Commit | Line | Data |
---|---|---|
51580e79 | 1 | /* Copyright (c) 2011-2014 PLUMgrid, http://plumgrid.com |
969bf05e | 2 | * Copyright (c) 2016 Facebook |
51580e79 AS |
3 | * |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | */ | |
13 | #include <linux/kernel.h> | |
14 | #include <linux/types.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/bpf.h> | |
58e2af8b | 17 | #include <linux/bpf_verifier.h> |
51580e79 AS |
18 | #include <linux/filter.h> |
19 | #include <net/netlink.h> | |
20 | #include <linux/file.h> | |
21 | #include <linux/vmalloc.h> | |
ebb676da | 22 | #include <linux/stringify.h> |
51580e79 AS |
23 | |
24 | /* bpf_check() is a static code analyzer that walks eBPF program | |
25 | * instruction by instruction and updates register/stack state. | |
26 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
27 | * | |
28 | * The first pass is depth-first-search to check that the program is a DAG. | |
29 | * It rejects the following programs: | |
30 | * - larger than BPF_MAXINSNS insns | |
31 | * - if loop is present (detected via back-edge) | |
32 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
33 | * - out of bounds or malformed jumps | |
34 | * The second pass is all possible path descent from the 1st insn. | |
35 | * Since it's analyzing all pathes through the program, the length of the | |
eba38a96 | 36 | * analysis is limited to 64k insn, which may be hit even if total number of |
51580e79 AS |
37 | * insn is less then 4K, but there are too many branches that change stack/regs. |
38 | * Number of 'branches to be analyzed' is limited to 1k | |
39 | * | |
40 | * On entry to each instruction, each register has a type, and the instruction | |
41 | * changes the types of the registers depending on instruction semantics. | |
42 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
43 | * copied to R1. | |
44 | * | |
45 | * All registers are 64-bit. | |
46 | * R0 - return register | |
47 | * R1-R5 argument passing registers | |
48 | * R6-R9 callee saved registers | |
49 | * R10 - frame pointer read-only | |
50 | * | |
51 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
52 | * and has type PTR_TO_CTX. | |
53 | * | |
54 | * Verifier tracks arithmetic operations on pointers in case: | |
55 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
56 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
57 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
58 | * and 2nd arithmetic instruction is pattern matched to recognize | |
59 | * that it wants to construct a pointer to some element within stack. | |
60 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
61 | * (and -20 constant is saved for further stack bounds checking). | |
62 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
63 | * | |
64 | * Most of the time the registers have UNKNOWN_VALUE type, which | |
65 | * means the register has some value, but it's not a valid pointer. | |
66 | * (like pointer plus pointer becomes UNKNOWN_VALUE type) | |
67 | * | |
68 | * When verifier sees load or store instructions the type of base register | |
69 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer | |
70 | * types recognized by check_mem_access() function. | |
71 | * | |
72 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
73 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
74 | * | |
75 | * registers used to pass values to function calls are checked against | |
76 | * function argument constraints. | |
77 | * | |
78 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
79 | * It means that the register type passed to this function must be | |
80 | * PTR_TO_STACK and it will be used inside the function as | |
81 | * 'pointer to map element key' | |
82 | * | |
83 | * For example the argument constraints for bpf_map_lookup_elem(): | |
84 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
85 | * .arg1_type = ARG_CONST_MAP_PTR, | |
86 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
87 | * | |
88 | * ret_type says that this function returns 'pointer to map elem value or null' | |
89 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
90 | * 2nd argument should be a pointer to stack, which will be used inside | |
91 | * the helper function as a pointer to map element key. | |
92 | * | |
93 | * On the kernel side the helper function looks like: | |
94 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
95 | * { | |
96 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
97 | * void *key = (void *) (unsigned long) r2; | |
98 | * void *value; | |
99 | * | |
100 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
101 | * [key, key + map->key_size) bytes are valid and were initialized on | |
102 | * the stack of eBPF program. | |
103 | * } | |
104 | * | |
105 | * Corresponding eBPF program may look like: | |
106 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
107 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
108 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
109 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
110 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
111 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
112 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
113 | * | |
114 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
115 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
116 | * and were initialized prior to this call. | |
117 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
118 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
119 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
120 | * returns ether pointer to map value or NULL. | |
121 | * | |
122 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
123 | * insn, the register holding that pointer in the true branch changes state to | |
124 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
125 | * branch. See check_cond_jmp_op(). | |
126 | * | |
127 | * After the call R0 is set to return type of the function and registers R1-R5 | |
128 | * are set to NOT_INIT to indicate that they are no longer readable. | |
129 | */ | |
130 | ||
17a52670 | 131 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 132 | struct bpf_verifier_stack_elem { |
17a52670 AS |
133 | /* verifer state is 'st' |
134 | * before processing instruction 'insn_idx' | |
135 | * and after processing instruction 'prev_insn_idx' | |
136 | */ | |
58e2af8b | 137 | struct bpf_verifier_state st; |
17a52670 AS |
138 | int insn_idx; |
139 | int prev_insn_idx; | |
58e2af8b | 140 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
141 | }; |
142 | ||
07016151 DB |
143 | #define BPF_COMPLEXITY_LIMIT_INSNS 65536 |
144 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 | |
145 | ||
fad73a1a MKL |
146 | #define BPF_MAP_PTR_POISON ((void *)0xeB9F + POISON_POINTER_DELTA) |
147 | ||
33ff9823 DB |
148 | struct bpf_call_arg_meta { |
149 | struct bpf_map *map_ptr; | |
435faee1 | 150 | bool raw_mode; |
36bbef52 | 151 | bool pkt_access; |
435faee1 DB |
152 | int regno; |
153 | int access_size; | |
33ff9823 DB |
154 | }; |
155 | ||
cbd35700 AS |
156 | /* verbose verifier prints what it's seeing |
157 | * bpf_check() is called under lock, so no race to access these global vars | |
158 | */ | |
159 | static u32 log_level, log_size, log_len; | |
160 | static char *log_buf; | |
161 | ||
162 | static DEFINE_MUTEX(bpf_verifier_lock); | |
163 | ||
164 | /* log_level controls verbosity level of eBPF verifier. | |
165 | * verbose() is used to dump the verification trace to the log, so the user | |
166 | * can figure out what's wrong with the program | |
167 | */ | |
1d056d9c | 168 | static __printf(1, 2) void verbose(const char *fmt, ...) |
cbd35700 AS |
169 | { |
170 | va_list args; | |
171 | ||
172 | if (log_level == 0 || log_len >= log_size - 1) | |
173 | return; | |
174 | ||
175 | va_start(args, fmt); | |
176 | log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); | |
177 | va_end(args); | |
178 | } | |
179 | ||
17a52670 AS |
180 | /* string representation of 'enum bpf_reg_type' */ |
181 | static const char * const reg_type_str[] = { | |
182 | [NOT_INIT] = "?", | |
183 | [UNKNOWN_VALUE] = "inv", | |
184 | [PTR_TO_CTX] = "ctx", | |
185 | [CONST_PTR_TO_MAP] = "map_ptr", | |
186 | [PTR_TO_MAP_VALUE] = "map_value", | |
187 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
48461135 | 188 | [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj", |
17a52670 AS |
189 | [FRAME_PTR] = "fp", |
190 | [PTR_TO_STACK] = "fp", | |
191 | [CONST_IMM] = "imm", | |
969bf05e AS |
192 | [PTR_TO_PACKET] = "pkt", |
193 | [PTR_TO_PACKET_END] = "pkt_end", | |
17a52670 AS |
194 | }; |
195 | ||
ebb676da TG |
196 | #define __BPF_FUNC_STR_FN(x) [BPF_FUNC_ ## x] = __stringify(bpf_ ## x) |
197 | static const char * const func_id_str[] = { | |
198 | __BPF_FUNC_MAPPER(__BPF_FUNC_STR_FN) | |
199 | }; | |
200 | #undef __BPF_FUNC_STR_FN | |
201 | ||
202 | static const char *func_id_name(int id) | |
203 | { | |
204 | BUILD_BUG_ON(ARRAY_SIZE(func_id_str) != __BPF_FUNC_MAX_ID); | |
205 | ||
206 | if (id >= 0 && id < __BPF_FUNC_MAX_ID && func_id_str[id]) | |
207 | return func_id_str[id]; | |
208 | else | |
209 | return "unknown"; | |
210 | } | |
211 | ||
58e2af8b | 212 | static void print_verifier_state(struct bpf_verifier_state *state) |
17a52670 | 213 | { |
58e2af8b | 214 | struct bpf_reg_state *reg; |
17a52670 AS |
215 | enum bpf_reg_type t; |
216 | int i; | |
217 | ||
218 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
219 | reg = &state->regs[i]; |
220 | t = reg->type; | |
17a52670 AS |
221 | if (t == NOT_INIT) |
222 | continue; | |
223 | verbose(" R%d=%s", i, reg_type_str[t]); | |
224 | if (t == CONST_IMM || t == PTR_TO_STACK) | |
969bf05e AS |
225 | verbose("%lld", reg->imm); |
226 | else if (t == PTR_TO_PACKET) | |
227 | verbose("(id=%d,off=%d,r=%d)", | |
228 | reg->id, reg->off, reg->range); | |
229 | else if (t == UNKNOWN_VALUE && reg->imm) | |
230 | verbose("%lld", reg->imm); | |
17a52670 | 231 | else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || |
48461135 JB |
232 | t == PTR_TO_MAP_VALUE_OR_NULL || |
233 | t == PTR_TO_MAP_VALUE_ADJ) | |
57a09bf0 | 234 | verbose("(ks=%d,vs=%d,id=%u)", |
1a0dc1ac | 235 | reg->map_ptr->key_size, |
57a09bf0 TG |
236 | reg->map_ptr->value_size, |
237 | reg->id); | |
48461135 | 238 | if (reg->min_value != BPF_REGISTER_MIN_RANGE) |
f23cc643 JB |
239 | verbose(",min_value=%lld", |
240 | (long long)reg->min_value); | |
48461135 JB |
241 | if (reg->max_value != BPF_REGISTER_MAX_RANGE) |
242 | verbose(",max_value=%llu", | |
243 | (unsigned long long)reg->max_value); | |
17a52670 | 244 | } |
9c399760 | 245 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { |
1a0dc1ac | 246 | if (state->stack_slot_type[i] == STACK_SPILL) |
17a52670 | 247 | verbose(" fp%d=%s", -MAX_BPF_STACK + i, |
1a0dc1ac | 248 | reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); |
17a52670 AS |
249 | } |
250 | verbose("\n"); | |
251 | } | |
252 | ||
cbd35700 AS |
253 | static const char *const bpf_class_string[] = { |
254 | [BPF_LD] = "ld", | |
255 | [BPF_LDX] = "ldx", | |
256 | [BPF_ST] = "st", | |
257 | [BPF_STX] = "stx", | |
258 | [BPF_ALU] = "alu", | |
259 | [BPF_JMP] = "jmp", | |
260 | [BPF_RET] = "BUG", | |
261 | [BPF_ALU64] = "alu64", | |
262 | }; | |
263 | ||
687f0715 | 264 | static const char *const bpf_alu_string[16] = { |
cbd35700 AS |
265 | [BPF_ADD >> 4] = "+=", |
266 | [BPF_SUB >> 4] = "-=", | |
267 | [BPF_MUL >> 4] = "*=", | |
268 | [BPF_DIV >> 4] = "/=", | |
269 | [BPF_OR >> 4] = "|=", | |
270 | [BPF_AND >> 4] = "&=", | |
271 | [BPF_LSH >> 4] = "<<=", | |
272 | [BPF_RSH >> 4] = ">>=", | |
273 | [BPF_NEG >> 4] = "neg", | |
274 | [BPF_MOD >> 4] = "%=", | |
275 | [BPF_XOR >> 4] = "^=", | |
276 | [BPF_MOV >> 4] = "=", | |
277 | [BPF_ARSH >> 4] = "s>>=", | |
278 | [BPF_END >> 4] = "endian", | |
279 | }; | |
280 | ||
281 | static const char *const bpf_ldst_string[] = { | |
282 | [BPF_W >> 3] = "u32", | |
283 | [BPF_H >> 3] = "u16", | |
284 | [BPF_B >> 3] = "u8", | |
285 | [BPF_DW >> 3] = "u64", | |
286 | }; | |
287 | ||
687f0715 | 288 | static const char *const bpf_jmp_string[16] = { |
cbd35700 AS |
289 | [BPF_JA >> 4] = "jmp", |
290 | [BPF_JEQ >> 4] = "==", | |
291 | [BPF_JGT >> 4] = ">", | |
292 | [BPF_JGE >> 4] = ">=", | |
293 | [BPF_JSET >> 4] = "&", | |
294 | [BPF_JNE >> 4] = "!=", | |
295 | [BPF_JSGT >> 4] = "s>", | |
296 | [BPF_JSGE >> 4] = "s>=", | |
297 | [BPF_CALL >> 4] = "call", | |
298 | [BPF_EXIT >> 4] = "exit", | |
299 | }; | |
300 | ||
0d0e5769 DB |
301 | static void print_bpf_insn(const struct bpf_verifier_env *env, |
302 | const struct bpf_insn *insn) | |
cbd35700 AS |
303 | { |
304 | u8 class = BPF_CLASS(insn->code); | |
305 | ||
306 | if (class == BPF_ALU || class == BPF_ALU64) { | |
307 | if (BPF_SRC(insn->code) == BPF_X) | |
308 | verbose("(%02x) %sr%d %s %sr%d\n", | |
309 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
310 | insn->dst_reg, | |
311 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
312 | class == BPF_ALU ? "(u32) " : "", | |
313 | insn->src_reg); | |
314 | else | |
315 | verbose("(%02x) %sr%d %s %s%d\n", | |
316 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
317 | insn->dst_reg, | |
318 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
319 | class == BPF_ALU ? "(u32) " : "", | |
320 | insn->imm); | |
321 | } else if (class == BPF_STX) { | |
322 | if (BPF_MODE(insn->code) == BPF_MEM) | |
323 | verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", | |
324 | insn->code, | |
325 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
326 | insn->dst_reg, | |
327 | insn->off, insn->src_reg); | |
328 | else if (BPF_MODE(insn->code) == BPF_XADD) | |
329 | verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", | |
330 | insn->code, | |
331 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
332 | insn->dst_reg, insn->off, | |
333 | insn->src_reg); | |
334 | else | |
335 | verbose("BUG_%02x\n", insn->code); | |
336 | } else if (class == BPF_ST) { | |
337 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
338 | verbose("BUG_st_%02x\n", insn->code); | |
339 | return; | |
340 | } | |
341 | verbose("(%02x) *(%s *)(r%d %+d) = %d\n", | |
342 | insn->code, | |
343 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
344 | insn->dst_reg, | |
345 | insn->off, insn->imm); | |
346 | } else if (class == BPF_LDX) { | |
347 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
348 | verbose("BUG_ldx_%02x\n", insn->code); | |
349 | return; | |
350 | } | |
351 | verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", | |
352 | insn->code, insn->dst_reg, | |
353 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
354 | insn->src_reg, insn->off); | |
355 | } else if (class == BPF_LD) { | |
356 | if (BPF_MODE(insn->code) == BPF_ABS) { | |
357 | verbose("(%02x) r0 = *(%s *)skb[%d]\n", | |
358 | insn->code, | |
359 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
360 | insn->imm); | |
361 | } else if (BPF_MODE(insn->code) == BPF_IND) { | |
362 | verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", | |
363 | insn->code, | |
364 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
365 | insn->src_reg, insn->imm); | |
0d0e5769 DB |
366 | } else if (BPF_MODE(insn->code) == BPF_IMM && |
367 | BPF_SIZE(insn->code) == BPF_DW) { | |
368 | /* At this point, we already made sure that the second | |
369 | * part of the ldimm64 insn is accessible. | |
370 | */ | |
371 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; | |
372 | bool map_ptr = insn->src_reg == BPF_PSEUDO_MAP_FD; | |
373 | ||
374 | if (map_ptr && !env->allow_ptr_leaks) | |
375 | imm = 0; | |
376 | ||
377 | verbose("(%02x) r%d = 0x%llx\n", insn->code, | |
378 | insn->dst_reg, (unsigned long long)imm); | |
cbd35700 AS |
379 | } else { |
380 | verbose("BUG_ld_%02x\n", insn->code); | |
381 | return; | |
382 | } | |
383 | } else if (class == BPF_JMP) { | |
384 | u8 opcode = BPF_OP(insn->code); | |
385 | ||
386 | if (opcode == BPF_CALL) { | |
ebb676da TG |
387 | verbose("(%02x) call %s#%d\n", insn->code, |
388 | func_id_name(insn->imm), insn->imm); | |
cbd35700 AS |
389 | } else if (insn->code == (BPF_JMP | BPF_JA)) { |
390 | verbose("(%02x) goto pc%+d\n", | |
391 | insn->code, insn->off); | |
392 | } else if (insn->code == (BPF_JMP | BPF_EXIT)) { | |
393 | verbose("(%02x) exit\n", insn->code); | |
394 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
395 | verbose("(%02x) if r%d %s r%d goto pc%+d\n", | |
396 | insn->code, insn->dst_reg, | |
397 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
398 | insn->src_reg, insn->off); | |
399 | } else { | |
400 | verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", | |
401 | insn->code, insn->dst_reg, | |
402 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
403 | insn->imm, insn->off); | |
404 | } | |
405 | } else { | |
406 | verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); | |
407 | } | |
408 | } | |
409 | ||
58e2af8b | 410 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) |
17a52670 | 411 | { |
58e2af8b | 412 | struct bpf_verifier_stack_elem *elem; |
17a52670 AS |
413 | int insn_idx; |
414 | ||
415 | if (env->head == NULL) | |
416 | return -1; | |
417 | ||
418 | memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); | |
419 | insn_idx = env->head->insn_idx; | |
420 | if (prev_insn_idx) | |
421 | *prev_insn_idx = env->head->prev_insn_idx; | |
422 | elem = env->head->next; | |
423 | kfree(env->head); | |
424 | env->head = elem; | |
425 | env->stack_size--; | |
426 | return insn_idx; | |
427 | } | |
428 | ||
58e2af8b JK |
429 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
430 | int insn_idx, int prev_insn_idx) | |
17a52670 | 431 | { |
58e2af8b | 432 | struct bpf_verifier_stack_elem *elem; |
17a52670 | 433 | |
58e2af8b | 434 | elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
435 | if (!elem) |
436 | goto err; | |
437 | ||
438 | memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); | |
439 | elem->insn_idx = insn_idx; | |
440 | elem->prev_insn_idx = prev_insn_idx; | |
441 | elem->next = env->head; | |
442 | env->head = elem; | |
443 | env->stack_size++; | |
07016151 | 444 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
17a52670 AS |
445 | verbose("BPF program is too complex\n"); |
446 | goto err; | |
447 | } | |
448 | return &elem->st; | |
449 | err: | |
450 | /* pop all elements and return */ | |
451 | while (pop_stack(env, NULL) >= 0); | |
452 | return NULL; | |
453 | } | |
454 | ||
455 | #define CALLER_SAVED_REGS 6 | |
456 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
457 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
458 | }; | |
459 | ||
58e2af8b | 460 | static void init_reg_state(struct bpf_reg_state *regs) |
17a52670 AS |
461 | { |
462 | int i; | |
463 | ||
464 | for (i = 0; i < MAX_BPF_REG; i++) { | |
465 | regs[i].type = NOT_INIT; | |
466 | regs[i].imm = 0; | |
48461135 JB |
467 | regs[i].min_value = BPF_REGISTER_MIN_RANGE; |
468 | regs[i].max_value = BPF_REGISTER_MAX_RANGE; | |
17a52670 AS |
469 | } |
470 | ||
471 | /* frame pointer */ | |
472 | regs[BPF_REG_FP].type = FRAME_PTR; | |
473 | ||
474 | /* 1st arg to a function */ | |
475 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
476 | } | |
477 | ||
6760bf2d | 478 | static void __mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) |
17a52670 | 479 | { |
17a52670 | 480 | regs[regno].type = UNKNOWN_VALUE; |
57a09bf0 | 481 | regs[regno].id = 0; |
17a52670 | 482 | regs[regno].imm = 0; |
17a52670 AS |
483 | } |
484 | ||
6760bf2d DB |
485 | static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) |
486 | { | |
487 | BUG_ON(regno >= MAX_BPF_REG); | |
488 | __mark_reg_unknown_value(regs, regno); | |
489 | } | |
490 | ||
48461135 JB |
491 | static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno) |
492 | { | |
493 | regs[regno].min_value = BPF_REGISTER_MIN_RANGE; | |
494 | regs[regno].max_value = BPF_REGISTER_MAX_RANGE; | |
495 | } | |
496 | ||
f0318d01 GB |
497 | static void mark_reg_unknown_value_and_range(struct bpf_reg_state *regs, |
498 | u32 regno) | |
499 | { | |
500 | mark_reg_unknown_value(regs, regno); | |
501 | reset_reg_range_values(regs, regno); | |
502 | } | |
503 | ||
17a52670 AS |
504 | enum reg_arg_type { |
505 | SRC_OP, /* register is used as source operand */ | |
506 | DST_OP, /* register is used as destination operand */ | |
507 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
508 | }; | |
509 | ||
58e2af8b | 510 | static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, |
17a52670 AS |
511 | enum reg_arg_type t) |
512 | { | |
513 | if (regno >= MAX_BPF_REG) { | |
514 | verbose("R%d is invalid\n", regno); | |
515 | return -EINVAL; | |
516 | } | |
517 | ||
518 | if (t == SRC_OP) { | |
519 | /* check whether register used as source operand can be read */ | |
520 | if (regs[regno].type == NOT_INIT) { | |
521 | verbose("R%d !read_ok\n", regno); | |
522 | return -EACCES; | |
523 | } | |
524 | } else { | |
525 | /* check whether register used as dest operand can be written to */ | |
526 | if (regno == BPF_REG_FP) { | |
527 | verbose("frame pointer is read only\n"); | |
528 | return -EACCES; | |
529 | } | |
530 | if (t == DST_OP) | |
531 | mark_reg_unknown_value(regs, regno); | |
532 | } | |
533 | return 0; | |
534 | } | |
535 | ||
536 | static int bpf_size_to_bytes(int bpf_size) | |
537 | { | |
538 | if (bpf_size == BPF_W) | |
539 | return 4; | |
540 | else if (bpf_size == BPF_H) | |
541 | return 2; | |
542 | else if (bpf_size == BPF_B) | |
543 | return 1; | |
544 | else if (bpf_size == BPF_DW) | |
545 | return 8; | |
546 | else | |
547 | return -EINVAL; | |
548 | } | |
549 | ||
1be7f75d AS |
550 | static bool is_spillable_regtype(enum bpf_reg_type type) |
551 | { | |
552 | switch (type) { | |
553 | case PTR_TO_MAP_VALUE: | |
554 | case PTR_TO_MAP_VALUE_OR_NULL: | |
f0318d01 | 555 | case PTR_TO_MAP_VALUE_ADJ: |
1be7f75d AS |
556 | case PTR_TO_STACK: |
557 | case PTR_TO_CTX: | |
969bf05e AS |
558 | case PTR_TO_PACKET: |
559 | case PTR_TO_PACKET_END: | |
1be7f75d AS |
560 | case FRAME_PTR: |
561 | case CONST_PTR_TO_MAP: | |
562 | return true; | |
563 | default: | |
564 | return false; | |
565 | } | |
566 | } | |
567 | ||
17a52670 AS |
568 | /* check_stack_read/write functions track spill/fill of registers, |
569 | * stack boundary and alignment are checked in check_mem_access() | |
570 | */ | |
58e2af8b JK |
571 | static int check_stack_write(struct bpf_verifier_state *state, int off, |
572 | int size, int value_regno) | |
17a52670 | 573 | { |
17a52670 | 574 | int i; |
9c399760 AS |
575 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
576 | * so it's aligned access and [off, off + size) are within stack limits | |
577 | */ | |
17a52670 AS |
578 | |
579 | if (value_regno >= 0 && | |
1be7f75d | 580 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
581 | |
582 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 583 | if (size != BPF_REG_SIZE) { |
17a52670 AS |
584 | verbose("invalid size of register spill\n"); |
585 | return -EACCES; | |
586 | } | |
587 | ||
17a52670 | 588 | /* save register state */ |
9c399760 AS |
589 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
590 | state->regs[value_regno]; | |
17a52670 | 591 | |
9c399760 AS |
592 | for (i = 0; i < BPF_REG_SIZE; i++) |
593 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; | |
594 | } else { | |
17a52670 | 595 | /* regular write of data into stack */ |
9c399760 | 596 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
58e2af8b | 597 | (struct bpf_reg_state) {}; |
9c399760 AS |
598 | |
599 | for (i = 0; i < size; i++) | |
600 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; | |
17a52670 AS |
601 | } |
602 | return 0; | |
603 | } | |
604 | ||
58e2af8b | 605 | static int check_stack_read(struct bpf_verifier_state *state, int off, int size, |
17a52670 AS |
606 | int value_regno) |
607 | { | |
9c399760 | 608 | u8 *slot_type; |
17a52670 | 609 | int i; |
17a52670 | 610 | |
9c399760 | 611 | slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; |
17a52670 | 612 | |
9c399760 AS |
613 | if (slot_type[0] == STACK_SPILL) { |
614 | if (size != BPF_REG_SIZE) { | |
17a52670 AS |
615 | verbose("invalid size of register spill\n"); |
616 | return -EACCES; | |
617 | } | |
9c399760 AS |
618 | for (i = 1; i < BPF_REG_SIZE; i++) { |
619 | if (slot_type[i] != STACK_SPILL) { | |
17a52670 AS |
620 | verbose("corrupted spill memory\n"); |
621 | return -EACCES; | |
622 | } | |
623 | } | |
624 | ||
625 | if (value_regno >= 0) | |
626 | /* restore register state from stack */ | |
9c399760 AS |
627 | state->regs[value_regno] = |
628 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; | |
17a52670 AS |
629 | return 0; |
630 | } else { | |
631 | for (i = 0; i < size; i++) { | |
9c399760 | 632 | if (slot_type[i] != STACK_MISC) { |
17a52670 AS |
633 | verbose("invalid read from stack off %d+%d size %d\n", |
634 | off, i, size); | |
635 | return -EACCES; | |
636 | } | |
637 | } | |
638 | if (value_regno >= 0) | |
639 | /* have read misc data from the stack */ | |
f0318d01 GB |
640 | mark_reg_unknown_value_and_range(state->regs, |
641 | value_regno); | |
17a52670 AS |
642 | return 0; |
643 | } | |
644 | } | |
645 | ||
646 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
58e2af8b | 647 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
648 | int size) |
649 | { | |
650 | struct bpf_map *map = env->cur_state.regs[regno].map_ptr; | |
651 | ||
5722569b | 652 | if (off < 0 || size <= 0 || off + size > map->value_size) { |
17a52670 AS |
653 | verbose("invalid access to map value, value_size=%d off=%d size=%d\n", |
654 | map->value_size, off, size); | |
655 | return -EACCES; | |
656 | } | |
657 | return 0; | |
658 | } | |
659 | ||
dbcfe5f7 GB |
660 | /* check read/write into an adjusted map element */ |
661 | static int check_map_access_adj(struct bpf_verifier_env *env, u32 regno, | |
662 | int off, int size) | |
663 | { | |
664 | struct bpf_verifier_state *state = &env->cur_state; | |
665 | struct bpf_reg_state *reg = &state->regs[regno]; | |
666 | int err; | |
667 | ||
668 | /* We adjusted the register to this map value, so we | |
669 | * need to change off and size to min_value and max_value | |
670 | * respectively to make sure our theoretical access will be | |
671 | * safe. | |
672 | */ | |
673 | if (log_level) | |
674 | print_verifier_state(state); | |
675 | env->varlen_map_value_access = true; | |
676 | /* The minimum value is only important with signed | |
677 | * comparisons where we can't assume the floor of a | |
678 | * value is 0. If we are using signed variables for our | |
679 | * index'es we need to make sure that whatever we use | |
680 | * will have a set floor within our range. | |
681 | */ | |
682 | if (reg->min_value < 0) { | |
683 | verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
684 | regno); | |
685 | return -EACCES; | |
686 | } | |
687 | err = check_map_access(env, regno, reg->min_value + off, size); | |
688 | if (err) { | |
689 | verbose("R%d min value is outside of the array range\n", | |
690 | regno); | |
691 | return err; | |
692 | } | |
693 | ||
694 | /* If we haven't set a max value then we need to bail | |
695 | * since we can't be sure we won't do bad things. | |
696 | */ | |
697 | if (reg->max_value == BPF_REGISTER_MAX_RANGE) { | |
698 | verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n", | |
699 | regno); | |
700 | return -EACCES; | |
701 | } | |
702 | return check_map_access(env, regno, reg->max_value + off, size); | |
703 | } | |
704 | ||
969bf05e AS |
705 | #define MAX_PACKET_OFF 0xffff |
706 | ||
58e2af8b | 707 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
3a0af8fd TG |
708 | const struct bpf_call_arg_meta *meta, |
709 | enum bpf_access_type t) | |
4acf6c0b | 710 | { |
36bbef52 | 711 | switch (env->prog->type) { |
3a0af8fd TG |
712 | case BPF_PROG_TYPE_LWT_IN: |
713 | case BPF_PROG_TYPE_LWT_OUT: | |
714 | /* dst_input() and dst_output() can't write for now */ | |
715 | if (t == BPF_WRITE) | |
716 | return false; | |
7e57fbb2 | 717 | /* fallthrough */ |
36bbef52 DB |
718 | case BPF_PROG_TYPE_SCHED_CLS: |
719 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 720 | case BPF_PROG_TYPE_XDP: |
3a0af8fd | 721 | case BPF_PROG_TYPE_LWT_XMIT: |
36bbef52 DB |
722 | if (meta) |
723 | return meta->pkt_access; | |
724 | ||
725 | env->seen_direct_write = true; | |
4acf6c0b BB |
726 | return true; |
727 | default: | |
728 | return false; | |
729 | } | |
730 | } | |
731 | ||
58e2af8b | 732 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
969bf05e AS |
733 | int size) |
734 | { | |
58e2af8b JK |
735 | struct bpf_reg_state *regs = env->cur_state.regs; |
736 | struct bpf_reg_state *reg = ®s[regno]; | |
969bf05e | 737 | |
d91b28ed | 738 | off += reg->off; |
b399cf64 | 739 | if (off < 0 || size <= 0 || off + size > reg->range) { |
d91b28ed AS |
740 | verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
741 | off, size, regno, reg->id, reg->off, reg->range); | |
969bf05e AS |
742 | return -EACCES; |
743 | } | |
744 | return 0; | |
745 | } | |
746 | ||
17a52670 | 747 | /* check access to 'struct bpf_context' fields */ |
58e2af8b | 748 | static int check_ctx_access(struct bpf_verifier_env *env, int off, int size, |
19de99f7 | 749 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 750 | { |
13a27dfc JK |
751 | /* for analyzer ctx accesses are already validated and converted */ |
752 | if (env->analyzer_ops) | |
753 | return 0; | |
754 | ||
17a52670 | 755 | if (env->prog->aux->ops->is_valid_access && |
19de99f7 | 756 | env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) { |
32bbe007 AS |
757 | /* remember the offset of last byte accessed in ctx */ |
758 | if (env->prog->aux->max_ctx_offset < off + size) | |
759 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 760 | return 0; |
32bbe007 | 761 | } |
17a52670 AS |
762 | |
763 | verbose("invalid bpf_context access off=%d size=%d\n", off, size); | |
764 | return -EACCES; | |
765 | } | |
766 | ||
58e2af8b | 767 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1be7f75d AS |
768 | { |
769 | if (env->allow_ptr_leaks) | |
770 | return false; | |
771 | ||
772 | switch (env->cur_state.regs[regno].type) { | |
773 | case UNKNOWN_VALUE: | |
774 | case CONST_IMM: | |
775 | return false; | |
776 | default: | |
777 | return true; | |
778 | } | |
779 | } | |
780 | ||
79adffcd DB |
781 | static int check_pkt_ptr_alignment(const struct bpf_reg_state *reg, |
782 | int off, int size) | |
969bf05e | 783 | { |
969bf05e | 784 | if (reg->id && size != 1) { |
79adffcd | 785 | verbose("Unknown alignment. Only byte-sized access allowed in packet access.\n"); |
969bf05e AS |
786 | return -EACCES; |
787 | } | |
788 | ||
789 | /* skb->data is NET_IP_ALIGN-ed */ | |
79adffcd | 790 | if ((NET_IP_ALIGN + reg->off + off) % size != 0) { |
969bf05e AS |
791 | verbose("misaligned packet access off %d+%d+%d size %d\n", |
792 | NET_IP_ALIGN, reg->off, off, size); | |
793 | return -EACCES; | |
794 | } | |
79adffcd | 795 | |
969bf05e AS |
796 | return 0; |
797 | } | |
798 | ||
79adffcd DB |
799 | static int check_val_ptr_alignment(const struct bpf_reg_state *reg, |
800 | int size) | |
801 | { | |
802 | if (size != 1) { | |
803 | verbose("Unknown alignment. Only byte-sized access allowed in value access.\n"); | |
804 | return -EACCES; | |
805 | } | |
806 | ||
969bf05e AS |
807 | return 0; |
808 | } | |
809 | ||
79adffcd DB |
810 | static int check_ptr_alignment(const struct bpf_reg_state *reg, |
811 | int off, int size) | |
812 | { | |
813 | switch (reg->type) { | |
814 | case PTR_TO_PACKET: | |
815 | return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 : | |
816 | check_pkt_ptr_alignment(reg, off, size); | |
817 | case PTR_TO_MAP_VALUE_ADJ: | |
818 | return IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) ? 0 : | |
819 | check_val_ptr_alignment(reg, size); | |
820 | default: | |
821 | if (off % size != 0) { | |
822 | verbose("misaligned access off %d size %d\n", | |
823 | off, size); | |
824 | return -EACCES; | |
825 | } | |
826 | ||
827 | return 0; | |
828 | } | |
829 | } | |
830 | ||
17a52670 AS |
831 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
832 | * if t==write, value_regno is a register which value is stored into memory | |
833 | * if t==read, value_regno is a register which will receive the value from memory | |
834 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
835 | * if t==read && value_regno==-1, don't care what we read from memory | |
836 | */ | |
58e2af8b | 837 | static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
838 | int bpf_size, enum bpf_access_type t, |
839 | int value_regno) | |
840 | { | |
58e2af8b JK |
841 | struct bpf_verifier_state *state = &env->cur_state; |
842 | struct bpf_reg_state *reg = &state->regs[regno]; | |
17a52670 AS |
843 | int size, err = 0; |
844 | ||
1a0dc1ac AS |
845 | if (reg->type == PTR_TO_STACK) |
846 | off += reg->imm; | |
24b4d2ab | 847 | |
17a52670 AS |
848 | size = bpf_size_to_bytes(bpf_size); |
849 | if (size < 0) | |
850 | return size; | |
851 | ||
79adffcd | 852 | err = check_ptr_alignment(reg, off, size); |
969bf05e AS |
853 | if (err) |
854 | return err; | |
17a52670 | 855 | |
48461135 JB |
856 | if (reg->type == PTR_TO_MAP_VALUE || |
857 | reg->type == PTR_TO_MAP_VALUE_ADJ) { | |
1be7f75d AS |
858 | if (t == BPF_WRITE && value_regno >= 0 && |
859 | is_pointer_value(env, value_regno)) { | |
860 | verbose("R%d leaks addr into map\n", value_regno); | |
861 | return -EACCES; | |
862 | } | |
48461135 | 863 | |
dbcfe5f7 GB |
864 | if (reg->type == PTR_TO_MAP_VALUE_ADJ) |
865 | err = check_map_access_adj(env, regno, off, size); | |
866 | else | |
867 | err = check_map_access(env, regno, off, size); | |
17a52670 | 868 | if (!err && t == BPF_READ && value_regno >= 0) |
f0318d01 GB |
869 | mark_reg_unknown_value_and_range(state->regs, |
870 | value_regno); | |
17a52670 | 871 | |
1a0dc1ac | 872 | } else if (reg->type == PTR_TO_CTX) { |
19de99f7 AS |
873 | enum bpf_reg_type reg_type = UNKNOWN_VALUE; |
874 | ||
1be7f75d AS |
875 | if (t == BPF_WRITE && value_regno >= 0 && |
876 | is_pointer_value(env, value_regno)) { | |
877 | verbose("R%d leaks addr into ctx\n", value_regno); | |
878 | return -EACCES; | |
879 | } | |
19de99f7 | 880 | err = check_ctx_access(env, off, size, t, ®_type); |
969bf05e | 881 | if (!err && t == BPF_READ && value_regno >= 0) { |
f0318d01 GB |
882 | mark_reg_unknown_value_and_range(state->regs, |
883 | value_regno); | |
1955351d MS |
884 | /* note that reg.[id|off|range] == 0 */ |
885 | state->regs[value_regno].type = reg_type; | |
969bf05e | 886 | } |
17a52670 | 887 | |
1a0dc1ac | 888 | } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { |
17a52670 AS |
889 | if (off >= 0 || off < -MAX_BPF_STACK) { |
890 | verbose("invalid stack off=%d size=%d\n", off, size); | |
891 | return -EACCES; | |
892 | } | |
1be7f75d AS |
893 | if (t == BPF_WRITE) { |
894 | if (!env->allow_ptr_leaks && | |
895 | state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && | |
896 | size != BPF_REG_SIZE) { | |
897 | verbose("attempt to corrupt spilled pointer on stack\n"); | |
898 | return -EACCES; | |
899 | } | |
17a52670 | 900 | err = check_stack_write(state, off, size, value_regno); |
1be7f75d | 901 | } else { |
17a52670 | 902 | err = check_stack_read(state, off, size, value_regno); |
1be7f75d | 903 | } |
969bf05e | 904 | } else if (state->regs[regno].type == PTR_TO_PACKET) { |
3a0af8fd | 905 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL, t)) { |
969bf05e AS |
906 | verbose("cannot write into packet\n"); |
907 | return -EACCES; | |
908 | } | |
4acf6c0b BB |
909 | if (t == BPF_WRITE && value_regno >= 0 && |
910 | is_pointer_value(env, value_regno)) { | |
911 | verbose("R%d leaks addr into packet\n", value_regno); | |
912 | return -EACCES; | |
913 | } | |
969bf05e AS |
914 | err = check_packet_access(env, regno, off, size); |
915 | if (!err && t == BPF_READ && value_regno >= 0) | |
f0318d01 GB |
916 | mark_reg_unknown_value_and_range(state->regs, |
917 | value_regno); | |
17a52670 AS |
918 | } else { |
919 | verbose("R%d invalid mem access '%s'\n", | |
1a0dc1ac | 920 | regno, reg_type_str[reg->type]); |
17a52670 AS |
921 | return -EACCES; |
922 | } | |
969bf05e AS |
923 | |
924 | if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && | |
925 | state->regs[value_regno].type == UNKNOWN_VALUE) { | |
926 | /* 1 or 2 byte load zero-extends, determine the number of | |
927 | * zero upper bits. Not doing it fo 4 byte load, since | |
928 | * such values cannot be added to ptr_to_packet anyway. | |
929 | */ | |
930 | state->regs[value_regno].imm = 64 - size * 8; | |
931 | } | |
17a52670 AS |
932 | return err; |
933 | } | |
934 | ||
58e2af8b | 935 | static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 936 | { |
58e2af8b | 937 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
938 | int err; |
939 | ||
940 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
941 | insn->imm != 0) { | |
942 | verbose("BPF_XADD uses reserved fields\n"); | |
943 | return -EINVAL; | |
944 | } | |
945 | ||
946 | /* check src1 operand */ | |
947 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
948 | if (err) | |
949 | return err; | |
950 | ||
951 | /* check src2 operand */ | |
952 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
953 | if (err) | |
954 | return err; | |
955 | ||
956 | /* check whether atomic_add can read the memory */ | |
957 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
958 | BPF_SIZE(insn->code), BPF_READ, -1); | |
959 | if (err) | |
960 | return err; | |
961 | ||
962 | /* check whether atomic_add can write into the same memory */ | |
963 | return check_mem_access(env, insn->dst_reg, insn->off, | |
964 | BPF_SIZE(insn->code), BPF_WRITE, -1); | |
965 | } | |
966 | ||
967 | /* when register 'regno' is passed into function that will read 'access_size' | |
968 | * bytes from that pointer, make sure that it's within stack boundary | |
969 | * and all elements of stack are initialized | |
970 | */ | |
58e2af8b | 971 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
972 | int access_size, bool zero_size_allowed, |
973 | struct bpf_call_arg_meta *meta) | |
17a52670 | 974 | { |
58e2af8b JK |
975 | struct bpf_verifier_state *state = &env->cur_state; |
976 | struct bpf_reg_state *regs = state->regs; | |
17a52670 AS |
977 | int off, i; |
978 | ||
8e2fe1d9 DB |
979 | if (regs[regno].type != PTR_TO_STACK) { |
980 | if (zero_size_allowed && access_size == 0 && | |
981 | regs[regno].type == CONST_IMM && | |
982 | regs[regno].imm == 0) | |
983 | return 0; | |
984 | ||
985 | verbose("R%d type=%s expected=%s\n", regno, | |
986 | reg_type_str[regs[regno].type], | |
987 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 988 | return -EACCES; |
8e2fe1d9 | 989 | } |
17a52670 AS |
990 | |
991 | off = regs[regno].imm; | |
992 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || | |
993 | access_size <= 0) { | |
994 | verbose("invalid stack type R%d off=%d access_size=%d\n", | |
995 | regno, off, access_size); | |
996 | return -EACCES; | |
997 | } | |
998 | ||
435faee1 DB |
999 | if (meta && meta->raw_mode) { |
1000 | meta->access_size = access_size; | |
1001 | meta->regno = regno; | |
1002 | return 0; | |
1003 | } | |
1004 | ||
17a52670 | 1005 | for (i = 0; i < access_size; i++) { |
9c399760 | 1006 | if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { |
17a52670 AS |
1007 | verbose("invalid indirect read from stack off %d+%d size %d\n", |
1008 | off, i, access_size); | |
1009 | return -EACCES; | |
1010 | } | |
1011 | } | |
1012 | return 0; | |
1013 | } | |
1014 | ||
06c1c049 GB |
1015 | static int check_helper_mem_access(struct bpf_verifier_env *env, int regno, |
1016 | int access_size, bool zero_size_allowed, | |
1017 | struct bpf_call_arg_meta *meta) | |
1018 | { | |
1019 | struct bpf_reg_state *regs = env->cur_state.regs; | |
1020 | ||
1021 | switch (regs[regno].type) { | |
1022 | case PTR_TO_PACKET: | |
1023 | return check_packet_access(env, regno, 0, access_size); | |
1024 | case PTR_TO_MAP_VALUE: | |
1025 | return check_map_access(env, regno, 0, access_size); | |
1026 | case PTR_TO_MAP_VALUE_ADJ: | |
1027 | return check_map_access_adj(env, regno, 0, access_size); | |
1028 | default: /* const_imm|ptr_to_stack or invalid ptr */ | |
1029 | return check_stack_boundary(env, regno, access_size, | |
1030 | zero_size_allowed, meta); | |
1031 | } | |
1032 | } | |
1033 | ||
58e2af8b | 1034 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
1035 | enum bpf_arg_type arg_type, |
1036 | struct bpf_call_arg_meta *meta) | |
17a52670 | 1037 | { |
58e2af8b | 1038 | struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; |
6841de8b | 1039 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
1040 | int err = 0; |
1041 | ||
80f1d68c | 1042 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
1043 | return 0; |
1044 | ||
6841de8b | 1045 | if (type == NOT_INIT) { |
17a52670 AS |
1046 | verbose("R%d !read_ok\n", regno); |
1047 | return -EACCES; | |
1048 | } | |
1049 | ||
1be7f75d AS |
1050 | if (arg_type == ARG_ANYTHING) { |
1051 | if (is_pointer_value(env, regno)) { | |
1052 | verbose("R%d leaks addr into helper function\n", regno); | |
1053 | return -EACCES; | |
1054 | } | |
80f1d68c | 1055 | return 0; |
1be7f75d | 1056 | } |
80f1d68c | 1057 | |
3a0af8fd TG |
1058 | if (type == PTR_TO_PACKET && |
1059 | !may_access_direct_pkt_data(env, meta, BPF_READ)) { | |
36bbef52 | 1060 | verbose("helper access to the packet is not allowed\n"); |
6841de8b AS |
1061 | return -EACCES; |
1062 | } | |
1063 | ||
8e2fe1d9 | 1064 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
1065 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
1066 | expected_type = PTR_TO_STACK; | |
6841de8b AS |
1067 | if (type != PTR_TO_PACKET && type != expected_type) |
1068 | goto err_type; | |
39f19ebb AS |
1069 | } else if (arg_type == ARG_CONST_SIZE || |
1070 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
17a52670 | 1071 | expected_type = CONST_IMM; |
06c1c049 GB |
1072 | /* One exception. Allow UNKNOWN_VALUE registers when the |
1073 | * boundaries are known and don't cause unsafe memory accesses | |
1074 | */ | |
1075 | if (type != UNKNOWN_VALUE && type != expected_type) | |
6841de8b | 1076 | goto err_type; |
17a52670 AS |
1077 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
1078 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
1079 | if (type != expected_type) |
1080 | goto err_type; | |
608cd71a AS |
1081 | } else if (arg_type == ARG_PTR_TO_CTX) { |
1082 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
1083 | if (type != expected_type) |
1084 | goto err_type; | |
39f19ebb AS |
1085 | } else if (arg_type == ARG_PTR_TO_MEM || |
1086 | arg_type == ARG_PTR_TO_UNINIT_MEM) { | |
8e2fe1d9 DB |
1087 | expected_type = PTR_TO_STACK; |
1088 | /* One exception here. In case function allows for NULL to be | |
1089 | * passed in as argument, it's a CONST_IMM type. Final test | |
1090 | * happens during stack boundary checking. | |
1091 | */ | |
6841de8b AS |
1092 | if (type == CONST_IMM && reg->imm == 0) |
1093 | /* final test in check_stack_boundary() */; | |
5722569b GB |
1094 | else if (type != PTR_TO_PACKET && type != PTR_TO_MAP_VALUE && |
1095 | type != PTR_TO_MAP_VALUE_ADJ && type != expected_type) | |
6841de8b | 1096 | goto err_type; |
39f19ebb | 1097 | meta->raw_mode = arg_type == ARG_PTR_TO_UNINIT_MEM; |
17a52670 AS |
1098 | } else { |
1099 | verbose("unsupported arg_type %d\n", arg_type); | |
1100 | return -EFAULT; | |
1101 | } | |
1102 | ||
17a52670 AS |
1103 | if (arg_type == ARG_CONST_MAP_PTR) { |
1104 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 1105 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
1106 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
1107 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1108 | * check that [key, key + map->key_size) are within | |
1109 | * stack limits and initialized | |
1110 | */ | |
33ff9823 | 1111 | if (!meta->map_ptr) { |
17a52670 AS |
1112 | /* in function declaration map_ptr must come before |
1113 | * map_key, so that it's verified and known before | |
1114 | * we have to check map_key here. Otherwise it means | |
1115 | * that kernel subsystem misconfigured verifier | |
1116 | */ | |
1117 | verbose("invalid map_ptr to access map->key\n"); | |
1118 | return -EACCES; | |
1119 | } | |
6841de8b AS |
1120 | if (type == PTR_TO_PACKET) |
1121 | err = check_packet_access(env, regno, 0, | |
1122 | meta->map_ptr->key_size); | |
1123 | else | |
1124 | err = check_stack_boundary(env, regno, | |
1125 | meta->map_ptr->key_size, | |
1126 | false, NULL); | |
17a52670 AS |
1127 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1128 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1129 | * check [value, value + map->value_size) validity | |
1130 | */ | |
33ff9823 | 1131 | if (!meta->map_ptr) { |
17a52670 AS |
1132 | /* kernel subsystem misconfigured verifier */ |
1133 | verbose("invalid map_ptr to access map->value\n"); | |
1134 | return -EACCES; | |
1135 | } | |
6841de8b AS |
1136 | if (type == PTR_TO_PACKET) |
1137 | err = check_packet_access(env, regno, 0, | |
1138 | meta->map_ptr->value_size); | |
1139 | else | |
1140 | err = check_stack_boundary(env, regno, | |
1141 | meta->map_ptr->value_size, | |
1142 | false, NULL); | |
39f19ebb AS |
1143 | } else if (arg_type == ARG_CONST_SIZE || |
1144 | arg_type == ARG_CONST_SIZE_OR_ZERO) { | |
1145 | bool zero_size_allowed = (arg_type == ARG_CONST_SIZE_OR_ZERO); | |
17a52670 | 1146 | |
17a52670 AS |
1147 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1148 | * from stack pointer 'buf'. Check it | |
1149 | * note: regno == len, regno - 1 == buf | |
1150 | */ | |
1151 | if (regno == 0) { | |
1152 | /* kernel subsystem misconfigured verifier */ | |
39f19ebb | 1153 | verbose("ARG_CONST_SIZE cannot be first argument\n"); |
17a52670 AS |
1154 | return -EACCES; |
1155 | } | |
06c1c049 GB |
1156 | |
1157 | /* If the register is UNKNOWN_VALUE, the access check happens | |
1158 | * using its boundaries. Otherwise, just use its imm | |
1159 | */ | |
1160 | if (type == UNKNOWN_VALUE) { | |
1161 | /* For unprivileged variable accesses, disable raw | |
1162 | * mode so that the program is required to | |
1163 | * initialize all the memory that the helper could | |
1164 | * just partially fill up. | |
1165 | */ | |
1166 | meta = NULL; | |
1167 | ||
1168 | if (reg->min_value < 0) { | |
1169 | verbose("R%d min value is negative, either use unsigned or 'var &= const'\n", | |
1170 | regno); | |
1171 | return -EACCES; | |
1172 | } | |
1173 | ||
1174 | if (reg->min_value == 0) { | |
1175 | err = check_helper_mem_access(env, regno - 1, 0, | |
1176 | zero_size_allowed, | |
1177 | meta); | |
1178 | if (err) | |
1179 | return err; | |
1180 | } | |
1181 | ||
1182 | if (reg->max_value == BPF_REGISTER_MAX_RANGE) { | |
1183 | verbose("R%d unbounded memory access, use 'var &= const' or 'if (var < const)'\n", | |
1184 | regno); | |
1185 | return -EACCES; | |
1186 | } | |
1187 | err = check_helper_mem_access(env, regno - 1, | |
1188 | reg->max_value, | |
1189 | zero_size_allowed, meta); | |
1190 | if (err) | |
1191 | return err; | |
1192 | } else { | |
1193 | /* register is CONST_IMM */ | |
1194 | err = check_helper_mem_access(env, regno - 1, reg->imm, | |
1195 | zero_size_allowed, meta); | |
1196 | } | |
17a52670 AS |
1197 | } |
1198 | ||
1199 | return err; | |
6841de8b AS |
1200 | err_type: |
1201 | verbose("R%d type=%s expected=%s\n", regno, | |
1202 | reg_type_str[type], reg_type_str[expected_type]); | |
1203 | return -EACCES; | |
17a52670 AS |
1204 | } |
1205 | ||
35578d79 KX |
1206 | static int check_map_func_compatibility(struct bpf_map *map, int func_id) |
1207 | { | |
35578d79 KX |
1208 | if (!map) |
1209 | return 0; | |
1210 | ||
6aff67c8 AS |
1211 | /* We need a two way check, first is from map perspective ... */ |
1212 | switch (map->map_type) { | |
1213 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1214 | if (func_id != BPF_FUNC_tail_call) | |
1215 | goto error; | |
1216 | break; | |
1217 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1218 | if (func_id != BPF_FUNC_perf_event_read && | |
1219 | func_id != BPF_FUNC_perf_event_output) | |
1220 | goto error; | |
1221 | break; | |
1222 | case BPF_MAP_TYPE_STACK_TRACE: | |
1223 | if (func_id != BPF_FUNC_get_stackid) | |
1224 | goto error; | |
1225 | break; | |
4ed8ec52 | 1226 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1227 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1228 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1229 | goto error; |
1230 | break; | |
56f668df | 1231 | case BPF_MAP_TYPE_ARRAY_OF_MAPS: |
bcc6b1b7 | 1232 | case BPF_MAP_TYPE_HASH_OF_MAPS: |
56f668df MKL |
1233 | if (func_id != BPF_FUNC_map_lookup_elem) |
1234 | goto error; | |
6aff67c8 AS |
1235 | default: |
1236 | break; | |
1237 | } | |
1238 | ||
1239 | /* ... and second from the function itself. */ | |
1240 | switch (func_id) { | |
1241 | case BPF_FUNC_tail_call: | |
1242 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1243 | goto error; | |
1244 | break; | |
1245 | case BPF_FUNC_perf_event_read: | |
1246 | case BPF_FUNC_perf_event_output: | |
1247 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) | |
1248 | goto error; | |
1249 | break; | |
1250 | case BPF_FUNC_get_stackid: | |
1251 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1252 | goto error; | |
1253 | break; | |
60d20f91 | 1254 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1255 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1256 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1257 | goto error; | |
1258 | break; | |
6aff67c8 AS |
1259 | default: |
1260 | break; | |
35578d79 KX |
1261 | } |
1262 | ||
1263 | return 0; | |
6aff67c8 | 1264 | error: |
ebb676da TG |
1265 | verbose("cannot pass map_type %d into func %s#%d\n", |
1266 | map->map_type, func_id_name(func_id), func_id); | |
6aff67c8 | 1267 | return -EINVAL; |
35578d79 KX |
1268 | } |
1269 | ||
435faee1 DB |
1270 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1271 | { | |
1272 | int count = 0; | |
1273 | ||
39f19ebb | 1274 | if (fn->arg1_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1275 | count++; |
39f19ebb | 1276 | if (fn->arg2_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1277 | count++; |
39f19ebb | 1278 | if (fn->arg3_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1279 | count++; |
39f19ebb | 1280 | if (fn->arg4_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 | 1281 | count++; |
39f19ebb | 1282 | if (fn->arg5_type == ARG_PTR_TO_UNINIT_MEM) |
435faee1 DB |
1283 | count++; |
1284 | ||
1285 | return count > 1 ? -EINVAL : 0; | |
1286 | } | |
1287 | ||
58e2af8b | 1288 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1289 | { |
58e2af8b JK |
1290 | struct bpf_verifier_state *state = &env->cur_state; |
1291 | struct bpf_reg_state *regs = state->regs, *reg; | |
969bf05e AS |
1292 | int i; |
1293 | ||
1294 | for (i = 0; i < MAX_BPF_REG; i++) | |
1295 | if (regs[i].type == PTR_TO_PACKET || | |
1296 | regs[i].type == PTR_TO_PACKET_END) | |
1297 | mark_reg_unknown_value(regs, i); | |
1298 | ||
1299 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1300 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1301 | continue; | |
1302 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1303 | if (reg->type != PTR_TO_PACKET && | |
1304 | reg->type != PTR_TO_PACKET_END) | |
1305 | continue; | |
1306 | reg->type = UNKNOWN_VALUE; | |
1307 | reg->imm = 0; | |
1308 | } | |
1309 | } | |
1310 | ||
81ed18ab | 1311 | static int check_call(struct bpf_verifier_env *env, int func_id, int insn_idx) |
17a52670 | 1312 | { |
58e2af8b | 1313 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 1314 | const struct bpf_func_proto *fn = NULL; |
58e2af8b JK |
1315 | struct bpf_reg_state *regs = state->regs; |
1316 | struct bpf_reg_state *reg; | |
33ff9823 | 1317 | struct bpf_call_arg_meta meta; |
969bf05e | 1318 | bool changes_data; |
17a52670 AS |
1319 | int i, err; |
1320 | ||
1321 | /* find function prototype */ | |
1322 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
ebb676da | 1323 | verbose("invalid func %s#%d\n", func_id_name(func_id), func_id); |
17a52670 AS |
1324 | return -EINVAL; |
1325 | } | |
1326 | ||
1327 | if (env->prog->aux->ops->get_func_proto) | |
1328 | fn = env->prog->aux->ops->get_func_proto(func_id); | |
1329 | ||
1330 | if (!fn) { | |
ebb676da | 1331 | verbose("unknown func %s#%d\n", func_id_name(func_id), func_id); |
17a52670 AS |
1332 | return -EINVAL; |
1333 | } | |
1334 | ||
1335 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1336 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
17a52670 AS |
1337 | verbose("cannot call GPL only function from proprietary program\n"); |
1338 | return -EINVAL; | |
1339 | } | |
1340 | ||
17bedab2 | 1341 | changes_data = bpf_helper_changes_pkt_data(fn->func); |
969bf05e | 1342 | |
33ff9823 | 1343 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1344 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1345 | |
435faee1 DB |
1346 | /* We only support one arg being in raw mode at the moment, which |
1347 | * is sufficient for the helper functions we have right now. | |
1348 | */ | |
1349 | err = check_raw_mode(fn); | |
1350 | if (err) { | |
ebb676da TG |
1351 | verbose("kernel subsystem misconfigured func %s#%d\n", |
1352 | func_id_name(func_id), func_id); | |
435faee1 DB |
1353 | return err; |
1354 | } | |
1355 | ||
17a52670 | 1356 | /* check args */ |
33ff9823 | 1357 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1358 | if (err) |
1359 | return err; | |
33ff9823 | 1360 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1361 | if (err) |
1362 | return err; | |
33ff9823 | 1363 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1364 | if (err) |
1365 | return err; | |
33ff9823 | 1366 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1367 | if (err) |
1368 | return err; | |
33ff9823 | 1369 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1370 | if (err) |
1371 | return err; | |
1372 | ||
435faee1 DB |
1373 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1374 | * is inferred from register state. | |
1375 | */ | |
1376 | for (i = 0; i < meta.access_size; i++) { | |
1377 | err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1); | |
1378 | if (err) | |
1379 | return err; | |
1380 | } | |
1381 | ||
17a52670 AS |
1382 | /* reset caller saved regs */ |
1383 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
1384 | reg = regs + caller_saved[i]; | |
1385 | reg->type = NOT_INIT; | |
1386 | reg->imm = 0; | |
1387 | } | |
1388 | ||
1389 | /* update return register */ | |
1390 | if (fn->ret_type == RET_INTEGER) { | |
1391 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
1392 | } else if (fn->ret_type == RET_VOID) { | |
1393 | regs[BPF_REG_0].type = NOT_INIT; | |
1394 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
fad73a1a MKL |
1395 | struct bpf_insn_aux_data *insn_aux; |
1396 | ||
17a52670 | 1397 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; |
48461135 | 1398 | regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0; |
17a52670 AS |
1399 | /* remember map_ptr, so that check_map_access() |
1400 | * can check 'value_size' boundary of memory access | |
1401 | * to map element returned from bpf_map_lookup_elem() | |
1402 | */ | |
33ff9823 | 1403 | if (meta.map_ptr == NULL) { |
17a52670 AS |
1404 | verbose("kernel subsystem misconfigured verifier\n"); |
1405 | return -EINVAL; | |
1406 | } | |
33ff9823 | 1407 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1408 | regs[BPF_REG_0].id = ++env->id_gen; |
fad73a1a MKL |
1409 | insn_aux = &env->insn_aux_data[insn_idx]; |
1410 | if (!insn_aux->map_ptr) | |
1411 | insn_aux->map_ptr = meta.map_ptr; | |
1412 | else if (insn_aux->map_ptr != meta.map_ptr) | |
1413 | insn_aux->map_ptr = BPF_MAP_PTR_POISON; | |
17a52670 | 1414 | } else { |
ebb676da TG |
1415 | verbose("unknown return type %d of func %s#%d\n", |
1416 | fn->ret_type, func_id_name(func_id), func_id); | |
17a52670 AS |
1417 | return -EINVAL; |
1418 | } | |
04fd61ab | 1419 | |
33ff9823 | 1420 | err = check_map_func_compatibility(meta.map_ptr, func_id); |
35578d79 KX |
1421 | if (err) |
1422 | return err; | |
04fd61ab | 1423 | |
969bf05e AS |
1424 | if (changes_data) |
1425 | clear_all_pkt_pointers(env); | |
1426 | return 0; | |
1427 | } | |
1428 | ||
58e2af8b JK |
1429 | static int check_packet_ptr_add(struct bpf_verifier_env *env, |
1430 | struct bpf_insn *insn) | |
969bf05e | 1431 | { |
58e2af8b JK |
1432 | struct bpf_reg_state *regs = env->cur_state.regs; |
1433 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1434 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
1435 | struct bpf_reg_state tmp_reg; | |
969bf05e AS |
1436 | s32 imm; |
1437 | ||
1438 | if (BPF_SRC(insn->code) == BPF_K) { | |
1439 | /* pkt_ptr += imm */ | |
1440 | imm = insn->imm; | |
1441 | ||
1442 | add_imm: | |
63dfef75 | 1443 | if (imm < 0) { |
969bf05e AS |
1444 | verbose("addition of negative constant to packet pointer is not allowed\n"); |
1445 | return -EACCES; | |
1446 | } | |
1447 | if (imm >= MAX_PACKET_OFF || | |
1448 | imm + dst_reg->off >= MAX_PACKET_OFF) { | |
1449 | verbose("constant %d is too large to add to packet pointer\n", | |
1450 | imm); | |
1451 | return -EACCES; | |
1452 | } | |
1453 | /* a constant was added to pkt_ptr. | |
1454 | * Remember it while keeping the same 'id' | |
1455 | */ | |
1456 | dst_reg->off += imm; | |
1457 | } else { | |
1b9b69ec AS |
1458 | if (src_reg->type == PTR_TO_PACKET) { |
1459 | /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ | |
1460 | tmp_reg = *dst_reg; /* save r7 state */ | |
1461 | *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ | |
1462 | src_reg = &tmp_reg; /* pretend it's src_reg state */ | |
1463 | /* if the checks below reject it, the copy won't matter, | |
1464 | * since we're rejecting the whole program. If all ok, | |
1465 | * then imm22 state will be added to r7 | |
1466 | * and r7 will be pkt(id=0,off=22,r=62) while | |
1467 | * r6 will stay as pkt(id=0,off=0,r=62) | |
1468 | */ | |
1469 | } | |
1470 | ||
969bf05e AS |
1471 | if (src_reg->type == CONST_IMM) { |
1472 | /* pkt_ptr += reg where reg is known constant */ | |
1473 | imm = src_reg->imm; | |
1474 | goto add_imm; | |
1475 | } | |
1476 | /* disallow pkt_ptr += reg | |
1477 | * if reg is not uknown_value with guaranteed zero upper bits | |
1478 | * otherwise pkt_ptr may overflow and addition will become | |
1479 | * subtraction which is not allowed | |
1480 | */ | |
1481 | if (src_reg->type != UNKNOWN_VALUE) { | |
1482 | verbose("cannot add '%s' to ptr_to_packet\n", | |
1483 | reg_type_str[src_reg->type]); | |
1484 | return -EACCES; | |
1485 | } | |
1486 | if (src_reg->imm < 48) { | |
1487 | verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", | |
1488 | src_reg->imm); | |
1489 | return -EACCES; | |
1490 | } | |
1491 | /* dst_reg stays as pkt_ptr type and since some positive | |
1492 | * integer value was added to the pointer, increment its 'id' | |
1493 | */ | |
1f415a74 | 1494 | dst_reg->id = ++env->id_gen; |
969bf05e AS |
1495 | |
1496 | /* something was added to pkt_ptr, set range and off to zero */ | |
1497 | dst_reg->off = 0; | |
1498 | dst_reg->range = 0; | |
1499 | } | |
1500 | return 0; | |
1501 | } | |
1502 | ||
58e2af8b | 1503 | static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) |
969bf05e | 1504 | { |
58e2af8b JK |
1505 | struct bpf_reg_state *regs = env->cur_state.regs; |
1506 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
969bf05e AS |
1507 | u8 opcode = BPF_OP(insn->code); |
1508 | s64 imm_log2; | |
1509 | ||
1510 | /* for type == UNKNOWN_VALUE: | |
1511 | * imm > 0 -> number of zero upper bits | |
1512 | * imm == 0 -> don't track which is the same as all bits can be non-zero | |
1513 | */ | |
1514 | ||
1515 | if (BPF_SRC(insn->code) == BPF_X) { | |
58e2af8b | 1516 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
969bf05e AS |
1517 | |
1518 | if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && | |
1519 | dst_reg->imm && opcode == BPF_ADD) { | |
1520 | /* dreg += sreg | |
1521 | * where both have zero upper bits. Adding them | |
1522 | * can only result making one more bit non-zero | |
1523 | * in the larger value. | |
1524 | * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) | |
1525 | * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) | |
1526 | */ | |
1527 | dst_reg->imm = min(dst_reg->imm, src_reg->imm); | |
1528 | dst_reg->imm--; | |
1529 | return 0; | |
1530 | } | |
1531 | if (src_reg->type == CONST_IMM && src_reg->imm > 0 && | |
1532 | dst_reg->imm && opcode == BPF_ADD) { | |
1533 | /* dreg += sreg | |
1534 | * where dreg has zero upper bits and sreg is const. | |
1535 | * Adding them can only result making one more bit | |
1536 | * non-zero in the larger value. | |
1537 | */ | |
1538 | imm_log2 = __ilog2_u64((long long)src_reg->imm); | |
1539 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1540 | dst_reg->imm--; | |
1541 | return 0; | |
1542 | } | |
1543 | /* all other cases non supported yet, just mark dst_reg */ | |
1544 | dst_reg->imm = 0; | |
1545 | return 0; | |
1546 | } | |
1547 | ||
1548 | /* sign extend 32-bit imm into 64-bit to make sure that | |
1549 | * negative values occupy bit 63. Note ilog2() would have | |
1550 | * been incorrect, since sizeof(insn->imm) == 4 | |
1551 | */ | |
1552 | imm_log2 = __ilog2_u64((long long)insn->imm); | |
1553 | ||
1554 | if (dst_reg->imm && opcode == BPF_LSH) { | |
1555 | /* reg <<= imm | |
1556 | * if reg was a result of 2 byte load, then its imm == 48 | |
1557 | * which means that upper 48 bits are zero and shifting this reg | |
1558 | * left by 4 would mean that upper 44 bits are still zero | |
1559 | */ | |
1560 | dst_reg->imm -= insn->imm; | |
1561 | } else if (dst_reg->imm && opcode == BPF_MUL) { | |
1562 | /* reg *= imm | |
1563 | * if multiplying by 14 subtract 4 | |
1564 | * This is conservative calculation of upper zero bits. | |
1565 | * It's not trying to special case insn->imm == 1 or 0 cases | |
1566 | */ | |
1567 | dst_reg->imm -= imm_log2 + 1; | |
1568 | } else if (opcode == BPF_AND) { | |
1569 | /* reg &= imm */ | |
1570 | dst_reg->imm = 63 - imm_log2; | |
1571 | } else if (dst_reg->imm && opcode == BPF_ADD) { | |
1572 | /* reg += imm */ | |
1573 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1574 | dst_reg->imm--; | |
1575 | } else if (opcode == BPF_RSH) { | |
1576 | /* reg >>= imm | |
1577 | * which means that after right shift, upper bits will be zero | |
1578 | * note that verifier already checked that | |
1579 | * 0 <= imm < 64 for shift insn | |
1580 | */ | |
1581 | dst_reg->imm += insn->imm; | |
1582 | if (unlikely(dst_reg->imm > 64)) | |
1583 | /* some dumb code did: | |
1584 | * r2 = *(u32 *)mem; | |
1585 | * r2 >>= 32; | |
1586 | * and all bits are zero now */ | |
1587 | dst_reg->imm = 64; | |
1588 | } else { | |
1589 | /* all other alu ops, means that we don't know what will | |
1590 | * happen to the value, mark it with unknown number of zero bits | |
1591 | */ | |
1592 | dst_reg->imm = 0; | |
1593 | } | |
1594 | ||
1595 | if (dst_reg->imm < 0) { | |
1596 | /* all 64 bits of the register can contain non-zero bits | |
1597 | * and such value cannot be added to ptr_to_packet, since it | |
1598 | * may overflow, mark it as unknown to avoid further eval | |
1599 | */ | |
1600 | dst_reg->imm = 0; | |
1601 | } | |
1602 | return 0; | |
1603 | } | |
1604 | ||
58e2af8b JK |
1605 | static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, |
1606 | struct bpf_insn *insn) | |
969bf05e | 1607 | { |
58e2af8b JK |
1608 | struct bpf_reg_state *regs = env->cur_state.regs; |
1609 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1610 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
969bf05e | 1611 | u8 opcode = BPF_OP(insn->code); |
3fadc801 | 1612 | u64 dst_imm = dst_reg->imm; |
969bf05e | 1613 | |
3fadc801 DB |
1614 | /* dst_reg->type == CONST_IMM here. Simulate execution of insns |
1615 | * containing ALU ops. Don't care about overflow or negative | |
1616 | * values, just add/sub/... them; registers are in u64. | |
969bf05e | 1617 | */ |
3fadc801 DB |
1618 | if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) { |
1619 | dst_imm += insn->imm; | |
1620 | } else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && | |
1621 | src_reg->type == CONST_IMM) { | |
1622 | dst_imm += src_reg->imm; | |
1623 | } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_K) { | |
1624 | dst_imm -= insn->imm; | |
1625 | } else if (opcode == BPF_SUB && BPF_SRC(insn->code) == BPF_X && | |
1626 | src_reg->type == CONST_IMM) { | |
1627 | dst_imm -= src_reg->imm; | |
1628 | } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_K) { | |
1629 | dst_imm *= insn->imm; | |
1630 | } else if (opcode == BPF_MUL && BPF_SRC(insn->code) == BPF_X && | |
1631 | src_reg->type == CONST_IMM) { | |
1632 | dst_imm *= src_reg->imm; | |
1633 | } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_K) { | |
1634 | dst_imm |= insn->imm; | |
1635 | } else if (opcode == BPF_OR && BPF_SRC(insn->code) == BPF_X && | |
1636 | src_reg->type == CONST_IMM) { | |
1637 | dst_imm |= src_reg->imm; | |
1638 | } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_K) { | |
1639 | dst_imm &= insn->imm; | |
1640 | } else if (opcode == BPF_AND && BPF_SRC(insn->code) == BPF_X && | |
1641 | src_reg->type == CONST_IMM) { | |
1642 | dst_imm &= src_reg->imm; | |
1643 | } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_K) { | |
1644 | dst_imm >>= insn->imm; | |
1645 | } else if (opcode == BPF_RSH && BPF_SRC(insn->code) == BPF_X && | |
1646 | src_reg->type == CONST_IMM) { | |
1647 | dst_imm >>= src_reg->imm; | |
1648 | } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_K) { | |
1649 | dst_imm <<= insn->imm; | |
1650 | } else if (opcode == BPF_LSH && BPF_SRC(insn->code) == BPF_X && | |
1651 | src_reg->type == CONST_IMM) { | |
1652 | dst_imm <<= src_reg->imm; | |
1653 | } else { | |
969bf05e | 1654 | mark_reg_unknown_value(regs, insn->dst_reg); |
3fadc801 DB |
1655 | goto out; |
1656 | } | |
1657 | ||
1658 | dst_reg->imm = dst_imm; | |
1659 | out: | |
17a52670 AS |
1660 | return 0; |
1661 | } | |
1662 | ||
48461135 JB |
1663 | static void check_reg_overflow(struct bpf_reg_state *reg) |
1664 | { | |
1665 | if (reg->max_value > BPF_REGISTER_MAX_RANGE) | |
1666 | reg->max_value = BPF_REGISTER_MAX_RANGE; | |
f23cc643 JB |
1667 | if (reg->min_value < BPF_REGISTER_MIN_RANGE || |
1668 | reg->min_value > BPF_REGISTER_MAX_RANGE) | |
48461135 JB |
1669 | reg->min_value = BPF_REGISTER_MIN_RANGE; |
1670 | } | |
1671 | ||
1672 | static void adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
1673 | struct bpf_insn *insn) | |
1674 | { | |
1675 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; | |
f23cc643 JB |
1676 | s64 min_val = BPF_REGISTER_MIN_RANGE; |
1677 | u64 max_val = BPF_REGISTER_MAX_RANGE; | |
48461135 JB |
1678 | u8 opcode = BPF_OP(insn->code); |
1679 | ||
1680 | dst_reg = ®s[insn->dst_reg]; | |
1681 | if (BPF_SRC(insn->code) == BPF_X) { | |
1682 | check_reg_overflow(®s[insn->src_reg]); | |
1683 | min_val = regs[insn->src_reg].min_value; | |
1684 | max_val = regs[insn->src_reg].max_value; | |
1685 | ||
1686 | /* If the source register is a random pointer then the | |
1687 | * min_value/max_value values represent the range of the known | |
1688 | * accesses into that value, not the actual min/max value of the | |
1689 | * register itself. In this case we have to reset the reg range | |
1690 | * values so we know it is not safe to look at. | |
1691 | */ | |
1692 | if (regs[insn->src_reg].type != CONST_IMM && | |
1693 | regs[insn->src_reg].type != UNKNOWN_VALUE) { | |
1694 | min_val = BPF_REGISTER_MIN_RANGE; | |
1695 | max_val = BPF_REGISTER_MAX_RANGE; | |
1696 | } | |
1697 | } else if (insn->imm < BPF_REGISTER_MAX_RANGE && | |
1698 | (s64)insn->imm > BPF_REGISTER_MIN_RANGE) { | |
1699 | min_val = max_val = insn->imm; | |
48461135 JB |
1700 | } |
1701 | ||
1702 | /* We don't know anything about what was done to this register, mark it | |
1703 | * as unknown. | |
1704 | */ | |
1705 | if (min_val == BPF_REGISTER_MIN_RANGE && | |
1706 | max_val == BPF_REGISTER_MAX_RANGE) { | |
1707 | reset_reg_range_values(regs, insn->dst_reg); | |
1708 | return; | |
1709 | } | |
1710 | ||
f23cc643 JB |
1711 | /* If one of our values was at the end of our ranges then we can't just |
1712 | * do our normal operations to the register, we need to set the values | |
1713 | * to the min/max since they are undefined. | |
1714 | */ | |
1715 | if (min_val == BPF_REGISTER_MIN_RANGE) | |
1716 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1717 | if (max_val == BPF_REGISTER_MAX_RANGE) | |
1718 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1719 | ||
48461135 JB |
1720 | switch (opcode) { |
1721 | case BPF_ADD: | |
f23cc643 JB |
1722 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
1723 | dst_reg->min_value += min_val; | |
1724 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1725 | dst_reg->max_value += max_val; | |
48461135 JB |
1726 | break; |
1727 | case BPF_SUB: | |
f23cc643 JB |
1728 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
1729 | dst_reg->min_value -= min_val; | |
1730 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1731 | dst_reg->max_value -= max_val; | |
48461135 JB |
1732 | break; |
1733 | case BPF_MUL: | |
f23cc643 JB |
1734 | if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
1735 | dst_reg->min_value *= min_val; | |
1736 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1737 | dst_reg->max_value *= max_val; | |
48461135 JB |
1738 | break; |
1739 | case BPF_AND: | |
f23cc643 JB |
1740 | /* Disallow AND'ing of negative numbers, ain't nobody got time |
1741 | * for that. Otherwise the minimum is 0 and the max is the max | |
1742 | * value we could AND against. | |
1743 | */ | |
1744 | if (min_val < 0) | |
1745 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1746 | else | |
1747 | dst_reg->min_value = 0; | |
48461135 JB |
1748 | dst_reg->max_value = max_val; |
1749 | break; | |
1750 | case BPF_LSH: | |
1751 | /* Gotta have special overflow logic here, if we're shifting | |
1752 | * more than MAX_RANGE then just assume we have an invalid | |
1753 | * range. | |
1754 | */ | |
1755 | if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) | |
1756 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
f23cc643 | 1757 | else if (dst_reg->min_value != BPF_REGISTER_MIN_RANGE) |
48461135 JB |
1758 | dst_reg->min_value <<= min_val; |
1759 | ||
1760 | if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) | |
1761 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
f23cc643 | 1762 | else if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) |
48461135 JB |
1763 | dst_reg->max_value <<= max_val; |
1764 | break; | |
1765 | case BPF_RSH: | |
f23cc643 JB |
1766 | /* RSH by a negative number is undefined, and the BPF_RSH is an |
1767 | * unsigned shift, so make the appropriate casts. | |
48461135 | 1768 | */ |
f23cc643 JB |
1769 | if (min_val < 0 || dst_reg->min_value < 0) |
1770 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1771 | else | |
1772 | dst_reg->min_value = | |
1773 | (u64)(dst_reg->min_value) >> min_val; | |
1774 | if (dst_reg->max_value != BPF_REGISTER_MAX_RANGE) | |
1775 | dst_reg->max_value >>= max_val; | |
48461135 JB |
1776 | break; |
1777 | default: | |
1778 | reset_reg_range_values(regs, insn->dst_reg); | |
1779 | break; | |
1780 | } | |
1781 | ||
1782 | check_reg_overflow(dst_reg); | |
1783 | } | |
1784 | ||
17a52670 | 1785 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 1786 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 1787 | { |
58e2af8b | 1788 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; |
17a52670 AS |
1789 | u8 opcode = BPF_OP(insn->code); |
1790 | int err; | |
1791 | ||
1792 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
1793 | if (opcode == BPF_NEG) { | |
1794 | if (BPF_SRC(insn->code) != 0 || | |
1795 | insn->src_reg != BPF_REG_0 || | |
1796 | insn->off != 0 || insn->imm != 0) { | |
1797 | verbose("BPF_NEG uses reserved fields\n"); | |
1798 | return -EINVAL; | |
1799 | } | |
1800 | } else { | |
1801 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
1802 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { | |
1803 | verbose("BPF_END uses reserved fields\n"); | |
1804 | return -EINVAL; | |
1805 | } | |
1806 | } | |
1807 | ||
1808 | /* check src operand */ | |
1809 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1810 | if (err) | |
1811 | return err; | |
1812 | ||
1be7f75d AS |
1813 | if (is_pointer_value(env, insn->dst_reg)) { |
1814 | verbose("R%d pointer arithmetic prohibited\n", | |
1815 | insn->dst_reg); | |
1816 | return -EACCES; | |
1817 | } | |
1818 | ||
17a52670 AS |
1819 | /* check dest operand */ |
1820 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1821 | if (err) | |
1822 | return err; | |
1823 | ||
1824 | } else if (opcode == BPF_MOV) { | |
1825 | ||
1826 | if (BPF_SRC(insn->code) == BPF_X) { | |
1827 | if (insn->imm != 0 || insn->off != 0) { | |
1828 | verbose("BPF_MOV uses reserved fields\n"); | |
1829 | return -EINVAL; | |
1830 | } | |
1831 | ||
1832 | /* check src operand */ | |
1833 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1834 | if (err) | |
1835 | return err; | |
1836 | } else { | |
1837 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1838 | verbose("BPF_MOV uses reserved fields\n"); | |
1839 | return -EINVAL; | |
1840 | } | |
1841 | } | |
1842 | ||
1843 | /* check dest operand */ | |
1844 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1845 | if (err) | |
1846 | return err; | |
1847 | ||
48461135 JB |
1848 | /* we are setting our register to something new, we need to |
1849 | * reset its range values. | |
1850 | */ | |
1851 | reset_reg_range_values(regs, insn->dst_reg); | |
1852 | ||
17a52670 AS |
1853 | if (BPF_SRC(insn->code) == BPF_X) { |
1854 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
1855 | /* case: R1 = R2 | |
1856 | * copy register state to dest reg | |
1857 | */ | |
1858 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
1859 | } else { | |
1be7f75d AS |
1860 | if (is_pointer_value(env, insn->src_reg)) { |
1861 | verbose("R%d partial copy of pointer\n", | |
1862 | insn->src_reg); | |
1863 | return -EACCES; | |
1864 | } | |
57a09bf0 | 1865 | mark_reg_unknown_value(regs, insn->dst_reg); |
17a52670 AS |
1866 | } |
1867 | } else { | |
1868 | /* case: R = imm | |
1869 | * remember the value we stored into this reg | |
1870 | */ | |
1871 | regs[insn->dst_reg].type = CONST_IMM; | |
1872 | regs[insn->dst_reg].imm = insn->imm; | |
48461135 JB |
1873 | regs[insn->dst_reg].max_value = insn->imm; |
1874 | regs[insn->dst_reg].min_value = insn->imm; | |
17a52670 AS |
1875 | } |
1876 | ||
1877 | } else if (opcode > BPF_END) { | |
1878 | verbose("invalid BPF_ALU opcode %x\n", opcode); | |
1879 | return -EINVAL; | |
1880 | ||
1881 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
1882 | ||
17a52670 AS |
1883 | if (BPF_SRC(insn->code) == BPF_X) { |
1884 | if (insn->imm != 0 || insn->off != 0) { | |
1885 | verbose("BPF_ALU uses reserved fields\n"); | |
1886 | return -EINVAL; | |
1887 | } | |
1888 | /* check src1 operand */ | |
1889 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1890 | if (err) | |
1891 | return err; | |
1892 | } else { | |
1893 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1894 | verbose("BPF_ALU uses reserved fields\n"); | |
1895 | return -EINVAL; | |
1896 | } | |
1897 | } | |
1898 | ||
1899 | /* check src2 operand */ | |
1900 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1901 | if (err) | |
1902 | return err; | |
1903 | ||
1904 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
1905 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
1906 | verbose("div by zero\n"); | |
1907 | return -EINVAL; | |
1908 | } | |
1909 | ||
229394e8 RV |
1910 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
1911 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
1912 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
1913 | ||
1914 | if (insn->imm < 0 || insn->imm >= size) { | |
1915 | verbose("invalid shift %d\n", insn->imm); | |
1916 | return -EINVAL; | |
1917 | } | |
1918 | } | |
1919 | ||
1a0dc1ac AS |
1920 | /* check dest operand */ |
1921 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
1922 | if (err) | |
1923 | return err; | |
1924 | ||
1925 | dst_reg = ®s[insn->dst_reg]; | |
1926 | ||
48461135 JB |
1927 | /* first we want to adjust our ranges. */ |
1928 | adjust_reg_min_max_vals(env, insn); | |
1929 | ||
17a52670 AS |
1930 | /* pattern match 'bpf_add Rx, imm' instruction */ |
1931 | if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && | |
1a0dc1ac AS |
1932 | dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { |
1933 | dst_reg->type = PTR_TO_STACK; | |
1934 | dst_reg->imm = insn->imm; | |
1935 | return 0; | |
332270fd YS |
1936 | } else if (opcode == BPF_ADD && |
1937 | BPF_CLASS(insn->code) == BPF_ALU64 && | |
1938 | dst_reg->type == PTR_TO_STACK && | |
1939 | ((BPF_SRC(insn->code) == BPF_X && | |
1940 | regs[insn->src_reg].type == CONST_IMM) || | |
1941 | BPF_SRC(insn->code) == BPF_K)) { | |
1942 | if (BPF_SRC(insn->code) == BPF_X) | |
1943 | dst_reg->imm += regs[insn->src_reg].imm; | |
1944 | else | |
1945 | dst_reg->imm += insn->imm; | |
1946 | return 0; | |
969bf05e AS |
1947 | } else if (opcode == BPF_ADD && |
1948 | BPF_CLASS(insn->code) == BPF_ALU64 && | |
1b9b69ec AS |
1949 | (dst_reg->type == PTR_TO_PACKET || |
1950 | (BPF_SRC(insn->code) == BPF_X && | |
1951 | regs[insn->src_reg].type == PTR_TO_PACKET))) { | |
969bf05e AS |
1952 | /* ptr_to_packet += K|X */ |
1953 | return check_packet_ptr_add(env, insn); | |
1954 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1955 | dst_reg->type == UNKNOWN_VALUE && | |
1956 | env->allow_ptr_leaks) { | |
1957 | /* unknown += K|X */ | |
1958 | return evaluate_reg_alu(env, insn); | |
1959 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1960 | dst_reg->type == CONST_IMM && | |
1961 | env->allow_ptr_leaks) { | |
1962 | /* reg_imm += K|X */ | |
1963 | return evaluate_reg_imm_alu(env, insn); | |
1be7f75d AS |
1964 | } else if (is_pointer_value(env, insn->dst_reg)) { |
1965 | verbose("R%d pointer arithmetic prohibited\n", | |
1966 | insn->dst_reg); | |
1967 | return -EACCES; | |
1968 | } else if (BPF_SRC(insn->code) == BPF_X && | |
1969 | is_pointer_value(env, insn->src_reg)) { | |
1970 | verbose("R%d pointer arithmetic prohibited\n", | |
1971 | insn->src_reg); | |
1972 | return -EACCES; | |
1973 | } | |
17a52670 | 1974 | |
48461135 JB |
1975 | /* If we did pointer math on a map value then just set it to our |
1976 | * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or | |
1977 | * loads to this register appropriately, otherwise just mark the | |
1978 | * register as unknown. | |
1979 | */ | |
1980 | if (env->allow_ptr_leaks && | |
fce366a9 | 1981 | BPF_CLASS(insn->code) == BPF_ALU64 && opcode == BPF_ADD && |
48461135 JB |
1982 | (dst_reg->type == PTR_TO_MAP_VALUE || |
1983 | dst_reg->type == PTR_TO_MAP_VALUE_ADJ)) | |
1984 | dst_reg->type = PTR_TO_MAP_VALUE_ADJ; | |
1985 | else | |
1986 | mark_reg_unknown_value(regs, insn->dst_reg); | |
17a52670 AS |
1987 | } |
1988 | ||
1989 | return 0; | |
1990 | } | |
1991 | ||
58e2af8b JK |
1992 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
1993 | struct bpf_reg_state *dst_reg) | |
969bf05e | 1994 | { |
58e2af8b | 1995 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e | 1996 | int i; |
2d2be8ca DB |
1997 | |
1998 | /* LLVM can generate two kind of checks: | |
1999 | * | |
2000 | * Type 1: | |
2001 | * | |
2002 | * r2 = r3; | |
2003 | * r2 += 8; | |
2004 | * if (r2 > pkt_end) goto <handle exception> | |
2005 | * <access okay> | |
2006 | * | |
2007 | * Where: | |
2008 | * r2 == dst_reg, pkt_end == src_reg | |
2009 | * r2=pkt(id=n,off=8,r=0) | |
2010 | * r3=pkt(id=n,off=0,r=0) | |
2011 | * | |
2012 | * Type 2: | |
2013 | * | |
2014 | * r2 = r3; | |
2015 | * r2 += 8; | |
2016 | * if (pkt_end >= r2) goto <access okay> | |
2017 | * <handle exception> | |
2018 | * | |
2019 | * Where: | |
2020 | * pkt_end == dst_reg, r2 == src_reg | |
2021 | * r2=pkt(id=n,off=8,r=0) | |
2022 | * r3=pkt(id=n,off=0,r=0) | |
2023 | * | |
2024 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
2025 | * so that range of bytes [r3, r3 + 8) is safe to access. | |
969bf05e | 2026 | */ |
2d2be8ca | 2027 | |
969bf05e AS |
2028 | for (i = 0; i < MAX_BPF_REG; i++) |
2029 | if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) | |
b1977682 AS |
2030 | /* keep the maximum range already checked */ |
2031 | regs[i].range = max(regs[i].range, dst_reg->off); | |
969bf05e AS |
2032 | |
2033 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
2034 | if (state->stack_slot_type[i] != STACK_SPILL) | |
2035 | continue; | |
2036 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
2037 | if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) | |
b1977682 | 2038 | reg->range = max(reg->range, dst_reg->off); |
969bf05e AS |
2039 | } |
2040 | } | |
2041 | ||
48461135 JB |
2042 | /* Adjusts the register min/max values in the case that the dst_reg is the |
2043 | * variable register that we are working on, and src_reg is a constant or we're | |
2044 | * simply doing a BPF_K check. | |
2045 | */ | |
2046 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
2047 | struct bpf_reg_state *false_reg, u64 val, | |
2048 | u8 opcode) | |
2049 | { | |
2050 | switch (opcode) { | |
2051 | case BPF_JEQ: | |
2052 | /* If this is false then we know nothing Jon Snow, but if it is | |
2053 | * true then we know for sure. | |
2054 | */ | |
2055 | true_reg->max_value = true_reg->min_value = val; | |
2056 | break; | |
2057 | case BPF_JNE: | |
2058 | /* If this is true we know nothing Jon Snow, but if it is false | |
2059 | * we know the value for sure; | |
2060 | */ | |
2061 | false_reg->max_value = false_reg->min_value = val; | |
2062 | break; | |
2063 | case BPF_JGT: | |
2064 | /* Unsigned comparison, the minimum value is 0. */ | |
2065 | false_reg->min_value = 0; | |
7e57fbb2 | 2066 | /* fallthrough */ |
48461135 JB |
2067 | case BPF_JSGT: |
2068 | /* If this is false then we know the maximum val is val, | |
2069 | * otherwise we know the min val is val+1. | |
2070 | */ | |
2071 | false_reg->max_value = val; | |
2072 | true_reg->min_value = val + 1; | |
2073 | break; | |
2074 | case BPF_JGE: | |
2075 | /* Unsigned comparison, the minimum value is 0. */ | |
2076 | false_reg->min_value = 0; | |
7e57fbb2 | 2077 | /* fallthrough */ |
48461135 JB |
2078 | case BPF_JSGE: |
2079 | /* If this is false then we know the maximum value is val - 1, | |
2080 | * otherwise we know the mimimum value is val. | |
2081 | */ | |
2082 | false_reg->max_value = val - 1; | |
2083 | true_reg->min_value = val; | |
2084 | break; | |
2085 | default: | |
2086 | break; | |
2087 | } | |
2088 | ||
2089 | check_reg_overflow(false_reg); | |
2090 | check_reg_overflow(true_reg); | |
2091 | } | |
2092 | ||
2093 | /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg | |
2094 | * is the variable reg. | |
2095 | */ | |
2096 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
2097 | struct bpf_reg_state *false_reg, u64 val, | |
2098 | u8 opcode) | |
2099 | { | |
2100 | switch (opcode) { | |
2101 | case BPF_JEQ: | |
2102 | /* If this is false then we know nothing Jon Snow, but if it is | |
2103 | * true then we know for sure. | |
2104 | */ | |
2105 | true_reg->max_value = true_reg->min_value = val; | |
2106 | break; | |
2107 | case BPF_JNE: | |
2108 | /* If this is true we know nothing Jon Snow, but if it is false | |
2109 | * we know the value for sure; | |
2110 | */ | |
2111 | false_reg->max_value = false_reg->min_value = val; | |
2112 | break; | |
2113 | case BPF_JGT: | |
2114 | /* Unsigned comparison, the minimum value is 0. */ | |
2115 | true_reg->min_value = 0; | |
7e57fbb2 | 2116 | /* fallthrough */ |
48461135 JB |
2117 | case BPF_JSGT: |
2118 | /* | |
2119 | * If this is false, then the val is <= the register, if it is | |
2120 | * true the register <= to the val. | |
2121 | */ | |
2122 | false_reg->min_value = val; | |
2123 | true_reg->max_value = val - 1; | |
2124 | break; | |
2125 | case BPF_JGE: | |
2126 | /* Unsigned comparison, the minimum value is 0. */ | |
2127 | true_reg->min_value = 0; | |
7e57fbb2 | 2128 | /* fallthrough */ |
48461135 JB |
2129 | case BPF_JSGE: |
2130 | /* If this is false then constant < register, if it is true then | |
2131 | * the register < constant. | |
2132 | */ | |
2133 | false_reg->min_value = val + 1; | |
2134 | true_reg->max_value = val; | |
2135 | break; | |
2136 | default: | |
2137 | break; | |
2138 | } | |
2139 | ||
2140 | check_reg_overflow(false_reg); | |
2141 | check_reg_overflow(true_reg); | |
2142 | } | |
2143 | ||
57a09bf0 TG |
2144 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
2145 | enum bpf_reg_type type) | |
2146 | { | |
2147 | struct bpf_reg_state *reg = ®s[regno]; | |
2148 | ||
2149 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
56f668df MKL |
2150 | if (type == UNKNOWN_VALUE) { |
2151 | __mark_reg_unknown_value(regs, regno); | |
2152 | } else if (reg->map_ptr->inner_map_meta) { | |
2153 | reg->type = CONST_PTR_TO_MAP; | |
2154 | reg->map_ptr = reg->map_ptr->inner_map_meta; | |
2155 | } else { | |
2156 | reg->type = type; | |
2157 | } | |
a08dd0da DB |
2158 | /* We don't need id from this point onwards anymore, thus we |
2159 | * should better reset it, so that state pruning has chances | |
2160 | * to take effect. | |
2161 | */ | |
2162 | reg->id = 0; | |
57a09bf0 TG |
2163 | } |
2164 | } | |
2165 | ||
2166 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
2167 | * be folded together at some point. | |
2168 | */ | |
2169 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
2170 | enum bpf_reg_type type) | |
2171 | { | |
2172 | struct bpf_reg_state *regs = state->regs; | |
a08dd0da | 2173 | u32 id = regs[regno].id; |
57a09bf0 TG |
2174 | int i; |
2175 | ||
2176 | for (i = 0; i < MAX_BPF_REG; i++) | |
a08dd0da | 2177 | mark_map_reg(regs, i, id, type); |
57a09bf0 TG |
2178 | |
2179 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
2180 | if (state->stack_slot_type[i] != STACK_SPILL) | |
2181 | continue; | |
a08dd0da | 2182 | mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, id, type); |
57a09bf0 TG |
2183 | } |
2184 | } | |
2185 | ||
58e2af8b | 2186 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
2187 | struct bpf_insn *insn, int *insn_idx) |
2188 | { | |
58e2af8b JK |
2189 | struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; |
2190 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; | |
17a52670 AS |
2191 | u8 opcode = BPF_OP(insn->code); |
2192 | int err; | |
2193 | ||
2194 | if (opcode > BPF_EXIT) { | |
2195 | verbose("invalid BPF_JMP opcode %x\n", opcode); | |
2196 | return -EINVAL; | |
2197 | } | |
2198 | ||
2199 | if (BPF_SRC(insn->code) == BPF_X) { | |
2200 | if (insn->imm != 0) { | |
2201 | verbose("BPF_JMP uses reserved fields\n"); | |
2202 | return -EINVAL; | |
2203 | } | |
2204 | ||
2205 | /* check src1 operand */ | |
2206 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2207 | if (err) | |
2208 | return err; | |
1be7f75d AS |
2209 | |
2210 | if (is_pointer_value(env, insn->src_reg)) { | |
2211 | verbose("R%d pointer comparison prohibited\n", | |
2212 | insn->src_reg); | |
2213 | return -EACCES; | |
2214 | } | |
17a52670 AS |
2215 | } else { |
2216 | if (insn->src_reg != BPF_REG_0) { | |
2217 | verbose("BPF_JMP uses reserved fields\n"); | |
2218 | return -EINVAL; | |
2219 | } | |
2220 | } | |
2221 | ||
2222 | /* check src2 operand */ | |
2223 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2224 | if (err) | |
2225 | return err; | |
2226 | ||
1a0dc1ac AS |
2227 | dst_reg = ®s[insn->dst_reg]; |
2228 | ||
17a52670 AS |
2229 | /* detect if R == 0 where R was initialized to zero earlier */ |
2230 | if (BPF_SRC(insn->code) == BPF_K && | |
2231 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
1a0dc1ac | 2232 | dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { |
17a52670 AS |
2233 | if (opcode == BPF_JEQ) { |
2234 | /* if (imm == imm) goto pc+off; | |
2235 | * only follow the goto, ignore fall-through | |
2236 | */ | |
2237 | *insn_idx += insn->off; | |
2238 | return 0; | |
2239 | } else { | |
2240 | /* if (imm != imm) goto pc+off; | |
2241 | * only follow fall-through branch, since | |
2242 | * that's where the program will go | |
2243 | */ | |
2244 | return 0; | |
2245 | } | |
2246 | } | |
2247 | ||
2248 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
2249 | if (!other_branch) | |
2250 | return -EFAULT; | |
2251 | ||
48461135 JB |
2252 | /* detect if we are comparing against a constant value so we can adjust |
2253 | * our min/max values for our dst register. | |
2254 | */ | |
2255 | if (BPF_SRC(insn->code) == BPF_X) { | |
2256 | if (regs[insn->src_reg].type == CONST_IMM) | |
2257 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2258 | dst_reg, regs[insn->src_reg].imm, | |
2259 | opcode); | |
2260 | else if (dst_reg->type == CONST_IMM) | |
2261 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
2262 | ®s[insn->src_reg], dst_reg->imm, | |
2263 | opcode); | |
2264 | } else { | |
2265 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2266 | dst_reg, insn->imm, opcode); | |
2267 | } | |
2268 | ||
58e2af8b | 2269 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 2270 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
2271 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
2272 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
2273 | /* Mark all identical map registers in each branch as either |
2274 | * safe or unknown depending R == 0 or R != 0 conditional. | |
2275 | */ | |
2276 | mark_map_regs(this_branch, insn->dst_reg, | |
2277 | opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE); | |
2278 | mark_map_regs(other_branch, insn->dst_reg, | |
2279 | opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE); | |
969bf05e AS |
2280 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && |
2281 | dst_reg->type == PTR_TO_PACKET && | |
2282 | regs[insn->src_reg].type == PTR_TO_PACKET_END) { | |
2d2be8ca DB |
2283 | find_good_pkt_pointers(this_branch, dst_reg); |
2284 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && | |
2285 | dst_reg->type == PTR_TO_PACKET_END && | |
2286 | regs[insn->src_reg].type == PTR_TO_PACKET) { | |
2287 | find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); | |
1be7f75d AS |
2288 | } else if (is_pointer_value(env, insn->dst_reg)) { |
2289 | verbose("R%d pointer comparison prohibited\n", insn->dst_reg); | |
2290 | return -EACCES; | |
17a52670 AS |
2291 | } |
2292 | if (log_level) | |
2d2be8ca | 2293 | print_verifier_state(this_branch); |
17a52670 AS |
2294 | return 0; |
2295 | } | |
2296 | ||
0246e64d AS |
2297 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
2298 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
2299 | { | |
2300 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
2301 | ||
2302 | return (struct bpf_map *) (unsigned long) imm64; | |
2303 | } | |
2304 | ||
17a52670 | 2305 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 2306 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2307 | { |
58e2af8b | 2308 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
2309 | int err; |
2310 | ||
2311 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
2312 | verbose("invalid BPF_LD_IMM insn\n"); | |
2313 | return -EINVAL; | |
2314 | } | |
2315 | if (insn->off != 0) { | |
2316 | verbose("BPF_LD_IMM64 uses reserved fields\n"); | |
2317 | return -EINVAL; | |
2318 | } | |
2319 | ||
2320 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
2321 | if (err) | |
2322 | return err; | |
2323 | ||
6b173873 | 2324 | if (insn->src_reg == 0) { |
6b173873 JK |
2325 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; |
2326 | ||
6b173873 JK |
2327 | regs[insn->dst_reg].type = CONST_IMM; |
2328 | regs[insn->dst_reg].imm = imm; | |
17a52670 | 2329 | return 0; |
6b173873 | 2330 | } |
17a52670 AS |
2331 | |
2332 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
2333 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
2334 | ||
2335 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
2336 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
2337 | return 0; | |
2338 | } | |
2339 | ||
96be4325 DB |
2340 | static bool may_access_skb(enum bpf_prog_type type) |
2341 | { | |
2342 | switch (type) { | |
2343 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
2344 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 2345 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
2346 | return true; |
2347 | default: | |
2348 | return false; | |
2349 | } | |
2350 | } | |
2351 | ||
ddd872bc AS |
2352 | /* verify safety of LD_ABS|LD_IND instructions: |
2353 | * - they can only appear in the programs where ctx == skb | |
2354 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
2355 | * preserve R6-R9, and store return value into R0 | |
2356 | * | |
2357 | * Implicit input: | |
2358 | * ctx == skb == R6 == CTX | |
2359 | * | |
2360 | * Explicit input: | |
2361 | * SRC == any register | |
2362 | * IMM == 32-bit immediate | |
2363 | * | |
2364 | * Output: | |
2365 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
2366 | */ | |
58e2af8b | 2367 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 2368 | { |
58e2af8b | 2369 | struct bpf_reg_state *regs = env->cur_state.regs; |
ddd872bc | 2370 | u8 mode = BPF_MODE(insn->code); |
58e2af8b | 2371 | struct bpf_reg_state *reg; |
ddd872bc AS |
2372 | int i, err; |
2373 | ||
24701ece | 2374 | if (!may_access_skb(env->prog->type)) { |
1a0dc1ac | 2375 | verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
2376 | return -EINVAL; |
2377 | } | |
2378 | ||
2379 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 2380 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 2381 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
1a0dc1ac | 2382 | verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
2383 | return -EINVAL; |
2384 | } | |
2385 | ||
2386 | /* check whether implicit source operand (register R6) is readable */ | |
2387 | err = check_reg_arg(regs, BPF_REG_6, SRC_OP); | |
2388 | if (err) | |
2389 | return err; | |
2390 | ||
2391 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
2392 | verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
2393 | return -EINVAL; | |
2394 | } | |
2395 | ||
2396 | if (mode == BPF_IND) { | |
2397 | /* check explicit source operand */ | |
2398 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2399 | if (err) | |
2400 | return err; | |
2401 | } | |
2402 | ||
2403 | /* reset caller saved regs to unreadable */ | |
2404 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
2405 | reg = regs + caller_saved[i]; | |
2406 | reg->type = NOT_INIT; | |
2407 | reg->imm = 0; | |
2408 | } | |
2409 | ||
2410 | /* mark destination R0 register as readable, since it contains | |
2411 | * the value fetched from the packet | |
2412 | */ | |
2413 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
2414 | return 0; | |
2415 | } | |
2416 | ||
475fb78f AS |
2417 | /* non-recursive DFS pseudo code |
2418 | * 1 procedure DFS-iterative(G,v): | |
2419 | * 2 label v as discovered | |
2420 | * 3 let S be a stack | |
2421 | * 4 S.push(v) | |
2422 | * 5 while S is not empty | |
2423 | * 6 t <- S.pop() | |
2424 | * 7 if t is what we're looking for: | |
2425 | * 8 return t | |
2426 | * 9 for all edges e in G.adjacentEdges(t) do | |
2427 | * 10 if edge e is already labelled | |
2428 | * 11 continue with the next edge | |
2429 | * 12 w <- G.adjacentVertex(t,e) | |
2430 | * 13 if vertex w is not discovered and not explored | |
2431 | * 14 label e as tree-edge | |
2432 | * 15 label w as discovered | |
2433 | * 16 S.push(w) | |
2434 | * 17 continue at 5 | |
2435 | * 18 else if vertex w is discovered | |
2436 | * 19 label e as back-edge | |
2437 | * 20 else | |
2438 | * 21 // vertex w is explored | |
2439 | * 22 label e as forward- or cross-edge | |
2440 | * 23 label t as explored | |
2441 | * 24 S.pop() | |
2442 | * | |
2443 | * convention: | |
2444 | * 0x10 - discovered | |
2445 | * 0x11 - discovered and fall-through edge labelled | |
2446 | * 0x12 - discovered and fall-through and branch edges labelled | |
2447 | * 0x20 - explored | |
2448 | */ | |
2449 | ||
2450 | enum { | |
2451 | DISCOVERED = 0x10, | |
2452 | EXPLORED = 0x20, | |
2453 | FALLTHROUGH = 1, | |
2454 | BRANCH = 2, | |
2455 | }; | |
2456 | ||
58e2af8b | 2457 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 2458 | |
475fb78f AS |
2459 | static int *insn_stack; /* stack of insns to process */ |
2460 | static int cur_stack; /* current stack index */ | |
2461 | static int *insn_state; | |
2462 | ||
2463 | /* t, w, e - match pseudo-code above: | |
2464 | * t - index of current instruction | |
2465 | * w - next instruction | |
2466 | * e - edge | |
2467 | */ | |
58e2af8b | 2468 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
2469 | { |
2470 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
2471 | return 0; | |
2472 | ||
2473 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
2474 | return 0; | |
2475 | ||
2476 | if (w < 0 || w >= env->prog->len) { | |
2477 | verbose("jump out of range from insn %d to %d\n", t, w); | |
2478 | return -EINVAL; | |
2479 | } | |
2480 | ||
f1bca824 AS |
2481 | if (e == BRANCH) |
2482 | /* mark branch target for state pruning */ | |
2483 | env->explored_states[w] = STATE_LIST_MARK; | |
2484 | ||
475fb78f AS |
2485 | if (insn_state[w] == 0) { |
2486 | /* tree-edge */ | |
2487 | insn_state[t] = DISCOVERED | e; | |
2488 | insn_state[w] = DISCOVERED; | |
2489 | if (cur_stack >= env->prog->len) | |
2490 | return -E2BIG; | |
2491 | insn_stack[cur_stack++] = w; | |
2492 | return 1; | |
2493 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
2494 | verbose("back-edge from insn %d to %d\n", t, w); | |
2495 | return -EINVAL; | |
2496 | } else if (insn_state[w] == EXPLORED) { | |
2497 | /* forward- or cross-edge */ | |
2498 | insn_state[t] = DISCOVERED | e; | |
2499 | } else { | |
2500 | verbose("insn state internal bug\n"); | |
2501 | return -EFAULT; | |
2502 | } | |
2503 | return 0; | |
2504 | } | |
2505 | ||
2506 | /* non-recursive depth-first-search to detect loops in BPF program | |
2507 | * loop == back-edge in directed graph | |
2508 | */ | |
58e2af8b | 2509 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
2510 | { |
2511 | struct bpf_insn *insns = env->prog->insnsi; | |
2512 | int insn_cnt = env->prog->len; | |
2513 | int ret = 0; | |
2514 | int i, t; | |
2515 | ||
2516 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2517 | if (!insn_state) | |
2518 | return -ENOMEM; | |
2519 | ||
2520 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2521 | if (!insn_stack) { | |
2522 | kfree(insn_state); | |
2523 | return -ENOMEM; | |
2524 | } | |
2525 | ||
2526 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
2527 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
2528 | cur_stack = 1; | |
2529 | ||
2530 | peek_stack: | |
2531 | if (cur_stack == 0) | |
2532 | goto check_state; | |
2533 | t = insn_stack[cur_stack - 1]; | |
2534 | ||
2535 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
2536 | u8 opcode = BPF_OP(insns[t].code); | |
2537 | ||
2538 | if (opcode == BPF_EXIT) { | |
2539 | goto mark_explored; | |
2540 | } else if (opcode == BPF_CALL) { | |
2541 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2542 | if (ret == 1) | |
2543 | goto peek_stack; | |
2544 | else if (ret < 0) | |
2545 | goto err_free; | |
07016151 DB |
2546 | if (t + 1 < insn_cnt) |
2547 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2548 | } else if (opcode == BPF_JA) { |
2549 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
2550 | ret = -EINVAL; | |
2551 | goto err_free; | |
2552 | } | |
2553 | /* unconditional jump with single edge */ | |
2554 | ret = push_insn(t, t + insns[t].off + 1, | |
2555 | FALLTHROUGH, env); | |
2556 | if (ret == 1) | |
2557 | goto peek_stack; | |
2558 | else if (ret < 0) | |
2559 | goto err_free; | |
f1bca824 AS |
2560 | /* tell verifier to check for equivalent states |
2561 | * after every call and jump | |
2562 | */ | |
c3de6317 AS |
2563 | if (t + 1 < insn_cnt) |
2564 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2565 | } else { |
2566 | /* conditional jump with two edges */ | |
2567 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2568 | if (ret == 1) | |
2569 | goto peek_stack; | |
2570 | else if (ret < 0) | |
2571 | goto err_free; | |
2572 | ||
2573 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
2574 | if (ret == 1) | |
2575 | goto peek_stack; | |
2576 | else if (ret < 0) | |
2577 | goto err_free; | |
2578 | } | |
2579 | } else { | |
2580 | /* all other non-branch instructions with single | |
2581 | * fall-through edge | |
2582 | */ | |
2583 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2584 | if (ret == 1) | |
2585 | goto peek_stack; | |
2586 | else if (ret < 0) | |
2587 | goto err_free; | |
2588 | } | |
2589 | ||
2590 | mark_explored: | |
2591 | insn_state[t] = EXPLORED; | |
2592 | if (cur_stack-- <= 0) { | |
2593 | verbose("pop stack internal bug\n"); | |
2594 | ret = -EFAULT; | |
2595 | goto err_free; | |
2596 | } | |
2597 | goto peek_stack; | |
2598 | ||
2599 | check_state: | |
2600 | for (i = 0; i < insn_cnt; i++) { | |
2601 | if (insn_state[i] != EXPLORED) { | |
2602 | verbose("unreachable insn %d\n", i); | |
2603 | ret = -EINVAL; | |
2604 | goto err_free; | |
2605 | } | |
2606 | } | |
2607 | ret = 0; /* cfg looks good */ | |
2608 | ||
2609 | err_free: | |
2610 | kfree(insn_state); | |
2611 | kfree(insn_stack); | |
2612 | return ret; | |
2613 | } | |
2614 | ||
969bf05e AS |
2615 | /* the following conditions reduce the number of explored insns |
2616 | * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet | |
2617 | */ | |
58e2af8b JK |
2618 | static bool compare_ptrs_to_packet(struct bpf_reg_state *old, |
2619 | struct bpf_reg_state *cur) | |
969bf05e AS |
2620 | { |
2621 | if (old->id != cur->id) | |
2622 | return false; | |
2623 | ||
2624 | /* old ptr_to_packet is more conservative, since it allows smaller | |
2625 | * range. Ex: | |
2626 | * old(off=0,r=10) is equal to cur(off=0,r=20), because | |
2627 | * old(off=0,r=10) means that with range=10 the verifier proceeded | |
2628 | * further and found no issues with the program. Now we're in the same | |
2629 | * spot with cur(off=0,r=20), so we're safe too, since anything further | |
2630 | * will only be looking at most 10 bytes after this pointer. | |
2631 | */ | |
2632 | if (old->off == cur->off && old->range < cur->range) | |
2633 | return true; | |
2634 | ||
2635 | /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) | |
2636 | * since both cannot be used for packet access and safe(old) | |
2637 | * pointer has smaller off that could be used for further | |
2638 | * 'if (ptr > data_end)' check | |
2639 | * Ex: | |
2640 | * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean | |
2641 | * that we cannot access the packet. | |
2642 | * The safe range is: | |
2643 | * [ptr, ptr + range - off) | |
2644 | * so whenever off >=range, it means no safe bytes from this pointer. | |
2645 | * When comparing old->off <= cur->off, it means that older code | |
2646 | * went with smaller offset and that offset was later | |
2647 | * used to figure out the safe range after 'if (ptr > data_end)' check | |
2648 | * Say, 'old' state was explored like: | |
2649 | * ... R3(off=0, r=0) | |
2650 | * R4 = R3 + 20 | |
2651 | * ... now R4(off=20,r=0) <-- here | |
2652 | * if (R4 > data_end) | |
2653 | * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. | |
2654 | * ... the code further went all the way to bpf_exit. | |
2655 | * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). | |
2656 | * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier | |
2657 | * goes further, such cur_R4 will give larger safe packet range after | |
2658 | * 'if (R4 > data_end)' and all further insn were already good with r=20, | |
2659 | * so they will be good with r=30 and we can prune the search. | |
2660 | */ | |
2661 | if (old->off <= cur->off && | |
2662 | old->off >= old->range && cur->off >= cur->range) | |
2663 | return true; | |
2664 | ||
2665 | return false; | |
2666 | } | |
2667 | ||
f1bca824 AS |
2668 | /* compare two verifier states |
2669 | * | |
2670 | * all states stored in state_list are known to be valid, since | |
2671 | * verifier reached 'bpf_exit' instruction through them | |
2672 | * | |
2673 | * this function is called when verifier exploring different branches of | |
2674 | * execution popped from the state stack. If it sees an old state that has | |
2675 | * more strict register state and more strict stack state then this execution | |
2676 | * branch doesn't need to be explored further, since verifier already | |
2677 | * concluded that more strict state leads to valid finish. | |
2678 | * | |
2679 | * Therefore two states are equivalent if register state is more conservative | |
2680 | * and explored stack state is more conservative than the current one. | |
2681 | * Example: | |
2682 | * explored current | |
2683 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
2684 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
2685 | * | |
2686 | * In other words if current stack state (one being explored) has more | |
2687 | * valid slots than old one that already passed validation, it means | |
2688 | * the verifier can stop exploring and conclude that current state is valid too | |
2689 | * | |
2690 | * Similarly with registers. If explored state has register type as invalid | |
2691 | * whereas register type in current state is meaningful, it means that | |
2692 | * the current state will reach 'bpf_exit' instruction safely | |
2693 | */ | |
48461135 JB |
2694 | static bool states_equal(struct bpf_verifier_env *env, |
2695 | struct bpf_verifier_state *old, | |
58e2af8b | 2696 | struct bpf_verifier_state *cur) |
f1bca824 | 2697 | { |
e2d2afe1 | 2698 | bool varlen_map_access = env->varlen_map_value_access; |
58e2af8b | 2699 | struct bpf_reg_state *rold, *rcur; |
f1bca824 AS |
2700 | int i; |
2701 | ||
2702 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
2703 | rold = &old->regs[i]; |
2704 | rcur = &cur->regs[i]; | |
2705 | ||
2706 | if (memcmp(rold, rcur, sizeof(*rold)) == 0) | |
2707 | continue; | |
2708 | ||
48461135 JB |
2709 | /* If the ranges were not the same, but everything else was and |
2710 | * we didn't do a variable access into a map then we are a-ok. | |
2711 | */ | |
e2d2afe1 | 2712 | if (!varlen_map_access && |
d2a4dd37 | 2713 | memcmp(rold, rcur, offsetofend(struct bpf_reg_state, id)) == 0) |
48461135 JB |
2714 | continue; |
2715 | ||
e2d2afe1 JB |
2716 | /* If we didn't map access then again we don't care about the |
2717 | * mismatched range values and it's ok if our old type was | |
2718 | * UNKNOWN and we didn't go to a NOT_INIT'ed reg. | |
2719 | */ | |
1a0dc1ac | 2720 | if (rold->type == NOT_INIT || |
e2d2afe1 JB |
2721 | (!varlen_map_access && rold->type == UNKNOWN_VALUE && |
2722 | rcur->type != NOT_INIT)) | |
1a0dc1ac AS |
2723 | continue; |
2724 | ||
969bf05e AS |
2725 | if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && |
2726 | compare_ptrs_to_packet(rold, rcur)) | |
2727 | continue; | |
2728 | ||
1a0dc1ac | 2729 | return false; |
f1bca824 AS |
2730 | } |
2731 | ||
2732 | for (i = 0; i < MAX_BPF_STACK; i++) { | |
9c399760 AS |
2733 | if (old->stack_slot_type[i] == STACK_INVALID) |
2734 | continue; | |
2735 | if (old->stack_slot_type[i] != cur->stack_slot_type[i]) | |
2736 | /* Ex: old explored (safe) state has STACK_SPILL in | |
2737 | * this stack slot, but current has has STACK_MISC -> | |
2738 | * this verifier states are not equivalent, | |
2739 | * return false to continue verification of this path | |
2740 | */ | |
f1bca824 | 2741 | return false; |
9c399760 AS |
2742 | if (i % BPF_REG_SIZE) |
2743 | continue; | |
2744 | if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], | |
2745 | &cur->spilled_regs[i / BPF_REG_SIZE], | |
2746 | sizeof(old->spilled_regs[0]))) | |
2747 | /* when explored and current stack slot types are | |
2748 | * the same, check that stored pointers types | |
2749 | * are the same as well. | |
2750 | * Ex: explored safe path could have stored | |
58e2af8b | 2751 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} |
9c399760 | 2752 | * but current path has stored: |
58e2af8b | 2753 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} |
9c399760 AS |
2754 | * such verifier states are not equivalent. |
2755 | * return false to continue verification of this path | |
2756 | */ | |
2757 | return false; | |
2758 | else | |
2759 | continue; | |
f1bca824 AS |
2760 | } |
2761 | return true; | |
2762 | } | |
2763 | ||
58e2af8b | 2764 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 2765 | { |
58e2af8b JK |
2766 | struct bpf_verifier_state_list *new_sl; |
2767 | struct bpf_verifier_state_list *sl; | |
f1bca824 AS |
2768 | |
2769 | sl = env->explored_states[insn_idx]; | |
2770 | if (!sl) | |
2771 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
2772 | * be doing state search here | |
2773 | */ | |
2774 | return 0; | |
2775 | ||
2776 | while (sl != STATE_LIST_MARK) { | |
48461135 | 2777 | if (states_equal(env, &sl->state, &env->cur_state)) |
f1bca824 AS |
2778 | /* reached equivalent register/stack state, |
2779 | * prune the search | |
2780 | */ | |
2781 | return 1; | |
2782 | sl = sl->next; | |
2783 | } | |
2784 | ||
2785 | /* there were no equivalent states, remember current one. | |
2786 | * technically the current state is not proven to be safe yet, | |
2787 | * but it will either reach bpf_exit (which means it's safe) or | |
2788 | * it will be rejected. Since there are no loops, we won't be | |
2789 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
2790 | */ | |
58e2af8b | 2791 | new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); |
f1bca824 AS |
2792 | if (!new_sl) |
2793 | return -ENOMEM; | |
2794 | ||
2795 | /* add new state to the head of linked list */ | |
2796 | memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); | |
2797 | new_sl->next = env->explored_states[insn_idx]; | |
2798 | env->explored_states[insn_idx] = new_sl; | |
2799 | return 0; | |
2800 | } | |
2801 | ||
13a27dfc JK |
2802 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
2803 | int insn_idx, int prev_insn_idx) | |
2804 | { | |
2805 | if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) | |
2806 | return 0; | |
2807 | ||
2808 | return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
2809 | } | |
2810 | ||
58e2af8b | 2811 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 2812 | { |
58e2af8b | 2813 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 2814 | struct bpf_insn *insns = env->prog->insnsi; |
58e2af8b | 2815 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
2816 | int insn_cnt = env->prog->len; |
2817 | int insn_idx, prev_insn_idx = 0; | |
2818 | int insn_processed = 0; | |
2819 | bool do_print_state = false; | |
2820 | ||
2821 | init_reg_state(regs); | |
2822 | insn_idx = 0; | |
48461135 | 2823 | env->varlen_map_value_access = false; |
17a52670 AS |
2824 | for (;;) { |
2825 | struct bpf_insn *insn; | |
2826 | u8 class; | |
2827 | int err; | |
2828 | ||
2829 | if (insn_idx >= insn_cnt) { | |
2830 | verbose("invalid insn idx %d insn_cnt %d\n", | |
2831 | insn_idx, insn_cnt); | |
2832 | return -EFAULT; | |
2833 | } | |
2834 | ||
2835 | insn = &insns[insn_idx]; | |
2836 | class = BPF_CLASS(insn->code); | |
2837 | ||
07016151 | 2838 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
bc1750f3 | 2839 | verbose("BPF program is too large. Processed %d insn\n", |
17a52670 AS |
2840 | insn_processed); |
2841 | return -E2BIG; | |
2842 | } | |
2843 | ||
f1bca824 AS |
2844 | err = is_state_visited(env, insn_idx); |
2845 | if (err < 0) | |
2846 | return err; | |
2847 | if (err == 1) { | |
2848 | /* found equivalent state, can prune the search */ | |
2849 | if (log_level) { | |
2850 | if (do_print_state) | |
2851 | verbose("\nfrom %d to %d: safe\n", | |
2852 | prev_insn_idx, insn_idx); | |
2853 | else | |
2854 | verbose("%d: safe\n", insn_idx); | |
2855 | } | |
2856 | goto process_bpf_exit; | |
2857 | } | |
2858 | ||
17a52670 AS |
2859 | if (log_level && do_print_state) { |
2860 | verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); | |
1a0dc1ac | 2861 | print_verifier_state(&env->cur_state); |
17a52670 AS |
2862 | do_print_state = false; |
2863 | } | |
2864 | ||
2865 | if (log_level) { | |
2866 | verbose("%d: ", insn_idx); | |
0d0e5769 | 2867 | print_bpf_insn(env, insn); |
17a52670 AS |
2868 | } |
2869 | ||
13a27dfc JK |
2870 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
2871 | if (err) | |
2872 | return err; | |
2873 | ||
17a52670 | 2874 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 2875 | err = check_alu_op(env, insn); |
17a52670 AS |
2876 | if (err) |
2877 | return err; | |
2878 | ||
2879 | } else if (class == BPF_LDX) { | |
3df126f3 | 2880 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
2881 | |
2882 | /* check for reserved fields is already done */ | |
2883 | ||
17a52670 AS |
2884 | /* check src operand */ |
2885 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2886 | if (err) | |
2887 | return err; | |
2888 | ||
2889 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
2890 | if (err) | |
2891 | return err; | |
2892 | ||
725f9dcd AS |
2893 | src_reg_type = regs[insn->src_reg].type; |
2894 | ||
17a52670 AS |
2895 | /* check that memory (src_reg + off) is readable, |
2896 | * the state of dst_reg will be updated by this func | |
2897 | */ | |
2898 | err = check_mem_access(env, insn->src_reg, insn->off, | |
2899 | BPF_SIZE(insn->code), BPF_READ, | |
2900 | insn->dst_reg); | |
2901 | if (err) | |
2902 | return err; | |
2903 | ||
ea2e7ce5 AS |
2904 | if (BPF_SIZE(insn->code) != BPF_W && |
2905 | BPF_SIZE(insn->code) != BPF_DW) { | |
725f9dcd AS |
2906 | insn_idx++; |
2907 | continue; | |
2908 | } | |
9bac3d6d | 2909 | |
3df126f3 JK |
2910 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
2911 | ||
2912 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
2913 | /* saw a valid insn |
2914 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 2915 | * save type to validate intersecting paths |
9bac3d6d | 2916 | */ |
3df126f3 | 2917 | *prev_src_type = src_reg_type; |
9bac3d6d | 2918 | |
3df126f3 | 2919 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 2920 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 2921 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
2922 | /* ABuser program is trying to use the same insn |
2923 | * dst_reg = *(u32*) (src_reg + off) | |
2924 | * with different pointer types: | |
2925 | * src_reg == ctx in one branch and | |
2926 | * src_reg == stack|map in some other branch. | |
2927 | * Reject it. | |
2928 | */ | |
2929 | verbose("same insn cannot be used with different pointers\n"); | |
2930 | return -EINVAL; | |
2931 | } | |
2932 | ||
17a52670 | 2933 | } else if (class == BPF_STX) { |
3df126f3 | 2934 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 2935 | |
17a52670 AS |
2936 | if (BPF_MODE(insn->code) == BPF_XADD) { |
2937 | err = check_xadd(env, insn); | |
2938 | if (err) | |
2939 | return err; | |
2940 | insn_idx++; | |
2941 | continue; | |
2942 | } | |
2943 | ||
17a52670 AS |
2944 | /* check src1 operand */ |
2945 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2946 | if (err) | |
2947 | return err; | |
2948 | /* check src2 operand */ | |
2949 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2950 | if (err) | |
2951 | return err; | |
2952 | ||
d691f9e8 AS |
2953 | dst_reg_type = regs[insn->dst_reg].type; |
2954 | ||
17a52670 AS |
2955 | /* check that memory (dst_reg + off) is writeable */ |
2956 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2957 | BPF_SIZE(insn->code), BPF_WRITE, | |
2958 | insn->src_reg); | |
2959 | if (err) | |
2960 | return err; | |
2961 | ||
3df126f3 JK |
2962 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
2963 | ||
2964 | if (*prev_dst_type == NOT_INIT) { | |
2965 | *prev_dst_type = dst_reg_type; | |
2966 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 2967 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 2968 | *prev_dst_type == PTR_TO_CTX)) { |
d691f9e8 AS |
2969 | verbose("same insn cannot be used with different pointers\n"); |
2970 | return -EINVAL; | |
2971 | } | |
2972 | ||
17a52670 AS |
2973 | } else if (class == BPF_ST) { |
2974 | if (BPF_MODE(insn->code) != BPF_MEM || | |
2975 | insn->src_reg != BPF_REG_0) { | |
2976 | verbose("BPF_ST uses reserved fields\n"); | |
2977 | return -EINVAL; | |
2978 | } | |
2979 | /* check src operand */ | |
2980 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2981 | if (err) | |
2982 | return err; | |
2983 | ||
2984 | /* check that memory (dst_reg + off) is writeable */ | |
2985 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2986 | BPF_SIZE(insn->code), BPF_WRITE, | |
2987 | -1); | |
2988 | if (err) | |
2989 | return err; | |
2990 | ||
2991 | } else if (class == BPF_JMP) { | |
2992 | u8 opcode = BPF_OP(insn->code); | |
2993 | ||
2994 | if (opcode == BPF_CALL) { | |
2995 | if (BPF_SRC(insn->code) != BPF_K || | |
2996 | insn->off != 0 || | |
2997 | insn->src_reg != BPF_REG_0 || | |
2998 | insn->dst_reg != BPF_REG_0) { | |
2999 | verbose("BPF_CALL uses reserved fields\n"); | |
3000 | return -EINVAL; | |
3001 | } | |
3002 | ||
81ed18ab | 3003 | err = check_call(env, insn->imm, insn_idx); |
17a52670 AS |
3004 | if (err) |
3005 | return err; | |
3006 | ||
3007 | } else if (opcode == BPF_JA) { | |
3008 | if (BPF_SRC(insn->code) != BPF_K || | |
3009 | insn->imm != 0 || | |
3010 | insn->src_reg != BPF_REG_0 || | |
3011 | insn->dst_reg != BPF_REG_0) { | |
3012 | verbose("BPF_JA uses reserved fields\n"); | |
3013 | return -EINVAL; | |
3014 | } | |
3015 | ||
3016 | insn_idx += insn->off + 1; | |
3017 | continue; | |
3018 | ||
3019 | } else if (opcode == BPF_EXIT) { | |
3020 | if (BPF_SRC(insn->code) != BPF_K || | |
3021 | insn->imm != 0 || | |
3022 | insn->src_reg != BPF_REG_0 || | |
3023 | insn->dst_reg != BPF_REG_0) { | |
3024 | verbose("BPF_EXIT uses reserved fields\n"); | |
3025 | return -EINVAL; | |
3026 | } | |
3027 | ||
3028 | /* eBPF calling convetion is such that R0 is used | |
3029 | * to return the value from eBPF program. | |
3030 | * Make sure that it's readable at this time | |
3031 | * of bpf_exit, which means that program wrote | |
3032 | * something into it earlier | |
3033 | */ | |
3034 | err = check_reg_arg(regs, BPF_REG_0, SRC_OP); | |
3035 | if (err) | |
3036 | return err; | |
3037 | ||
1be7f75d AS |
3038 | if (is_pointer_value(env, BPF_REG_0)) { |
3039 | verbose("R0 leaks addr as return value\n"); | |
3040 | return -EACCES; | |
3041 | } | |
3042 | ||
f1bca824 | 3043 | process_bpf_exit: |
17a52670 AS |
3044 | insn_idx = pop_stack(env, &prev_insn_idx); |
3045 | if (insn_idx < 0) { | |
3046 | break; | |
3047 | } else { | |
3048 | do_print_state = true; | |
3049 | continue; | |
3050 | } | |
3051 | } else { | |
3052 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
3053 | if (err) | |
3054 | return err; | |
3055 | } | |
3056 | } else if (class == BPF_LD) { | |
3057 | u8 mode = BPF_MODE(insn->code); | |
3058 | ||
3059 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
3060 | err = check_ld_abs(env, insn); |
3061 | if (err) | |
3062 | return err; | |
3063 | ||
17a52670 AS |
3064 | } else if (mode == BPF_IMM) { |
3065 | err = check_ld_imm(env, insn); | |
3066 | if (err) | |
3067 | return err; | |
3068 | ||
3069 | insn_idx++; | |
3070 | } else { | |
3071 | verbose("invalid BPF_LD mode\n"); | |
3072 | return -EINVAL; | |
3073 | } | |
48461135 | 3074 | reset_reg_range_values(regs, insn->dst_reg); |
17a52670 AS |
3075 | } else { |
3076 | verbose("unknown insn class %d\n", class); | |
3077 | return -EINVAL; | |
3078 | } | |
3079 | ||
3080 | insn_idx++; | |
3081 | } | |
3082 | ||
1a0dc1ac | 3083 | verbose("processed %d insns\n", insn_processed); |
17a52670 AS |
3084 | return 0; |
3085 | } | |
3086 | ||
56f668df MKL |
3087 | static int check_map_prealloc(struct bpf_map *map) |
3088 | { | |
3089 | return (map->map_type != BPF_MAP_TYPE_HASH && | |
bcc6b1b7 MKL |
3090 | map->map_type != BPF_MAP_TYPE_PERCPU_HASH && |
3091 | map->map_type != BPF_MAP_TYPE_HASH_OF_MAPS) || | |
56f668df MKL |
3092 | !(map->map_flags & BPF_F_NO_PREALLOC); |
3093 | } | |
3094 | ||
fdc15d38 AS |
3095 | static int check_map_prog_compatibility(struct bpf_map *map, |
3096 | struct bpf_prog *prog) | |
3097 | ||
3098 | { | |
56f668df MKL |
3099 | /* Make sure that BPF_PROG_TYPE_PERF_EVENT programs only use |
3100 | * preallocated hash maps, since doing memory allocation | |
3101 | * in overflow_handler can crash depending on where nmi got | |
3102 | * triggered. | |
3103 | */ | |
3104 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) { | |
3105 | if (!check_map_prealloc(map)) { | |
3106 | verbose("perf_event programs can only use preallocated hash map\n"); | |
3107 | return -EINVAL; | |
3108 | } | |
3109 | if (map->inner_map_meta && | |
3110 | !check_map_prealloc(map->inner_map_meta)) { | |
3111 | verbose("perf_event programs can only use preallocated inner hash map\n"); | |
3112 | return -EINVAL; | |
3113 | } | |
fdc15d38 AS |
3114 | } |
3115 | return 0; | |
3116 | } | |
3117 | ||
0246e64d AS |
3118 | /* look for pseudo eBPF instructions that access map FDs and |
3119 | * replace them with actual map pointers | |
3120 | */ | |
58e2af8b | 3121 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
3122 | { |
3123 | struct bpf_insn *insn = env->prog->insnsi; | |
3124 | int insn_cnt = env->prog->len; | |
fdc15d38 | 3125 | int i, j, err; |
0246e64d | 3126 | |
f1f7714e | 3127 | err = bpf_prog_calc_tag(env->prog); |
aafe6ae9 DB |
3128 | if (err) |
3129 | return err; | |
3130 | ||
0246e64d | 3131 | for (i = 0; i < insn_cnt; i++, insn++) { |
9bac3d6d | 3132 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 3133 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
9bac3d6d AS |
3134 | verbose("BPF_LDX uses reserved fields\n"); |
3135 | return -EINVAL; | |
3136 | } | |
3137 | ||
d691f9e8 AS |
3138 | if (BPF_CLASS(insn->code) == BPF_STX && |
3139 | ((BPF_MODE(insn->code) != BPF_MEM && | |
3140 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
3141 | verbose("BPF_STX uses reserved fields\n"); | |
3142 | return -EINVAL; | |
3143 | } | |
3144 | ||
0246e64d AS |
3145 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
3146 | struct bpf_map *map; | |
3147 | struct fd f; | |
3148 | ||
3149 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
3150 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
3151 | insn[1].off != 0) { | |
3152 | verbose("invalid bpf_ld_imm64 insn\n"); | |
3153 | return -EINVAL; | |
3154 | } | |
3155 | ||
3156 | if (insn->src_reg == 0) | |
3157 | /* valid generic load 64-bit imm */ | |
3158 | goto next_insn; | |
3159 | ||
3160 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
3161 | verbose("unrecognized bpf_ld_imm64 insn\n"); | |
3162 | return -EINVAL; | |
3163 | } | |
3164 | ||
3165 | f = fdget(insn->imm); | |
c2101297 | 3166 | map = __bpf_map_get(f); |
0246e64d AS |
3167 | if (IS_ERR(map)) { |
3168 | verbose("fd %d is not pointing to valid bpf_map\n", | |
3169 | insn->imm); | |
0246e64d AS |
3170 | return PTR_ERR(map); |
3171 | } | |
3172 | ||
fdc15d38 AS |
3173 | err = check_map_prog_compatibility(map, env->prog); |
3174 | if (err) { | |
3175 | fdput(f); | |
3176 | return err; | |
3177 | } | |
3178 | ||
0246e64d AS |
3179 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
3180 | insn[0].imm = (u32) (unsigned long) map; | |
3181 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
3182 | ||
3183 | /* check whether we recorded this map already */ | |
3184 | for (j = 0; j < env->used_map_cnt; j++) | |
3185 | if (env->used_maps[j] == map) { | |
3186 | fdput(f); | |
3187 | goto next_insn; | |
3188 | } | |
3189 | ||
3190 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
3191 | fdput(f); | |
3192 | return -E2BIG; | |
3193 | } | |
3194 | ||
0246e64d AS |
3195 | /* hold the map. If the program is rejected by verifier, |
3196 | * the map will be released by release_maps() or it | |
3197 | * will be used by the valid program until it's unloaded | |
3198 | * and all maps are released in free_bpf_prog_info() | |
3199 | */ | |
92117d84 AS |
3200 | map = bpf_map_inc(map, false); |
3201 | if (IS_ERR(map)) { | |
3202 | fdput(f); | |
3203 | return PTR_ERR(map); | |
3204 | } | |
3205 | env->used_maps[env->used_map_cnt++] = map; | |
3206 | ||
0246e64d AS |
3207 | fdput(f); |
3208 | next_insn: | |
3209 | insn++; | |
3210 | i++; | |
3211 | } | |
3212 | } | |
3213 | ||
3214 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
3215 | * 'struct bpf_map *' into a register instead of user map_fd. | |
3216 | * These pointers will be used later by verifier to validate map access. | |
3217 | */ | |
3218 | return 0; | |
3219 | } | |
3220 | ||
3221 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 3222 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
3223 | { |
3224 | int i; | |
3225 | ||
3226 | for (i = 0; i < env->used_map_cnt; i++) | |
3227 | bpf_map_put(env->used_maps[i]); | |
3228 | } | |
3229 | ||
3230 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 3231 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
3232 | { |
3233 | struct bpf_insn *insn = env->prog->insnsi; | |
3234 | int insn_cnt = env->prog->len; | |
3235 | int i; | |
3236 | ||
3237 | for (i = 0; i < insn_cnt; i++, insn++) | |
3238 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
3239 | insn->src_reg = 0; | |
3240 | } | |
3241 | ||
8041902d AS |
3242 | /* single env->prog->insni[off] instruction was replaced with the range |
3243 | * insni[off, off + cnt). Adjust corresponding insn_aux_data by copying | |
3244 | * [0, off) and [off, end) to new locations, so the patched range stays zero | |
3245 | */ | |
3246 | static int adjust_insn_aux_data(struct bpf_verifier_env *env, u32 prog_len, | |
3247 | u32 off, u32 cnt) | |
3248 | { | |
3249 | struct bpf_insn_aux_data *new_data, *old_data = env->insn_aux_data; | |
3250 | ||
3251 | if (cnt == 1) | |
3252 | return 0; | |
3253 | new_data = vzalloc(sizeof(struct bpf_insn_aux_data) * prog_len); | |
3254 | if (!new_data) | |
3255 | return -ENOMEM; | |
3256 | memcpy(new_data, old_data, sizeof(struct bpf_insn_aux_data) * off); | |
3257 | memcpy(new_data + off + cnt - 1, old_data + off, | |
3258 | sizeof(struct bpf_insn_aux_data) * (prog_len - off - cnt + 1)); | |
3259 | env->insn_aux_data = new_data; | |
3260 | vfree(old_data); | |
3261 | return 0; | |
3262 | } | |
3263 | ||
3264 | static struct bpf_prog *bpf_patch_insn_data(struct bpf_verifier_env *env, u32 off, | |
3265 | const struct bpf_insn *patch, u32 len) | |
3266 | { | |
3267 | struct bpf_prog *new_prog; | |
3268 | ||
3269 | new_prog = bpf_patch_insn_single(env->prog, off, patch, len); | |
3270 | if (!new_prog) | |
3271 | return NULL; | |
3272 | if (adjust_insn_aux_data(env, new_prog->len, off, len)) | |
3273 | return NULL; | |
3274 | return new_prog; | |
3275 | } | |
3276 | ||
9bac3d6d AS |
3277 | /* convert load instructions that access fields of 'struct __sk_buff' |
3278 | * into sequence of instructions that access fields of 'struct sk_buff' | |
3279 | */ | |
58e2af8b | 3280 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 3281 | { |
36bbef52 | 3282 | const struct bpf_verifier_ops *ops = env->prog->aux->ops; |
3df126f3 | 3283 | const int insn_cnt = env->prog->len; |
36bbef52 | 3284 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 3285 | struct bpf_prog *new_prog; |
d691f9e8 | 3286 | enum bpf_access_type type; |
3df126f3 | 3287 | int i, cnt, delta = 0; |
9bac3d6d | 3288 | |
36bbef52 DB |
3289 | if (ops->gen_prologue) { |
3290 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
3291 | env->prog); | |
3292 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
3293 | verbose("bpf verifier is misconfigured\n"); | |
3294 | return -EINVAL; | |
3295 | } else if (cnt) { | |
8041902d | 3296 | new_prog = bpf_patch_insn_data(env, 0, insn_buf, cnt); |
36bbef52 DB |
3297 | if (!new_prog) |
3298 | return -ENOMEM; | |
8041902d | 3299 | |
36bbef52 | 3300 | env->prog = new_prog; |
3df126f3 | 3301 | delta += cnt - 1; |
36bbef52 DB |
3302 | } |
3303 | } | |
3304 | ||
3305 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
3306 | return 0; |
3307 | ||
3df126f3 | 3308 | insn = env->prog->insnsi + delta; |
36bbef52 | 3309 | |
9bac3d6d | 3310 | for (i = 0; i < insn_cnt; i++, insn++) { |
62c7989b DB |
3311 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_B) || |
3312 | insn->code == (BPF_LDX | BPF_MEM | BPF_H) || | |
3313 | insn->code == (BPF_LDX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 3314 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) |
d691f9e8 | 3315 | type = BPF_READ; |
62c7989b DB |
3316 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_B) || |
3317 | insn->code == (BPF_STX | BPF_MEM | BPF_H) || | |
3318 | insn->code == (BPF_STX | BPF_MEM | BPF_W) || | |
ea2e7ce5 | 3319 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) |
d691f9e8 AS |
3320 | type = BPF_WRITE; |
3321 | else | |
9bac3d6d AS |
3322 | continue; |
3323 | ||
8041902d | 3324 | if (env->insn_aux_data[i + delta].ptr_type != PTR_TO_CTX) |
9bac3d6d | 3325 | continue; |
9bac3d6d | 3326 | |
6b8cc1d1 | 3327 | cnt = ops->convert_ctx_access(type, insn, insn_buf, env->prog); |
9bac3d6d AS |
3328 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { |
3329 | verbose("bpf verifier is misconfigured\n"); | |
3330 | return -EINVAL; | |
3331 | } | |
3332 | ||
8041902d | 3333 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, cnt); |
9bac3d6d AS |
3334 | if (!new_prog) |
3335 | return -ENOMEM; | |
3336 | ||
3df126f3 | 3337 | delta += cnt - 1; |
9bac3d6d AS |
3338 | |
3339 | /* keep walking new program and skip insns we just inserted */ | |
3340 | env->prog = new_prog; | |
3df126f3 | 3341 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
3342 | } |
3343 | ||
3344 | return 0; | |
3345 | } | |
3346 | ||
79741b3b | 3347 | /* fixup insn->imm field of bpf_call instructions |
81ed18ab | 3348 | * and inline eligible helpers as explicit sequence of BPF instructions |
e245c5c6 AS |
3349 | * |
3350 | * this function is called after eBPF program passed verification | |
3351 | */ | |
79741b3b | 3352 | static int fixup_bpf_calls(struct bpf_verifier_env *env) |
e245c5c6 | 3353 | { |
79741b3b AS |
3354 | struct bpf_prog *prog = env->prog; |
3355 | struct bpf_insn *insn = prog->insnsi; | |
e245c5c6 | 3356 | const struct bpf_func_proto *fn; |
79741b3b | 3357 | const int insn_cnt = prog->len; |
81ed18ab AS |
3358 | struct bpf_insn insn_buf[16]; |
3359 | struct bpf_prog *new_prog; | |
3360 | struct bpf_map *map_ptr; | |
3361 | int i, cnt, delta = 0; | |
e245c5c6 | 3362 | |
79741b3b AS |
3363 | for (i = 0; i < insn_cnt; i++, insn++) { |
3364 | if (insn->code != (BPF_JMP | BPF_CALL)) | |
3365 | continue; | |
e245c5c6 | 3366 | |
79741b3b AS |
3367 | if (insn->imm == BPF_FUNC_get_route_realm) |
3368 | prog->dst_needed = 1; | |
3369 | if (insn->imm == BPF_FUNC_get_prandom_u32) | |
3370 | bpf_user_rnd_init_once(); | |
79741b3b | 3371 | if (insn->imm == BPF_FUNC_tail_call) { |
7b9f6da1 DM |
3372 | /* If we tail call into other programs, we |
3373 | * cannot make any assumptions since they can | |
3374 | * be replaced dynamically during runtime in | |
3375 | * the program array. | |
3376 | */ | |
3377 | prog->cb_access = 1; | |
7b9f6da1 | 3378 | |
79741b3b AS |
3379 | /* mark bpf_tail_call as different opcode to avoid |
3380 | * conditional branch in the interpeter for every normal | |
3381 | * call and to prevent accidental JITing by JIT compiler | |
3382 | * that doesn't support bpf_tail_call yet | |
e245c5c6 | 3383 | */ |
79741b3b AS |
3384 | insn->imm = 0; |
3385 | insn->code |= BPF_X; | |
3386 | continue; | |
3387 | } | |
e245c5c6 | 3388 | |
81ed18ab AS |
3389 | if (ebpf_jit_enabled() && insn->imm == BPF_FUNC_map_lookup_elem) { |
3390 | map_ptr = env->insn_aux_data[i + delta].map_ptr; | |
fad73a1a MKL |
3391 | if (map_ptr == BPF_MAP_PTR_POISON || |
3392 | !map_ptr->ops->map_gen_lookup) | |
81ed18ab AS |
3393 | goto patch_call_imm; |
3394 | ||
3395 | cnt = map_ptr->ops->map_gen_lookup(map_ptr, insn_buf); | |
3396 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { | |
3397 | verbose("bpf verifier is misconfigured\n"); | |
3398 | return -EINVAL; | |
3399 | } | |
3400 | ||
3401 | new_prog = bpf_patch_insn_data(env, i + delta, insn_buf, | |
3402 | cnt); | |
3403 | if (!new_prog) | |
3404 | return -ENOMEM; | |
3405 | ||
3406 | delta += cnt - 1; | |
3407 | ||
3408 | /* keep walking new program and skip insns we just inserted */ | |
3409 | env->prog = prog = new_prog; | |
3410 | insn = new_prog->insnsi + i + delta; | |
3411 | continue; | |
3412 | } | |
3413 | ||
3414 | patch_call_imm: | |
79741b3b AS |
3415 | fn = prog->aux->ops->get_func_proto(insn->imm); |
3416 | /* all functions that have prototype and verifier allowed | |
3417 | * programs to call them, must be real in-kernel functions | |
3418 | */ | |
3419 | if (!fn->func) { | |
3420 | verbose("kernel subsystem misconfigured func %s#%d\n", | |
3421 | func_id_name(insn->imm), insn->imm); | |
3422 | return -EFAULT; | |
e245c5c6 | 3423 | } |
79741b3b | 3424 | insn->imm = fn->func - __bpf_call_base; |
e245c5c6 | 3425 | } |
e245c5c6 | 3426 | |
79741b3b AS |
3427 | return 0; |
3428 | } | |
e245c5c6 | 3429 | |
58e2af8b | 3430 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 3431 | { |
58e2af8b | 3432 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
3433 | int i; |
3434 | ||
3435 | if (!env->explored_states) | |
3436 | return; | |
3437 | ||
3438 | for (i = 0; i < env->prog->len; i++) { | |
3439 | sl = env->explored_states[i]; | |
3440 | ||
3441 | if (sl) | |
3442 | while (sl != STATE_LIST_MARK) { | |
3443 | sln = sl->next; | |
3444 | kfree(sl); | |
3445 | sl = sln; | |
3446 | } | |
3447 | } | |
3448 | ||
3449 | kfree(env->explored_states); | |
3450 | } | |
3451 | ||
9bac3d6d | 3452 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 3453 | { |
cbd35700 | 3454 | char __user *log_ubuf = NULL; |
58e2af8b | 3455 | struct bpf_verifier_env *env; |
51580e79 AS |
3456 | int ret = -EINVAL; |
3457 | ||
58e2af8b | 3458 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
3459 | * allocate/free it every time bpf_check() is called |
3460 | */ | |
58e2af8b | 3461 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
3462 | if (!env) |
3463 | return -ENOMEM; | |
3464 | ||
3df126f3 JK |
3465 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
3466 | (*prog)->len); | |
3467 | ret = -ENOMEM; | |
3468 | if (!env->insn_aux_data) | |
3469 | goto err_free_env; | |
9bac3d6d | 3470 | env->prog = *prog; |
0246e64d | 3471 | |
cbd35700 AS |
3472 | /* grab the mutex to protect few globals used by verifier */ |
3473 | mutex_lock(&bpf_verifier_lock); | |
3474 | ||
3475 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
3476 | /* user requested verbose verifier output | |
3477 | * and supplied buffer to store the verification trace | |
3478 | */ | |
3479 | log_level = attr->log_level; | |
3480 | log_ubuf = (char __user *) (unsigned long) attr->log_buf; | |
3481 | log_size = attr->log_size; | |
3482 | log_len = 0; | |
3483 | ||
3484 | ret = -EINVAL; | |
3485 | /* log_* values have to be sane */ | |
3486 | if (log_size < 128 || log_size > UINT_MAX >> 8 || | |
3487 | log_level == 0 || log_ubuf == NULL) | |
3df126f3 | 3488 | goto err_unlock; |
cbd35700 AS |
3489 | |
3490 | ret = -ENOMEM; | |
3491 | log_buf = vmalloc(log_size); | |
3492 | if (!log_buf) | |
3df126f3 | 3493 | goto err_unlock; |
cbd35700 AS |
3494 | } else { |
3495 | log_level = 0; | |
3496 | } | |
3497 | ||
0246e64d AS |
3498 | ret = replace_map_fd_with_map_ptr(env); |
3499 | if (ret < 0) | |
3500 | goto skip_full_check; | |
3501 | ||
9bac3d6d | 3502 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 3503 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
3504 | GFP_USER); |
3505 | ret = -ENOMEM; | |
3506 | if (!env->explored_states) | |
3507 | goto skip_full_check; | |
3508 | ||
475fb78f AS |
3509 | ret = check_cfg(env); |
3510 | if (ret < 0) | |
3511 | goto skip_full_check; | |
3512 | ||
1be7f75d AS |
3513 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
3514 | ||
17a52670 | 3515 | ret = do_check(env); |
cbd35700 | 3516 | |
0246e64d | 3517 | skip_full_check: |
17a52670 | 3518 | while (pop_stack(env, NULL) >= 0); |
f1bca824 | 3519 | free_states(env); |
0246e64d | 3520 | |
9bac3d6d AS |
3521 | if (ret == 0) |
3522 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
3523 | ret = convert_ctx_accesses(env); | |
3524 | ||
e245c5c6 | 3525 | if (ret == 0) |
79741b3b | 3526 | ret = fixup_bpf_calls(env); |
e245c5c6 | 3527 | |
cbd35700 AS |
3528 | if (log_level && log_len >= log_size - 1) { |
3529 | BUG_ON(log_len >= log_size); | |
3530 | /* verifier log exceeded user supplied buffer */ | |
3531 | ret = -ENOSPC; | |
3532 | /* fall through to return what was recorded */ | |
3533 | } | |
3534 | ||
3535 | /* copy verifier log back to user space including trailing zero */ | |
3536 | if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { | |
3537 | ret = -EFAULT; | |
3538 | goto free_log_buf; | |
3539 | } | |
3540 | ||
0246e64d AS |
3541 | if (ret == 0 && env->used_map_cnt) { |
3542 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
3543 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
3544 | sizeof(env->used_maps[0]), | |
3545 | GFP_KERNEL); | |
0246e64d | 3546 | |
9bac3d6d | 3547 | if (!env->prog->aux->used_maps) { |
0246e64d AS |
3548 | ret = -ENOMEM; |
3549 | goto free_log_buf; | |
3550 | } | |
3551 | ||
9bac3d6d | 3552 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 3553 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 3554 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
3555 | |
3556 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
3557 | * bpf_ld_imm64 instructions | |
3558 | */ | |
3559 | convert_pseudo_ld_imm64(env); | |
3560 | } | |
cbd35700 AS |
3561 | |
3562 | free_log_buf: | |
3563 | if (log_level) | |
3564 | vfree(log_buf); | |
9bac3d6d | 3565 | if (!env->prog->aux->used_maps) |
0246e64d AS |
3566 | /* if we didn't copy map pointers into bpf_prog_info, release |
3567 | * them now. Otherwise free_bpf_prog_info() will release them. | |
3568 | */ | |
3569 | release_maps(env); | |
9bac3d6d | 3570 | *prog = env->prog; |
3df126f3 | 3571 | err_unlock: |
cbd35700 | 3572 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
3573 | vfree(env->insn_aux_data); |
3574 | err_free_env: | |
3575 | kfree(env); | |
51580e79 AS |
3576 | return ret; |
3577 | } | |
13a27dfc JK |
3578 | |
3579 | int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, | |
3580 | void *priv) | |
3581 | { | |
3582 | struct bpf_verifier_env *env; | |
3583 | int ret; | |
3584 | ||
3585 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); | |
3586 | if (!env) | |
3587 | return -ENOMEM; | |
3588 | ||
3589 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * | |
3590 | prog->len); | |
3591 | ret = -ENOMEM; | |
3592 | if (!env->insn_aux_data) | |
3593 | goto err_free_env; | |
3594 | env->prog = prog; | |
3595 | env->analyzer_ops = ops; | |
3596 | env->analyzer_priv = priv; | |
3597 | ||
3598 | /* grab the mutex to protect few globals used by verifier */ | |
3599 | mutex_lock(&bpf_verifier_lock); | |
3600 | ||
3601 | log_level = 0; | |
3602 | ||
3603 | env->explored_states = kcalloc(env->prog->len, | |
3604 | sizeof(struct bpf_verifier_state_list *), | |
3605 | GFP_KERNEL); | |
3606 | ret = -ENOMEM; | |
3607 | if (!env->explored_states) | |
3608 | goto skip_full_check; | |
3609 | ||
3610 | ret = check_cfg(env); | |
3611 | if (ret < 0) | |
3612 | goto skip_full_check; | |
3613 | ||
3614 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); | |
3615 | ||
3616 | ret = do_check(env); | |
3617 | ||
3618 | skip_full_check: | |
3619 | while (pop_stack(env, NULL) >= 0); | |
3620 | free_states(env); | |
3621 | ||
3622 | mutex_unlock(&bpf_verifier_lock); | |
3623 | vfree(env->insn_aux_data); | |
3624 | err_free_env: | |
3625 | kfree(env); | |
3626 | return ret; | |
3627 | } | |
3628 | EXPORT_SYMBOL_GPL(bpf_analyzer); |