { |
10919 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
10920 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
10921 | sizeof(env->used_maps[0]), | |
10922 | GFP_KERNEL); | |
0246e64d | 10923 | |
9bac3d6d | 10924 | if (!env->prog->aux->used_maps) { |
0246e64d | 10925 | ret = -ENOMEM; |
a2a7d570 | 10926 | goto err_release_maps; |
0246e64d AS |
10927 | } |
10928 | ||
9bac3d6d | 10929 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 10930 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 10931 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
10932 | |
10933 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
10934 | * bpf_ld_imm64 instructions | |
10935 | */ | |
10936 | convert_pseudo_ld_imm64(env); | |
10937 | } | |
cbd35700 | 10938 | |
ba64e7d8 YS |
10939 | if (ret == 0) |
10940 | adjust_btf_func(env); | |
10941 | ||
a2a7d570 | 10942 | err_release_maps: |
9bac3d6d | 10943 | if (!env->prog->aux->used_maps) |
0246e64d | 10944 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 10945 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
10946 | */ |
10947 | release_maps(env); | |
03f87c0b THJ |
10948 | |
10949 | /* extension progs temporarily inherit the attach_type of their targets | |
10950 | for verification purposes, so set it back to zero before returning | |
10951 | */ | |
10952 | if (env->prog->type == BPF_PROG_TYPE_EXT) | |
10953 | env->prog->expected_attach_type = 0; | |
10954 | ||
9bac3d6d | 10955 | *prog = env->prog; |
3df126f3 | 10956 | err_unlock: |
45a73c17 AS |
10957 | if (!is_priv) |
10958 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
10959 | vfree(env->insn_aux_data); |
10960 | err_free_env: | |
10961 | kfree(env); | |
51580e79 AS |
10962 | return ret; |
10963 | } |