Commit | Line | Data |
---|---|---|
51580e79 | 1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 2 | * Copyright (c) 2016 Facebook |
fd978bf7 | 3 | * Copyright (c) 2018 Covalent IO, Inc. http://covalent.io |
51580e79 AS |
4 | * |
5 | * This program is free software; you can redistribute it and/or | |
6 | * modify it under the terms of version 2 of the GNU General Public | |
7 | * License as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but | |
10 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
12 | * General Public License for more details. | |
13 | */ | |
838e9690 | 14 | #include <uapi/linux/btf.h> |
51580e79 AS |
15 | #include <linux/kernel.h> |
16 | #include <linux/types.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/bpf.h> | |
838e9690 | 19 | #include <linux/btf.h> |
58e2af8b | 20 | #include <linux/bpf_verifier.h> |
51580e79 AS |
21 | #include <linux/filter.h> |
22 | #include <net/netlink.h> | |
23 | #include <linux/file.h> | |
24 | #include <linux/vmalloc.h> | |
ebb676da | 25 | #include <linux/stringify.h> |
cc8b0b92 AS |
26 | #include <linux/bsearch.h> |
27 | #include <linux/sort.h> | |
c195651e | 28 | #include <linux/perf_event.h> |
d9762e84 | 29 | #include <linux/ctype.h> |
51580e79 | 30 | |
f4ac7e0b JK |
31 | #include "disasm.h" |
32 | ||
00176a34 JK |
33 | static const struct bpf_verifier_ops * const bpf_verifier_ops[] = { |
34 | #define BPF_PROG_TYPE(_id, _name) \ | |
35 | [_id] = & _name ## _verifier_ops, | |
36 | #define BPF_MAP_TYPE(_id, _ops) | |
37 | #include <linux/bpf_types.h> | |
38 | #undef BPF_PROG_TYPE | |
39 | #undef BPF_MAP_TYPE | |
40 | }; | |
41 | ||
51580e79 AS |
42 | /* bpf_check() is a static code analyzer that walks eBPF program |
43 | * instruction by instruction and updates register/stack state. | |
44 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
45 | * | |
46 | * The first pass is depth-first-search to check that the program is a DAG. | |
47 | * It rejects the following programs: | |
48 | * - larger than BPF_MAXINSNS insns | |
49 | * - if loop is present (detected via back-edge) | |
50 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
51 | * - out of bounds or malformed jumps | |
52 | * The second pass is all possible path descent from the 1st insn. | |
53 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 54 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
55 | * insn is less then 4K, but there are too many branches that change stack/regs. |
56 | * Number of 'branches to be analyzed' is limited to 1k | |
57 | * | |
58 | * On entry to each instruction, each register has a type, and the instruction | |
59 | * changes the types of the registers depending on instruction semantics. | |
60 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
61 | * copied to R1. | |
62 | * | |
63 | * All registers are 64-bit. | |
64 | * R0 - return register | |
65 | * R1-R5 argument passing registers | |
66 | * R6-R9 callee saved registers | |
67 | * R10 - frame pointer read-only | |
68 | * | |
69 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
70 | * and has type PTR_TO_CTX. | |
71 | * | |
72 | * Verifier tracks arithmetic operations on pointers in case: | |
73 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
74 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
75 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
76 | * and 2nd arithmetic instruction is pattern matched to recognize | |
77 | * that it wants to construct a pointer to some element within stack. | |
78 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
79 | * (and -20 constant is saved for further stack bounds checking). | |
80 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
81 | * | |
f1174f77 | 82 | * Most of the time the registers have SCALAR_VALUE type, which |
51580e79 | 83 | * means the register has some value, but it's not a valid pointer. |
f1174f77 | 84 | * (like pointer plus pointer becomes SCALAR_VALUE type) |
51580e79 AS |
85 | * |
86 | * When verifier sees load or store instructions the type of base register | |
c64b7983 JS |
87 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, PTR_TO_STACK, PTR_TO_SOCKET. These are |
88 | * four pointer types recognized by check_mem_access() function. | |
51580e79 AS |
89 | * |
90 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
91 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
92 | * | |
93 | * registers used to pass values to function calls are checked against | |
94 | * function argument constraints. | |
95 | * | |
96 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
97 | * It means that the register type passed to this function must be | |
98 | * PTR_TO_STACK and it will be used inside the function as | |
99 | * 'pointer to map element key' | |
100 | * | |
101 | * For example the argument constraints for bpf_map_lookup_elem(): | |
102 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
103 | * .arg1_type = ARG_CONST_MAP_PTR, | |
104 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
105 | * | |
106 | * ret_type says that this function returns 'pointer to map elem value or null' | |
107 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
108 | * 2nd argument should be a pointer to stack, which will be used inside | |
109 | * the helper function as a pointer to map element key. | |
110 | * | |
111 | * On the kernel side the helper function looks like: | |
112 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
113 | * { | |
114 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
115 | * void *key = (void *) (unsigned long) r2; | |
116 | * void *value; | |
117 | * | |
118 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
119 | * [key, key + map->key_size) bytes are valid and were initialized on | |
120 | * the stack of eBPF program. | |
121 | * } | |
122 | * | |
123 | * Corresponding eBPF program may look like: | |
124 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
125 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
126 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
127 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
128 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
129 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
130 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
131 | * | |
132 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
133 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
134 | * and were initialized prior to this call. | |
135 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
136 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
137 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
138 | * returns ether pointer to map value or NULL. | |
139 | * | |
140 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
141 | * insn, the register holding that pointer in the true branch changes state to | |
142 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
143 | * branch. See check_cond_jmp_op(). | |
144 | * | |
145 | * After the call R0 is set to return type of the function and registers R1-R5 | |
146 | * are set to NOT_INIT to indicate that they are no longer readable. | |
fd978bf7 JS |
147 | * |
148 | * The following reference types represent a potential reference to a kernel | |
149 | * resource which, after first being allocated, must be checked and freed by | |
150 | * the BPF program: | |
151 | * - PTR_TO_SOCKET_OR_NULL, PTR_TO_SOCKET | |
152 | * | |
153 | * When the verifier sees a helper call return a reference type, it allocates a | |
154 | * pointer id for the reference and stores it in the current function state. | |
155 | * Similar to the way that PTR_TO_MAP_VALUE_OR_NULL is converted into | |
156 | * PTR_TO_MAP_VALUE, PTR_TO_SOCKET_OR_NULL becomes PTR_TO_SOCKET when the type | |
157 | * passes through a NULL-check conditional. For the branch wherein the state is | |
158 | * changed to CONST_IMM, the verifier releases the reference. | |
6acc9b43 JS |
159 | * |
160 | * For each helper function that allocates a reference, such as | |
161 | * bpf_sk_lookup_tcp(), there is a corresponding release function, such as | |
162 | * bpf_sk_release(). When a reference type passes into the release function, | |
163 | * the verifier also releases the reference. If any unchecked or unreleased | |
164 | * reference remains at the end of the program, the verifier rejects it. | |
51580e79 AS |
165 | */ |
166 | ||
17a52670 | 167 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 168 | struct bpf_verifier_stack_elem { |
17a52670 AS |
169 | /* verifer state is 'st' |
170 | * before processing instruction 'insn_idx' | |
171 | * and after processing instruction 'prev_insn_idx' | |
172 | */ | |
58e2af8b | 173 | struct bpf_verifier_state st; |
17a52670 AS |
174 | int insn_idx; |
175 | int prev_insn_idx; | |
58e2af8b | 176 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
177 | }; |
178 | ||
b285fcb7 | 179 | #define BPF_COMPLEXITY_LIMIT_JMP_SEQ 8192 |
ceefbc96 | 180 | #define BPF_COMPLEXITY_LIMIT_STATES 64 |
07016151 | 181 | |
c93552c4 DB |
182 | #define BPF_MAP_PTR_UNPRIV 1UL |
183 | #define BPF_MAP_PTR_POISON ((void *)((0xeB9FUL << 1) + \ | |
184 | POISON_POINTER_DELTA)) | |
185 | #define BPF_MAP_PTR(X) ((struct bpf_map *)((X) & ~BPF_MAP_PTR_UNPRIV)) | |
186 | ||
187 | static bool bpf_map_ptr_poisoned(const struct bpf_insn_aux_data *aux) | |
188 | { | |
189 | return BPF_MAP_PTR(aux->map_state) == BPF_MAP_PTR_POISON; | |
190 | } | |
191 | ||
192 | static bool bpf_map_ptr_unpriv(const struct bpf_insn_aux_data *aux) | |
193 | { | |
194 | return aux->map_state & BPF_MAP_PTR_UNPRIV; | |
195 | } | |
196 | ||
197 | static void bpf_map_ptr_store(struct bpf_insn_aux_data *aux, | |
198 | const struct bpf_map *map, bool unpriv) | |
199 | { | |
200 | BUILD_BUG_ON((unsigned long)BPF_MAP_PTR_POISON & BPF_MAP_PTR_UNPRIV); | |
201 | unpriv |= bpf_map_ptr_unpriv(aux); | |
202 | aux->map_state = (unsigned long)map | | |
203 | (unpriv ? BPF_MAP_PTR_UNPRIV : 0UL); | |
204 | } | |
fad73a1a | 205 | |
33ff9823 DB |
206 | struct bpf_call_arg_meta { |
207 | struct bpf_map *map_ptr; | |
435faee1 | 208 | bool raw_mode; |
36bbef52 | 209 | bool pkt_access; |
435faee1 DB |
210 | int regno; |
211 | int access_size; | |
849fa506 YS |
212 | s64 msize_smax_value; |
213 | u64 msize_umax_value; | |
1b986589 | 214 | int ref_obj_id; |
d83525ca | 215 | int func_id; |
33ff9823 DB |
216 | }; |
217 | ||
cbd35700 AS |
218 | static DEFINE_MUTEX(bpf_verifier_lock); |
219 | ||
d9762e84 MKL |
220 | static const struct bpf_line_info * |
221 | find_linfo(const struct bpf_verifier_env *env, u32 insn_off) | |
222 | { | |
223 | const struct bpf_line_info *linfo; | |
224 | const struct bpf_prog *prog; | |
225 | u32 i, nr_linfo; | |
226 | ||
227 | prog = env->prog; | |
228 | nr_linfo = prog->aux->nr_linfo; | |
229 | ||
230 | if (!nr_linfo || insn_off >= prog->len) | |
231 | return NULL; | |
232 | ||
233 | linfo = prog->aux->linfo; | |
234 | for (i = 1; i < nr_linfo; i++) | |
235 | if (insn_off < linfo[i].insn_off) | |
236 | break; | |
237 | ||
238 | return &linfo[i - 1]; | |
239 | } | |
240 | ||
77d2e05a MKL |
241 | void bpf_verifier_vlog(struct bpf_verifier_log *log, const char *fmt, |
242 | va_list args) | |
cbd35700 | 243 | { |
a2a7d570 | 244 | unsigned int n; |
cbd35700 | 245 | |
a2a7d570 | 246 | n = vscnprintf(log->kbuf, BPF_VERIFIER_TMP_LOG_SIZE, fmt, args); |
a2a7d570 JK |
247 | |
248 | WARN_ONCE(n >= BPF_VERIFIER_TMP_LOG_SIZE - 1, | |
249 | "verifier log line truncated - local buffer too short\n"); | |
250 | ||
251 | n = min(log->len_total - log->len_used - 1, n); | |
252 | log->kbuf[n] = '\0'; | |
253 | ||
254 | if (!copy_to_user(log->ubuf + log->len_used, log->kbuf, n + 1)) | |
255 | log->len_used += n; | |
256 | else | |
257 | log->ubuf = NULL; | |
cbd35700 | 258 | } |
abe08840 JO |
259 | |
260 | /* log_level controls verbosity level of eBPF verifier. | |
261 | * bpf_verifier_log_write() is used to dump the verification trace to the log, | |
262 | * so the user can figure out what's wrong with the program | |
430e68d1 | 263 | */ |
abe08840 JO |
264 | __printf(2, 3) void bpf_verifier_log_write(struct bpf_verifier_env *env, |
265 | const char *fmt, ...) | |
266 | { | |
267 | va_list args; | |
268 | ||
77d2e05a MKL |
269 | if (!bpf_verifier_log_needed(&env->log)) |
270 | return; | |
271 | ||
abe08840 | 272 | va_start(args, fmt); |
77d2e05a | 273 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
274 | va_end(args); |
275 | } | |
276 | EXPORT_SYMBOL_GPL(bpf_verifier_log_write); | |
277 | ||
278 | __printf(2, 3) static void verbose(void *private_data, const char *fmt, ...) | |
279 | { | |
77d2e05a | 280 | struct bpf_verifier_env *env = private_data; |
abe08840 JO |
281 | va_list args; |
282 | ||
77d2e05a MKL |
283 | if (!bpf_verifier_log_needed(&env->log)) |
284 | return; | |
285 | ||
abe08840 | 286 | va_start(args, fmt); |
77d2e05a | 287 | bpf_verifier_vlog(&env->log, fmt, args); |
abe08840 JO |
288 | va_end(args); |
289 | } | |
cbd35700 | 290 | |
d9762e84 MKL |
291 | static const char *ltrim(const char *s) |
292 | { | |
293 | while (isspace(*s)) | |
294 | s++; | |
295 | ||
296 | return s; | |
297 | } | |
298 | ||
299 | __printf(3, 4) static void verbose_linfo(struct bpf_verifier_env *env, | |
300 | u32 insn_off, | |
301 | const char *prefix_fmt, ...) | |
302 | { | |
303 | const struct bpf_line_info *linfo; | |
304 | ||
305 | if (!bpf_verifier_log_needed(&env->log)) | |
306 | return; | |
307 | ||
308 | linfo = find_linfo(env, insn_off); | |
309 | if (!linfo || linfo == env->prev_linfo) | |
310 | return; | |
311 | ||
312 | if (prefix_fmt) { | |
313 | va_list args; | |
314 | ||
315 | va_start(args, prefix_fmt); | |
316 | bpf_verifier_vlog(&env->log, prefix_fmt, args); | |
317 | va_end(args); | |
318 | } | |
319 | ||
320 | verbose(env, "%s\n", | |
321 | ltrim(btf_name_by_offset(env->prog->aux->btf, | |
322 | linfo->line_off))); | |
323 | ||
324 | env->prev_linfo = linfo; | |
325 | } | |
326 | ||
de8f3a83 DB |
327 | static bool type_is_pkt_pointer(enum bpf_reg_type type) |
328 | { | |
329 | return type == PTR_TO_PACKET || | |
330 | type == PTR_TO_PACKET_META; | |
331 | } | |
332 | ||
46f8bc92 MKL |
333 | static bool type_is_sk_pointer(enum bpf_reg_type type) |
334 | { | |
335 | return type == PTR_TO_SOCKET || | |
655a51e5 | 336 | type == PTR_TO_SOCK_COMMON || |
fada7fdc JL |
337 | type == PTR_TO_TCP_SOCK || |
338 | type == PTR_TO_XDP_SOCK; | |
46f8bc92 MKL |
339 | } |
340 | ||
840b9615 JS |
341 | static bool reg_type_may_be_null(enum bpf_reg_type type) |
342 | { | |
fd978bf7 | 343 | return type == PTR_TO_MAP_VALUE_OR_NULL || |
46f8bc92 | 344 | type == PTR_TO_SOCKET_OR_NULL || |
655a51e5 MKL |
345 | type == PTR_TO_SOCK_COMMON_OR_NULL || |
346 | type == PTR_TO_TCP_SOCK_OR_NULL; | |
fd978bf7 JS |
347 | } |
348 | ||
d83525ca AS |
349 | static bool reg_may_point_to_spin_lock(const struct bpf_reg_state *reg) |
350 | { | |
351 | return reg->type == PTR_TO_MAP_VALUE && | |
352 | map_value_has_spin_lock(reg->map_ptr); | |
353 | } | |
354 | ||
cba368c1 MKL |
355 | static bool reg_type_may_be_refcounted_or_null(enum bpf_reg_type type) |
356 | { | |
357 | return type == PTR_TO_SOCKET || | |
358 | type == PTR_TO_SOCKET_OR_NULL || | |
359 | type == PTR_TO_TCP_SOCK || | |
360 | type == PTR_TO_TCP_SOCK_OR_NULL; | |
361 | } | |
362 | ||
1b986589 | 363 | static bool arg_type_may_be_refcounted(enum bpf_arg_type type) |
fd978bf7 | 364 | { |
1b986589 | 365 | return type == ARG_PTR_TO_SOCK_COMMON; |
fd978bf7 JS |
366 | } |
367 | ||
368 | /* Determine whether the function releases some resources allocated by another | |
369 | * function call. The first reference type argument will be assumed to be | |
370 | * released by release_reference(). | |
371 | */ | |
372 | static bool is_release_function(enum bpf_func_id func_id) | |
373 | { | |
6acc9b43 | 374 | return func_id == BPF_FUNC_sk_release; |
840b9615 JS |
375 | } |
376 | ||
46f8bc92 MKL |
377 | static bool is_acquire_function(enum bpf_func_id func_id) |
378 | { | |
379 | return func_id == BPF_FUNC_sk_lookup_tcp || | |
edbf8c01 LB |
380 | func_id == BPF_FUNC_sk_lookup_udp || |
381 | func_id == BPF_FUNC_skc_lookup_tcp; | |
46f8bc92 MKL |
382 | } |
383 | ||
1b986589 MKL |
384 | static bool is_ptr_cast_function(enum bpf_func_id func_id) |
385 | { | |
386 | return func_id == BPF_FUNC_tcp_sock || | |
387 | func_id == BPF_FUNC_sk_fullsock; | |
388 | } | |
389 | ||
17a52670 AS |
390 | /* string representation of 'enum bpf_reg_type' */ |
391 | static const char * const reg_type_str[] = { | |
392 | [NOT_INIT] = "?", | |
f1174f77 | 393 | [SCALAR_VALUE] = "inv", |
17a52670 AS |
394 | [PTR_TO_CTX] = "ctx", |
395 | [CONST_PTR_TO_MAP] = "map_ptr", | |
396 | [PTR_TO_MAP_VALUE] = "map_value", | |
397 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
17a52670 | 398 | [PTR_TO_STACK] = "fp", |
969bf05e | 399 | [PTR_TO_PACKET] = "pkt", |
de8f3a83 | 400 | [PTR_TO_PACKET_META] = "pkt_meta", |
969bf05e | 401 | [PTR_TO_PACKET_END] = "pkt_end", |
d58e468b | 402 | [PTR_TO_FLOW_KEYS] = "flow_keys", |
c64b7983 JS |
403 | [PTR_TO_SOCKET] = "sock", |
404 | [PTR_TO_SOCKET_OR_NULL] = "sock_or_null", | |
46f8bc92 MKL |
405 | [PTR_TO_SOCK_COMMON] = "sock_common", |
406 | [PTR_TO_SOCK_COMMON_OR_NULL] = "sock_common_or_null", | |
655a51e5 MKL |
407 | [PTR_TO_TCP_SOCK] = "tcp_sock", |
408 | [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", | |
9df1c28b | 409 | [PTR_TO_TP_BUFFER] = "tp_buffer", |
fada7fdc | 410 | [PTR_TO_XDP_SOCK] = "xdp_sock", |
17a52670 AS |
411 | }; |
412 | ||
8efea21d EC |
413 | static char slot_type_char[] = { |
414 | [STACK_INVALID] = '?', | |
415 | [STACK_SPILL] = 'r', | |
416 | [STACK_MISC] = 'm', | |
417 | [STACK_ZERO] = '0', | |
418 | }; | |
419 | ||
4e92024a AS |
420 | static void print_liveness(struct bpf_verifier_env *env, |
421 | enum bpf_reg_liveness live) | |
422 | { | |
9242b5f5 | 423 | if (live & (REG_LIVE_READ | REG_LIVE_WRITTEN | REG_LIVE_DONE)) |
4e92024a AS |
424 | verbose(env, "_"); |
425 | if (live & REG_LIVE_READ) | |
426 | verbose(env, "r"); | |
427 | if (live & REG_LIVE_WRITTEN) | |
428 | verbose(env, "w"); | |
9242b5f5 AS |
429 | if (live & REG_LIVE_DONE) |
430 | verbose(env, "D"); | |
4e92024a AS |
431 | } |
432 | ||
f4d7e40a AS |
433 | static struct bpf_func_state *func(struct bpf_verifier_env *env, |
434 | const struct bpf_reg_state *reg) | |
435 | { | |
436 | struct bpf_verifier_state *cur = env->cur_state; | |
437 | ||
438 | return cur->frame[reg->frameno]; | |
439 | } | |
440 | ||
61bd5218 | 441 | static void print_verifier_state(struct bpf_verifier_env *env, |
f4d7e40a | 442 | const struct bpf_func_state *state) |
17a52670 | 443 | { |
f4d7e40a | 444 | const struct bpf_reg_state *reg; |
17a52670 AS |
445 | enum bpf_reg_type t; |
446 | int i; | |
447 | ||
f4d7e40a AS |
448 | if (state->frameno) |
449 | verbose(env, " frame%d:", state->frameno); | |
17a52670 | 450 | for (i = 0; i < MAX_BPF_REG; i++) { |
1a0dc1ac AS |
451 | reg = &state->regs[i]; |
452 | t = reg->type; | |
17a52670 AS |
453 | if (t == NOT_INIT) |
454 | continue; | |
4e92024a AS |
455 | verbose(env, " R%d", i); |
456 | print_liveness(env, reg->live); | |
457 | verbose(env, "=%s", reg_type_str[t]); | |
f1174f77 EC |
458 | if ((t == SCALAR_VALUE || t == PTR_TO_STACK) && |
459 | tnum_is_const(reg->var_off)) { | |
460 | /* reg->off should be 0 for SCALAR_VALUE */ | |
61bd5218 | 461 | verbose(env, "%lld", reg->var_off.value + reg->off); |
f4d7e40a AS |
462 | if (t == PTR_TO_STACK) |
463 | verbose(env, ",call_%d", func(env, reg)->callsite); | |
f1174f77 | 464 | } else { |
cba368c1 MKL |
465 | verbose(env, "(id=%d", reg->id); |
466 | if (reg_type_may_be_refcounted_or_null(t)) | |
467 | verbose(env, ",ref_obj_id=%d", reg->ref_obj_id); | |
f1174f77 | 468 | if (t != SCALAR_VALUE) |
61bd5218 | 469 | verbose(env, ",off=%d", reg->off); |
de8f3a83 | 470 | if (type_is_pkt_pointer(t)) |
61bd5218 | 471 | verbose(env, ",r=%d", reg->range); |
f1174f77 EC |
472 | else if (t == CONST_PTR_TO_MAP || |
473 | t == PTR_TO_MAP_VALUE || | |
474 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
61bd5218 | 475 | verbose(env, ",ks=%d,vs=%d", |
f1174f77 EC |
476 | reg->map_ptr->key_size, |
477 | reg->map_ptr->value_size); | |
7d1238f2 EC |
478 | if (tnum_is_const(reg->var_off)) { |
479 | /* Typically an immediate SCALAR_VALUE, but | |
480 | * could be a pointer whose offset is too big | |
481 | * for reg->off | |
482 | */ | |
61bd5218 | 483 | verbose(env, ",imm=%llx", reg->var_off.value); |
7d1238f2 EC |
484 | } else { |
485 | if (reg->smin_value != reg->umin_value && | |
486 | reg->smin_value != S64_MIN) | |
61bd5218 | 487 | verbose(env, ",smin_value=%lld", |
7d1238f2 EC |
488 | (long long)reg->smin_value); |
489 | if (reg->smax_value != reg->umax_value && | |
490 | reg->smax_value != S64_MAX) | |
61bd5218 | 491 | verbose(env, ",smax_value=%lld", |
7d1238f2 EC |
492 | (long long)reg->smax_value); |
493 | if (reg->umin_value != 0) | |
61bd5218 | 494 | verbose(env, ",umin_value=%llu", |
7d1238f2 EC |
495 | (unsigned long long)reg->umin_value); |
496 | if (reg->umax_value != U64_MAX) | |
61bd5218 | 497 | verbose(env, ",umax_value=%llu", |
7d1238f2 EC |
498 | (unsigned long long)reg->umax_value); |
499 | if (!tnum_is_unknown(reg->var_off)) { | |
500 | char tn_buf[48]; | |
f1174f77 | 501 | |
7d1238f2 | 502 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 503 | verbose(env, ",var_off=%s", tn_buf); |
7d1238f2 | 504 | } |
f1174f77 | 505 | } |
61bd5218 | 506 | verbose(env, ")"); |
f1174f77 | 507 | } |
17a52670 | 508 | } |
638f5b90 | 509 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE; i++) { |
8efea21d EC |
510 | char types_buf[BPF_REG_SIZE + 1]; |
511 | bool valid = false; | |
512 | int j; | |
513 | ||
514 | for (j = 0; j < BPF_REG_SIZE; j++) { | |
515 | if (state->stack[i].slot_type[j] != STACK_INVALID) | |
516 | valid = true; | |
517 | types_buf[j] = slot_type_char[ | |
518 | state->stack[i].slot_type[j]]; | |
519 | } | |
520 | types_buf[BPF_REG_SIZE] = 0; | |
521 | if (!valid) | |
522 | continue; | |
523 | verbose(env, " fp%d", (-i - 1) * BPF_REG_SIZE); | |
524 | print_liveness(env, state->stack[i].spilled_ptr.live); | |
525 | if (state->stack[i].slot_type[0] == STACK_SPILL) | |
4e92024a | 526 | verbose(env, "=%s", |
638f5b90 | 527 | reg_type_str[state->stack[i].spilled_ptr.type]); |
8efea21d EC |
528 | else |
529 | verbose(env, "=%s", types_buf); | |
17a52670 | 530 | } |
fd978bf7 JS |
531 | if (state->acquired_refs && state->refs[0].id) { |
532 | verbose(env, " refs=%d", state->refs[0].id); | |
533 | for (i = 1; i < state->acquired_refs; i++) | |
534 | if (state->refs[i].id) | |
535 | verbose(env, ",%d", state->refs[i].id); | |
536 | } | |
61bd5218 | 537 | verbose(env, "\n"); |
17a52670 AS |
538 | } |
539 | ||
84dbf350 JS |
540 | #define COPY_STATE_FN(NAME, COUNT, FIELD, SIZE) \ |
541 | static int copy_##NAME##_state(struct bpf_func_state *dst, \ | |
542 | const struct bpf_func_state *src) \ | |
543 | { \ | |
544 | if (!src->FIELD) \ | |
545 | return 0; \ | |
546 | if (WARN_ON_ONCE(dst->COUNT < src->COUNT)) { \ | |
547 | /* internal bug, make state invalid to reject the program */ \ | |
548 | memset(dst, 0, sizeof(*dst)); \ | |
549 | return -EFAULT; \ | |
550 | } \ | |
551 | memcpy(dst->FIELD, src->FIELD, \ | |
552 | sizeof(*src->FIELD) * (src->COUNT / SIZE)); \ | |
553 | return 0; \ | |
638f5b90 | 554 | } |
fd978bf7 JS |
555 | /* copy_reference_state() */ |
556 | COPY_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
557 | /* copy_stack_state() */ |
558 | COPY_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
559 | #undef COPY_STATE_FN | |
560 | ||
561 | #define REALLOC_STATE_FN(NAME, COUNT, FIELD, SIZE) \ | |
562 | static int realloc_##NAME##_state(struct bpf_func_state *state, int size, \ | |
563 | bool copy_old) \ | |
564 | { \ | |
565 | u32 old_size = state->COUNT; \ | |
566 | struct bpf_##NAME##_state *new_##FIELD; \ | |
567 | int slot = size / SIZE; \ | |
568 | \ | |
569 | if (size <= old_size || !size) { \ | |
570 | if (copy_old) \ | |
571 | return 0; \ | |
572 | state->COUNT = slot * SIZE; \ | |
573 | if (!size && old_size) { \ | |
574 | kfree(state->FIELD); \ | |
575 | state->FIELD = NULL; \ | |
576 | } \ | |
577 | return 0; \ | |
578 | } \ | |
579 | new_##FIELD = kmalloc_array(slot, sizeof(struct bpf_##NAME##_state), \ | |
580 | GFP_KERNEL); \ | |
581 | if (!new_##FIELD) \ | |
582 | return -ENOMEM; \ | |
583 | if (copy_old) { \ | |
584 | if (state->FIELD) \ | |
585 | memcpy(new_##FIELD, state->FIELD, \ | |
586 | sizeof(*new_##FIELD) * (old_size / SIZE)); \ | |
587 | memset(new_##FIELD + old_size / SIZE, 0, \ | |
588 | sizeof(*new_##FIELD) * (size - old_size) / SIZE); \ | |
589 | } \ | |
590 | state->COUNT = slot * SIZE; \ | |
591 | kfree(state->FIELD); \ | |
592 | state->FIELD = new_##FIELD; \ | |
593 | return 0; \ | |
594 | } | |
fd978bf7 JS |
595 | /* realloc_reference_state() */ |
596 | REALLOC_STATE_FN(reference, acquired_refs, refs, 1) | |
84dbf350 JS |
597 | /* realloc_stack_state() */ |
598 | REALLOC_STATE_FN(stack, allocated_stack, stack, BPF_REG_SIZE) | |
599 | #undef REALLOC_STATE_FN | |
638f5b90 AS |
600 | |
601 | /* do_check() starts with zero-sized stack in struct bpf_verifier_state to | |
602 | * make it consume minimal amount of memory. check_stack_write() access from | |
f4d7e40a | 603 | * the program calls into realloc_func_state() to grow the stack size. |
84dbf350 JS |
604 | * Note there is a non-zero 'parent' pointer inside bpf_verifier_state |
605 | * which realloc_stack_state() copies over. It points to previous | |
606 | * bpf_verifier_state which is never reallocated. | |
638f5b90 | 607 | */ |
fd978bf7 JS |
608 | static int realloc_func_state(struct bpf_func_state *state, int stack_size, |
609 | int refs_size, bool copy_old) | |
638f5b90 | 610 | { |
fd978bf7 JS |
611 | int err = realloc_reference_state(state, refs_size, copy_old); |
612 | if (err) | |
613 | return err; | |
614 | return realloc_stack_state(state, stack_size, copy_old); | |
615 | } | |
616 | ||
617 | /* Acquire a pointer id from the env and update the state->refs to include | |
618 | * this new pointer reference. | |
619 | * On success, returns a valid pointer id to associate with the register | |
620 | * On failure, returns a negative errno. | |
638f5b90 | 621 | */ |
fd978bf7 | 622 | static int acquire_reference_state(struct bpf_verifier_env *env, int insn_idx) |
638f5b90 | 623 | { |
fd978bf7 JS |
624 | struct bpf_func_state *state = cur_func(env); |
625 | int new_ofs = state->acquired_refs; | |
626 | int id, err; | |
627 | ||
628 | err = realloc_reference_state(state, state->acquired_refs + 1, true); | |
629 | if (err) | |
630 | return err; | |
631 | id = ++env->id_gen; | |
632 | state->refs[new_ofs].id = id; | |
633 | state->refs[new_ofs].insn_idx = insn_idx; | |
638f5b90 | 634 | |
fd978bf7 JS |
635 | return id; |
636 | } | |
637 | ||
638 | /* release function corresponding to acquire_reference_state(). Idempotent. */ | |
46f8bc92 | 639 | static int release_reference_state(struct bpf_func_state *state, int ptr_id) |
fd978bf7 JS |
640 | { |
641 | int i, last_idx; | |
642 | ||
fd978bf7 JS |
643 | last_idx = state->acquired_refs - 1; |
644 | for (i = 0; i < state->acquired_refs; i++) { | |
645 | if (state->refs[i].id == ptr_id) { | |
646 | if (last_idx && i != last_idx) | |
647 | memcpy(&state->refs[i], &state->refs[last_idx], | |
648 | sizeof(*state->refs)); | |
649 | memset(&state->refs[last_idx], 0, sizeof(*state->refs)); | |
650 | state->acquired_refs--; | |
638f5b90 | 651 | return 0; |
638f5b90 | 652 | } |
638f5b90 | 653 | } |
46f8bc92 | 654 | return -EINVAL; |
fd978bf7 JS |
655 | } |
656 | ||
657 | static int transfer_reference_state(struct bpf_func_state *dst, | |
658 | struct bpf_func_state *src) | |
659 | { | |
660 | int err = realloc_reference_state(dst, src->acquired_refs, false); | |
661 | if (err) | |
662 | return err; | |
663 | err = copy_reference_state(dst, src); | |
664 | if (err) | |
665 | return err; | |
638f5b90 AS |
666 | return 0; |
667 | } | |
668 | ||
f4d7e40a AS |
669 | static void free_func_state(struct bpf_func_state *state) |
670 | { | |
5896351e AS |
671 | if (!state) |
672 | return; | |
fd978bf7 | 673 | kfree(state->refs); |
f4d7e40a AS |
674 | kfree(state->stack); |
675 | kfree(state); | |
676 | } | |
677 | ||
1969db47 AS |
678 | static void free_verifier_state(struct bpf_verifier_state *state, |
679 | bool free_self) | |
638f5b90 | 680 | { |
f4d7e40a AS |
681 | int i; |
682 | ||
683 | for (i = 0; i <= state->curframe; i++) { | |
684 | free_func_state(state->frame[i]); | |
685 | state->frame[i] = NULL; | |
686 | } | |
1969db47 AS |
687 | if (free_self) |
688 | kfree(state); | |
638f5b90 AS |
689 | } |
690 | ||
691 | /* copy verifier state from src to dst growing dst stack space | |
692 | * when necessary to accommodate larger src stack | |
693 | */ | |
f4d7e40a AS |
694 | static int copy_func_state(struct bpf_func_state *dst, |
695 | const struct bpf_func_state *src) | |
638f5b90 AS |
696 | { |
697 | int err; | |
698 | ||
fd978bf7 JS |
699 | err = realloc_func_state(dst, src->allocated_stack, src->acquired_refs, |
700 | false); | |
701 | if (err) | |
702 | return err; | |
703 | memcpy(dst, src, offsetof(struct bpf_func_state, acquired_refs)); | |
704 | err = copy_reference_state(dst, src); | |
638f5b90 AS |
705 | if (err) |
706 | return err; | |
638f5b90 AS |
707 | return copy_stack_state(dst, src); |
708 | } | |
709 | ||
f4d7e40a AS |
710 | static int copy_verifier_state(struct bpf_verifier_state *dst_state, |
711 | const struct bpf_verifier_state *src) | |
712 | { | |
713 | struct bpf_func_state *dst; | |
714 | int i, err; | |
715 | ||
716 | /* if dst has more stack frames then src frame, free them */ | |
717 | for (i = src->curframe + 1; i <= dst_state->curframe; i++) { | |
718 | free_func_state(dst_state->frame[i]); | |
719 | dst_state->frame[i] = NULL; | |
720 | } | |
979d63d5 | 721 | dst_state->speculative = src->speculative; |
f4d7e40a | 722 | dst_state->curframe = src->curframe; |
d83525ca | 723 | dst_state->active_spin_lock = src->active_spin_lock; |
f4d7e40a AS |
724 | for (i = 0; i <= src->curframe; i++) { |
725 | dst = dst_state->frame[i]; | |
726 | if (!dst) { | |
727 | dst = kzalloc(sizeof(*dst), GFP_KERNEL); | |
728 | if (!dst) | |
729 | return -ENOMEM; | |
730 | dst_state->frame[i] = dst; | |
731 | } | |
732 | err = copy_func_state(dst, src->frame[i]); | |
733 | if (err) | |
734 | return err; | |
735 | } | |
736 | return 0; | |
737 | } | |
738 | ||
638f5b90 AS |
739 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx, |
740 | int *insn_idx) | |
741 | { | |
742 | struct bpf_verifier_state *cur = env->cur_state; | |
743 | struct bpf_verifier_stack_elem *elem, *head = env->head; | |
744 | int err; | |
17a52670 AS |
745 | |
746 | if (env->head == NULL) | |
638f5b90 | 747 | return -ENOENT; |
17a52670 | 748 | |
638f5b90 AS |
749 | if (cur) { |
750 | err = copy_verifier_state(cur, &head->st); | |
751 | if (err) | |
752 | return err; | |
753 | } | |
754 | if (insn_idx) | |
755 | *insn_idx = head->insn_idx; | |
17a52670 | 756 | if (prev_insn_idx) |
638f5b90 AS |
757 | *prev_insn_idx = head->prev_insn_idx; |
758 | elem = head->next; | |
1969db47 | 759 | free_verifier_state(&head->st, false); |
638f5b90 | 760 | kfree(head); |
17a52670 AS |
761 | env->head = elem; |
762 | env->stack_size--; | |
638f5b90 | 763 | return 0; |
17a52670 AS |
764 | } |
765 | ||
58e2af8b | 766 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
979d63d5 DB |
767 | int insn_idx, int prev_insn_idx, |
768 | bool speculative) | |
17a52670 | 769 | { |
638f5b90 | 770 | struct bpf_verifier_state *cur = env->cur_state; |
58e2af8b | 771 | struct bpf_verifier_stack_elem *elem; |
638f5b90 | 772 | int err; |
17a52670 | 773 | |
638f5b90 | 774 | elem = kzalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
775 | if (!elem) |
776 | goto err; | |
777 | ||
17a52670 AS |
778 | elem->insn_idx = insn_idx; |
779 | elem->prev_insn_idx = prev_insn_idx; | |
780 | elem->next = env->head; | |
781 | env->head = elem; | |
782 | env->stack_size++; | |
1969db47 AS |
783 | err = copy_verifier_state(&elem->st, cur); |
784 | if (err) | |
785 | goto err; | |
979d63d5 | 786 | elem->st.speculative |= speculative; |
b285fcb7 AS |
787 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_JMP_SEQ) { |
788 | verbose(env, "The sequence of %d jumps is too complex.\n", | |
789 | env->stack_size); | |
17a52670 AS |
790 | goto err; |
791 | } | |
792 | return &elem->st; | |
793 | err: | |
5896351e AS |
794 | free_verifier_state(env->cur_state, true); |
795 | env->cur_state = NULL; | |
17a52670 | 796 | /* pop all elements and return */ |
638f5b90 | 797 | while (!pop_stack(env, NULL, NULL)); |
17a52670 AS |
798 | return NULL; |
799 | } | |
800 | ||
801 | #define CALLER_SAVED_REGS 6 | |
802 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
803 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
804 | }; | |
805 | ||
f1174f77 EC |
806 | static void __mark_reg_not_init(struct bpf_reg_state *reg); |
807 | ||
b03c9f9f EC |
808 | /* Mark the unknown part of a register (variable offset or scalar value) as |
809 | * known to have the value @imm. | |
810 | */ | |
811 | static void __mark_reg_known(struct bpf_reg_state *reg, u64 imm) | |
812 | { | |
a9c676bc AS |
813 | /* Clear id, off, and union(map_ptr, range) */ |
814 | memset(((u8 *)reg) + sizeof(reg->type), 0, | |
815 | offsetof(struct bpf_reg_state, var_off) - sizeof(reg->type)); | |
b03c9f9f EC |
816 | reg->var_off = tnum_const(imm); |
817 | reg->smin_value = (s64)imm; | |
818 | reg->smax_value = (s64)imm; | |
819 | reg->umin_value = imm; | |
820 | reg->umax_value = imm; | |
821 | } | |
822 | ||
f1174f77 EC |
823 | /* Mark the 'variable offset' part of a register as zero. This should be |
824 | * used only on registers holding a pointer type. | |
825 | */ | |
826 | static void __mark_reg_known_zero(struct bpf_reg_state *reg) | |
a9789ef9 | 827 | { |
b03c9f9f | 828 | __mark_reg_known(reg, 0); |
f1174f77 | 829 | } |
a9789ef9 | 830 | |
cc2b14d5 AS |
831 | static void __mark_reg_const_zero(struct bpf_reg_state *reg) |
832 | { | |
833 | __mark_reg_known(reg, 0); | |
cc2b14d5 AS |
834 | reg->type = SCALAR_VALUE; |
835 | } | |
836 | ||
61bd5218 JK |
837 | static void mark_reg_known_zero(struct bpf_verifier_env *env, |
838 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
839 | { |
840 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 841 | verbose(env, "mark_reg_known_zero(regs, %u)\n", regno); |
f1174f77 EC |
842 | /* Something bad happened, let's kill all regs */ |
843 | for (regno = 0; regno < MAX_BPF_REG; regno++) | |
844 | __mark_reg_not_init(regs + regno); | |
845 | return; | |
846 | } | |
847 | __mark_reg_known_zero(regs + regno); | |
848 | } | |
849 | ||
de8f3a83 DB |
850 | static bool reg_is_pkt_pointer(const struct bpf_reg_state *reg) |
851 | { | |
852 | return type_is_pkt_pointer(reg->type); | |
853 | } | |
854 | ||
855 | static bool reg_is_pkt_pointer_any(const struct bpf_reg_state *reg) | |
856 | { | |
857 | return reg_is_pkt_pointer(reg) || | |
858 | reg->type == PTR_TO_PACKET_END; | |
859 | } | |
860 | ||
861 | /* Unmodified PTR_TO_PACKET[_META,_END] register from ctx access. */ | |
862 | static bool reg_is_init_pkt_pointer(const struct bpf_reg_state *reg, | |
863 | enum bpf_reg_type which) | |
864 | { | |
865 | /* The register can already have a range from prior markings. | |
866 | * This is fine as long as it hasn't been advanced from its | |
867 | * origin. | |
868 | */ | |
869 | return reg->type == which && | |
870 | reg->id == 0 && | |
871 | reg->off == 0 && | |
872 | tnum_equals_const(reg->var_off, 0); | |
873 | } | |
874 | ||
b03c9f9f EC |
875 | /* Attempts to improve min/max values based on var_off information */ |
876 | static void __update_reg_bounds(struct bpf_reg_state *reg) | |
877 | { | |
878 | /* min signed is max(sign bit) | min(other bits) */ | |
879 | reg->smin_value = max_t(s64, reg->smin_value, | |
880 | reg->var_off.value | (reg->var_off.mask & S64_MIN)); | |
881 | /* max signed is min(sign bit) | max(other bits) */ | |
882 | reg->smax_value = min_t(s64, reg->smax_value, | |
883 | reg->var_off.value | (reg->var_off.mask & S64_MAX)); | |
884 | reg->umin_value = max(reg->umin_value, reg->var_off.value); | |
885 | reg->umax_value = min(reg->umax_value, | |
886 | reg->var_off.value | reg->var_off.mask); | |
887 | } | |
888 | ||
889 | /* Uses signed min/max values to inform unsigned, and vice-versa */ | |
890 | static void __reg_deduce_bounds(struct bpf_reg_state *reg) | |
891 | { | |
892 | /* Learn sign from signed bounds. | |
893 | * If we cannot cross the sign boundary, then signed and unsigned bounds | |
894 | * are the same, so combine. This works even in the negative case, e.g. | |
895 | * -3 s<= x s<= -1 implies 0xf...fd u<= x u<= 0xf...ff. | |
896 | */ | |
897 | if (reg->smin_value >= 0 || reg->smax_value < 0) { | |
898 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
899 | reg->umin_value); | |
900 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
901 | reg->umax_value); | |
902 | return; | |
903 | } | |
904 | /* Learn sign from unsigned bounds. Signed bounds cross the sign | |
905 | * boundary, so we must be careful. | |
906 | */ | |
907 | if ((s64)reg->umax_value >= 0) { | |
908 | /* Positive. We can't learn anything from the smin, but smax | |
909 | * is positive, hence safe. | |
910 | */ | |
911 | reg->smin_value = reg->umin_value; | |
912 | reg->smax_value = reg->umax_value = min_t(u64, reg->smax_value, | |
913 | reg->umax_value); | |
914 | } else if ((s64)reg->umin_value < 0) { | |
915 | /* Negative. We can't learn anything from the smax, but smin | |
916 | * is negative, hence safe. | |
917 | */ | |
918 | reg->smin_value = reg->umin_value = max_t(u64, reg->smin_value, | |
919 | reg->umin_value); | |
920 | reg->smax_value = reg->umax_value; | |
921 | } | |
922 | } | |
923 | ||
924 | /* Attempts to improve var_off based on unsigned min/max information */ | |
925 | static void __reg_bound_offset(struct bpf_reg_state *reg) | |
926 | { | |
927 | reg->var_off = tnum_intersect(reg->var_off, | |
928 | tnum_range(reg->umin_value, | |
929 | reg->umax_value)); | |
930 | } | |
931 | ||
932 | /* Reset the min/max bounds of a register */ | |
933 | static void __mark_reg_unbounded(struct bpf_reg_state *reg) | |
934 | { | |
935 | reg->smin_value = S64_MIN; | |
936 | reg->smax_value = S64_MAX; | |
937 | reg->umin_value = 0; | |
938 | reg->umax_value = U64_MAX; | |
939 | } | |
940 | ||
f1174f77 EC |
941 | /* Mark a register as having a completely unknown (scalar) value. */ |
942 | static void __mark_reg_unknown(struct bpf_reg_state *reg) | |
943 | { | |
a9c676bc AS |
944 | /* |
945 | * Clear type, id, off, and union(map_ptr, range) and | |
946 | * padding between 'type' and union | |
947 | */ | |
948 | memset(reg, 0, offsetof(struct bpf_reg_state, var_off)); | |
f1174f77 | 949 | reg->type = SCALAR_VALUE; |
f1174f77 | 950 | reg->var_off = tnum_unknown; |
f4d7e40a | 951 | reg->frameno = 0; |
b03c9f9f | 952 | __mark_reg_unbounded(reg); |
f1174f77 EC |
953 | } |
954 | ||
61bd5218 JK |
955 | static void mark_reg_unknown(struct bpf_verifier_env *env, |
956 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
957 | { |
958 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 959 | verbose(env, "mark_reg_unknown(regs, %u)\n", regno); |
19ceb417 AS |
960 | /* Something bad happened, let's kill all regs except FP */ |
961 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
962 | __mark_reg_not_init(regs + regno); |
963 | return; | |
964 | } | |
965 | __mark_reg_unknown(regs + regno); | |
966 | } | |
967 | ||
968 | static void __mark_reg_not_init(struct bpf_reg_state *reg) | |
969 | { | |
970 | __mark_reg_unknown(reg); | |
971 | reg->type = NOT_INIT; | |
972 | } | |
973 | ||
61bd5218 JK |
974 | static void mark_reg_not_init(struct bpf_verifier_env *env, |
975 | struct bpf_reg_state *regs, u32 regno) | |
f1174f77 EC |
976 | { |
977 | if (WARN_ON(regno >= MAX_BPF_REG)) { | |
61bd5218 | 978 | verbose(env, "mark_reg_not_init(regs, %u)\n", regno); |
19ceb417 AS |
979 | /* Something bad happened, let's kill all regs except FP */ |
980 | for (regno = 0; regno < BPF_REG_FP; regno++) | |
f1174f77 EC |
981 | __mark_reg_not_init(regs + regno); |
982 | return; | |
983 | } | |
984 | __mark_reg_not_init(regs + regno); | |
a9789ef9 DB |
985 | } |
986 | ||
5327ed3d | 987 | #define DEF_NOT_SUBREG (0) |
61bd5218 | 988 | static void init_reg_state(struct bpf_verifier_env *env, |
f4d7e40a | 989 | struct bpf_func_state *state) |
17a52670 | 990 | { |
f4d7e40a | 991 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
992 | int i; |
993 | ||
dc503a8a | 994 | for (i = 0; i < MAX_BPF_REG; i++) { |
61bd5218 | 995 | mark_reg_not_init(env, regs, i); |
dc503a8a | 996 | regs[i].live = REG_LIVE_NONE; |
679c782d | 997 | regs[i].parent = NULL; |
5327ed3d | 998 | regs[i].subreg_def = DEF_NOT_SUBREG; |
dc503a8a | 999 | } |
17a52670 AS |
1000 | |
1001 | /* frame pointer */ | |
f1174f77 | 1002 | regs[BPF_REG_FP].type = PTR_TO_STACK; |
61bd5218 | 1003 | mark_reg_known_zero(env, regs, BPF_REG_FP); |
f4d7e40a | 1004 | regs[BPF_REG_FP].frameno = state->frameno; |
17a52670 AS |
1005 | |
1006 | /* 1st arg to a function */ | |
1007 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
61bd5218 | 1008 | mark_reg_known_zero(env, regs, BPF_REG_1); |
6760bf2d DB |
1009 | } |
1010 | ||
f4d7e40a AS |
1011 | #define BPF_MAIN_FUNC (-1) |
1012 | static void init_func_state(struct bpf_verifier_env *env, | |
1013 | struct bpf_func_state *state, | |
1014 | int callsite, int frameno, int subprogno) | |
1015 | { | |
1016 | state->callsite = callsite; | |
1017 | state->frameno = frameno; | |
1018 | state->subprogno = subprogno; | |
1019 | init_reg_state(env, state); | |
1020 | } | |
1021 | ||
17a52670 AS |
1022 | enum reg_arg_type { |
1023 | SRC_OP, /* register is used as source operand */ | |
1024 | DST_OP, /* register is used as destination operand */ | |
1025 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
1026 | }; | |
1027 | ||
cc8b0b92 AS |
1028 | static int cmp_subprogs(const void *a, const void *b) |
1029 | { | |
9c8105bd JW |
1030 | return ((struct bpf_subprog_info *)a)->start - |
1031 | ((struct bpf_subprog_info *)b)->start; | |
cc8b0b92 AS |
1032 | } |
1033 | ||
1034 | static int find_subprog(struct bpf_verifier_env *env, int off) | |
1035 | { | |
9c8105bd | 1036 | struct bpf_subprog_info *p; |
cc8b0b92 | 1037 | |
9c8105bd JW |
1038 | p = bsearch(&off, env->subprog_info, env->subprog_cnt, |
1039 | sizeof(env->subprog_info[0]), cmp_subprogs); | |
cc8b0b92 AS |
1040 | if (!p) |
1041 | return -ENOENT; | |
9c8105bd | 1042 | return p - env->subprog_info; |
cc8b0b92 AS |
1043 | |
1044 | } | |
1045 | ||
1046 | static int add_subprog(struct bpf_verifier_env *env, int off) | |
1047 | { | |
1048 | int insn_cnt = env->prog->len; | |
1049 | int ret; | |
1050 | ||
1051 | if (off >= insn_cnt || off < 0) { | |
1052 | verbose(env, "call to invalid destination\n"); | |
1053 | return -EINVAL; | |
1054 | } | |
1055 | ret = find_subprog(env, off); | |
1056 | if (ret >= 0) | |
1057 | return 0; | |
4cb3d99c | 1058 | if (env->subprog_cnt >= BPF_MAX_SUBPROGS) { |
cc8b0b92 AS |
1059 | verbose(env, "too many subprograms\n"); |
1060 | return -E2BIG; | |
1061 | } | |
9c8105bd JW |
1062 | env->subprog_info[env->subprog_cnt++].start = off; |
1063 | sort(env->subprog_info, env->subprog_cnt, | |
1064 | sizeof(env->subprog_info[0]), cmp_subprogs, NULL); | |
cc8b0b92 AS |
1065 | return 0; |
1066 | } | |
1067 | ||
1068 | static int check_subprogs(struct bpf_verifier_env *env) | |
1069 | { | |
1070 | int i, ret, subprog_start, subprog_end, off, cur_subprog = 0; | |
9c8105bd | 1071 | struct bpf_subprog_info *subprog = env->subprog_info; |
cc8b0b92 AS |
1072 | struct bpf_insn *insn = env->prog->insnsi; |
1073 | int insn_cnt = env->prog->len; | |
1074 | ||
f910cefa JW |
1075 | /* Add entry function. */ |
1076 | ret = add_subprog(env, 0); | |
1077 | if (ret < 0) | |
1078 | return ret; | |
1079 | ||
cc8b0b92 AS |
1080 | /* determine subprog starts. The end is one before the next starts */ |
1081 | for (i = 0; i < insn_cnt; i++) { | |
1082 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
1083 | continue; | |
1084 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
1085 | continue; | |
1086 | if (!env->allow_ptr_leaks) { | |
1087 | verbose(env, "function calls to other bpf functions are allowed for root only\n"); | |
1088 | return -EPERM; | |
1089 | } | |
cc8b0b92 AS |
1090 | ret = add_subprog(env, i + insn[i].imm + 1); |
1091 | if (ret < 0) | |
1092 | return ret; | |
1093 | } | |
1094 | ||
4cb3d99c JW |
1095 | /* Add a fake 'exit' subprog which could simplify subprog iteration |
1096 | * logic. 'subprog_cnt' should not be increased. | |
1097 | */ | |
1098 | subprog[env->subprog_cnt].start = insn_cnt; | |
1099 | ||
06ee7115 | 1100 | if (env->log.level & BPF_LOG_LEVEL2) |
cc8b0b92 | 1101 | for (i = 0; i < env->subprog_cnt; i++) |
9c8105bd | 1102 | verbose(env, "func#%d @%d\n", i, subprog[i].start); |
cc8b0b92 AS |
1103 | |
1104 | /* now check that all jumps are within the same subprog */ | |
4cb3d99c JW |
1105 | subprog_start = subprog[cur_subprog].start; |
1106 | subprog_end = subprog[cur_subprog + 1].start; | |
cc8b0b92 AS |
1107 | for (i = 0; i < insn_cnt; i++) { |
1108 | u8 code = insn[i].code; | |
1109 | ||
092ed096 | 1110 | if (BPF_CLASS(code) != BPF_JMP && BPF_CLASS(code) != BPF_JMP32) |
cc8b0b92 AS |
1111 | goto next; |
1112 | if (BPF_OP(code) == BPF_EXIT || BPF_OP(code) == BPF_CALL) | |
1113 | goto next; | |
1114 | off = i + insn[i].off + 1; | |
1115 | if (off < subprog_start || off >= subprog_end) { | |
1116 | verbose(env, "jump out of range from insn %d to %d\n", i, off); | |
1117 | return -EINVAL; | |
1118 | } | |
1119 | next: | |
1120 | if (i == subprog_end - 1) { | |
1121 | /* to avoid fall-through from one subprog into another | |
1122 | * the last insn of the subprog should be either exit | |
1123 | * or unconditional jump back | |
1124 | */ | |
1125 | if (code != (BPF_JMP | BPF_EXIT) && | |
1126 | code != (BPF_JMP | BPF_JA)) { | |
1127 | verbose(env, "last insn is not an exit or jmp\n"); | |
1128 | return -EINVAL; | |
1129 | } | |
1130 | subprog_start = subprog_end; | |
4cb3d99c JW |
1131 | cur_subprog++; |
1132 | if (cur_subprog < env->subprog_cnt) | |
9c8105bd | 1133 | subprog_end = subprog[cur_subprog + 1].start; |
cc8b0b92 AS |
1134 | } |
1135 | } | |
1136 | return 0; | |
1137 | } | |
1138 | ||
679c782d EC |
1139 | /* Parentage chain of this register (or stack slot) should take care of all |
1140 | * issues like callee-saved registers, stack slot allocation time, etc. | |
1141 | */ | |
f4d7e40a | 1142 | static int mark_reg_read(struct bpf_verifier_env *env, |
679c782d | 1143 | const struct bpf_reg_state *state, |
5327ed3d | 1144 | struct bpf_reg_state *parent, u8 flag) |
f4d7e40a AS |
1145 | { |
1146 | bool writes = parent == state->parent; /* Observe write marks */ | |
06ee7115 | 1147 | int cnt = 0; |
dc503a8a EC |
1148 | |
1149 | while (parent) { | |
1150 | /* if read wasn't screened by an earlier write ... */ | |
679c782d | 1151 | if (writes && state->live & REG_LIVE_WRITTEN) |
dc503a8a | 1152 | break; |
9242b5f5 AS |
1153 | if (parent->live & REG_LIVE_DONE) { |
1154 | verbose(env, "verifier BUG type %s var_off %lld off %d\n", | |
1155 | reg_type_str[parent->type], | |
1156 | parent->var_off.value, parent->off); | |
1157 | return -EFAULT; | |
1158 | } | |
5327ed3d JW |
1159 | /* The first condition is more likely to be true than the |
1160 | * second, checked it first. | |
1161 | */ | |
1162 | if ((parent->live & REG_LIVE_READ) == flag || | |
1163 | parent->live & REG_LIVE_READ64) | |
25af32da AS |
1164 | /* The parentage chain never changes and |
1165 | * this parent was already marked as LIVE_READ. | |
1166 | * There is no need to keep walking the chain again and | |
1167 | * keep re-marking all parents as LIVE_READ. | |
1168 | * This case happens when the same register is read | |
1169 | * multiple times without writes into it in-between. | |
5327ed3d JW |
1170 | * Also, if parent has the stronger REG_LIVE_READ64 set, |
1171 | * then no need to set the weak REG_LIVE_READ32. | |
25af32da AS |
1172 | */ |
1173 | break; | |
dc503a8a | 1174 | /* ... then we depend on parent's value */ |
5327ed3d JW |
1175 | parent->live |= flag; |
1176 | /* REG_LIVE_READ64 overrides REG_LIVE_READ32. */ | |
1177 | if (flag == REG_LIVE_READ64) | |
1178 | parent->live &= ~REG_LIVE_READ32; | |
dc503a8a EC |
1179 | state = parent; |
1180 | parent = state->parent; | |
f4d7e40a | 1181 | writes = true; |
06ee7115 | 1182 | cnt++; |
dc503a8a | 1183 | } |
06ee7115 AS |
1184 | |
1185 | if (env->longest_mark_read_walk < cnt) | |
1186 | env->longest_mark_read_walk = cnt; | |
f4d7e40a | 1187 | return 0; |
dc503a8a EC |
1188 | } |
1189 | ||
5327ed3d JW |
1190 | /* This function is supposed to be used by the following 32-bit optimization |
1191 | * code only. It returns TRUE if the source or destination register operates | |
1192 | * on 64-bit, otherwise return FALSE. | |
1193 | */ | |
1194 | static bool is_reg64(struct bpf_verifier_env *env, struct bpf_insn *insn, | |
1195 | u32 regno, struct bpf_reg_state *reg, enum reg_arg_type t) | |
1196 | { | |
1197 | u8 code, class, op; | |
1198 | ||
1199 | code = insn->code; | |
1200 | class = BPF_CLASS(code); | |
1201 | op = BPF_OP(code); | |
1202 | if (class == BPF_JMP) { | |
1203 | /* BPF_EXIT for "main" will reach here. Return TRUE | |
1204 | * conservatively. | |
1205 | */ | |
1206 | if (op == BPF_EXIT) | |
1207 | return true; | |
1208 | if (op == BPF_CALL) { | |
1209 | /* BPF to BPF call will reach here because of marking | |
1210 | * caller saved clobber with DST_OP_NO_MARK for which we | |
1211 | * don't care the register def because they are anyway | |
1212 | * marked as NOT_INIT already. | |
1213 | */ | |
1214 | if (insn->src_reg == BPF_PSEUDO_CALL) | |
1215 | return false; | |
1216 | /* Helper call will reach here because of arg type | |
1217 | * check, conservatively return TRUE. | |
1218 | */ | |
1219 | if (t == SRC_OP) | |
1220 | return true; | |
1221 | ||
1222 | return false; | |
1223 | } | |
1224 | } | |
1225 | ||
1226 | if (class == BPF_ALU64 || class == BPF_JMP || | |
1227 | /* BPF_END always use BPF_ALU class. */ | |
1228 | (class == BPF_ALU && op == BPF_END && insn->imm == 64)) | |
1229 | return true; | |
1230 | ||
1231 | if (class == BPF_ALU || class == BPF_JMP32) | |
1232 | return false; | |
1233 | ||
1234 | if (class == BPF_LDX) { | |
1235 | if (t != SRC_OP) | |
1236 | return BPF_SIZE(code) == BPF_DW; | |
1237 | /* LDX source must be ptr. */ | |
1238 | return true; | |
1239 | } | |
1240 | ||
1241 | if (class == BPF_STX) { | |
1242 | if (reg->type != SCALAR_VALUE) | |
1243 | return true; | |
1244 | return BPF_SIZE(code) == BPF_DW; | |
1245 | } | |
1246 | ||
1247 | if (class == BPF_LD) { | |
1248 | u8 mode = BPF_MODE(code); | |
1249 | ||
1250 | /* LD_IMM64 */ | |
1251 | if (mode == BPF_IMM) | |
1252 | return true; | |
1253 | ||
1254 | /* Both LD_IND and LD_ABS return 32-bit data. */ | |
1255 | if (t != SRC_OP) | |
1256 | return false; | |
1257 | ||
1258 | /* Implicit ctx ptr. */ | |
1259 | if (regno == BPF_REG_6) | |
1260 | return true; | |
1261 | ||
1262 | /* Explicit source could be any width. */ | |
1263 | return true; | |
1264 | } | |
1265 | ||
1266 | if (class == BPF_ST) | |
1267 | /* The only source register for BPF_ST is a ptr. */ | |
1268 | return true; | |
1269 | ||
1270 | /* Conservatively return true at default. */ | |
1271 | return true; | |
1272 | } | |
1273 | ||
b325fbca JW |
1274 | /* Return TRUE if INSN doesn't have explicit value define. */ |
1275 | static bool insn_no_def(struct bpf_insn *insn) | |
1276 | { | |
1277 | u8 class = BPF_CLASS(insn->code); | |
1278 | ||
1279 | return (class == BPF_JMP || class == BPF_JMP32 || | |
1280 | class == BPF_STX || class == BPF_ST); | |
1281 | } | |
1282 | ||
1283 | /* Return TRUE if INSN has defined any 32-bit value explicitly. */ | |
1284 | static bool insn_has_def32(struct bpf_verifier_env *env, struct bpf_insn *insn) | |
1285 | { | |
1286 | if (insn_no_def(insn)) | |
1287 | return false; | |
1288 | ||
1289 | return !is_reg64(env, insn, insn->dst_reg, NULL, DST_OP); | |
1290 | } | |
1291 | ||
5327ed3d JW |
1292 | static void mark_insn_zext(struct bpf_verifier_env *env, |
1293 | struct bpf_reg_state *reg) | |
1294 | { | |
1295 | s32 def_idx = reg->subreg_def; | |
1296 | ||
1297 | if (def_idx == DEF_NOT_SUBREG) | |
1298 | return; | |
1299 | ||
1300 | env->insn_aux_data[def_idx - 1].zext_dst = true; | |
1301 | /* The dst will be zero extended, so won't be sub-register anymore. */ | |
1302 | reg->subreg_def = DEF_NOT_SUBREG; | |
1303 | } | |
1304 | ||
dc503a8a | 1305 | static int check_reg_arg(struct bpf_verifier_env *env, u32 regno, |
17a52670 AS |
1306 | enum reg_arg_type t) |
1307 | { | |
f4d7e40a AS |
1308 | struct bpf_verifier_state *vstate = env->cur_state; |
1309 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
5327ed3d | 1310 | struct bpf_insn *insn = env->prog->insnsi + env->insn_idx; |
c342dc10 | 1311 | struct bpf_reg_state *reg, *regs = state->regs; |
5327ed3d | 1312 | bool rw64; |
dc503a8a | 1313 | |
17a52670 | 1314 | if (regno >= MAX_BPF_REG) { |
61bd5218 | 1315 | verbose(env, "R%d is invalid\n", regno); |
17a52670 AS |
1316 | return -EINVAL; |
1317 | } | |
1318 | ||
c342dc10 | 1319 | reg = ®s[regno]; |
5327ed3d | 1320 | rw64 = is_reg64(env, insn, regno, reg, t); |
17a52670 AS |
1321 | if (t == SRC_OP) { |
1322 | /* check whether register used as source operand can be read */ | |
c342dc10 | 1323 | if (reg->type == NOT_INIT) { |
61bd5218 | 1324 | verbose(env, "R%d !read_ok\n", regno); |
17a52670 AS |
1325 | return -EACCES; |
1326 | } | |
679c782d | 1327 | /* We don't need to worry about FP liveness because it's read-only */ |
c342dc10 JW |
1328 | if (regno == BPF_REG_FP) |
1329 | return 0; | |
1330 | ||
5327ed3d JW |
1331 | if (rw64) |
1332 | mark_insn_zext(env, reg); | |
1333 | ||
1334 | return mark_reg_read(env, reg, reg->parent, | |
1335 | rw64 ? REG_LIVE_READ64 : REG_LIVE_READ32); | |
17a52670 AS |
1336 | } else { |
1337 | /* check whether register used as dest operand can be written to */ | |
1338 | if (regno == BPF_REG_FP) { | |
61bd5218 | 1339 | verbose(env, "frame pointer is read only\n"); |
17a52670 AS |
1340 | return -EACCES; |
1341 | } | |
c342dc10 | 1342 | reg->live |= REG_LIVE_WRITTEN; |
5327ed3d | 1343 | reg->subreg_def = rw64 ? DEF_NOT_SUBREG : env->insn_idx + 1; |
17a52670 | 1344 | if (t == DST_OP) |
61bd5218 | 1345 | mark_reg_unknown(env, regs, regno); |
17a52670 AS |
1346 | } |
1347 | return 0; | |
1348 | } | |
1349 | ||
1be7f75d AS |
1350 | static bool is_spillable_regtype(enum bpf_reg_type type) |
1351 | { | |
1352 | switch (type) { | |
1353 | case PTR_TO_MAP_VALUE: | |
1354 | case PTR_TO_MAP_VALUE_OR_NULL: | |
1355 | case PTR_TO_STACK: | |
1356 | case PTR_TO_CTX: | |
969bf05e | 1357 | case PTR_TO_PACKET: |
de8f3a83 | 1358 | case PTR_TO_PACKET_META: |
969bf05e | 1359 | case PTR_TO_PACKET_END: |
d58e468b | 1360 | case PTR_TO_FLOW_KEYS: |
1be7f75d | 1361 | case CONST_PTR_TO_MAP: |
c64b7983 JS |
1362 | case PTR_TO_SOCKET: |
1363 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
1364 | case PTR_TO_SOCK_COMMON: |
1365 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
1366 | case PTR_TO_TCP_SOCK: |
1367 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 1368 | case PTR_TO_XDP_SOCK: |
1be7f75d AS |
1369 | return true; |
1370 | default: | |
1371 | return false; | |
1372 | } | |
1373 | } | |
1374 | ||
cc2b14d5 AS |
1375 | /* Does this register contain a constant zero? */ |
1376 | static bool register_is_null(struct bpf_reg_state *reg) | |
1377 | { | |
1378 | return reg->type == SCALAR_VALUE && tnum_equals_const(reg->var_off, 0); | |
1379 | } | |
1380 | ||
f7cf25b2 AS |
1381 | static bool register_is_const(struct bpf_reg_state *reg) |
1382 | { | |
1383 | return reg->type == SCALAR_VALUE && tnum_is_const(reg->var_off); | |
1384 | } | |
1385 | ||
1386 | static void save_register_state(struct bpf_func_state *state, | |
1387 | int spi, struct bpf_reg_state *reg) | |
1388 | { | |
1389 | int i; | |
1390 | ||
1391 | state->stack[spi].spilled_ptr = *reg; | |
1392 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1393 | ||
1394 | for (i = 0; i < BPF_REG_SIZE; i++) | |
1395 | state->stack[spi].slot_type[i] = STACK_SPILL; | |
1396 | } | |
1397 | ||
17a52670 AS |
1398 | /* check_stack_read/write functions track spill/fill of registers, |
1399 | * stack boundary and alignment are checked in check_mem_access() | |
1400 | */ | |
61bd5218 | 1401 | static int check_stack_write(struct bpf_verifier_env *env, |
f4d7e40a | 1402 | struct bpf_func_state *state, /* func where register points to */ |
af86ca4e | 1403 | int off, int size, int value_regno, int insn_idx) |
17a52670 | 1404 | { |
f4d7e40a | 1405 | struct bpf_func_state *cur; /* state of the current function */ |
638f5b90 | 1406 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE, err; |
f7cf25b2 | 1407 | struct bpf_reg_state *reg = NULL; |
638f5b90 | 1408 | |
f4d7e40a | 1409 | err = realloc_func_state(state, round_up(slot + 1, BPF_REG_SIZE), |
fd978bf7 | 1410 | state->acquired_refs, true); |
638f5b90 AS |
1411 | if (err) |
1412 | return err; | |
9c399760 AS |
1413 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
1414 | * so it's aligned access and [off, off + size) are within stack limits | |
1415 | */ | |
638f5b90 AS |
1416 | if (!env->allow_ptr_leaks && |
1417 | state->stack[spi].slot_type[0] == STACK_SPILL && | |
1418 | size != BPF_REG_SIZE) { | |
1419 | verbose(env, "attempt to corrupt spilled pointer on stack\n"); | |
1420 | return -EACCES; | |
1421 | } | |
17a52670 | 1422 | |
f4d7e40a | 1423 | cur = env->cur_state->frame[env->cur_state->curframe]; |
f7cf25b2 AS |
1424 | if (value_regno >= 0) |
1425 | reg = &cur->regs[value_regno]; | |
17a52670 | 1426 | |
f7cf25b2 AS |
1427 | if (reg && size == BPF_REG_SIZE && register_is_const(reg) && |
1428 | !register_is_null(reg) && env->allow_ptr_leaks) { | |
1429 | save_register_state(state, spi, reg); | |
1430 | } else if (reg && is_spillable_regtype(reg->type)) { | |
17a52670 | 1431 | /* register containing pointer is being spilled into stack */ |
9c399760 | 1432 | if (size != BPF_REG_SIZE) { |
f7cf25b2 | 1433 | verbose_linfo(env, insn_idx, "; "); |
61bd5218 | 1434 | verbose(env, "invalid size of register spill\n"); |
17a52670 AS |
1435 | return -EACCES; |
1436 | } | |
1437 | ||
f7cf25b2 | 1438 | if (state != cur && reg->type == PTR_TO_STACK) { |
f4d7e40a AS |
1439 | verbose(env, "cannot spill pointers to stack into stack frame of the caller\n"); |
1440 | return -EINVAL; | |
1441 | } | |
1442 | ||
f7cf25b2 AS |
1443 | if (!env->allow_ptr_leaks) { |
1444 | bool sanitize = false; | |
17a52670 | 1445 | |
f7cf25b2 AS |
1446 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
1447 | register_is_const(&state->stack[spi].spilled_ptr)) | |
1448 | sanitize = true; | |
1449 | for (i = 0; i < BPF_REG_SIZE; i++) | |
1450 | if (state->stack[spi].slot_type[i] == STACK_MISC) { | |
1451 | sanitize = true; | |
1452 | break; | |
1453 | } | |
1454 | if (sanitize) { | |
af86ca4e AS |
1455 | int *poff = &env->insn_aux_data[insn_idx].sanitize_stack_off; |
1456 | int soff = (-spi - 1) * BPF_REG_SIZE; | |
1457 | ||
1458 | /* detected reuse of integer stack slot with a pointer | |
1459 | * which means either llvm is reusing stack slot or | |
1460 | * an attacker is trying to exploit CVE-2018-3639 | |
1461 | * (speculative store bypass) | |
1462 | * Have to sanitize that slot with preemptive | |
1463 | * store of zero. | |
1464 | */ | |
1465 | if (*poff && *poff != soff) { | |
1466 | /* disallow programs where single insn stores | |
1467 | * into two different stack slots, since verifier | |
1468 | * cannot sanitize them | |
1469 | */ | |
1470 | verbose(env, | |
1471 | "insn %d cannot access two stack slots fp%d and fp%d", | |
1472 | insn_idx, *poff, soff); | |
1473 | return -EINVAL; | |
1474 | } | |
1475 | *poff = soff; | |
1476 | } | |
af86ca4e | 1477 | } |
f7cf25b2 | 1478 | save_register_state(state, spi, reg); |
9c399760 | 1479 | } else { |
cc2b14d5 AS |
1480 | u8 type = STACK_MISC; |
1481 | ||
679c782d EC |
1482 | /* regular write of data into stack destroys any spilled ptr */ |
1483 | state->stack[spi].spilled_ptr.type = NOT_INIT; | |
0bae2d4d JW |
1484 | /* Mark slots as STACK_MISC if they belonged to spilled ptr. */ |
1485 | if (state->stack[spi].slot_type[0] == STACK_SPILL) | |
1486 | for (i = 0; i < BPF_REG_SIZE; i++) | |
1487 | state->stack[spi].slot_type[i] = STACK_MISC; | |
9c399760 | 1488 | |
cc2b14d5 AS |
1489 | /* only mark the slot as written if all 8 bytes were written |
1490 | * otherwise read propagation may incorrectly stop too soon | |
1491 | * when stack slots are partially written. | |
1492 | * This heuristic means that read propagation will be | |
1493 | * conservative, since it will add reg_live_read marks | |
1494 | * to stack slots all the way to first state when programs | |
1495 | * writes+reads less than 8 bytes | |
1496 | */ | |
1497 | if (size == BPF_REG_SIZE) | |
1498 | state->stack[spi].spilled_ptr.live |= REG_LIVE_WRITTEN; | |
1499 | ||
1500 | /* when we zero initialize stack slots mark them as such */ | |
f7cf25b2 | 1501 | if (reg && register_is_null(reg)) |
cc2b14d5 AS |
1502 | type = STACK_ZERO; |
1503 | ||
0bae2d4d | 1504 | /* Mark slots affected by this stack write. */ |
9c399760 | 1505 | for (i = 0; i < size; i++) |
638f5b90 | 1506 | state->stack[spi].slot_type[(slot - i) % BPF_REG_SIZE] = |
cc2b14d5 | 1507 | type; |
17a52670 AS |
1508 | } |
1509 | return 0; | |
1510 | } | |
1511 | ||
61bd5218 | 1512 | static int check_stack_read(struct bpf_verifier_env *env, |
f4d7e40a AS |
1513 | struct bpf_func_state *reg_state /* func where register points to */, |
1514 | int off, int size, int value_regno) | |
17a52670 | 1515 | { |
f4d7e40a AS |
1516 | struct bpf_verifier_state *vstate = env->cur_state; |
1517 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
638f5b90 | 1518 | int i, slot = -off - 1, spi = slot / BPF_REG_SIZE; |
f7cf25b2 | 1519 | struct bpf_reg_state *reg; |
638f5b90 | 1520 | u8 *stype; |
17a52670 | 1521 | |
f4d7e40a | 1522 | if (reg_state->allocated_stack <= slot) { |
638f5b90 AS |
1523 | verbose(env, "invalid read from stack off %d+0 size %d\n", |
1524 | off, size); | |
1525 | return -EACCES; | |
1526 | } | |
f4d7e40a | 1527 | stype = reg_state->stack[spi].slot_type; |
f7cf25b2 | 1528 | reg = ®_state->stack[spi].spilled_ptr; |
17a52670 | 1529 | |
638f5b90 | 1530 | if (stype[0] == STACK_SPILL) { |
9c399760 | 1531 | if (size != BPF_REG_SIZE) { |
f7cf25b2 AS |
1532 | if (reg->type != SCALAR_VALUE) { |
1533 | verbose_linfo(env, env->insn_idx, "; "); | |
1534 | verbose(env, "invalid size of register fill\n"); | |
1535 | return -EACCES; | |
1536 | } | |
1537 | if (value_regno >= 0) { | |
1538 | mark_reg_unknown(env, state->regs, value_regno); | |
1539 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
1540 | } | |
1541 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); | |
1542 | return 0; | |
17a52670 | 1543 | } |
9c399760 | 1544 | for (i = 1; i < BPF_REG_SIZE; i++) { |
638f5b90 | 1545 | if (stype[(slot - i) % BPF_REG_SIZE] != STACK_SPILL) { |
61bd5218 | 1546 | verbose(env, "corrupted spill memory\n"); |
17a52670 AS |
1547 | return -EACCES; |
1548 | } | |
1549 | } | |
1550 | ||
dc503a8a | 1551 | if (value_regno >= 0) { |
17a52670 | 1552 | /* restore register state from stack */ |
f7cf25b2 | 1553 | state->regs[value_regno] = *reg; |
2f18f62e AS |
1554 | /* mark reg as written since spilled pointer state likely |
1555 | * has its liveness marks cleared by is_state_visited() | |
1556 | * which resets stack/reg liveness for state transitions | |
1557 | */ | |
1558 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
dc503a8a | 1559 | } |
f7cf25b2 | 1560 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
17a52670 | 1561 | } else { |
cc2b14d5 AS |
1562 | int zeros = 0; |
1563 | ||
17a52670 | 1564 | for (i = 0; i < size; i++) { |
cc2b14d5 AS |
1565 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_MISC) |
1566 | continue; | |
1567 | if (stype[(slot - i) % BPF_REG_SIZE] == STACK_ZERO) { | |
1568 | zeros++; | |
1569 | continue; | |
17a52670 | 1570 | } |
cc2b14d5 AS |
1571 | verbose(env, "invalid read from stack off %d+%d size %d\n", |
1572 | off, i, size); | |
1573 | return -EACCES; | |
1574 | } | |
f7cf25b2 | 1575 | mark_reg_read(env, reg, reg->parent, REG_LIVE_READ64); |
cc2b14d5 AS |
1576 | if (value_regno >= 0) { |
1577 | if (zeros == size) { | |
1578 | /* any size read into register is zero extended, | |
1579 | * so the whole register == const_zero | |
1580 | */ | |
1581 | __mark_reg_const_zero(&state->regs[value_regno]); | |
1582 | } else { | |
1583 | /* have read misc data from the stack */ | |
1584 | mark_reg_unknown(env, state->regs, value_regno); | |
1585 | } | |
1586 | state->regs[value_regno].live |= REG_LIVE_WRITTEN; | |
17a52670 | 1587 | } |
17a52670 | 1588 | } |
f7cf25b2 | 1589 | return 0; |
17a52670 AS |
1590 | } |
1591 | ||
e4298d25 DB |
1592 | static int check_stack_access(struct bpf_verifier_env *env, |
1593 | const struct bpf_reg_state *reg, | |
1594 | int off, int size) | |
1595 | { | |
1596 | /* Stack accesses must be at a fixed offset, so that we | |
1597 | * can determine what type of data were returned. See | |
1598 | * check_stack_read(). | |
1599 | */ | |
1600 | if (!tnum_is_const(reg->var_off)) { | |
1601 | char tn_buf[48]; | |
1602 | ||
1603 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
1fbd20f8 | 1604 | verbose(env, "variable stack access var_off=%s off=%d size=%d\n", |
e4298d25 DB |
1605 | tn_buf, off, size); |
1606 | return -EACCES; | |
1607 | } | |
1608 | ||
1609 | if (off >= 0 || off < -MAX_BPF_STACK) { | |
1610 | verbose(env, "invalid stack off=%d size=%d\n", off, size); | |
1611 | return -EACCES; | |
1612 | } | |
1613 | ||
1614 | return 0; | |
1615 | } | |
1616 | ||
591fe988 DB |
1617 | static int check_map_access_type(struct bpf_verifier_env *env, u32 regno, |
1618 | int off, int size, enum bpf_access_type type) | |
1619 | { | |
1620 | struct bpf_reg_state *regs = cur_regs(env); | |
1621 | struct bpf_map *map = regs[regno].map_ptr; | |
1622 | u32 cap = bpf_map_flags_to_cap(map); | |
1623 | ||
1624 | if (type == BPF_WRITE && !(cap & BPF_MAP_CAN_WRITE)) { | |
1625 | verbose(env, "write into map forbidden, value_size=%d off=%d size=%d\n", | |
1626 | map->value_size, off, size); | |
1627 | return -EACCES; | |
1628 | } | |
1629 | ||
1630 | if (type == BPF_READ && !(cap & BPF_MAP_CAN_READ)) { | |
1631 | verbose(env, "read from map forbidden, value_size=%d off=%d size=%d\n", | |
1632 | map->value_size, off, size); | |
1633 | return -EACCES; | |
1634 | } | |
1635 | ||
1636 | return 0; | |
1637 | } | |
1638 | ||
17a52670 | 1639 | /* check read/write into map element returned by bpf_map_lookup_elem() */ |
f1174f77 | 1640 | static int __check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1641 | int size, bool zero_size_allowed) |
17a52670 | 1642 | { |
638f5b90 AS |
1643 | struct bpf_reg_state *regs = cur_regs(env); |
1644 | struct bpf_map *map = regs[regno].map_ptr; | |
17a52670 | 1645 | |
9fd29c08 YS |
1646 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1647 | off + size > map->value_size) { | |
61bd5218 | 1648 | verbose(env, "invalid access to map value, value_size=%d off=%d size=%d\n", |
17a52670 AS |
1649 | map->value_size, off, size); |
1650 | return -EACCES; | |
1651 | } | |
1652 | return 0; | |
1653 | } | |
1654 | ||
f1174f77 EC |
1655 | /* check read/write into a map element with possible variable offset */ |
1656 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, | |
9fd29c08 | 1657 | int off, int size, bool zero_size_allowed) |
dbcfe5f7 | 1658 | { |
f4d7e40a AS |
1659 | struct bpf_verifier_state *vstate = env->cur_state; |
1660 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
dbcfe5f7 GB |
1661 | struct bpf_reg_state *reg = &state->regs[regno]; |
1662 | int err; | |
1663 | ||
f1174f77 EC |
1664 | /* We may have adjusted the register to this map value, so we |
1665 | * need to try adding each of min_value and max_value to off | |
1666 | * to make sure our theoretical access will be safe. | |
dbcfe5f7 | 1667 | */ |
06ee7115 | 1668 | if (env->log.level & BPF_LOG_LEVEL) |
61bd5218 | 1669 | print_verifier_state(env, state); |
b7137c4e | 1670 | |
dbcfe5f7 GB |
1671 | /* The minimum value is only important with signed |
1672 | * comparisons where we can't assume the floor of a | |
1673 | * value is 0. If we are using signed variables for our | |
1674 | * index'es we need to make sure that whatever we use | |
1675 | * will have a set floor within our range. | |
1676 | */ | |
b7137c4e DB |
1677 | if (reg->smin_value < 0 && |
1678 | (reg->smin_value == S64_MIN || | |
1679 | (off + reg->smin_value != (s64)(s32)(off + reg->smin_value)) || | |
1680 | reg->smin_value + off < 0)) { | |
61bd5218 | 1681 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
dbcfe5f7 GB |
1682 | regno); |
1683 | return -EACCES; | |
1684 | } | |
9fd29c08 YS |
1685 | err = __check_map_access(env, regno, reg->smin_value + off, size, |
1686 | zero_size_allowed); | |
dbcfe5f7 | 1687 | if (err) { |
61bd5218 JK |
1688 | verbose(env, "R%d min value is outside of the array range\n", |
1689 | regno); | |
dbcfe5f7 GB |
1690 | return err; |
1691 | } | |
1692 | ||
b03c9f9f EC |
1693 | /* If we haven't set a max value then we need to bail since we can't be |
1694 | * sure we won't do bad things. | |
1695 | * If reg->umax_value + off could overflow, treat that as unbounded too. | |
dbcfe5f7 | 1696 | */ |
b03c9f9f | 1697 | if (reg->umax_value >= BPF_MAX_VAR_OFF) { |
61bd5218 | 1698 | verbose(env, "R%d unbounded memory access, make sure to bounds check any array access into a map\n", |
dbcfe5f7 GB |
1699 | regno); |
1700 | return -EACCES; | |
1701 | } | |
9fd29c08 YS |
1702 | err = __check_map_access(env, regno, reg->umax_value + off, size, |
1703 | zero_size_allowed); | |
f1174f77 | 1704 | if (err) |
61bd5218 JK |
1705 | verbose(env, "R%d max value is outside of the array range\n", |
1706 | regno); | |
d83525ca AS |
1707 | |
1708 | if (map_value_has_spin_lock(reg->map_ptr)) { | |
1709 | u32 lock = reg->map_ptr->spin_lock_off; | |
1710 | ||
1711 | /* if any part of struct bpf_spin_lock can be touched by | |
1712 | * load/store reject this program. | |
1713 | * To check that [x1, x2) overlaps with [y1, y2) | |
1714 | * it is sufficient to check x1 < y2 && y1 < x2. | |
1715 | */ | |
1716 | if (reg->smin_value + off < lock + sizeof(struct bpf_spin_lock) && | |
1717 | lock < reg->umax_value + off + size) { | |
1718 | verbose(env, "bpf_spin_lock cannot be accessed directly by load/store\n"); | |
1719 | return -EACCES; | |
1720 | } | |
1721 | } | |
f1174f77 | 1722 | return err; |
dbcfe5f7 GB |
1723 | } |
1724 | ||
969bf05e AS |
1725 | #define MAX_PACKET_OFF 0xffff |
1726 | ||
58e2af8b | 1727 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
1728 | const struct bpf_call_arg_meta *meta, |
1729 | enum bpf_access_type t) | |
4acf6c0b | 1730 | { |
36bbef52 | 1731 | switch (env->prog->type) { |
5d66fa7d | 1732 | /* Program types only with direct read access go here! */ |
3a0af8fd TG |
1733 | case BPF_PROG_TYPE_LWT_IN: |
1734 | case BPF_PROG_TYPE_LWT_OUT: | |
004d4b27 | 1735 | case BPF_PROG_TYPE_LWT_SEG6LOCAL: |
2dbb9b9e | 1736 | case BPF_PROG_TYPE_SK_REUSEPORT: |
5d66fa7d | 1737 | case BPF_PROG_TYPE_FLOW_DISSECTOR: |
d5563d36 | 1738 | case BPF_PROG_TYPE_CGROUP_SKB: |
3a0af8fd TG |
1739 | if (t == BPF_WRITE) |
1740 | return false; | |
7e57fbb2 | 1741 | /* fallthrough */ |
5d66fa7d DB |
1742 | |
1743 | /* Program types with direct read + write access go here! */ | |
36bbef52 DB |
1744 | case BPF_PROG_TYPE_SCHED_CLS: |
1745 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 1746 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 1747 | case BPF_PROG_TYPE_LWT_XMIT: |
8a31db56 | 1748 | case BPF_PROG_TYPE_SK_SKB: |
4f738adb | 1749 | case BPF_PROG_TYPE_SK_MSG: |
36bbef52 DB |
1750 | if (meta) |
1751 | return meta->pkt_access; | |
1752 | ||
1753 | env->seen_direct_write = true; | |
4acf6c0b BB |
1754 | return true; |
1755 | default: | |
1756 | return false; | |
1757 | } | |
1758 | } | |
1759 | ||
f1174f77 | 1760 | static int __check_packet_access(struct bpf_verifier_env *env, u32 regno, |
9fd29c08 | 1761 | int off, int size, bool zero_size_allowed) |
969bf05e | 1762 | { |
638f5b90 | 1763 | struct bpf_reg_state *regs = cur_regs(env); |
58e2af8b | 1764 | struct bpf_reg_state *reg = ®s[regno]; |
969bf05e | 1765 | |
9fd29c08 YS |
1766 | if (off < 0 || size < 0 || (size == 0 && !zero_size_allowed) || |
1767 | (u64)off + size > reg->range) { | |
61bd5218 | 1768 | verbose(env, "invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
d91b28ed | 1769 | off, size, regno, reg->id, reg->off, reg->range); |
969bf05e AS |
1770 | return -EACCES; |
1771 | } | |
1772 | return 0; | |
1773 | } | |
1774 | ||
f1174f77 | 1775 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
9fd29c08 | 1776 | int size, bool zero_size_allowed) |
f1174f77 | 1777 | { |
638f5b90 | 1778 | struct bpf_reg_state *regs = cur_regs(env); |
f1174f77 EC |
1779 | struct bpf_reg_state *reg = ®s[regno]; |
1780 | int err; | |
1781 | ||
1782 | /* We may have added a variable offset to the packet pointer; but any | |
1783 | * reg->range we have comes after that. We are only checking the fixed | |
1784 | * offset. | |
1785 | */ | |
1786 | ||
1787 | /* We don't allow negative numbers, because we aren't tracking enough | |
1788 | * detail to prove they're safe. | |
1789 | */ | |
b03c9f9f | 1790 | if (reg->smin_value < 0) { |
61bd5218 | 1791 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", |
f1174f77 EC |
1792 | regno); |
1793 | return -EACCES; | |
1794 | } | |
9fd29c08 | 1795 | err = __check_packet_access(env, regno, off, size, zero_size_allowed); |
f1174f77 | 1796 | if (err) { |
61bd5218 | 1797 | verbose(env, "R%d offset is outside of the packet\n", regno); |
f1174f77 EC |
1798 | return err; |
1799 | } | |
e647815a JW |
1800 | |
1801 | /* __check_packet_access has made sure "off + size - 1" is within u16. | |
1802 | * reg->umax_value can't be bigger than MAX_PACKET_OFF which is 0xffff, | |
1803 | * otherwise find_good_pkt_pointers would have refused to set range info | |
1804 | * that __check_packet_access would have rejected this pkt access. | |
1805 | * Therefore, "off + reg->umax_value + size - 1" won't overflow u32. | |
1806 | */ | |
1807 | env->prog->aux->max_pkt_offset = | |
1808 | max_t(u32, env->prog->aux->max_pkt_offset, | |
1809 | off + reg->umax_value + size - 1); | |
1810 | ||
f1174f77 EC |
1811 | return err; |
1812 | } | |
1813 | ||
1814 | /* check access to 'struct bpf_context' fields. Supports fixed offsets only */ | |
31fd8581 | 1815 | static int check_ctx_access(struct bpf_verifier_env *env, int insn_idx, int off, int size, |
19de99f7 | 1816 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 1817 | { |
f96da094 DB |
1818 | struct bpf_insn_access_aux info = { |
1819 | .reg_type = *reg_type, | |
1820 | }; | |
31fd8581 | 1821 | |
4f9218aa | 1822 | if (env->ops->is_valid_access && |
5e43f899 | 1823 | env->ops->is_valid_access(off, size, t, env->prog, &info)) { |
f96da094 DB |
1824 | /* A non zero info.ctx_field_size indicates that this field is a |
1825 | * candidate for later verifier transformation to load the whole | |
1826 | * field and then apply a mask when accessed with a narrower | |
1827 | * access than actual ctx access size. A zero info.ctx_field_size | |
1828 | * will only allow for whole field access and rejects any other | |
1829 | * type of narrower access. | |
31fd8581 | 1830 | */ |
23994631 | 1831 | *reg_type = info.reg_type; |
31fd8581 | 1832 | |
4f9218aa | 1833 | env->insn_aux_data[insn_idx].ctx_field_size = info.ctx_field_size; |
32bbe007 AS |
1834 | /* remember the offset of last byte accessed in ctx */ |
1835 | if (env->prog->aux->max_ctx_offset < off + size) | |
1836 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 1837 | return 0; |
32bbe007 | 1838 | } |
17a52670 | 1839 | |
61bd5218 | 1840 | verbose(env, "invalid bpf_context access off=%d size=%d\n", off, size); |
17a52670 AS |
1841 | return -EACCES; |
1842 | } | |
1843 | ||
d58e468b PP |
1844 | static int check_flow_keys_access(struct bpf_verifier_env *env, int off, |
1845 | int size) | |
1846 | { | |
1847 | if (size < 0 || off < 0 || | |
1848 | (u64)off + size > sizeof(struct bpf_flow_keys)) { | |
1849 | verbose(env, "invalid access to flow keys off=%d size=%d\n", | |
1850 | off, size); | |
1851 | return -EACCES; | |
1852 | } | |
1853 | return 0; | |
1854 | } | |
1855 | ||
5f456649 MKL |
1856 | static int check_sock_access(struct bpf_verifier_env *env, int insn_idx, |
1857 | u32 regno, int off, int size, | |
1858 | enum bpf_access_type t) | |
c64b7983 JS |
1859 | { |
1860 | struct bpf_reg_state *regs = cur_regs(env); | |
1861 | struct bpf_reg_state *reg = ®s[regno]; | |
5f456649 | 1862 | struct bpf_insn_access_aux info = {}; |
46f8bc92 | 1863 | bool valid; |
c64b7983 JS |
1864 | |
1865 | if (reg->smin_value < 0) { | |
1866 | verbose(env, "R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
1867 | regno); | |
1868 | return -EACCES; | |
1869 | } | |
1870 | ||
46f8bc92 MKL |
1871 | switch (reg->type) { |
1872 | case PTR_TO_SOCK_COMMON: | |
1873 | valid = bpf_sock_common_is_valid_access(off, size, t, &info); | |
1874 | break; | |
1875 | case PTR_TO_SOCKET: | |
1876 | valid = bpf_sock_is_valid_access(off, size, t, &info); | |
1877 | break; | |
655a51e5 MKL |
1878 | case PTR_TO_TCP_SOCK: |
1879 | valid = bpf_tcp_sock_is_valid_access(off, size, t, &info); | |
1880 | break; | |
fada7fdc JL |
1881 | case PTR_TO_XDP_SOCK: |
1882 | valid = bpf_xdp_sock_is_valid_access(off, size, t, &info); | |
1883 | break; | |
46f8bc92 MKL |
1884 | default: |
1885 | valid = false; | |
c64b7983 JS |
1886 | } |
1887 | ||
5f456649 | 1888 | |
46f8bc92 MKL |
1889 | if (valid) { |
1890 | env->insn_aux_data[insn_idx].ctx_field_size = | |
1891 | info.ctx_field_size; | |
1892 | return 0; | |
1893 | } | |
1894 | ||
1895 | verbose(env, "R%d invalid %s access off=%d size=%d\n", | |
1896 | regno, reg_type_str[reg->type], off, size); | |
1897 | ||
1898 | return -EACCES; | |
c64b7983 JS |
1899 | } |
1900 | ||
4cabc5b1 DB |
1901 | static bool __is_pointer_value(bool allow_ptr_leaks, |
1902 | const struct bpf_reg_state *reg) | |
1be7f75d | 1903 | { |
4cabc5b1 | 1904 | if (allow_ptr_leaks) |
1be7f75d AS |
1905 | return false; |
1906 | ||
f1174f77 | 1907 | return reg->type != SCALAR_VALUE; |
1be7f75d AS |
1908 | } |
1909 | ||
2a159c6f DB |
1910 | static struct bpf_reg_state *reg_state(struct bpf_verifier_env *env, int regno) |
1911 | { | |
1912 | return cur_regs(env) + regno; | |
1913 | } | |
1914 | ||
4cabc5b1 DB |
1915 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1916 | { | |
2a159c6f | 1917 | return __is_pointer_value(env->allow_ptr_leaks, reg_state(env, regno)); |
4cabc5b1 DB |
1918 | } |
1919 | ||
f37a8cb8 DB |
1920 | static bool is_ctx_reg(struct bpf_verifier_env *env, int regno) |
1921 | { | |
2a159c6f | 1922 | const struct bpf_reg_state *reg = reg_state(env, regno); |
f37a8cb8 | 1923 | |
46f8bc92 MKL |
1924 | return reg->type == PTR_TO_CTX; |
1925 | } | |
1926 | ||
1927 | static bool is_sk_reg(struct bpf_verifier_env *env, int regno) | |
1928 | { | |
1929 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
1930 | ||
1931 | return type_is_sk_pointer(reg->type); | |
f37a8cb8 DB |
1932 | } |
1933 | ||
ca369602 DB |
1934 | static bool is_pkt_reg(struct bpf_verifier_env *env, int regno) |
1935 | { | |
2a159c6f | 1936 | const struct bpf_reg_state *reg = reg_state(env, regno); |
ca369602 DB |
1937 | |
1938 | return type_is_pkt_pointer(reg->type); | |
1939 | } | |
1940 | ||
4b5defde DB |
1941 | static bool is_flow_key_reg(struct bpf_verifier_env *env, int regno) |
1942 | { | |
1943 | const struct bpf_reg_state *reg = reg_state(env, regno); | |
1944 | ||
1945 | /* Separate to is_ctx_reg() since we still want to allow BPF_ST here. */ | |
1946 | return reg->type == PTR_TO_FLOW_KEYS; | |
1947 | } | |
1948 | ||
61bd5218 JK |
1949 | static int check_pkt_ptr_alignment(struct bpf_verifier_env *env, |
1950 | const struct bpf_reg_state *reg, | |
d1174416 | 1951 | int off, int size, bool strict) |
969bf05e | 1952 | { |
f1174f77 | 1953 | struct tnum reg_off; |
e07b98d9 | 1954 | int ip_align; |
d1174416 DM |
1955 | |
1956 | /* Byte size accesses are always allowed. */ | |
1957 | if (!strict || size == 1) | |
1958 | return 0; | |
1959 | ||
e4eda884 DM |
1960 | /* For platforms that do not have a Kconfig enabling |
1961 | * CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS the value of | |
1962 | * NET_IP_ALIGN is universally set to '2'. And on platforms | |
1963 | * that do set CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS, we get | |
1964 | * to this code only in strict mode where we want to emulate | |
1965 | * the NET_IP_ALIGN==2 checking. Therefore use an | |
1966 | * unconditional IP align value of '2'. | |
e07b98d9 | 1967 | */ |
e4eda884 | 1968 | ip_align = 2; |
f1174f77 EC |
1969 | |
1970 | reg_off = tnum_add(reg->var_off, tnum_const(ip_align + reg->off + off)); | |
1971 | if (!tnum_is_aligned(reg_off, size)) { | |
1972 | char tn_buf[48]; | |
1973 | ||
1974 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 JK |
1975 | verbose(env, |
1976 | "misaligned packet access off %d+%s+%d+%d size %d\n", | |
f1174f77 | 1977 | ip_align, tn_buf, reg->off, off, size); |
969bf05e AS |
1978 | return -EACCES; |
1979 | } | |
79adffcd | 1980 | |
969bf05e AS |
1981 | return 0; |
1982 | } | |
1983 | ||
61bd5218 JK |
1984 | static int check_generic_ptr_alignment(struct bpf_verifier_env *env, |
1985 | const struct bpf_reg_state *reg, | |
f1174f77 EC |
1986 | const char *pointer_desc, |
1987 | int off, int size, bool strict) | |
79adffcd | 1988 | { |
f1174f77 EC |
1989 | struct tnum reg_off; |
1990 | ||
1991 | /* Byte size accesses are always allowed. */ | |
1992 | if (!strict || size == 1) | |
1993 | return 0; | |
1994 | ||
1995 | reg_off = tnum_add(reg->var_off, tnum_const(reg->off + off)); | |
1996 | if (!tnum_is_aligned(reg_off, size)) { | |
1997 | char tn_buf[48]; | |
1998 | ||
1999 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
61bd5218 | 2000 | verbose(env, "misaligned %saccess off %s+%d+%d size %d\n", |
f1174f77 | 2001 | pointer_desc, tn_buf, reg->off, off, size); |
79adffcd DB |
2002 | return -EACCES; |
2003 | } | |
2004 | ||
969bf05e AS |
2005 | return 0; |
2006 | } | |
2007 | ||
e07b98d9 | 2008 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
ca369602 DB |
2009 | const struct bpf_reg_state *reg, int off, |
2010 | int size, bool strict_alignment_once) | |
79adffcd | 2011 | { |
ca369602 | 2012 | bool strict = env->strict_alignment || strict_alignment_once; |
f1174f77 | 2013 | const char *pointer_desc = ""; |
d1174416 | 2014 | |
79adffcd DB |
2015 | switch (reg->type) { |
2016 | case PTR_TO_PACKET: | |
de8f3a83 DB |
2017 | case PTR_TO_PACKET_META: |
2018 | /* Special case, because of NET_IP_ALIGN. Given metadata sits | |
2019 | * right in front, treat it the very same way. | |
2020 | */ | |
61bd5218 | 2021 | return check_pkt_ptr_alignment(env, reg, off, size, strict); |
d58e468b PP |
2022 | case PTR_TO_FLOW_KEYS: |
2023 | pointer_desc = "flow keys "; | |
2024 | break; | |
f1174f77 EC |
2025 | case PTR_TO_MAP_VALUE: |
2026 | pointer_desc = "value "; | |
2027 | break; | |
2028 | case PTR_TO_CTX: | |
2029 | pointer_desc = "context "; | |
2030 | break; | |
2031 | case PTR_TO_STACK: | |
2032 | pointer_desc = "stack "; | |
a5ec6ae1 JH |
2033 | /* The stack spill tracking logic in check_stack_write() |
2034 | * and check_stack_read() relies on stack accesses being | |
2035 | * aligned. | |
2036 | */ | |
2037 | strict = true; | |
f1174f77 | 2038 | break; |
c64b7983 JS |
2039 | case PTR_TO_SOCKET: |
2040 | pointer_desc = "sock "; | |
2041 | break; | |
46f8bc92 MKL |
2042 | case PTR_TO_SOCK_COMMON: |
2043 | pointer_desc = "sock_common "; | |
2044 | break; | |
655a51e5 MKL |
2045 | case PTR_TO_TCP_SOCK: |
2046 | pointer_desc = "tcp_sock "; | |
2047 | break; | |
fada7fdc JL |
2048 | case PTR_TO_XDP_SOCK: |
2049 | pointer_desc = "xdp_sock "; | |
2050 | break; | |
79adffcd | 2051 | default: |
f1174f77 | 2052 | break; |
79adffcd | 2053 | } |
61bd5218 JK |
2054 | return check_generic_ptr_alignment(env, reg, pointer_desc, off, size, |
2055 | strict); | |
79adffcd DB |
2056 | } |
2057 | ||
f4d7e40a AS |
2058 | static int update_stack_depth(struct bpf_verifier_env *env, |
2059 | const struct bpf_func_state *func, | |
2060 | int off) | |
2061 | { | |
9c8105bd | 2062 | u16 stack = env->subprog_info[func->subprogno].stack_depth; |
f4d7e40a AS |
2063 | |
2064 | if (stack >= -off) | |
2065 | return 0; | |
2066 | ||
2067 | /* update known max for given subprogram */ | |
9c8105bd | 2068 | env->subprog_info[func->subprogno].stack_depth = -off; |
70a87ffe AS |
2069 | return 0; |
2070 | } | |
f4d7e40a | 2071 | |
70a87ffe AS |
2072 | /* starting from main bpf function walk all instructions of the function |
2073 | * and recursively walk all callees that given function can call. | |
2074 | * Ignore jump and exit insns. | |
2075 | * Since recursion is prevented by check_cfg() this algorithm | |
2076 | * only needs a local stack of MAX_CALL_FRAMES to remember callsites | |
2077 | */ | |
2078 | static int check_max_stack_depth(struct bpf_verifier_env *env) | |
2079 | { | |
9c8105bd JW |
2080 | int depth = 0, frame = 0, idx = 0, i = 0, subprog_end; |
2081 | struct bpf_subprog_info *subprog = env->subprog_info; | |
70a87ffe | 2082 | struct bpf_insn *insn = env->prog->insnsi; |
70a87ffe AS |
2083 | int ret_insn[MAX_CALL_FRAMES]; |
2084 | int ret_prog[MAX_CALL_FRAMES]; | |
f4d7e40a | 2085 | |
70a87ffe AS |
2086 | process_func: |
2087 | /* round up to 32-bytes, since this is granularity | |
2088 | * of interpreter stack size | |
2089 | */ | |
9c8105bd | 2090 | depth += round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe | 2091 | if (depth > MAX_BPF_STACK) { |
f4d7e40a | 2092 | verbose(env, "combined stack size of %d calls is %d. Too large\n", |
70a87ffe | 2093 | frame + 1, depth); |
f4d7e40a AS |
2094 | return -EACCES; |
2095 | } | |
70a87ffe | 2096 | continue_func: |
4cb3d99c | 2097 | subprog_end = subprog[idx + 1].start; |
70a87ffe AS |
2098 | for (; i < subprog_end; i++) { |
2099 | if (insn[i].code != (BPF_JMP | BPF_CALL)) | |
2100 | continue; | |
2101 | if (insn[i].src_reg != BPF_PSEUDO_CALL) | |
2102 | continue; | |
2103 | /* remember insn and function to return to */ | |
2104 | ret_insn[frame] = i + 1; | |
9c8105bd | 2105 | ret_prog[frame] = idx; |
70a87ffe AS |
2106 | |
2107 | /* find the callee */ | |
2108 | i = i + insn[i].imm + 1; | |
9c8105bd JW |
2109 | idx = find_subprog(env, i); |
2110 | if (idx < 0) { | |
70a87ffe AS |
2111 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", |
2112 | i); | |
2113 | return -EFAULT; | |
2114 | } | |
70a87ffe AS |
2115 | frame++; |
2116 | if (frame >= MAX_CALL_FRAMES) { | |
927cb781 PC |
2117 | verbose(env, "the call stack of %d frames is too deep !\n", |
2118 | frame); | |
2119 | return -E2BIG; | |
70a87ffe AS |
2120 | } |
2121 | goto process_func; | |
2122 | } | |
2123 | /* end of for() loop means the last insn of the 'subprog' | |
2124 | * was reached. Doesn't matter whether it was JA or EXIT | |
2125 | */ | |
2126 | if (frame == 0) | |
2127 | return 0; | |
9c8105bd | 2128 | depth -= round_up(max_t(u32, subprog[idx].stack_depth, 1), 32); |
70a87ffe AS |
2129 | frame--; |
2130 | i = ret_insn[frame]; | |
9c8105bd | 2131 | idx = ret_prog[frame]; |
70a87ffe | 2132 | goto continue_func; |
f4d7e40a AS |
2133 | } |
2134 | ||
19d28fbd | 2135 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
2136 | static int get_callee_stack_depth(struct bpf_verifier_env *env, |
2137 | const struct bpf_insn *insn, int idx) | |
2138 | { | |
2139 | int start = idx + insn->imm + 1, subprog; | |
2140 | ||
2141 | subprog = find_subprog(env, start); | |
2142 | if (subprog < 0) { | |
2143 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
2144 | start); | |
2145 | return -EFAULT; | |
2146 | } | |
9c8105bd | 2147 | return env->subprog_info[subprog].stack_depth; |
1ea47e01 | 2148 | } |
19d28fbd | 2149 | #endif |
1ea47e01 | 2150 | |
58990d1f DB |
2151 | static int check_ctx_reg(struct bpf_verifier_env *env, |
2152 | const struct bpf_reg_state *reg, int regno) | |
2153 | { | |
2154 | /* Access to ctx or passing it to a helper is only allowed in | |
2155 | * its original, unmodified form. | |
2156 | */ | |
2157 | ||
2158 | if (reg->off) { | |
2159 | verbose(env, "dereference of modified ctx ptr R%d off=%d disallowed\n", | |
2160 | regno, reg->off); | |
2161 | return -EACCES; | |
2162 | } | |
2163 | ||
2164 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
2165 | char tn_buf[48]; | |
2166 | ||
2167 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
2168 | verbose(env, "variable ctx access var_off=%s disallowed\n", tn_buf); | |
2169 | return -EACCES; | |
2170 | } | |
2171 | ||
2172 | return 0; | |
2173 | } | |
2174 | ||
9df1c28b MM |
2175 | static int check_tp_buffer_access(struct bpf_verifier_env *env, |
2176 | const struct bpf_reg_state *reg, | |
2177 | int regno, int off, int size) | |
2178 | { | |
2179 | if (off < 0) { | |
2180 | verbose(env, | |
2181 | "R%d invalid tracepoint buffer access: off=%d, size=%d", | |
2182 | regno, off, size); | |
2183 | return -EACCES; | |
2184 | } | |
2185 | if (!tnum_is_const(reg->var_off) || reg->var_off.value) { | |
2186 | char tn_buf[48]; | |
2187 | ||
2188 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
2189 | verbose(env, | |
2190 | "R%d invalid variable buffer offset: off=%d, var_off=%s", | |
2191 | regno, off, tn_buf); | |
2192 | return -EACCES; | |
2193 | } | |
2194 | if (off + size > env->prog->aux->max_tp_access) | |
2195 | env->prog->aux->max_tp_access = off + size; | |
2196 | ||
2197 | return 0; | |
2198 | } | |
2199 | ||
2200 | ||
0c17d1d2 JH |
2201 | /* truncate register to smaller size (in bytes) |
2202 | * must be called with size < BPF_REG_SIZE | |
2203 | */ | |
2204 | static void coerce_reg_to_size(struct bpf_reg_state *reg, int size) | |
2205 | { | |
2206 | u64 mask; | |
2207 | ||
2208 | /* clear high bits in bit representation */ | |
2209 | reg->var_off = tnum_cast(reg->var_off, size); | |
2210 | ||
2211 | /* fix arithmetic bounds */ | |
2212 | mask = ((u64)1 << (size * 8)) - 1; | |
2213 | if ((reg->umin_value & ~mask) == (reg->umax_value & ~mask)) { | |
2214 | reg->umin_value &= mask; | |
2215 | reg->umax_value &= mask; | |
2216 | } else { | |
2217 | reg->umin_value = 0; | |
2218 | reg->umax_value = mask; | |
2219 | } | |
2220 | reg->smin_value = reg->umin_value; | |
2221 | reg->smax_value = reg->umax_value; | |
2222 | } | |
2223 | ||
17a52670 AS |
2224 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
2225 | * if t==write, value_regno is a register which value is stored into memory | |
2226 | * if t==read, value_regno is a register which will receive the value from memory | |
2227 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
2228 | * if t==read && value_regno==-1, don't care what we read from memory | |
2229 | */ | |
ca369602 DB |
2230 | static int check_mem_access(struct bpf_verifier_env *env, int insn_idx, u32 regno, |
2231 | int off, int bpf_size, enum bpf_access_type t, | |
2232 | int value_regno, bool strict_alignment_once) | |
17a52670 | 2233 | { |
638f5b90 AS |
2234 | struct bpf_reg_state *regs = cur_regs(env); |
2235 | struct bpf_reg_state *reg = regs + regno; | |
f4d7e40a | 2236 | struct bpf_func_state *state; |
17a52670 AS |
2237 | int size, err = 0; |
2238 | ||
2239 | size = bpf_size_to_bytes(bpf_size); | |
2240 | if (size < 0) | |
2241 | return size; | |
2242 | ||
f1174f77 | 2243 | /* alignment checks will add in reg->off themselves */ |
ca369602 | 2244 | err = check_ptr_alignment(env, reg, off, size, strict_alignment_once); |
969bf05e AS |
2245 | if (err) |
2246 | return err; | |
17a52670 | 2247 | |
f1174f77 EC |
2248 | /* for access checks, reg->off is just part of off */ |
2249 | off += reg->off; | |
2250 | ||
2251 | if (reg->type == PTR_TO_MAP_VALUE) { | |
1be7f75d AS |
2252 | if (t == BPF_WRITE && value_regno >= 0 && |
2253 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 2254 | verbose(env, "R%d leaks addr into map\n", value_regno); |
1be7f75d AS |
2255 | return -EACCES; |
2256 | } | |
591fe988 DB |
2257 | err = check_map_access_type(env, regno, off, size, t); |
2258 | if (err) | |
2259 | return err; | |
9fd29c08 | 2260 | err = check_map_access(env, regno, off, size, false); |
17a52670 | 2261 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 2262 | mark_reg_unknown(env, regs, value_regno); |
17a52670 | 2263 | |
1a0dc1ac | 2264 | } else if (reg->type == PTR_TO_CTX) { |
f1174f77 | 2265 | enum bpf_reg_type reg_type = SCALAR_VALUE; |
19de99f7 | 2266 | |
1be7f75d AS |
2267 | if (t == BPF_WRITE && value_regno >= 0 && |
2268 | is_pointer_value(env, value_regno)) { | |
61bd5218 | 2269 | verbose(env, "R%d leaks addr into ctx\n", value_regno); |
1be7f75d AS |
2270 | return -EACCES; |
2271 | } | |
f1174f77 | 2272 | |
58990d1f DB |
2273 | err = check_ctx_reg(env, reg, regno); |
2274 | if (err < 0) | |
2275 | return err; | |
2276 | ||
31fd8581 | 2277 | err = check_ctx_access(env, insn_idx, off, size, t, ®_type); |
969bf05e | 2278 | if (!err && t == BPF_READ && value_regno >= 0) { |
f1174f77 | 2279 | /* ctx access returns either a scalar, or a |
de8f3a83 DB |
2280 | * PTR_TO_PACKET[_META,_END]. In the latter |
2281 | * case, we know the offset is zero. | |
f1174f77 | 2282 | */ |
46f8bc92 | 2283 | if (reg_type == SCALAR_VALUE) { |
638f5b90 | 2284 | mark_reg_unknown(env, regs, value_regno); |
46f8bc92 | 2285 | } else { |
638f5b90 | 2286 | mark_reg_known_zero(env, regs, |
61bd5218 | 2287 | value_regno); |
46f8bc92 MKL |
2288 | if (reg_type_may_be_null(reg_type)) |
2289 | regs[value_regno].id = ++env->id_gen; | |
5327ed3d JW |
2290 | /* A load of ctx field could have different |
2291 | * actual load size with the one encoded in the | |
2292 | * insn. When the dst is PTR, it is for sure not | |
2293 | * a sub-register. | |
2294 | */ | |
2295 | regs[value_regno].subreg_def = DEF_NOT_SUBREG; | |
46f8bc92 | 2296 | } |
638f5b90 | 2297 | regs[value_regno].type = reg_type; |
969bf05e | 2298 | } |
17a52670 | 2299 | |
f1174f77 | 2300 | } else if (reg->type == PTR_TO_STACK) { |
f1174f77 | 2301 | off += reg->var_off.value; |
e4298d25 DB |
2302 | err = check_stack_access(env, reg, off, size); |
2303 | if (err) | |
2304 | return err; | |
8726679a | 2305 | |
f4d7e40a AS |
2306 | state = func(env, reg); |
2307 | err = update_stack_depth(env, state, off); | |
2308 | if (err) | |
2309 | return err; | |
8726679a | 2310 | |
638f5b90 | 2311 | if (t == BPF_WRITE) |
61bd5218 | 2312 | err = check_stack_write(env, state, off, size, |
af86ca4e | 2313 | value_regno, insn_idx); |
638f5b90 | 2314 | else |
61bd5218 JK |
2315 | err = check_stack_read(env, state, off, size, |
2316 | value_regno); | |
de8f3a83 | 2317 | } else if (reg_is_pkt_pointer(reg)) { |
3a0af8fd | 2318 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
61bd5218 | 2319 | verbose(env, "cannot write into packet\n"); |
969bf05e AS |
2320 | return -EACCES; |
2321 | } | |
4acf6c0b BB |
2322 | if (t == BPF_WRITE && value_regno >= 0 && |
2323 | is_pointer_value(env, value_regno)) { | |
61bd5218 JK |
2324 | verbose(env, "R%d leaks addr into packet\n", |
2325 | value_regno); | |
4acf6c0b BB |
2326 | return -EACCES; |
2327 | } | |
9fd29c08 | 2328 | err = check_packet_access(env, regno, off, size, false); |
969bf05e | 2329 | if (!err && t == BPF_READ && value_regno >= 0) |
638f5b90 | 2330 | mark_reg_unknown(env, regs, value_regno); |
d58e468b PP |
2331 | } else if (reg->type == PTR_TO_FLOW_KEYS) { |
2332 | if (t == BPF_WRITE && value_regno >= 0 && | |
2333 | is_pointer_value(env, value_regno)) { | |
2334 | verbose(env, "R%d leaks addr into flow keys\n", | |
2335 | value_regno); | |
2336 | return -EACCES; | |
2337 | } | |
2338 | ||
2339 | err = check_flow_keys_access(env, off, size); | |
2340 | if (!err && t == BPF_READ && value_regno >= 0) | |
2341 | mark_reg_unknown(env, regs, value_regno); | |
46f8bc92 | 2342 | } else if (type_is_sk_pointer(reg->type)) { |
c64b7983 | 2343 | if (t == BPF_WRITE) { |
46f8bc92 MKL |
2344 | verbose(env, "R%d cannot write into %s\n", |
2345 | regno, reg_type_str[reg->type]); | |
c64b7983 JS |
2346 | return -EACCES; |
2347 | } | |
5f456649 | 2348 | err = check_sock_access(env, insn_idx, regno, off, size, t); |
c64b7983 JS |
2349 | if (!err && value_regno >= 0) |
2350 | mark_reg_unknown(env, regs, value_regno); | |
9df1c28b MM |
2351 | } else if (reg->type == PTR_TO_TP_BUFFER) { |
2352 | err = check_tp_buffer_access(env, reg, regno, off, size); | |
2353 | if (!err && t == BPF_READ && value_regno >= 0) | |
2354 | mark_reg_unknown(env, regs, value_regno); | |
17a52670 | 2355 | } else { |
61bd5218 JK |
2356 | verbose(env, "R%d invalid mem access '%s'\n", regno, |
2357 | reg_type_str[reg->type]); | |
17a52670 AS |
2358 | return -EACCES; |
2359 | } | |
969bf05e | 2360 | |
f1174f77 | 2361 | if (!err && size < BPF_REG_SIZE && value_regno >= 0 && t == BPF_READ && |
638f5b90 | 2362 | regs[value_regno].type == SCALAR_VALUE) { |
f1174f77 | 2363 | /* b/h/w load zero-extends, mark upper bits as known 0 */ |
0c17d1d2 | 2364 | coerce_reg_to_size(®s[value_regno], size); |
969bf05e | 2365 | } |
17a52670 AS |
2366 | return err; |
2367 | } | |
2368 | ||
31fd8581 | 2369 | static int check_xadd(struct bpf_verifier_env *env, int insn_idx, struct bpf_insn *insn) |
17a52670 | 2370 | { |
17a52670 AS |
2371 | int err; |
2372 | ||
2373 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
2374 | insn->imm != 0) { | |
61bd5218 | 2375 | verbose(env, "BPF_XADD uses reserved fields\n"); |
17a52670 AS |
2376 | return -EINVAL; |
2377 | } | |
2378 | ||
2379 | /* check src1 operand */ | |
dc503a8a | 2380 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
2381 | if (err) |
2382 | return err; | |
2383 | ||
2384 | /* check src2 operand */ | |
dc503a8a | 2385 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
2386 | if (err) |
2387 | return err; | |
2388 | ||
6bdf6abc | 2389 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 | 2390 | verbose(env, "R%d leaks addr into mem\n", insn->src_reg); |
6bdf6abc DB |
2391 | return -EACCES; |
2392 | } | |
2393 | ||
ca369602 | 2394 | if (is_ctx_reg(env, insn->dst_reg) || |
4b5defde | 2395 | is_pkt_reg(env, insn->dst_reg) || |
46f8bc92 MKL |
2396 | is_flow_key_reg(env, insn->dst_reg) || |
2397 | is_sk_reg(env, insn->dst_reg)) { | |
ca369602 | 2398 | verbose(env, "BPF_XADD stores into R%d %s is not allowed\n", |
2a159c6f DB |
2399 | insn->dst_reg, |
2400 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
2401 | return -EACCES; |
2402 | } | |
2403 | ||
17a52670 | 2404 | /* check whether atomic_add can read the memory */ |
31fd8581 | 2405 | err = check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 2406 | BPF_SIZE(insn->code), BPF_READ, -1, true); |
17a52670 AS |
2407 | if (err) |
2408 | return err; | |
2409 | ||
2410 | /* check whether atomic_add can write into the same memory */ | |
31fd8581 | 2411 | return check_mem_access(env, insn_idx, insn->dst_reg, insn->off, |
ca369602 | 2412 | BPF_SIZE(insn->code), BPF_WRITE, -1, true); |
17a52670 AS |
2413 | } |
2414 | ||
2011fccf AI |
2415 | static int __check_stack_boundary(struct bpf_verifier_env *env, u32 regno, |
2416 | int off, int access_size, | |
2417 | bool zero_size_allowed) | |
2418 | { | |
2419 | struct bpf_reg_state *reg = reg_state(env, regno); | |
2420 | ||
2421 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || | |
2422 | access_size < 0 || (access_size == 0 && !zero_size_allowed)) { | |
2423 | if (tnum_is_const(reg->var_off)) { | |
2424 | verbose(env, "invalid stack type R%d off=%d access_size=%d\n", | |
2425 | regno, off, access_size); | |
2426 | } else { | |
2427 | char tn_buf[48]; | |
2428 | ||
2429 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
2430 | verbose(env, "invalid stack type R%d var_off=%s access_size=%d\n", | |
2431 | regno, tn_buf, access_size); | |
2432 | } | |
2433 | return -EACCES; | |
2434 | } | |
2435 | return 0; | |
2436 | } | |
2437 | ||
17a52670 AS |
2438 | /* when register 'regno' is passed into function that will read 'access_size' |
2439 | * bytes from that pointer, make sure that it's within stack boundary | |
f1174f77 EC |
2440 | * and all elements of stack are initialized. |
2441 | * Unlike most pointer bounds-checking functions, this one doesn't take an | |
2442 | * 'off' argument, so it has to add in reg->off itself. | |
17a52670 | 2443 | */ |
58e2af8b | 2444 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
2445 | int access_size, bool zero_size_allowed, |
2446 | struct bpf_call_arg_meta *meta) | |
17a52670 | 2447 | { |
2a159c6f | 2448 | struct bpf_reg_state *reg = reg_state(env, regno); |
f4d7e40a | 2449 | struct bpf_func_state *state = func(env, reg); |
f7cf25b2 | 2450 | int err, min_off, max_off, i, j, slot, spi; |
17a52670 | 2451 | |
914cb781 | 2452 | if (reg->type != PTR_TO_STACK) { |
f1174f77 | 2453 | /* Allow zero-byte read from NULL, regardless of pointer type */ |
8e2fe1d9 | 2454 | if (zero_size_allowed && access_size == 0 && |
914cb781 | 2455 | register_is_null(reg)) |
8e2fe1d9 DB |
2456 | return 0; |
2457 | ||
61bd5218 | 2458 | verbose(env, "R%d type=%s expected=%s\n", regno, |
914cb781 | 2459 | reg_type_str[reg->type], |
8e2fe1d9 | 2460 | reg_type_str[PTR_TO_STACK]); |
17a52670 | 2461 | return -EACCES; |
8e2fe1d9 | 2462 | } |
17a52670 | 2463 | |
2011fccf AI |
2464 | if (tnum_is_const(reg->var_off)) { |
2465 | min_off = max_off = reg->var_off.value + reg->off; | |
2466 | err = __check_stack_boundary(env, regno, min_off, access_size, | |
2467 | zero_size_allowed); | |
2468 | if (err) | |
2469 | return err; | |
2470 | } else { | |
088ec26d AI |
2471 | /* Variable offset is prohibited for unprivileged mode for |
2472 | * simplicity since it requires corresponding support in | |
2473 | * Spectre masking for stack ALU. | |
2474 | * See also retrieve_ptr_limit(). | |
2475 | */ | |
2476 | if (!env->allow_ptr_leaks) { | |
2477 | char tn_buf[48]; | |
f1174f77 | 2478 | |
088ec26d AI |
2479 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
2480 | verbose(env, "R%d indirect variable offset stack access prohibited for !root, var_off=%s\n", | |
2481 | regno, tn_buf); | |
2482 | return -EACCES; | |
2483 | } | |
f2bcd05e AI |
2484 | /* Only initialized buffer on stack is allowed to be accessed |
2485 | * with variable offset. With uninitialized buffer it's hard to | |
2486 | * guarantee that whole memory is marked as initialized on | |
2487 | * helper return since specific bounds are unknown what may | |
2488 | * cause uninitialized stack leaking. | |
2489 | */ | |
2490 | if (meta && meta->raw_mode) | |
2491 | meta = NULL; | |
2492 | ||
107c26a7 AI |
2493 | if (reg->smax_value >= BPF_MAX_VAR_OFF || |
2494 | reg->smax_value <= -BPF_MAX_VAR_OFF) { | |
2495 | verbose(env, "R%d unbounded indirect variable offset stack access\n", | |
2496 | regno); | |
2497 | return -EACCES; | |
2498 | } | |
2011fccf | 2499 | min_off = reg->smin_value + reg->off; |
107c26a7 | 2500 | max_off = reg->smax_value + reg->off; |
2011fccf AI |
2501 | err = __check_stack_boundary(env, regno, min_off, access_size, |
2502 | zero_size_allowed); | |
107c26a7 AI |
2503 | if (err) { |
2504 | verbose(env, "R%d min value is outside of stack bound\n", | |
2505 | regno); | |
2011fccf | 2506 | return err; |
107c26a7 | 2507 | } |
2011fccf AI |
2508 | err = __check_stack_boundary(env, regno, max_off, access_size, |
2509 | zero_size_allowed); | |
107c26a7 AI |
2510 | if (err) { |
2511 | verbose(env, "R%d max value is outside of stack bound\n", | |
2512 | regno); | |
2011fccf | 2513 | return err; |
107c26a7 | 2514 | } |
17a52670 AS |
2515 | } |
2516 | ||
435faee1 DB |
2517 | if (meta && meta->raw_mode) { |
2518 | meta->access_size = access_size; | |
2519 | meta->regno = regno; | |
2520 | return 0; | |
2521 | } | |
2522 | ||
2011fccf | 2523 | for (i = min_off; i < max_off + access_size; i++) { |
cc2b14d5 AS |
2524 | u8 *stype; |
2525 | ||
2011fccf | 2526 | slot = -i - 1; |
638f5b90 | 2527 | spi = slot / BPF_REG_SIZE; |
cc2b14d5 AS |
2528 | if (state->allocated_stack <= slot) |
2529 | goto err; | |
2530 | stype = &state->stack[spi].slot_type[slot % BPF_REG_SIZE]; | |
2531 | if (*stype == STACK_MISC) | |
2532 | goto mark; | |
2533 | if (*stype == STACK_ZERO) { | |
2534 | /* helper can write anything into the stack */ | |
2535 | *stype = STACK_MISC; | |
2536 | goto mark; | |
17a52670 | 2537 | } |
f7cf25b2 AS |
2538 | if (state->stack[spi].slot_type[0] == STACK_SPILL && |
2539 | state->stack[spi].spilled_ptr.type == SCALAR_VALUE) { | |
2540 | __mark_reg_unknown(&state->stack[spi].spilled_ptr); | |
2541 | for (j = 0; j < BPF_REG_SIZE; j++) | |
2542 | state->stack[spi].slot_type[j] = STACK_MISC; | |
2543 | goto mark; | |
2544 | } | |
2545 | ||
cc2b14d5 | 2546 | err: |
2011fccf AI |
2547 | if (tnum_is_const(reg->var_off)) { |
2548 | verbose(env, "invalid indirect read from stack off %d+%d size %d\n", | |
2549 | min_off, i - min_off, access_size); | |
2550 | } else { | |
2551 | char tn_buf[48]; | |
2552 | ||
2553 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); | |
2554 | verbose(env, "invalid indirect read from stack var_off %s+%d size %d\n", | |
2555 | tn_buf, i - min_off, access_size); | |
2556 | } | |
cc2b14d5 AS |
2557 | return -EACCES; |
2558 | mark: | |
2559 | /* reading any byte out of 8-byte 'spill_slot' will cause | |
2560 | * the whole slot to be marked as 'read' | |
2561 | */ | |
679c782d | 2562 | mark_reg_read(env, &state->stack[spi].spilled_ptr, |
5327ed3d JW |
2563 | state->stack[spi].spilled_ptr.parent, |
2564 | REG_LIVE_READ64); | |
17a52670 | 2565 | } |
2011fccf | 2566 | return update_stack_depth(env, state, min_off); |
17a52670 AS |
2567 | } |
2568 | ||
06c1c049 GB |
2569 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
2570 | int access_size, bool zero_size_allowed, | |
2571 | struct bpf_call_arg_meta *meta) | |
2572 | { | |
638f5b90 | 2573 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
06c1c049 | 2574 | |
f1174f77 | 2575 | switch (reg->type) { |
06c1c049 | 2576 | case PTR_TO_PACKET: |
de8f3a83 | 2577 | case PTR_TO_PACKET_META: |
9fd29c08 YS |
2578 | return check_packet_access(env, regno, reg->off, access_size, |
2579 | zero_size_allowed); | |
06c1c049 | 2580 | case PTR_TO_MAP_VALUE: |
591fe988 DB |
2581 | if (check_map_access_type(env, regno, reg->off, access_size, |
2582 | meta && meta->raw_mode ? BPF_WRITE : | |
2583 | BPF_READ)) | |
2584 | return -EACCES; | |
9fd29c08 YS |
2585 | return check_map_access(env, regno, reg->off, access_size, |
2586 | zero_size_allowed); | |
f1174f77 | 2587 | default: /* scalar_value|ptr_to_stack or invalid ptr */ |
06c1c049 GB |
2588 | return check_stack_boundary(env, regno, access_size, |
2589 | zero_size_allowed, meta); | |
2590 | } | |
2591 | } | |
2592 | ||
d83525ca AS |
2593 | /* Implementation details: |
2594 | * bpf_map_lookup returns PTR_TO_MAP_VALUE_OR_NULL | |
2595 | * Two bpf_map_lookups (even with the same key) will have different reg->id. | |
2596 | * For traditional PTR_TO_MAP_VALUE the verifier clears reg->id after | |
2597 | * value_or_null->value transition, since the verifier only cares about | |
2598 | * the range of access to valid map value pointer and doesn't care about actual | |
2599 | * address of the map element. | |
2600 | * For maps with 'struct bpf_spin_lock' inside map value the verifier keeps | |
2601 | * reg->id > 0 after value_or_null->value transition. By doing so | |
2602 | * two bpf_map_lookups will be considered two different pointers that | |
2603 | * point to different bpf_spin_locks. | |
2604 | * The verifier allows taking only one bpf_spin_lock at a time to avoid | |
2605 | * dead-locks. | |
2606 | * Since only one bpf_spin_lock is allowed the checks are simpler than | |
2607 | * reg_is_refcounted() logic. The verifier needs to remember only | |
2608 | * one spin_lock instead of array of acquired_refs. | |
2609 | * cur_state->active_spin_lock remembers which map value element got locked | |
2610 | * and clears it after bpf_spin_unlock. | |
2611 | */ | |
2612 | static int process_spin_lock(struct bpf_verifier_env *env, int regno, | |
2613 | bool is_lock) | |
2614 | { | |
2615 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; | |
2616 | struct bpf_verifier_state *cur = env->cur_state; | |
2617 | bool is_const = tnum_is_const(reg->var_off); | |
2618 | struct bpf_map *map = reg->map_ptr; | |
2619 | u64 val = reg->var_off.value; | |
2620 | ||
2621 | if (reg->type != PTR_TO_MAP_VALUE) { | |
2622 | verbose(env, "R%d is not a pointer to map_value\n", regno); | |
2623 | return -EINVAL; | |
2624 | } | |
2625 | if (!is_const) { | |
2626 | verbose(env, | |
2627 | "R%d doesn't have constant offset. bpf_spin_lock has to be at the constant offset\n", | |
2628 | regno); | |
2629 | return -EINVAL; | |
2630 | } | |
2631 | if (!map->btf) { | |
2632 | verbose(env, | |
2633 | "map '%s' has to have BTF in order to use bpf_spin_lock\n", | |
2634 | map->name); | |
2635 | return -EINVAL; | |
2636 | } | |
2637 | if (!map_value_has_spin_lock(map)) { | |
2638 | if (map->spin_lock_off == -E2BIG) | |
2639 | verbose(env, | |
2640 | "map '%s' has more than one 'struct bpf_spin_lock'\n", | |
2641 | map->name); | |
2642 | else if (map->spin_lock_off == -ENOENT) | |
2643 | verbose(env, | |
2644 | "map '%s' doesn't have 'struct bpf_spin_lock'\n", | |
2645 | map->name); | |
2646 | else | |
2647 | verbose(env, | |
2648 | "map '%s' is not a struct type or bpf_spin_lock is mangled\n", | |
2649 | map->name); | |
2650 | return -EINVAL; | |
2651 | } | |
2652 | if (map->spin_lock_off != val + reg->off) { | |
2653 | verbose(env, "off %lld doesn't point to 'struct bpf_spin_lock'\n", | |
2654 | val + reg->off); | |
2655 | return -EINVAL; | |
2656 | } | |
2657 | if (is_lock) { | |
2658 | if (cur->active_spin_lock) { | |
2659 | verbose(env, | |
2660 | "Locking two bpf_spin_locks are not allowed\n"); | |
2661 | return -EINVAL; | |
2662 | } | |
2663 | cur->active_spin_lock = reg->id; | |
2664 | } else { | |
2665 | if (!cur->active_spin_lock) { | |
2666 | verbose(env, "bpf_spin_unlock without taking a lock\n"); | |
2667 | return -EINVAL; | |
2668 | } | |
2669 | if (cur->active_spin_lock != reg->id) { | |
2670 | verbose(env, "bpf_spin_unlock of different lock\n"); | |
2671 | return -EINVAL; | |
2672 | } | |
2673 | cur->active_spin_lock = 0; | |
2674 | } | |
2675 | return 0; | |
2676 | } | |
2677 | ||
90133415 DB |
2678 | static bool arg_type_is_mem_ptr(enum bpf_arg_type type) |
2679 | { | |
2680 | return type == ARG_PTR_TO_MEM || | |
2681 | type == ARG_PTR_TO_MEM_OR_NULL || | |
2682 | type == ARG_PTR_TO_UNINIT_MEM; | |
2683 | } | |
2684 | ||
2685 | static bool arg_type_is_mem_size(enum bpf_arg_type type) | |
2686 | { | |
2687 | return type == ARG_CONST_SIZE || | |
2688 | type == ARG_CONST_SIZE_OR_ZERO; | |
2689 | } | |
2690 | ||
57c3bb72 AI |
2691 | static bool arg_type_is_int_ptr(enum bpf_arg_type type) |
2692 | { | |
2693 | return type == ARG_PTR_TO_INT || | |
2694 | type == ARG_PTR_TO_LONG; | |
2695 | } | |
2696 | ||
2697 | static int int_ptr_type_to_size(enum bpf_arg_type type) | |
2698 | { | |
2699 | if (type == ARG_PTR_TO_INT) | |
2700 | return sizeof(u32); | |
2701 | else if (type == ARG_PTR_TO_LONG) | |
2702 | return sizeof(u64); | |
2703 | ||
2704 | return -EINVAL; | |
2705 | } | |
2706 | ||
58e2af8b | 2707 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
2708 | enum bpf_arg_type arg_type, |
2709 | struct bpf_call_arg_meta *meta) | |
17a52670 | 2710 | { |
638f5b90 | 2711 | struct bpf_reg_state *regs = cur_regs(env), *reg = ®s[regno]; |
6841de8b | 2712 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
2713 | int err = 0; |
2714 | ||
80f1d68c | 2715 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
2716 | return 0; |
2717 | ||
dc503a8a EC |
2718 | err = check_reg_arg(env, regno, SRC_OP); |
2719 | if (err) | |
2720 | return err; | |
17a52670 | 2721 | |
1be7f75d AS |
2722 | if (arg_type == ARG_ANYTHING) { |
2723 | if (is_pointer_value(env, regno)) { | |
61bd5218 JK |
2724 | verbose(env, "R%d leaks addr into helper function\n", |
2725 | regno); | |
1be7f75d AS |
2726 | return -EACCES; |
2727 | } | |
80f1d68c | 2728 | return 0; |
1be7f75d | 2729 | } |
80f1d68c | 2730 | |
de8f3a83 | 2731 | if (type_is_pkt_pointer(type) && |
3a0af8fd | 2732 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { |
61bd5218 | 2733 | verbose(env, "helper access to the packet is not allowed\n"); |
6841de8b AS |
2734 | return -EACCES; |
2735 | } | |
2736 | ||
8e2fe1d9 | 2737 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
2ea864c5 | 2738 | arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
2739 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE || |
2740 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) { | |
17a52670 | 2741 | expected_type = PTR_TO_STACK; |
6ac99e8f MKL |
2742 | if (register_is_null(reg) && |
2743 | arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL) | |
2744 | /* final test in check_stack_boundary() */; | |
2745 | else if (!type_is_pkt_pointer(type) && | |
2746 | type != PTR_TO_MAP_VALUE && | |
2747 | type != expected_type) | |
6841de8b | 2748 | goto err_type; |
39f19ebb AS |
2749 | } else if (arg_type == ARG_CONST_SIZE || |
2750 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
f1174f77 EC |
2751 | expected_type = SCALAR_VALUE; |
2752 | if (type != expected_type) | |
6841de8b | 2753 | goto err_type; |
17a52670 AS |
2754 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
2755 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
2756 | if (type != expected_type) |
2757 | goto err_type; | |
608cd71a AS |
2758 | } else if (arg_type == ARG_PTR_TO_CTX) { |
2759 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
2760 | if (type != expected_type) |
2761 | goto err_type; | |
58990d1f DB |
2762 | err = check_ctx_reg(env, reg, regno); |
2763 | if (err < 0) | |
2764 | return err; | |
46f8bc92 MKL |
2765 | } else if (arg_type == ARG_PTR_TO_SOCK_COMMON) { |
2766 | expected_type = PTR_TO_SOCK_COMMON; | |
2767 | /* Any sk pointer can be ARG_PTR_TO_SOCK_COMMON */ | |
2768 | if (!type_is_sk_pointer(type)) | |
2769 | goto err_type; | |
1b986589 MKL |
2770 | if (reg->ref_obj_id) { |
2771 | if (meta->ref_obj_id) { | |
2772 | verbose(env, "verifier internal error: more than one arg with ref_obj_id R%d %u %u\n", | |
2773 | regno, reg->ref_obj_id, | |
2774 | meta->ref_obj_id); | |
2775 | return -EFAULT; | |
2776 | } | |
2777 | meta->ref_obj_id = reg->ref_obj_id; | |
fd978bf7 | 2778 | } |
6ac99e8f MKL |
2779 | } else if (arg_type == ARG_PTR_TO_SOCKET) { |
2780 | expected_type = PTR_TO_SOCKET; | |
2781 | if (type != expected_type) | |
2782 | goto err_type; | |
d83525ca AS |
2783 | } else if (arg_type == ARG_PTR_TO_SPIN_LOCK) { |
2784 | if (meta->func_id == BPF_FUNC_spin_lock) { | |
2785 | if (process_spin_lock(env, regno, true)) | |
2786 | return -EACCES; | |
2787 | } else if (meta->func_id == BPF_FUNC_spin_unlock) { | |
2788 | if (process_spin_lock(env, regno, false)) | |
2789 | return -EACCES; | |
2790 | } else { | |
2791 | verbose(env, "verifier internal error\n"); | |
2792 | return -EFAULT; | |
2793 | } | |
90133415 | 2794 | } else if (arg_type_is_mem_ptr(arg_type)) { |
8e2fe1d9 DB |
2795 | expected_type = PTR_TO_STACK; |
2796 | /* One exception here. In case function allows for NULL to be | |
f1174f77 | 2797 | * passed in as argument, it's a SCALAR_VALUE type. Final test |
8e2fe1d9 DB |
2798 | * happens during stack boundary checking. |
2799 | */ | |
914cb781 | 2800 | if (register_is_null(reg) && |
db1ac496 | 2801 | arg_type == ARG_PTR_TO_MEM_OR_NULL) |
6841de8b | 2802 | /* final test in check_stack_boundary() */; |
de8f3a83 DB |
2803 | else if (!type_is_pkt_pointer(type) && |
2804 | type != PTR_TO_MAP_VALUE && | |
f1174f77 | 2805 | type != expected_type) |
6841de8b | 2806 | goto err_type; |
39f19ebb | 2807 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
57c3bb72 AI |
2808 | } else if (arg_type_is_int_ptr(arg_type)) { |
2809 | expected_type = PTR_TO_STACK; | |
2810 | if (!type_is_pkt_pointer(type) && | |
2811 | type != PTR_TO_MAP_VALUE && | |
2812 | type != expected_type) | |
2813 | goto err_type; | |
17a52670 | 2814 | } else { |
61bd5218 | 2815 | verbose(env, "unsupported arg_type %d\n", arg_type); |
17a52670 AS |
2816 | return -EFAULT; |
2817 | } | |
2818 | ||
17a52670 AS |
2819 | if (arg_type == ARG_CONST_MAP_PTR) { |
2820 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 2821 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
2822 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
2823 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
2824 | * check that [key, key + map->key_size) are within | |
2825 | * stack limits and initialized | |
2826 | */ | |
33ff9823 | 2827 | if (!meta->map_ptr) { |
17a52670 AS |
2828 | /* in function declaration map_ptr must come before |
2829 | * map_key, so that it's verified and known before | |
2830 | * we have to check map_key here. Otherwise it means | |
2831 | * that kernel subsystem misconfigured verifier | |
2832 | */ | |
61bd5218 | 2833 | verbose(env, "invalid map_ptr to access map->key\n"); |
17a52670 AS |
2834 | return -EACCES; |
2835 | } | |
d71962f3 PC |
2836 | err = check_helper_mem_access(env, regno, |
2837 | meta->map_ptr->key_size, false, | |
2838 | NULL); | |
2ea864c5 | 2839 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE || |
6ac99e8f MKL |
2840 | (arg_type == ARG_PTR_TO_MAP_VALUE_OR_NULL && |
2841 | !register_is_null(reg)) || | |
2ea864c5 | 2842 | arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE) { |
17a52670 AS |
2843 | /* bpf_map_xxx(..., map_ptr, ..., value) call: |
2844 | * check [value, value + map->value_size) validity | |
2845 | */ | |
33ff9823 | 2846 | if (!meta->map_ptr) { |
17a52670 | 2847 | /* kernel subsystem misconfigured verifier */ |
61bd5218 | 2848 | verbose(env, "invalid map_ptr to access map->value\n"); |
17a52670 AS |
2849 | return -EACCES; |
2850 | } | |
2ea864c5 | 2851 | meta->raw_mode = (arg_type == ARG_PTR_TO_UNINIT_MAP_VALUE); |
d71962f3 PC |
2852 | err = check_helper_mem_access(env, regno, |
2853 | meta->map_ptr->value_size, false, | |
2ea864c5 | 2854 | meta); |
90133415 | 2855 | } else if (arg_type_is_mem_size(arg_type)) { |
39f19ebb | 2856 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); |
17a52670 | 2857 | |
849fa506 YS |
2858 | /* remember the mem_size which may be used later |
2859 | * to refine return values. | |
2860 | */ | |
2861 | meta->msize_smax_value = reg->smax_value; | |
2862 | meta->msize_umax_value = reg->umax_value; | |
2863 | ||
f1174f77 EC |
2864 | /* The register is SCALAR_VALUE; the access check |
2865 | * happens using its boundaries. | |
06c1c049 | 2866 | */ |
f1174f77 | 2867 | if (!tnum_is_const(reg->var_off)) |
06c1c049 GB |
2868 | /* For unprivileged variable accesses, disable raw |
2869 | * mode so that the program is required to | |
2870 | * initialize all the memory that the helper could | |
2871 | * just partially fill up. | |
2872 | */ | |
2873 | meta = NULL; | |
2874 | ||
b03c9f9f | 2875 | if (reg->smin_value < 0) { |
61bd5218 | 2876 | verbose(env, "R%d min value is negative, either use unsigned or 'var &= const'\n", |
f1174f77 EC |
2877 | regno); |
2878 | return -EACCES; | |
2879 | } | |
06c1c049 | 2880 | |
b03c9f9f | 2881 | if (reg->umin_value == 0) { |
f1174f77 EC |
2882 | err = check_helper_mem_access(env, regno - 1, 0, |
2883 | zero_size_allowed, | |
2884 | meta); | |
06c1c049 GB |
2885 | if (err) |
2886 | return err; | |
06c1c049 | 2887 | } |
f1174f77 | 2888 | |
b03c9f9f | 2889 | if (reg->umax_value >= BPF_MAX_VAR_SIZ) { |
61bd5218 | 2890 | verbose(env, "R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", |
f1174f77 EC |
2891 | regno); |
2892 | return -EACCES; | |
2893 | } | |
2894 | err = check_helper_mem_access(env, regno - 1, | |
b03c9f9f | 2895 | reg->umax_value, |
f1174f77 | 2896 | zero_size_allowed, meta); |
57c3bb72 AI |
2897 | } else if (arg_type_is_int_ptr(arg_type)) { |
2898 | int size = int_ptr_type_to_size(arg_type); | |
2899 | ||
2900 | err = check_helper_mem_access(env, regno, size, false, meta); | |
2901 | if (err) | |
2902 | return err; | |
2903 | err = check_ptr_alignment(env, reg, 0, size, true); | |
17a52670 AS |
2904 | } |
2905 | ||
2906 | return err; | |
6841de8b | 2907 | err_type: |
61bd5218 | 2908 | verbose(env, "R%d type=%s expected=%s\n", regno, |
6841de8b AS |
2909 | reg_type_str[type], reg_type_str[expected_type]); |
2910 | return -EACCES; | |
17a52670 AS |
2911 | } |
2912 | ||
61bd5218 JK |
2913 | static int check_map_func_compatibility(struct bpf_verifier_env *env, |
2914 | struct bpf_map *map, int func_id) | |
35578d79 | 2915 | { |
35578d79 KX |
2916 | if (!map) |
2917 | return 0; | |
2918 | ||
6aff67c8 AS |
2919 | /* We need a two way check, first is from map perspective ... */ |
2920 | switch (map->map_type) { | |
2921 | case BPF_MAP_TYPE_PROG_ARRAY: | |
2922 | if (func_id != BPF_FUNC_tail_call) | |
2923 | goto error; | |
2924 | break; | |
2925 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
2926 | if (func_id != BPF_FUNC_perf_event_read && | |
908432ca YS |
2927 | func_id != BPF_FUNC_perf_event_output && |
2928 | func_id != BPF_FUNC_perf_event_read_value) | |
6aff67c8 AS |
2929 | goto error; |
2930 | break; | |
2931 | case BPF_MAP_TYPE_STACK_TRACE: | |
2932 | if (func_id != BPF_FUNC_get_stackid) | |
2933 | goto error; | |
2934 | break; | |
4ed8ec52 | 2935 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 2936 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 2937 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
2938 | goto error; |
2939 | break; | |
cd339431 | 2940 | case BPF_MAP_TYPE_CGROUP_STORAGE: |
b741f163 | 2941 | case BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE: |
cd339431 RG |
2942 | if (func_id != BPF_FUNC_get_local_storage) |
2943 | goto error; | |
2944 | break; | |
546ac1ff JF |
2945 | /* devmap returns a pointer to a live net_device ifindex that we cannot |
2946 | * allow to be modified from bpf side. So do not allow lookup elements | |
2947 | * for now. | |
2948 | */ | |
2949 | case BPF_MAP_TYPE_DEVMAP: | |
2ddf71e2 | 2950 | if (func_id != BPF_FUNC_redirect_map) |
546ac1ff JF |
2951 | goto error; |
2952 | break; | |
fbfc504a BT |
2953 | /* Restrict bpf side of cpumap and xskmap, open when use-cases |
2954 | * appear. | |
2955 | */ | |
6710e112 JDB |
2956 | case BPF_MAP_TYPE_CPUMAP: |
2957 | if (func_id != BPF_FUNC_redirect_map) | |
2958 | goto error; | |
2959 | break; | |
fada7fdc JL |
2960 | case BPF_MAP_TYPE_XSKMAP: |
2961 | if (func_id != BPF_FUNC_redirect_map && | |
2962 | func_id != BPF_FUNC_map_lookup_elem) | |
2963 | goto error; | |
2964 | break; | |
56f668df | 2965 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 2966 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
2967 | if (func_id != BPF_FUNC_map_lookup_elem) |
2968 | goto error; | |
16a43625 | 2969 | break; |
174a79ff JF |
2970 | case BPF_MAP_TYPE_SOCKMAP: |
2971 | if (func_id != BPF_FUNC_sk_redirect_map && | |
2972 | func_id != BPF_FUNC_sock_map_update && | |
4f738adb JF |
2973 | func_id != BPF_FUNC_map_delete_elem && |
2974 | func_id != BPF_FUNC_msg_redirect_map) | |
174a79ff JF |
2975 | goto error; |
2976 | break; | |
81110384 JF |
2977 | case BPF_MAP_TYPE_SOCKHASH: |
2978 | if (func_id != BPF_FUNC_sk_redirect_hash && | |
2979 | func_id != BPF_FUNC_sock_hash_update && | |
2980 | func_id != BPF_FUNC_map_delete_elem && | |
2981 | func_id != BPF_FUNC_msg_redirect_hash) | |
2982 | goto error; | |
2983 | break; | |
2dbb9b9e MKL |
2984 | case BPF_MAP_TYPE_REUSEPORT_SOCKARRAY: |
2985 | if (func_id != BPF_FUNC_sk_select_reuseport) | |
2986 | goto error; | |
2987 | break; | |
f1a2e44a MV |
2988 | case BPF_MAP_TYPE_QUEUE: |
2989 | case BPF_MAP_TYPE_STACK: | |
2990 | if (func_id != BPF_FUNC_map_peek_elem && | |
2991 | func_id != BPF_FUNC_map_pop_elem && | |
2992 | func_id != BPF_FUNC_map_push_elem) | |
2993 | goto error; | |
2994 | break; | |
6ac99e8f MKL |
2995 | case BPF_MAP_TYPE_SK_STORAGE: |
2996 | if (func_id != BPF_FUNC_sk_storage_get && | |
2997 | func_id != BPF_FUNC_sk_storage_delete) | |
2998 | goto error; | |
2999 | break; | |
6aff67c8 AS |
3000 | default: |
3001 | break; | |
3002 | } | |
3003 | ||
3004 | /* ... and second from the function itself. */ | |
3005 | switch (func_id) { | |
3006 | case BPF_FUNC_tail_call: | |
3007 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
3008 | goto error; | |
f910cefa | 3009 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
3010 | verbose(env, "tail_calls are not allowed in programs with bpf-to-bpf calls\n"); |
3011 | return -EINVAL; | |
3012 | } | |
6aff67c8 AS |
3013 | break; |
3014 | case BPF_FUNC_perf_event_read: | |
3015 | case BPF_FUNC_perf_event_output: | |
908432ca | 3016 | case BPF_FUNC_perf_event_read_value: |
6aff67c8 AS |
3017 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) |
3018 | goto error; | |
3019 | break; | |
3020 | case BPF_FUNC_get_stackid: | |
3021 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
3022 | goto error; | |
3023 | break; | |
60d20f91 | 3024 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 3025 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
3026 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
3027 | goto error; | |
3028 | break; | |
97f91a7c | 3029 | case BPF_FUNC_redirect_map: |
9c270af3 | 3030 | if (map->map_type != BPF_MAP_TYPE_DEVMAP && |
fbfc504a BT |
3031 | map->map_type != BPF_MAP_TYPE_CPUMAP && |
3032 | map->map_type != BPF_MAP_TYPE_XSKMAP) | |
97f91a7c JF |
3033 | goto error; |
3034 | break; | |
174a79ff | 3035 | case BPF_FUNC_sk_redirect_map: |
4f738adb | 3036 | case BPF_FUNC_msg_redirect_map: |
81110384 | 3037 | case BPF_FUNC_sock_map_update: |
174a79ff JF |
3038 | if (map->map_type != BPF_MAP_TYPE_SOCKMAP) |
3039 | goto error; | |
3040 | break; | |
81110384 JF |
3041 | case BPF_FUNC_sk_redirect_hash: |
3042 | case BPF_FUNC_msg_redirect_hash: | |
3043 | case BPF_FUNC_sock_hash_update: | |
3044 | if (map->map_type != BPF_MAP_TYPE_SOCKHASH) | |
174a79ff JF |
3045 | goto error; |
3046 | break; | |
cd339431 | 3047 | case BPF_FUNC_get_local_storage: |
b741f163 RG |
3048 | if (map->map_type != BPF_MAP_TYPE_CGROUP_STORAGE && |
3049 | map->map_type != BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE) | |
cd339431 RG |
3050 | goto error; |
3051 | break; | |
2dbb9b9e MKL |
3052 | case BPF_FUNC_sk_select_reuseport: |
3053 | if (map->map_type != BPF_MAP_TYPE_REUSEPORT_SOCKARRAY) | |
3054 | goto error; | |
3055 | break; | |
f1a2e44a MV |
3056 | case BPF_FUNC_map_peek_elem: |
3057 | case BPF_FUNC_map_pop_elem: | |
3058 | case BPF_FUNC_map_push_elem: | |
3059 | if (map->map_type != BPF_MAP_TYPE_QUEUE && | |
3060 | map->map_type != BPF_MAP_TYPE_STACK) | |
3061 | goto error; | |
3062 | break; | |
6ac99e8f MKL |
3063 | case BPF_FUNC_sk_storage_get: |
3064 | case BPF_FUNC_sk_storage_delete: | |
3065 | if (map->map_type != BPF_MAP_TYPE_SK_STORAGE) | |
3066 | goto error; | |
3067 | break; | |
6aff67c8 AS |
3068 | default: |
3069 | break; | |
35578d79 KX |
3070 | } |
3071 | ||
3072 | return 0; | |
6aff67c8 | 3073 | error: |
61bd5218 | 3074 | verbose(env, "cannot pass map_type %d into func %s#%d\n", |
ebb676da | 3075 | map->map_type, func_id_name(func_id), func_id); |
6aff67c8 | 3076 | return -EINVAL; |
35578d79 KX |
3077 | } |
3078 | ||
90133415 | 3079 | static bool check_raw_mode_ok(const struct bpf_func_proto *fn) |
435faee1 DB |
3080 | { |
3081 | int count = 0; | |
3082 | ||
39f19ebb | 3083 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 3084 | count++; |
39f19ebb | 3085 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 3086 | count++; |
39f19ebb | 3087 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 3088 | count++; |
39f19ebb | 3089 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 3090 | count++; |
39f19ebb | 3091 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
3092 | count++; |
3093 | ||
90133415 DB |
3094 | /* We only support one arg being in raw mode at the moment, |
3095 | * which is sufficient for the helper functions we have | |
3096 | * right now. | |
3097 | */ | |
3098 | return count <= 1; | |
3099 | } | |
3100 | ||
3101 | static bool check_args_pair_invalid(enum bpf_arg_type arg_curr, | |
3102 | enum bpf_arg_type arg_next) | |
3103 | { | |
3104 | return (arg_type_is_mem_ptr(arg_curr) && | |
3105 | !arg_type_is_mem_size(arg_next)) || | |
3106 | (!arg_type_is_mem_ptr(arg_curr) && | |
3107 | arg_type_is_mem_size(arg_next)); | |
3108 | } | |
3109 | ||
3110 | static bool check_arg_pair_ok(const struct bpf_func_proto *fn) | |
3111 | { | |
3112 | /* bpf_xxx(..., buf, len) call will access 'len' | |
3113 | * bytes from memory 'buf'. Both arg types need | |
3114 | * to be paired, so make sure there's no buggy | |
3115 | * helper function specification. | |
3116 | */ | |
3117 | if (arg_type_is_mem_size(fn->arg1_type) || | |
3118 | arg_type_is_mem_ptr(fn->arg5_type) || | |
3119 | check_args_pair_invalid(fn->arg1_type, fn->arg2_type) || | |
3120 | check_args_pair_invalid(fn->arg2_type, fn->arg3_type) || | |
3121 | check_args_pair_invalid(fn->arg3_type, fn->arg4_type) || | |
3122 | check_args_pair_invalid(fn->arg4_type, fn->arg5_type)) | |
3123 | return false; | |
3124 | ||
3125 | return true; | |
3126 | } | |
3127 | ||
1b986589 | 3128 | static bool check_refcount_ok(const struct bpf_func_proto *fn, int func_id) |
fd978bf7 JS |
3129 | { |
3130 | int count = 0; | |
3131 | ||
1b986589 | 3132 | if (arg_type_may_be_refcounted(fn->arg1_type)) |
fd978bf7 | 3133 | count++; |
1b986589 | 3134 | if (arg_type_may_be_refcounted(fn->arg2_type)) |
fd978bf7 | 3135 | count++; |
1b986589 | 3136 | if (arg_type_may_be_refcounted(fn->arg3_type)) |
fd978bf7 | 3137 | count++; |
1b986589 | 3138 | if (arg_type_may_be_refcounted(fn->arg4_type)) |
fd978bf7 | 3139 | count++; |
1b986589 | 3140 | if (arg_type_may_be_refcounted(fn->arg5_type)) |
fd978bf7 JS |
3141 | count++; |
3142 | ||
1b986589 MKL |
3143 | /* A reference acquiring function cannot acquire |
3144 | * another refcounted ptr. | |
3145 | */ | |
3146 | if (is_acquire_function(func_id) && count) | |
3147 | return false; | |
3148 | ||
fd978bf7 JS |
3149 | /* We only support one arg being unreferenced at the moment, |
3150 | * which is sufficient for the helper functions we have right now. | |
3151 | */ | |
3152 | return count <= 1; | |
3153 | } | |
3154 | ||
1b986589 | 3155 | static int check_func_proto(const struct bpf_func_proto *fn, int func_id) |
90133415 DB |
3156 | { |
3157 | return check_raw_mode_ok(fn) && | |
fd978bf7 | 3158 | check_arg_pair_ok(fn) && |
1b986589 | 3159 | check_refcount_ok(fn, func_id) ? 0 : -EINVAL; |
435faee1 DB |
3160 | } |
3161 | ||
de8f3a83 DB |
3162 | /* Packet data might have moved, any old PTR_TO_PACKET[_META,_END] |
3163 | * are now invalid, so turn them into unknown SCALAR_VALUE. | |
f1174f77 | 3164 | */ |
f4d7e40a AS |
3165 | static void __clear_all_pkt_pointers(struct bpf_verifier_env *env, |
3166 | struct bpf_func_state *state) | |
969bf05e | 3167 | { |
58e2af8b | 3168 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e AS |
3169 | int i; |
3170 | ||
3171 | for (i = 0; i < MAX_BPF_REG; i++) | |
de8f3a83 | 3172 | if (reg_is_pkt_pointer_any(®s[i])) |
61bd5218 | 3173 | mark_reg_unknown(env, regs, i); |
969bf05e | 3174 | |
f3709f69 JS |
3175 | bpf_for_each_spilled_reg(i, state, reg) { |
3176 | if (!reg) | |
969bf05e | 3177 | continue; |
de8f3a83 DB |
3178 | if (reg_is_pkt_pointer_any(reg)) |
3179 | __mark_reg_unknown(reg); | |
969bf05e AS |
3180 | } |
3181 | } | |
3182 | ||
f4d7e40a AS |
3183 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
3184 | { | |
3185 | struct bpf_verifier_state *vstate = env->cur_state; | |
3186 | int i; | |
3187 | ||
3188 | for (i = 0; i <= vstate->curframe; i++) | |
3189 | __clear_all_pkt_pointers(env, vstate->frame[i]); | |
3190 | } | |
3191 | ||
fd978bf7 | 3192 | static void release_reg_references(struct bpf_verifier_env *env, |
1b986589 MKL |
3193 | struct bpf_func_state *state, |
3194 | int ref_obj_id) | |
fd978bf7 JS |
3195 | { |
3196 | struct bpf_reg_state *regs = state->regs, *reg; | |
3197 | int i; | |
3198 | ||
3199 | for (i = 0; i < MAX_BPF_REG; i++) | |
1b986589 | 3200 | if (regs[i].ref_obj_id == ref_obj_id) |
fd978bf7 JS |
3201 | mark_reg_unknown(env, regs, i); |
3202 | ||
3203 | bpf_for_each_spilled_reg(i, state, reg) { | |
3204 | if (!reg) | |
3205 | continue; | |
1b986589 | 3206 | if (reg->ref_obj_id == ref_obj_id) |
fd978bf7 JS |
3207 | __mark_reg_unknown(reg); |
3208 | } | |
3209 | } | |
3210 | ||
3211 | /* The pointer with the specified id has released its reference to kernel | |
3212 | * resources. Identify all copies of the same pointer and clear the reference. | |
3213 | */ | |
3214 | static int release_reference(struct bpf_verifier_env *env, | |
1b986589 | 3215 | int ref_obj_id) |
fd978bf7 JS |
3216 | { |
3217 | struct bpf_verifier_state *vstate = env->cur_state; | |
1b986589 | 3218 | int err; |
fd978bf7 JS |
3219 | int i; |
3220 | ||
1b986589 MKL |
3221 | err = release_reference_state(cur_func(env), ref_obj_id); |
3222 | if (err) | |
3223 | return err; | |
3224 | ||
fd978bf7 | 3225 | for (i = 0; i <= vstate->curframe; i++) |
1b986589 | 3226 | release_reg_references(env, vstate->frame[i], ref_obj_id); |
fd978bf7 | 3227 | |
1b986589 | 3228 | return 0; |
fd978bf7 JS |
3229 | } |
3230 | ||
f4d7e40a AS |
3231 | static int check_func_call(struct bpf_verifier_env *env, struct bpf_insn *insn, |
3232 | int *insn_idx) | |
3233 | { | |
3234 | struct bpf_verifier_state *state = env->cur_state; | |
3235 | struct bpf_func_state *caller, *callee; | |
fd978bf7 | 3236 | int i, err, subprog, target_insn; |
f4d7e40a | 3237 | |
aada9ce6 | 3238 | if (state->curframe + 1 >= MAX_CALL_FRAMES) { |
f4d7e40a | 3239 | verbose(env, "the call stack of %d frames is too deep\n", |
aada9ce6 | 3240 | state->curframe + 2); |
f4d7e40a AS |
3241 | return -E2BIG; |
3242 | } | |
3243 | ||
3244 | target_insn = *insn_idx + insn->imm; | |
3245 | subprog = find_subprog(env, target_insn + 1); | |
3246 | if (subprog < 0) { | |
3247 | verbose(env, "verifier bug. No program starts at insn %d\n", | |
3248 | target_insn + 1); | |
3249 | return -EFAULT; | |
3250 | } | |
3251 | ||
3252 | caller = state->frame[state->curframe]; | |
3253 | if (state->frame[state->curframe + 1]) { | |
3254 | verbose(env, "verifier bug. Frame %d already allocated\n", | |
3255 | state->curframe + 1); | |
3256 | return -EFAULT; | |
3257 | } | |
3258 | ||
3259 | callee = kzalloc(sizeof(*callee), GFP_KERNEL); | |
3260 | if (!callee) | |
3261 | return -ENOMEM; | |
3262 | state->frame[state->curframe + 1] = callee; | |
3263 | ||
3264 | /* callee cannot access r0, r6 - r9 for reading and has to write | |
3265 | * into its own stack before reading from it. | |
3266 | * callee can read/write into caller's stack | |
3267 | */ | |
3268 | init_func_state(env, callee, | |
3269 | /* remember the callsite, it will be used by bpf_exit */ | |
3270 | *insn_idx /* callsite */, | |
3271 | state->curframe + 1 /* frameno within this callchain */, | |
f910cefa | 3272 | subprog /* subprog number within this prog */); |
f4d7e40a | 3273 | |
fd978bf7 JS |
3274 | /* Transfer references to the callee */ |
3275 | err = transfer_reference_state(callee, caller); | |
3276 | if (err) | |
3277 | return err; | |
3278 | ||
679c782d EC |
3279 | /* copy r1 - r5 args that callee can access. The copy includes parent |
3280 | * pointers, which connects us up to the liveness chain | |
3281 | */ | |
f4d7e40a AS |
3282 | for (i = BPF_REG_1; i <= BPF_REG_5; i++) |
3283 | callee->regs[i] = caller->regs[i]; | |
3284 | ||
679c782d | 3285 | /* after the call registers r0 - r5 were scratched */ |
f4d7e40a AS |
3286 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
3287 | mark_reg_not_init(env, caller->regs, caller_saved[i]); | |
3288 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); | |
3289 | } | |
3290 | ||
3291 | /* only increment it after check_reg_arg() finished */ | |
3292 | state->curframe++; | |
3293 | ||
3294 | /* and go analyze first insn of the callee */ | |
3295 | *insn_idx = target_insn; | |
3296 | ||
06ee7115 | 3297 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
3298 | verbose(env, "caller:\n"); |
3299 | print_verifier_state(env, caller); | |
3300 | verbose(env, "callee:\n"); | |
3301 | print_verifier_state(env, callee); | |
3302 | } | |
3303 | return 0; | |
3304 | } | |
3305 | ||
3306 | static int prepare_func_exit(struct bpf_verifier_env *env, int *insn_idx) | |
3307 | { | |
3308 | struct bpf_verifier_state *state = env->cur_state; | |
3309 | struct bpf_func_state *caller, *callee; | |
3310 | struct bpf_reg_state *r0; | |
fd978bf7 | 3311 | int err; |
f4d7e40a AS |
3312 | |
3313 | callee = state->frame[state->curframe]; | |
3314 | r0 = &callee->regs[BPF_REG_0]; | |
3315 | if (r0->type == PTR_TO_STACK) { | |
3316 | /* technically it's ok to return caller's stack pointer | |
3317 | * (or caller's caller's pointer) back to the caller, | |
3318 | * since these pointers are valid. Only current stack | |
3319 | * pointer will be invalid as soon as function exits, | |
3320 | * but let's be conservative | |
3321 | */ | |
3322 | verbose(env, "cannot return stack pointer to the caller\n"); | |
3323 | return -EINVAL; | |
3324 | } | |
3325 | ||
3326 | state->curframe--; | |
3327 | caller = state->frame[state->curframe]; | |
3328 | /* return to the caller whatever r0 had in the callee */ | |
3329 | caller->regs[BPF_REG_0] = *r0; | |
3330 | ||
fd978bf7 JS |
3331 | /* Transfer references to the caller */ |
3332 | err = transfer_reference_state(caller, callee); | |
3333 | if (err) | |
3334 | return err; | |
3335 | ||
f4d7e40a | 3336 | *insn_idx = callee->callsite + 1; |
06ee7115 | 3337 | if (env->log.level & BPF_LOG_LEVEL) { |
f4d7e40a AS |
3338 | verbose(env, "returning from callee:\n"); |
3339 | print_verifier_state(env, callee); | |
3340 | verbose(env, "to caller at %d:\n", *insn_idx); | |
3341 | print_verifier_state(env, caller); | |
3342 | } | |
3343 | /* clear everything in the callee */ | |
3344 | free_func_state(callee); | |
3345 | state->frame[state->curframe + 1] = NULL; | |
3346 | return 0; | |
3347 | } | |
3348 | ||
849fa506 YS |
3349 | static void do_refine_retval_range(struct bpf_reg_state *regs, int ret_type, |
3350 | int func_id, | |
3351 | struct bpf_call_arg_meta *meta) | |
3352 | { | |
3353 | struct bpf_reg_state *ret_reg = ®s[BPF_REG_0]; | |
3354 | ||
3355 | if (ret_type != RET_INTEGER || | |
3356 | (func_id != BPF_FUNC_get_stack && | |
3357 | func_id != BPF_FUNC_probe_read_str)) | |
3358 | return; | |
3359 | ||
3360 | ret_reg->smax_value = meta->msize_smax_value; | |
3361 | ret_reg->umax_value = meta->msize_umax_value; | |
3362 | __reg_deduce_bounds(ret_reg); | |
3363 | __reg_bound_offset(ret_reg); | |
3364 | } | |
3365 | ||
c93552c4 DB |
3366 | static int |
3367 | record_func_map(struct bpf_verifier_env *env, struct bpf_call_arg_meta *meta, | |
3368 | int func_id, int insn_idx) | |
3369 | { | |
3370 | struct bpf_insn_aux_data *aux = &env->insn_aux_data[insn_idx]; | |
591fe988 | 3371 | struct bpf_map *map = meta->map_ptr; |
c93552c4 DB |
3372 | |
3373 | if (func_id != BPF_FUNC_tail_call && | |
09772d92 DB |
3374 | func_id != BPF_FUNC_map_lookup_elem && |
3375 | func_id != BPF_FUNC_map_update_elem && | |
f1a2e44a MV |
3376 | func_id != BPF_FUNC_map_delete_elem && |
3377 | func_id != BPF_FUNC_map_push_elem && | |
3378 | func_id != BPF_FUNC_map_pop_elem && | |
3379 | func_id != BPF_FUNC_map_peek_elem) | |
c93552c4 | 3380 | return 0; |
09772d92 | 3381 | |
591fe988 | 3382 | if (map == NULL) { |
c93552c4 DB |
3383 | verbose(env, "kernel subsystem misconfigured verifier\n"); |
3384 | return -EINVAL; | |
3385 | } | |
3386 | ||
591fe988 DB |
3387 | /* In case of read-only, some additional restrictions |
3388 | * need to be applied in order to prevent altering the | |
3389 | * state of the map from program side. | |
3390 | */ | |
3391 | if ((map->map_flags & BPF_F_RDONLY_PROG) && | |
3392 | (func_id == BPF_FUNC_map_delete_elem || | |
3393 | func_id == BPF_FUNC_map_update_elem || | |
3394 | func_id == BPF_FUNC_map_push_elem || | |
3395 | func_id == BPF_FUNC_map_pop_elem)) { | |
3396 | verbose(env, "write into map forbidden\n"); | |
3397 | return -EACCES; | |
3398 | } | |
3399 | ||
c93552c4 DB |
3400 | if (!BPF_MAP_PTR(aux->map_state)) |
3401 | bpf_map_ptr_store(aux, meta->map_ptr, | |
3402 | meta->map_ptr->unpriv_array); | |
3403 | else if (BPF_MAP_PTR(aux->map_state) != meta->map_ptr) | |
3404 | bpf_map_ptr_store(aux, BPF_MAP_PTR_POISON, | |
3405 | meta->map_ptr->unpriv_array); | |
3406 | return 0; | |
3407 | } | |
3408 | ||
fd978bf7 JS |
3409 | static int check_reference_leak(struct bpf_verifier_env *env) |
3410 | { | |
3411 | struct bpf_func_state *state = cur_func(env); | |
3412 | int i; | |
3413 | ||
3414 | for (i = 0; i < state->acquired_refs; i++) { | |
3415 | verbose(env, "Unreleased reference id=%d alloc_insn=%d\n", | |
3416 | state->refs[i].id, state->refs[i].insn_idx); | |
3417 | } | |
3418 | return state->acquired_refs ? -EINVAL : 0; | |
3419 | } | |
3420 | ||
f4d7e40a | 3421 | static int check_helper_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 3422 | { |
17a52670 | 3423 | const struct bpf_func_proto *fn = NULL; |
638f5b90 | 3424 | struct bpf_reg_state *regs; |
33ff9823 | 3425 | struct bpf_call_arg_meta meta; |
969bf05e | 3426 | bool changes_data; |
17a52670 AS |
3427 | int i, err; |
3428 | ||
3429 | /* find function prototype */ | |
3430 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
61bd5218 JK |
3431 | verbose(env, "invalid func %s#%d\n", func_id_name(func_id), |
3432 | func_id); | |
17a52670 AS |
3433 | return -EINVAL; |
3434 | } | |
3435 | ||
00176a34 | 3436 | if (env->ops->get_func_proto) |
5e43f899 | 3437 | fn = env->ops->get_func_proto(func_id, env->prog); |
17a52670 | 3438 | if (!fn) { |
61bd5218 JK |
3439 | verbose(env, "unknown func %s#%d\n", func_id_name(func_id), |
3440 | func_id); | |
17a52670 AS |
3441 | return -EINVAL; |
3442 | } | |
3443 | ||
3444 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 3445 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
3fe2867c | 3446 | verbose(env, "cannot call GPL-restricted function from non-GPL compatible program\n"); |
17a52670 AS |
3447 | return -EINVAL; |
3448 | } | |
3449 | ||
04514d13 | 3450 | /* With LD_ABS/IND some JITs save/restore skb from r1. */ |
17bedab2 | 3451 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
04514d13 DB |
3452 | if (changes_data && fn->arg1_type != ARG_PTR_TO_CTX) { |
3453 | verbose(env, "kernel subsystem misconfigured func %s#%d: r1 != ctx\n", | |
3454 | func_id_name(func_id), func_id); | |
3455 | return -EINVAL; | |
3456 | } | |
969bf05e | 3457 | |
33ff9823 | 3458 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 3459 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 3460 | |
1b986589 | 3461 | err = check_func_proto(fn, func_id); |
435faee1 | 3462 | if (err) { |
61bd5218 | 3463 | verbose(env, "kernel subsystem misconfigured func %s#%d\n", |
ebb676da | 3464 | func_id_name(func_id), func_id); |
435faee1 DB |
3465 | return err; |
3466 | } | |
3467 | ||
d83525ca | 3468 | meta.func_id = func_id; |
17a52670 | 3469 | /* check args */ |
33ff9823 | 3470 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
3471 | if (err) |
3472 | return err; | |
33ff9823 | 3473 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
3474 | if (err) |
3475 | return err; | |
33ff9823 | 3476 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
3477 | if (err) |
3478 | return err; | |
33ff9823 | 3479 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
3480 | if (err) |
3481 | return err; | |
33ff9823 | 3482 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
3483 | if (err) |
3484 | return err; | |
3485 | ||
c93552c4 DB |
3486 | err = record_func_map(env, &meta, func_id, insn_idx); |
3487 | if (err) | |
3488 | return err; | |
3489 | ||
435faee1 DB |
3490 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
3491 | * is inferred from register state. | |
3492 | */ | |
3493 | for (i = 0; i < meta.access_size; i++) { | |
ca369602 DB |
3494 | err = check_mem_access(env, insn_idx, meta.regno, i, BPF_B, |
3495 | BPF_WRITE, -1, false); | |
435faee1 DB |
3496 | if (err) |
3497 | return err; | |
3498 | } | |
3499 | ||
fd978bf7 JS |
3500 | if (func_id == BPF_FUNC_tail_call) { |
3501 | err = check_reference_leak(env); | |
3502 | if (err) { | |
3503 | verbose(env, "tail_call would lead to reference leak\n"); | |
3504 | return err; | |
3505 | } | |
3506 | } else if (is_release_function(func_id)) { | |
1b986589 | 3507 | err = release_reference(env, meta.ref_obj_id); |
46f8bc92 MKL |
3508 | if (err) { |
3509 | verbose(env, "func %s#%d reference has not been acquired before\n", | |
3510 | func_id_name(func_id), func_id); | |
fd978bf7 | 3511 | return err; |
46f8bc92 | 3512 | } |
fd978bf7 JS |
3513 | } |
3514 | ||
638f5b90 | 3515 | regs = cur_regs(env); |
cd339431 RG |
3516 | |
3517 | /* check that flags argument in get_local_storage(map, flags) is 0, | |
3518 | * this is required because get_local_storage() can't return an error. | |
3519 | */ | |
3520 | if (func_id == BPF_FUNC_get_local_storage && | |
3521 | !register_is_null(®s[BPF_REG_2])) { | |
3522 | verbose(env, "get_local_storage() doesn't support non-zero flags\n"); | |
3523 | return -EINVAL; | |
3524 | } | |
3525 | ||
17a52670 | 3526 | /* reset caller saved regs */ |
dc503a8a | 3527 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 3528 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
3529 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
3530 | } | |
17a52670 | 3531 | |
5327ed3d JW |
3532 | /* helper call returns 64-bit value. */ |
3533 | regs[BPF_REG_0].subreg_def = DEF_NOT_SUBREG; | |
3534 | ||
dc503a8a | 3535 | /* update return register (already marked as written above) */ |
17a52670 | 3536 | if (fn->ret_type == RET_INTEGER) { |
f1174f77 | 3537 | /* sets type to SCALAR_VALUE */ |
61bd5218 | 3538 | mark_reg_unknown(env, regs, BPF_REG_0); |
17a52670 AS |
3539 | } else if (fn->ret_type == RET_VOID) { |
3540 | regs[BPF_REG_0].type = NOT_INIT; | |
3e6a4b3e RG |
3541 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL || |
3542 | fn->ret_type == RET_PTR_TO_MAP_VALUE) { | |
f1174f77 | 3543 | /* There is no offset yet applied, variable or fixed */ |
61bd5218 | 3544 | mark_reg_known_zero(env, regs, BPF_REG_0); |
17a52670 AS |
3545 | /* remember map_ptr, so that check_map_access() |
3546 | * can check 'value_size' boundary of memory access | |
3547 | * to map element returned from bpf_map_lookup_elem() | |
3548 | */ | |
33ff9823 | 3549 | if (meta.map_ptr == NULL) { |
61bd5218 JK |
3550 | verbose(env, |
3551 | "kernel subsystem misconfigured verifier\n"); | |
17a52670 AS |
3552 | return -EINVAL; |
3553 | } | |
33ff9823 | 3554 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
4d31f301 DB |
3555 | if (fn->ret_type == RET_PTR_TO_MAP_VALUE) { |
3556 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE; | |
e16d2f1a AS |
3557 | if (map_value_has_spin_lock(meta.map_ptr)) |
3558 | regs[BPF_REG_0].id = ++env->id_gen; | |
4d31f301 DB |
3559 | } else { |
3560 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
3561 | regs[BPF_REG_0].id = ++env->id_gen; | |
3562 | } | |
c64b7983 JS |
3563 | } else if (fn->ret_type == RET_PTR_TO_SOCKET_OR_NULL) { |
3564 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
3565 | regs[BPF_REG_0].type = PTR_TO_SOCKET_OR_NULL; | |
0f3adc28 | 3566 | regs[BPF_REG_0].id = ++env->id_gen; |
85a51f8c LB |
3567 | } else if (fn->ret_type == RET_PTR_TO_SOCK_COMMON_OR_NULL) { |
3568 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
3569 | regs[BPF_REG_0].type = PTR_TO_SOCK_COMMON_OR_NULL; | |
3570 | regs[BPF_REG_0].id = ++env->id_gen; | |
655a51e5 MKL |
3571 | } else if (fn->ret_type == RET_PTR_TO_TCP_SOCK_OR_NULL) { |
3572 | mark_reg_known_zero(env, regs, BPF_REG_0); | |
3573 | regs[BPF_REG_0].type = PTR_TO_TCP_SOCK_OR_NULL; | |
3574 | regs[BPF_REG_0].id = ++env->id_gen; | |
17a52670 | 3575 | } else { |
61bd5218 | 3576 | verbose(env, "unknown return type %d of func %s#%d\n", |
ebb676da | 3577 | fn->ret_type, func_id_name(func_id), func_id); |
17a52670 AS |
3578 | return -EINVAL; |
3579 | } | |
04fd61ab | 3580 | |
0f3adc28 | 3581 | if (is_ptr_cast_function(func_id)) { |
1b986589 MKL |
3582 | /* For release_reference() */ |
3583 | regs[BPF_REG_0].ref_obj_id = meta.ref_obj_id; | |
0f3adc28 LB |
3584 | } else if (is_acquire_function(func_id)) { |
3585 | int id = acquire_reference_state(env, insn_idx); | |
3586 | ||
3587 | if (id < 0) | |
3588 | return id; | |
3589 | /* For mark_ptr_or_null_reg() */ | |
3590 | regs[BPF_REG_0].id = id; | |
3591 | /* For release_reference() */ | |
3592 | regs[BPF_REG_0].ref_obj_id = id; | |
3593 | } | |
1b986589 | 3594 | |
849fa506 YS |
3595 | do_refine_retval_range(regs, fn->ret_type, func_id, &meta); |
3596 | ||
61bd5218 | 3597 | err = check_map_func_compatibility(env, meta.map_ptr, func_id); |
35578d79 KX |
3598 | if (err) |
3599 | return err; | |
04fd61ab | 3600 | |
c195651e YS |
3601 | if (func_id == BPF_FUNC_get_stack && !env->prog->has_callchain_buf) { |
3602 | const char *err_str; | |
3603 | ||
3604 | #ifdef CONFIG_PERF_EVENTS | |
3605 | err = get_callchain_buffers(sysctl_perf_event_max_stack); | |
3606 | err_str = "cannot get callchain buffer for func %s#%d\n"; | |
3607 | #else | |
3608 | err = -ENOTSUPP; | |
3609 | err_str = "func %s#%d not supported without CONFIG_PERF_EVENTS\n"; | |
3610 | #endif | |
3611 | if (err) { | |
3612 | verbose(env, err_str, func_id_name(func_id), func_id); | |
3613 | return err; | |
3614 | } | |
3615 | ||
3616 | env->prog->has_callchain_buf = true; | |
3617 | } | |
3618 | ||
969bf05e AS |
3619 | if (changes_data) |
3620 | clear_all_pkt_pointers(env); | |
3621 | return 0; | |
3622 | } | |
3623 | ||
b03c9f9f EC |
3624 | static bool signed_add_overflows(s64 a, s64 b) |
3625 | { | |
3626 | /* Do the add in u64, where overflow is well-defined */ | |
3627 | s64 res = (s64)((u64)a + (u64)b); | |
3628 | ||
3629 | if (b < 0) | |
3630 | return res > a; | |
3631 | return res < a; | |
3632 | } | |
3633 | ||
3634 | static bool signed_sub_overflows(s64 a, s64 b) | |
3635 | { | |
3636 | /* Do the sub in u64, where overflow is well-defined */ | |
3637 | s64 res = (s64)((u64)a - (u64)b); | |
3638 | ||
3639 | if (b < 0) | |
3640 | return res < a; | |
3641 | return res > a; | |
969bf05e AS |
3642 | } |
3643 | ||
bb7f0f98 AS |
3644 | static bool check_reg_sane_offset(struct bpf_verifier_env *env, |
3645 | const struct bpf_reg_state *reg, | |
3646 | enum bpf_reg_type type) | |
3647 | { | |
3648 | bool known = tnum_is_const(reg->var_off); | |
3649 | s64 val = reg->var_off.value; | |
3650 | s64 smin = reg->smin_value; | |
3651 | ||
3652 | if (known && (val >= BPF_MAX_VAR_OFF || val <= -BPF_MAX_VAR_OFF)) { | |
3653 | verbose(env, "math between %s pointer and %lld is not allowed\n", | |
3654 | reg_type_str[type], val); | |
3655 | return false; | |
3656 | } | |
3657 | ||
3658 | if (reg->off >= BPF_MAX_VAR_OFF || reg->off <= -BPF_MAX_VAR_OFF) { | |
3659 | verbose(env, "%s pointer offset %d is not allowed\n", | |
3660 | reg_type_str[type], reg->off); | |
3661 | return false; | |
3662 | } | |
3663 | ||
3664 | if (smin == S64_MIN) { | |
3665 | verbose(env, "math between %s pointer and register with unbounded min value is not allowed\n", | |
3666 | reg_type_str[type]); | |
3667 | return false; | |
3668 | } | |
3669 | ||
3670 | if (smin >= BPF_MAX_VAR_OFF || smin <= -BPF_MAX_VAR_OFF) { | |
3671 | verbose(env, "value %lld makes %s pointer be out of bounds\n", | |
3672 | smin, reg_type_str[type]); | |
3673 | return false; | |
3674 | } | |
3675 | ||
3676 | return true; | |
3677 | } | |
3678 | ||
979d63d5 DB |
3679 | static struct bpf_insn_aux_data *cur_aux(struct bpf_verifier_env *env) |
3680 | { | |
3681 | return &env->insn_aux_data[env->insn_idx]; | |
3682 | } | |
3683 | ||
3684 | static int retrieve_ptr_limit(const struct bpf_reg_state *ptr_reg, | |
3685 | u32 *ptr_limit, u8 opcode, bool off_is_neg) | |
3686 | { | |
3687 | bool mask_to_left = (opcode == BPF_ADD && off_is_neg) || | |
3688 | (opcode == BPF_SUB && !off_is_neg); | |
3689 | u32 off; | |
3690 | ||
3691 | switch (ptr_reg->type) { | |
3692 | case PTR_TO_STACK: | |
088ec26d AI |
3693 | /* Indirect variable offset stack access is prohibited in |
3694 | * unprivileged mode so it's not handled here. | |
3695 | */ | |
979d63d5 DB |
3696 | off = ptr_reg->off + ptr_reg->var_off.value; |
3697 | if (mask_to_left) | |
3698 | *ptr_limit = MAX_BPF_STACK + off; | |
3699 | else | |
3700 | *ptr_limit = -off; | |
3701 | return 0; | |
3702 | case PTR_TO_MAP_VALUE: | |
3703 | if (mask_to_left) { | |
3704 | *ptr_limit = ptr_reg->umax_value + ptr_reg->off; | |
3705 | } else { | |
3706 | off = ptr_reg->smin_value + ptr_reg->off; | |
3707 | *ptr_limit = ptr_reg->map_ptr->value_size - off; | |
3708 | } | |
3709 | return 0; | |
3710 | default: | |
3711 | return -EINVAL; | |
3712 | } | |
3713 | } | |
3714 | ||
d3bd7413 DB |
3715 | static bool can_skip_alu_sanitation(const struct bpf_verifier_env *env, |
3716 | const struct bpf_insn *insn) | |
3717 | { | |
3718 | return env->allow_ptr_leaks || BPF_SRC(insn->code) == BPF_K; | |
3719 | } | |
3720 | ||
3721 | static int update_alu_sanitation_state(struct bpf_insn_aux_data *aux, | |
3722 | u32 alu_state, u32 alu_limit) | |
3723 | { | |
3724 | /* If we arrived here from different branches with different | |
3725 | * state or limits to sanitize, then this won't work. | |
3726 | */ | |
3727 | if (aux->alu_state && | |
3728 | (aux->alu_state != alu_state || | |
3729 | aux->alu_limit != alu_limit)) | |
3730 | return -EACCES; | |
3731 | ||
3732 | /* Corresponding fixup done in fixup_bpf_calls(). */ | |
3733 | aux->alu_state = alu_state; | |
3734 | aux->alu_limit = alu_limit; | |
3735 | return 0; | |
3736 | } | |
3737 | ||
3738 | static int sanitize_val_alu(struct bpf_verifier_env *env, | |
3739 | struct bpf_insn *insn) | |
3740 | { | |
3741 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
3742 | ||
3743 | if (can_skip_alu_sanitation(env, insn)) | |
3744 | return 0; | |
3745 | ||
3746 | return update_alu_sanitation_state(aux, BPF_ALU_NON_POINTER, 0); | |
3747 | } | |
3748 | ||
979d63d5 DB |
3749 | static int sanitize_ptr_alu(struct bpf_verifier_env *env, |
3750 | struct bpf_insn *insn, | |
3751 | const struct bpf_reg_state *ptr_reg, | |
3752 | struct bpf_reg_state *dst_reg, | |
3753 | bool off_is_neg) | |
3754 | { | |
3755 | struct bpf_verifier_state *vstate = env->cur_state; | |
3756 | struct bpf_insn_aux_data *aux = cur_aux(env); | |
3757 | bool ptr_is_dst_reg = ptr_reg == dst_reg; | |
3758 | u8 opcode = BPF_OP(insn->code); | |
3759 | u32 alu_state, alu_limit; | |
3760 | struct bpf_reg_state tmp; | |
3761 | bool ret; | |
3762 | ||
d3bd7413 | 3763 | if (can_skip_alu_sanitation(env, insn)) |
979d63d5 DB |
3764 | return 0; |
3765 | ||
3766 | /* We already marked aux for masking from non-speculative | |
3767 | * paths, thus we got here in the first place. We only care | |
3768 | * to explore bad access from here. | |
3769 | */ | |
3770 | if (vstate->speculative) | |
3771 | goto do_sim; | |
3772 | ||
3773 | alu_state = off_is_neg ? BPF_ALU_NEG_VALUE : 0; | |
3774 | alu_state |= ptr_is_dst_reg ? | |
3775 | BPF_ALU_SANITIZE_SRC : BPF_ALU_SANITIZE_DST; | |
3776 | ||
3777 | if (retrieve_ptr_limit(ptr_reg, &alu_limit, opcode, off_is_neg)) | |
3778 | return 0; | |
d3bd7413 | 3779 | if (update_alu_sanitation_state(aux, alu_state, alu_limit)) |
979d63d5 | 3780 | return -EACCES; |
979d63d5 DB |
3781 | do_sim: |
3782 | /* Simulate and find potential out-of-bounds access under | |
3783 | * speculative execution from truncation as a result of | |
3784 | * masking when off was not within expected range. If off | |
3785 | * sits in dst, then we temporarily need to move ptr there | |
3786 | * to simulate dst (== 0) +/-= ptr. Needed, for example, | |
3787 | * for cases where we use K-based arithmetic in one direction | |
3788 | * and truncated reg-based in the other in order to explore | |
3789 | * bad access. | |
3790 | */ | |
3791 | if (!ptr_is_dst_reg) { | |
3792 | tmp = *dst_reg; | |
3793 | *dst_reg = *ptr_reg; | |
3794 | } | |
3795 | ret = push_stack(env, env->insn_idx + 1, env->insn_idx, true); | |
0803278b | 3796 | if (!ptr_is_dst_reg && ret) |
979d63d5 DB |
3797 | *dst_reg = tmp; |
3798 | return !ret ? -EFAULT : 0; | |
3799 | } | |
3800 | ||
f1174f77 | 3801 | /* Handles arithmetic on a pointer and a scalar: computes new min/max and var_off. |
f1174f77 EC |
3802 | * Caller should also handle BPF_MOV case separately. |
3803 | * If we return -EACCES, caller may want to try again treating pointer as a | |
3804 | * scalar. So we only emit a diagnostic if !env->allow_ptr_leaks. | |
3805 | */ | |
3806 | static int adjust_ptr_min_max_vals(struct bpf_verifier_env *env, | |
3807 | struct bpf_insn *insn, | |
3808 | const struct bpf_reg_state *ptr_reg, | |
3809 | const struct bpf_reg_state *off_reg) | |
969bf05e | 3810 | { |
f4d7e40a AS |
3811 | struct bpf_verifier_state *vstate = env->cur_state; |
3812 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
3813 | struct bpf_reg_state *regs = state->regs, *dst_reg; | |
f1174f77 | 3814 | bool known = tnum_is_const(off_reg->var_off); |
b03c9f9f EC |
3815 | s64 smin_val = off_reg->smin_value, smax_val = off_reg->smax_value, |
3816 | smin_ptr = ptr_reg->smin_value, smax_ptr = ptr_reg->smax_value; | |
3817 | u64 umin_val = off_reg->umin_value, umax_val = off_reg->umax_value, | |
3818 | umin_ptr = ptr_reg->umin_value, umax_ptr = ptr_reg->umax_value; | |
9d7eceed | 3819 | u32 dst = insn->dst_reg, src = insn->src_reg; |
969bf05e | 3820 | u8 opcode = BPF_OP(insn->code); |
979d63d5 | 3821 | int ret; |
969bf05e | 3822 | |
f1174f77 | 3823 | dst_reg = ®s[dst]; |
969bf05e | 3824 | |
6f16101e DB |
3825 | if ((known && (smin_val != smax_val || umin_val != umax_val)) || |
3826 | smin_val > smax_val || umin_val > umax_val) { | |
3827 | /* Taint dst register if offset had invalid bounds derived from | |
3828 | * e.g. dead branches. | |
3829 | */ | |
3830 | __mark_reg_unknown(dst_reg); | |
3831 | return 0; | |
f1174f77 EC |
3832 | } |
3833 | ||
3834 | if (BPF_CLASS(insn->code) != BPF_ALU64) { | |
3835 | /* 32-bit ALU ops on pointers produce (meaningless) scalars */ | |
82abbf8d AS |
3836 | verbose(env, |
3837 | "R%d 32-bit pointer arithmetic prohibited\n", | |
3838 | dst); | |
f1174f77 | 3839 | return -EACCES; |
969bf05e AS |
3840 | } |
3841 | ||
aad2eeaf JS |
3842 | switch (ptr_reg->type) { |
3843 | case PTR_TO_MAP_VALUE_OR_NULL: | |
3844 | verbose(env, "R%d pointer arithmetic on %s prohibited, null-check it first\n", | |
3845 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 3846 | return -EACCES; |
aad2eeaf JS |
3847 | case CONST_PTR_TO_MAP: |
3848 | case PTR_TO_PACKET_END: | |
c64b7983 JS |
3849 | case PTR_TO_SOCKET: |
3850 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
3851 | case PTR_TO_SOCK_COMMON: |
3852 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
3853 | case PTR_TO_TCP_SOCK: |
3854 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 3855 | case PTR_TO_XDP_SOCK: |
aad2eeaf JS |
3856 | verbose(env, "R%d pointer arithmetic on %s prohibited\n", |
3857 | dst, reg_type_str[ptr_reg->type]); | |
f1174f77 | 3858 | return -EACCES; |
9d7eceed DB |
3859 | case PTR_TO_MAP_VALUE: |
3860 | if (!env->allow_ptr_leaks && !known && (smin_val < 0) != (smax_val < 0)) { | |
3861 | verbose(env, "R%d has unknown scalar with mixed signed bounds, pointer arithmetic with it prohibited for !root\n", | |
3862 | off_reg == dst_reg ? dst : src); | |
3863 | return -EACCES; | |
3864 | } | |
3865 | /* fall-through */ | |
aad2eeaf JS |
3866 | default: |
3867 | break; | |
f1174f77 EC |
3868 | } |
3869 | ||
3870 | /* In case of 'scalar += pointer', dst_reg inherits pointer type and id. | |
3871 | * The id may be overwritten later if we create a new variable offset. | |
969bf05e | 3872 | */ |
f1174f77 EC |
3873 | dst_reg->type = ptr_reg->type; |
3874 | dst_reg->id = ptr_reg->id; | |
969bf05e | 3875 | |
bb7f0f98 AS |
3876 | if (!check_reg_sane_offset(env, off_reg, ptr_reg->type) || |
3877 | !check_reg_sane_offset(env, ptr_reg, ptr_reg->type)) | |
3878 | return -EINVAL; | |
3879 | ||
f1174f77 EC |
3880 | switch (opcode) { |
3881 | case BPF_ADD: | |
979d63d5 DB |
3882 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
3883 | if (ret < 0) { | |
3884 | verbose(env, "R%d tried to add from different maps or paths\n", dst); | |
3885 | return ret; | |
3886 | } | |
f1174f77 EC |
3887 | /* We can take a fixed offset as long as it doesn't overflow |
3888 | * the s32 'off' field | |
969bf05e | 3889 | */ |
b03c9f9f EC |
3890 | if (known && (ptr_reg->off + smin_val == |
3891 | (s64)(s32)(ptr_reg->off + smin_val))) { | |
f1174f77 | 3892 | /* pointer += K. Accumulate it into fixed offset */ |
b03c9f9f EC |
3893 | dst_reg->smin_value = smin_ptr; |
3894 | dst_reg->smax_value = smax_ptr; | |
3895 | dst_reg->umin_value = umin_ptr; | |
3896 | dst_reg->umax_value = umax_ptr; | |
f1174f77 | 3897 | dst_reg->var_off = ptr_reg->var_off; |
b03c9f9f | 3898 | dst_reg->off = ptr_reg->off + smin_val; |
0962590e | 3899 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
3900 | break; |
3901 | } | |
f1174f77 EC |
3902 | /* A new variable offset is created. Note that off_reg->off |
3903 | * == 0, since it's a scalar. | |
3904 | * dst_reg gets the pointer type and since some positive | |
3905 | * integer value was added to the pointer, give it a new 'id' | |
3906 | * if it's a PTR_TO_PACKET. | |
3907 | * this creates a new 'base' pointer, off_reg (variable) gets | |
3908 | * added into the variable offset, and we copy the fixed offset | |
3909 | * from ptr_reg. | |
969bf05e | 3910 | */ |
b03c9f9f EC |
3911 | if (signed_add_overflows(smin_ptr, smin_val) || |
3912 | signed_add_overflows(smax_ptr, smax_val)) { | |
3913 | dst_reg->smin_value = S64_MIN; | |
3914 | dst_reg->smax_value = S64_MAX; | |
3915 | } else { | |
3916 | dst_reg->smin_value = smin_ptr + smin_val; | |
3917 | dst_reg->smax_value = smax_ptr + smax_val; | |
3918 | } | |
3919 | if (umin_ptr + umin_val < umin_ptr || | |
3920 | umax_ptr + umax_val < umax_ptr) { | |
3921 | dst_reg->umin_value = 0; | |
3922 | dst_reg->umax_value = U64_MAX; | |
3923 | } else { | |
3924 | dst_reg->umin_value = umin_ptr + umin_val; | |
3925 | dst_reg->umax_value = umax_ptr + umax_val; | |
3926 | } | |
f1174f77 EC |
3927 | dst_reg->var_off = tnum_add(ptr_reg->var_off, off_reg->var_off); |
3928 | dst_reg->off = ptr_reg->off; | |
0962590e | 3929 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 3930 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
3931 | dst_reg->id = ++env->id_gen; |
3932 | /* something was added to pkt_ptr, set range to zero */ | |
0962590e | 3933 | dst_reg->raw = 0; |
f1174f77 EC |
3934 | } |
3935 | break; | |
3936 | case BPF_SUB: | |
979d63d5 DB |
3937 | ret = sanitize_ptr_alu(env, insn, ptr_reg, dst_reg, smin_val < 0); |
3938 | if (ret < 0) { | |
3939 | verbose(env, "R%d tried to sub from different maps or paths\n", dst); | |
3940 | return ret; | |
3941 | } | |
f1174f77 EC |
3942 | if (dst_reg == off_reg) { |
3943 | /* scalar -= pointer. Creates an unknown scalar */ | |
82abbf8d AS |
3944 | verbose(env, "R%d tried to subtract pointer from scalar\n", |
3945 | dst); | |
f1174f77 EC |
3946 | return -EACCES; |
3947 | } | |
3948 | /* We don't allow subtraction from FP, because (according to | |
3949 | * test_verifier.c test "invalid fp arithmetic", JITs might not | |
3950 | * be able to deal with it. | |
969bf05e | 3951 | */ |
f1174f77 | 3952 | if (ptr_reg->type == PTR_TO_STACK) { |
82abbf8d AS |
3953 | verbose(env, "R%d subtraction from stack pointer prohibited\n", |
3954 | dst); | |
f1174f77 EC |
3955 | return -EACCES; |
3956 | } | |
b03c9f9f EC |
3957 | if (known && (ptr_reg->off - smin_val == |
3958 | (s64)(s32)(ptr_reg->off - smin_val))) { | |
f1174f77 | 3959 | /* pointer -= K. Subtract it from fixed offset */ |
b03c9f9f EC |
3960 | dst_reg->smin_value = smin_ptr; |
3961 | dst_reg->smax_value = smax_ptr; | |
3962 | dst_reg->umin_value = umin_ptr; | |
3963 | dst_reg->umax_value = umax_ptr; | |
f1174f77 EC |
3964 | dst_reg->var_off = ptr_reg->var_off; |
3965 | dst_reg->id = ptr_reg->id; | |
b03c9f9f | 3966 | dst_reg->off = ptr_reg->off - smin_val; |
0962590e | 3967 | dst_reg->raw = ptr_reg->raw; |
f1174f77 EC |
3968 | break; |
3969 | } | |
f1174f77 EC |
3970 | /* A new variable offset is created. If the subtrahend is known |
3971 | * nonnegative, then any reg->range we had before is still good. | |
969bf05e | 3972 | */ |
b03c9f9f EC |
3973 | if (signed_sub_overflows(smin_ptr, smax_val) || |
3974 | signed_sub_overflows(smax_ptr, smin_val)) { | |
3975 | /* Overflow possible, we know nothing */ | |
3976 | dst_reg->smin_value = S64_MIN; | |
3977 | dst_reg->smax_value = S64_MAX; | |
3978 | } else { | |
3979 | dst_reg->smin_value = smin_ptr - smax_val; | |
3980 | dst_reg->smax_value = smax_ptr - smin_val; | |
3981 | } | |
3982 | if (umin_ptr < umax_val) { | |
3983 | /* Overflow possible, we know nothing */ | |
3984 | dst_reg->umin_value = 0; | |
3985 | dst_reg->umax_value = U64_MAX; | |
3986 | } else { | |
3987 | /* Cannot overflow (as long as bounds are consistent) */ | |
3988 | dst_reg->umin_value = umin_ptr - umax_val; | |
3989 | dst_reg->umax_value = umax_ptr - umin_val; | |
3990 | } | |
f1174f77 EC |
3991 | dst_reg->var_off = tnum_sub(ptr_reg->var_off, off_reg->var_off); |
3992 | dst_reg->off = ptr_reg->off; | |
0962590e | 3993 | dst_reg->raw = ptr_reg->raw; |
de8f3a83 | 3994 | if (reg_is_pkt_pointer(ptr_reg)) { |
f1174f77 EC |
3995 | dst_reg->id = ++env->id_gen; |
3996 | /* something was added to pkt_ptr, set range to zero */ | |
b03c9f9f | 3997 | if (smin_val < 0) |
0962590e | 3998 | dst_reg->raw = 0; |
43188702 | 3999 | } |
f1174f77 EC |
4000 | break; |
4001 | case BPF_AND: | |
4002 | case BPF_OR: | |
4003 | case BPF_XOR: | |
82abbf8d AS |
4004 | /* bitwise ops on pointers are troublesome, prohibit. */ |
4005 | verbose(env, "R%d bitwise operator %s on pointer prohibited\n", | |
4006 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 EC |
4007 | return -EACCES; |
4008 | default: | |
4009 | /* other operators (e.g. MUL,LSH) produce non-pointer results */ | |
82abbf8d AS |
4010 | verbose(env, "R%d pointer arithmetic with %s operator prohibited\n", |
4011 | dst, bpf_alu_string[opcode >> 4]); | |
f1174f77 | 4012 | return -EACCES; |
43188702 JF |
4013 | } |
4014 | ||
bb7f0f98 AS |
4015 | if (!check_reg_sane_offset(env, dst_reg, ptr_reg->type)) |
4016 | return -EINVAL; | |
4017 | ||
b03c9f9f EC |
4018 | __update_reg_bounds(dst_reg); |
4019 | __reg_deduce_bounds(dst_reg); | |
4020 | __reg_bound_offset(dst_reg); | |
0d6303db DB |
4021 | |
4022 | /* For unprivileged we require that resulting offset must be in bounds | |
4023 | * in order to be able to sanitize access later on. | |
4024 | */ | |
e4298d25 DB |
4025 | if (!env->allow_ptr_leaks) { |
4026 | if (dst_reg->type == PTR_TO_MAP_VALUE && | |
4027 | check_map_access(env, dst, dst_reg->off, 1, false)) { | |
4028 | verbose(env, "R%d pointer arithmetic of map value goes out of range, " | |
4029 | "prohibited for !root\n", dst); | |
4030 | return -EACCES; | |
4031 | } else if (dst_reg->type == PTR_TO_STACK && | |
4032 | check_stack_access(env, dst_reg, dst_reg->off + | |
4033 | dst_reg->var_off.value, 1)) { | |
4034 | verbose(env, "R%d stack pointer arithmetic goes out of range, " | |
4035 | "prohibited for !root\n", dst); | |
4036 | return -EACCES; | |
4037 | } | |
0d6303db DB |
4038 | } |
4039 | ||
43188702 JF |
4040 | return 0; |
4041 | } | |
4042 | ||
468f6eaf JH |
4043 | /* WARNING: This function does calculations on 64-bit values, but the actual |
4044 | * execution may occur on 32-bit values. Therefore, things like bitshifts | |
4045 | * need extra checks in the 32-bit case. | |
4046 | */ | |
f1174f77 EC |
4047 | static int adjust_scalar_min_max_vals(struct bpf_verifier_env *env, |
4048 | struct bpf_insn *insn, | |
4049 | struct bpf_reg_state *dst_reg, | |
4050 | struct bpf_reg_state src_reg) | |
969bf05e | 4051 | { |
638f5b90 | 4052 | struct bpf_reg_state *regs = cur_regs(env); |
48461135 | 4053 | u8 opcode = BPF_OP(insn->code); |
f1174f77 | 4054 | bool src_known, dst_known; |
b03c9f9f EC |
4055 | s64 smin_val, smax_val; |
4056 | u64 umin_val, umax_val; | |
468f6eaf | 4057 | u64 insn_bitness = (BPF_CLASS(insn->code) == BPF_ALU64) ? 64 : 32; |
d3bd7413 DB |
4058 | u32 dst = insn->dst_reg; |
4059 | int ret; | |
48461135 | 4060 | |
b799207e JH |
4061 | if (insn_bitness == 32) { |
4062 | /* Relevant for 32-bit RSH: Information can propagate towards | |
4063 | * LSB, so it isn't sufficient to only truncate the output to | |
4064 | * 32 bits. | |
4065 | */ | |
4066 | coerce_reg_to_size(dst_reg, 4); | |
4067 | coerce_reg_to_size(&src_reg, 4); | |
4068 | } | |
4069 | ||
b03c9f9f EC |
4070 | smin_val = src_reg.smin_value; |
4071 | smax_val = src_reg.smax_value; | |
4072 | umin_val = src_reg.umin_value; | |
4073 | umax_val = src_reg.umax_value; | |
f1174f77 EC |
4074 | src_known = tnum_is_const(src_reg.var_off); |
4075 | dst_known = tnum_is_const(dst_reg->var_off); | |
f23cc643 | 4076 | |
6f16101e DB |
4077 | if ((src_known && (smin_val != smax_val || umin_val != umax_val)) || |
4078 | smin_val > smax_val || umin_val > umax_val) { | |
4079 | /* Taint dst register if offset had invalid bounds derived from | |
4080 | * e.g. dead branches. | |
4081 | */ | |
4082 | __mark_reg_unknown(dst_reg); | |
4083 | return 0; | |
4084 | } | |
4085 | ||
bb7f0f98 AS |
4086 | if (!src_known && |
4087 | opcode != BPF_ADD && opcode != BPF_SUB && opcode != BPF_AND) { | |
4088 | __mark_reg_unknown(dst_reg); | |
4089 | return 0; | |
4090 | } | |
4091 | ||
48461135 JB |
4092 | switch (opcode) { |
4093 | case BPF_ADD: | |
d3bd7413 DB |
4094 | ret = sanitize_val_alu(env, insn); |
4095 | if (ret < 0) { | |
4096 | verbose(env, "R%d tried to add from different pointers or scalars\n", dst); | |
4097 | return ret; | |
4098 | } | |
b03c9f9f EC |
4099 | if (signed_add_overflows(dst_reg->smin_value, smin_val) || |
4100 | signed_add_overflows(dst_reg->smax_value, smax_val)) { | |
4101 | dst_reg->smin_value = S64_MIN; | |
4102 | dst_reg->smax_value = S64_MAX; | |
4103 | } else { | |
4104 | dst_reg->smin_value += smin_val; | |
4105 | dst_reg->smax_value += smax_val; | |
4106 | } | |
4107 | if (dst_reg->umin_value + umin_val < umin_val || | |
4108 | dst_reg->umax_value + umax_val < umax_val) { | |
4109 | dst_reg->umin_value = 0; | |
4110 | dst_reg->umax_value = U64_MAX; | |
4111 | } else { | |
4112 | dst_reg->umin_value += umin_val; | |
4113 | dst_reg->umax_value += umax_val; | |
4114 | } | |
f1174f77 | 4115 | dst_reg->var_off = tnum_add(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
4116 | break; |
4117 | case BPF_SUB: | |
d3bd7413 DB |
4118 | ret = sanitize_val_alu(env, insn); |
4119 | if (ret < 0) { | |
4120 | verbose(env, "R%d tried to sub from different pointers or scalars\n", dst); | |
4121 | return ret; | |
4122 | } | |
b03c9f9f EC |
4123 | if (signed_sub_overflows(dst_reg->smin_value, smax_val) || |
4124 | signed_sub_overflows(dst_reg->smax_value, smin_val)) { | |
4125 | /* Overflow possible, we know nothing */ | |
4126 | dst_reg->smin_value = S64_MIN; | |
4127 | dst_reg->smax_value = S64_MAX; | |
4128 | } else { | |
4129 | dst_reg->smin_value -= smax_val; | |
4130 | dst_reg->smax_value -= smin_val; | |
4131 | } | |
4132 | if (dst_reg->umin_value < umax_val) { | |
4133 | /* Overflow possible, we know nothing */ | |
4134 | dst_reg->umin_value = 0; | |
4135 | dst_reg->umax_value = U64_MAX; | |
4136 | } else { | |
4137 | /* Cannot overflow (as long as bounds are consistent) */ | |
4138 | dst_reg->umin_value -= umax_val; | |
4139 | dst_reg->umax_value -= umin_val; | |
4140 | } | |
f1174f77 | 4141 | dst_reg->var_off = tnum_sub(dst_reg->var_off, src_reg.var_off); |
48461135 JB |
4142 | break; |
4143 | case BPF_MUL: | |
b03c9f9f EC |
4144 | dst_reg->var_off = tnum_mul(dst_reg->var_off, src_reg.var_off); |
4145 | if (smin_val < 0 || dst_reg->smin_value < 0) { | |
f1174f77 | 4146 | /* Ain't nobody got time to multiply that sign */ |
b03c9f9f EC |
4147 | __mark_reg_unbounded(dst_reg); |
4148 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
4149 | break; |
4150 | } | |
b03c9f9f EC |
4151 | /* Both values are positive, so we can work with unsigned and |
4152 | * copy the result to signed (unless it exceeds S64_MAX). | |
f1174f77 | 4153 | */ |
b03c9f9f EC |
4154 | if (umax_val > U32_MAX || dst_reg->umax_value > U32_MAX) { |
4155 | /* Potential overflow, we know nothing */ | |
4156 | __mark_reg_unbounded(dst_reg); | |
4157 | /* (except what we can learn from the var_off) */ | |
4158 | __update_reg_bounds(dst_reg); | |
4159 | break; | |
4160 | } | |
4161 | dst_reg->umin_value *= umin_val; | |
4162 | dst_reg->umax_value *= umax_val; | |
4163 | if (dst_reg->umax_value > S64_MAX) { | |
4164 | /* Overflow possible, we know nothing */ | |
4165 | dst_reg->smin_value = S64_MIN; | |
4166 | dst_reg->smax_value = S64_MAX; | |
4167 | } else { | |
4168 | dst_reg->smin_value = dst_reg->umin_value; | |
4169 | dst_reg->smax_value = dst_reg->umax_value; | |
4170 | } | |
48461135 JB |
4171 | break; |
4172 | case BPF_AND: | |
f1174f77 | 4173 | if (src_known && dst_known) { |
b03c9f9f EC |
4174 | __mark_reg_known(dst_reg, dst_reg->var_off.value & |
4175 | src_reg.var_off.value); | |
f1174f77 EC |
4176 | break; |
4177 | } | |
b03c9f9f EC |
4178 | /* We get our minimum from the var_off, since that's inherently |
4179 | * bitwise. Our maximum is the minimum of the operands' maxima. | |
f23cc643 | 4180 | */ |
f1174f77 | 4181 | dst_reg->var_off = tnum_and(dst_reg->var_off, src_reg.var_off); |
b03c9f9f EC |
4182 | dst_reg->umin_value = dst_reg->var_off.value; |
4183 | dst_reg->umax_value = min(dst_reg->umax_value, umax_val); | |
4184 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
4185 | /* Lose signed bounds when ANDing negative numbers, | |
4186 | * ain't nobody got time for that. | |
4187 | */ | |
4188 | dst_reg->smin_value = S64_MIN; | |
4189 | dst_reg->smax_value = S64_MAX; | |
4190 | } else { | |
4191 | /* ANDing two positives gives a positive, so safe to | |
4192 | * cast result into s64. | |
4193 | */ | |
4194 | dst_reg->smin_value = dst_reg->umin_value; | |
4195 | dst_reg->smax_value = dst_reg->umax_value; | |
4196 | } | |
4197 | /* We may learn something more from the var_off */ | |
4198 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
4199 | break; |
4200 | case BPF_OR: | |
4201 | if (src_known && dst_known) { | |
b03c9f9f EC |
4202 | __mark_reg_known(dst_reg, dst_reg->var_off.value | |
4203 | src_reg.var_off.value); | |
f1174f77 EC |
4204 | break; |
4205 | } | |
b03c9f9f EC |
4206 | /* We get our maximum from the var_off, and our minimum is the |
4207 | * maximum of the operands' minima | |
f1174f77 EC |
4208 | */ |
4209 | dst_reg->var_off = tnum_or(dst_reg->var_off, src_reg.var_off); | |
b03c9f9f EC |
4210 | dst_reg->umin_value = max(dst_reg->umin_value, umin_val); |
4211 | dst_reg->umax_value = dst_reg->var_off.value | | |
4212 | dst_reg->var_off.mask; | |
4213 | if (dst_reg->smin_value < 0 || smin_val < 0) { | |
4214 | /* Lose signed bounds when ORing negative numbers, | |
4215 | * ain't nobody got time for that. | |
4216 | */ | |
4217 | dst_reg->smin_value = S64_MIN; | |
4218 | dst_reg->smax_value = S64_MAX; | |
f1174f77 | 4219 | } else { |
b03c9f9f EC |
4220 | /* ORing two positives gives a positive, so safe to |
4221 | * cast result into s64. | |
4222 | */ | |
4223 | dst_reg->smin_value = dst_reg->umin_value; | |
4224 | dst_reg->smax_value = dst_reg->umax_value; | |
f1174f77 | 4225 | } |
b03c9f9f EC |
4226 | /* We may learn something more from the var_off */ |
4227 | __update_reg_bounds(dst_reg); | |
48461135 JB |
4228 | break; |
4229 | case BPF_LSH: | |
468f6eaf JH |
4230 | if (umax_val >= insn_bitness) { |
4231 | /* Shifts greater than 31 or 63 are undefined. | |
4232 | * This includes shifts by a negative number. | |
b03c9f9f | 4233 | */ |
61bd5218 | 4234 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
4235 | break; |
4236 | } | |
b03c9f9f EC |
4237 | /* We lose all sign bit information (except what we can pick |
4238 | * up from var_off) | |
48461135 | 4239 | */ |
b03c9f9f EC |
4240 | dst_reg->smin_value = S64_MIN; |
4241 | dst_reg->smax_value = S64_MAX; | |
4242 | /* If we might shift our top bit out, then we know nothing */ | |
4243 | if (dst_reg->umax_value > 1ULL << (63 - umax_val)) { | |
4244 | dst_reg->umin_value = 0; | |
4245 | dst_reg->umax_value = U64_MAX; | |
d1174416 | 4246 | } else { |
b03c9f9f EC |
4247 | dst_reg->umin_value <<= umin_val; |
4248 | dst_reg->umax_value <<= umax_val; | |
d1174416 | 4249 | } |
afbe1a5b | 4250 | dst_reg->var_off = tnum_lshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
4251 | /* We may learn something more from the var_off */ |
4252 | __update_reg_bounds(dst_reg); | |
48461135 JB |
4253 | break; |
4254 | case BPF_RSH: | |
468f6eaf JH |
4255 | if (umax_val >= insn_bitness) { |
4256 | /* Shifts greater than 31 or 63 are undefined. | |
4257 | * This includes shifts by a negative number. | |
b03c9f9f | 4258 | */ |
61bd5218 | 4259 | mark_reg_unknown(env, regs, insn->dst_reg); |
f1174f77 EC |
4260 | break; |
4261 | } | |
4374f256 EC |
4262 | /* BPF_RSH is an unsigned shift. If the value in dst_reg might |
4263 | * be negative, then either: | |
4264 | * 1) src_reg might be zero, so the sign bit of the result is | |
4265 | * unknown, so we lose our signed bounds | |
4266 | * 2) it's known negative, thus the unsigned bounds capture the | |
4267 | * signed bounds | |
4268 | * 3) the signed bounds cross zero, so they tell us nothing | |
4269 | * about the result | |
4270 | * If the value in dst_reg is known nonnegative, then again the | |
4271 | * unsigned bounts capture the signed bounds. | |
4272 | * Thus, in all cases it suffices to blow away our signed bounds | |
4273 | * and rely on inferring new ones from the unsigned bounds and | |
4274 | * var_off of the result. | |
4275 | */ | |
4276 | dst_reg->smin_value = S64_MIN; | |
4277 | dst_reg->smax_value = S64_MAX; | |
afbe1a5b | 4278 | dst_reg->var_off = tnum_rshift(dst_reg->var_off, umin_val); |
b03c9f9f EC |
4279 | dst_reg->umin_value >>= umax_val; |
4280 | dst_reg->umax_value >>= umin_val; | |
4281 | /* We may learn something more from the var_off */ | |
4282 | __update_reg_bounds(dst_reg); | |
48461135 | 4283 | break; |
9cbe1f5a YS |
4284 | case BPF_ARSH: |
4285 | if (umax_val >= insn_bitness) { | |
4286 | /* Shifts greater than 31 or 63 are undefined. | |
4287 | * This includes shifts by a negative number. | |
4288 | */ | |
4289 | mark_reg_unknown(env, regs, insn->dst_reg); | |
4290 | break; | |
4291 | } | |
4292 | ||
4293 | /* Upon reaching here, src_known is true and | |
4294 | * umax_val is equal to umin_val. | |
4295 | */ | |
4296 | dst_reg->smin_value >>= umin_val; | |
4297 | dst_reg->smax_value >>= umin_val; | |
4298 | dst_reg->var_off = tnum_arshift(dst_reg->var_off, umin_val); | |
4299 | ||
4300 | /* blow away the dst_reg umin_value/umax_value and rely on | |
4301 | * dst_reg var_off to refine the result. | |
4302 | */ | |
4303 | dst_reg->umin_value = 0; | |
4304 | dst_reg->umax_value = U64_MAX; | |
4305 | __update_reg_bounds(dst_reg); | |
4306 | break; | |
48461135 | 4307 | default: |
61bd5218 | 4308 | mark_reg_unknown(env, regs, insn->dst_reg); |
48461135 JB |
4309 | break; |
4310 | } | |
4311 | ||
468f6eaf JH |
4312 | if (BPF_CLASS(insn->code) != BPF_ALU64) { |
4313 | /* 32-bit ALU ops are (32,32)->32 */ | |
4314 | coerce_reg_to_size(dst_reg, 4); | |
468f6eaf JH |
4315 | } |
4316 | ||
b03c9f9f EC |
4317 | __reg_deduce_bounds(dst_reg); |
4318 | __reg_bound_offset(dst_reg); | |
f1174f77 EC |
4319 | return 0; |
4320 | } | |
4321 | ||
4322 | /* Handles ALU ops other than BPF_END, BPF_NEG and BPF_MOV: computes new min/max | |
4323 | * and var_off. | |
4324 | */ | |
4325 | static int adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
4326 | struct bpf_insn *insn) | |
4327 | { | |
f4d7e40a AS |
4328 | struct bpf_verifier_state *vstate = env->cur_state; |
4329 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; | |
4330 | struct bpf_reg_state *regs = state->regs, *dst_reg, *src_reg; | |
f1174f77 EC |
4331 | struct bpf_reg_state *ptr_reg = NULL, off_reg = {0}; |
4332 | u8 opcode = BPF_OP(insn->code); | |
f1174f77 EC |
4333 | |
4334 | dst_reg = ®s[insn->dst_reg]; | |
f1174f77 EC |
4335 | src_reg = NULL; |
4336 | if (dst_reg->type != SCALAR_VALUE) | |
4337 | ptr_reg = dst_reg; | |
4338 | if (BPF_SRC(insn->code) == BPF_X) { | |
4339 | src_reg = ®s[insn->src_reg]; | |
f1174f77 EC |
4340 | if (src_reg->type != SCALAR_VALUE) { |
4341 | if (dst_reg->type != SCALAR_VALUE) { | |
4342 | /* Combining two pointers by any ALU op yields | |
82abbf8d AS |
4343 | * an arbitrary scalar. Disallow all math except |
4344 | * pointer subtraction | |
f1174f77 | 4345 | */ |
dd066823 | 4346 | if (opcode == BPF_SUB && env->allow_ptr_leaks) { |
82abbf8d AS |
4347 | mark_reg_unknown(env, regs, insn->dst_reg); |
4348 | return 0; | |
f1174f77 | 4349 | } |
82abbf8d AS |
4350 | verbose(env, "R%d pointer %s pointer prohibited\n", |
4351 | insn->dst_reg, | |
4352 | bpf_alu_string[opcode >> 4]); | |
4353 | return -EACCES; | |
f1174f77 EC |
4354 | } else { |
4355 | /* scalar += pointer | |
4356 | * This is legal, but we have to reverse our | |
4357 | * src/dest handling in computing the range | |
4358 | */ | |
82abbf8d AS |
4359 | return adjust_ptr_min_max_vals(env, insn, |
4360 | src_reg, dst_reg); | |
f1174f77 EC |
4361 | } |
4362 | } else if (ptr_reg) { | |
4363 | /* pointer += scalar */ | |
82abbf8d AS |
4364 | return adjust_ptr_min_max_vals(env, insn, |
4365 | dst_reg, src_reg); | |
f1174f77 EC |
4366 | } |
4367 | } else { | |
4368 | /* Pretend the src is a reg with a known value, since we only | |
4369 | * need to be able to read from this state. | |
4370 | */ | |
4371 | off_reg.type = SCALAR_VALUE; | |
b03c9f9f | 4372 | __mark_reg_known(&off_reg, insn->imm); |
f1174f77 | 4373 | src_reg = &off_reg; |
82abbf8d AS |
4374 | if (ptr_reg) /* pointer += K */ |
4375 | return adjust_ptr_min_max_vals(env, insn, | |
4376 | ptr_reg, src_reg); | |
f1174f77 EC |
4377 | } |
4378 | ||
4379 | /* Got here implies adding two SCALAR_VALUEs */ | |
4380 | if (WARN_ON_ONCE(ptr_reg)) { | |
f4d7e40a | 4381 | print_verifier_state(env, state); |
61bd5218 | 4382 | verbose(env, "verifier internal error: unexpected ptr_reg\n"); |
f1174f77 EC |
4383 | return -EINVAL; |
4384 | } | |
4385 | if (WARN_ON(!src_reg)) { | |
f4d7e40a | 4386 | print_verifier_state(env, state); |
61bd5218 | 4387 | verbose(env, "verifier internal error: no src_reg\n"); |
f1174f77 EC |
4388 | return -EINVAL; |
4389 | } | |
4390 | return adjust_scalar_min_max_vals(env, insn, dst_reg, *src_reg); | |
48461135 JB |
4391 | } |
4392 | ||
17a52670 | 4393 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 4394 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 4395 | { |
638f5b90 | 4396 | struct bpf_reg_state *regs = cur_regs(env); |
17a52670 AS |
4397 | u8 opcode = BPF_OP(insn->code); |
4398 | int err; | |
4399 | ||
4400 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
4401 | if (opcode == BPF_NEG) { | |
4402 | if (BPF_SRC(insn->code) != 0 || | |
4403 | insn->src_reg != BPF_REG_0 || | |
4404 | insn->off != 0 || insn->imm != 0) { | |
61bd5218 | 4405 | verbose(env, "BPF_NEG uses reserved fields\n"); |
17a52670 AS |
4406 | return -EINVAL; |
4407 | } | |
4408 | } else { | |
4409 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
e67b8a68 EC |
4410 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64) || |
4411 | BPF_CLASS(insn->code) == BPF_ALU64) { | |
61bd5218 | 4412 | verbose(env, "BPF_END uses reserved fields\n"); |
17a52670 AS |
4413 | return -EINVAL; |
4414 | } | |
4415 | } | |
4416 | ||
4417 | /* check src operand */ | |
dc503a8a | 4418 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4419 | if (err) |
4420 | return err; | |
4421 | ||
1be7f75d | 4422 | if (is_pointer_value(env, insn->dst_reg)) { |
61bd5218 | 4423 | verbose(env, "R%d pointer arithmetic prohibited\n", |
1be7f75d AS |
4424 | insn->dst_reg); |
4425 | return -EACCES; | |
4426 | } | |
4427 | ||
17a52670 | 4428 | /* check dest operand */ |
dc503a8a | 4429 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
4430 | if (err) |
4431 | return err; | |
4432 | ||
4433 | } else if (opcode == BPF_MOV) { | |
4434 | ||
4435 | if (BPF_SRC(insn->code) == BPF_X) { | |
4436 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 4437 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
4438 | return -EINVAL; |
4439 | } | |
4440 | ||
4441 | /* check src operand */ | |
dc503a8a | 4442 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4443 | if (err) |
4444 | return err; | |
4445 | } else { | |
4446 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 4447 | verbose(env, "BPF_MOV uses reserved fields\n"); |
17a52670 AS |
4448 | return -EINVAL; |
4449 | } | |
4450 | } | |
4451 | ||
fbeb1603 AF |
4452 | /* check dest operand, mark as required later */ |
4453 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); | |
17a52670 AS |
4454 | if (err) |
4455 | return err; | |
4456 | ||
4457 | if (BPF_SRC(insn->code) == BPF_X) { | |
e434b8cd JW |
4458 | struct bpf_reg_state *src_reg = regs + insn->src_reg; |
4459 | struct bpf_reg_state *dst_reg = regs + insn->dst_reg; | |
4460 | ||
17a52670 AS |
4461 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
4462 | /* case: R1 = R2 | |
4463 | * copy register state to dest reg | |
4464 | */ | |
e434b8cd JW |
4465 | *dst_reg = *src_reg; |
4466 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 4467 | dst_reg->subreg_def = DEF_NOT_SUBREG; |
17a52670 | 4468 | } else { |
f1174f77 | 4469 | /* R1 = (u32) R2 */ |
1be7f75d | 4470 | if (is_pointer_value(env, insn->src_reg)) { |
61bd5218 JK |
4471 | verbose(env, |
4472 | "R%d partial copy of pointer\n", | |
1be7f75d AS |
4473 | insn->src_reg); |
4474 | return -EACCES; | |
e434b8cd JW |
4475 | } else if (src_reg->type == SCALAR_VALUE) { |
4476 | *dst_reg = *src_reg; | |
4477 | dst_reg->live |= REG_LIVE_WRITTEN; | |
5327ed3d | 4478 | dst_reg->subreg_def = env->insn_idx + 1; |
e434b8cd JW |
4479 | } else { |
4480 | mark_reg_unknown(env, regs, | |
4481 | insn->dst_reg); | |
1be7f75d | 4482 | } |
e434b8cd | 4483 | coerce_reg_to_size(dst_reg, 4); |
17a52670 AS |
4484 | } |
4485 | } else { | |
4486 | /* case: R = imm | |
4487 | * remember the value we stored into this reg | |
4488 | */ | |
fbeb1603 AF |
4489 | /* clear any state __mark_reg_known doesn't set */ |
4490 | mark_reg_unknown(env, regs, insn->dst_reg); | |
f1174f77 | 4491 | regs[insn->dst_reg].type = SCALAR_VALUE; |
95a762e2 JH |
4492 | if (BPF_CLASS(insn->code) == BPF_ALU64) { |
4493 | __mark_reg_known(regs + insn->dst_reg, | |
4494 | insn->imm); | |
4495 | } else { | |
4496 | __mark_reg_known(regs + insn->dst_reg, | |
4497 | (u32)insn->imm); | |
4498 | } | |
17a52670 AS |
4499 | } |
4500 | ||
4501 | } else if (opcode > BPF_END) { | |
61bd5218 | 4502 | verbose(env, "invalid BPF_ALU opcode %x\n", opcode); |
17a52670 AS |
4503 | return -EINVAL; |
4504 | ||
4505 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
4506 | ||
17a52670 AS |
4507 | if (BPF_SRC(insn->code) == BPF_X) { |
4508 | if (insn->imm != 0 || insn->off != 0) { | |
61bd5218 | 4509 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
4510 | return -EINVAL; |
4511 | } | |
4512 | /* check src1 operand */ | |
dc503a8a | 4513 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
4514 | if (err) |
4515 | return err; | |
4516 | } else { | |
4517 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
61bd5218 | 4518 | verbose(env, "BPF_ALU uses reserved fields\n"); |
17a52670 AS |
4519 | return -EINVAL; |
4520 | } | |
4521 | } | |
4522 | ||
4523 | /* check src2 operand */ | |
dc503a8a | 4524 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
4525 | if (err) |
4526 | return err; | |
4527 | ||
4528 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
4529 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
61bd5218 | 4530 | verbose(env, "div by zero\n"); |
17a52670 AS |
4531 | return -EINVAL; |
4532 | } | |
4533 | ||
229394e8 RV |
4534 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
4535 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
4536 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
4537 | ||
4538 | if (insn->imm < 0 || insn->imm >= size) { | |
61bd5218 | 4539 | verbose(env, "invalid shift %d\n", insn->imm); |
229394e8 RV |
4540 | return -EINVAL; |
4541 | } | |
4542 | } | |
4543 | ||
1a0dc1ac | 4544 | /* check dest operand */ |
dc503a8a | 4545 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
1a0dc1ac AS |
4546 | if (err) |
4547 | return err; | |
4548 | ||
f1174f77 | 4549 | return adjust_reg_min_max_vals(env, insn); |
17a52670 AS |
4550 | } |
4551 | ||
4552 | return 0; | |
4553 | } | |
4554 | ||
c6a9efa1 PC |
4555 | static void __find_good_pkt_pointers(struct bpf_func_state *state, |
4556 | struct bpf_reg_state *dst_reg, | |
4557 | enum bpf_reg_type type, u16 new_range) | |
4558 | { | |
4559 | struct bpf_reg_state *reg; | |
4560 | int i; | |
4561 | ||
4562 | for (i = 0; i < MAX_BPF_REG; i++) { | |
4563 | reg = &state->regs[i]; | |
4564 | if (reg->type == type && reg->id == dst_reg->id) | |
4565 | /* keep the maximum range already checked */ | |
4566 | reg->range = max(reg->range, new_range); | |
4567 | } | |
4568 | ||
4569 | bpf_for_each_spilled_reg(i, state, reg) { | |
4570 | if (!reg) | |
4571 | continue; | |
4572 | if (reg->type == type && reg->id == dst_reg->id) | |
4573 | reg->range = max(reg->range, new_range); | |
4574 | } | |
4575 | } | |
4576 | ||
f4d7e40a | 4577 | static void find_good_pkt_pointers(struct bpf_verifier_state *vstate, |
de8f3a83 | 4578 | struct bpf_reg_state *dst_reg, |
f8ddadc4 | 4579 | enum bpf_reg_type type, |
fb2a311a | 4580 | bool range_right_open) |
969bf05e | 4581 | { |
fb2a311a | 4582 | u16 new_range; |
c6a9efa1 | 4583 | int i; |
2d2be8ca | 4584 | |
fb2a311a DB |
4585 | if (dst_reg->off < 0 || |
4586 | (dst_reg->off == 0 && range_right_open)) | |
f1174f77 EC |
4587 | /* This doesn't give us any range */ |
4588 | return; | |
4589 | ||
b03c9f9f EC |
4590 | if (dst_reg->umax_value > MAX_PACKET_OFF || |
4591 | dst_reg->umax_value + dst_reg->off > MAX_PACKET_OFF) | |
f1174f77 EC |
4592 | /* Risk of overflow. For instance, ptr + (1<<63) may be less |
4593 | * than pkt_end, but that's because it's also less than pkt. | |
4594 | */ | |
4595 | return; | |
4596 | ||
fb2a311a DB |
4597 | new_range = dst_reg->off; |
4598 | if (range_right_open) | |
4599 | new_range--; | |
4600 | ||
4601 | /* Examples for register markings: | |
2d2be8ca | 4602 | * |
fb2a311a | 4603 | * pkt_data in dst register: |
2d2be8ca DB |
4604 | * |
4605 | * r2 = r3; | |
4606 | * r2 += 8; | |
4607 | * if (r2 > pkt_end) goto <handle exception> | |
4608 | * <access okay> | |
4609 | * | |
b4e432f1 DB |
4610 | * r2 = r3; |
4611 | * r2 += 8; | |
4612 | * if (r2 < pkt_end) goto <access okay> | |
4613 | * <handle exception> | |
4614 | * | |
2d2be8ca DB |
4615 | * Where: |
4616 | * r2 == dst_reg, pkt_end == src_reg | |
4617 | * r2=pkt(id=n,off=8,r=0) | |
4618 | * r3=pkt(id=n,off=0,r=0) | |
4619 | * | |
fb2a311a | 4620 | * pkt_data in src register: |
2d2be8ca DB |
4621 | * |
4622 | * r2 = r3; | |
4623 | * r2 += 8; | |
4624 | * if (pkt_end >= r2) goto <access okay> | |
4625 | * <handle exception> | |
4626 | * | |
b4e432f1 DB |
4627 | * r2 = r3; |
4628 | * r2 += 8; | |
4629 | * if (pkt_end <= r2) goto <handle exception> | |
4630 | * <access okay> | |
4631 | * | |
2d2be8ca DB |
4632 | * Where: |
4633 | * pkt_end == dst_reg, r2 == src_reg | |
4634 | * r2=pkt(id=n,off=8,r=0) | |
4635 | * r3=pkt(id=n,off=0,r=0) | |
4636 | * | |
4637 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
fb2a311a DB |
4638 | * or r3=pkt(id=n,off=0,r=8-1), so that range of bytes [r3, r3 + 8) |
4639 | * and [r3, r3 + 8-1) respectively is safe to access depending on | |
4640 | * the check. | |
969bf05e | 4641 | */ |
2d2be8ca | 4642 | |
f1174f77 EC |
4643 | /* If our ids match, then we must have the same max_value. And we |
4644 | * don't care about the other reg's fixed offset, since if it's too big | |
4645 | * the range won't allow anything. | |
4646 | * dst_reg->off is known < MAX_PACKET_OFF, therefore it fits in a u16. | |
4647 | */ | |
c6a9efa1 PC |
4648 | for (i = 0; i <= vstate->curframe; i++) |
4649 | __find_good_pkt_pointers(vstate->frame[i], dst_reg, type, | |
4650 | new_range); | |
969bf05e AS |
4651 | } |
4652 | ||
4f7b3e82 AS |
4653 | /* compute branch direction of the expression "if (reg opcode val) goto target;" |
4654 | * and return: | |
4655 | * 1 - branch will be taken and "goto target" will be executed | |
4656 | * 0 - branch will not be taken and fall-through to next insn | |
4657 | * -1 - unknown. Example: "if (reg < 5)" is unknown when register value range [0,10] | |
4658 | */ | |
092ed096 JW |
4659 | static int is_branch_taken(struct bpf_reg_state *reg, u64 val, u8 opcode, |
4660 | bool is_jmp32) | |
4f7b3e82 | 4661 | { |
092ed096 | 4662 | struct bpf_reg_state reg_lo; |
a72dafaf JW |
4663 | s64 sval; |
4664 | ||
4f7b3e82 AS |
4665 | if (__is_pointer_value(false, reg)) |
4666 | return -1; | |
4667 | ||
092ed096 JW |
4668 | if (is_jmp32) { |
4669 | reg_lo = *reg; | |
4670 | reg = ®_lo; | |
4671 | /* For JMP32, only low 32 bits are compared, coerce_reg_to_size | |
4672 | * could truncate high bits and update umin/umax according to | |
4673 | * information of low bits. | |
4674 | */ | |
4675 | coerce_reg_to_size(reg, 4); | |
4676 | /* smin/smax need special handling. For example, after coerce, | |
4677 | * if smin_value is 0x00000000ffffffffLL, the value is -1 when | |
4678 | * used as operand to JMP32. It is a negative number from s32's | |
4679 | * point of view, while it is a positive number when seen as | |
4680 | * s64. The smin/smax are kept as s64, therefore, when used with | |
4681 | * JMP32, they need to be transformed into s32, then sign | |
4682 | * extended back to s64. | |
4683 | * | |
4684 | * Also, smin/smax were copied from umin/umax. If umin/umax has | |
4685 | * different sign bit, then min/max relationship doesn't | |
4686 | * maintain after casting into s32, for this case, set smin/smax | |
4687 | * to safest range. | |
4688 | */ | |
4689 | if ((reg->umax_value ^ reg->umin_value) & | |
4690 | (1ULL << 31)) { | |
4691 | reg->smin_value = S32_MIN; | |
4692 | reg->smax_value = S32_MAX; | |
4693 | } | |
4694 | reg->smin_value = (s64)(s32)reg->smin_value; | |
4695 | reg->smax_value = (s64)(s32)reg->smax_value; | |
4696 | ||
4697 | val = (u32)val; | |
4698 | sval = (s64)(s32)val; | |
4699 | } else { | |
4700 | sval = (s64)val; | |
4701 | } | |
a72dafaf | 4702 | |
4f7b3e82 AS |
4703 | switch (opcode) { |
4704 | case BPF_JEQ: | |
4705 | if (tnum_is_const(reg->var_off)) | |
4706 | return !!tnum_equals_const(reg->var_off, val); | |
4707 | break; | |
4708 | case BPF_JNE: | |
4709 | if (tnum_is_const(reg->var_off)) | |
4710 | return !tnum_equals_const(reg->var_off, val); | |
4711 | break; | |
960ea056 JK |
4712 | case BPF_JSET: |
4713 | if ((~reg->var_off.mask & reg->var_off.value) & val) | |
4714 | return 1; | |
4715 | if (!((reg->var_off.mask | reg->var_off.value) & val)) | |
4716 | return 0; | |
4717 | break; | |
4f7b3e82 AS |
4718 | case BPF_JGT: |
4719 | if (reg->umin_value > val) | |
4720 | return 1; | |
4721 | else if (reg->umax_value <= val) | |
4722 | return 0; | |
4723 | break; | |
4724 | case BPF_JSGT: | |
a72dafaf | 4725 | if (reg->smin_value > sval) |
4f7b3e82 | 4726 | return 1; |
a72dafaf | 4727 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
4728 | return 0; |
4729 | break; | |
4730 | case BPF_JLT: | |
4731 | if (reg->umax_value < val) | |
4732 | return 1; | |
4733 | else if (reg->umin_value >= val) | |
4734 | return 0; | |
4735 | break; | |
4736 | case BPF_JSLT: | |
a72dafaf | 4737 | if (reg->smax_value < sval) |
4f7b3e82 | 4738 | return 1; |
a72dafaf | 4739 | else if (reg->smin_value >= sval) |
4f7b3e82 AS |
4740 | return 0; |
4741 | break; | |
4742 | case BPF_JGE: | |
4743 | if (reg->umin_value >= val) | |
4744 | return 1; | |
4745 | else if (reg->umax_value < val) | |
4746 | return 0; | |
4747 | break; | |
4748 | case BPF_JSGE: | |
a72dafaf | 4749 | if (reg->smin_value >= sval) |
4f7b3e82 | 4750 | return 1; |
a72dafaf | 4751 | else if (reg->smax_value < sval) |
4f7b3e82 AS |
4752 | return 0; |
4753 | break; | |
4754 | case BPF_JLE: | |
4755 | if (reg->umax_value <= val) | |
4756 | return 1; | |
4757 | else if (reg->umin_value > val) | |
4758 | return 0; | |
4759 | break; | |
4760 | case BPF_JSLE: | |
a72dafaf | 4761 | if (reg->smax_value <= sval) |
4f7b3e82 | 4762 | return 1; |
a72dafaf | 4763 | else if (reg->smin_value > sval) |
4f7b3e82 AS |
4764 | return 0; |
4765 | break; | |
4766 | } | |
4767 | ||
4768 | return -1; | |
4769 | } | |
4770 | ||
092ed096 JW |
4771 | /* Generate min value of the high 32-bit from TNUM info. */ |
4772 | static u64 gen_hi_min(struct tnum var) | |
4773 | { | |
4774 | return var.value & ~0xffffffffULL; | |
4775 | } | |
4776 | ||
4777 | /* Generate max value of the high 32-bit from TNUM info. */ | |
4778 | static u64 gen_hi_max(struct tnum var) | |
4779 | { | |
4780 | return (var.value | var.mask) & ~0xffffffffULL; | |
4781 | } | |
4782 | ||
4783 | /* Return true if VAL is compared with a s64 sign extended from s32, and they | |
4784 | * are with the same signedness. | |
4785 | */ | |
4786 | static bool cmp_val_with_extended_s64(s64 sval, struct bpf_reg_state *reg) | |
4787 | { | |
4788 | return ((s32)sval >= 0 && | |
4789 | reg->smin_value >= 0 && reg->smax_value <= S32_MAX) || | |
4790 | ((s32)sval < 0 && | |
4791 | reg->smax_value <= 0 && reg->smin_value >= S32_MIN); | |
4792 | } | |
4793 | ||
48461135 JB |
4794 | /* Adjusts the register min/max values in the case that the dst_reg is the |
4795 | * variable register that we are working on, and src_reg is a constant or we're | |
4796 | * simply doing a BPF_K check. | |
f1174f77 | 4797 | * In JEQ/JNE cases we also adjust the var_off values. |
48461135 JB |
4798 | */ |
4799 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
4800 | struct bpf_reg_state *false_reg, u64 val, | |
092ed096 | 4801 | u8 opcode, bool is_jmp32) |
48461135 | 4802 | { |
a72dafaf JW |
4803 | s64 sval; |
4804 | ||
f1174f77 EC |
4805 | /* If the dst_reg is a pointer, we can't learn anything about its |
4806 | * variable offset from the compare (unless src_reg were a pointer into | |
4807 | * the same object, but we don't bother with that. | |
4808 | * Since false_reg and true_reg have the same type by construction, we | |
4809 | * only need to check one of them for pointerness. | |
4810 | */ | |
4811 | if (__is_pointer_value(false, false_reg)) | |
4812 | return; | |
4cabc5b1 | 4813 | |
092ed096 JW |
4814 | val = is_jmp32 ? (u32)val : val; |
4815 | sval = is_jmp32 ? (s64)(s32)val : (s64)val; | |
a72dafaf | 4816 | |
48461135 JB |
4817 | switch (opcode) { |
4818 | case BPF_JEQ: | |
48461135 | 4819 | case BPF_JNE: |
a72dafaf JW |
4820 | { |
4821 | struct bpf_reg_state *reg = | |
4822 | opcode == BPF_JEQ ? true_reg : false_reg; | |
4823 | ||
4824 | /* For BPF_JEQ, if this is false we know nothing Jon Snow, but | |
4825 | * if it is true we know the value for sure. Likewise for | |
4826 | * BPF_JNE. | |
48461135 | 4827 | */ |
092ed096 JW |
4828 | if (is_jmp32) { |
4829 | u64 old_v = reg->var_off.value; | |
4830 | u64 hi_mask = ~0xffffffffULL; | |
4831 | ||
4832 | reg->var_off.value = (old_v & hi_mask) | val; | |
4833 | reg->var_off.mask &= hi_mask; | |
4834 | } else { | |
4835 | __mark_reg_known(reg, val); | |
4836 | } | |
48461135 | 4837 | break; |
a72dafaf | 4838 | } |
960ea056 JK |
4839 | case BPF_JSET: |
4840 | false_reg->var_off = tnum_and(false_reg->var_off, | |
4841 | tnum_const(~val)); | |
4842 | if (is_power_of_2(val)) | |
4843 | true_reg->var_off = tnum_or(true_reg->var_off, | |
4844 | tnum_const(val)); | |
4845 | break; | |
48461135 | 4846 | case BPF_JGE: |
a72dafaf JW |
4847 | case BPF_JGT: |
4848 | { | |
4849 | u64 false_umax = opcode == BPF_JGT ? val : val - 1; | |
4850 | u64 true_umin = opcode == BPF_JGT ? val + 1 : val; | |
4851 | ||
092ed096 JW |
4852 | if (is_jmp32) { |
4853 | false_umax += gen_hi_max(false_reg->var_off); | |
4854 | true_umin += gen_hi_min(true_reg->var_off); | |
4855 | } | |
a72dafaf JW |
4856 | false_reg->umax_value = min(false_reg->umax_value, false_umax); |
4857 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
b03c9f9f | 4858 | break; |
a72dafaf | 4859 | } |
48461135 | 4860 | case BPF_JSGE: |
a72dafaf JW |
4861 | case BPF_JSGT: |
4862 | { | |
4863 | s64 false_smax = opcode == BPF_JSGT ? sval : sval - 1; | |
4864 | s64 true_smin = opcode == BPF_JSGT ? sval + 1 : sval; | |
4865 | ||
092ed096 JW |
4866 | /* If the full s64 was not sign-extended from s32 then don't |
4867 | * deduct further info. | |
4868 | */ | |
4869 | if (is_jmp32 && !cmp_val_with_extended_s64(sval, false_reg)) | |
4870 | break; | |
a72dafaf JW |
4871 | false_reg->smax_value = min(false_reg->smax_value, false_smax); |
4872 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
48461135 | 4873 | break; |
a72dafaf | 4874 | } |
b4e432f1 | 4875 | case BPF_JLE: |
a72dafaf JW |
4876 | case BPF_JLT: |
4877 | { | |
4878 | u64 false_umin = opcode == BPF_JLT ? val : val + 1; | |
4879 | u64 true_umax = opcode == BPF_JLT ? val - 1 : val; | |
4880 | ||
092ed096 JW |
4881 | if (is_jmp32) { |
4882 | false_umin += gen_hi_min(false_reg->var_off); | |
4883 | true_umax += gen_hi_max(true_reg->var_off); | |
4884 | } | |
a72dafaf JW |
4885 | false_reg->umin_value = max(false_reg->umin_value, false_umin); |
4886 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
b4e432f1 | 4887 | break; |
a72dafaf | 4888 | } |
b4e432f1 | 4889 | case BPF_JSLE: |
a72dafaf JW |
4890 | case BPF_JSLT: |
4891 | { | |
4892 | s64 false_smin = opcode == BPF_JSLT ? sval : sval + 1; | |
4893 | s64 true_smax = opcode == BPF_JSLT ? sval - 1 : sval; | |
4894 | ||
092ed096 JW |
4895 | if (is_jmp32 && !cmp_val_with_extended_s64(sval, false_reg)) |
4896 | break; | |
a72dafaf JW |
4897 | false_reg->smin_value = max(false_reg->smin_value, false_smin); |
4898 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
b4e432f1 | 4899 | break; |
a72dafaf | 4900 | } |
48461135 JB |
4901 | default: |
4902 | break; | |
4903 | } | |
4904 | ||
b03c9f9f EC |
4905 | __reg_deduce_bounds(false_reg); |
4906 | __reg_deduce_bounds(true_reg); | |
4907 | /* We might have learned some bits from the bounds. */ | |
4908 | __reg_bound_offset(false_reg); | |
4909 | __reg_bound_offset(true_reg); | |
4910 | /* Intersecting with the old var_off might have improved our bounds | |
4911 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
4912 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
4913 | */ | |
4914 | __update_reg_bounds(false_reg); | |
4915 | __update_reg_bounds(true_reg); | |
48461135 JB |
4916 | } |
4917 | ||
f1174f77 EC |
4918 | /* Same as above, but for the case that dst_reg holds a constant and src_reg is |
4919 | * the variable reg. | |
48461135 JB |
4920 | */ |
4921 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
4922 | struct bpf_reg_state *false_reg, u64 val, | |
092ed096 | 4923 | u8 opcode, bool is_jmp32) |
48461135 | 4924 | { |
a72dafaf JW |
4925 | s64 sval; |
4926 | ||
f1174f77 EC |
4927 | if (__is_pointer_value(false, false_reg)) |
4928 | return; | |
4cabc5b1 | 4929 | |
092ed096 JW |
4930 | val = is_jmp32 ? (u32)val : val; |
4931 | sval = is_jmp32 ? (s64)(s32)val : (s64)val; | |
a72dafaf | 4932 | |
48461135 JB |
4933 | switch (opcode) { |
4934 | case BPF_JEQ: | |
48461135 | 4935 | case BPF_JNE: |
a72dafaf JW |
4936 | { |
4937 | struct bpf_reg_state *reg = | |
4938 | opcode == BPF_JEQ ? true_reg : false_reg; | |
4939 | ||
092ed096 JW |
4940 | if (is_jmp32) { |
4941 | u64 old_v = reg->var_off.value; | |
4942 | u64 hi_mask = ~0xffffffffULL; | |
4943 | ||
4944 | reg->var_off.value = (old_v & hi_mask) | val; | |
4945 | reg->var_off.mask &= hi_mask; | |
4946 | } else { | |
4947 | __mark_reg_known(reg, val); | |
4948 | } | |
48461135 | 4949 | break; |
a72dafaf | 4950 | } |
960ea056 JK |
4951 | case BPF_JSET: |
4952 | false_reg->var_off = tnum_and(false_reg->var_off, | |
4953 | tnum_const(~val)); | |
4954 | if (is_power_of_2(val)) | |
4955 | true_reg->var_off = tnum_or(true_reg->var_off, | |
4956 | tnum_const(val)); | |
4957 | break; | |
48461135 | 4958 | case BPF_JGE: |
a72dafaf JW |
4959 | case BPF_JGT: |
4960 | { | |
4961 | u64 false_umin = opcode == BPF_JGT ? val : val + 1; | |
4962 | u64 true_umax = opcode == BPF_JGT ? val - 1 : val; | |
4963 | ||
092ed096 JW |
4964 | if (is_jmp32) { |
4965 | false_umin += gen_hi_min(false_reg->var_off); | |
4966 | true_umax += gen_hi_max(true_reg->var_off); | |
4967 | } | |
a72dafaf JW |
4968 | false_reg->umin_value = max(false_reg->umin_value, false_umin); |
4969 | true_reg->umax_value = min(true_reg->umax_value, true_umax); | |
b03c9f9f | 4970 | break; |
a72dafaf | 4971 | } |
48461135 | 4972 | case BPF_JSGE: |
a72dafaf JW |
4973 | case BPF_JSGT: |
4974 | { | |
4975 | s64 false_smin = opcode == BPF_JSGT ? sval : sval + 1; | |
4976 | s64 true_smax = opcode == BPF_JSGT ? sval - 1 : sval; | |
4977 | ||
092ed096 JW |
4978 | if (is_jmp32 && !cmp_val_with_extended_s64(sval, false_reg)) |
4979 | break; | |
a72dafaf JW |
4980 | false_reg->smin_value = max(false_reg->smin_value, false_smin); |
4981 | true_reg->smax_value = min(true_reg->smax_value, true_smax); | |
48461135 | 4982 | break; |
a72dafaf | 4983 | } |
b4e432f1 | 4984 | case BPF_JLE: |
a72dafaf JW |
4985 | case BPF_JLT: |
4986 | { | |
4987 | u64 false_umax = opcode == BPF_JLT ? val : val - 1; | |
4988 | u64 true_umin = opcode == BPF_JLT ? val + 1 : val; | |
4989 | ||
092ed096 JW |
4990 | if (is_jmp32) { |
4991 | false_umax += gen_hi_max(false_reg->var_off); | |
4992 | true_umin += gen_hi_min(true_reg->var_off); | |
4993 | } | |
a72dafaf JW |
4994 | false_reg->umax_value = min(false_reg->umax_value, false_umax); |
4995 | true_reg->umin_value = max(true_reg->umin_value, true_umin); | |
b4e432f1 | 4996 | break; |
a72dafaf | 4997 | } |
b4e432f1 | 4998 | case BPF_JSLE: |
a72dafaf JW |
4999 | case BPF_JSLT: |
5000 | { | |
5001 | s64 false_smax = opcode == BPF_JSLT ? sval : sval - 1; | |
5002 | s64 true_smin = opcode == BPF_JSLT ? sval + 1 : sval; | |
5003 | ||
092ed096 JW |
5004 | if (is_jmp32 && !cmp_val_with_extended_s64(sval, false_reg)) |
5005 | break; | |
a72dafaf JW |
5006 | false_reg->smax_value = min(false_reg->smax_value, false_smax); |
5007 | true_reg->smin_value = max(true_reg->smin_value, true_smin); | |
b4e432f1 | 5008 | break; |
a72dafaf | 5009 | } |
48461135 JB |
5010 | default: |
5011 | break; | |
5012 | } | |
5013 | ||
b03c9f9f EC |
5014 | __reg_deduce_bounds(false_reg); |
5015 | __reg_deduce_bounds(true_reg); | |
5016 | /* We might have learned some bits from the bounds. */ | |
5017 | __reg_bound_offset(false_reg); | |
5018 | __reg_bound_offset(true_reg); | |
5019 | /* Intersecting with the old var_off might have improved our bounds | |
5020 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
5021 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
5022 | */ | |
5023 | __update_reg_bounds(false_reg); | |
5024 | __update_reg_bounds(true_reg); | |
f1174f77 EC |
5025 | } |
5026 | ||
5027 | /* Regs are known to be equal, so intersect their min/max/var_off */ | |
5028 | static void __reg_combine_min_max(struct bpf_reg_state *src_reg, | |
5029 | struct bpf_reg_state *dst_reg) | |
5030 | { | |
b03c9f9f EC |
5031 | src_reg->umin_value = dst_reg->umin_value = max(src_reg->umin_value, |
5032 | dst_reg->umin_value); | |
5033 | src_reg->umax_value = dst_reg->umax_value = min(src_reg->umax_value, | |
5034 | dst_reg->umax_value); | |
5035 | src_reg->smin_value = dst_reg->smin_value = max(src_reg->smin_value, | |
5036 | dst_reg->smin_value); | |
5037 | src_reg->smax_value = dst_reg->smax_value = min(src_reg->smax_value, | |
5038 | dst_reg->smax_value); | |
f1174f77 EC |
5039 | src_reg->var_off = dst_reg->var_off = tnum_intersect(src_reg->var_off, |
5040 | dst_reg->var_off); | |
b03c9f9f EC |
5041 | /* We might have learned new bounds from the var_off. */ |
5042 | __update_reg_bounds(src_reg); | |
5043 | __update_reg_bounds(dst_reg); | |
5044 | /* We might have learned something about the sign bit. */ | |
5045 | __reg_deduce_bounds(src_reg); | |
5046 | __reg_deduce_bounds(dst_reg); | |
5047 | /* We might have learned some bits from the bounds. */ | |
5048 | __reg_bound_offset(src_reg); | |
5049 | __reg_bound_offset(dst_reg); | |
5050 | /* Intersecting with the old var_off might have improved our bounds | |
5051 | * slightly. e.g. if umax was 0x7f...f and var_off was (0; 0xf...fc), | |
5052 | * then new var_off is (0; 0x7f...fc) which improves our umax. | |
5053 | */ | |
5054 | __update_reg_bounds(src_reg); | |
5055 | __update_reg_bounds(dst_reg); | |
f1174f77 EC |
5056 | } |
5057 | ||
5058 | static void reg_combine_min_max(struct bpf_reg_state *true_src, | |
5059 | struct bpf_reg_state *true_dst, | |
5060 | struct bpf_reg_state *false_src, | |
5061 | struct bpf_reg_state *false_dst, | |
5062 | u8 opcode) | |
5063 | { | |
5064 | switch (opcode) { | |
5065 | case BPF_JEQ: | |
5066 | __reg_combine_min_max(true_src, true_dst); | |
5067 | break; | |
5068 | case BPF_JNE: | |
5069 | __reg_combine_min_max(false_src, false_dst); | |
b03c9f9f | 5070 | break; |
4cabc5b1 | 5071 | } |
48461135 JB |
5072 | } |
5073 | ||
fd978bf7 JS |
5074 | static void mark_ptr_or_null_reg(struct bpf_func_state *state, |
5075 | struct bpf_reg_state *reg, u32 id, | |
840b9615 | 5076 | bool is_null) |
57a09bf0 | 5077 | { |
840b9615 | 5078 | if (reg_type_may_be_null(reg->type) && reg->id == id) { |
f1174f77 EC |
5079 | /* Old offset (both fixed and variable parts) should |
5080 | * have been known-zero, because we don't allow pointer | |
5081 | * arithmetic on pointers that might be NULL. | |
5082 | */ | |
b03c9f9f EC |
5083 | if (WARN_ON_ONCE(reg->smin_value || reg->smax_value || |
5084 | !tnum_equals_const(reg->var_off, 0) || | |
f1174f77 | 5085 | reg->off)) { |
b03c9f9f EC |
5086 | __mark_reg_known_zero(reg); |
5087 | reg->off = 0; | |
f1174f77 EC |
5088 | } |
5089 | if (is_null) { | |
5090 | reg->type = SCALAR_VALUE; | |
840b9615 JS |
5091 | } else if (reg->type == PTR_TO_MAP_VALUE_OR_NULL) { |
5092 | if (reg->map_ptr->inner_map_meta) { | |
5093 | reg->type = CONST_PTR_TO_MAP; | |
5094 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
fada7fdc JL |
5095 | } else if (reg->map_ptr->map_type == |
5096 | BPF_MAP_TYPE_XSKMAP) { | |
5097 | reg->type = PTR_TO_XDP_SOCK; | |
840b9615 JS |
5098 | } else { |
5099 | reg->type = PTR_TO_MAP_VALUE; | |
5100 | } | |
c64b7983 JS |
5101 | } else if (reg->type == PTR_TO_SOCKET_OR_NULL) { |
5102 | reg->type = PTR_TO_SOCKET; | |
46f8bc92 MKL |
5103 | } else if (reg->type == PTR_TO_SOCK_COMMON_OR_NULL) { |
5104 | reg->type = PTR_TO_SOCK_COMMON; | |
655a51e5 MKL |
5105 | } else if (reg->type == PTR_TO_TCP_SOCK_OR_NULL) { |
5106 | reg->type = PTR_TO_TCP_SOCK; | |
56f668df | 5107 | } |
1b986589 MKL |
5108 | if (is_null) { |
5109 | /* We don't need id and ref_obj_id from this point | |
5110 | * onwards anymore, thus we should better reset it, | |
5111 | * so that state pruning has chances to take effect. | |
5112 | */ | |
5113 | reg->id = 0; | |
5114 | reg->ref_obj_id = 0; | |
5115 | } else if (!reg_may_point_to_spin_lock(reg)) { | |
5116 | /* For not-NULL ptr, reg->ref_obj_id will be reset | |
5117 | * in release_reg_references(). | |
5118 | * | |
5119 | * reg->id is still used by spin_lock ptr. Other | |
5120 | * than spin_lock ptr type, reg->id can be reset. | |
fd978bf7 JS |
5121 | */ |
5122 | reg->id = 0; | |
56f668df | 5123 | } |
57a09bf0 TG |
5124 | } |
5125 | } | |
5126 | ||
c6a9efa1 PC |
5127 | static void __mark_ptr_or_null_regs(struct bpf_func_state *state, u32 id, |
5128 | bool is_null) | |
5129 | { | |
5130 | struct bpf_reg_state *reg; | |
5131 | int i; | |
5132 | ||
5133 | for (i = 0; i < MAX_BPF_REG; i++) | |
5134 | mark_ptr_or_null_reg(state, &state->regs[i], id, is_null); | |
5135 | ||
5136 | bpf_for_each_spilled_reg(i, state, reg) { | |
5137 | if (!reg) | |
5138 | continue; | |
5139 | mark_ptr_or_null_reg(state, reg, id, is_null); | |
5140 | } | |
5141 | } | |
5142 | ||
57a09bf0 TG |
5143 | /* The logic is similar to find_good_pkt_pointers(), both could eventually |
5144 | * be folded together at some point. | |
5145 | */ | |
840b9615 JS |
5146 | static void mark_ptr_or_null_regs(struct bpf_verifier_state *vstate, u32 regno, |
5147 | bool is_null) | |
57a09bf0 | 5148 | { |
f4d7e40a | 5149 | struct bpf_func_state *state = vstate->frame[vstate->curframe]; |
c6a9efa1 | 5150 | struct bpf_reg_state *regs = state->regs; |
1b986589 | 5151 | u32 ref_obj_id = regs[regno].ref_obj_id; |
a08dd0da | 5152 | u32 id = regs[regno].id; |
c6a9efa1 | 5153 | int i; |
57a09bf0 | 5154 | |
1b986589 MKL |
5155 | if (ref_obj_id && ref_obj_id == id && is_null) |
5156 | /* regs[regno] is in the " == NULL" branch. | |
5157 | * No one could have freed the reference state before | |
5158 | * doing the NULL check. | |
5159 | */ | |
5160 | WARN_ON_ONCE(release_reference_state(state, id)); | |
fd978bf7 | 5161 | |
c6a9efa1 PC |
5162 | for (i = 0; i <= vstate->curframe; i++) |
5163 | __mark_ptr_or_null_regs(vstate->frame[i], id, is_null); | |
57a09bf0 TG |
5164 | } |
5165 | ||
5beca081 DB |
5166 | static bool try_match_pkt_pointers(const struct bpf_insn *insn, |
5167 | struct bpf_reg_state *dst_reg, | |
5168 | struct bpf_reg_state *src_reg, | |
5169 | struct bpf_verifier_state *this_branch, | |
5170 | struct bpf_verifier_state *other_branch) | |
5171 | { | |
5172 | if (BPF_SRC(insn->code) != BPF_X) | |
5173 | return false; | |
5174 | ||
092ed096 JW |
5175 | /* Pointers are always 64-bit. */ |
5176 | if (BPF_CLASS(insn->code) == BPF_JMP32) | |
5177 | return false; | |
5178 | ||
5beca081 DB |
5179 | switch (BPF_OP(insn->code)) { |
5180 | case BPF_JGT: | |
5181 | if ((dst_reg->type == PTR_TO_PACKET && | |
5182 | src_reg->type == PTR_TO_PACKET_END) || | |
5183 | (dst_reg->type == PTR_TO_PACKET_META && | |
5184 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
5185 | /* pkt_data' > pkt_end, pkt_meta' > pkt_data */ | |
5186 | find_good_pkt_pointers(this_branch, dst_reg, | |
5187 | dst_reg->type, false); | |
5188 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
5189 | src_reg->type == PTR_TO_PACKET) || | |
5190 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
5191 | src_reg->type == PTR_TO_PACKET_META)) { | |
5192 | /* pkt_end > pkt_data', pkt_data > pkt_meta' */ | |
5193 | find_good_pkt_pointers(other_branch, src_reg, | |
5194 | src_reg->type, true); | |
5195 | } else { | |
5196 | return false; | |
5197 | } | |
5198 | break; | |
5199 | case BPF_JLT: | |
5200 | if ((dst_reg->type == PTR_TO_PACKET && | |
5201 | src_reg->type == PTR_TO_PACKET_END) || | |
5202 | (dst_reg->type == PTR_TO_PACKET_META && | |
5203 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
5204 | /* pkt_data' < pkt_end, pkt_meta' < pkt_data */ | |
5205 | find_good_pkt_pointers(other_branch, dst_reg, | |
5206 | dst_reg->type, true); | |
5207 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
5208 | src_reg->type == PTR_TO_PACKET) || | |
5209 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
5210 | src_reg->type == PTR_TO_PACKET_META)) { | |
5211 | /* pkt_end < pkt_data', pkt_data > pkt_meta' */ | |
5212 | find_good_pkt_pointers(this_branch, src_reg, | |
5213 | src_reg->type, false); | |
5214 | } else { | |
5215 | return false; | |
5216 | } | |
5217 | break; | |
5218 | case BPF_JGE: | |
5219 | if ((dst_reg->type == PTR_TO_PACKET && | |
5220 | src_reg->type == PTR_TO_PACKET_END) || | |
5221 | (dst_reg->type == PTR_TO_PACKET_META && | |
5222 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
5223 | /* pkt_data' >= pkt_end, pkt_meta' >= pkt_data */ | |
5224 | find_good_pkt_pointers(this_branch, dst_reg, | |
5225 | dst_reg->type, true); | |
5226 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
5227 | src_reg->type == PTR_TO_PACKET) || | |
5228 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
5229 | src_reg->type == PTR_TO_PACKET_META)) { | |
5230 | /* pkt_end >= pkt_data', pkt_data >= pkt_meta' */ | |
5231 | find_good_pkt_pointers(other_branch, src_reg, | |
5232 | src_reg->type, false); | |
5233 | } else { | |
5234 | return false; | |
5235 | } | |
5236 | break; | |
5237 | case BPF_JLE: | |
5238 | if ((dst_reg->type == PTR_TO_PACKET && | |
5239 | src_reg->type == PTR_TO_PACKET_END) || | |
5240 | (dst_reg->type == PTR_TO_PACKET_META && | |
5241 | reg_is_init_pkt_pointer(src_reg, PTR_TO_PACKET))) { | |
5242 | /* pkt_data' <= pkt_end, pkt_meta' <= pkt_data */ | |
5243 | find_good_pkt_pointers(other_branch, dst_reg, | |
5244 | dst_reg->type, false); | |
5245 | } else if ((dst_reg->type == PTR_TO_PACKET_END && | |
5246 | src_reg->type == PTR_TO_PACKET) || | |
5247 | (reg_is_init_pkt_pointer(dst_reg, PTR_TO_PACKET) && | |
5248 | src_reg->type == PTR_TO_PACKET_META)) { | |
5249 | /* pkt_end <= pkt_data', pkt_data <= pkt_meta' */ | |
5250 | find_good_pkt_pointers(this_branch, src_reg, | |
5251 | src_reg->type, true); | |
5252 | } else { | |
5253 | return false; | |
5254 | } | |
5255 | break; | |
5256 | default: | |
5257 | return false; | |
5258 | } | |
5259 | ||
5260 | return true; | |
5261 | } | |
5262 | ||
58e2af8b | 5263 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
5264 | struct bpf_insn *insn, int *insn_idx) |
5265 | { | |
f4d7e40a AS |
5266 | struct bpf_verifier_state *this_branch = env->cur_state; |
5267 | struct bpf_verifier_state *other_branch; | |
5268 | struct bpf_reg_state *regs = this_branch->frame[this_branch->curframe]->regs; | |
fb8d251e | 5269 | struct bpf_reg_state *dst_reg, *other_branch_regs, *src_reg = NULL; |
17a52670 | 5270 | u8 opcode = BPF_OP(insn->code); |
092ed096 | 5271 | bool is_jmp32; |
fb8d251e | 5272 | int pred = -1; |
17a52670 AS |
5273 | int err; |
5274 | ||
092ed096 JW |
5275 | /* Only conditional jumps are expected to reach here. */ |
5276 | if (opcode == BPF_JA || opcode > BPF_JSLE) { | |
5277 | verbose(env, "invalid BPF_JMP/JMP32 opcode %x\n", opcode); | |
17a52670 AS |
5278 | return -EINVAL; |
5279 | } | |
5280 | ||
5281 | if (BPF_SRC(insn->code) == BPF_X) { | |
5282 | if (insn->imm != 0) { | |
092ed096 | 5283 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
5284 | return -EINVAL; |
5285 | } | |
5286 | ||
5287 | /* check src1 operand */ | |
dc503a8a | 5288 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
5289 | if (err) |
5290 | return err; | |
1be7f75d AS |
5291 | |
5292 | if (is_pointer_value(env, insn->src_reg)) { | |
61bd5218 | 5293 | verbose(env, "R%d pointer comparison prohibited\n", |
1be7f75d AS |
5294 | insn->src_reg); |
5295 | return -EACCES; | |
5296 | } | |
fb8d251e | 5297 | src_reg = ®s[insn->src_reg]; |
17a52670 AS |
5298 | } else { |
5299 | if (insn->src_reg != BPF_REG_0) { | |
092ed096 | 5300 | verbose(env, "BPF_JMP/JMP32 uses reserved fields\n"); |
17a52670 AS |
5301 | return -EINVAL; |
5302 | } | |
5303 | } | |
5304 | ||
5305 | /* check src2 operand */ | |
dc503a8a | 5306 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
5307 | if (err) |
5308 | return err; | |
5309 | ||
1a0dc1ac | 5310 | dst_reg = ®s[insn->dst_reg]; |
092ed096 | 5311 | is_jmp32 = BPF_CLASS(insn->code) == BPF_JMP32; |
1a0dc1ac | 5312 | |
fb8d251e AS |
5313 | if (BPF_SRC(insn->code) == BPF_K) |
5314 | pred = is_branch_taken(dst_reg, insn->imm, | |
5315 | opcode, is_jmp32); | |
5316 | else if (src_reg->type == SCALAR_VALUE && | |
5317 | tnum_is_const(src_reg->var_off)) | |
5318 | pred = is_branch_taken(dst_reg, src_reg->var_off.value, | |
5319 | opcode, is_jmp32); | |
5320 | if (pred == 1) { | |
5321 | /* only follow the goto, ignore fall-through */ | |
5322 | *insn_idx += insn->off; | |
5323 | return 0; | |
5324 | } else if (pred == 0) { | |
5325 | /* only follow fall-through branch, since | |
5326 | * that's where the program will go | |
5327 | */ | |
5328 | return 0; | |
17a52670 AS |
5329 | } |
5330 | ||
979d63d5 DB |
5331 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx, |
5332 | false); | |
17a52670 AS |
5333 | if (!other_branch) |
5334 | return -EFAULT; | |
f4d7e40a | 5335 | other_branch_regs = other_branch->frame[other_branch->curframe]->regs; |
17a52670 | 5336 | |
48461135 JB |
5337 | /* detect if we are comparing against a constant value so we can adjust |
5338 | * our min/max values for our dst register. | |
f1174f77 EC |
5339 | * this is only legit if both are scalars (or pointers to the same |
5340 | * object, I suppose, but we don't support that right now), because | |
5341 | * otherwise the different base pointers mean the offsets aren't | |
5342 | * comparable. | |
48461135 JB |
5343 | */ |
5344 | if (BPF_SRC(insn->code) == BPF_X) { | |
092ed096 JW |
5345 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
5346 | struct bpf_reg_state lo_reg0 = *dst_reg; | |
5347 | struct bpf_reg_state lo_reg1 = *src_reg; | |
5348 | struct bpf_reg_state *src_lo, *dst_lo; | |
5349 | ||
5350 | dst_lo = &lo_reg0; | |
5351 | src_lo = &lo_reg1; | |
5352 | coerce_reg_to_size(dst_lo, 4); | |
5353 | coerce_reg_to_size(src_lo, 4); | |
5354 | ||
f1174f77 | 5355 | if (dst_reg->type == SCALAR_VALUE && |
092ed096 JW |
5356 | src_reg->type == SCALAR_VALUE) { |
5357 | if (tnum_is_const(src_reg->var_off) || | |
5358 | (is_jmp32 && tnum_is_const(src_lo->var_off))) | |
f4d7e40a | 5359 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 JW |
5360 | dst_reg, |
5361 | is_jmp32 | |
5362 | ? src_lo->var_off.value | |
5363 | : src_reg->var_off.value, | |
5364 | opcode, is_jmp32); | |
5365 | else if (tnum_is_const(dst_reg->var_off) || | |
5366 | (is_jmp32 && tnum_is_const(dst_lo->var_off))) | |
f4d7e40a | 5367 | reg_set_min_max_inv(&other_branch_regs[insn->src_reg], |
092ed096 JW |
5368 | src_reg, |
5369 | is_jmp32 | |
5370 | ? dst_lo->var_off.value | |
5371 | : dst_reg->var_off.value, | |
5372 | opcode, is_jmp32); | |
5373 | else if (!is_jmp32 && | |
5374 | (opcode == BPF_JEQ || opcode == BPF_JNE)) | |
f1174f77 | 5375 | /* Comparing for equality, we can combine knowledge */ |
f4d7e40a AS |
5376 | reg_combine_min_max(&other_branch_regs[insn->src_reg], |
5377 | &other_branch_regs[insn->dst_reg], | |
092ed096 | 5378 | src_reg, dst_reg, opcode); |
f1174f77 EC |
5379 | } |
5380 | } else if (dst_reg->type == SCALAR_VALUE) { | |
f4d7e40a | 5381 | reg_set_min_max(&other_branch_regs[insn->dst_reg], |
092ed096 | 5382 | dst_reg, insn->imm, opcode, is_jmp32); |
48461135 JB |
5383 | } |
5384 | ||
092ed096 JW |
5385 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem(). |
5386 | * NOTE: these optimizations below are related with pointer comparison | |
5387 | * which will never be JMP32. | |
5388 | */ | |
5389 | if (!is_jmp32 && BPF_SRC(insn->code) == BPF_K && | |
1a0dc1ac | 5390 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
840b9615 JS |
5391 | reg_type_may_be_null(dst_reg->type)) { |
5392 | /* Mark all identical registers in each branch as either | |
57a09bf0 TG |
5393 | * safe or unknown depending R == 0 or R != 0 conditional. |
5394 | */ | |
840b9615 JS |
5395 | mark_ptr_or_null_regs(this_branch, insn->dst_reg, |
5396 | opcode == BPF_JNE); | |
5397 | mark_ptr_or_null_regs(other_branch, insn->dst_reg, | |
5398 | opcode == BPF_JEQ); | |
5beca081 DB |
5399 | } else if (!try_match_pkt_pointers(insn, dst_reg, ®s[insn->src_reg], |
5400 | this_branch, other_branch) && | |
5401 | is_pointer_value(env, insn->dst_reg)) { | |
61bd5218 JK |
5402 | verbose(env, "R%d pointer comparison prohibited\n", |
5403 | insn->dst_reg); | |
1be7f75d | 5404 | return -EACCES; |
17a52670 | 5405 | } |
06ee7115 | 5406 | if (env->log.level & BPF_LOG_LEVEL) |
f4d7e40a | 5407 | print_verifier_state(env, this_branch->frame[this_branch->curframe]); |
17a52670 AS |
5408 | return 0; |
5409 | } | |
5410 | ||
17a52670 | 5411 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 5412 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 5413 | { |
d8eca5bb | 5414 | struct bpf_insn_aux_data *aux = cur_aux(env); |
638f5b90 | 5415 | struct bpf_reg_state *regs = cur_regs(env); |
d8eca5bb | 5416 | struct bpf_map *map; |
17a52670 AS |
5417 | int err; |
5418 | ||
5419 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
61bd5218 | 5420 | verbose(env, "invalid BPF_LD_IMM insn\n"); |
17a52670 AS |
5421 | return -EINVAL; |
5422 | } | |
5423 | if (insn->off != 0) { | |
61bd5218 | 5424 | verbose(env, "BPF_LD_IMM64 uses reserved fields\n"); |
17a52670 AS |
5425 | return -EINVAL; |
5426 | } | |
5427 | ||
dc503a8a | 5428 | err = check_reg_arg(env, insn->dst_reg, DST_OP); |
17a52670 AS |
5429 | if (err) |
5430 | return err; | |
5431 | ||
6b173873 | 5432 | if (insn->src_reg == 0) { |
6b173873 JK |
5433 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
5434 | ||
f1174f77 | 5435 | regs[insn->dst_reg].type = SCALAR_VALUE; |
b03c9f9f | 5436 | __mark_reg_known(®s[insn->dst_reg], imm); |
17a52670 | 5437 | return 0; |
6b173873 | 5438 | } |
17a52670 | 5439 | |
d8eca5bb DB |
5440 | map = env->used_maps[aux->map_index]; |
5441 | mark_reg_known_zero(env, regs, insn->dst_reg); | |
5442 | regs[insn->dst_reg].map_ptr = map; | |
5443 | ||
5444 | if (insn->src_reg == BPF_PSEUDO_MAP_VALUE) { | |
5445 | regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; | |
5446 | regs[insn->dst_reg].off = aux->map_off; | |
5447 | if (map_value_has_spin_lock(map)) | |
5448 | regs[insn->dst_reg].id = ++env->id_gen; | |
5449 | } else if (insn->src_reg == BPF_PSEUDO_MAP_FD) { | |
5450 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
5451 | } else { | |
5452 | verbose(env, "bpf verifier is misconfigured\n"); | |
5453 | return -EINVAL; | |
5454 | } | |
17a52670 | 5455 | |
17a52670 AS |
5456 | return 0; |
5457 | } | |
5458 | ||
96be4325 DB |
5459 | static bool may_access_skb(enum bpf_prog_type type) |
5460 | { | |
5461 | switch (type) { | |
5462 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
5463 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 5464 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
5465 | return true; |
5466 | default: | |
5467 | return false; | |
5468 | } | |
5469 | } | |
5470 | ||
ddd872bc AS |
5471 | /* verify safety of LD_ABS|LD_IND instructions: |
5472 | * - they can only appear in the programs where ctx == skb | |
5473 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
5474 | * preserve R6-R9, and store return value into R0 | |
5475 | * | |
5476 | * Implicit input: | |
5477 | * ctx == skb == R6 == CTX | |
5478 | * | |
5479 | * Explicit input: | |
5480 | * SRC == any register | |
5481 | * IMM == 32-bit immediate | |
5482 | * | |
5483 | * Output: | |
5484 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
5485 | */ | |
58e2af8b | 5486 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 5487 | { |
638f5b90 | 5488 | struct bpf_reg_state *regs = cur_regs(env); |
ddd872bc | 5489 | u8 mode = BPF_MODE(insn->code); |
ddd872bc AS |
5490 | int i, err; |
5491 | ||
24701ece | 5492 | if (!may_access_skb(env->prog->type)) { |
61bd5218 | 5493 | verbose(env, "BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
5494 | return -EINVAL; |
5495 | } | |
5496 | ||
e0cea7ce DB |
5497 | if (!env->ops->gen_ld_abs) { |
5498 | verbose(env, "bpf verifier is misconfigured\n"); | |
5499 | return -EINVAL; | |
5500 | } | |
5501 | ||
f910cefa | 5502 | if (env->subprog_cnt > 1) { |
f4d7e40a AS |
5503 | /* when program has LD_ABS insn JITs and interpreter assume |
5504 | * that r1 == ctx == skb which is not the case for callees | |
5505 | * that can have arbitrary arguments. It's problematic | |
5506 | * for main prog as well since JITs would need to analyze | |
5507 | * all functions in order to make proper register save/restore | |
5508 | * decisions in the main prog. Hence disallow LD_ABS with calls | |
5509 | */ | |
5510 | verbose(env, "BPF_LD_[ABS|IND] instructions cannot be mixed with bpf-to-bpf calls\n"); | |
5511 | return -EINVAL; | |
5512 | } | |
5513 | ||
ddd872bc | 5514 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || |
d82bccc6 | 5515 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 5516 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
61bd5218 | 5517 | verbose(env, "BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
5518 | return -EINVAL; |
5519 | } | |
5520 | ||
5521 | /* check whether implicit source operand (register R6) is readable */ | |
dc503a8a | 5522 | err = check_reg_arg(env, BPF_REG_6, SRC_OP); |
ddd872bc AS |
5523 | if (err) |
5524 | return err; | |
5525 | ||
fd978bf7 JS |
5526 | /* Disallow usage of BPF_LD_[ABS|IND] with reference tracking, as |
5527 | * gen_ld_abs() may terminate the program at runtime, leading to | |
5528 | * reference leak. | |
5529 | */ | |
5530 | err = check_reference_leak(env); | |
5531 | if (err) { | |
5532 | verbose(env, "BPF_LD_[ABS|IND] cannot be mixed with socket references\n"); | |
5533 | return err; | |
5534 | } | |
5535 | ||
d83525ca AS |
5536 | if (env->cur_state->active_spin_lock) { |
5537 | verbose(env, "BPF_LD_[ABS|IND] cannot be used inside bpf_spin_lock-ed region\n"); | |
5538 | return -EINVAL; | |
5539 | } | |
5540 | ||
ddd872bc | 5541 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { |
61bd5218 JK |
5542 | verbose(env, |
5543 | "at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
ddd872bc AS |
5544 | return -EINVAL; |
5545 | } | |
5546 | ||
5547 | if (mode == BPF_IND) { | |
5548 | /* check explicit source operand */ | |
dc503a8a | 5549 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
ddd872bc AS |
5550 | if (err) |
5551 | return err; | |
5552 | } | |
5553 | ||
5554 | /* reset caller saved regs to unreadable */ | |
dc503a8a | 5555 | for (i = 0; i < CALLER_SAVED_REGS; i++) { |
61bd5218 | 5556 | mark_reg_not_init(env, regs, caller_saved[i]); |
dc503a8a EC |
5557 | check_reg_arg(env, caller_saved[i], DST_OP_NO_MARK); |
5558 | } | |
ddd872bc AS |
5559 | |
5560 | /* mark destination R0 register as readable, since it contains | |
dc503a8a EC |
5561 | * the value fetched from the packet. |
5562 | * Already marked as written above. | |
ddd872bc | 5563 | */ |
61bd5218 | 5564 | mark_reg_unknown(env, regs, BPF_REG_0); |
5327ed3d JW |
5565 | /* ld_abs load up to 32-bit skb data. */ |
5566 | regs[BPF_REG_0].subreg_def = env->insn_idx + 1; | |
ddd872bc AS |
5567 | return 0; |
5568 | } | |
5569 | ||
390ee7e2 AS |
5570 | static int check_return_code(struct bpf_verifier_env *env) |
5571 | { | |
5cf1e914 | 5572 | struct tnum enforce_attach_type_range = tnum_unknown; |
390ee7e2 AS |
5573 | struct bpf_reg_state *reg; |
5574 | struct tnum range = tnum_range(0, 1); | |
5575 | ||
5576 | switch (env->prog->type) { | |
5577 | case BPF_PROG_TYPE_CGROUP_SKB: | |
5cf1e914 | 5578 | if (env->prog->expected_attach_type == BPF_CGROUP_INET_EGRESS) { |
5579 | range = tnum_range(0, 3); | |
5580 | enforce_attach_type_range = tnum_range(2, 3); | |
5581 | } | |
390ee7e2 | 5582 | case BPF_PROG_TYPE_CGROUP_SOCK: |
4fbac77d | 5583 | case BPF_PROG_TYPE_CGROUP_SOCK_ADDR: |
390ee7e2 | 5584 | case BPF_PROG_TYPE_SOCK_OPS: |
ebc614f6 | 5585 | case BPF_PROG_TYPE_CGROUP_DEVICE: |
7b146ceb | 5586 | case BPF_PROG_TYPE_CGROUP_SYSCTL: |
390ee7e2 AS |
5587 | break; |
5588 | default: | |
5589 | return 0; | |
5590 | } | |
5591 | ||
638f5b90 | 5592 | reg = cur_regs(env) + BPF_REG_0; |
390ee7e2 | 5593 | if (reg->type != SCALAR_VALUE) { |
61bd5218 | 5594 | verbose(env, "At program exit the register R0 is not a known value (%s)\n", |
390ee7e2 AS |
5595 | reg_type_str[reg->type]); |
5596 | return -EINVAL; | |
5597 | } | |
5598 | ||
5599 | if (!tnum_in(range, reg->var_off)) { | |
5cf1e914 | 5600 | char tn_buf[48]; |
5601 | ||
61bd5218 | 5602 | verbose(env, "At program exit the register R0 "); |
390ee7e2 | 5603 | if (!tnum_is_unknown(reg->var_off)) { |
390ee7e2 | 5604 | tnum_strn(tn_buf, sizeof(tn_buf), reg->var_off); |
61bd5218 | 5605 | verbose(env, "has value %s", tn_buf); |
390ee7e2 | 5606 | } else { |
61bd5218 | 5607 | verbose(env, "has unknown scalar value"); |
390ee7e2 | 5608 | } |
5cf1e914 | 5609 | tnum_strn(tn_buf, sizeof(tn_buf), range); |
5610 | verbose(env, " should have been %s\n", tn_buf); | |
390ee7e2 AS |
5611 | return -EINVAL; |
5612 | } | |
5cf1e914 | 5613 | |
5614 | if (!tnum_is_unknown(enforce_attach_type_range) && | |
5615 | tnum_in(enforce_attach_type_range, reg->var_off)) | |
5616 | env->prog->enforce_expected_attach_type = 1; | |
390ee7e2 AS |
5617 | return 0; |
5618 | } | |
5619 | ||
475fb78f AS |
5620 | /* non-recursive DFS pseudo code |
5621 | * 1 procedure DFS-iterative(G,v): | |
5622 | * 2 label v as discovered | |
5623 | * 3 let S be a stack | |
5624 | * 4 S.push(v) | |
5625 | * 5 while S is not empty | |
5626 | * 6 t <- S.pop() | |
5627 | * 7 if t is what we're looking for: | |
5628 | * 8 return t | |
5629 | * 9 for all edges e in G.adjacentEdges(t) do | |
5630 | * 10 if edge e is already labelled | |
5631 | * 11 continue with the next edge | |
5632 | * 12 w <- G.adjacentVertex(t,e) | |
5633 | * 13 if vertex w is not discovered and not explored | |
5634 | * 14 label e as tree-edge | |
5635 | * 15 label w as discovered | |
5636 | * 16 S.push(w) | |
5637 | * 17 continue at 5 | |
5638 | * 18 else if vertex w is discovered | |
5639 | * 19 label e as back-edge | |
5640 | * 20 else | |
5641 | * 21 // vertex w is explored | |
5642 | * 22 label e as forward- or cross-edge | |
5643 | * 23 label t as explored | |
5644 | * 24 S.pop() | |
5645 | * | |
5646 | * convention: | |
5647 | * 0x10 - discovered | |
5648 | * 0x11 - discovered and fall-through edge labelled | |
5649 | * 0x12 - discovered and fall-through and branch edges labelled | |
5650 | * 0x20 - explored | |
5651 | */ | |
5652 | ||
5653 | enum { | |
5654 | DISCOVERED = 0x10, | |
5655 | EXPLORED = 0x20, | |
5656 | FALLTHROUGH = 1, | |
5657 | BRANCH = 2, | |
5658 | }; | |
5659 | ||
dc2a4ebc AS |
5660 | static u32 state_htab_size(struct bpf_verifier_env *env) |
5661 | { | |
5662 | return env->prog->len; | |
5663 | } | |
5664 | ||
5d839021 AS |
5665 | static struct bpf_verifier_state_list **explored_state( |
5666 | struct bpf_verifier_env *env, | |
5667 | int idx) | |
5668 | { | |
dc2a4ebc AS |
5669 | struct bpf_verifier_state *cur = env->cur_state; |
5670 | struct bpf_func_state *state = cur->frame[cur->curframe]; | |
5671 | ||
5672 | return &env->explored_states[(idx ^ state->callsite) % state_htab_size(env)]; | |
5d839021 AS |
5673 | } |
5674 | ||
5675 | static void init_explored_state(struct bpf_verifier_env *env, int idx) | |
5676 | { | |
a8f500af | 5677 | env->insn_aux_data[idx].prune_point = true; |
5d839021 | 5678 | } |
f1bca824 | 5679 | |
475fb78f AS |
5680 | /* t, w, e - match pseudo-code above: |
5681 | * t - index of current instruction | |
5682 | * w - next instruction | |
5683 | * e - edge | |
5684 | */ | |
58e2af8b | 5685 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f | 5686 | { |
7df737e9 AS |
5687 | int *insn_stack = env->cfg.insn_stack; |
5688 | int *insn_state = env->cfg.insn_state; | |
5689 | ||
475fb78f AS |
5690 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) |
5691 | return 0; | |
5692 | ||
5693 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
5694 | return 0; | |
5695 | ||
5696 | if (w < 0 || w >= env->prog->len) { | |
d9762e84 | 5697 | verbose_linfo(env, t, "%d: ", t); |
61bd5218 | 5698 | verbose(env, "jump out of range from insn %d to %d\n", t, w); |
475fb78f AS |
5699 | return -EINVAL; |
5700 | } | |
5701 | ||
f1bca824 AS |
5702 | if (e == BRANCH) |
5703 | /* mark branch target for state pruning */ | |
5d839021 | 5704 | init_explored_state(env, w); |
f1bca824 | 5705 | |
475fb78f AS |
5706 | if (insn_state[w] == 0) { |
5707 | /* tree-edge */ | |
5708 | insn_state[t] = DISCOVERED | e; | |
5709 | insn_state[w] = DISCOVERED; | |
7df737e9 | 5710 | if (env->cfg.cur_stack >= env->prog->len) |
475fb78f | 5711 | return -E2BIG; |
7df737e9 | 5712 | insn_stack[env->cfg.cur_stack++] = w; |
475fb78f AS |
5713 | return 1; |
5714 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
d9762e84 MKL |
5715 | verbose_linfo(env, t, "%d: ", t); |
5716 | verbose_linfo(env, w, "%d: ", w); | |
61bd5218 | 5717 | verbose(env, "back-edge from insn %d to %d\n", t, w); |
475fb78f AS |
5718 | return -EINVAL; |
5719 | } else if (insn_state[w] == EXPLORED) { | |
5720 | /* forward- or cross-edge */ | |
5721 | insn_state[t] = DISCOVERED | e; | |
5722 | } else { | |
61bd5218 | 5723 | verbose(env, "insn state internal bug\n"); |
475fb78f AS |
5724 | return -EFAULT; |
5725 | } | |
5726 | return 0; | |
5727 | } | |
5728 | ||
5729 | /* non-recursive depth-first-search to detect loops in BPF program | |
5730 | * loop == back-edge in directed graph | |
5731 | */ | |
58e2af8b | 5732 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
5733 | { |
5734 | struct bpf_insn *insns = env->prog->insnsi; | |
5735 | int insn_cnt = env->prog->len; | |
7df737e9 | 5736 | int *insn_stack, *insn_state; |
475fb78f AS |
5737 | int ret = 0; |
5738 | int i, t; | |
5739 | ||
7df737e9 | 5740 | insn_state = env->cfg.insn_state = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f AS |
5741 | if (!insn_state) |
5742 | return -ENOMEM; | |
5743 | ||
7df737e9 | 5744 | insn_stack = env->cfg.insn_stack = kvcalloc(insn_cnt, sizeof(int), GFP_KERNEL); |
475fb78f | 5745 | if (!insn_stack) { |
71dde681 | 5746 | kvfree(insn_state); |
475fb78f AS |
5747 | return -ENOMEM; |
5748 | } | |
5749 | ||
5750 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
5751 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
7df737e9 | 5752 | env->cfg.cur_stack = 1; |
475fb78f AS |
5753 | |
5754 | peek_stack: | |
7df737e9 | 5755 | if (env->cfg.cur_stack == 0) |
475fb78f | 5756 | goto check_state; |
7df737e9 | 5757 | t = insn_stack[env->cfg.cur_stack - 1]; |
475fb78f | 5758 | |
092ed096 JW |
5759 | if (BPF_CLASS(insns[t].code) == BPF_JMP || |
5760 | BPF_CLASS(insns[t].code) == BPF_JMP32) { | |
475fb78f AS |
5761 | u8 opcode = BPF_OP(insns[t].code); |
5762 | ||
5763 | if (opcode == BPF_EXIT) { | |
5764 | goto mark_explored; | |
5765 | } else if (opcode == BPF_CALL) { | |
5766 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
5767 | if (ret == 1) | |
5768 | goto peek_stack; | |
5769 | else if (ret < 0) | |
5770 | goto err_free; | |
07016151 | 5771 | if (t + 1 < insn_cnt) |
5d839021 | 5772 | init_explored_state(env, t + 1); |
cc8b0b92 | 5773 | if (insns[t].src_reg == BPF_PSEUDO_CALL) { |
5d839021 | 5774 | init_explored_state(env, t); |
cc8b0b92 AS |
5775 | ret = push_insn(t, t + insns[t].imm + 1, BRANCH, env); |
5776 | if (ret == 1) | |
5777 | goto peek_stack; | |
5778 | else if (ret < 0) | |
5779 | goto err_free; | |
5780 | } | |
475fb78f AS |
5781 | } else if (opcode == BPF_JA) { |
5782 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
5783 | ret = -EINVAL; | |
5784 | goto err_free; | |
5785 | } | |
5786 | /* unconditional jump with single edge */ | |
5787 | ret = push_insn(t, t + insns[t].off + 1, | |
5788 | FALLTHROUGH, env); | |
5789 | if (ret == 1) | |
5790 | goto peek_stack; | |
5791 | else if (ret < 0) | |
5792 | goto err_free; | |
f1bca824 AS |
5793 | /* tell verifier to check for equivalent states |
5794 | * after every call and jump | |
5795 | */ | |
c3de6317 | 5796 | if (t + 1 < insn_cnt) |
5d839021 | 5797 | init_explored_state(env, t + 1); |
475fb78f AS |
5798 | } else { |
5799 | /* conditional jump with two edges */ | |
5d839021 | 5800 | init_explored_state(env, t); |
475fb78f AS |
5801 | ret = push_insn(t, t + 1, FALLTHROUGH, env); |
5802 | if (ret == 1) | |
5803 | goto peek_stack; | |
5804 | else if (ret < 0) | |
5805 | goto err_free; | |
5806 | ||
5807 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
5808 | if (ret == 1) | |
5809 | goto peek_stack; | |
5810 | else if (ret < 0) | |
5811 | goto err_free; | |
5812 | } | |
5813 | } else { | |
5814 | /* all other non-branch instructions with single | |
5815 | * fall-through edge | |
5816 | */ | |
5817 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
5818 | if (ret == 1) | |
5819 | goto peek_stack; | |
5820 | else if (ret < 0) | |
5821 | goto err_free; | |
5822 | } | |
5823 | ||
5824 | mark_explored: | |
5825 | insn_state[t] = EXPLORED; | |
7df737e9 | 5826 | if (env->cfg.cur_stack-- <= 0) { |
61bd5218 | 5827 | verbose(env, "pop stack internal bug\n"); |
475fb78f AS |
5828 | ret = -EFAULT; |
5829 | goto err_free; | |
5830 | } | |
5831 | goto peek_stack; | |
5832 | ||
5833 | check_state: | |
5834 | for (i = 0; i < insn_cnt; i++) { | |
5835 | if (insn_state[i] != EXPLORED) { | |
61bd5218 | 5836 | verbose(env, "unreachable insn %d\n", i); |
475fb78f AS |
5837 | ret = -EINVAL; |
5838 | goto err_free; | |
5839 | } | |
5840 | } | |
5841 | ret = 0; /* cfg looks good */ | |
5842 | ||
5843 | err_free: | |
71dde681 AS |
5844 | kvfree(insn_state); |
5845 | kvfree(insn_stack); | |
7df737e9 | 5846 | env->cfg.insn_state = env->cfg.insn_stack = NULL; |
475fb78f AS |
5847 | return ret; |
5848 | } | |
5849 | ||
838e9690 YS |
5850 | /* The minimum supported BTF func info size */ |
5851 | #define MIN_BPF_FUNCINFO_SIZE 8 | |
5852 | #define MAX_FUNCINFO_REC_SIZE 252 | |
5853 | ||
c454a46b MKL |
5854 | static int check_btf_func(struct bpf_verifier_env *env, |
5855 | const union bpf_attr *attr, | |
5856 | union bpf_attr __user *uattr) | |
838e9690 | 5857 | { |
d0b2818e | 5858 | u32 i, nfuncs, urec_size, min_size; |
838e9690 | 5859 | u32 krec_size = sizeof(struct bpf_func_info); |
c454a46b | 5860 | struct bpf_func_info *krecord; |
838e9690 | 5861 | const struct btf_type *type; |
c454a46b MKL |
5862 | struct bpf_prog *prog; |
5863 | const struct btf *btf; | |
838e9690 | 5864 | void __user *urecord; |
d0b2818e | 5865 | u32 prev_offset = 0; |
838e9690 YS |
5866 | int ret = 0; |
5867 | ||
5868 | nfuncs = attr->func_info_cnt; | |
5869 | if (!nfuncs) | |
5870 | return 0; | |
5871 | ||
5872 | if (nfuncs != env->subprog_cnt) { | |
5873 | verbose(env, "number of funcs in func_info doesn't match number of subprogs\n"); | |
5874 | return -EINVAL; | |
5875 | } | |
5876 | ||
5877 | urec_size = attr->func_info_rec_size; | |
5878 | if (urec_size < MIN_BPF_FUNCINFO_SIZE || | |
5879 | urec_size > MAX_FUNCINFO_REC_SIZE || | |
5880 | urec_size % sizeof(u32)) { | |
5881 | verbose(env, "invalid func info rec size %u\n", urec_size); | |
5882 | return -EINVAL; | |
5883 | } | |
5884 | ||
c454a46b MKL |
5885 | prog = env->prog; |
5886 | btf = prog->aux->btf; | |
838e9690 YS |
5887 | |
5888 | urecord = u64_to_user_ptr(attr->func_info); | |
5889 | min_size = min_t(u32, krec_size, urec_size); | |
5890 | ||
ba64e7d8 | 5891 | krecord = kvcalloc(nfuncs, krec_size, GFP_KERNEL | __GFP_NOWARN); |
c454a46b MKL |
5892 | if (!krecord) |
5893 | return -ENOMEM; | |
ba64e7d8 | 5894 | |
838e9690 YS |
5895 | for (i = 0; i < nfuncs; i++) { |
5896 | ret = bpf_check_uarg_tail_zero(urecord, krec_size, urec_size); | |
5897 | if (ret) { | |
5898 | if (ret == -E2BIG) { | |
5899 | verbose(env, "nonzero tailing record in func info"); | |
5900 | /* set the size kernel expects so loader can zero | |
5901 | * out the rest of the record. | |
5902 | */ | |
5903 | if (put_user(min_size, &uattr->func_info_rec_size)) | |
5904 | ret = -EFAULT; | |
5905 | } | |
c454a46b | 5906 | goto err_free; |
838e9690 YS |
5907 | } |
5908 | ||
ba64e7d8 | 5909 | if (copy_from_user(&krecord[i], urecord, min_size)) { |
838e9690 | 5910 | ret = -EFAULT; |
c454a46b | 5911 | goto err_free; |
838e9690 YS |
5912 | } |
5913 | ||
d30d42e0 | 5914 | /* check insn_off */ |
838e9690 | 5915 | if (i == 0) { |
d30d42e0 | 5916 | if (krecord[i].insn_off) { |
838e9690 | 5917 | verbose(env, |
d30d42e0 MKL |
5918 | "nonzero insn_off %u for the first func info record", |
5919 | krecord[i].insn_off); | |
838e9690 | 5920 | ret = -EINVAL; |
c454a46b | 5921 | goto err_free; |
838e9690 | 5922 | } |
d30d42e0 | 5923 | } else if (krecord[i].insn_off <= prev_offset) { |
838e9690 YS |
5924 | verbose(env, |
5925 | "same or smaller insn offset (%u) than previous func info record (%u)", | |
d30d42e0 | 5926 | krecord[i].insn_off, prev_offset); |
838e9690 | 5927 | ret = -EINVAL; |
c454a46b | 5928 | goto err_free; |
838e9690 YS |
5929 | } |
5930 | ||
d30d42e0 | 5931 | if (env->subprog_info[i].start != krecord[i].insn_off) { |
838e9690 YS |
5932 | verbose(env, "func_info BTF section doesn't match subprog layout in BPF program\n"); |
5933 | ret = -EINVAL; | |
c454a46b | 5934 | goto err_free; |
838e9690 YS |
5935 | } |
5936 | ||
5937 | /* check type_id */ | |
ba64e7d8 | 5938 | type = btf_type_by_id(btf, krecord[i].type_id); |
838e9690 YS |
5939 | if (!type || BTF_INFO_KIND(type->info) != BTF_KIND_FUNC) { |
5940 | verbose(env, "invalid type id %d in func info", | |
ba64e7d8 | 5941 | krecord[i].type_id); |
838e9690 | 5942 | ret = -EINVAL; |
c454a46b | 5943 | goto err_free; |
838e9690 YS |
5944 | } |
5945 | ||
d30d42e0 | 5946 | prev_offset = krecord[i].insn_off; |
838e9690 YS |
5947 | urecord += urec_size; |
5948 | } | |
5949 | ||
ba64e7d8 YS |
5950 | prog->aux->func_info = krecord; |
5951 | prog->aux->func_info_cnt = nfuncs; | |
838e9690 YS |
5952 | return 0; |
5953 | ||
c454a46b | 5954 | err_free: |
ba64e7d8 | 5955 | kvfree(krecord); |
838e9690 YS |
5956 | return ret; |
5957 | } | |
5958 | ||
ba64e7d8 YS |
5959 | static void adjust_btf_func(struct bpf_verifier_env *env) |
5960 | { | |
5961 | int i; | |
5962 | ||
5963 | if (!env->prog->aux->func_info) | |
5964 | return; | |
5965 | ||
5966 | for (i = 0; i < env->subprog_cnt; i++) | |
d30d42e0 | 5967 | env->prog->aux->func_info[i].insn_off = env->subprog_info[i].start; |
ba64e7d8 YS |
5968 | } |
5969 | ||
c454a46b MKL |
5970 | #define MIN_BPF_LINEINFO_SIZE (offsetof(struct bpf_line_info, line_col) + \ |
5971 | sizeof(((struct bpf_line_info *)(0))->line_col)) | |
5972 | #define MAX_LINEINFO_REC_SIZE MAX_FUNCINFO_REC_SIZE | |
5973 | ||
5974 | static int check_btf_line(struct bpf_verifier_env *env, | |
5975 | const union bpf_attr *attr, | |
5976 | union bpf_attr __user *uattr) | |
5977 | { | |
5978 | u32 i, s, nr_linfo, ncopy, expected_size, rec_size, prev_offset = 0; | |
5979 | struct bpf_subprog_info *sub; | |
5980 | struct bpf_line_info *linfo; | |
5981 | struct bpf_prog *prog; | |
5982 | const struct btf *btf; | |
5983 | void __user *ulinfo; | |
5984 | int err; | |
5985 | ||
5986 | nr_linfo = attr->line_info_cnt; | |
5987 | if (!nr_linfo) | |
5988 | return 0; | |
5989 | ||
5990 | rec_size = attr->line_info_rec_size; | |
5991 | if (rec_size < MIN_BPF_LINEINFO_SIZE || | |
5992 | rec_size > MAX_LINEINFO_REC_SIZE || | |
5993 | rec_size & (sizeof(u32) - 1)) | |
5994 | return -EINVAL; | |
5995 | ||
5996 | /* Need to zero it in case the userspace may | |
5997 | * pass in a smaller bpf_line_info object. | |
5998 | */ | |
5999 | linfo = kvcalloc(nr_linfo, sizeof(struct bpf_line_info), | |
6000 | GFP_KERNEL | __GFP_NOWARN); | |
6001 | if (!linfo) | |
6002 | return -ENOMEM; | |
6003 | ||
6004 | prog = env->prog; | |
6005 | btf = prog->aux->btf; | |
6006 | ||
6007 | s = 0; | |
6008 | sub = env->subprog_info; | |
6009 | ulinfo = u64_to_user_ptr(attr->line_info); | |
6010 | expected_size = sizeof(struct bpf_line_info); | |
6011 | ncopy = min_t(u32, expected_size, rec_size); | |
6012 | for (i = 0; i < nr_linfo; i++) { | |
6013 | err = bpf_check_uarg_tail_zero(ulinfo, expected_size, rec_size); | |
6014 | if (err) { | |
6015 | if (err == -E2BIG) { | |
6016 | verbose(env, "nonzero tailing record in line_info"); | |
6017 | if (put_user(expected_size, | |
6018 | &uattr->line_info_rec_size)) | |
6019 | err = -EFAULT; | |
6020 | } | |
6021 | goto err_free; | |
6022 | } | |
6023 | ||
6024 | if (copy_from_user(&linfo[i], ulinfo, ncopy)) { | |
6025 | err = -EFAULT; | |
6026 | goto err_free; | |
6027 | } | |
6028 | ||
6029 | /* | |
6030 | * Check insn_off to ensure | |
6031 | * 1) strictly increasing AND | |
6032 | * 2) bounded by prog->len | |
6033 | * | |
6034 | * The linfo[0].insn_off == 0 check logically falls into | |
6035 | * the later "missing bpf_line_info for func..." case | |
6036 | * because the first linfo[0].insn_off must be the | |
6037 | * first sub also and the first sub must have | |
6038 | * subprog_info[0].start == 0. | |
6039 | */ | |
6040 | if ((i && linfo[i].insn_off <= prev_offset) || | |
6041 | linfo[i].insn_off >= prog->len) { | |
6042 | verbose(env, "Invalid line_info[%u].insn_off:%u (prev_offset:%u prog->len:%u)\n", | |
6043 | i, linfo[i].insn_off, prev_offset, | |
6044 | prog->len); | |
6045 | err = -EINVAL; | |
6046 | goto err_free; | |
6047 | } | |
6048 | ||
fdbaa0be MKL |
6049 | if (!prog->insnsi[linfo[i].insn_off].code) { |
6050 | verbose(env, | |
6051 | "Invalid insn code at line_info[%u].insn_off\n", | |
6052 | i); | |
6053 | err = -EINVAL; | |
6054 | goto err_free; | |
6055 | } | |
6056 | ||
23127b33 MKL |
6057 | if (!btf_name_by_offset(btf, linfo[i].line_off) || |
6058 | !btf_name_by_offset(btf, linfo[i].file_name_off)) { | |
c454a46b MKL |
6059 | verbose(env, "Invalid line_info[%u].line_off or .file_name_off\n", i); |
6060 | err = -EINVAL; | |
6061 | goto err_free; | |
6062 | } | |
6063 | ||
6064 | if (s != env->subprog_cnt) { | |
6065 | if (linfo[i].insn_off == sub[s].start) { | |
6066 | sub[s].linfo_idx = i; | |
6067 | s++; | |
6068 | } else if (sub[s].start < linfo[i].insn_off) { | |
6069 | verbose(env, "missing bpf_line_info for func#%u\n", s); | |
6070 | err = -EINVAL; | |
6071 | goto err_free; | |
6072 | } | |
6073 | } | |
6074 | ||
6075 | prev_offset = linfo[i].insn_off; | |
6076 | ulinfo += rec_size; | |
6077 | } | |
6078 | ||
6079 | if (s != env->subprog_cnt) { | |
6080 | verbose(env, "missing bpf_line_info for %u funcs starting from func#%u\n", | |
6081 | env->subprog_cnt - s, s); | |
6082 | err = -EINVAL; | |
6083 | goto err_free; | |
6084 | } | |
6085 | ||
6086 | prog->aux->linfo = linfo; | |
6087 | prog->aux->nr_linfo = nr_linfo; | |
6088 | ||
6089 | return 0; | |
6090 | ||
6091 | err_free: | |
6092 | kvfree(linfo); | |
6093 | return err; | |
6094 | } | |
6095 | ||
6096 | static int check_btf_info(struct bpf_verifier_env *env, | |
6097 | const union bpf_attr *attr, | |
6098 | union bpf_attr __user *uattr) | |
6099 | { | |
6100 | struct btf *btf; | |
6101 | int err; | |
6102 | ||
6103 | if (!attr->func_info_cnt && !attr->line_info_cnt) | |
6104 | return 0; | |
6105 | ||
6106 | btf = btf_get_by_fd(attr->prog_btf_fd); | |
6107 | if (IS_ERR(btf)) | |
6108 | return PTR_ERR(btf); | |
6109 | env->prog->aux->btf = btf; | |
6110 | ||
6111 | err = check_btf_func(env, attr, uattr); | |
6112 | if (err) | |
6113 | return err; | |
6114 | ||
6115 | err = check_btf_line(env, attr, uattr); | |
6116 | if (err) | |
6117 | return err; | |
6118 | ||
6119 | return 0; | |
ba64e7d8 YS |
6120 | } |
6121 | ||
f1174f77 EC |
6122 | /* check %cur's range satisfies %old's */ |
6123 | static bool range_within(struct bpf_reg_state *old, | |
6124 | struct bpf_reg_state *cur) | |
6125 | { | |
b03c9f9f EC |
6126 | return old->umin_value <= cur->umin_value && |
6127 | old->umax_value >= cur->umax_value && | |
6128 | old->smin_value <= cur->smin_value && | |
6129 | old->smax_value >= cur->smax_value; | |
f1174f77 EC |
6130 | } |
6131 | ||
6132 | /* Maximum number of register states that can exist at once */ | |
6133 | #define ID_MAP_SIZE (MAX_BPF_REG + MAX_BPF_STACK / BPF_REG_SIZE) | |
6134 | struct idpair { | |
6135 | u32 old; | |
6136 | u32 cur; | |
6137 | }; | |
6138 | ||
6139 | /* If in the old state two registers had the same id, then they need to have | |
6140 | * the same id in the new state as well. But that id could be different from | |
6141 | * the old state, so we need to track the mapping from old to new ids. | |
6142 | * Once we have seen that, say, a reg with old id 5 had new id 9, any subsequent | |
6143 | * regs with old id 5 must also have new id 9 for the new state to be safe. But | |
6144 | * regs with a different old id could still have new id 9, we don't care about | |
6145 | * that. | |
6146 | * So we look through our idmap to see if this old id has been seen before. If | |
6147 | * so, we require the new id to match; otherwise, we add the id pair to the map. | |
969bf05e | 6148 | */ |
f1174f77 | 6149 | static bool check_ids(u32 old_id, u32 cur_id, struct idpair *idmap) |
969bf05e | 6150 | { |
f1174f77 | 6151 | unsigned int i; |
969bf05e | 6152 | |
f1174f77 EC |
6153 | for (i = 0; i < ID_MAP_SIZE; i++) { |
6154 | if (!idmap[i].old) { | |
6155 | /* Reached an empty slot; haven't seen this id before */ | |
6156 | idmap[i].old = old_id; | |
6157 | idmap[i].cur = cur_id; | |
6158 | return true; | |
6159 | } | |
6160 | if (idmap[i].old == old_id) | |
6161 | return idmap[i].cur == cur_id; | |
6162 | } | |
6163 | /* We ran out of idmap slots, which should be impossible */ | |
6164 | WARN_ON_ONCE(1); | |
6165 | return false; | |
6166 | } | |
6167 | ||
9242b5f5 AS |
6168 | static void clean_func_state(struct bpf_verifier_env *env, |
6169 | struct bpf_func_state *st) | |
6170 | { | |
6171 | enum bpf_reg_liveness live; | |
6172 | int i, j; | |
6173 | ||
6174 | for (i = 0; i < BPF_REG_FP; i++) { | |
6175 | live = st->regs[i].live; | |
6176 | /* liveness must not touch this register anymore */ | |
6177 | st->regs[i].live |= REG_LIVE_DONE; | |
6178 | if (!(live & REG_LIVE_READ)) | |
6179 | /* since the register is unused, clear its state | |
6180 | * to make further comparison simpler | |
6181 | */ | |
6182 | __mark_reg_not_init(&st->regs[i]); | |
6183 | } | |
6184 | ||
6185 | for (i = 0; i < st->allocated_stack / BPF_REG_SIZE; i++) { | |
6186 | live = st->stack[i].spilled_ptr.live; | |
6187 | /* liveness must not touch this stack slot anymore */ | |
6188 | st->stack[i].spilled_ptr.live |= REG_LIVE_DONE; | |
6189 | if (!(live & REG_LIVE_READ)) { | |
6190 | __mark_reg_not_init(&st->stack[i].spilled_ptr); | |
6191 | for (j = 0; j < BPF_REG_SIZE; j++) | |
6192 | st->stack[i].slot_type[j] = STACK_INVALID; | |
6193 | } | |
6194 | } | |
6195 | } | |
6196 | ||
6197 | static void clean_verifier_state(struct bpf_verifier_env *env, | |
6198 | struct bpf_verifier_state *st) | |
6199 | { | |
6200 | int i; | |
6201 | ||
6202 | if (st->frame[0]->regs[0].live & REG_LIVE_DONE) | |
6203 | /* all regs in this state in all frames were already marked */ | |
6204 | return; | |
6205 | ||
6206 | for (i = 0; i <= st->curframe; i++) | |
6207 | clean_func_state(env, st->frame[i]); | |
6208 | } | |
6209 | ||
6210 | /* the parentage chains form a tree. | |
6211 | * the verifier states are added to state lists at given insn and | |
6212 | * pushed into state stack for future exploration. | |
6213 | * when the verifier reaches bpf_exit insn some of the verifer states | |
6214 | * stored in the state lists have their final liveness state already, | |
6215 | * but a lot of states will get revised from liveness point of view when | |
6216 | * the verifier explores other branches. | |
6217 | * Example: | |
6218 | * 1: r0 = 1 | |
6219 | * 2: if r1 == 100 goto pc+1 | |
6220 | * 3: r0 = 2 | |
6221 | * 4: exit | |
6222 | * when the verifier reaches exit insn the register r0 in the state list of | |
6223 | * insn 2 will be seen as !REG_LIVE_READ. Then the verifier pops the other_branch | |
6224 | * of insn 2 and goes exploring further. At the insn 4 it will walk the | |
6225 | * parentage chain from insn 4 into insn 2 and will mark r0 as REG_LIVE_READ. | |
6226 | * | |
6227 | * Since the verifier pushes the branch states as it sees them while exploring | |
6228 | * the program the condition of walking the branch instruction for the second | |
6229 | * time means that all states below this branch were already explored and | |
6230 | * their final liveness markes are already propagated. | |
6231 | * Hence when the verifier completes the search of state list in is_state_visited() | |
6232 | * we can call this clean_live_states() function to mark all liveness states | |
6233 | * as REG_LIVE_DONE to indicate that 'parent' pointers of 'struct bpf_reg_state' | |
6234 | * will not be used. | |
6235 | * This function also clears the registers and stack for states that !READ | |
6236 | * to simplify state merging. | |
6237 | * | |
6238 | * Important note here that walking the same branch instruction in the callee | |
6239 | * doesn't meant that the states are DONE. The verifier has to compare | |
6240 | * the callsites | |
6241 | */ | |
6242 | static void clean_live_states(struct bpf_verifier_env *env, int insn, | |
6243 | struct bpf_verifier_state *cur) | |
6244 | { | |
6245 | struct bpf_verifier_state_list *sl; | |
6246 | int i; | |
6247 | ||
5d839021 | 6248 | sl = *explored_state(env, insn); |
a8f500af | 6249 | while (sl) { |
dc2a4ebc AS |
6250 | if (sl->state.insn_idx != insn || |
6251 | sl->state.curframe != cur->curframe) | |
9242b5f5 AS |
6252 | goto next; |
6253 | for (i = 0; i <= cur->curframe; i++) | |
6254 | if (sl->state.frame[i]->callsite != cur->frame[i]->callsite) | |
6255 | goto next; | |
6256 | clean_verifier_state(env, &sl->state); | |
6257 | next: | |
6258 | sl = sl->next; | |
6259 | } | |
6260 | } | |
6261 | ||
f1174f77 | 6262 | /* Returns true if (rold safe implies rcur safe) */ |
1b688a19 EC |
6263 | static bool regsafe(struct bpf_reg_state *rold, struct bpf_reg_state *rcur, |
6264 | struct idpair *idmap) | |
f1174f77 | 6265 | { |
f4d7e40a AS |
6266 | bool equal; |
6267 | ||
dc503a8a EC |
6268 | if (!(rold->live & REG_LIVE_READ)) |
6269 | /* explored state didn't use this */ | |
6270 | return true; | |
6271 | ||
679c782d | 6272 | equal = memcmp(rold, rcur, offsetof(struct bpf_reg_state, parent)) == 0; |
f4d7e40a AS |
6273 | |
6274 | if (rold->type == PTR_TO_STACK) | |
6275 | /* two stack pointers are equal only if they're pointing to | |
6276 | * the same stack frame, since fp-8 in foo != fp-8 in bar | |
6277 | */ | |
6278 | return equal && rold->frameno == rcur->frameno; | |
6279 | ||
6280 | if (equal) | |
969bf05e AS |
6281 | return true; |
6282 | ||
f1174f77 EC |
6283 | if (rold->type == NOT_INIT) |
6284 | /* explored state can't have used this */ | |
969bf05e | 6285 | return true; |
f1174f77 EC |
6286 | if (rcur->type == NOT_INIT) |
6287 | return false; | |
6288 | switch (rold->type) { | |
6289 | case SCALAR_VALUE: | |
6290 | if (rcur->type == SCALAR_VALUE) { | |
6291 | /* new val must satisfy old val knowledge */ | |
6292 | return range_within(rold, rcur) && | |
6293 | tnum_in(rold->var_off, rcur->var_off); | |
6294 | } else { | |
179d1c56 JH |
6295 | /* We're trying to use a pointer in place of a scalar. |
6296 | * Even if the scalar was unbounded, this could lead to | |
6297 | * pointer leaks because scalars are allowed to leak | |
6298 | * while pointers are not. We could make this safe in | |
6299 | * special cases if root is calling us, but it's | |
6300 | * probably not worth the hassle. | |
f1174f77 | 6301 | */ |
179d1c56 | 6302 | return false; |
f1174f77 EC |
6303 | } |
6304 | case PTR_TO_MAP_VALUE: | |
1b688a19 EC |
6305 | /* If the new min/max/var_off satisfy the old ones and |
6306 | * everything else matches, we are OK. | |
d83525ca AS |
6307 | * 'id' is not compared, since it's only used for maps with |
6308 | * bpf_spin_lock inside map element and in such cases if | |
6309 | * the rest of the prog is valid for one map element then | |
6310 | * it's valid for all map elements regardless of the key | |
6311 | * used in bpf_map_lookup() | |
1b688a19 EC |
6312 | */ |
6313 | return memcmp(rold, rcur, offsetof(struct bpf_reg_state, id)) == 0 && | |
6314 | range_within(rold, rcur) && | |
6315 | tnum_in(rold->var_off, rcur->var_off); | |
f1174f77 EC |
6316 | case PTR_TO_MAP_VALUE_OR_NULL: |
6317 | /* a PTR_TO_MAP_VALUE could be safe to use as a | |
6318 | * PTR_TO_MAP_VALUE_OR_NULL into the same map. | |
6319 | * However, if the old PTR_TO_MAP_VALUE_OR_NULL then got NULL- | |
6320 | * checked, doing so could have affected others with the same | |
6321 | * id, and we can't check for that because we lost the id when | |
6322 | * we converted to a PTR_TO_MAP_VALUE. | |
6323 | */ | |
6324 | if (rcur->type != PTR_TO_MAP_VALUE_OR_NULL) | |
6325 | return false; | |
6326 | if (memcmp(rold, rcur, offsetof(struct bpf_reg_state, id))) | |
6327 | return false; | |
6328 | /* Check our ids match any regs they're supposed to */ | |
6329 | return check_ids(rold->id, rcur->id, idmap); | |
de8f3a83 | 6330 | case PTR_TO_PACKET_META: |
f1174f77 | 6331 | case PTR_TO_PACKET: |
de8f3a83 | 6332 | if (rcur->type != rold->type) |
f1174f77 EC |
6333 | return false; |
6334 | /* We must have at least as much range as the old ptr | |
6335 | * did, so that any accesses which were safe before are | |
6336 | * still safe. This is true even if old range < old off, | |
6337 | * since someone could have accessed through (ptr - k), or | |
6338 | * even done ptr -= k in a register, to get a safe access. | |
6339 | */ | |
6340 | if (rold->range > rcur->range) | |
6341 | return false; | |
6342 | /* If the offsets don't match, we can't trust our alignment; | |
6343 | * nor can we be sure that we won't fall out of range. | |
6344 | */ | |
6345 | if (rold->off != rcur->off) | |
6346 | return false; | |
6347 | /* id relations must be preserved */ | |
6348 | if (rold->id && !check_ids(rold->id, rcur->id, idmap)) | |
6349 | return false; | |
6350 | /* new val must satisfy old val knowledge */ | |
6351 | return range_within(rold, rcur) && | |
6352 | tnum_in(rold->var_off, rcur->var_off); | |
6353 | case PTR_TO_CTX: | |
6354 | case CONST_PTR_TO_MAP: | |
f1174f77 | 6355 | case PTR_TO_PACKET_END: |
d58e468b | 6356 | case PTR_TO_FLOW_KEYS: |
c64b7983 JS |
6357 | case PTR_TO_SOCKET: |
6358 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
6359 | case PTR_TO_SOCK_COMMON: |
6360 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
6361 | case PTR_TO_TCP_SOCK: |
6362 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 6363 | case PTR_TO_XDP_SOCK: |
f1174f77 EC |
6364 | /* Only valid matches are exact, which memcmp() above |
6365 | * would have accepted | |
6366 | */ | |
6367 | default: | |
6368 | /* Don't know what's going on, just say it's not safe */ | |
6369 | return false; | |
6370 | } | |
969bf05e | 6371 | |
f1174f77 EC |
6372 | /* Shouldn't get here; if we do, say it's not safe */ |
6373 | WARN_ON_ONCE(1); | |
969bf05e AS |
6374 | return false; |
6375 | } | |
6376 | ||
f4d7e40a AS |
6377 | static bool stacksafe(struct bpf_func_state *old, |
6378 | struct bpf_func_state *cur, | |
638f5b90 AS |
6379 | struct idpair *idmap) |
6380 | { | |
6381 | int i, spi; | |
6382 | ||
638f5b90 AS |
6383 | /* walk slots of the explored stack and ignore any additional |
6384 | * slots in the current stack, since explored(safe) state | |
6385 | * didn't use them | |
6386 | */ | |
6387 | for (i = 0; i < old->allocated_stack; i++) { | |
6388 | spi = i / BPF_REG_SIZE; | |
6389 | ||
b233920c AS |
6390 | if (!(old->stack[spi].spilled_ptr.live & REG_LIVE_READ)) { |
6391 | i += BPF_REG_SIZE - 1; | |
cc2b14d5 | 6392 | /* explored state didn't use this */ |
fd05e57b | 6393 | continue; |
b233920c | 6394 | } |
cc2b14d5 | 6395 | |
638f5b90 AS |
6396 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_INVALID) |
6397 | continue; | |
19e2dbb7 AS |
6398 | |
6399 | /* explored stack has more populated slots than current stack | |
6400 | * and these slots were used | |
6401 | */ | |
6402 | if (i >= cur->allocated_stack) | |
6403 | return false; | |
6404 | ||
cc2b14d5 AS |
6405 | /* if old state was safe with misc data in the stack |
6406 | * it will be safe with zero-initialized stack. | |
6407 | * The opposite is not true | |
6408 | */ | |
6409 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_MISC && | |
6410 | cur->stack[spi].slot_type[i % BPF_REG_SIZE] == STACK_ZERO) | |
6411 | continue; | |
638f5b90 AS |
6412 | if (old->stack[spi].slot_type[i % BPF_REG_SIZE] != |
6413 | cur->stack[spi].slot_type[i % BPF_REG_SIZE]) | |
6414 | /* Ex: old explored (safe) state has STACK_SPILL in | |
6415 | * this stack slot, but current has has STACK_MISC -> | |
6416 | * this verifier states are not equivalent, | |
6417 | * return false to continue verification of this path | |
6418 | */ | |
6419 | return false; | |
6420 | if (i % BPF_REG_SIZE) | |
6421 | continue; | |
6422 | if (old->stack[spi].slot_type[0] != STACK_SPILL) | |
6423 | continue; | |
6424 | if (!regsafe(&old->stack[spi].spilled_ptr, | |
6425 | &cur->stack[spi].spilled_ptr, | |
6426 | idmap)) | |
6427 | /* when explored and current stack slot are both storing | |
6428 | * spilled registers, check that stored pointers types | |
6429 | * are the same as well. | |
6430 | * Ex: explored safe path could have stored | |
6431 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -8} | |
6432 | * but current path has stored: | |
6433 | * (bpf_reg_state) {.type = PTR_TO_STACK, .off = -16} | |
6434 | * such verifier states are not equivalent. | |
6435 | * return false to continue verification of this path | |
6436 | */ | |
6437 | return false; | |
6438 | } | |
6439 | return true; | |
6440 | } | |
6441 | ||
fd978bf7 JS |
6442 | static bool refsafe(struct bpf_func_state *old, struct bpf_func_state *cur) |
6443 | { | |
6444 | if (old->acquired_refs != cur->acquired_refs) | |
6445 | return false; | |
6446 | return !memcmp(old->refs, cur->refs, | |
6447 | sizeof(*old->refs) * old->acquired_refs); | |
6448 | } | |
6449 | ||
f1bca824 AS |
6450 | /* compare two verifier states |
6451 | * | |
6452 | * all states stored in state_list are known to be valid, since | |
6453 | * verifier reached 'bpf_exit' instruction through them | |
6454 | * | |
6455 | * this function is called when verifier exploring different branches of | |
6456 | * execution popped from the state stack. If it sees an old state that has | |
6457 | * more strict register state and more strict stack state then this execution | |
6458 | * branch doesn't need to be explored further, since verifier already | |
6459 | * concluded that more strict state leads to valid finish. | |
6460 | * | |
6461 | * Therefore two states are equivalent if register state is more conservative | |
6462 | * and explored stack state is more conservative than the current one. | |
6463 | * Example: | |
6464 | * explored current | |
6465 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
6466 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
6467 | * | |
6468 | * In other words if current stack state (one being explored) has more | |
6469 | * valid slots than old one that already passed validation, it means | |
6470 | * the verifier can stop exploring and conclude that current state is valid too | |
6471 | * | |
6472 | * Similarly with registers. If explored state has register type as invalid | |
6473 | * whereas register type in current state is meaningful, it means that | |
6474 | * the current state will reach 'bpf_exit' instruction safely | |
6475 | */ | |
f4d7e40a AS |
6476 | static bool func_states_equal(struct bpf_func_state *old, |
6477 | struct bpf_func_state *cur) | |
f1bca824 | 6478 | { |
f1174f77 EC |
6479 | struct idpair *idmap; |
6480 | bool ret = false; | |
f1bca824 AS |
6481 | int i; |
6482 | ||
f1174f77 EC |
6483 | idmap = kcalloc(ID_MAP_SIZE, sizeof(struct idpair), GFP_KERNEL); |
6484 | /* If we failed to allocate the idmap, just say it's not safe */ | |
6485 | if (!idmap) | |
1a0dc1ac | 6486 | return false; |
f1174f77 EC |
6487 | |
6488 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1b688a19 | 6489 | if (!regsafe(&old->regs[i], &cur->regs[i], idmap)) |
f1174f77 | 6490 | goto out_free; |
f1bca824 AS |
6491 | } |
6492 | ||
638f5b90 AS |
6493 | if (!stacksafe(old, cur, idmap)) |
6494 | goto out_free; | |
fd978bf7 JS |
6495 | |
6496 | if (!refsafe(old, cur)) | |
6497 | goto out_free; | |
f1174f77 EC |
6498 | ret = true; |
6499 | out_free: | |
6500 | kfree(idmap); | |
6501 | return ret; | |
f1bca824 AS |
6502 | } |
6503 | ||
f4d7e40a AS |
6504 | static bool states_equal(struct bpf_verifier_env *env, |
6505 | struct bpf_verifier_state *old, | |
6506 | struct bpf_verifier_state *cur) | |
6507 | { | |
6508 | int i; | |
6509 | ||
6510 | if (old->curframe != cur->curframe) | |
6511 | return false; | |
6512 | ||
979d63d5 DB |
6513 | /* Verification state from speculative execution simulation |
6514 | * must never prune a non-speculative execution one. | |
6515 | */ | |
6516 | if (old->speculative && !cur->speculative) | |
6517 | return false; | |
6518 | ||
d83525ca AS |
6519 | if (old->active_spin_lock != cur->active_spin_lock) |
6520 | return false; | |
6521 | ||
f4d7e40a AS |
6522 | /* for states to be equal callsites have to be the same |
6523 | * and all frame states need to be equivalent | |
6524 | */ | |
6525 | for (i = 0; i <= old->curframe; i++) { | |
6526 | if (old->frame[i]->callsite != cur->frame[i]->callsite) | |
6527 | return false; | |
6528 | if (!func_states_equal(old->frame[i], cur->frame[i])) | |
6529 | return false; | |
6530 | } | |
6531 | return true; | |
6532 | } | |
6533 | ||
5327ed3d JW |
6534 | /* Return 0 if no propagation happened. Return negative error code if error |
6535 | * happened. Otherwise, return the propagated bit. | |
6536 | */ | |
55e7f3b5 JW |
6537 | static int propagate_liveness_reg(struct bpf_verifier_env *env, |
6538 | struct bpf_reg_state *reg, | |
6539 | struct bpf_reg_state *parent_reg) | |
6540 | { | |
5327ed3d JW |
6541 | u8 parent_flag = parent_reg->live & REG_LIVE_READ; |
6542 | u8 flag = reg->live & REG_LIVE_READ; | |
55e7f3b5 JW |
6543 | int err; |
6544 | ||
5327ed3d JW |
6545 | /* When comes here, read flags of PARENT_REG or REG could be any of |
6546 | * REG_LIVE_READ64, REG_LIVE_READ32, REG_LIVE_NONE. There is no need | |
6547 | * of propagation if PARENT_REG has strongest REG_LIVE_READ64. | |
6548 | */ | |
6549 | if (parent_flag == REG_LIVE_READ64 || | |
6550 | /* Or if there is no read flag from REG. */ | |
6551 | !flag || | |
6552 | /* Or if the read flag from REG is the same as PARENT_REG. */ | |
6553 | parent_flag == flag) | |
55e7f3b5 JW |
6554 | return 0; |
6555 | ||
5327ed3d | 6556 | err = mark_reg_read(env, reg, parent_reg, flag); |
55e7f3b5 JW |
6557 | if (err) |
6558 | return err; | |
6559 | ||
5327ed3d | 6560 | return flag; |
55e7f3b5 JW |
6561 | } |
6562 | ||
8e9cd9ce | 6563 | /* A write screens off any subsequent reads; but write marks come from the |
f4d7e40a AS |
6564 | * straight-line code between a state and its parent. When we arrive at an |
6565 | * equivalent state (jump target or such) we didn't arrive by the straight-line | |
6566 | * code, so read marks in the state must propagate to the parent regardless | |
6567 | * of the state's write marks. That's what 'parent == state->parent' comparison | |
679c782d | 6568 | * in mark_reg_read() is for. |
8e9cd9ce | 6569 | */ |
f4d7e40a AS |
6570 | static int propagate_liveness(struct bpf_verifier_env *env, |
6571 | const struct bpf_verifier_state *vstate, | |
6572 | struct bpf_verifier_state *vparent) | |
dc503a8a | 6573 | { |
3f8cafa4 | 6574 | struct bpf_reg_state *state_reg, *parent_reg; |
f4d7e40a | 6575 | struct bpf_func_state *state, *parent; |
3f8cafa4 | 6576 | int i, frame, err = 0; |
dc503a8a | 6577 | |
f4d7e40a AS |
6578 | if (vparent->curframe != vstate->curframe) { |
6579 | WARN(1, "propagate_live: parent frame %d current frame %d\n", | |
6580 | vparent->curframe, vstate->curframe); | |
6581 | return -EFAULT; | |
6582 | } | |
dc503a8a EC |
6583 | /* Propagate read liveness of registers... */ |
6584 | BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); | |
83d16312 | 6585 | for (frame = 0; frame <= vstate->curframe; frame++) { |
3f8cafa4 JW |
6586 | parent = vparent->frame[frame]; |
6587 | state = vstate->frame[frame]; | |
6588 | parent_reg = parent->regs; | |
6589 | state_reg = state->regs; | |
83d16312 JK |
6590 | /* We don't need to worry about FP liveness, it's read-only */ |
6591 | for (i = frame < vstate->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) { | |
55e7f3b5 JW |
6592 | err = propagate_liveness_reg(env, &state_reg[i], |
6593 | &parent_reg[i]); | |
5327ed3d | 6594 | if (err < 0) |
3f8cafa4 | 6595 | return err; |
5327ed3d JW |
6596 | if (err == REG_LIVE_READ64) |
6597 | mark_insn_zext(env, &parent_reg[i]); | |
dc503a8a | 6598 | } |
f4d7e40a | 6599 | |
1b04aee7 | 6600 | /* Propagate stack slots. */ |
f4d7e40a AS |
6601 | for (i = 0; i < state->allocated_stack / BPF_REG_SIZE && |
6602 | i < parent->allocated_stack / BPF_REG_SIZE; i++) { | |
3f8cafa4 JW |
6603 | parent_reg = &parent->stack[i].spilled_ptr; |
6604 | state_reg = &state->stack[i].spilled_ptr; | |
55e7f3b5 JW |
6605 | err = propagate_liveness_reg(env, state_reg, |
6606 | parent_reg); | |
5327ed3d | 6607 | if (err < 0) |
3f8cafa4 | 6608 | return err; |
dc503a8a EC |
6609 | } |
6610 | } | |
5327ed3d | 6611 | return 0; |
dc503a8a EC |
6612 | } |
6613 | ||
58e2af8b | 6614 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 6615 | { |
58e2af8b | 6616 | struct bpf_verifier_state_list *new_sl; |
9f4686c4 | 6617 | struct bpf_verifier_state_list *sl, **pprev; |
679c782d | 6618 | struct bpf_verifier_state *cur = env->cur_state, *new; |
ceefbc96 | 6619 | int i, j, err, states_cnt = 0; |
f1bca824 | 6620 | |
a8f500af | 6621 | if (!env->insn_aux_data[insn_idx].prune_point) |
f1bca824 AS |
6622 | /* this 'insn_idx' instruction wasn't marked, so we will not |
6623 | * be doing state search here | |
6624 | */ | |
6625 | return 0; | |
6626 | ||
a8f500af AS |
6627 | pprev = explored_state(env, insn_idx); |
6628 | sl = *pprev; | |
6629 | ||
9242b5f5 AS |
6630 | clean_live_states(env, insn_idx, cur); |
6631 | ||
a8f500af | 6632 | while (sl) { |
dc2a4ebc AS |
6633 | states_cnt++; |
6634 | if (sl->state.insn_idx != insn_idx) | |
6635 | goto next; | |
638f5b90 | 6636 | if (states_equal(env, &sl->state, cur)) { |
9f4686c4 | 6637 | sl->hit_cnt++; |
f1bca824 | 6638 | /* reached equivalent register/stack state, |
dc503a8a EC |
6639 | * prune the search. |
6640 | * Registers read by the continuation are read by us. | |
8e9cd9ce EC |
6641 | * If we have any write marks in env->cur_state, they |
6642 | * will prevent corresponding reads in the continuation | |
6643 | * from reaching our parent (an explored_state). Our | |
6644 | * own state will get the read marks recorded, but | |
6645 | * they'll be immediately forgotten as we're pruning | |
6646 | * this state and will pop a new one. | |
f1bca824 | 6647 | */ |
f4d7e40a AS |
6648 | err = propagate_liveness(env, &sl->state, cur); |
6649 | if (err) | |
6650 | return err; | |
f1bca824 | 6651 | return 1; |
dc503a8a | 6652 | } |
9f4686c4 AS |
6653 | sl->miss_cnt++; |
6654 | /* heuristic to determine whether this state is beneficial | |
6655 | * to keep checking from state equivalence point of view. | |
6656 | * Higher numbers increase max_states_per_insn and verification time, | |
6657 | * but do not meaningfully decrease insn_processed. | |
6658 | */ | |
6659 | if (sl->miss_cnt > sl->hit_cnt * 3 + 3) { | |
6660 | /* the state is unlikely to be useful. Remove it to | |
6661 | * speed up verification | |
6662 | */ | |
6663 | *pprev = sl->next; | |
6664 | if (sl->state.frame[0]->regs[0].live & REG_LIVE_DONE) { | |
6665 | free_verifier_state(&sl->state, false); | |
6666 | kfree(sl); | |
6667 | env->peak_states--; | |
6668 | } else { | |
6669 | /* cannot free this state, since parentage chain may | |
6670 | * walk it later. Add it for free_list instead to | |
6671 | * be freed at the end of verification | |
6672 | */ | |
6673 | sl->next = env->free_list; | |
6674 | env->free_list = sl; | |
6675 | } | |
6676 | sl = *pprev; | |
6677 | continue; | |
6678 | } | |
dc2a4ebc | 6679 | next: |
9f4686c4 AS |
6680 | pprev = &sl->next; |
6681 | sl = *pprev; | |
f1bca824 AS |
6682 | } |
6683 | ||
06ee7115 AS |
6684 | if (env->max_states_per_insn < states_cnt) |
6685 | env->max_states_per_insn = states_cnt; | |
6686 | ||
ceefbc96 AS |
6687 | if (!env->allow_ptr_leaks && states_cnt > BPF_COMPLEXITY_LIMIT_STATES) |
6688 | return 0; | |
6689 | ||
f1bca824 AS |
6690 | /* there were no equivalent states, remember current one. |
6691 | * technically the current state is not proven to be safe yet, | |
f4d7e40a AS |
6692 | * but it will either reach outer most bpf_exit (which means it's safe) |
6693 | * or it will be rejected. Since there are no loops, we won't be | |
6694 | * seeing this tuple (frame[0].callsite, frame[1].callsite, .. insn_idx) | |
6695 | * again on the way to bpf_exit | |
f1bca824 | 6696 | */ |
638f5b90 | 6697 | new_sl = kzalloc(sizeof(struct bpf_verifier_state_list), GFP_KERNEL); |
f1bca824 AS |
6698 | if (!new_sl) |
6699 | return -ENOMEM; | |
06ee7115 AS |
6700 | env->total_states++; |
6701 | env->peak_states++; | |
f1bca824 AS |
6702 | |
6703 | /* add new state to the head of linked list */ | |
679c782d EC |
6704 | new = &new_sl->state; |
6705 | err = copy_verifier_state(new, cur); | |
1969db47 | 6706 | if (err) { |
679c782d | 6707 | free_verifier_state(new, false); |
1969db47 AS |
6708 | kfree(new_sl); |
6709 | return err; | |
6710 | } | |
dc2a4ebc | 6711 | new->insn_idx = insn_idx; |
5d839021 AS |
6712 | new_sl->next = *explored_state(env, insn_idx); |
6713 | *explored_state(env, insn_idx) = new_sl; | |
7640ead9 JK |
6714 | /* connect new state to parentage chain. Current frame needs all |
6715 | * registers connected. Only r6 - r9 of the callers are alive (pushed | |
6716 | * to the stack implicitly by JITs) so in callers' frames connect just | |
6717 | * r6 - r9 as an optimization. Callers will have r1 - r5 connected to | |
6718 | * the state of the call instruction (with WRITTEN set), and r0 comes | |
6719 | * from callee with its full parentage chain, anyway. | |
6720 | */ | |
6721 | for (j = 0; j <= cur->curframe; j++) | |
6722 | for (i = j < cur->curframe ? BPF_REG_6 : 0; i < BPF_REG_FP; i++) | |
6723 | cur->frame[j]->regs[i].parent = &new->frame[j]->regs[i]; | |
8e9cd9ce EC |
6724 | /* clear write marks in current state: the writes we did are not writes |
6725 | * our child did, so they don't screen off its reads from us. | |
6726 | * (There are no read marks in current state, because reads always mark | |
6727 | * their parent and current state never has children yet. Only | |
6728 | * explored_states can get read marks.) | |
6729 | */ | |
dc503a8a | 6730 | for (i = 0; i < BPF_REG_FP; i++) |
f4d7e40a AS |
6731 | cur->frame[cur->curframe]->regs[i].live = REG_LIVE_NONE; |
6732 | ||
6733 | /* all stack frames are accessible from callee, clear them all */ | |
6734 | for (j = 0; j <= cur->curframe; j++) { | |
6735 | struct bpf_func_state *frame = cur->frame[j]; | |
679c782d | 6736 | struct bpf_func_state *newframe = new->frame[j]; |
f4d7e40a | 6737 | |
679c782d | 6738 | for (i = 0; i < frame->allocated_stack / BPF_REG_SIZE; i++) { |
cc2b14d5 | 6739 | frame->stack[i].spilled_ptr.live = REG_LIVE_NONE; |
679c782d EC |
6740 | frame->stack[i].spilled_ptr.parent = |
6741 | &newframe->stack[i].spilled_ptr; | |
6742 | } | |
f4d7e40a | 6743 | } |
f1bca824 AS |
6744 | return 0; |
6745 | } | |
6746 | ||
c64b7983 JS |
6747 | /* Return true if it's OK to have the same insn return a different type. */ |
6748 | static bool reg_type_mismatch_ok(enum bpf_reg_type type) | |
6749 | { | |
6750 | switch (type) { | |
6751 | case PTR_TO_CTX: | |
6752 | case PTR_TO_SOCKET: | |
6753 | case PTR_TO_SOCKET_OR_NULL: | |
46f8bc92 MKL |
6754 | case PTR_TO_SOCK_COMMON: |
6755 | case PTR_TO_SOCK_COMMON_OR_NULL: | |
655a51e5 MKL |
6756 | case PTR_TO_TCP_SOCK: |
6757 | case PTR_TO_TCP_SOCK_OR_NULL: | |
fada7fdc | 6758 | case PTR_TO_XDP_SOCK: |
c64b7983 JS |
6759 | return false; |
6760 | default: | |
6761 | return true; | |
6762 | } | |
6763 | } | |
6764 | ||
6765 | /* If an instruction was previously used with particular pointer types, then we | |
6766 | * need to be careful to avoid cases such as the below, where it may be ok | |
6767 | * for one branch accessing the pointer, but not ok for the other branch: | |
6768 | * | |
6769 | * R1 = sock_ptr | |
6770 | * goto X; | |
6771 | * ... | |
6772 | * R1 = some_other_valid_ptr; | |
6773 | * goto X; | |
6774 | * ... | |
6775 | * R2 = *(u32 *)(R1 + 0); | |
6776 | */ | |
6777 | static bool reg_type_mismatch(enum bpf_reg_type src, enum bpf_reg_type prev) | |
6778 | { | |
6779 | return src != prev && (!reg_type_mismatch_ok(src) || | |
6780 | !reg_type_mismatch_ok(prev)); | |
6781 | } | |
6782 | ||
58e2af8b | 6783 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 6784 | { |
638f5b90 | 6785 | struct bpf_verifier_state *state; |
17a52670 | 6786 | struct bpf_insn *insns = env->prog->insnsi; |
638f5b90 | 6787 | struct bpf_reg_state *regs; |
06ee7115 | 6788 | int insn_cnt = env->prog->len; |
17a52670 AS |
6789 | bool do_print_state = false; |
6790 | ||
d9762e84 MKL |
6791 | env->prev_linfo = NULL; |
6792 | ||
638f5b90 AS |
6793 | state = kzalloc(sizeof(struct bpf_verifier_state), GFP_KERNEL); |
6794 | if (!state) | |
6795 | return -ENOMEM; | |
f4d7e40a | 6796 | state->curframe = 0; |
979d63d5 | 6797 | state->speculative = false; |
f4d7e40a AS |
6798 | state->frame[0] = kzalloc(sizeof(struct bpf_func_state), GFP_KERNEL); |
6799 | if (!state->frame[0]) { | |
6800 | kfree(state); | |
6801 | return -ENOMEM; | |
6802 | } | |
6803 | env->cur_state = state; | |
6804 | init_func_state(env, state->frame[0], | |
6805 | BPF_MAIN_FUNC /* callsite */, | |
6806 | 0 /* frameno */, | |
6807 | 0 /* subprogno, zero == main subprog */); | |
c08435ec | 6808 | |
17a52670 AS |
6809 | for (;;) { |
6810 | struct bpf_insn *insn; | |
6811 | u8 class; | |
6812 | int err; | |
6813 | ||
c08435ec | 6814 | if (env->insn_idx >= insn_cnt) { |
61bd5218 | 6815 | verbose(env, "invalid insn idx %d insn_cnt %d\n", |
c08435ec | 6816 | env->insn_idx, insn_cnt); |
17a52670 AS |
6817 | return -EFAULT; |
6818 | } | |
6819 | ||
c08435ec | 6820 | insn = &insns[env->insn_idx]; |
17a52670 AS |
6821 | class = BPF_CLASS(insn->code); |
6822 | ||
06ee7115 | 6823 | if (++env->insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
61bd5218 JK |
6824 | verbose(env, |
6825 | "BPF program is too large. Processed %d insn\n", | |
06ee7115 | 6826 | env->insn_processed); |
17a52670 AS |
6827 | return -E2BIG; |
6828 | } | |
6829 | ||
c08435ec | 6830 | err = is_state_visited(env, env->insn_idx); |
f1bca824 AS |
6831 | if (err < 0) |
6832 | return err; | |
6833 | if (err == 1) { | |
6834 | /* found equivalent state, can prune the search */ | |
06ee7115 | 6835 | if (env->log.level & BPF_LOG_LEVEL) { |
f1bca824 | 6836 | if (do_print_state) |
979d63d5 DB |
6837 | verbose(env, "\nfrom %d to %d%s: safe\n", |
6838 | env->prev_insn_idx, env->insn_idx, | |
6839 | env->cur_state->speculative ? | |
6840 | " (speculative execution)" : ""); | |
f1bca824 | 6841 | else |
c08435ec | 6842 | verbose(env, "%d: safe\n", env->insn_idx); |
f1bca824 AS |
6843 | } |
6844 | goto process_bpf_exit; | |
6845 | } | |
6846 | ||
c3494801 AS |
6847 | if (signal_pending(current)) |
6848 | return -EAGAIN; | |
6849 | ||
3c2ce60b DB |
6850 | if (need_resched()) |
6851 | cond_resched(); | |
6852 | ||
06ee7115 AS |
6853 | if (env->log.level & BPF_LOG_LEVEL2 || |
6854 | (env->log.level & BPF_LOG_LEVEL && do_print_state)) { | |
6855 | if (env->log.level & BPF_LOG_LEVEL2) | |
c08435ec | 6856 | verbose(env, "%d:", env->insn_idx); |
c5fc9692 | 6857 | else |
979d63d5 DB |
6858 | verbose(env, "\nfrom %d to %d%s:", |
6859 | env->prev_insn_idx, env->insn_idx, | |
6860 | env->cur_state->speculative ? | |
6861 | " (speculative execution)" : ""); | |
f4d7e40a | 6862 | print_verifier_state(env, state->frame[state->curframe]); |
17a52670 AS |
6863 | do_print_state = false; |
6864 | } | |
6865 | ||
06ee7115 | 6866 | if (env->log.level & BPF_LOG_LEVEL) { |
7105e828 DB |
6867 | const struct bpf_insn_cbs cbs = { |
6868 | .cb_print = verbose, | |
abe08840 | 6869 | .private_data = env, |
7105e828 DB |
6870 | }; |
6871 | ||
c08435ec DB |
6872 | verbose_linfo(env, env->insn_idx, "; "); |
6873 | verbose(env, "%d: ", env->insn_idx); | |
abe08840 | 6874 | print_bpf_insn(&cbs, insn, env->allow_ptr_leaks); |
17a52670 AS |
6875 | } |
6876 | ||
cae1927c | 6877 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
c08435ec DB |
6878 | err = bpf_prog_offload_verify_insn(env, env->insn_idx, |
6879 | env->prev_insn_idx); | |
cae1927c JK |
6880 | if (err) |
6881 | return err; | |
6882 | } | |
13a27dfc | 6883 | |
638f5b90 | 6884 | regs = cur_regs(env); |
c08435ec | 6885 | env->insn_aux_data[env->insn_idx].seen = true; |
fd978bf7 | 6886 | |
17a52670 | 6887 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 6888 | err = check_alu_op(env, insn); |
17a52670 AS |
6889 | if (err) |
6890 | return err; | |
6891 | ||
6892 | } else if (class == BPF_LDX) { | |
3df126f3 | 6893 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
6894 | |
6895 | /* check for reserved fields is already done */ | |
6896 | ||
17a52670 | 6897 | /* check src operand */ |
dc503a8a | 6898 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6899 | if (err) |
6900 | return err; | |
6901 | ||
dc503a8a | 6902 | err = check_reg_arg(env, insn->dst_reg, DST_OP_NO_MARK); |
17a52670 AS |
6903 | if (err) |
6904 | return err; | |
6905 | ||
725f9dcd AS |
6906 | src_reg_type = regs[insn->src_reg].type; |
6907 | ||
17a52670 AS |
6908 | /* check that memory (src_reg + off) is readable, |
6909 | * the state of dst_reg will be updated by this func | |
6910 | */ | |
c08435ec DB |
6911 | err = check_mem_access(env, env->insn_idx, insn->src_reg, |
6912 | insn->off, BPF_SIZE(insn->code), | |
6913 | BPF_READ, insn->dst_reg, false); | |
17a52670 AS |
6914 | if (err) |
6915 | return err; | |
6916 | ||
c08435ec | 6917 | prev_src_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
6918 | |
6919 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
6920 | /* saw a valid insn |
6921 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 6922 | * save type to validate intersecting paths |
9bac3d6d | 6923 | */ |
3df126f3 | 6924 | *prev_src_type = src_reg_type; |
9bac3d6d | 6925 | |
c64b7983 | 6926 | } else if (reg_type_mismatch(src_reg_type, *prev_src_type)) { |
9bac3d6d AS |
6927 | /* ABuser program is trying to use the same insn |
6928 | * dst_reg = *(u32*) (src_reg + off) | |
6929 | * with different pointer types: | |
6930 | * src_reg == ctx in one branch and | |
6931 | * src_reg == stack|map in some other branch. | |
6932 | * Reject it. | |
6933 | */ | |
61bd5218 | 6934 | verbose(env, "same insn cannot be used with different pointers\n"); |
9bac3d6d AS |
6935 | return -EINVAL; |
6936 | } | |
6937 | ||
17a52670 | 6938 | } else if (class == BPF_STX) { |
3df126f3 | 6939 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 6940 | |
17a52670 | 6941 | if (BPF_MODE(insn->code) == BPF_XADD) { |
c08435ec | 6942 | err = check_xadd(env, env->insn_idx, insn); |
17a52670 AS |
6943 | if (err) |
6944 | return err; | |
c08435ec | 6945 | env->insn_idx++; |
17a52670 AS |
6946 | continue; |
6947 | } | |
6948 | ||
17a52670 | 6949 | /* check src1 operand */ |
dc503a8a | 6950 | err = check_reg_arg(env, insn->src_reg, SRC_OP); |
17a52670 AS |
6951 | if (err) |
6952 | return err; | |
6953 | /* check src2 operand */ | |
dc503a8a | 6954 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6955 | if (err) |
6956 | return err; | |
6957 | ||
d691f9e8 AS |
6958 | dst_reg_type = regs[insn->dst_reg].type; |
6959 | ||
17a52670 | 6960 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
6961 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
6962 | insn->off, BPF_SIZE(insn->code), | |
6963 | BPF_WRITE, insn->src_reg, false); | |
17a52670 AS |
6964 | if (err) |
6965 | return err; | |
6966 | ||
c08435ec | 6967 | prev_dst_type = &env->insn_aux_data[env->insn_idx].ptr_type; |
3df126f3 JK |
6968 | |
6969 | if (*prev_dst_type == NOT_INIT) { | |
6970 | *prev_dst_type = dst_reg_type; | |
c64b7983 | 6971 | } else if (reg_type_mismatch(dst_reg_type, *prev_dst_type)) { |
61bd5218 | 6972 | verbose(env, "same insn cannot be used with different pointers\n"); |
d691f9e8 AS |
6973 | return -EINVAL; |
6974 | } | |
6975 | ||
17a52670 AS |
6976 | } else if (class == BPF_ST) { |
6977 | if (BPF_MODE(insn->code) != BPF_MEM || | |
6978 | insn->src_reg != BPF_REG_0) { | |
61bd5218 | 6979 | verbose(env, "BPF_ST uses reserved fields\n"); |
17a52670 AS |
6980 | return -EINVAL; |
6981 | } | |
6982 | /* check src operand */ | |
dc503a8a | 6983 | err = check_reg_arg(env, insn->dst_reg, SRC_OP); |
17a52670 AS |
6984 | if (err) |
6985 | return err; | |
6986 | ||
f37a8cb8 | 6987 | if (is_ctx_reg(env, insn->dst_reg)) { |
9d2be44a | 6988 | verbose(env, "BPF_ST stores into R%d %s is not allowed\n", |
2a159c6f DB |
6989 | insn->dst_reg, |
6990 | reg_type_str[reg_state(env, insn->dst_reg)->type]); | |
f37a8cb8 DB |
6991 | return -EACCES; |
6992 | } | |
6993 | ||
17a52670 | 6994 | /* check that memory (dst_reg + off) is writeable */ |
c08435ec DB |
6995 | err = check_mem_access(env, env->insn_idx, insn->dst_reg, |
6996 | insn->off, BPF_SIZE(insn->code), | |
6997 | BPF_WRITE, -1, false); | |
17a52670 AS |
6998 | if (err) |
6999 | return err; | |
7000 | ||
092ed096 | 7001 | } else if (class == BPF_JMP || class == BPF_JMP32) { |
17a52670 AS |
7002 | u8 opcode = BPF_OP(insn->code); |
7003 | ||
7004 | if (opcode == BPF_CALL) { | |
7005 | if (BPF_SRC(insn->code) != BPF_K || | |
7006 | insn->off != 0 || | |
f4d7e40a AS |
7007 | (insn->src_reg != BPF_REG_0 && |
7008 | insn->src_reg != BPF_PSEUDO_CALL) || | |
092ed096 JW |
7009 | insn->dst_reg != BPF_REG_0 || |
7010 | class == BPF_JMP32) { | |
61bd5218 | 7011 | verbose(env, "BPF_CALL uses reserved fields\n"); |
17a52670 AS |
7012 | return -EINVAL; |
7013 | } | |
7014 | ||
d83525ca AS |
7015 | if (env->cur_state->active_spin_lock && |
7016 | (insn->src_reg == BPF_PSEUDO_CALL || | |
7017 | insn->imm != BPF_FUNC_spin_unlock)) { | |
7018 | verbose(env, "function calls are not allowed while holding a lock\n"); | |
7019 | return -EINVAL; | |
7020 | } | |
f4d7e40a | 7021 | if (insn->src_reg == BPF_PSEUDO_CALL) |
c08435ec | 7022 | err = check_func_call(env, insn, &env->insn_idx); |
f4d7e40a | 7023 | else |
c08435ec | 7024 | err = check_helper_call(env, insn->imm, env->insn_idx); |
17a52670 AS |
7025 | if (err) |
7026 | return err; | |
7027 | ||
7028 | } else if (opcode == BPF_JA) { | |
7029 | if (BPF_SRC(insn->code) != BPF_K || | |
7030 | insn->imm != 0 || | |
7031 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
7032 | insn->dst_reg != BPF_REG_0 || |
7033 | class == BPF_JMP32) { | |
61bd5218 | 7034 | verbose(env, "BPF_JA uses reserved fields\n"); |
17a52670 AS |
7035 | return -EINVAL; |
7036 | } | |
7037 | ||
c08435ec | 7038 | env->insn_idx += insn->off + 1; |
17a52670 AS |
7039 | continue; |
7040 | ||
7041 | } else if (opcode == BPF_EXIT) { | |
7042 | if (BPF_SRC(insn->code) != BPF_K || | |
7043 | insn->imm != 0 || | |
7044 | insn->src_reg != BPF_REG_0 || | |
092ed096 JW |
7045 | insn->dst_reg != BPF_REG_0 || |
7046 | class == BPF_JMP32) { | |
61bd5218 | 7047 | verbose(env, "BPF_EXIT uses reserved fields\n"); |
17a52670 AS |
7048 | return -EINVAL; |
7049 | } | |
7050 | ||
d83525ca AS |
7051 | if (env->cur_state->active_spin_lock) { |
7052 | verbose(env, "bpf_spin_unlock is missing\n"); | |
7053 | return -EINVAL; | |
7054 | } | |
7055 | ||
f4d7e40a AS |
7056 | if (state->curframe) { |
7057 | /* exit from nested function */ | |
c08435ec DB |
7058 | env->prev_insn_idx = env->insn_idx; |
7059 | err = prepare_func_exit(env, &env->insn_idx); | |
f4d7e40a AS |
7060 | if (err) |
7061 | return err; | |
7062 | do_print_state = true; | |
7063 | continue; | |
7064 | } | |
7065 | ||
fd978bf7 JS |
7066 | err = check_reference_leak(env); |
7067 | if (err) | |
7068 | return err; | |
7069 | ||
17a52670 AS |
7070 | /* eBPF calling convetion is such that R0 is used |
7071 | * to return the value from eBPF program. | |
7072 | * Make sure that it's readable at this time | |
7073 | * of bpf_exit, which means that program wrote | |
7074 | * something into it earlier | |
7075 | */ | |
dc503a8a | 7076 | err = check_reg_arg(env, BPF_REG_0, SRC_OP); |
17a52670 AS |
7077 | if (err) |
7078 | return err; | |
7079 | ||
1be7f75d | 7080 | if (is_pointer_value(env, BPF_REG_0)) { |
61bd5218 | 7081 | verbose(env, "R0 leaks addr as return value\n"); |
1be7f75d AS |
7082 | return -EACCES; |
7083 | } | |
7084 | ||
390ee7e2 AS |
7085 | err = check_return_code(env); |
7086 | if (err) | |
7087 | return err; | |
f1bca824 | 7088 | process_bpf_exit: |
c08435ec DB |
7089 | err = pop_stack(env, &env->prev_insn_idx, |
7090 | &env->insn_idx); | |
638f5b90 AS |
7091 | if (err < 0) { |
7092 | if (err != -ENOENT) | |
7093 | return err; | |
17a52670 AS |
7094 | break; |
7095 | } else { | |
7096 | do_print_state = true; | |
7097 | continue; | |
7098 | } | |
7099 | } else { | |
c08435ec | 7100 | err = check_cond_jmp_op(env, insn, &env->insn_idx); |
17a52670 AS |
7101 | if (err) |
7102 | return err; | |
7103 | } | |
7104 | } else if (class == BPF_LD) { | |
7105 | u8 mode = BPF_MODE(insn->code); | |
7106 | ||
7107 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
7108 | err = check_ld_abs(env, insn); |
7109 | if (err) | |
7110 | return err; | |
7111 | ||
17a52670 AS |
7112 | } else if (mode == BPF_IMM) { |
7113 | err = check_ld_imm(env, insn); | |
7114 | if (err) | |
7115 | return err; | |
7116 | ||
c08435ec DB |
7117 | env->insn_idx++; |
7118 | env->insn_aux_data[env->insn_idx].seen = true; | |
17a52670 | 7119 | } else { |
61bd5218 | 7120 | verbose(env, "invalid BPF_LD mode\n"); |
17a52670 AS |
7121 | return -EINVAL; |
7122 | } | |
7123 | } else { | |
61bd5218 | 7124 | verbose(env, "unknown insn class %d\n", class); |
17a52670 AS |
7125 | return -EINVAL; |
7126 | } | |
7127 | ||
c08435ec | 7128 | env->insn_idx++; |
17a52670 AS |
7129 | } |
7130 | ||
9c8105bd | 7131 | env->prog->aux->stack_depth = env->subprog_info[0].stack_depth; |
17a52670 AS |
7132 | return 0; |
7133 | } | |
7134 | ||
56f668df MKL |
7135 | static int check_map_prealloc(struct bpf_map *map) |
7136 | { | |
7137 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
7138 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
7139 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
7140 | !(map->map_flags & BPF_F_NO_PREALLOC); |
7141 | } | |
7142 | ||
d83525ca AS |
7143 | static bool is_tracing_prog_type(enum bpf_prog_type type) |
7144 | { | |
7145 | switch (type) { | |
7146 | case BPF_PROG_TYPE_KPROBE: | |
7147 | case BPF_PROG_TYPE_TRACEPOINT: | |
7148 | case BPF_PROG_TYPE_PERF_EVENT: | |
7149 | case BPF_PROG_TYPE_RAW_TRACEPOINT: | |
7150 | return true; | |
7151 | default: | |
7152 | return false; | |
7153 | } | |
7154 | } | |
7155 | ||
61bd5218 JK |
7156 | static int check_map_prog_compatibility(struct bpf_verifier_env *env, |
7157 | struct bpf_map *map, | |
fdc15d38 AS |
7158 | struct bpf_prog *prog) |
7159 | ||
7160 | { | |
56f668df MKL |
7161 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
7162 | * preallocated hash maps, since doing memory allocation | |
7163 | * in overflow_handler can crash depending on where nmi got | |
7164 | * triggered. | |
7165 | */ | |
7166 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
7167 | if (!check_map_prealloc(map)) { | |
61bd5218 | 7168 | verbose(env, "perf_event programs can only use preallocated hash map\n"); |
56f668df MKL |
7169 | return -EINVAL; |
7170 | } | |
7171 | if (map->inner_map_meta && | |
7172 | !check_map_prealloc(map->inner_map_meta)) { | |
61bd5218 | 7173 | verbose(env, "perf_event programs can only use preallocated inner hash map\n"); |
56f668df MKL |
7174 | return -EINVAL; |
7175 | } | |
fdc15d38 | 7176 | } |
a3884572 | 7177 | |
d83525ca AS |
7178 | if ((is_tracing_prog_type(prog->type) || |
7179 | prog->type == BPF_PROG_TYPE_SOCKET_FILTER) && | |
7180 | map_value_has_spin_lock(map)) { | |
7181 | verbose(env, "tracing progs cannot use bpf_spin_lock yet\n"); | |
7182 | return -EINVAL; | |
7183 | } | |
7184 | ||
a3884572 | 7185 | if ((bpf_prog_is_dev_bound(prog->aux) || bpf_map_is_dev_bound(map)) && |
09728266 | 7186 | !bpf_offload_prog_map_match(prog, map)) { |
a3884572 JK |
7187 | verbose(env, "offload device mismatch between prog and map\n"); |
7188 | return -EINVAL; | |
7189 | } | |
7190 | ||
fdc15d38 AS |
7191 | return 0; |
7192 | } | |
7193 | ||
b741f163 RG |
7194 | static bool bpf_map_is_cgroup_storage(struct bpf_map *map) |
7195 | { | |
7196 | return (map->map_type == BPF_MAP_TYPE_CGROUP_STORAGE || | |
7197 | map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE); | |
7198 | } | |
7199 | ||
0246e64d AS |
7200 | /* look for pseudo eBPF instructions that access map FDs and |
7201 | * replace them with actual map pointers | |
7202 | */ | |
58e2af8b | 7203 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
7204 | { |
7205 | struct bpf_insn *insn = env->prog->insnsi; | |
7206 | int insn_cnt = env->prog->len; | |
fdc15d38 | 7207 | int i, j, err; |
0246e64d | 7208 | |
f1f7714e | 7209 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
7210 | if (err) |
7211 | return err; | |
7212 | ||
0246e64d | 7213 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 7214 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 7215 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
61bd5218 | 7216 | verbose(env, "BPF_LDX uses reserved fields\n"); |
9bac3d6d AS |
7217 | return -EINVAL; |
7218 | } | |
7219 | ||
d691f9e8 AS |
7220 | if (BPF_CLASS(insn->code) == BPF_STX && |
7221 | ((BPF_MODE(insn->code) != BPF_MEM && | |
7222 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
61bd5218 | 7223 | verbose(env, "BPF_STX uses reserved fields\n"); |
d691f9e8 AS |
7224 | return -EINVAL; |
7225 | } | |
7226 | ||
0246e64d | 7227 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
d8eca5bb | 7228 | struct bpf_insn_aux_data *aux; |
0246e64d AS |
7229 | struct bpf_map *map; |
7230 | struct fd f; | |
d8eca5bb | 7231 | u64 addr; |
0246e64d AS |
7232 | |
7233 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
7234 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
7235 | insn[1].off != 0) { | |
61bd5218 | 7236 | verbose(env, "invalid bpf_ld_imm64 insn\n"); |
0246e64d AS |
7237 | return -EINVAL; |
7238 | } | |
7239 | ||
d8eca5bb | 7240 | if (insn[0].src_reg == 0) |
0246e64d AS |
7241 | /* valid generic load 64-bit imm */ |
7242 | goto next_insn; | |
7243 | ||
d8eca5bb DB |
7244 | /* In final convert_pseudo_ld_imm64() step, this is |
7245 | * converted into regular 64-bit imm load insn. | |
7246 | */ | |
7247 | if ((insn[0].src_reg != BPF_PSEUDO_MAP_FD && | |
7248 | insn[0].src_reg != BPF_PSEUDO_MAP_VALUE) || | |
7249 | (insn[0].src_reg == BPF_PSEUDO_MAP_FD && | |
7250 | insn[1].imm != 0)) { | |
7251 | verbose(env, | |
7252 | "unrecognized bpf_ld_imm64 insn\n"); | |
0246e64d AS |
7253 | return -EINVAL; |
7254 | } | |
7255 | ||
20182390 | 7256 | f = fdget(insn[0].imm); |
c2101297 | 7257 | map = __bpf_map_get(f); |
0246e64d | 7258 | if (IS_ERR(map)) { |
61bd5218 | 7259 | verbose(env, "fd %d is not pointing to valid bpf_map\n", |
20182390 | 7260 | insn[0].imm); |
0246e64d AS |
7261 | return PTR_ERR(map); |
7262 | } | |
7263 | ||
61bd5218 | 7264 | err = check_map_prog_compatibility(env, map, env->prog); |
fdc15d38 AS |
7265 | if (err) { |
7266 | fdput(f); | |
7267 | return err; | |
7268 | } | |
7269 | ||
d8eca5bb DB |
7270 | aux = &env->insn_aux_data[i]; |
7271 | if (insn->src_reg == BPF_PSEUDO_MAP_FD) { | |
7272 | addr = (unsigned long)map; | |
7273 | } else { | |
7274 | u32 off = insn[1].imm; | |
7275 | ||
7276 | if (off >= BPF_MAX_VAR_OFF) { | |
7277 | verbose(env, "direct value offset of %u is not allowed\n", off); | |
7278 | fdput(f); | |
7279 | return -EINVAL; | |
7280 | } | |
7281 | ||
7282 | if (!map->ops->map_direct_value_addr) { | |
7283 | verbose(env, "no direct value access support for this map type\n"); | |
7284 | fdput(f); | |
7285 | return -EINVAL; | |
7286 | } | |
7287 | ||
7288 | err = map->ops->map_direct_value_addr(map, &addr, off); | |
7289 | if (err) { | |
7290 | verbose(env, "invalid access to map value pointer, value_size=%u off=%u\n", | |
7291 | map->value_size, off); | |
7292 | fdput(f); | |
7293 | return err; | |
7294 | } | |
7295 | ||
7296 | aux->map_off = off; | |
7297 | addr += off; | |
7298 | } | |
7299 | ||
7300 | insn[0].imm = (u32)addr; | |
7301 | insn[1].imm = addr >> 32; | |
0246e64d AS |
7302 | |
7303 | /* check whether we recorded this map already */ | |
d8eca5bb | 7304 | for (j = 0; j < env->used_map_cnt; j++) { |
0246e64d | 7305 | if (env->used_maps[j] == map) { |
d8eca5bb | 7306 | aux->map_index = j; |
0246e64d AS |
7307 | fdput(f); |
7308 | goto next_insn; | |
7309 | } | |
d8eca5bb | 7310 | } |
0246e64d AS |
7311 | |
7312 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
7313 | fdput(f); | |
7314 | return -E2BIG; | |
7315 | } | |
7316 | ||
0246e64d AS |
7317 | /* hold the map. If the program is rejected by verifier, |
7318 | * the map will be released by release_maps() or it | |
7319 | * will be used by the valid program until it's unloaded | |
ab7f5bf0 | 7320 | * and all maps are released in free_used_maps() |
0246e64d | 7321 | */ |
92117d84 AS |
7322 | map = bpf_map_inc(map, false); |
7323 | if (IS_ERR(map)) { | |
7324 | fdput(f); | |
7325 | return PTR_ERR(map); | |
7326 | } | |
d8eca5bb DB |
7327 | |
7328 | aux->map_index = env->used_map_cnt; | |
92117d84 AS |
7329 | env->used_maps[env->used_map_cnt++] = map; |
7330 | ||
b741f163 | 7331 | if (bpf_map_is_cgroup_storage(map) && |
de9cbbaa | 7332 | bpf_cgroup_storage_assign(env->prog, map)) { |
b741f163 | 7333 | verbose(env, "only one cgroup storage of each type is allowed\n"); |
de9cbbaa RG |
7334 | fdput(f); |
7335 | return -EBUSY; | |
7336 | } | |
7337 | ||
0246e64d AS |
7338 | fdput(f); |
7339 | next_insn: | |
7340 | insn++; | |
7341 | i++; | |
5e581dad DB |
7342 | continue; |
7343 | } | |
7344 | ||
7345 | /* Basic sanity check before we invest more work here. */ | |
7346 | if (!bpf_opcode_in_insntable(insn->code)) { | |
7347 | verbose(env, "unknown opcode %02x\n", insn->code); | |
7348 | return -EINVAL; | |
0246e64d AS |
7349 | } |
7350 | } | |
7351 | ||
7352 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
7353 | * 'struct bpf_map *' into a register instead of user map_fd. | |
7354 | * These pointers will be used later by verifier to validate map access. | |
7355 | */ | |
7356 | return 0; | |
7357 | } | |
7358 | ||
7359 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 7360 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d | 7361 | { |
8bad74f9 | 7362 | enum bpf_cgroup_storage_type stype; |
0246e64d AS |
7363 | int i; |
7364 | ||
8bad74f9 RG |
7365 | for_each_cgroup_storage_type(stype) { |
7366 | if (!env->prog->aux->cgroup_storage[stype]) | |
7367 | continue; | |
de9cbbaa | 7368 | bpf_cgroup_storage_release(env->prog, |
8bad74f9 RG |
7369 | env->prog->aux->cgroup_storage[stype]); |
7370 | } | |
de9cbbaa | 7371 | |
0246e64d AS |
7372 | for (i = 0; i < env->used_map_cnt; i++) |
7373 | bpf_map_put(env->used_maps[i]); | |
7374 | } | |
7375 | ||
7376 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 7377 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
7378 | { |
7379 | struct bpf_insn *insn = env->prog->insnsi; | |
7380 | int insn_cnt = env->prog->len; | |
7381 | int i; | |
7382 | ||
7383 | for (i = 0; i < insn_cnt; i++, insn++) | |
7384 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
7385 | insn->src_reg = 0; | |
7386 | } | |
7387 | ||
8041902d AS |
7388 | /* single env->prog->insni[off] instruction was replaced with the range |
7389 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
7390 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
7391 | */ | |
b325fbca JW |
7392 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, |
7393 | struct bpf_prog *new_prog, u32 off, u32 cnt) | |
8041902d AS |
7394 | { |
7395 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
b325fbca JW |
7396 | struct bpf_insn *insn = new_prog->insnsi; |
7397 | u32 prog_len; | |
c131187d | 7398 | int i; |
8041902d | 7399 | |
b325fbca JW |
7400 | /* aux info at OFF always needs adjustment, no matter fast path |
7401 | * (cnt == 1) is taken or not. There is no guarantee INSN at OFF is the | |
7402 | * original insn at old prog. | |
7403 | */ | |
7404 | old_data[off].zext_dst = insn_has_def32(env, insn + off + cnt - 1); | |
7405 | ||
8041902d AS |
7406 | if (cnt == 1) |
7407 | return 0; | |
b325fbca | 7408 | prog_len = new_prog->len; |
fad953ce KC |
7409 | new_data = vzalloc(array_size(prog_len, |
7410 | sizeof(struct bpf_insn_aux_data))); | |
8041902d AS |
7411 | if (!new_data) |
7412 | return -ENOMEM; | |
7413 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
7414 | memcpy(new_data + off + cnt - 1, old_data + off, | |
7415 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
b325fbca | 7416 | for (i = off; i < off + cnt - 1; i++) { |
c131187d | 7417 | new_data[i].seen = true; |
b325fbca JW |
7418 | new_data[i].zext_dst = insn_has_def32(env, insn + i); |
7419 | } | |
8041902d AS |
7420 | env->insn_aux_data = new_data; |
7421 | vfree(old_data); | |
7422 | return 0; | |
7423 | } | |
7424 | ||
cc8b0b92 AS |
7425 | static void adjust_subprog_starts(struct bpf_verifier_env *env, u32 off, u32 len) |
7426 | { | |
7427 | int i; | |
7428 | ||
7429 | if (len == 1) | |
7430 | return; | |
4cb3d99c JW |
7431 | /* NOTE: fake 'exit' subprog should be updated as well. */ |
7432 | for (i = 0; i <= env->subprog_cnt; i++) { | |
afd59424 | 7433 | if (env->subprog_info[i].start <= off) |
cc8b0b92 | 7434 | continue; |
9c8105bd | 7435 | env->subprog_info[i].start += len - 1; |
cc8b0b92 AS |
7436 | } |
7437 | } | |
7438 | ||
8041902d AS |
7439 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, |
7440 | const struct bpf_insn *patch, u32 len) | |
7441 | { | |
7442 | struct bpf_prog *new_prog; | |
7443 | ||
7444 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
4f73379e AS |
7445 | if (IS_ERR(new_prog)) { |
7446 | if (PTR_ERR(new_prog) == -ERANGE) | |
7447 | verbose(env, | |
7448 | "insn %d cannot be patched due to 16-bit range\n", | |
7449 | env->insn_aux_data[off].orig_idx); | |
8041902d | 7450 | return NULL; |
4f73379e | 7451 | } |
b325fbca | 7452 | if (adjust_insn_aux_data(env, new_prog, off, len)) |
8041902d | 7453 | return NULL; |
cc8b0b92 | 7454 | adjust_subprog_starts(env, off, len); |
8041902d AS |
7455 | return new_prog; |
7456 | } | |
7457 | ||
52875a04 JK |
7458 | static int adjust_subprog_starts_after_remove(struct bpf_verifier_env *env, |
7459 | u32 off, u32 cnt) | |
7460 | { | |
7461 | int i, j; | |
7462 | ||
7463 | /* find first prog starting at or after off (first to remove) */ | |
7464 | for (i = 0; i < env->subprog_cnt; i++) | |
7465 | if (env->subprog_info[i].start >= off) | |
7466 | break; | |
7467 | /* find first prog starting at or after off + cnt (first to stay) */ | |
7468 | for (j = i; j < env->subprog_cnt; j++) | |
7469 | if (env->subprog_info[j].start >= off + cnt) | |
7470 | break; | |
7471 | /* if j doesn't start exactly at off + cnt, we are just removing | |
7472 | * the front of previous prog | |
7473 | */ | |
7474 | if (env->subprog_info[j].start != off + cnt) | |
7475 | j--; | |
7476 | ||
7477 | if (j > i) { | |
7478 | struct bpf_prog_aux *aux = env->prog->aux; | |
7479 | int move; | |
7480 | ||
7481 | /* move fake 'exit' subprog as well */ | |
7482 | move = env->subprog_cnt + 1 - j; | |
7483 | ||
7484 | memmove(env->subprog_info + i, | |
7485 | env->subprog_info + j, | |
7486 | sizeof(*env->subprog_info) * move); | |
7487 | env->subprog_cnt -= j - i; | |
7488 | ||
7489 | /* remove func_info */ | |
7490 | if (aux->func_info) { | |
7491 | move = aux->func_info_cnt - j; | |
7492 | ||
7493 | memmove(aux->func_info + i, | |
7494 | aux->func_info + j, | |
7495 | sizeof(*aux->func_info) * move); | |
7496 | aux->func_info_cnt -= j - i; | |
7497 | /* func_info->insn_off is set after all code rewrites, | |
7498 | * in adjust_btf_func() - no need to adjust | |
7499 | */ | |
7500 | } | |
7501 | } else { | |
7502 | /* convert i from "first prog to remove" to "first to adjust" */ | |
7503 | if (env->subprog_info[i].start == off) | |
7504 | i++; | |
7505 | } | |
7506 | ||
7507 | /* update fake 'exit' subprog as well */ | |
7508 | for (; i <= env->subprog_cnt; i++) | |
7509 | env->subprog_info[i].start -= cnt; | |
7510 | ||
7511 | return 0; | |
7512 | } | |
7513 | ||
7514 | static int bpf_adj_linfo_after_remove(struct bpf_verifier_env *env, u32 off, | |
7515 | u32 cnt) | |
7516 | { | |
7517 | struct bpf_prog *prog = env->prog; | |
7518 | u32 i, l_off, l_cnt, nr_linfo; | |
7519 | struct bpf_line_info *linfo; | |
7520 | ||
7521 | nr_linfo = prog->aux->nr_linfo; | |
7522 | if (!nr_linfo) | |
7523 | return 0; | |
7524 | ||
7525 | linfo = prog->aux->linfo; | |
7526 | ||
7527 | /* find first line info to remove, count lines to be removed */ | |
7528 | for (i = 0; i < nr_linfo; i++) | |
7529 | if (linfo[i].insn_off >= off) | |
7530 | break; | |
7531 | ||
7532 | l_off = i; | |
7533 | l_cnt = 0; | |
7534 | for (; i < nr_linfo; i++) | |
7535 | if (linfo[i].insn_off < off + cnt) | |
7536 | l_cnt++; | |
7537 | else | |
7538 | break; | |
7539 | ||
7540 | /* First live insn doesn't match first live linfo, it needs to "inherit" | |
7541 | * last removed linfo. prog is already modified, so prog->len == off | |
7542 | * means no live instructions after (tail of the program was removed). | |
7543 | */ | |
7544 | if (prog->len != off && l_cnt && | |
7545 | (i == nr_linfo || linfo[i].insn_off != off + cnt)) { | |
7546 | l_cnt--; | |
7547 | linfo[--i].insn_off = off + cnt; | |
7548 | } | |
7549 | ||
7550 | /* remove the line info which refer to the removed instructions */ | |
7551 | if (l_cnt) { | |
7552 | memmove(linfo + l_off, linfo + i, | |
7553 | sizeof(*linfo) * (nr_linfo - i)); | |
7554 | ||
7555 | prog->aux->nr_linfo -= l_cnt; | |
7556 | nr_linfo = prog->aux->nr_linfo; | |
7557 | } | |
7558 | ||
7559 | /* pull all linfo[i].insn_off >= off + cnt in by cnt */ | |
7560 | for (i = l_off; i < nr_linfo; i++) | |
7561 | linfo[i].insn_off -= cnt; | |
7562 | ||
7563 | /* fix up all subprogs (incl. 'exit') which start >= off */ | |
7564 | for (i = 0; i <= env->subprog_cnt; i++) | |
7565 | if (env->subprog_info[i].linfo_idx > l_off) { | |
7566 | /* program may have started in the removed region but | |
7567 | * may not be fully removed | |
7568 | */ | |
7569 | if (env->subprog_info[i].linfo_idx >= l_off + l_cnt) | |
7570 | env->subprog_info[i].linfo_idx -= l_cnt; | |
7571 | else | |
7572 | env->subprog_info[i].linfo_idx = l_off; | |
7573 | } | |
7574 | ||
7575 | return 0; | |
7576 | } | |
7577 | ||
7578 | static int verifier_remove_insns(struct bpf_verifier_env *env, u32 off, u32 cnt) | |
7579 | { | |
7580 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
7581 | unsigned int orig_prog_len = env->prog->len; | |
7582 | int err; | |
7583 | ||
08ca90af JK |
7584 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
7585 | bpf_prog_offload_remove_insns(env, off, cnt); | |
7586 | ||
52875a04 JK |
7587 | err = bpf_remove_insns(env->prog, off, cnt); |
7588 | if (err) | |
7589 | return err; | |
7590 | ||
7591 | err = adjust_subprog_starts_after_remove(env, off, cnt); | |
7592 | if (err) | |
7593 | return err; | |
7594 | ||
7595 | err = bpf_adj_linfo_after_remove(env, off, cnt); | |
7596 | if (err) | |
7597 | return err; | |
7598 | ||
7599 | memmove(aux_data + off, aux_data + off + cnt, | |
7600 | sizeof(*aux_data) * (orig_prog_len - off - cnt)); | |
7601 | ||
7602 | return 0; | |
7603 | } | |
7604 | ||
2a5418a1 DB |
7605 | /* The verifier does more data flow analysis than llvm and will not |
7606 | * explore branches that are dead at run time. Malicious programs can | |
7607 | * have dead code too. Therefore replace all dead at-run-time code | |
7608 | * with 'ja -1'. | |
7609 | * | |
7610 | * Just nops are not optimal, e.g. if they would sit at the end of the | |
7611 | * program and through another bug we would manage to jump there, then | |
7612 | * we'd execute beyond program memory otherwise. Returning exception | |
7613 | * code also wouldn't work since we can have subprogs where the dead | |
7614 | * code could be located. | |
c131187d AS |
7615 | */ |
7616 | static void sanitize_dead_code(struct bpf_verifier_env *env) | |
7617 | { | |
7618 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
2a5418a1 | 7619 | struct bpf_insn trap = BPF_JMP_IMM(BPF_JA, 0, 0, -1); |
c131187d AS |
7620 | struct bpf_insn *insn = env->prog->insnsi; |
7621 | const int insn_cnt = env->prog->len; | |
7622 | int i; | |
7623 | ||
7624 | for (i = 0; i < insn_cnt; i++) { | |
7625 | if (aux_data[i].seen) | |
7626 | continue; | |
2a5418a1 | 7627 | memcpy(insn + i, &trap, sizeof(trap)); |
c131187d AS |
7628 | } |
7629 | } | |
7630 | ||
e2ae4ca2 JK |
7631 | static bool insn_is_cond_jump(u8 code) |
7632 | { | |
7633 | u8 op; | |
7634 | ||
092ed096 JW |
7635 | if (BPF_CLASS(code) == BPF_JMP32) |
7636 | return true; | |
7637 | ||
e2ae4ca2 JK |
7638 | if (BPF_CLASS(code) != BPF_JMP) |
7639 | return false; | |
7640 | ||
7641 | op = BPF_OP(code); | |
7642 | return op != BPF_JA && op != BPF_EXIT && op != BPF_CALL; | |
7643 | } | |
7644 | ||
7645 | static void opt_hard_wire_dead_code_branches(struct bpf_verifier_env *env) | |
7646 | { | |
7647 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
7648 | struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
7649 | struct bpf_insn *insn = env->prog->insnsi; | |
7650 | const int insn_cnt = env->prog->len; | |
7651 | int i; | |
7652 | ||
7653 | for (i = 0; i < insn_cnt; i++, insn++) { | |
7654 | if (!insn_is_cond_jump(insn->code)) | |
7655 | continue; | |
7656 | ||
7657 | if (!aux_data[i + 1].seen) | |
7658 | ja.off = insn->off; | |
7659 | else if (!aux_data[i + 1 + insn->off].seen) | |
7660 | ja.off = 0; | |
7661 | else | |
7662 | continue; | |
7663 | ||
08ca90af JK |
7664 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
7665 | bpf_prog_offload_replace_insn(env, i, &ja); | |
7666 | ||
e2ae4ca2 JK |
7667 | memcpy(insn, &ja, sizeof(ja)); |
7668 | } | |
7669 | } | |
7670 | ||
52875a04 JK |
7671 | static int opt_remove_dead_code(struct bpf_verifier_env *env) |
7672 | { | |
7673 | struct bpf_insn_aux_data *aux_data = env->insn_aux_data; | |
7674 | int insn_cnt = env->prog->len; | |
7675 | int i, err; | |
7676 | ||
7677 | for (i = 0; i < insn_cnt; i++) { | |
7678 | int j; | |
7679 | ||
7680 | j = 0; | |
7681 | while (i + j < insn_cnt && !aux_data[i + j].seen) | |
7682 | j++; | |
7683 | if (!j) | |
7684 | continue; | |
7685 | ||
7686 | err = verifier_remove_insns(env, i, j); | |
7687 | if (err) | |
7688 | return err; | |
7689 | insn_cnt = env->prog->len; | |
7690 | } | |
7691 | ||
7692 | return 0; | |
7693 | } | |
7694 | ||
a1b14abc JK |
7695 | static int opt_remove_nops(struct bpf_verifier_env *env) |
7696 | { | |
7697 | const struct bpf_insn ja = BPF_JMP_IMM(BPF_JA, 0, 0, 0); | |
7698 | struct bpf_insn *insn = env->prog->insnsi; | |
7699 | int insn_cnt = env->prog->len; | |
7700 | int i, err; | |
7701 | ||
7702 | for (i = 0; i < insn_cnt; i++) { | |
7703 | if (memcmp(&insn[i], &ja, sizeof(ja))) | |
7704 | continue; | |
7705 | ||
7706 | err = verifier_remove_insns(env, i, 1); | |
7707 | if (err) | |
7708 | return err; | |
7709 | insn_cnt--; | |
7710 | i--; | |
7711 | } | |
7712 | ||
7713 | return 0; | |
7714 | } | |
7715 | ||
d6c2308c JW |
7716 | static int opt_subreg_zext_lo32_rnd_hi32(struct bpf_verifier_env *env, |
7717 | const union bpf_attr *attr) | |
a4b1d3c1 | 7718 | { |
d6c2308c | 7719 | struct bpf_insn *patch, zext_patch[2], rnd_hi32_patch[4]; |
a4b1d3c1 | 7720 | struct bpf_insn_aux_data *aux = env->insn_aux_data; |
d6c2308c | 7721 | int i, patch_len, delta = 0, len = env->prog->len; |
a4b1d3c1 | 7722 | struct bpf_insn *insns = env->prog->insnsi; |
a4b1d3c1 | 7723 | struct bpf_prog *new_prog; |
d6c2308c | 7724 | bool rnd_hi32; |
a4b1d3c1 | 7725 | |
d6c2308c | 7726 | rnd_hi32 = attr->prog_flags & BPF_F_TEST_RND_HI32; |
a4b1d3c1 | 7727 | zext_patch[1] = BPF_ZEXT_REG(0); |
d6c2308c JW |
7728 | rnd_hi32_patch[1] = BPF_ALU64_IMM(BPF_MOV, BPF_REG_AX, 0); |
7729 | rnd_hi32_patch[2] = BPF_ALU64_IMM(BPF_LSH, BPF_REG_AX, 32); | |
7730 | rnd_hi32_patch[3] = BPF_ALU64_REG(BPF_OR, 0, BPF_REG_AX); | |
a4b1d3c1 JW |
7731 | for (i = 0; i < len; i++) { |
7732 | int adj_idx = i + delta; | |
7733 | struct bpf_insn insn; | |
7734 | ||
d6c2308c JW |
7735 | insn = insns[adj_idx]; |
7736 | if (!aux[adj_idx].zext_dst) { | |
7737 | u8 code, class; | |
7738 | u32 imm_rnd; | |
7739 | ||
7740 | if (!rnd_hi32) | |
7741 | continue; | |
7742 | ||
7743 | code = insn.code; | |
7744 | class = BPF_CLASS(code); | |
7745 | if (insn_no_def(&insn)) | |
7746 | continue; | |
7747 | ||
7748 | /* NOTE: arg "reg" (the fourth one) is only used for | |
7749 | * BPF_STX which has been ruled out in above | |
7750 | * check, it is safe to pass NULL here. | |
7751 | */ | |
7752 | if (is_reg64(env, &insn, insn.dst_reg, NULL, DST_OP)) { | |
7753 | if (class == BPF_LD && | |
7754 | BPF_MODE(code) == BPF_IMM) | |
7755 | i++; | |
7756 | continue; | |
7757 | } | |
7758 | ||
7759 | /* ctx load could be transformed into wider load. */ | |
7760 | if (class == BPF_LDX && | |
7761 | aux[adj_idx].ptr_type == PTR_TO_CTX) | |
7762 | continue; | |
7763 | ||
7764 | imm_rnd = get_random_int(); | |
7765 | rnd_hi32_patch[0] = insn; | |
7766 | rnd_hi32_patch[1].imm = imm_rnd; | |
7767 | rnd_hi32_patch[3].dst_reg = insn.dst_reg; | |
7768 | patch = rnd_hi32_patch; | |
7769 | patch_len = 4; | |
7770 | goto apply_patch_buffer; | |
7771 | } | |
7772 | ||
7773 | if (!bpf_jit_needs_zext()) | |
a4b1d3c1 JW |
7774 | continue; |
7775 | ||
a4b1d3c1 JW |
7776 | zext_patch[0] = insn; |
7777 | zext_patch[1].dst_reg = insn.dst_reg; | |
7778 | zext_patch[1].src_reg = insn.dst_reg; | |
d6c2308c JW |
7779 | patch = zext_patch; |
7780 | patch_len = 2; | |
7781 | apply_patch_buffer: | |
7782 | new_prog = bpf_patch_insn_data(env, adj_idx, patch, patch_len); | |
a4b1d3c1 JW |
7783 | if (!new_prog) |
7784 | return -ENOMEM; | |
7785 | env->prog = new_prog; | |
7786 | insns = new_prog->insnsi; | |
7787 | aux = env->insn_aux_data; | |
d6c2308c | 7788 | delta += patch_len - 1; |
a4b1d3c1 JW |
7789 | } |
7790 | ||
7791 | return 0; | |
7792 | } | |
7793 | ||
c64b7983 JS |
7794 | /* convert load instructions that access fields of a context type into a |
7795 | * sequence of instructions that access fields of the underlying structure: | |
7796 | * struct __sk_buff -> struct sk_buff | |
7797 | * struct bpf_sock_ops -> struct sock | |
9bac3d6d | 7798 | */ |
58e2af8b | 7799 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 7800 | { |
00176a34 | 7801 | const struct bpf_verifier_ops *ops = env->ops; |
f96da094 | 7802 | int i, cnt, size, ctx_field_size, delta = 0; |
3df126f3 | 7803 | const int insn_cnt = env->prog->len; |
36bbef52 | 7804 | struct bpf_insn insn_buf[16], *insn; |
46f53a65 | 7805 | u32 target_size, size_default, off; |
9bac3d6d | 7806 | struct bpf_prog *new_prog; |
d691f9e8 | 7807 | enum bpf_access_type type; |
f96da094 | 7808 | bool is_narrower_load; |
9bac3d6d | 7809 | |
b09928b9 DB |
7810 | if (ops->gen_prologue || env->seen_direct_write) { |
7811 | if (!ops->gen_prologue) { | |
7812 | verbose(env, "bpf verifier is misconfigured\n"); | |
7813 | return -EINVAL; | |
7814 | } | |
36bbef52 DB |
7815 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, |
7816 | env->prog); | |
7817 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
61bd5218 | 7818 | verbose(env, "bpf verifier is misconfigured\n"); |
36bbef52 DB |
7819 | return -EINVAL; |
7820 | } else if (cnt) { | |
8041902d | 7821 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
7822 | if (!new_prog) |
7823 | return -ENOMEM; | |
8041902d | 7824 | |
36bbef52 | 7825 | env->prog = new_prog; |
3df126f3 | 7826 | delta += cnt - 1; |
36bbef52 DB |
7827 | } |
7828 | } | |
7829 | ||
c64b7983 | 7830 | if (bpf_prog_is_dev_bound(env->prog->aux)) |
9bac3d6d AS |
7831 | return 0; |
7832 | ||
3df126f3 | 7833 | insn = env->prog->insnsi + delta; |
36bbef52 | 7834 | |
9bac3d6d | 7835 | for (i = 0; i < insn_cnt; i++, insn++) { |
c64b7983 JS |
7836 | bpf_convert_ctx_access_t convert_ctx_access; |
7837 | ||
62c7989b DB |
7838 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
7839 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
7840 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 7841 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 7842 | type = BPF_READ; |
62c7989b DB |
7843 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
7844 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
7845 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 7846 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
7847 | type = BPF_WRITE; |
7848 | else | |
9bac3d6d AS |
7849 | continue; |
7850 | ||
af86ca4e AS |
7851 | if (type == BPF_WRITE && |
7852 | env->insn_aux_data[i + delta].sanitize_stack_off) { | |
7853 | struct bpf_insn patch[] = { | |
7854 | /* Sanitize suspicious stack slot with zero. | |
7855 | * There are no memory dependencies for this store, | |
7856 | * since it's only using frame pointer and immediate | |
7857 | * constant of zero | |
7858 | */ | |
7859 | BPF_ST_MEM(BPF_DW, BPF_REG_FP, | |
7860 | env->insn_aux_data[i + delta].sanitize_stack_off, | |
7861 | 0), | |
7862 | /* the original STX instruction will immediately | |
7863 | * overwrite the same stack slot with appropriate value | |
7864 | */ | |
7865 | *insn, | |
7866 | }; | |
7867 | ||
7868 | cnt = ARRAY_SIZE(patch); | |
7869 | new_prog = bpf_patch_insn_data(env, i + delta, patch, cnt); | |
7870 | if (!new_prog) | |
7871 | return -ENOMEM; | |
7872 | ||
7873 | delta += cnt - 1; | |
7874 | env->prog = new_prog; | |
7875 | insn = new_prog->insnsi + i + delta; | |
7876 | continue; | |
7877 | } | |
7878 | ||
c64b7983 JS |
7879 | switch (env->insn_aux_data[i + delta].ptr_type) { |
7880 | case PTR_TO_CTX: | |
7881 | if (!ops->convert_ctx_access) | |
7882 | continue; | |
7883 | convert_ctx_access = ops->convert_ctx_access; | |
7884 | break; | |
7885 | case PTR_TO_SOCKET: | |
46f8bc92 | 7886 | case PTR_TO_SOCK_COMMON: |
c64b7983 JS |
7887 | convert_ctx_access = bpf_sock_convert_ctx_access; |
7888 | break; | |
655a51e5 MKL |
7889 | case PTR_TO_TCP_SOCK: |
7890 | convert_ctx_access = bpf_tcp_sock_convert_ctx_access; | |
7891 | break; | |
fada7fdc JL |
7892 | case PTR_TO_XDP_SOCK: |
7893 | convert_ctx_access = bpf_xdp_sock_convert_ctx_access; | |
7894 | break; | |
c64b7983 | 7895 | default: |
9bac3d6d | 7896 | continue; |
c64b7983 | 7897 | } |
9bac3d6d | 7898 | |
31fd8581 | 7899 | ctx_field_size = env->insn_aux_data[i + delta].ctx_field_size; |
f96da094 | 7900 | size = BPF_LDST_BYTES(insn); |
31fd8581 YS |
7901 | |
7902 | /* If the read access is a narrower load of the field, | |
7903 | * convert to a 4/8-byte load, to minimum program type specific | |
7904 | * convert_ctx_access changes. If conversion is successful, | |
7905 | * we will apply proper mask to the result. | |
7906 | */ | |
f96da094 | 7907 | is_narrower_load = size < ctx_field_size; |
46f53a65 AI |
7908 | size_default = bpf_ctx_off_adjust_machine(ctx_field_size); |
7909 | off = insn->off; | |
31fd8581 | 7910 | if (is_narrower_load) { |
f96da094 DB |
7911 | u8 size_code; |
7912 | ||
7913 | if (type == BPF_WRITE) { | |
61bd5218 | 7914 | verbose(env, "bpf verifier narrow ctx access misconfigured\n"); |
f96da094 DB |
7915 | return -EINVAL; |
7916 | } | |
31fd8581 | 7917 | |
f96da094 | 7918 | size_code = BPF_H; |
31fd8581 YS |
7919 | if (ctx_field_size == 4) |
7920 | size_code = BPF_W; | |
7921 | else if (ctx_field_size == 8) | |
7922 | size_code = BPF_DW; | |
f96da094 | 7923 | |
bc23105c | 7924 | insn->off = off & ~(size_default - 1); |
31fd8581 YS |
7925 | insn->code = BPF_LDX | BPF_MEM | size_code; |
7926 | } | |
f96da094 DB |
7927 | |
7928 | target_size = 0; | |
c64b7983 JS |
7929 | cnt = convert_ctx_access(type, insn, insn_buf, env->prog, |
7930 | &target_size); | |
f96da094 DB |
7931 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf) || |
7932 | (ctx_field_size && !target_size)) { | |
61bd5218 | 7933 | verbose(env, "bpf verifier is misconfigured\n"); |
9bac3d6d AS |
7934 | return -EINVAL; |
7935 | } | |
f96da094 DB |
7936 | |
7937 | if (is_narrower_load && size < target_size) { | |
46f53a65 AI |
7938 | u8 shift = (off & (size_default - 1)) * 8; |
7939 | ||
7940 | if (ctx_field_size <= 4) { | |
7941 | if (shift) | |
7942 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_RSH, | |
7943 | insn->dst_reg, | |
7944 | shift); | |
31fd8581 | 7945 | insn_buf[cnt++] = BPF_ALU32_IMM(BPF_AND, insn->dst_reg, |
f96da094 | 7946 | (1 << size * 8) - 1); |
46f53a65 AI |
7947 | } else { |
7948 | if (shift) | |
7949 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_RSH, | |
7950 | insn->dst_reg, | |
7951 | shift); | |
31fd8581 | 7952 | insn_buf[cnt++] = BPF_ALU64_IMM(BPF_AND, insn->dst_reg, |
e2f7fc0a | 7953 | (1ULL << size * 8) - 1); |
46f53a65 | 7954 | } |
31fd8581 | 7955 | } |
9bac3d6d | 7956 | |
8041902d | 7957 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
7958 | if (!new_prog) |
7959 | return -ENOMEM; | |
7960 | ||
3df126f3 | 7961 | delta += cnt - 1; |
9bac3d6d AS |
7962 | |
7963 | /* keep walking new program and skip insns we just inserted */ | |
7964 | env->prog = new_prog; | |
3df126f3 | 7965 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
7966 | } |
7967 | ||
7968 | return 0; | |
7969 | } | |
7970 | ||
1c2a088a AS |
7971 | static int jit_subprogs(struct bpf_verifier_env *env) |
7972 | { | |
7973 | struct bpf_prog *prog = env->prog, **func, *tmp; | |
7974 | int i, j, subprog_start, subprog_end = 0, len, subprog; | |
7105e828 | 7975 | struct bpf_insn *insn; |
1c2a088a | 7976 | void *old_bpf_func; |
c454a46b | 7977 | int err; |
1c2a088a | 7978 | |
f910cefa | 7979 | if (env->subprog_cnt <= 1) |
1c2a088a AS |
7980 | return 0; |
7981 | ||
7105e828 | 7982 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { |
1c2a088a AS |
7983 | if (insn->code != (BPF_JMP | BPF_CALL) || |
7984 | insn->src_reg != BPF_PSEUDO_CALL) | |
7985 | continue; | |
c7a89784 DB |
7986 | /* Upon error here we cannot fall back to interpreter but |
7987 | * need a hard reject of the program. Thus -EFAULT is | |
7988 | * propagated in any case. | |
7989 | */ | |
1c2a088a AS |
7990 | subprog = find_subprog(env, i + insn->imm + 1); |
7991 | if (subprog < 0) { | |
7992 | WARN_ONCE(1, "verifier bug. No program starts at insn %d\n", | |
7993 | i + insn->imm + 1); | |
7994 | return -EFAULT; | |
7995 | } | |
7996 | /* temporarily remember subprog id inside insn instead of | |
7997 | * aux_data, since next loop will split up all insns into funcs | |
7998 | */ | |
f910cefa | 7999 | insn->off = subprog; |
1c2a088a AS |
8000 | /* remember original imm in case JIT fails and fallback |
8001 | * to interpreter will be needed | |
8002 | */ | |
8003 | env->insn_aux_data[i].call_imm = insn->imm; | |
8004 | /* point imm to __bpf_call_base+1 from JITs point of view */ | |
8005 | insn->imm = 1; | |
8006 | } | |
8007 | ||
c454a46b MKL |
8008 | err = bpf_prog_alloc_jited_linfo(prog); |
8009 | if (err) | |
8010 | goto out_undo_insn; | |
8011 | ||
8012 | err = -ENOMEM; | |
6396bb22 | 8013 | func = kcalloc(env->subprog_cnt, sizeof(prog), GFP_KERNEL); |
1c2a088a | 8014 | if (!func) |
c7a89784 | 8015 | goto out_undo_insn; |
1c2a088a | 8016 | |
f910cefa | 8017 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a | 8018 | subprog_start = subprog_end; |
4cb3d99c | 8019 | subprog_end = env->subprog_info[i + 1].start; |
1c2a088a AS |
8020 | |
8021 | len = subprog_end - subprog_start; | |
492ecee8 AS |
8022 | /* BPF_PROG_RUN doesn't call subprogs directly, |
8023 | * hence main prog stats include the runtime of subprogs. | |
8024 | * subprogs don't have IDs and not reachable via prog_get_next_id | |
8025 | * func[i]->aux->stats will never be accessed and stays NULL | |
8026 | */ | |
8027 | func[i] = bpf_prog_alloc_no_stats(bpf_prog_size(len), GFP_USER); | |
1c2a088a AS |
8028 | if (!func[i]) |
8029 | goto out_free; | |
8030 | memcpy(func[i]->insnsi, &prog->insnsi[subprog_start], | |
8031 | len * sizeof(struct bpf_insn)); | |
4f74d809 | 8032 | func[i]->type = prog->type; |
1c2a088a | 8033 | func[i]->len = len; |
4f74d809 DB |
8034 | if (bpf_prog_calc_tag(func[i])) |
8035 | goto out_free; | |
1c2a088a | 8036 | func[i]->is_func = 1; |
ba64e7d8 YS |
8037 | func[i]->aux->func_idx = i; |
8038 | /* the btf and func_info will be freed only at prog->aux */ | |
8039 | func[i]->aux->btf = prog->aux->btf; | |
8040 | func[i]->aux->func_info = prog->aux->func_info; | |
8041 | ||
1c2a088a AS |
8042 | /* Use bpf_prog_F_tag to indicate functions in stack traces. |
8043 | * Long term would need debug info to populate names | |
8044 | */ | |
8045 | func[i]->aux->name[0] = 'F'; | |
9c8105bd | 8046 | func[i]->aux->stack_depth = env->subprog_info[i].stack_depth; |
1c2a088a | 8047 | func[i]->jit_requested = 1; |
c454a46b MKL |
8048 | func[i]->aux->linfo = prog->aux->linfo; |
8049 | func[i]->aux->nr_linfo = prog->aux->nr_linfo; | |
8050 | func[i]->aux->jited_linfo = prog->aux->jited_linfo; | |
8051 | func[i]->aux->linfo_idx = env->subprog_info[i].linfo_idx; | |
1c2a088a AS |
8052 | func[i] = bpf_int_jit_compile(func[i]); |
8053 | if (!func[i]->jited) { | |
8054 | err = -ENOTSUPP; | |
8055 | goto out_free; | |
8056 | } | |
8057 | cond_resched(); | |
8058 | } | |
8059 | /* at this point all bpf functions were successfully JITed | |
8060 | * now populate all bpf_calls with correct addresses and | |
8061 | * run last pass of JIT | |
8062 | */ | |
f910cefa | 8063 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
8064 | insn = func[i]->insnsi; |
8065 | for (j = 0; j < func[i]->len; j++, insn++) { | |
8066 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
8067 | insn->src_reg != BPF_PSEUDO_CALL) | |
8068 | continue; | |
8069 | subprog = insn->off; | |
0d306c31 PB |
8070 | insn->imm = BPF_CAST_CALL(func[subprog]->bpf_func) - |
8071 | __bpf_call_base; | |
1c2a088a | 8072 | } |
2162fed4 SD |
8073 | |
8074 | /* we use the aux data to keep a list of the start addresses | |
8075 | * of the JITed images for each function in the program | |
8076 | * | |
8077 | * for some architectures, such as powerpc64, the imm field | |
8078 | * might not be large enough to hold the offset of the start | |
8079 | * address of the callee's JITed image from __bpf_call_base | |
8080 | * | |
8081 | * in such cases, we can lookup the start address of a callee | |
8082 | * by using its subprog id, available from the off field of | |
8083 | * the call instruction, as an index for this list | |
8084 | */ | |
8085 | func[i]->aux->func = func; | |
8086 | func[i]->aux->func_cnt = env->subprog_cnt; | |
1c2a088a | 8087 | } |
f910cefa | 8088 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
8089 | old_bpf_func = func[i]->bpf_func; |
8090 | tmp = bpf_int_jit_compile(func[i]); | |
8091 | if (tmp != func[i] || func[i]->bpf_func != old_bpf_func) { | |
8092 | verbose(env, "JIT doesn't support bpf-to-bpf calls\n"); | |
c7a89784 | 8093 | err = -ENOTSUPP; |
1c2a088a AS |
8094 | goto out_free; |
8095 | } | |
8096 | cond_resched(); | |
8097 | } | |
8098 | ||
8099 | /* finally lock prog and jit images for all functions and | |
8100 | * populate kallsysm | |
8101 | */ | |
f910cefa | 8102 | for (i = 0; i < env->subprog_cnt; i++) { |
1c2a088a AS |
8103 | bpf_prog_lock_ro(func[i]); |
8104 | bpf_prog_kallsyms_add(func[i]); | |
8105 | } | |
7105e828 DB |
8106 | |
8107 | /* Last step: make now unused interpreter insns from main | |
8108 | * prog consistent for later dump requests, so they can | |
8109 | * later look the same as if they were interpreted only. | |
8110 | */ | |
8111 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
7105e828 DB |
8112 | if (insn->code != (BPF_JMP | BPF_CALL) || |
8113 | insn->src_reg != BPF_PSEUDO_CALL) | |
8114 | continue; | |
8115 | insn->off = env->insn_aux_data[i].call_imm; | |
8116 | subprog = find_subprog(env, i + insn->off + 1); | |
dbecd738 | 8117 | insn->imm = subprog; |
7105e828 DB |
8118 | } |
8119 | ||
1c2a088a AS |
8120 | prog->jited = 1; |
8121 | prog->bpf_func = func[0]->bpf_func; | |
8122 | prog->aux->func = func; | |
f910cefa | 8123 | prog->aux->func_cnt = env->subprog_cnt; |
c454a46b | 8124 | bpf_prog_free_unused_jited_linfo(prog); |
1c2a088a AS |
8125 | return 0; |
8126 | out_free: | |
f910cefa | 8127 | for (i = 0; i < env->subprog_cnt; i++) |
1c2a088a AS |
8128 | if (func[i]) |
8129 | bpf_jit_free(func[i]); | |
8130 | kfree(func); | |
c7a89784 | 8131 | out_undo_insn: |
1c2a088a AS |
8132 | /* cleanup main prog to be interpreted */ |
8133 | prog->jit_requested = 0; | |
8134 | for (i = 0, insn = prog->insnsi; i < prog->len; i++, insn++) { | |
8135 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
8136 | insn->src_reg != BPF_PSEUDO_CALL) | |
8137 | continue; | |
8138 | insn->off = 0; | |
8139 | insn->imm = env->insn_aux_data[i].call_imm; | |
8140 | } | |
c454a46b | 8141 | bpf_prog_free_jited_linfo(prog); |
1c2a088a AS |
8142 | return err; |
8143 | } | |
8144 | ||
1ea47e01 AS |
8145 | static int fixup_call_args(struct bpf_verifier_env *env) |
8146 | { | |
19d28fbd | 8147 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON |
1ea47e01 AS |
8148 | struct bpf_prog *prog = env->prog; |
8149 | struct bpf_insn *insn = prog->insnsi; | |
8150 | int i, depth; | |
19d28fbd | 8151 | #endif |
e4052d06 | 8152 | int err = 0; |
1ea47e01 | 8153 | |
e4052d06 QM |
8154 | if (env->prog->jit_requested && |
8155 | !bpf_prog_is_dev_bound(env->prog->aux)) { | |
19d28fbd DM |
8156 | err = jit_subprogs(env); |
8157 | if (err == 0) | |
1c2a088a | 8158 | return 0; |
c7a89784 DB |
8159 | if (err == -EFAULT) |
8160 | return err; | |
19d28fbd DM |
8161 | } |
8162 | #ifndef CONFIG_BPF_JIT_ALWAYS_ON | |
1ea47e01 AS |
8163 | for (i = 0; i < prog->len; i++, insn++) { |
8164 | if (insn->code != (BPF_JMP | BPF_CALL) || | |
8165 | insn->src_reg != BPF_PSEUDO_CALL) | |
8166 | continue; | |
8167 | depth = get_callee_stack_depth(env, insn, i); | |
8168 | if (depth < 0) | |
8169 | return depth; | |
8170 | bpf_patch_call_args(insn, depth); | |
8171 | } | |
19d28fbd DM |
8172 | err = 0; |
8173 | #endif | |
8174 | return err; | |
1ea47e01 AS |
8175 | } |
8176 | ||
79741b3b | 8177 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 8178 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
8179 | * |
8180 | * this function is called after eBPF program passed verification | |
8181 | */ | |
79741b3b | 8182 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 8183 | { |
79741b3b AS |
8184 | struct bpf_prog *prog = env->prog; |
8185 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 8186 | const struct bpf_func_proto *fn; |
79741b3b | 8187 | const int insn_cnt = prog->len; |
09772d92 | 8188 | const struct bpf_map_ops *ops; |
c93552c4 | 8189 | struct bpf_insn_aux_data *aux; |
81ed18ab AS |
8190 | struct bpf_insn insn_buf[16]; |
8191 | struct bpf_prog *new_prog; | |
8192 | struct bpf_map *map_ptr; | |
8193 | int i, cnt, delta = 0; | |
e245c5c6 | 8194 | |
79741b3b | 8195 | for (i = 0; i < insn_cnt; i++, insn++) { |
f6b1b3bf DB |
8196 | if (insn->code == (BPF_ALU64 | BPF_MOD | BPF_X) || |
8197 | insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
8198 | insn->code == (BPF_ALU | BPF_MOD | BPF_X) || | |
68fda450 | 8199 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { |
f6b1b3bf DB |
8200 | bool is64 = BPF_CLASS(insn->code) == BPF_ALU64; |
8201 | struct bpf_insn mask_and_div[] = { | |
8202 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
8203 | /* Rx div 0 -> 0 */ | |
8204 | BPF_JMP_IMM(BPF_JNE, insn->src_reg, 0, 2), | |
8205 | BPF_ALU32_REG(BPF_XOR, insn->dst_reg, insn->dst_reg), | |
8206 | BPF_JMP_IMM(BPF_JA, 0, 0, 1), | |
8207 | *insn, | |
8208 | }; | |
8209 | struct bpf_insn mask_and_mod[] = { | |
8210 | BPF_MOV32_REG(insn->src_reg, insn->src_reg), | |
8211 | /* Rx mod 0 -> Rx */ | |
8212 | BPF_JMP_IMM(BPF_JEQ, insn->src_reg, 0, 1), | |
8213 | *insn, | |
8214 | }; | |
8215 | struct bpf_insn *patchlet; | |
8216 | ||
8217 | if (insn->code == (BPF_ALU64 | BPF_DIV | BPF_X) || | |
8218 | insn->code == (BPF_ALU | BPF_DIV | BPF_X)) { | |
8219 | patchlet = mask_and_div + (is64 ? 1 : 0); | |
8220 | cnt = ARRAY_SIZE(mask_and_div) - (is64 ? 1 : 0); | |
8221 | } else { | |
8222 | patchlet = mask_and_mod + (is64 ? 1 : 0); | |
8223 | cnt = ARRAY_SIZE(mask_and_mod) - (is64 ? 1 : 0); | |
8224 | } | |
8225 | ||
8226 | new_prog = bpf_patch_insn_data(env, i + delta, patchlet, cnt); | |
68fda450 AS |
8227 | if (!new_prog) |
8228 | return -ENOMEM; | |
8229 | ||
8230 | delta += cnt - 1; | |
8231 | env->prog = prog = new_prog; | |
8232 | insn = new_prog->insnsi + i + delta; | |
8233 | continue; | |
8234 | } | |
8235 | ||
e0cea7ce DB |
8236 | if (BPF_CLASS(insn->code) == BPF_LD && |
8237 | (BPF_MODE(insn->code) == BPF_ABS || | |
8238 | BPF_MODE(insn->code) == BPF_IND)) { | |
8239 | cnt = env->ops->gen_ld_abs(insn, insn_buf); | |
8240 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
8241 | verbose(env, "bpf verifier is misconfigured\n"); | |
8242 | return -EINVAL; | |
8243 | } | |
8244 | ||
8245 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
8246 | if (!new_prog) | |
8247 | return -ENOMEM; | |
8248 | ||
8249 | delta += cnt - 1; | |
8250 | env->prog = prog = new_prog; | |
8251 | insn = new_prog->insnsi + i + delta; | |
8252 | continue; | |
8253 | } | |
8254 | ||
979d63d5 DB |
8255 | if (insn->code == (BPF_ALU64 | BPF_ADD | BPF_X) || |
8256 | insn->code == (BPF_ALU64 | BPF_SUB | BPF_X)) { | |
8257 | const u8 code_add = BPF_ALU64 | BPF_ADD | BPF_X; | |
8258 | const u8 code_sub = BPF_ALU64 | BPF_SUB | BPF_X; | |
8259 | struct bpf_insn insn_buf[16]; | |
8260 | struct bpf_insn *patch = &insn_buf[0]; | |
8261 | bool issrc, isneg; | |
8262 | u32 off_reg; | |
8263 | ||
8264 | aux = &env->insn_aux_data[i + delta]; | |
3612af78 DB |
8265 | if (!aux->alu_state || |
8266 | aux->alu_state == BPF_ALU_NON_POINTER) | |
979d63d5 DB |
8267 | continue; |
8268 | ||
8269 | isneg = aux->alu_state & BPF_ALU_NEG_VALUE; | |
8270 | issrc = (aux->alu_state & BPF_ALU_SANITIZE) == | |
8271 | BPF_ALU_SANITIZE_SRC; | |
8272 | ||
8273 | off_reg = issrc ? insn->src_reg : insn->dst_reg; | |
8274 | if (isneg) | |
8275 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
8276 | *patch++ = BPF_MOV32_IMM(BPF_REG_AX, aux->alu_limit - 1); | |
8277 | *patch++ = BPF_ALU64_REG(BPF_SUB, BPF_REG_AX, off_reg); | |
8278 | *patch++ = BPF_ALU64_REG(BPF_OR, BPF_REG_AX, off_reg); | |
8279 | *patch++ = BPF_ALU64_IMM(BPF_NEG, BPF_REG_AX, 0); | |
8280 | *patch++ = BPF_ALU64_IMM(BPF_ARSH, BPF_REG_AX, 63); | |
8281 | if (issrc) { | |
8282 | *patch++ = BPF_ALU64_REG(BPF_AND, BPF_REG_AX, | |
8283 | off_reg); | |
8284 | insn->src_reg = BPF_REG_AX; | |
8285 | } else { | |
8286 | *patch++ = BPF_ALU64_REG(BPF_AND, off_reg, | |
8287 | BPF_REG_AX); | |
8288 | } | |
8289 | if (isneg) | |
8290 | insn->code = insn->code == code_add ? | |
8291 | code_sub : code_add; | |
8292 | *patch++ = *insn; | |
8293 | if (issrc && isneg) | |
8294 | *patch++ = BPF_ALU64_IMM(BPF_MUL, off_reg, -1); | |
8295 | cnt = patch - insn_buf; | |
8296 | ||
8297 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
8298 | if (!new_prog) | |
8299 | return -ENOMEM; | |
8300 | ||
8301 | delta += cnt - 1; | |
8302 | env->prog = prog = new_prog; | |
8303 | insn = new_prog->insnsi + i + delta; | |
8304 | continue; | |
8305 | } | |
8306 | ||
79741b3b AS |
8307 | if (insn->code != (BPF_JMP | BPF_CALL)) |
8308 | continue; | |
cc8b0b92 AS |
8309 | if (insn->src_reg == BPF_PSEUDO_CALL) |
8310 | continue; | |
e245c5c6 | 8311 | |
79741b3b AS |
8312 | if (insn->imm == BPF_FUNC_get_route_realm) |
8313 | prog->dst_needed = 1; | |
8314 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
8315 | bpf_user_rnd_init_once(); | |
9802d865 JB |
8316 | if (insn->imm == BPF_FUNC_override_return) |
8317 | prog->kprobe_override = 1; | |
79741b3b | 8318 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
8319 | /* If we tail call into other programs, we |
8320 | * cannot make any assumptions since they can | |
8321 | * be replaced dynamically during runtime in | |
8322 | * the program array. | |
8323 | */ | |
8324 | prog->cb_access = 1; | |
80a58d02 | 8325 | env->prog->aux->stack_depth = MAX_BPF_STACK; |
e647815a | 8326 | env->prog->aux->max_pkt_offset = MAX_PACKET_OFF; |
7b9f6da1 | 8327 | |
79741b3b AS |
8328 | /* mark bpf_tail_call as different opcode to avoid |
8329 | * conditional branch in the interpeter for every normal | |
8330 | * call and to prevent accidental JITing by JIT compiler | |
8331 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 8332 | */ |
79741b3b | 8333 | insn->imm = 0; |
71189fa9 | 8334 | insn->code = BPF_JMP | BPF_TAIL_CALL; |
b2157399 | 8335 | |
c93552c4 DB |
8336 | aux = &env->insn_aux_data[i + delta]; |
8337 | if (!bpf_map_ptr_unpriv(aux)) | |
8338 | continue; | |
8339 | ||
b2157399 AS |
8340 | /* instead of changing every JIT dealing with tail_call |
8341 | * emit two extra insns: | |
8342 | * if (index >= max_entries) goto out; | |
8343 | * index &= array->index_mask; | |
8344 | * to avoid out-of-bounds cpu speculation | |
8345 | */ | |
c93552c4 | 8346 | if (bpf_map_ptr_poisoned(aux)) { |
40950343 | 8347 | verbose(env, "tail_call abusing map_ptr\n"); |
b2157399 AS |
8348 | return -EINVAL; |
8349 | } | |
c93552c4 DB |
8350 | |
8351 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
b2157399 AS |
8352 | insn_buf[0] = BPF_JMP_IMM(BPF_JGE, BPF_REG_3, |
8353 | map_ptr->max_entries, 2); | |
8354 | insn_buf[1] = BPF_ALU32_IMM(BPF_AND, BPF_REG_3, | |
8355 | container_of(map_ptr, | |
8356 | struct bpf_array, | |
8357 | map)->index_mask); | |
8358 | insn_buf[2] = *insn; | |
8359 | cnt = 3; | |
8360 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); | |
8361 | if (!new_prog) | |
8362 | return -ENOMEM; | |
8363 | ||
8364 | delta += cnt - 1; | |
8365 | env->prog = prog = new_prog; | |
8366 | insn = new_prog->insnsi + i + delta; | |
79741b3b AS |
8367 | continue; |
8368 | } | |
e245c5c6 | 8369 | |
89c63074 | 8370 | /* BPF_EMIT_CALL() assumptions in some of the map_gen_lookup |
09772d92 DB |
8371 | * and other inlining handlers are currently limited to 64 bit |
8372 | * only. | |
89c63074 | 8373 | */ |
60b58afc | 8374 | if (prog->jit_requested && BITS_PER_LONG == 64 && |
09772d92 DB |
8375 | (insn->imm == BPF_FUNC_map_lookup_elem || |
8376 | insn->imm == BPF_FUNC_map_update_elem || | |
84430d42 DB |
8377 | insn->imm == BPF_FUNC_map_delete_elem || |
8378 | insn->imm == BPF_FUNC_map_push_elem || | |
8379 | insn->imm == BPF_FUNC_map_pop_elem || | |
8380 | insn->imm == BPF_FUNC_map_peek_elem)) { | |
c93552c4 DB |
8381 | aux = &env->insn_aux_data[i + delta]; |
8382 | if (bpf_map_ptr_poisoned(aux)) | |
8383 | goto patch_call_imm; | |
8384 | ||
8385 | map_ptr = BPF_MAP_PTR(aux->map_state); | |
09772d92 DB |
8386 | ops = map_ptr->ops; |
8387 | if (insn->imm == BPF_FUNC_map_lookup_elem && | |
8388 | ops->map_gen_lookup) { | |
8389 | cnt = ops->map_gen_lookup(map_ptr, insn_buf); | |
8390 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
8391 | verbose(env, "bpf verifier is misconfigured\n"); | |
8392 | return -EINVAL; | |
8393 | } | |
81ed18ab | 8394 | |
09772d92 DB |
8395 | new_prog = bpf_patch_insn_data(env, i + delta, |
8396 | insn_buf, cnt); | |
8397 | if (!new_prog) | |
8398 | return -ENOMEM; | |
81ed18ab | 8399 | |
09772d92 DB |
8400 | delta += cnt - 1; |
8401 | env->prog = prog = new_prog; | |
8402 | insn = new_prog->insnsi + i + delta; | |
8403 | continue; | |
8404 | } | |
81ed18ab | 8405 | |
09772d92 DB |
8406 | BUILD_BUG_ON(!__same_type(ops->map_lookup_elem, |
8407 | (void *(*)(struct bpf_map *map, void *key))NULL)); | |
8408 | BUILD_BUG_ON(!__same_type(ops->map_delete_elem, | |
8409 | (int (*)(struct bpf_map *map, void *key))NULL)); | |
8410 | BUILD_BUG_ON(!__same_type(ops->map_update_elem, | |
8411 | (int (*)(struct bpf_map *map, void *key, void *value, | |
8412 | u64 flags))NULL)); | |
84430d42 DB |
8413 | BUILD_BUG_ON(!__same_type(ops->map_push_elem, |
8414 | (int (*)(struct bpf_map *map, void *value, | |
8415 | u64 flags))NULL)); | |
8416 | BUILD_BUG_ON(!__same_type(ops->map_pop_elem, | |
8417 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
8418 | BUILD_BUG_ON(!__same_type(ops->map_peek_elem, | |
8419 | (int (*)(struct bpf_map *map, void *value))NULL)); | |
8420 | ||
09772d92 DB |
8421 | switch (insn->imm) { |
8422 | case BPF_FUNC_map_lookup_elem: | |
8423 | insn->imm = BPF_CAST_CALL(ops->map_lookup_elem) - | |
8424 | __bpf_call_base; | |
8425 | continue; | |
8426 | case BPF_FUNC_map_update_elem: | |
8427 | insn->imm = BPF_CAST_CALL(ops->map_update_elem) - | |
8428 | __bpf_call_base; | |
8429 | continue; | |
8430 | case BPF_FUNC_map_delete_elem: | |
8431 | insn->imm = BPF_CAST_CALL(ops->map_delete_elem) - | |
8432 | __bpf_call_base; | |
8433 | continue; | |
84430d42 DB |
8434 | case BPF_FUNC_map_push_elem: |
8435 | insn->imm = BPF_CAST_CALL(ops->map_push_elem) - | |
8436 | __bpf_call_base; | |
8437 | continue; | |
8438 | case BPF_FUNC_map_pop_elem: | |
8439 | insn->imm = BPF_CAST_CALL(ops->map_pop_elem) - | |
8440 | __bpf_call_base; | |
8441 | continue; | |
8442 | case BPF_FUNC_map_peek_elem: | |
8443 | insn->imm = BPF_CAST_CALL(ops->map_peek_elem) - | |
8444 | __bpf_call_base; | |
8445 | continue; | |
09772d92 | 8446 | } |
81ed18ab | 8447 | |
09772d92 | 8448 | goto patch_call_imm; |
81ed18ab AS |
8449 | } |
8450 | ||
8451 | patch_call_imm: | |
5e43f899 | 8452 | fn = env->ops->get_func_proto(insn->imm, env->prog); |
79741b3b AS |
8453 | /* all functions that have prototype and verifier allowed |
8454 | * programs to call them, must be real in-kernel functions | |
8455 | */ | |
8456 | if (!fn->func) { | |
61bd5218 JK |
8457 | verbose(env, |
8458 | "kernel subsystem misconfigured func %s#%d\n", | |
79741b3b AS |
8459 | func_id_name(insn->imm), insn->imm); |
8460 | return -EFAULT; | |
e245c5c6 | 8461 | } |
79741b3b | 8462 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 8463 | } |
e245c5c6 | 8464 | |
79741b3b AS |
8465 | return 0; |
8466 | } | |
e245c5c6 | 8467 | |
58e2af8b | 8468 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 8469 | { |
58e2af8b | 8470 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
8471 | int i; |
8472 | ||
9f4686c4 AS |
8473 | sl = env->free_list; |
8474 | while (sl) { | |
8475 | sln = sl->next; | |
8476 | free_verifier_state(&sl->state, false); | |
8477 | kfree(sl); | |
8478 | sl = sln; | |
8479 | } | |
8480 | ||
f1bca824 AS |
8481 | if (!env->explored_states) |
8482 | return; | |
8483 | ||
dc2a4ebc | 8484 | for (i = 0; i < state_htab_size(env); i++) { |
f1bca824 AS |
8485 | sl = env->explored_states[i]; |
8486 | ||
a8f500af AS |
8487 | while (sl) { |
8488 | sln = sl->next; | |
8489 | free_verifier_state(&sl->state, false); | |
8490 | kfree(sl); | |
8491 | sl = sln; | |
8492 | } | |
f1bca824 AS |
8493 | } |
8494 | ||
71dde681 | 8495 | kvfree(env->explored_states); |
f1bca824 AS |
8496 | } |
8497 | ||
06ee7115 AS |
8498 | static void print_verification_stats(struct bpf_verifier_env *env) |
8499 | { | |
8500 | int i; | |
8501 | ||
8502 | if (env->log.level & BPF_LOG_STATS) { | |
8503 | verbose(env, "verification time %lld usec\n", | |
8504 | div_u64(env->verification_time, 1000)); | |
8505 | verbose(env, "stack depth "); | |
8506 | for (i = 0; i < env->subprog_cnt; i++) { | |
8507 | u32 depth = env->subprog_info[i].stack_depth; | |
8508 | ||
8509 | verbose(env, "%d", depth); | |
8510 | if (i + 1 < env->subprog_cnt) | |
8511 | verbose(env, "+"); | |
8512 | } | |
8513 | verbose(env, "\n"); | |
8514 | } | |
8515 | verbose(env, "processed %d insns (limit %d) max_states_per_insn %d " | |
8516 | "total_states %d peak_states %d mark_read %d\n", | |
8517 | env->insn_processed, BPF_COMPLEXITY_LIMIT_INSNS, | |
8518 | env->max_states_per_insn, env->total_states, | |
8519 | env->peak_states, env->longest_mark_read_walk); | |
f1bca824 AS |
8520 | } |
8521 | ||
838e9690 YS |
8522 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr, |
8523 | union bpf_attr __user *uattr) | |
51580e79 | 8524 | { |
06ee7115 | 8525 | u64 start_time = ktime_get_ns(); |
58e2af8b | 8526 | struct bpf_verifier_env *env; |
b9193c1b | 8527 | struct bpf_verifier_log *log; |
9e4c24e7 | 8528 | int i, len, ret = -EINVAL; |
e2ae4ca2 | 8529 | bool is_priv; |
51580e79 | 8530 | |
eba0c929 AB |
8531 | /* no program is valid */ |
8532 | if (ARRAY_SIZE(bpf_verifier_ops) == 0) | |
8533 | return -EINVAL; | |
8534 | ||
58e2af8b | 8535 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
8536 | * allocate/free it every time bpf_check() is called |
8537 | */ | |
58e2af8b | 8538 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
8539 | if (!env) |
8540 | return -ENOMEM; | |
61bd5218 | 8541 | log = &env->log; |
cbd35700 | 8542 | |
9e4c24e7 | 8543 | len = (*prog)->len; |
fad953ce | 8544 | env->insn_aux_data = |
9e4c24e7 | 8545 | vzalloc(array_size(sizeof(struct bpf_insn_aux_data), len)); |
3df126f3 JK |
8546 | ret = -ENOMEM; |
8547 | if (!env->insn_aux_data) | |
8548 | goto err_free_env; | |
9e4c24e7 JK |
8549 | for (i = 0; i < len; i++) |
8550 | env->insn_aux_data[i].orig_idx = i; | |
9bac3d6d | 8551 | env->prog = *prog; |
00176a34 | 8552 | env->ops = bpf_verifier_ops[env->prog->type]; |
45a73c17 | 8553 | is_priv = capable(CAP_SYS_ADMIN); |
0246e64d | 8554 | |
cbd35700 | 8555 | /* grab the mutex to protect few globals used by verifier */ |
45a73c17 AS |
8556 | if (!is_priv) |
8557 | mutex_lock(&bpf_verifier_lock); | |
cbd35700 AS |
8558 | |
8559 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
8560 | /* user requested verbose verifier output | |
8561 | * and supplied buffer to store the verification trace | |
8562 | */ | |
e7bf8249 JK |
8563 | log->level = attr->log_level; |
8564 | log->ubuf = (char __user *) (unsigned long) attr->log_buf; | |
8565 | log->len_total = attr->log_size; | |
cbd35700 AS |
8566 | |
8567 | ret = -EINVAL; | |
e7bf8249 | 8568 | /* log attributes have to be sane */ |
7a9f5c65 | 8569 | if (log->len_total < 128 || log->len_total > UINT_MAX >> 2 || |
06ee7115 | 8570 | !log->level || !log->ubuf || log->level & ~BPF_LOG_MASK) |
3df126f3 | 8571 | goto err_unlock; |
cbd35700 | 8572 | } |
1ad2f583 DB |
8573 | |
8574 | env->strict_alignment = !!(attr->prog_flags & BPF_F_STRICT_ALIGNMENT); | |
8575 | if (!IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
e07b98d9 | 8576 | env->strict_alignment = true; |
e9ee9efc DM |
8577 | if (attr->prog_flags & BPF_F_ANY_ALIGNMENT) |
8578 | env->strict_alignment = false; | |
cbd35700 | 8579 | |
e2ae4ca2 JK |
8580 | env->allow_ptr_leaks = is_priv; |
8581 | ||
f4e3ec0d JK |
8582 | ret = replace_map_fd_with_map_ptr(env); |
8583 | if (ret < 0) | |
8584 | goto skip_full_check; | |
8585 | ||
cae1927c | 8586 | if (bpf_prog_is_dev_bound(env->prog->aux)) { |
a40a2632 | 8587 | ret = bpf_prog_offload_verifier_prep(env->prog); |
ab3f0063 | 8588 | if (ret) |
f4e3ec0d | 8589 | goto skip_full_check; |
ab3f0063 JK |
8590 | } |
8591 | ||
dc2a4ebc | 8592 | env->explored_states = kvcalloc(state_htab_size(env), |
58e2af8b | 8593 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
8594 | GFP_USER); |
8595 | ret = -ENOMEM; | |
8596 | if (!env->explored_states) | |
8597 | goto skip_full_check; | |
8598 | ||
d9762e84 | 8599 | ret = check_subprogs(env); |
475fb78f AS |
8600 | if (ret < 0) |
8601 | goto skip_full_check; | |
8602 | ||
c454a46b | 8603 | ret = check_btf_info(env, attr, uattr); |
838e9690 YS |
8604 | if (ret < 0) |
8605 | goto skip_full_check; | |
8606 | ||
d9762e84 MKL |
8607 | ret = check_cfg(env); |
8608 | if (ret < 0) | |
8609 | goto skip_full_check; | |
8610 | ||
17a52670 | 8611 | ret = do_check(env); |
8c01c4f8 CG |
8612 | if (env->cur_state) { |
8613 | free_verifier_state(env->cur_state, true); | |
8614 | env->cur_state = NULL; | |
8615 | } | |
cbd35700 | 8616 | |
c941ce9c QM |
8617 | if (ret == 0 && bpf_prog_is_dev_bound(env->prog->aux)) |
8618 | ret = bpf_prog_offload_finalize(env); | |
8619 | ||
0246e64d | 8620 | skip_full_check: |
638f5b90 | 8621 | while (!pop_stack(env, NULL, NULL)); |
f1bca824 | 8622 | free_states(env); |
0246e64d | 8623 | |
c131187d | 8624 | if (ret == 0) |
9b38c405 | 8625 | ret = check_max_stack_depth(env); |
c131187d | 8626 | |
9b38c405 | 8627 | /* instruction rewrites happen after this point */ |
e2ae4ca2 JK |
8628 | if (is_priv) { |
8629 | if (ret == 0) | |
8630 | opt_hard_wire_dead_code_branches(env); | |
52875a04 JK |
8631 | if (ret == 0) |
8632 | ret = opt_remove_dead_code(env); | |
a1b14abc JK |
8633 | if (ret == 0) |
8634 | ret = opt_remove_nops(env); | |
52875a04 JK |
8635 | } else { |
8636 | if (ret == 0) | |
8637 | sanitize_dead_code(env); | |
e2ae4ca2 JK |
8638 | } |
8639 | ||
9bac3d6d AS |
8640 | if (ret == 0) |
8641 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
8642 | ret = convert_ctx_accesses(env); | |
8643 | ||
e245c5c6 | 8644 | if (ret == 0) |
79741b3b | 8645 | ret = fixup_bpf_calls(env); |
e245c5c6 | 8646 | |
a4b1d3c1 JW |
8647 | /* do 32-bit optimization after insn patching has done so those patched |
8648 | * insns could be handled correctly. | |
8649 | */ | |
d6c2308c JW |
8650 | if (ret == 0 && !bpf_prog_is_dev_bound(env->prog->aux)) { |
8651 | ret = opt_subreg_zext_lo32_rnd_hi32(env, attr); | |
8652 | env->prog->aux->verifier_zext = bpf_jit_needs_zext() ? !ret | |
8653 | : false; | |
a4b1d3c1 JW |
8654 | } |
8655 | ||
1ea47e01 AS |
8656 | if (ret == 0) |
8657 | ret = fixup_call_args(env); | |
8658 | ||
06ee7115 AS |
8659 | env->verification_time = ktime_get_ns() - start_time; |
8660 | print_verification_stats(env); | |
8661 | ||
a2a7d570 | 8662 | if (log->level && bpf_verifier_log_full(log)) |
cbd35700 | 8663 | ret = -ENOSPC; |
a2a7d570 | 8664 | if (log->level && !log->ubuf) { |
cbd35700 | 8665 | ret = -EFAULT; |
a2a7d570 | 8666 | goto err_release_maps; |
cbd35700 AS |
8667 | } |
8668 | ||
0246e64d AS |
8669 | if (ret == 0 && env->used_map_cnt) { |
8670 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
8671 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
8672 | sizeof(env->used_maps[0]), | |
8673 | GFP_KERNEL); | |
0246e64d | 8674 | |
9bac3d6d | 8675 | if (!env->prog->aux->used_maps) { |
0246e64d | 8676 | ret = -ENOMEM; |
a2a7d570 | 8677 | goto err_release_maps; |
0246e64d AS |
8678 | } |
8679 | ||
9bac3d6d | 8680 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 8681 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 8682 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
8683 | |
8684 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
8685 | * bpf_ld_imm64 instructions | |
8686 | */ | |
8687 | convert_pseudo_ld_imm64(env); | |
8688 | } | |
cbd35700 | 8689 | |
ba64e7d8 YS |
8690 | if (ret == 0) |
8691 | adjust_btf_func(env); | |
8692 | ||
a2a7d570 | 8693 | err_release_maps: |
9bac3d6d | 8694 | if (!env->prog->aux->used_maps) |
0246e64d | 8695 | /* if we didn't copy map pointers into bpf_prog_info, release |
ab7f5bf0 | 8696 | * them now. Otherwise free_used_maps() will release them. |
0246e64d AS |
8697 | */ |
8698 | release_maps(env); | |
9bac3d6d | 8699 | *prog = env->prog; |
3df126f3 | 8700 | err_unlock: |
45a73c17 AS |
8701 | if (!is_priv) |
8702 | mutex_unlock(&bpf_verifier_lock); | |
3df126f3 JK |
8703 | vfree(env->insn_aux_data); |
8704 | err_free_env: | |
8705 | kfree(env); | |
51580e79 AS |
8706 | return ret; |
8707 | } |