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> | |
22 | ||
23 | /* bpf_check() is a static code analyzer that walks eBPF program | |
24 | * instruction by instruction and updates register/stack state. | |
25 | * All paths of conditional branches are analyzed until 'bpf_exit' insn. | |
26 | * | |
27 | * The first pass is depth-first-search to check that the program is a DAG. | |
28 | * It rejects the following programs: | |
29 | * - larger than BPF_MAXINSNS insns | |
30 | * - if loop is present (detected via back-edge) | |
31 | * - unreachable insns exist (shouldn't be a forest. program = one function) | |
32 | * - out of bounds or malformed jumps | |
33 | * The second pass is all possible path descent from the 1st insn. | |
34 | * Since it's analyzing all pathes through the program, the length of the | |
35 | * analysis is limited to 32k insn, which may be hit even if total number of | |
36 | * insn is less then 4K, but there are too many branches that change stack/regs. | |
37 | * Number of 'branches to be analyzed' is limited to 1k | |
38 | * | |
39 | * On entry to each instruction, each register has a type, and the instruction | |
40 | * changes the types of the registers depending on instruction semantics. | |
41 | * If instruction is BPF_MOV64_REG(BPF_REG_1, BPF_REG_5), then type of R5 is | |
42 | * copied to R1. | |
43 | * | |
44 | * All registers are 64-bit. | |
45 | * R0 - return register | |
46 | * R1-R5 argument passing registers | |
47 | * R6-R9 callee saved registers | |
48 | * R10 - frame pointer read-only | |
49 | * | |
50 | * At the start of BPF program the register R1 contains a pointer to bpf_context | |
51 | * and has type PTR_TO_CTX. | |
52 | * | |
53 | * Verifier tracks arithmetic operations on pointers in case: | |
54 | * BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), | |
55 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -20), | |
56 | * 1st insn copies R10 (which has FRAME_PTR) type into R1 | |
57 | * and 2nd arithmetic instruction is pattern matched to recognize | |
58 | * that it wants to construct a pointer to some element within stack. | |
59 | * So after 2nd insn, the register R1 has type PTR_TO_STACK | |
60 | * (and -20 constant is saved for further stack bounds checking). | |
61 | * Meaning that this reg is a pointer to stack plus known immediate constant. | |
62 | * | |
63 | * Most of the time the registers have UNKNOWN_VALUE type, which | |
64 | * means the register has some value, but it's not a valid pointer. | |
65 | * (like pointer plus pointer becomes UNKNOWN_VALUE type) | |
66 | * | |
67 | * When verifier sees load or store instructions the type of base register | |
68 | * can be: PTR_TO_MAP_VALUE, PTR_TO_CTX, FRAME_PTR. These are three pointer | |
69 | * types recognized by check_mem_access() function. | |
70 | * | |
71 | * PTR_TO_MAP_VALUE means that this register is pointing to 'map element value' | |
72 | * and the range of [ptr, ptr + map's value_size) is accessible. | |
73 | * | |
74 | * registers used to pass values to function calls are checked against | |
75 | * function argument constraints. | |
76 | * | |
77 | * ARG_PTR_TO_MAP_KEY is one of such argument constraints. | |
78 | * It means that the register type passed to this function must be | |
79 | * PTR_TO_STACK and it will be used inside the function as | |
80 | * 'pointer to map element key' | |
81 | * | |
82 | * For example the argument constraints for bpf_map_lookup_elem(): | |
83 | * .ret_type = RET_PTR_TO_MAP_VALUE_OR_NULL, | |
84 | * .arg1_type = ARG_CONST_MAP_PTR, | |
85 | * .arg2_type = ARG_PTR_TO_MAP_KEY, | |
86 | * | |
87 | * ret_type says that this function returns 'pointer to map elem value or null' | |
88 | * function expects 1st argument to be a const pointer to 'struct bpf_map' and | |
89 | * 2nd argument should be a pointer to stack, which will be used inside | |
90 | * the helper function as a pointer to map element key. | |
91 | * | |
92 | * On the kernel side the helper function looks like: | |
93 | * u64 bpf_map_lookup_elem(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) | |
94 | * { | |
95 | * struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; | |
96 | * void *key = (void *) (unsigned long) r2; | |
97 | * void *value; | |
98 | * | |
99 | * here kernel can access 'key' and 'map' pointers safely, knowing that | |
100 | * [key, key + map->key_size) bytes are valid and were initialized on | |
101 | * the stack of eBPF program. | |
102 | * } | |
103 | * | |
104 | * Corresponding eBPF program may look like: | |
105 | * BPF_MOV64_REG(BPF_REG_2, BPF_REG_10), // after this insn R2 type is FRAME_PTR | |
106 | * BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, -4), // after this insn R2 type is PTR_TO_STACK | |
107 | * BPF_LD_MAP_FD(BPF_REG_1, map_fd), // after this insn R1 type is CONST_PTR_TO_MAP | |
108 | * BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_map_lookup_elem), | |
109 | * here verifier looks at prototype of map_lookup_elem() and sees: | |
110 | * .arg1_type == ARG_CONST_MAP_PTR and R1->type == CONST_PTR_TO_MAP, which is ok, | |
111 | * Now verifier knows that this map has key of R1->map_ptr->key_size bytes | |
112 | * | |
113 | * Then .arg2_type == ARG_PTR_TO_MAP_KEY and R2->type == PTR_TO_STACK, ok so far, | |
114 | * Now verifier checks that [R2, R2 + map's key_size) are within stack limits | |
115 | * and were initialized prior to this call. | |
116 | * If it's ok, then verifier allows this BPF_CALL insn and looks at | |
117 | * .ret_type which is RET_PTR_TO_MAP_VALUE_OR_NULL, so it sets | |
118 | * R0->type = PTR_TO_MAP_VALUE_OR_NULL which means bpf_map_lookup_elem() function | |
119 | * returns ether pointer to map value or NULL. | |
120 | * | |
121 | * When type PTR_TO_MAP_VALUE_OR_NULL passes through 'if (reg != 0) goto +off' | |
122 | * insn, the register holding that pointer in the true branch changes state to | |
123 | * PTR_TO_MAP_VALUE and the same register changes state to CONST_IMM in the false | |
124 | * branch. See check_cond_jmp_op(). | |
125 | * | |
126 | * After the call R0 is set to return type of the function and registers R1-R5 | |
127 | * are set to NOT_INIT to indicate that they are no longer readable. | |
128 | */ | |
129 | ||
17a52670 | 130 | /* verifier_state + insn_idx are pushed to stack when branch is encountered */ |
58e2af8b | 131 | struct bpf_verifier_stack_elem { |
17a52670 AS |
132 | /* verifer state is 'st' |
133 | * before processing instruction 'insn_idx' | |
134 | * and after processing instruction 'prev_insn_idx' | |
135 | */ | |
58e2af8b | 136 | struct bpf_verifier_state st; |
17a52670 AS |
137 | int insn_idx; |
138 | int prev_insn_idx; | |
58e2af8b | 139 | struct bpf_verifier_stack_elem *next; |
cbd35700 AS |
140 | }; |
141 | ||
07016151 DB |
142 | #define BPF_COMPLEXITY_LIMIT_INSNS 65536 |
143 | #define BPF_COMPLEXITY_LIMIT_STACK 1024 | |
144 | ||
33ff9823 DB |
145 | struct bpf_call_arg_meta { |
146 | struct bpf_map *map_ptr; | |
435faee1 | 147 | bool raw_mode; |
36bbef52 | 148 | bool pkt_access; |
435faee1 DB |
149 | int regno; |
150 | int access_size; | |
33ff9823 DB |
151 | }; |
152 | ||
cbd35700 AS |
153 | /* verbose verifier prints what it's seeing |
154 | * bpf_check() is called under lock, so no race to access these global vars | |
155 | */ | |
156 | static u32 log_level, log_size, log_len; | |
157 | static char *log_buf; | |
158 | ||
159 | static DEFINE_MUTEX(bpf_verifier_lock); | |
160 | ||
161 | /* log_level controls verbosity level of eBPF verifier. | |
162 | * verbose() is used to dump the verification trace to the log, so the user | |
163 | * can figure out what's wrong with the program | |
164 | */ | |
1d056d9c | 165 | static __printf(1, 2) void verbose(const char *fmt, ...) |
cbd35700 AS |
166 | { |
167 | va_list args; | |
168 | ||
169 | if (log_level == 0 || log_len >= log_size - 1) | |
170 | return; | |
171 | ||
172 | va_start(args, fmt); | |
173 | log_len += vscnprintf(log_buf + log_len, log_size - log_len, fmt, args); | |
174 | va_end(args); | |
175 | } | |
176 | ||
17a52670 AS |
177 | /* string representation of 'enum bpf_reg_type' */ |
178 | static const char * const reg_type_str[] = { | |
179 | [NOT_INIT] = "?", | |
180 | [UNKNOWN_VALUE] = "inv", | |
181 | [PTR_TO_CTX] = "ctx", | |
182 | [CONST_PTR_TO_MAP] = "map_ptr", | |
183 | [PTR_TO_MAP_VALUE] = "map_value", | |
184 | [PTR_TO_MAP_VALUE_OR_NULL] = "map_value_or_null", | |
48461135 | 185 | [PTR_TO_MAP_VALUE_ADJ] = "map_value_adj", |
17a52670 AS |
186 | [FRAME_PTR] = "fp", |
187 | [PTR_TO_STACK] = "fp", | |
188 | [CONST_IMM] = "imm", | |
969bf05e AS |
189 | [PTR_TO_PACKET] = "pkt", |
190 | [PTR_TO_PACKET_END] = "pkt_end", | |
17a52670 AS |
191 | }; |
192 | ||
58e2af8b | 193 | static void print_verifier_state(struct bpf_verifier_state *state) |
17a52670 | 194 | { |
58e2af8b | 195 | struct bpf_reg_state *reg; |
17a52670 AS |
196 | enum bpf_reg_type t; |
197 | int i; | |
198 | ||
199 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
200 | reg = &state->regs[i]; |
201 | t = reg->type; | |
17a52670 AS |
202 | if (t == NOT_INIT) |
203 | continue; | |
204 | verbose(" R%d=%s", i, reg_type_str[t]); | |
205 | if (t == CONST_IMM || t == PTR_TO_STACK) | |
969bf05e AS |
206 | verbose("%lld", reg->imm); |
207 | else if (t == PTR_TO_PACKET) | |
208 | verbose("(id=%d,off=%d,r=%d)", | |
209 | reg->id, reg->off, reg->range); | |
210 | else if (t == UNKNOWN_VALUE && reg->imm) | |
211 | verbose("%lld", reg->imm); | |
17a52670 | 212 | else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || |
48461135 JB |
213 | t == PTR_TO_MAP_VALUE_OR_NULL || |
214 | t == PTR_TO_MAP_VALUE_ADJ) | |
57a09bf0 | 215 | verbose("(ks=%d,vs=%d,id=%u)", |
1a0dc1ac | 216 | reg->map_ptr->key_size, |
57a09bf0 TG |
217 | reg->map_ptr->value_size, |
218 | reg->id); | |
48461135 JB |
219 | if (reg->min_value != BPF_REGISTER_MIN_RANGE) |
220 | verbose(",min_value=%llu", | |
221 | (unsigned long long)reg->min_value); | |
222 | if (reg->max_value != BPF_REGISTER_MAX_RANGE) | |
223 | verbose(",max_value=%llu", | |
224 | (unsigned long long)reg->max_value); | |
17a52670 | 225 | } |
9c399760 | 226 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { |
1a0dc1ac | 227 | if (state->stack_slot_type[i] == STACK_SPILL) |
17a52670 | 228 | verbose(" fp%d=%s", -MAX_BPF_STACK + i, |
1a0dc1ac | 229 | reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); |
17a52670 AS |
230 | } |
231 | verbose("\n"); | |
232 | } | |
233 | ||
cbd35700 AS |
234 | static const char *const bpf_class_string[] = { |
235 | [BPF_LD] = "ld", | |
236 | [BPF_LDX] = "ldx", | |
237 | [BPF_ST] = "st", | |
238 | [BPF_STX] = "stx", | |
239 | [BPF_ALU] = "alu", | |
240 | [BPF_JMP] = "jmp", | |
241 | [BPF_RET] = "BUG", | |
242 | [BPF_ALU64] = "alu64", | |
243 | }; | |
244 | ||
687f0715 | 245 | static const char *const bpf_alu_string[16] = { |
cbd35700 AS |
246 | [BPF_ADD >> 4] = "+=", |
247 | [BPF_SUB >> 4] = "-=", | |
248 | [BPF_MUL >> 4] = "*=", | |
249 | [BPF_DIV >> 4] = "/=", | |
250 | [BPF_OR >> 4] = "|=", | |
251 | [BPF_AND >> 4] = "&=", | |
252 | [BPF_LSH >> 4] = "<<=", | |
253 | [BPF_RSH >> 4] = ">>=", | |
254 | [BPF_NEG >> 4] = "neg", | |
255 | [BPF_MOD >> 4] = "%=", | |
256 | [BPF_XOR >> 4] = "^=", | |
257 | [BPF_MOV >> 4] = "=", | |
258 | [BPF_ARSH >> 4] = "s>>=", | |
259 | [BPF_END >> 4] = "endian", | |
260 | }; | |
261 | ||
262 | static const char *const bpf_ldst_string[] = { | |
263 | [BPF_W >> 3] = "u32", | |
264 | [BPF_H >> 3] = "u16", | |
265 | [BPF_B >> 3] = "u8", | |
266 | [BPF_DW >> 3] = "u64", | |
267 | }; | |
268 | ||
687f0715 | 269 | static const char *const bpf_jmp_string[16] = { |
cbd35700 AS |
270 | [BPF_JA >> 4] = "jmp", |
271 | [BPF_JEQ >> 4] = "==", | |
272 | [BPF_JGT >> 4] = ">", | |
273 | [BPF_JGE >> 4] = ">=", | |
274 | [BPF_JSET >> 4] = "&", | |
275 | [BPF_JNE >> 4] = "!=", | |
276 | [BPF_JSGT >> 4] = "s>", | |
277 | [BPF_JSGE >> 4] = "s>=", | |
278 | [BPF_CALL >> 4] = "call", | |
279 | [BPF_EXIT >> 4] = "exit", | |
280 | }; | |
281 | ||
282 | static void print_bpf_insn(struct bpf_insn *insn) | |
283 | { | |
284 | u8 class = BPF_CLASS(insn->code); | |
285 | ||
286 | if (class == BPF_ALU || class == BPF_ALU64) { | |
287 | if (BPF_SRC(insn->code) == BPF_X) | |
288 | verbose("(%02x) %sr%d %s %sr%d\n", | |
289 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
290 | insn->dst_reg, | |
291 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
292 | class == BPF_ALU ? "(u32) " : "", | |
293 | insn->src_reg); | |
294 | else | |
295 | verbose("(%02x) %sr%d %s %s%d\n", | |
296 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
297 | insn->dst_reg, | |
298 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
299 | class == BPF_ALU ? "(u32) " : "", | |
300 | insn->imm); | |
301 | } else if (class == BPF_STX) { | |
302 | if (BPF_MODE(insn->code) == BPF_MEM) | |
303 | verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", | |
304 | insn->code, | |
305 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
306 | insn->dst_reg, | |
307 | insn->off, insn->src_reg); | |
308 | else if (BPF_MODE(insn->code) == BPF_XADD) | |
309 | verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", | |
310 | insn->code, | |
311 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
312 | insn->dst_reg, insn->off, | |
313 | insn->src_reg); | |
314 | else | |
315 | verbose("BUG_%02x\n", insn->code); | |
316 | } else if (class == BPF_ST) { | |
317 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
318 | verbose("BUG_st_%02x\n", insn->code); | |
319 | return; | |
320 | } | |
321 | verbose("(%02x) *(%s *)(r%d %+d) = %d\n", | |
322 | insn->code, | |
323 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
324 | insn->dst_reg, | |
325 | insn->off, insn->imm); | |
326 | } else if (class == BPF_LDX) { | |
327 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
328 | verbose("BUG_ldx_%02x\n", insn->code); | |
329 | return; | |
330 | } | |
331 | verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", | |
332 | insn->code, insn->dst_reg, | |
333 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
334 | insn->src_reg, insn->off); | |
335 | } else if (class == BPF_LD) { | |
336 | if (BPF_MODE(insn->code) == BPF_ABS) { | |
337 | verbose("(%02x) r0 = *(%s *)skb[%d]\n", | |
338 | insn->code, | |
339 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
340 | insn->imm); | |
341 | } else if (BPF_MODE(insn->code) == BPF_IND) { | |
342 | verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", | |
343 | insn->code, | |
344 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
345 | insn->src_reg, insn->imm); | |
346 | } else if (BPF_MODE(insn->code) == BPF_IMM) { | |
347 | verbose("(%02x) r%d = 0x%x\n", | |
348 | insn->code, insn->dst_reg, insn->imm); | |
349 | } else { | |
350 | verbose("BUG_ld_%02x\n", insn->code); | |
351 | return; | |
352 | } | |
353 | } else if (class == BPF_JMP) { | |
354 | u8 opcode = BPF_OP(insn->code); | |
355 | ||
356 | if (opcode == BPF_CALL) { | |
357 | verbose("(%02x) call %d\n", insn->code, insn->imm); | |
358 | } else if (insn->code == (BPF_JMP | BPF_JA)) { | |
359 | verbose("(%02x) goto pc%+d\n", | |
360 | insn->code, insn->off); | |
361 | } else if (insn->code == (BPF_JMP | BPF_EXIT)) { | |
362 | verbose("(%02x) exit\n", insn->code); | |
363 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
364 | verbose("(%02x) if r%d %s r%d goto pc%+d\n", | |
365 | insn->code, insn->dst_reg, | |
366 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
367 | insn->src_reg, insn->off); | |
368 | } else { | |
369 | verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", | |
370 | insn->code, insn->dst_reg, | |
371 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
372 | insn->imm, insn->off); | |
373 | } | |
374 | } else { | |
375 | verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); | |
376 | } | |
377 | } | |
378 | ||
58e2af8b | 379 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) |
17a52670 | 380 | { |
58e2af8b | 381 | struct bpf_verifier_stack_elem *elem; |
17a52670 AS |
382 | int insn_idx; |
383 | ||
384 | if (env->head == NULL) | |
385 | return -1; | |
386 | ||
387 | memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); | |
388 | insn_idx = env->head->insn_idx; | |
389 | if (prev_insn_idx) | |
390 | *prev_insn_idx = env->head->prev_insn_idx; | |
391 | elem = env->head->next; | |
392 | kfree(env->head); | |
393 | env->head = elem; | |
394 | env->stack_size--; | |
395 | return insn_idx; | |
396 | } | |
397 | ||
58e2af8b JK |
398 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
399 | int insn_idx, int prev_insn_idx) | |
17a52670 | 400 | { |
58e2af8b | 401 | struct bpf_verifier_stack_elem *elem; |
17a52670 | 402 | |
58e2af8b | 403 | elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
404 | if (!elem) |
405 | goto err; | |
406 | ||
407 | memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); | |
408 | elem->insn_idx = insn_idx; | |
409 | elem->prev_insn_idx = prev_insn_idx; | |
410 | elem->next = env->head; | |
411 | env->head = elem; | |
412 | env->stack_size++; | |
07016151 | 413 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
17a52670 AS |
414 | verbose("BPF program is too complex\n"); |
415 | goto err; | |
416 | } | |
417 | return &elem->st; | |
418 | err: | |
419 | /* pop all elements and return */ | |
420 | while (pop_stack(env, NULL) >= 0); | |
421 | return NULL; | |
422 | } | |
423 | ||
424 | #define CALLER_SAVED_REGS 6 | |
425 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
426 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
427 | }; | |
428 | ||
58e2af8b | 429 | static void init_reg_state(struct bpf_reg_state *regs) |
17a52670 AS |
430 | { |
431 | int i; | |
432 | ||
433 | for (i = 0; i < MAX_BPF_REG; i++) { | |
434 | regs[i].type = NOT_INIT; | |
435 | regs[i].imm = 0; | |
48461135 JB |
436 | regs[i].min_value = BPF_REGISTER_MIN_RANGE; |
437 | regs[i].max_value = BPF_REGISTER_MAX_RANGE; | |
17a52670 AS |
438 | } |
439 | ||
440 | /* frame pointer */ | |
441 | regs[BPF_REG_FP].type = FRAME_PTR; | |
442 | ||
443 | /* 1st arg to a function */ | |
444 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
445 | } | |
446 | ||
58e2af8b | 447 | static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) |
17a52670 AS |
448 | { |
449 | BUG_ON(regno >= MAX_BPF_REG); | |
450 | regs[regno].type = UNKNOWN_VALUE; | |
57a09bf0 | 451 | regs[regno].id = 0; |
17a52670 | 452 | regs[regno].imm = 0; |
17a52670 AS |
453 | } |
454 | ||
48461135 JB |
455 | static void reset_reg_range_values(struct bpf_reg_state *regs, u32 regno) |
456 | { | |
457 | regs[regno].min_value = BPF_REGISTER_MIN_RANGE; | |
458 | regs[regno].max_value = BPF_REGISTER_MAX_RANGE; | |
459 | } | |
460 | ||
17a52670 AS |
461 | enum reg_arg_type { |
462 | SRC_OP, /* register is used as source operand */ | |
463 | DST_OP, /* register is used as destination operand */ | |
464 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
465 | }; | |
466 | ||
58e2af8b | 467 | static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, |
17a52670 AS |
468 | enum reg_arg_type t) |
469 | { | |
470 | if (regno >= MAX_BPF_REG) { | |
471 | verbose("R%d is invalid\n", regno); | |
472 | return -EINVAL; | |
473 | } | |
474 | ||
475 | if (t == SRC_OP) { | |
476 | /* check whether register used as source operand can be read */ | |
477 | if (regs[regno].type == NOT_INIT) { | |
478 | verbose("R%d !read_ok\n", regno); | |
479 | return -EACCES; | |
480 | } | |
481 | } else { | |
482 | /* check whether register used as dest operand can be written to */ | |
483 | if (regno == BPF_REG_FP) { | |
484 | verbose("frame pointer is read only\n"); | |
485 | return -EACCES; | |
486 | } | |
487 | if (t == DST_OP) | |
488 | mark_reg_unknown_value(regs, regno); | |
489 | } | |
490 | return 0; | |
491 | } | |
492 | ||
493 | static int bpf_size_to_bytes(int bpf_size) | |
494 | { | |
495 | if (bpf_size == BPF_W) | |
496 | return 4; | |
497 | else if (bpf_size == BPF_H) | |
498 | return 2; | |
499 | else if (bpf_size == BPF_B) | |
500 | return 1; | |
501 | else if (bpf_size == BPF_DW) | |
502 | return 8; | |
503 | else | |
504 | return -EINVAL; | |
505 | } | |
506 | ||
1be7f75d AS |
507 | static bool is_spillable_regtype(enum bpf_reg_type type) |
508 | { | |
509 | switch (type) { | |
510 | case PTR_TO_MAP_VALUE: | |
511 | case PTR_TO_MAP_VALUE_OR_NULL: | |
512 | case PTR_TO_STACK: | |
513 | case PTR_TO_CTX: | |
969bf05e AS |
514 | case PTR_TO_PACKET: |
515 | case PTR_TO_PACKET_END: | |
1be7f75d AS |
516 | case FRAME_PTR: |
517 | case CONST_PTR_TO_MAP: | |
518 | return true; | |
519 | default: | |
520 | return false; | |
521 | } | |
522 | } | |
523 | ||
17a52670 AS |
524 | /* check_stack_read/write functions track spill/fill of registers, |
525 | * stack boundary and alignment are checked in check_mem_access() | |
526 | */ | |
58e2af8b JK |
527 | static int check_stack_write(struct bpf_verifier_state *state, int off, |
528 | int size, int value_regno) | |
17a52670 | 529 | { |
17a52670 | 530 | int i; |
9c399760 AS |
531 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
532 | * so it's aligned access and [off, off + size) are within stack limits | |
533 | */ | |
17a52670 AS |
534 | |
535 | if (value_regno >= 0 && | |
1be7f75d | 536 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
537 | |
538 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 539 | if (size != BPF_REG_SIZE) { |
17a52670 AS |
540 | verbose("invalid size of register spill\n"); |
541 | return -EACCES; | |
542 | } | |
543 | ||
17a52670 | 544 | /* save register state */ |
9c399760 AS |
545 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
546 | state->regs[value_regno]; | |
17a52670 | 547 | |
9c399760 AS |
548 | for (i = 0; i < BPF_REG_SIZE; i++) |
549 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; | |
550 | } else { | |
17a52670 | 551 | /* regular write of data into stack */ |
9c399760 | 552 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
58e2af8b | 553 | (struct bpf_reg_state) {}; |
9c399760 AS |
554 | |
555 | for (i = 0; i < size; i++) | |
556 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; | |
17a52670 AS |
557 | } |
558 | return 0; | |
559 | } | |
560 | ||
58e2af8b | 561 | static int check_stack_read(struct bpf_verifier_state *state, int off, int size, |
17a52670 AS |
562 | int value_regno) |
563 | { | |
9c399760 | 564 | u8 *slot_type; |
17a52670 | 565 | int i; |
17a52670 | 566 | |
9c399760 | 567 | slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; |
17a52670 | 568 | |
9c399760 AS |
569 | if (slot_type[0] == STACK_SPILL) { |
570 | if (size != BPF_REG_SIZE) { | |
17a52670 AS |
571 | verbose("invalid size of register spill\n"); |
572 | return -EACCES; | |
573 | } | |
9c399760 AS |
574 | for (i = 1; i < BPF_REG_SIZE; i++) { |
575 | if (slot_type[i] != STACK_SPILL) { | |
17a52670 AS |
576 | verbose("corrupted spill memory\n"); |
577 | return -EACCES; | |
578 | } | |
579 | } | |
580 | ||
581 | if (value_regno >= 0) | |
582 | /* restore register state from stack */ | |
9c399760 AS |
583 | state->regs[value_regno] = |
584 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; | |
17a52670 AS |
585 | return 0; |
586 | } else { | |
587 | for (i = 0; i < size; i++) { | |
9c399760 | 588 | if (slot_type[i] != STACK_MISC) { |
17a52670 AS |
589 | verbose("invalid read from stack off %d+%d size %d\n", |
590 | off, i, size); | |
591 | return -EACCES; | |
592 | } | |
593 | } | |
594 | if (value_regno >= 0) | |
595 | /* have read misc data from the stack */ | |
596 | mark_reg_unknown_value(state->regs, value_regno); | |
597 | return 0; | |
598 | } | |
599 | } | |
600 | ||
601 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
58e2af8b | 602 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
603 | int size) |
604 | { | |
605 | struct bpf_map *map = env->cur_state.regs[regno].map_ptr; | |
606 | ||
607 | if (off < 0 || off + size > map->value_size) { | |
608 | verbose("invalid access to map value, value_size=%d off=%d size=%d\n", | |
609 | map->value_size, off, size); | |
610 | return -EACCES; | |
611 | } | |
612 | return 0; | |
613 | } | |
614 | ||
969bf05e AS |
615 | #define MAX_PACKET_OFF 0xffff |
616 | ||
58e2af8b | 617 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
36bbef52 | 618 | const struct bpf_call_arg_meta *meta) |
4acf6c0b | 619 | { |
36bbef52 DB |
620 | switch (env->prog->type) { |
621 | case BPF_PROG_TYPE_SCHED_CLS: | |
622 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 623 | case BPF_PROG_TYPE_XDP: |
36bbef52 DB |
624 | if (meta) |
625 | return meta->pkt_access; | |
626 | ||
627 | env->seen_direct_write = true; | |
4acf6c0b BB |
628 | return true; |
629 | default: | |
630 | return false; | |
631 | } | |
632 | } | |
633 | ||
58e2af8b | 634 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
969bf05e AS |
635 | int size) |
636 | { | |
58e2af8b JK |
637 | struct bpf_reg_state *regs = env->cur_state.regs; |
638 | struct bpf_reg_state *reg = ®s[regno]; | |
969bf05e | 639 | |
d91b28ed | 640 | off += reg->off; |
b399cf64 | 641 | if (off < 0 || size <= 0 || off + size > reg->range) { |
d91b28ed AS |
642 | verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
643 | off, size, regno, reg->id, reg->off, reg->range); | |
969bf05e AS |
644 | return -EACCES; |
645 | } | |
646 | return 0; | |
647 | } | |
648 | ||
17a52670 | 649 | /* check access to 'struct bpf_context' fields */ |
58e2af8b | 650 | static int check_ctx_access(struct bpf_verifier_env *env, int off, int size, |
19de99f7 | 651 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 | 652 | { |
13a27dfc JK |
653 | /* for analyzer ctx accesses are already validated and converted */ |
654 | if (env->analyzer_ops) | |
655 | return 0; | |
656 | ||
17a52670 | 657 | if (env->prog->aux->ops->is_valid_access && |
19de99f7 | 658 | env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) { |
32bbe007 AS |
659 | /* remember the offset of last byte accessed in ctx */ |
660 | if (env->prog->aux->max_ctx_offset < off + size) | |
661 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 662 | return 0; |
32bbe007 | 663 | } |
17a52670 AS |
664 | |
665 | verbose("invalid bpf_context access off=%d size=%d\n", off, size); | |
666 | return -EACCES; | |
667 | } | |
668 | ||
58e2af8b | 669 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1be7f75d AS |
670 | { |
671 | if (env->allow_ptr_leaks) | |
672 | return false; | |
673 | ||
674 | switch (env->cur_state.regs[regno].type) { | |
675 | case UNKNOWN_VALUE: | |
676 | case CONST_IMM: | |
677 | return false; | |
678 | default: | |
679 | return true; | |
680 | } | |
681 | } | |
682 | ||
58e2af8b JK |
683 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
684 | struct bpf_reg_state *reg, int off, int size) | |
969bf05e | 685 | { |
48461135 | 686 | if (reg->type != PTR_TO_PACKET && reg->type != PTR_TO_MAP_VALUE_ADJ) { |
969bf05e | 687 | if (off % size != 0) { |
58e2af8b JK |
688 | verbose("misaligned access off %d size %d\n", |
689 | off, size); | |
969bf05e AS |
690 | return -EACCES; |
691 | } else { | |
692 | return 0; | |
693 | } | |
694 | } | |
695 | ||
969bf05e AS |
696 | if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) |
697 | /* misaligned access to packet is ok on x86,arm,arm64 */ | |
698 | return 0; | |
699 | ||
700 | if (reg->id && size != 1) { | |
701 | verbose("Unknown packet alignment. Only byte-sized access allowed\n"); | |
702 | return -EACCES; | |
703 | } | |
704 | ||
705 | /* skb->data is NET_IP_ALIGN-ed */ | |
48461135 JB |
706 | if (reg->type == PTR_TO_PACKET && |
707 | (NET_IP_ALIGN + reg->off + off) % size != 0) { | |
969bf05e AS |
708 | verbose("misaligned packet access off %d+%d+%d size %d\n", |
709 | NET_IP_ALIGN, reg->off, off, size); | |
710 | return -EACCES; | |
711 | } | |
712 | return 0; | |
713 | } | |
714 | ||
17a52670 AS |
715 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
716 | * if t==write, value_regno is a register which value is stored into memory | |
717 | * if t==read, value_regno is a register which will receive the value from memory | |
718 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
719 | * if t==read && value_regno==-1, don't care what we read from memory | |
720 | */ | |
58e2af8b | 721 | static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
722 | int bpf_size, enum bpf_access_type t, |
723 | int value_regno) | |
724 | { | |
58e2af8b JK |
725 | struct bpf_verifier_state *state = &env->cur_state; |
726 | struct bpf_reg_state *reg = &state->regs[regno]; | |
17a52670 AS |
727 | int size, err = 0; |
728 | ||
1a0dc1ac AS |
729 | if (reg->type == PTR_TO_STACK) |
730 | off += reg->imm; | |
24b4d2ab | 731 | |
17a52670 AS |
732 | size = bpf_size_to_bytes(bpf_size); |
733 | if (size < 0) | |
734 | return size; | |
735 | ||
969bf05e AS |
736 | err = check_ptr_alignment(env, reg, off, size); |
737 | if (err) | |
738 | return err; | |
17a52670 | 739 | |
48461135 JB |
740 | if (reg->type == PTR_TO_MAP_VALUE || |
741 | reg->type == PTR_TO_MAP_VALUE_ADJ) { | |
1be7f75d AS |
742 | if (t == BPF_WRITE && value_regno >= 0 && |
743 | is_pointer_value(env, value_regno)) { | |
744 | verbose("R%d leaks addr into map\n", value_regno); | |
745 | return -EACCES; | |
746 | } | |
48461135 JB |
747 | |
748 | /* If we adjusted the register to this map value at all then we | |
749 | * need to change off and size to min_value and max_value | |
750 | * respectively to make sure our theoretical access will be | |
751 | * safe. | |
752 | */ | |
753 | if (reg->type == PTR_TO_MAP_VALUE_ADJ) { | |
754 | if (log_level) | |
755 | print_verifier_state(state); | |
756 | env->varlen_map_value_access = true; | |
757 | /* The minimum value is only important with signed | |
758 | * comparisons where we can't assume the floor of a | |
759 | * value is 0. If we are using signed variables for our | |
760 | * index'es we need to make sure that whatever we use | |
761 | * will have a set floor within our range. | |
762 | */ | |
763 | if ((s64)reg->min_value < 0) { | |
764 | verbose("R%d min value is negative, either use unsigned index or do a if (index >=0) check.\n", | |
765 | regno); | |
766 | return -EACCES; | |
767 | } | |
768 | err = check_map_access(env, regno, reg->min_value + off, | |
769 | size); | |
770 | if (err) { | |
771 | verbose("R%d min value is outside of the array range\n", | |
772 | regno); | |
773 | return err; | |
774 | } | |
775 | ||
776 | /* If we haven't set a max value then we need to bail | |
777 | * since we can't be sure we won't do bad things. | |
778 | */ | |
779 | if (reg->max_value == BPF_REGISTER_MAX_RANGE) { | |
780 | verbose("R%d unbounded memory access, make sure to bounds check any array access into a map\n", | |
781 | regno); | |
782 | return -EACCES; | |
783 | } | |
784 | off += reg->max_value; | |
785 | } | |
17a52670 AS |
786 | err = check_map_access(env, regno, off, size); |
787 | if (!err && t == BPF_READ && value_regno >= 0) | |
788 | mark_reg_unknown_value(state->regs, value_regno); | |
789 | ||
1a0dc1ac | 790 | } else if (reg->type == PTR_TO_CTX) { |
19de99f7 AS |
791 | enum bpf_reg_type reg_type = UNKNOWN_VALUE; |
792 | ||
1be7f75d AS |
793 | if (t == BPF_WRITE && value_regno >= 0 && |
794 | is_pointer_value(env, value_regno)) { | |
795 | verbose("R%d leaks addr into ctx\n", value_regno); | |
796 | return -EACCES; | |
797 | } | |
19de99f7 | 798 | err = check_ctx_access(env, off, size, t, ®_type); |
969bf05e | 799 | if (!err && t == BPF_READ && value_regno >= 0) { |
17a52670 | 800 | mark_reg_unknown_value(state->regs, value_regno); |
1955351d MS |
801 | /* note that reg.[id|off|range] == 0 */ |
802 | state->regs[value_regno].type = reg_type; | |
969bf05e | 803 | } |
17a52670 | 804 | |
1a0dc1ac | 805 | } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { |
17a52670 AS |
806 | if (off >= 0 || off < -MAX_BPF_STACK) { |
807 | verbose("invalid stack off=%d size=%d\n", off, size); | |
808 | return -EACCES; | |
809 | } | |
1be7f75d AS |
810 | if (t == BPF_WRITE) { |
811 | if (!env->allow_ptr_leaks && | |
812 | state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && | |
813 | size != BPF_REG_SIZE) { | |
814 | verbose("attempt to corrupt spilled pointer on stack\n"); | |
815 | return -EACCES; | |
816 | } | |
17a52670 | 817 | err = check_stack_write(state, off, size, value_regno); |
1be7f75d | 818 | } else { |
17a52670 | 819 | err = check_stack_read(state, off, size, value_regno); |
1be7f75d | 820 | } |
969bf05e | 821 | } else if (state->regs[regno].type == PTR_TO_PACKET) { |
36bbef52 | 822 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL)) { |
969bf05e AS |
823 | verbose("cannot write into packet\n"); |
824 | return -EACCES; | |
825 | } | |
4acf6c0b BB |
826 | if (t == BPF_WRITE && value_regno >= 0 && |
827 | is_pointer_value(env, value_regno)) { | |
828 | verbose("R%d leaks addr into packet\n", value_regno); | |
829 | return -EACCES; | |
830 | } | |
969bf05e AS |
831 | err = check_packet_access(env, regno, off, size); |
832 | if (!err && t == BPF_READ && value_regno >= 0) | |
833 | mark_reg_unknown_value(state->regs, value_regno); | |
17a52670 AS |
834 | } else { |
835 | verbose("R%d invalid mem access '%s'\n", | |
1a0dc1ac | 836 | regno, reg_type_str[reg->type]); |
17a52670 AS |
837 | return -EACCES; |
838 | } | |
969bf05e AS |
839 | |
840 | if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && | |
841 | state->regs[value_regno].type == UNKNOWN_VALUE) { | |
842 | /* 1 or 2 byte load zero-extends, determine the number of | |
843 | * zero upper bits. Not doing it fo 4 byte load, since | |
844 | * such values cannot be added to ptr_to_packet anyway. | |
845 | */ | |
846 | state->regs[value_regno].imm = 64 - size * 8; | |
847 | } | |
17a52670 AS |
848 | return err; |
849 | } | |
850 | ||
58e2af8b | 851 | static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 852 | { |
58e2af8b | 853 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
854 | int err; |
855 | ||
856 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
857 | insn->imm != 0) { | |
858 | verbose("BPF_XADD uses reserved fields\n"); | |
859 | return -EINVAL; | |
860 | } | |
861 | ||
862 | /* check src1 operand */ | |
863 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
864 | if (err) | |
865 | return err; | |
866 | ||
867 | /* check src2 operand */ | |
868 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
869 | if (err) | |
870 | return err; | |
871 | ||
872 | /* check whether atomic_add can read the memory */ | |
873 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
874 | BPF_SIZE(insn->code), BPF_READ, -1); | |
875 | if (err) | |
876 | return err; | |
877 | ||
878 | /* check whether atomic_add can write into the same memory */ | |
879 | return check_mem_access(env, insn->dst_reg, insn->off, | |
880 | BPF_SIZE(insn->code), BPF_WRITE, -1); | |
881 | } | |
882 | ||
883 | /* when register 'regno' is passed into function that will read 'access_size' | |
884 | * bytes from that pointer, make sure that it's within stack boundary | |
885 | * and all elements of stack are initialized | |
886 | */ | |
58e2af8b | 887 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
888 | int access_size, bool zero_size_allowed, |
889 | struct bpf_call_arg_meta *meta) | |
17a52670 | 890 | { |
58e2af8b JK |
891 | struct bpf_verifier_state *state = &env->cur_state; |
892 | struct bpf_reg_state *regs = state->regs; | |
17a52670 AS |
893 | int off, i; |
894 | ||
8e2fe1d9 DB |
895 | if (regs[regno].type != PTR_TO_STACK) { |
896 | if (zero_size_allowed && access_size == 0 && | |
897 | regs[regno].type == CONST_IMM && | |
898 | regs[regno].imm == 0) | |
899 | return 0; | |
900 | ||
901 | verbose("R%d type=%s expected=%s\n", regno, | |
902 | reg_type_str[regs[regno].type], | |
903 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 904 | return -EACCES; |
8e2fe1d9 | 905 | } |
17a52670 AS |
906 | |
907 | off = regs[regno].imm; | |
908 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || | |
909 | access_size <= 0) { | |
910 | verbose("invalid stack type R%d off=%d access_size=%d\n", | |
911 | regno, off, access_size); | |
912 | return -EACCES; | |
913 | } | |
914 | ||
435faee1 DB |
915 | if (meta && meta->raw_mode) { |
916 | meta->access_size = access_size; | |
917 | meta->regno = regno; | |
918 | return 0; | |
919 | } | |
920 | ||
17a52670 | 921 | for (i = 0; i < access_size; i++) { |
9c399760 | 922 | if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { |
17a52670 AS |
923 | verbose("invalid indirect read from stack off %d+%d size %d\n", |
924 | off, i, access_size); | |
925 | return -EACCES; | |
926 | } | |
927 | } | |
928 | return 0; | |
929 | } | |
930 | ||
58e2af8b | 931 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
932 | enum bpf_arg_type arg_type, |
933 | struct bpf_call_arg_meta *meta) | |
17a52670 | 934 | { |
58e2af8b | 935 | struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; |
6841de8b | 936 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
937 | int err = 0; |
938 | ||
80f1d68c | 939 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
940 | return 0; |
941 | ||
6841de8b | 942 | if (type == NOT_INIT) { |
17a52670 AS |
943 | verbose("R%d !read_ok\n", regno); |
944 | return -EACCES; | |
945 | } | |
946 | ||
1be7f75d AS |
947 | if (arg_type == ARG_ANYTHING) { |
948 | if (is_pointer_value(env, regno)) { | |
949 | verbose("R%d leaks addr into helper function\n", regno); | |
950 | return -EACCES; | |
951 | } | |
80f1d68c | 952 | return 0; |
1be7f75d | 953 | } |
80f1d68c | 954 | |
36bbef52 DB |
955 | if (type == PTR_TO_PACKET && !may_access_direct_pkt_data(env, meta)) { |
956 | verbose("helper access to the packet is not allowed\n"); | |
6841de8b AS |
957 | return -EACCES; |
958 | } | |
959 | ||
8e2fe1d9 | 960 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
961 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
962 | expected_type = PTR_TO_STACK; | |
6841de8b AS |
963 | if (type != PTR_TO_PACKET && type != expected_type) |
964 | goto err_type; | |
8e2fe1d9 DB |
965 | } else if (arg_type == ARG_CONST_STACK_SIZE || |
966 | arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { | |
17a52670 | 967 | expected_type = CONST_IMM; |
6841de8b AS |
968 | if (type != expected_type) |
969 | goto err_type; | |
17a52670 AS |
970 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
971 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
972 | if (type != expected_type) |
973 | goto err_type; | |
608cd71a AS |
974 | } else if (arg_type == ARG_PTR_TO_CTX) { |
975 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
976 | if (type != expected_type) |
977 | goto err_type; | |
435faee1 DB |
978 | } else if (arg_type == ARG_PTR_TO_STACK || |
979 | arg_type == ARG_PTR_TO_RAW_STACK) { | |
8e2fe1d9 DB |
980 | expected_type = PTR_TO_STACK; |
981 | /* One exception here. In case function allows for NULL to be | |
982 | * passed in as argument, it's a CONST_IMM type. Final test | |
983 | * happens during stack boundary checking. | |
984 | */ | |
6841de8b AS |
985 | if (type == CONST_IMM && reg->imm == 0) |
986 | /* final test in check_stack_boundary() */; | |
987 | else if (type != PTR_TO_PACKET && type != expected_type) | |
988 | goto err_type; | |
435faee1 | 989 | meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK; |
17a52670 AS |
990 | } else { |
991 | verbose("unsupported arg_type %d\n", arg_type); | |
992 | return -EFAULT; | |
993 | } | |
994 | ||
17a52670 AS |
995 | if (arg_type == ARG_CONST_MAP_PTR) { |
996 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 997 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
998 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
999 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
1000 | * check that [key, key + map->key_size) are within | |
1001 | * stack limits and initialized | |
1002 | */ | |
33ff9823 | 1003 | if (!meta->map_ptr) { |
17a52670 AS |
1004 | /* in function declaration map_ptr must come before |
1005 | * map_key, so that it's verified and known before | |
1006 | * we have to check map_key here. Otherwise it means | |
1007 | * that kernel subsystem misconfigured verifier | |
1008 | */ | |
1009 | verbose("invalid map_ptr to access map->key\n"); | |
1010 | return -EACCES; | |
1011 | } | |
6841de8b AS |
1012 | if (type == PTR_TO_PACKET) |
1013 | err = check_packet_access(env, regno, 0, | |
1014 | meta->map_ptr->key_size); | |
1015 | else | |
1016 | err = check_stack_boundary(env, regno, | |
1017 | meta->map_ptr->key_size, | |
1018 | false, NULL); | |
17a52670 AS |
1019 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
1020 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
1021 | * check [value, value + map->value_size) validity | |
1022 | */ | |
33ff9823 | 1023 | if (!meta->map_ptr) { |
17a52670 AS |
1024 | /* kernel subsystem misconfigured verifier */ |
1025 | verbose("invalid map_ptr to access map->value\n"); | |
1026 | return -EACCES; | |
1027 | } | |
6841de8b AS |
1028 | if (type == PTR_TO_PACKET) |
1029 | err = check_packet_access(env, regno, 0, | |
1030 | meta->map_ptr->value_size); | |
1031 | else | |
1032 | err = check_stack_boundary(env, regno, | |
1033 | meta->map_ptr->value_size, | |
1034 | false, NULL); | |
8e2fe1d9 DB |
1035 | } else if (arg_type == ARG_CONST_STACK_SIZE || |
1036 | arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { | |
1037 | bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO); | |
17a52670 | 1038 | |
17a52670 AS |
1039 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
1040 | * from stack pointer 'buf'. Check it | |
1041 | * note: regno == len, regno - 1 == buf | |
1042 | */ | |
1043 | if (regno == 0) { | |
1044 | /* kernel subsystem misconfigured verifier */ | |
1045 | verbose("ARG_CONST_STACK_SIZE cannot be first argument\n"); | |
1046 | return -EACCES; | |
1047 | } | |
6841de8b AS |
1048 | if (regs[regno - 1].type == PTR_TO_PACKET) |
1049 | err = check_packet_access(env, regno - 1, 0, reg->imm); | |
1050 | else | |
1051 | err = check_stack_boundary(env, regno - 1, reg->imm, | |
1052 | zero_size_allowed, meta); | |
17a52670 AS |
1053 | } |
1054 | ||
1055 | return err; | |
6841de8b AS |
1056 | err_type: |
1057 | verbose("R%d type=%s expected=%s\n", regno, | |
1058 | reg_type_str[type], reg_type_str[expected_type]); | |
1059 | return -EACCES; | |
17a52670 AS |
1060 | } |
1061 | ||
35578d79 KX |
1062 | static int check_map_func_compatibility(struct bpf_map *map, int func_id) |
1063 | { | |
35578d79 KX |
1064 | if (!map) |
1065 | return 0; | |
1066 | ||
6aff67c8 AS |
1067 | /* We need a two way check, first is from map perspective ... */ |
1068 | switch (map->map_type) { | |
1069 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1070 | if (func_id != BPF_FUNC_tail_call) | |
1071 | goto error; | |
1072 | break; | |
1073 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1074 | if (func_id != BPF_FUNC_perf_event_read && | |
1075 | func_id != BPF_FUNC_perf_event_output) | |
1076 | goto error; | |
1077 | break; | |
1078 | case BPF_MAP_TYPE_STACK_TRACE: | |
1079 | if (func_id != BPF_FUNC_get_stackid) | |
1080 | goto error; | |
1081 | break; | |
4ed8ec52 | 1082 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1083 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1084 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1085 | goto error; |
1086 | break; | |
6aff67c8 AS |
1087 | default: |
1088 | break; | |
1089 | } | |
1090 | ||
1091 | /* ... and second from the function itself. */ | |
1092 | switch (func_id) { | |
1093 | case BPF_FUNC_tail_call: | |
1094 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1095 | goto error; | |
1096 | break; | |
1097 | case BPF_FUNC_perf_event_read: | |
1098 | case BPF_FUNC_perf_event_output: | |
1099 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) | |
1100 | goto error; | |
1101 | break; | |
1102 | case BPF_FUNC_get_stackid: | |
1103 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1104 | goto error; | |
1105 | break; | |
60d20f91 | 1106 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1107 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1108 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1109 | goto error; | |
1110 | break; | |
6aff67c8 AS |
1111 | default: |
1112 | break; | |
35578d79 KX |
1113 | } |
1114 | ||
1115 | return 0; | |
6aff67c8 AS |
1116 | error: |
1117 | verbose("cannot pass map_type %d into func %d\n", | |
1118 | map->map_type, func_id); | |
1119 | return -EINVAL; | |
35578d79 KX |
1120 | } |
1121 | ||
435faee1 DB |
1122 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1123 | { | |
1124 | int count = 0; | |
1125 | ||
1126 | if (fn->arg1_type == ARG_PTR_TO_RAW_STACK) | |
1127 | count++; | |
1128 | if (fn->arg2_type == ARG_PTR_TO_RAW_STACK) | |
1129 | count++; | |
1130 | if (fn->arg3_type == ARG_PTR_TO_RAW_STACK) | |
1131 | count++; | |
1132 | if (fn->arg4_type == ARG_PTR_TO_RAW_STACK) | |
1133 | count++; | |
1134 | if (fn->arg5_type == ARG_PTR_TO_RAW_STACK) | |
1135 | count++; | |
1136 | ||
1137 | return count > 1 ? -EINVAL : 0; | |
1138 | } | |
1139 | ||
58e2af8b | 1140 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1141 | { |
58e2af8b JK |
1142 | struct bpf_verifier_state *state = &env->cur_state; |
1143 | struct bpf_reg_state *regs = state->regs, *reg; | |
969bf05e AS |
1144 | int i; |
1145 | ||
1146 | for (i = 0; i < MAX_BPF_REG; i++) | |
1147 | if (regs[i].type == PTR_TO_PACKET || | |
1148 | regs[i].type == PTR_TO_PACKET_END) | |
1149 | mark_reg_unknown_value(regs, i); | |
1150 | ||
1151 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1152 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1153 | continue; | |
1154 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1155 | if (reg->type != PTR_TO_PACKET && | |
1156 | reg->type != PTR_TO_PACKET_END) | |
1157 | continue; | |
1158 | reg->type = UNKNOWN_VALUE; | |
1159 | reg->imm = 0; | |
1160 | } | |
1161 | } | |
1162 | ||
58e2af8b | 1163 | static int check_call(struct bpf_verifier_env *env, int func_id) |
17a52670 | 1164 | { |
58e2af8b | 1165 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 1166 | const struct bpf_func_proto *fn = NULL; |
58e2af8b JK |
1167 | struct bpf_reg_state *regs = state->regs; |
1168 | struct bpf_reg_state *reg; | |
33ff9823 | 1169 | struct bpf_call_arg_meta meta; |
969bf05e | 1170 | bool changes_data; |
17a52670 AS |
1171 | int i, err; |
1172 | ||
1173 | /* find function prototype */ | |
1174 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
1175 | verbose("invalid func %d\n", func_id); | |
1176 | return -EINVAL; | |
1177 | } | |
1178 | ||
1179 | if (env->prog->aux->ops->get_func_proto) | |
1180 | fn = env->prog->aux->ops->get_func_proto(func_id); | |
1181 | ||
1182 | if (!fn) { | |
1183 | verbose("unknown func %d\n", func_id); | |
1184 | return -EINVAL; | |
1185 | } | |
1186 | ||
1187 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1188 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
17a52670 AS |
1189 | verbose("cannot call GPL only function from proprietary program\n"); |
1190 | return -EINVAL; | |
1191 | } | |
1192 | ||
969bf05e AS |
1193 | changes_data = bpf_helper_changes_skb_data(fn->func); |
1194 | ||
33ff9823 | 1195 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1196 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1197 | |
435faee1 DB |
1198 | /* We only support one arg being in raw mode at the moment, which |
1199 | * is sufficient for the helper functions we have right now. | |
1200 | */ | |
1201 | err = check_raw_mode(fn); | |
1202 | if (err) { | |
1203 | verbose("kernel subsystem misconfigured func %d\n", func_id); | |
1204 | return err; | |
1205 | } | |
1206 | ||
17a52670 | 1207 | /* check args */ |
33ff9823 | 1208 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1209 | if (err) |
1210 | return err; | |
33ff9823 | 1211 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1212 | if (err) |
1213 | return err; | |
33ff9823 | 1214 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1215 | if (err) |
1216 | return err; | |
33ff9823 | 1217 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1218 | if (err) |
1219 | return err; | |
33ff9823 | 1220 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1221 | if (err) |
1222 | return err; | |
1223 | ||
435faee1 DB |
1224 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1225 | * is inferred from register state. | |
1226 | */ | |
1227 | for (i = 0; i < meta.access_size; i++) { | |
1228 | err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1); | |
1229 | if (err) | |
1230 | return err; | |
1231 | } | |
1232 | ||
17a52670 AS |
1233 | /* reset caller saved regs */ |
1234 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
1235 | reg = regs + caller_saved[i]; | |
1236 | reg->type = NOT_INIT; | |
1237 | reg->imm = 0; | |
1238 | } | |
1239 | ||
1240 | /* update return register */ | |
1241 | if (fn->ret_type == RET_INTEGER) { | |
1242 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
1243 | } else if (fn->ret_type == RET_VOID) { | |
1244 | regs[BPF_REG_0].type = NOT_INIT; | |
1245 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
1246 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
48461135 | 1247 | regs[BPF_REG_0].max_value = regs[BPF_REG_0].min_value = 0; |
17a52670 AS |
1248 | /* remember map_ptr, so that check_map_access() |
1249 | * can check 'value_size' boundary of memory access | |
1250 | * to map element returned from bpf_map_lookup_elem() | |
1251 | */ | |
33ff9823 | 1252 | if (meta.map_ptr == NULL) { |
17a52670 AS |
1253 | verbose("kernel subsystem misconfigured verifier\n"); |
1254 | return -EINVAL; | |
1255 | } | |
33ff9823 | 1256 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
57a09bf0 | 1257 | regs[BPF_REG_0].id = ++env->id_gen; |
17a52670 AS |
1258 | } else { |
1259 | verbose("unknown return type %d of func %d\n", | |
1260 | fn->ret_type, func_id); | |
1261 | return -EINVAL; | |
1262 | } | |
04fd61ab | 1263 | |
33ff9823 | 1264 | err = check_map_func_compatibility(meta.map_ptr, func_id); |
35578d79 KX |
1265 | if (err) |
1266 | return err; | |
04fd61ab | 1267 | |
969bf05e AS |
1268 | if (changes_data) |
1269 | clear_all_pkt_pointers(env); | |
1270 | return 0; | |
1271 | } | |
1272 | ||
58e2af8b JK |
1273 | static int check_packet_ptr_add(struct bpf_verifier_env *env, |
1274 | struct bpf_insn *insn) | |
969bf05e | 1275 | { |
58e2af8b JK |
1276 | struct bpf_reg_state *regs = env->cur_state.regs; |
1277 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1278 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
1279 | struct bpf_reg_state tmp_reg; | |
969bf05e AS |
1280 | s32 imm; |
1281 | ||
1282 | if (BPF_SRC(insn->code) == BPF_K) { | |
1283 | /* pkt_ptr += imm */ | |
1284 | imm = insn->imm; | |
1285 | ||
1286 | add_imm: | |
1287 | if (imm <= 0) { | |
1288 | verbose("addition of negative constant to packet pointer is not allowed\n"); | |
1289 | return -EACCES; | |
1290 | } | |
1291 | if (imm >= MAX_PACKET_OFF || | |
1292 | imm + dst_reg->off >= MAX_PACKET_OFF) { | |
1293 | verbose("constant %d is too large to add to packet pointer\n", | |
1294 | imm); | |
1295 | return -EACCES; | |
1296 | } | |
1297 | /* a constant was added to pkt_ptr. | |
1298 | * Remember it while keeping the same 'id' | |
1299 | */ | |
1300 | dst_reg->off += imm; | |
1301 | } else { | |
1b9b69ec AS |
1302 | if (src_reg->type == PTR_TO_PACKET) { |
1303 | /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ | |
1304 | tmp_reg = *dst_reg; /* save r7 state */ | |
1305 | *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ | |
1306 | src_reg = &tmp_reg; /* pretend it's src_reg state */ | |
1307 | /* if the checks below reject it, the copy won't matter, | |
1308 | * since we're rejecting the whole program. If all ok, | |
1309 | * then imm22 state will be added to r7 | |
1310 | * and r7 will be pkt(id=0,off=22,r=62) while | |
1311 | * r6 will stay as pkt(id=0,off=0,r=62) | |
1312 | */ | |
1313 | } | |
1314 | ||
969bf05e AS |
1315 | if (src_reg->type == CONST_IMM) { |
1316 | /* pkt_ptr += reg where reg is known constant */ | |
1317 | imm = src_reg->imm; | |
1318 | goto add_imm; | |
1319 | } | |
1320 | /* disallow pkt_ptr += reg | |
1321 | * if reg is not uknown_value with guaranteed zero upper bits | |
1322 | * otherwise pkt_ptr may overflow and addition will become | |
1323 | * subtraction which is not allowed | |
1324 | */ | |
1325 | if (src_reg->type != UNKNOWN_VALUE) { | |
1326 | verbose("cannot add '%s' to ptr_to_packet\n", | |
1327 | reg_type_str[src_reg->type]); | |
1328 | return -EACCES; | |
1329 | } | |
1330 | if (src_reg->imm < 48) { | |
1331 | verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", | |
1332 | src_reg->imm); | |
1333 | return -EACCES; | |
1334 | } | |
1335 | /* dst_reg stays as pkt_ptr type and since some positive | |
1336 | * integer value was added to the pointer, increment its 'id' | |
1337 | */ | |
1f415a74 | 1338 | dst_reg->id = ++env->id_gen; |
969bf05e AS |
1339 | |
1340 | /* something was added to pkt_ptr, set range and off to zero */ | |
1341 | dst_reg->off = 0; | |
1342 | dst_reg->range = 0; | |
1343 | } | |
1344 | return 0; | |
1345 | } | |
1346 | ||
58e2af8b | 1347 | static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) |
969bf05e | 1348 | { |
58e2af8b JK |
1349 | struct bpf_reg_state *regs = env->cur_state.regs; |
1350 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
969bf05e AS |
1351 | u8 opcode = BPF_OP(insn->code); |
1352 | s64 imm_log2; | |
1353 | ||
1354 | /* for type == UNKNOWN_VALUE: | |
1355 | * imm > 0 -> number of zero upper bits | |
1356 | * imm == 0 -> don't track which is the same as all bits can be non-zero | |
1357 | */ | |
1358 | ||
1359 | if (BPF_SRC(insn->code) == BPF_X) { | |
58e2af8b | 1360 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
969bf05e AS |
1361 | |
1362 | if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && | |
1363 | dst_reg->imm && opcode == BPF_ADD) { | |
1364 | /* dreg += sreg | |
1365 | * where both have zero upper bits. Adding them | |
1366 | * can only result making one more bit non-zero | |
1367 | * in the larger value. | |
1368 | * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) | |
1369 | * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) | |
1370 | */ | |
1371 | dst_reg->imm = min(dst_reg->imm, src_reg->imm); | |
1372 | dst_reg->imm--; | |
1373 | return 0; | |
1374 | } | |
1375 | if (src_reg->type == CONST_IMM && src_reg->imm > 0 && | |
1376 | dst_reg->imm && opcode == BPF_ADD) { | |
1377 | /* dreg += sreg | |
1378 | * where dreg has zero upper bits and sreg is const. | |
1379 | * Adding them can only result making one more bit | |
1380 | * non-zero in the larger value. | |
1381 | */ | |
1382 | imm_log2 = __ilog2_u64((long long)src_reg->imm); | |
1383 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1384 | dst_reg->imm--; | |
1385 | return 0; | |
1386 | } | |
1387 | /* all other cases non supported yet, just mark dst_reg */ | |
1388 | dst_reg->imm = 0; | |
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | /* sign extend 32-bit imm into 64-bit to make sure that | |
1393 | * negative values occupy bit 63. Note ilog2() would have | |
1394 | * been incorrect, since sizeof(insn->imm) == 4 | |
1395 | */ | |
1396 | imm_log2 = __ilog2_u64((long long)insn->imm); | |
1397 | ||
1398 | if (dst_reg->imm && opcode == BPF_LSH) { | |
1399 | /* reg <<= imm | |
1400 | * if reg was a result of 2 byte load, then its imm == 48 | |
1401 | * which means that upper 48 bits are zero and shifting this reg | |
1402 | * left by 4 would mean that upper 44 bits are still zero | |
1403 | */ | |
1404 | dst_reg->imm -= insn->imm; | |
1405 | } else if (dst_reg->imm && opcode == BPF_MUL) { | |
1406 | /* reg *= imm | |
1407 | * if multiplying by 14 subtract 4 | |
1408 | * This is conservative calculation of upper zero bits. | |
1409 | * It's not trying to special case insn->imm == 1 or 0 cases | |
1410 | */ | |
1411 | dst_reg->imm -= imm_log2 + 1; | |
1412 | } else if (opcode == BPF_AND) { | |
1413 | /* reg &= imm */ | |
1414 | dst_reg->imm = 63 - imm_log2; | |
1415 | } else if (dst_reg->imm && opcode == BPF_ADD) { | |
1416 | /* reg += imm */ | |
1417 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1418 | dst_reg->imm--; | |
1419 | } else if (opcode == BPF_RSH) { | |
1420 | /* reg >>= imm | |
1421 | * which means that after right shift, upper bits will be zero | |
1422 | * note that verifier already checked that | |
1423 | * 0 <= imm < 64 for shift insn | |
1424 | */ | |
1425 | dst_reg->imm += insn->imm; | |
1426 | if (unlikely(dst_reg->imm > 64)) | |
1427 | /* some dumb code did: | |
1428 | * r2 = *(u32 *)mem; | |
1429 | * r2 >>= 32; | |
1430 | * and all bits are zero now */ | |
1431 | dst_reg->imm = 64; | |
1432 | } else { | |
1433 | /* all other alu ops, means that we don't know what will | |
1434 | * happen to the value, mark it with unknown number of zero bits | |
1435 | */ | |
1436 | dst_reg->imm = 0; | |
1437 | } | |
1438 | ||
1439 | if (dst_reg->imm < 0) { | |
1440 | /* all 64 bits of the register can contain non-zero bits | |
1441 | * and such value cannot be added to ptr_to_packet, since it | |
1442 | * may overflow, mark it as unknown to avoid further eval | |
1443 | */ | |
1444 | dst_reg->imm = 0; | |
1445 | } | |
1446 | return 0; | |
1447 | } | |
1448 | ||
58e2af8b JK |
1449 | static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, |
1450 | struct bpf_insn *insn) | |
969bf05e | 1451 | { |
58e2af8b JK |
1452 | struct bpf_reg_state *regs = env->cur_state.regs; |
1453 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1454 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
969bf05e AS |
1455 | u8 opcode = BPF_OP(insn->code); |
1456 | ||
1457 | /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn. | |
1458 | * Don't care about overflow or negative values, just add them | |
1459 | */ | |
1460 | if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) | |
1461 | dst_reg->imm += insn->imm; | |
1462 | else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && | |
1463 | src_reg->type == CONST_IMM) | |
1464 | dst_reg->imm += src_reg->imm; | |
1465 | else | |
1466 | mark_reg_unknown_value(regs, insn->dst_reg); | |
17a52670 AS |
1467 | return 0; |
1468 | } | |
1469 | ||
48461135 JB |
1470 | static void check_reg_overflow(struct bpf_reg_state *reg) |
1471 | { | |
1472 | if (reg->max_value > BPF_REGISTER_MAX_RANGE) | |
1473 | reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1474 | if ((s64)reg->min_value < BPF_REGISTER_MIN_RANGE) | |
1475 | reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1476 | } | |
1477 | ||
1478 | static void adjust_reg_min_max_vals(struct bpf_verifier_env *env, | |
1479 | struct bpf_insn *insn) | |
1480 | { | |
1481 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; | |
1482 | u64 min_val = BPF_REGISTER_MIN_RANGE, max_val = BPF_REGISTER_MAX_RANGE; | |
1483 | bool min_set = false, max_set = false; | |
1484 | u8 opcode = BPF_OP(insn->code); | |
1485 | ||
1486 | dst_reg = ®s[insn->dst_reg]; | |
1487 | if (BPF_SRC(insn->code) == BPF_X) { | |
1488 | check_reg_overflow(®s[insn->src_reg]); | |
1489 | min_val = regs[insn->src_reg].min_value; | |
1490 | max_val = regs[insn->src_reg].max_value; | |
1491 | ||
1492 | /* If the source register is a random pointer then the | |
1493 | * min_value/max_value values represent the range of the known | |
1494 | * accesses into that value, not the actual min/max value of the | |
1495 | * register itself. In this case we have to reset the reg range | |
1496 | * values so we know it is not safe to look at. | |
1497 | */ | |
1498 | if (regs[insn->src_reg].type != CONST_IMM && | |
1499 | regs[insn->src_reg].type != UNKNOWN_VALUE) { | |
1500 | min_val = BPF_REGISTER_MIN_RANGE; | |
1501 | max_val = BPF_REGISTER_MAX_RANGE; | |
1502 | } | |
1503 | } else if (insn->imm < BPF_REGISTER_MAX_RANGE && | |
1504 | (s64)insn->imm > BPF_REGISTER_MIN_RANGE) { | |
1505 | min_val = max_val = insn->imm; | |
1506 | min_set = max_set = true; | |
1507 | } | |
1508 | ||
1509 | /* We don't know anything about what was done to this register, mark it | |
1510 | * as unknown. | |
1511 | */ | |
1512 | if (min_val == BPF_REGISTER_MIN_RANGE && | |
1513 | max_val == BPF_REGISTER_MAX_RANGE) { | |
1514 | reset_reg_range_values(regs, insn->dst_reg); | |
1515 | return; | |
1516 | } | |
1517 | ||
1518 | switch (opcode) { | |
1519 | case BPF_ADD: | |
1520 | dst_reg->min_value += min_val; | |
1521 | dst_reg->max_value += max_val; | |
1522 | break; | |
1523 | case BPF_SUB: | |
1524 | dst_reg->min_value -= min_val; | |
1525 | dst_reg->max_value -= max_val; | |
1526 | break; | |
1527 | case BPF_MUL: | |
1528 | dst_reg->min_value *= min_val; | |
1529 | dst_reg->max_value *= max_val; | |
1530 | break; | |
1531 | case BPF_AND: | |
1532 | /* & is special since it could end up with 0 bits set. */ | |
1533 | dst_reg->min_value &= min_val; | |
1534 | dst_reg->max_value = max_val; | |
1535 | break; | |
1536 | case BPF_LSH: | |
1537 | /* Gotta have special overflow logic here, if we're shifting | |
1538 | * more than MAX_RANGE then just assume we have an invalid | |
1539 | * range. | |
1540 | */ | |
1541 | if (min_val > ilog2(BPF_REGISTER_MAX_RANGE)) | |
1542 | dst_reg->min_value = BPF_REGISTER_MIN_RANGE; | |
1543 | else | |
1544 | dst_reg->min_value <<= min_val; | |
1545 | ||
1546 | if (max_val > ilog2(BPF_REGISTER_MAX_RANGE)) | |
1547 | dst_reg->max_value = BPF_REGISTER_MAX_RANGE; | |
1548 | else | |
1549 | dst_reg->max_value <<= max_val; | |
1550 | break; | |
1551 | case BPF_RSH: | |
1552 | dst_reg->min_value >>= min_val; | |
1553 | dst_reg->max_value >>= max_val; | |
1554 | break; | |
1555 | case BPF_MOD: | |
1556 | /* % is special since it is an unsigned modulus, so the floor | |
1557 | * will always be 0. | |
1558 | */ | |
1559 | dst_reg->min_value = 0; | |
1560 | dst_reg->max_value = max_val - 1; | |
1561 | break; | |
1562 | default: | |
1563 | reset_reg_range_values(regs, insn->dst_reg); | |
1564 | break; | |
1565 | } | |
1566 | ||
1567 | check_reg_overflow(dst_reg); | |
1568 | } | |
1569 | ||
17a52670 | 1570 | /* check validity of 32-bit and 64-bit arithmetic operations */ |
58e2af8b | 1571 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 1572 | { |
58e2af8b | 1573 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; |
17a52670 AS |
1574 | u8 opcode = BPF_OP(insn->code); |
1575 | int err; | |
1576 | ||
1577 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
1578 | if (opcode == BPF_NEG) { | |
1579 | if (BPF_SRC(insn->code) != 0 || | |
1580 | insn->src_reg != BPF_REG_0 || | |
1581 | insn->off != 0 || insn->imm != 0) { | |
1582 | verbose("BPF_NEG uses reserved fields\n"); | |
1583 | return -EINVAL; | |
1584 | } | |
1585 | } else { | |
1586 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
1587 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { | |
1588 | verbose("BPF_END uses reserved fields\n"); | |
1589 | return -EINVAL; | |
1590 | } | |
1591 | } | |
1592 | ||
1593 | /* check src operand */ | |
1594 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1595 | if (err) | |
1596 | return err; | |
1597 | ||
1be7f75d AS |
1598 | if (is_pointer_value(env, insn->dst_reg)) { |
1599 | verbose("R%d pointer arithmetic prohibited\n", | |
1600 | insn->dst_reg); | |
1601 | return -EACCES; | |
1602 | } | |
1603 | ||
17a52670 AS |
1604 | /* check dest operand */ |
1605 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1606 | if (err) | |
1607 | return err; | |
1608 | ||
1609 | } else if (opcode == BPF_MOV) { | |
1610 | ||
1611 | if (BPF_SRC(insn->code) == BPF_X) { | |
1612 | if (insn->imm != 0 || insn->off != 0) { | |
1613 | verbose("BPF_MOV uses reserved fields\n"); | |
1614 | return -EINVAL; | |
1615 | } | |
1616 | ||
1617 | /* check src operand */ | |
1618 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1619 | if (err) | |
1620 | return err; | |
1621 | } else { | |
1622 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1623 | verbose("BPF_MOV uses reserved fields\n"); | |
1624 | return -EINVAL; | |
1625 | } | |
1626 | } | |
1627 | ||
1628 | /* check dest operand */ | |
1629 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1630 | if (err) | |
1631 | return err; | |
1632 | ||
48461135 JB |
1633 | /* we are setting our register to something new, we need to |
1634 | * reset its range values. | |
1635 | */ | |
1636 | reset_reg_range_values(regs, insn->dst_reg); | |
1637 | ||
17a52670 AS |
1638 | if (BPF_SRC(insn->code) == BPF_X) { |
1639 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
1640 | /* case: R1 = R2 | |
1641 | * copy register state to dest reg | |
1642 | */ | |
1643 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
1644 | } else { | |
1be7f75d AS |
1645 | if (is_pointer_value(env, insn->src_reg)) { |
1646 | verbose("R%d partial copy of pointer\n", | |
1647 | insn->src_reg); | |
1648 | return -EACCES; | |
1649 | } | |
57a09bf0 | 1650 | mark_reg_unknown_value(regs, insn->dst_reg); |
17a52670 AS |
1651 | } |
1652 | } else { | |
1653 | /* case: R = imm | |
1654 | * remember the value we stored into this reg | |
1655 | */ | |
1656 | regs[insn->dst_reg].type = CONST_IMM; | |
1657 | regs[insn->dst_reg].imm = insn->imm; | |
48461135 JB |
1658 | regs[insn->dst_reg].max_value = insn->imm; |
1659 | regs[insn->dst_reg].min_value = insn->imm; | |
17a52670 AS |
1660 | } |
1661 | ||
1662 | } else if (opcode > BPF_END) { | |
1663 | verbose("invalid BPF_ALU opcode %x\n", opcode); | |
1664 | return -EINVAL; | |
1665 | ||
1666 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
1667 | ||
17a52670 AS |
1668 | if (BPF_SRC(insn->code) == BPF_X) { |
1669 | if (insn->imm != 0 || insn->off != 0) { | |
1670 | verbose("BPF_ALU uses reserved fields\n"); | |
1671 | return -EINVAL; | |
1672 | } | |
1673 | /* check src1 operand */ | |
1674 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1675 | if (err) | |
1676 | return err; | |
1677 | } else { | |
1678 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1679 | verbose("BPF_ALU uses reserved fields\n"); | |
1680 | return -EINVAL; | |
1681 | } | |
1682 | } | |
1683 | ||
1684 | /* check src2 operand */ | |
1685 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1686 | if (err) | |
1687 | return err; | |
1688 | ||
1689 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
1690 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
1691 | verbose("div by zero\n"); | |
1692 | return -EINVAL; | |
1693 | } | |
1694 | ||
229394e8 RV |
1695 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
1696 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
1697 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
1698 | ||
1699 | if (insn->imm < 0 || insn->imm >= size) { | |
1700 | verbose("invalid shift %d\n", insn->imm); | |
1701 | return -EINVAL; | |
1702 | } | |
1703 | } | |
1704 | ||
1a0dc1ac AS |
1705 | /* check dest operand */ |
1706 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
1707 | if (err) | |
1708 | return err; | |
1709 | ||
1710 | dst_reg = ®s[insn->dst_reg]; | |
1711 | ||
48461135 JB |
1712 | /* first we want to adjust our ranges. */ |
1713 | adjust_reg_min_max_vals(env, insn); | |
1714 | ||
17a52670 AS |
1715 | /* pattern match 'bpf_add Rx, imm' instruction */ |
1716 | if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && | |
1a0dc1ac AS |
1717 | dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { |
1718 | dst_reg->type = PTR_TO_STACK; | |
1719 | dst_reg->imm = insn->imm; | |
1720 | return 0; | |
969bf05e AS |
1721 | } else if (opcode == BPF_ADD && |
1722 | BPF_CLASS(insn->code) == BPF_ALU64 && | |
1b9b69ec AS |
1723 | (dst_reg->type == PTR_TO_PACKET || |
1724 | (BPF_SRC(insn->code) == BPF_X && | |
1725 | regs[insn->src_reg].type == PTR_TO_PACKET))) { | |
969bf05e AS |
1726 | /* ptr_to_packet += K|X */ |
1727 | return check_packet_ptr_add(env, insn); | |
1728 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1729 | dst_reg->type == UNKNOWN_VALUE && | |
1730 | env->allow_ptr_leaks) { | |
1731 | /* unknown += K|X */ | |
1732 | return evaluate_reg_alu(env, insn); | |
1733 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1734 | dst_reg->type == CONST_IMM && | |
1735 | env->allow_ptr_leaks) { | |
1736 | /* reg_imm += K|X */ | |
1737 | return evaluate_reg_imm_alu(env, insn); | |
1be7f75d AS |
1738 | } else if (is_pointer_value(env, insn->dst_reg)) { |
1739 | verbose("R%d pointer arithmetic prohibited\n", | |
1740 | insn->dst_reg); | |
1741 | return -EACCES; | |
1742 | } else if (BPF_SRC(insn->code) == BPF_X && | |
1743 | is_pointer_value(env, insn->src_reg)) { | |
1744 | verbose("R%d pointer arithmetic prohibited\n", | |
1745 | insn->src_reg); | |
1746 | return -EACCES; | |
1747 | } | |
17a52670 | 1748 | |
48461135 JB |
1749 | /* If we did pointer math on a map value then just set it to our |
1750 | * PTR_TO_MAP_VALUE_ADJ type so we can deal with any stores or | |
1751 | * loads to this register appropriately, otherwise just mark the | |
1752 | * register as unknown. | |
1753 | */ | |
1754 | if (env->allow_ptr_leaks && | |
1755 | (dst_reg->type == PTR_TO_MAP_VALUE || | |
1756 | dst_reg->type == PTR_TO_MAP_VALUE_ADJ)) | |
1757 | dst_reg->type = PTR_TO_MAP_VALUE_ADJ; | |
1758 | else | |
1759 | mark_reg_unknown_value(regs, insn->dst_reg); | |
17a52670 AS |
1760 | } |
1761 | ||
1762 | return 0; | |
1763 | } | |
1764 | ||
58e2af8b JK |
1765 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
1766 | struct bpf_reg_state *dst_reg) | |
969bf05e | 1767 | { |
58e2af8b | 1768 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e | 1769 | int i; |
2d2be8ca DB |
1770 | |
1771 | /* LLVM can generate two kind of checks: | |
1772 | * | |
1773 | * Type 1: | |
1774 | * | |
1775 | * r2 = r3; | |
1776 | * r2 += 8; | |
1777 | * if (r2 > pkt_end) goto <handle exception> | |
1778 | * <access okay> | |
1779 | * | |
1780 | * Where: | |
1781 | * r2 == dst_reg, pkt_end == src_reg | |
1782 | * r2=pkt(id=n,off=8,r=0) | |
1783 | * r3=pkt(id=n,off=0,r=0) | |
1784 | * | |
1785 | * Type 2: | |
1786 | * | |
1787 | * r2 = r3; | |
1788 | * r2 += 8; | |
1789 | * if (pkt_end >= r2) goto <access okay> | |
1790 | * <handle exception> | |
1791 | * | |
1792 | * Where: | |
1793 | * pkt_end == dst_reg, r2 == src_reg | |
1794 | * r2=pkt(id=n,off=8,r=0) | |
1795 | * r3=pkt(id=n,off=0,r=0) | |
1796 | * | |
1797 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
1798 | * so that range of bytes [r3, r3 + 8) is safe to access. | |
969bf05e | 1799 | */ |
2d2be8ca | 1800 | |
969bf05e AS |
1801 | for (i = 0; i < MAX_BPF_REG; i++) |
1802 | if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) | |
1803 | regs[i].range = dst_reg->off; | |
1804 | ||
1805 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1806 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1807 | continue; | |
1808 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1809 | if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) | |
1810 | reg->range = dst_reg->off; | |
1811 | } | |
1812 | } | |
1813 | ||
48461135 JB |
1814 | /* Adjusts the register min/max values in the case that the dst_reg is the |
1815 | * variable register that we are working on, and src_reg is a constant or we're | |
1816 | * simply doing a BPF_K check. | |
1817 | */ | |
1818 | static void reg_set_min_max(struct bpf_reg_state *true_reg, | |
1819 | struct bpf_reg_state *false_reg, u64 val, | |
1820 | u8 opcode) | |
1821 | { | |
1822 | switch (opcode) { | |
1823 | case BPF_JEQ: | |
1824 | /* If this is false then we know nothing Jon Snow, but if it is | |
1825 | * true then we know for sure. | |
1826 | */ | |
1827 | true_reg->max_value = true_reg->min_value = val; | |
1828 | break; | |
1829 | case BPF_JNE: | |
1830 | /* If this is true we know nothing Jon Snow, but if it is false | |
1831 | * we know the value for sure; | |
1832 | */ | |
1833 | false_reg->max_value = false_reg->min_value = val; | |
1834 | break; | |
1835 | case BPF_JGT: | |
1836 | /* Unsigned comparison, the minimum value is 0. */ | |
1837 | false_reg->min_value = 0; | |
1838 | case BPF_JSGT: | |
1839 | /* If this is false then we know the maximum val is val, | |
1840 | * otherwise we know the min val is val+1. | |
1841 | */ | |
1842 | false_reg->max_value = val; | |
1843 | true_reg->min_value = val + 1; | |
1844 | break; | |
1845 | case BPF_JGE: | |
1846 | /* Unsigned comparison, the minimum value is 0. */ | |
1847 | false_reg->min_value = 0; | |
1848 | case BPF_JSGE: | |
1849 | /* If this is false then we know the maximum value is val - 1, | |
1850 | * otherwise we know the mimimum value is val. | |
1851 | */ | |
1852 | false_reg->max_value = val - 1; | |
1853 | true_reg->min_value = val; | |
1854 | break; | |
1855 | default: | |
1856 | break; | |
1857 | } | |
1858 | ||
1859 | check_reg_overflow(false_reg); | |
1860 | check_reg_overflow(true_reg); | |
1861 | } | |
1862 | ||
1863 | /* Same as above, but for the case that dst_reg is a CONST_IMM reg and src_reg | |
1864 | * is the variable reg. | |
1865 | */ | |
1866 | static void reg_set_min_max_inv(struct bpf_reg_state *true_reg, | |
1867 | struct bpf_reg_state *false_reg, u64 val, | |
1868 | u8 opcode) | |
1869 | { | |
1870 | switch (opcode) { | |
1871 | case BPF_JEQ: | |
1872 | /* If this is false then we know nothing Jon Snow, but if it is | |
1873 | * true then we know for sure. | |
1874 | */ | |
1875 | true_reg->max_value = true_reg->min_value = val; | |
1876 | break; | |
1877 | case BPF_JNE: | |
1878 | /* If this is true we know nothing Jon Snow, but if it is false | |
1879 | * we know the value for sure; | |
1880 | */ | |
1881 | false_reg->max_value = false_reg->min_value = val; | |
1882 | break; | |
1883 | case BPF_JGT: | |
1884 | /* Unsigned comparison, the minimum value is 0. */ | |
1885 | true_reg->min_value = 0; | |
1886 | case BPF_JSGT: | |
1887 | /* | |
1888 | * If this is false, then the val is <= the register, if it is | |
1889 | * true the register <= to the val. | |
1890 | */ | |
1891 | false_reg->min_value = val; | |
1892 | true_reg->max_value = val - 1; | |
1893 | break; | |
1894 | case BPF_JGE: | |
1895 | /* Unsigned comparison, the minimum value is 0. */ | |
1896 | true_reg->min_value = 0; | |
1897 | case BPF_JSGE: | |
1898 | /* If this is false then constant < register, if it is true then | |
1899 | * the register < constant. | |
1900 | */ | |
1901 | false_reg->min_value = val + 1; | |
1902 | true_reg->max_value = val; | |
1903 | break; | |
1904 | default: | |
1905 | break; | |
1906 | } | |
1907 | ||
1908 | check_reg_overflow(false_reg); | |
1909 | check_reg_overflow(true_reg); | |
1910 | } | |
1911 | ||
57a09bf0 TG |
1912 | static void mark_map_reg(struct bpf_reg_state *regs, u32 regno, u32 id, |
1913 | enum bpf_reg_type type) | |
1914 | { | |
1915 | struct bpf_reg_state *reg = ®s[regno]; | |
1916 | ||
1917 | if (reg->type == PTR_TO_MAP_VALUE_OR_NULL && reg->id == id) { | |
1918 | reg->type = type; | |
1919 | if (type == UNKNOWN_VALUE) | |
1920 | mark_reg_unknown_value(regs, regno); | |
1921 | } | |
1922 | } | |
1923 | ||
1924 | /* The logic is similar to find_good_pkt_pointers(), both could eventually | |
1925 | * be folded together at some point. | |
1926 | */ | |
1927 | static void mark_map_regs(struct bpf_verifier_state *state, u32 regno, | |
1928 | enum bpf_reg_type type) | |
1929 | { | |
1930 | struct bpf_reg_state *regs = state->regs; | |
1931 | int i; | |
1932 | ||
1933 | for (i = 0; i < MAX_BPF_REG; i++) | |
1934 | mark_map_reg(regs, i, regs[regno].id, type); | |
1935 | ||
1936 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1937 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1938 | continue; | |
1939 | mark_map_reg(state->spilled_regs, i / BPF_REG_SIZE, | |
1940 | regs[regno].id, type); | |
1941 | } | |
1942 | } | |
1943 | ||
58e2af8b | 1944 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
1945 | struct bpf_insn *insn, int *insn_idx) |
1946 | { | |
58e2af8b JK |
1947 | struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; |
1948 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; | |
17a52670 AS |
1949 | u8 opcode = BPF_OP(insn->code); |
1950 | int err; | |
1951 | ||
1952 | if (opcode > BPF_EXIT) { | |
1953 | verbose("invalid BPF_JMP opcode %x\n", opcode); | |
1954 | return -EINVAL; | |
1955 | } | |
1956 | ||
1957 | if (BPF_SRC(insn->code) == BPF_X) { | |
1958 | if (insn->imm != 0) { | |
1959 | verbose("BPF_JMP uses reserved fields\n"); | |
1960 | return -EINVAL; | |
1961 | } | |
1962 | ||
1963 | /* check src1 operand */ | |
1964 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1965 | if (err) | |
1966 | return err; | |
1be7f75d AS |
1967 | |
1968 | if (is_pointer_value(env, insn->src_reg)) { | |
1969 | verbose("R%d pointer comparison prohibited\n", | |
1970 | insn->src_reg); | |
1971 | return -EACCES; | |
1972 | } | |
17a52670 AS |
1973 | } else { |
1974 | if (insn->src_reg != BPF_REG_0) { | |
1975 | verbose("BPF_JMP uses reserved fields\n"); | |
1976 | return -EINVAL; | |
1977 | } | |
1978 | } | |
1979 | ||
1980 | /* check src2 operand */ | |
1981 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1982 | if (err) | |
1983 | return err; | |
1984 | ||
1a0dc1ac AS |
1985 | dst_reg = ®s[insn->dst_reg]; |
1986 | ||
17a52670 AS |
1987 | /* detect if R == 0 where R was initialized to zero earlier */ |
1988 | if (BPF_SRC(insn->code) == BPF_K && | |
1989 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
1a0dc1ac | 1990 | dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { |
17a52670 AS |
1991 | if (opcode == BPF_JEQ) { |
1992 | /* if (imm == imm) goto pc+off; | |
1993 | * only follow the goto, ignore fall-through | |
1994 | */ | |
1995 | *insn_idx += insn->off; | |
1996 | return 0; | |
1997 | } else { | |
1998 | /* if (imm != imm) goto pc+off; | |
1999 | * only follow fall-through branch, since | |
2000 | * that's where the program will go | |
2001 | */ | |
2002 | return 0; | |
2003 | } | |
2004 | } | |
2005 | ||
2006 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
2007 | if (!other_branch) | |
2008 | return -EFAULT; | |
2009 | ||
48461135 JB |
2010 | /* detect if we are comparing against a constant value so we can adjust |
2011 | * our min/max values for our dst register. | |
2012 | */ | |
2013 | if (BPF_SRC(insn->code) == BPF_X) { | |
2014 | if (regs[insn->src_reg].type == CONST_IMM) | |
2015 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2016 | dst_reg, regs[insn->src_reg].imm, | |
2017 | opcode); | |
2018 | else if (dst_reg->type == CONST_IMM) | |
2019 | reg_set_min_max_inv(&other_branch->regs[insn->src_reg], | |
2020 | ®s[insn->src_reg], dst_reg->imm, | |
2021 | opcode); | |
2022 | } else { | |
2023 | reg_set_min_max(&other_branch->regs[insn->dst_reg], | |
2024 | dst_reg, insn->imm, opcode); | |
2025 | } | |
2026 | ||
58e2af8b | 2027 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 2028 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
2029 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
2030 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
57a09bf0 TG |
2031 | /* Mark all identical map registers in each branch as either |
2032 | * safe or unknown depending R == 0 or R != 0 conditional. | |
2033 | */ | |
2034 | mark_map_regs(this_branch, insn->dst_reg, | |
2035 | opcode == BPF_JEQ ? PTR_TO_MAP_VALUE : UNKNOWN_VALUE); | |
2036 | mark_map_regs(other_branch, insn->dst_reg, | |
2037 | opcode == BPF_JEQ ? UNKNOWN_VALUE : PTR_TO_MAP_VALUE); | |
969bf05e AS |
2038 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && |
2039 | dst_reg->type == PTR_TO_PACKET && | |
2040 | regs[insn->src_reg].type == PTR_TO_PACKET_END) { | |
2d2be8ca DB |
2041 | find_good_pkt_pointers(this_branch, dst_reg); |
2042 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && | |
2043 | dst_reg->type == PTR_TO_PACKET_END && | |
2044 | regs[insn->src_reg].type == PTR_TO_PACKET) { | |
2045 | find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); | |
1be7f75d AS |
2046 | } else if (is_pointer_value(env, insn->dst_reg)) { |
2047 | verbose("R%d pointer comparison prohibited\n", insn->dst_reg); | |
2048 | return -EACCES; | |
17a52670 AS |
2049 | } |
2050 | if (log_level) | |
2d2be8ca | 2051 | print_verifier_state(this_branch); |
17a52670 AS |
2052 | return 0; |
2053 | } | |
2054 | ||
0246e64d AS |
2055 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
2056 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
2057 | { | |
2058 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
2059 | ||
2060 | return (struct bpf_map *) (unsigned long) imm64; | |
2061 | } | |
2062 | ||
17a52670 | 2063 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 2064 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 2065 | { |
58e2af8b | 2066 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
2067 | int err; |
2068 | ||
2069 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
2070 | verbose("invalid BPF_LD_IMM insn\n"); | |
2071 | return -EINVAL; | |
2072 | } | |
2073 | if (insn->off != 0) { | |
2074 | verbose("BPF_LD_IMM64 uses reserved fields\n"); | |
2075 | return -EINVAL; | |
2076 | } | |
2077 | ||
2078 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
2079 | if (err) | |
2080 | return err; | |
2081 | ||
6b173873 JK |
2082 | if (insn->src_reg == 0) { |
2083 | /* generic move 64-bit immediate into a register, | |
2084 | * only analyzer needs to collect the ld_imm value. | |
2085 | */ | |
2086 | u64 imm = ((u64)(insn + 1)->imm << 32) | (u32)insn->imm; | |
2087 | ||
2088 | if (!env->analyzer_ops) | |
2089 | return 0; | |
2090 | ||
2091 | regs[insn->dst_reg].type = CONST_IMM; | |
2092 | regs[insn->dst_reg].imm = imm; | |
17a52670 | 2093 | return 0; |
6b173873 | 2094 | } |
17a52670 AS |
2095 | |
2096 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
2097 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
2098 | ||
2099 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
2100 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
2101 | return 0; | |
2102 | } | |
2103 | ||
96be4325 DB |
2104 | static bool may_access_skb(enum bpf_prog_type type) |
2105 | { | |
2106 | switch (type) { | |
2107 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
2108 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 2109 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
2110 | return true; |
2111 | default: | |
2112 | return false; | |
2113 | } | |
2114 | } | |
2115 | ||
ddd872bc AS |
2116 | /* verify safety of LD_ABS|LD_IND instructions: |
2117 | * - they can only appear in the programs where ctx == skb | |
2118 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
2119 | * preserve R6-R9, and store return value into R0 | |
2120 | * | |
2121 | * Implicit input: | |
2122 | * ctx == skb == R6 == CTX | |
2123 | * | |
2124 | * Explicit input: | |
2125 | * SRC == any register | |
2126 | * IMM == 32-bit immediate | |
2127 | * | |
2128 | * Output: | |
2129 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
2130 | */ | |
58e2af8b | 2131 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 2132 | { |
58e2af8b | 2133 | struct bpf_reg_state *regs = env->cur_state.regs; |
ddd872bc | 2134 | u8 mode = BPF_MODE(insn->code); |
58e2af8b | 2135 | struct bpf_reg_state *reg; |
ddd872bc AS |
2136 | int i, err; |
2137 | ||
24701ece | 2138 | if (!may_access_skb(env->prog->type)) { |
1a0dc1ac | 2139 | verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
2140 | return -EINVAL; |
2141 | } | |
2142 | ||
2143 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 2144 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 2145 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
1a0dc1ac | 2146 | verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
2147 | return -EINVAL; |
2148 | } | |
2149 | ||
2150 | /* check whether implicit source operand (register R6) is readable */ | |
2151 | err = check_reg_arg(regs, BPF_REG_6, SRC_OP); | |
2152 | if (err) | |
2153 | return err; | |
2154 | ||
2155 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
2156 | verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
2157 | return -EINVAL; | |
2158 | } | |
2159 | ||
2160 | if (mode == BPF_IND) { | |
2161 | /* check explicit source operand */ | |
2162 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2163 | if (err) | |
2164 | return err; | |
2165 | } | |
2166 | ||
2167 | /* reset caller saved regs to unreadable */ | |
2168 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
2169 | reg = regs + caller_saved[i]; | |
2170 | reg->type = NOT_INIT; | |
2171 | reg->imm = 0; | |
2172 | } | |
2173 | ||
2174 | /* mark destination R0 register as readable, since it contains | |
2175 | * the value fetched from the packet | |
2176 | */ | |
2177 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
2178 | return 0; | |
2179 | } | |
2180 | ||
475fb78f AS |
2181 | /* non-recursive DFS pseudo code |
2182 | * 1 procedure DFS-iterative(G,v): | |
2183 | * 2 label v as discovered | |
2184 | * 3 let S be a stack | |
2185 | * 4 S.push(v) | |
2186 | * 5 while S is not empty | |
2187 | * 6 t <- S.pop() | |
2188 | * 7 if t is what we're looking for: | |
2189 | * 8 return t | |
2190 | * 9 for all edges e in G.adjacentEdges(t) do | |
2191 | * 10 if edge e is already labelled | |
2192 | * 11 continue with the next edge | |
2193 | * 12 w <- G.adjacentVertex(t,e) | |
2194 | * 13 if vertex w is not discovered and not explored | |
2195 | * 14 label e as tree-edge | |
2196 | * 15 label w as discovered | |
2197 | * 16 S.push(w) | |
2198 | * 17 continue at 5 | |
2199 | * 18 else if vertex w is discovered | |
2200 | * 19 label e as back-edge | |
2201 | * 20 else | |
2202 | * 21 // vertex w is explored | |
2203 | * 22 label e as forward- or cross-edge | |
2204 | * 23 label t as explored | |
2205 | * 24 S.pop() | |
2206 | * | |
2207 | * convention: | |
2208 | * 0x10 - discovered | |
2209 | * 0x11 - discovered and fall-through edge labelled | |
2210 | * 0x12 - discovered and fall-through and branch edges labelled | |
2211 | * 0x20 - explored | |
2212 | */ | |
2213 | ||
2214 | enum { | |
2215 | DISCOVERED = 0x10, | |
2216 | EXPLORED = 0x20, | |
2217 | FALLTHROUGH = 1, | |
2218 | BRANCH = 2, | |
2219 | }; | |
2220 | ||
58e2af8b | 2221 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 2222 | |
475fb78f AS |
2223 | static int *insn_stack; /* stack of insns to process */ |
2224 | static int cur_stack; /* current stack index */ | |
2225 | static int *insn_state; | |
2226 | ||
2227 | /* t, w, e - match pseudo-code above: | |
2228 | * t - index of current instruction | |
2229 | * w - next instruction | |
2230 | * e - edge | |
2231 | */ | |
58e2af8b | 2232 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
2233 | { |
2234 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
2235 | return 0; | |
2236 | ||
2237 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
2238 | return 0; | |
2239 | ||
2240 | if (w < 0 || w >= env->prog->len) { | |
2241 | verbose("jump out of range from insn %d to %d\n", t, w); | |
2242 | return -EINVAL; | |
2243 | } | |
2244 | ||
f1bca824 AS |
2245 | if (e == BRANCH) |
2246 | /* mark branch target for state pruning */ | |
2247 | env->explored_states[w] = STATE_LIST_MARK; | |
2248 | ||
475fb78f AS |
2249 | if (insn_state[w] == 0) { |
2250 | /* tree-edge */ | |
2251 | insn_state[t] = DISCOVERED | e; | |
2252 | insn_state[w] = DISCOVERED; | |
2253 | if (cur_stack >= env->prog->len) | |
2254 | return -E2BIG; | |
2255 | insn_stack[cur_stack++] = w; | |
2256 | return 1; | |
2257 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
2258 | verbose("back-edge from insn %d to %d\n", t, w); | |
2259 | return -EINVAL; | |
2260 | } else if (insn_state[w] == EXPLORED) { | |
2261 | /* forward- or cross-edge */ | |
2262 | insn_state[t] = DISCOVERED | e; | |
2263 | } else { | |
2264 | verbose("insn state internal bug\n"); | |
2265 | return -EFAULT; | |
2266 | } | |
2267 | return 0; | |
2268 | } | |
2269 | ||
2270 | /* non-recursive depth-first-search to detect loops in BPF program | |
2271 | * loop == back-edge in directed graph | |
2272 | */ | |
58e2af8b | 2273 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
2274 | { |
2275 | struct bpf_insn *insns = env->prog->insnsi; | |
2276 | int insn_cnt = env->prog->len; | |
2277 | int ret = 0; | |
2278 | int i, t; | |
2279 | ||
2280 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2281 | if (!insn_state) | |
2282 | return -ENOMEM; | |
2283 | ||
2284 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
2285 | if (!insn_stack) { | |
2286 | kfree(insn_state); | |
2287 | return -ENOMEM; | |
2288 | } | |
2289 | ||
2290 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
2291 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
2292 | cur_stack = 1; | |
2293 | ||
2294 | peek_stack: | |
2295 | if (cur_stack == 0) | |
2296 | goto check_state; | |
2297 | t = insn_stack[cur_stack - 1]; | |
2298 | ||
2299 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
2300 | u8 opcode = BPF_OP(insns[t].code); | |
2301 | ||
2302 | if (opcode == BPF_EXIT) { | |
2303 | goto mark_explored; | |
2304 | } else if (opcode == BPF_CALL) { | |
2305 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2306 | if (ret == 1) | |
2307 | goto peek_stack; | |
2308 | else if (ret < 0) | |
2309 | goto err_free; | |
07016151 DB |
2310 | if (t + 1 < insn_cnt) |
2311 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2312 | } else if (opcode == BPF_JA) { |
2313 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
2314 | ret = -EINVAL; | |
2315 | goto err_free; | |
2316 | } | |
2317 | /* unconditional jump with single edge */ | |
2318 | ret = push_insn(t, t + insns[t].off + 1, | |
2319 | FALLTHROUGH, env); | |
2320 | if (ret == 1) | |
2321 | goto peek_stack; | |
2322 | else if (ret < 0) | |
2323 | goto err_free; | |
f1bca824 AS |
2324 | /* tell verifier to check for equivalent states |
2325 | * after every call and jump | |
2326 | */ | |
c3de6317 AS |
2327 | if (t + 1 < insn_cnt) |
2328 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2329 | } else { |
2330 | /* conditional jump with two edges */ | |
2331 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2332 | if (ret == 1) | |
2333 | goto peek_stack; | |
2334 | else if (ret < 0) | |
2335 | goto err_free; | |
2336 | ||
2337 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
2338 | if (ret == 1) | |
2339 | goto peek_stack; | |
2340 | else if (ret < 0) | |
2341 | goto err_free; | |
2342 | } | |
2343 | } else { | |
2344 | /* all other non-branch instructions with single | |
2345 | * fall-through edge | |
2346 | */ | |
2347 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2348 | if (ret == 1) | |
2349 | goto peek_stack; | |
2350 | else if (ret < 0) | |
2351 | goto err_free; | |
2352 | } | |
2353 | ||
2354 | mark_explored: | |
2355 | insn_state[t] = EXPLORED; | |
2356 | if (cur_stack-- <= 0) { | |
2357 | verbose("pop stack internal bug\n"); | |
2358 | ret = -EFAULT; | |
2359 | goto err_free; | |
2360 | } | |
2361 | goto peek_stack; | |
2362 | ||
2363 | check_state: | |
2364 | for (i = 0; i < insn_cnt; i++) { | |
2365 | if (insn_state[i] != EXPLORED) { | |
2366 | verbose("unreachable insn %d\n", i); | |
2367 | ret = -EINVAL; | |
2368 | goto err_free; | |
2369 | } | |
2370 | } | |
2371 | ret = 0; /* cfg looks good */ | |
2372 | ||
2373 | err_free: | |
2374 | kfree(insn_state); | |
2375 | kfree(insn_stack); | |
2376 | return ret; | |
2377 | } | |
2378 | ||
969bf05e AS |
2379 | /* the following conditions reduce the number of explored insns |
2380 | * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet | |
2381 | */ | |
58e2af8b JK |
2382 | static bool compare_ptrs_to_packet(struct bpf_reg_state *old, |
2383 | struct bpf_reg_state *cur) | |
969bf05e AS |
2384 | { |
2385 | if (old->id != cur->id) | |
2386 | return false; | |
2387 | ||
2388 | /* old ptr_to_packet is more conservative, since it allows smaller | |
2389 | * range. Ex: | |
2390 | * old(off=0,r=10) is equal to cur(off=0,r=20), because | |
2391 | * old(off=0,r=10) means that with range=10 the verifier proceeded | |
2392 | * further and found no issues with the program. Now we're in the same | |
2393 | * spot with cur(off=0,r=20), so we're safe too, since anything further | |
2394 | * will only be looking at most 10 bytes after this pointer. | |
2395 | */ | |
2396 | if (old->off == cur->off && old->range < cur->range) | |
2397 | return true; | |
2398 | ||
2399 | /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) | |
2400 | * since both cannot be used for packet access and safe(old) | |
2401 | * pointer has smaller off that could be used for further | |
2402 | * 'if (ptr > data_end)' check | |
2403 | * Ex: | |
2404 | * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean | |
2405 | * that we cannot access the packet. | |
2406 | * The safe range is: | |
2407 | * [ptr, ptr + range - off) | |
2408 | * so whenever off >=range, it means no safe bytes from this pointer. | |
2409 | * When comparing old->off <= cur->off, it means that older code | |
2410 | * went with smaller offset and that offset was later | |
2411 | * used to figure out the safe range after 'if (ptr > data_end)' check | |
2412 | * Say, 'old' state was explored like: | |
2413 | * ... R3(off=0, r=0) | |
2414 | * R4 = R3 + 20 | |
2415 | * ... now R4(off=20,r=0) <-- here | |
2416 | * if (R4 > data_end) | |
2417 | * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. | |
2418 | * ... the code further went all the way to bpf_exit. | |
2419 | * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). | |
2420 | * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier | |
2421 | * goes further, such cur_R4 will give larger safe packet range after | |
2422 | * 'if (R4 > data_end)' and all further insn were already good with r=20, | |
2423 | * so they will be good with r=30 and we can prune the search. | |
2424 | */ | |
2425 | if (old->off <= cur->off && | |
2426 | old->off >= old->range && cur->off >= cur->range) | |
2427 | return true; | |
2428 | ||
2429 | return false; | |
2430 | } | |
2431 | ||
f1bca824 AS |
2432 | /* compare two verifier states |
2433 | * | |
2434 | * all states stored in state_list are known to be valid, since | |
2435 | * verifier reached 'bpf_exit' instruction through them | |
2436 | * | |
2437 | * this function is called when verifier exploring different branches of | |
2438 | * execution popped from the state stack. If it sees an old state that has | |
2439 | * more strict register state and more strict stack state then this execution | |
2440 | * branch doesn't need to be explored further, since verifier already | |
2441 | * concluded that more strict state leads to valid finish. | |
2442 | * | |
2443 | * Therefore two states are equivalent if register state is more conservative | |
2444 | * and explored stack state is more conservative than the current one. | |
2445 | * Example: | |
2446 | * explored current | |
2447 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
2448 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
2449 | * | |
2450 | * In other words if current stack state (one being explored) has more | |
2451 | * valid slots than old one that already passed validation, it means | |
2452 | * the verifier can stop exploring and conclude that current state is valid too | |
2453 | * | |
2454 | * Similarly with registers. If explored state has register type as invalid | |
2455 | * whereas register type in current state is meaningful, it means that | |
2456 | * the current state will reach 'bpf_exit' instruction safely | |
2457 | */ | |
48461135 JB |
2458 | static bool states_equal(struct bpf_verifier_env *env, |
2459 | struct bpf_verifier_state *old, | |
58e2af8b | 2460 | struct bpf_verifier_state *cur) |
f1bca824 | 2461 | { |
58e2af8b | 2462 | struct bpf_reg_state *rold, *rcur; |
f1bca824 AS |
2463 | int i; |
2464 | ||
2465 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
2466 | rold = &old->regs[i]; |
2467 | rcur = &cur->regs[i]; | |
2468 | ||
2469 | if (memcmp(rold, rcur, sizeof(*rold)) == 0) | |
2470 | continue; | |
2471 | ||
48461135 JB |
2472 | /* If the ranges were not the same, but everything else was and |
2473 | * we didn't do a variable access into a map then we are a-ok. | |
2474 | */ | |
2475 | if (!env->varlen_map_value_access && | |
2476 | rold->type == rcur->type && rold->imm == rcur->imm) | |
2477 | continue; | |
2478 | ||
1a0dc1ac AS |
2479 | if (rold->type == NOT_INIT || |
2480 | (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT)) | |
2481 | continue; | |
2482 | ||
969bf05e AS |
2483 | if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && |
2484 | compare_ptrs_to_packet(rold, rcur)) | |
2485 | continue; | |
2486 | ||
1a0dc1ac | 2487 | return false; |
f1bca824 AS |
2488 | } |
2489 | ||
2490 | for (i = 0; i < MAX_BPF_STACK; i++) { | |
9c399760 AS |
2491 | if (old->stack_slot_type[i] == STACK_INVALID) |
2492 | continue; | |
2493 | if (old->stack_slot_type[i] != cur->stack_slot_type[i]) | |
2494 | /* Ex: old explored (safe) state has STACK_SPILL in | |
2495 | * this stack slot, but current has has STACK_MISC -> | |
2496 | * this verifier states are not equivalent, | |
2497 | * return false to continue verification of this path | |
2498 | */ | |
f1bca824 | 2499 | return false; |
9c399760 AS |
2500 | if (i % BPF_REG_SIZE) |
2501 | continue; | |
2502 | if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], | |
2503 | &cur->spilled_regs[i / BPF_REG_SIZE], | |
2504 | sizeof(old->spilled_regs[0]))) | |
2505 | /* when explored and current stack slot types are | |
2506 | * the same, check that stored pointers types | |
2507 | * are the same as well. | |
2508 | * Ex: explored safe path could have stored | |
58e2af8b | 2509 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} |
9c399760 | 2510 | * but current path has stored: |
58e2af8b | 2511 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} |
9c399760 AS |
2512 | * such verifier states are not equivalent. |
2513 | * return false to continue verification of this path | |
2514 | */ | |
2515 | return false; | |
2516 | else | |
2517 | continue; | |
f1bca824 AS |
2518 | } |
2519 | return true; | |
2520 | } | |
2521 | ||
58e2af8b | 2522 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 2523 | { |
58e2af8b JK |
2524 | struct bpf_verifier_state_list *new_sl; |
2525 | struct bpf_verifier_state_list *sl; | |
f1bca824 AS |
2526 | |
2527 | sl = env->explored_states[insn_idx]; | |
2528 | if (!sl) | |
2529 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
2530 | * be doing state search here | |
2531 | */ | |
2532 | return 0; | |
2533 | ||
2534 | while (sl != STATE_LIST_MARK) { | |
48461135 | 2535 | if (states_equal(env, &sl->state, &env->cur_state)) |
f1bca824 AS |
2536 | /* reached equivalent register/stack state, |
2537 | * prune the search | |
2538 | */ | |
2539 | return 1; | |
2540 | sl = sl->next; | |
2541 | } | |
2542 | ||
2543 | /* there were no equivalent states, remember current one. | |
2544 | * technically the current state is not proven to be safe yet, | |
2545 | * but it will either reach bpf_exit (which means it's safe) or | |
2546 | * it will be rejected. Since there are no loops, we won't be | |
2547 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
2548 | */ | |
58e2af8b | 2549 | new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); |
f1bca824 AS |
2550 | if (!new_sl) |
2551 | return -ENOMEM; | |
2552 | ||
2553 | /* add new state to the head of linked list */ | |
2554 | memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); | |
2555 | new_sl->next = env->explored_states[insn_idx]; | |
2556 | env->explored_states[insn_idx] = new_sl; | |
2557 | return 0; | |
2558 | } | |
2559 | ||
13a27dfc JK |
2560 | static int ext_analyzer_insn_hook(struct bpf_verifier_env *env, |
2561 | int insn_idx, int prev_insn_idx) | |
2562 | { | |
2563 | if (!env->analyzer_ops || !env->analyzer_ops->insn_hook) | |
2564 | return 0; | |
2565 | ||
2566 | return env->analyzer_ops->insn_hook(env, insn_idx, prev_insn_idx); | |
2567 | } | |
2568 | ||
58e2af8b | 2569 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 2570 | { |
58e2af8b | 2571 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 2572 | struct bpf_insn *insns = env->prog->insnsi; |
58e2af8b | 2573 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
2574 | int insn_cnt = env->prog->len; |
2575 | int insn_idx, prev_insn_idx = 0; | |
2576 | int insn_processed = 0; | |
2577 | bool do_print_state = false; | |
2578 | ||
2579 | init_reg_state(regs); | |
2580 | insn_idx = 0; | |
48461135 | 2581 | env->varlen_map_value_access = false; |
17a52670 AS |
2582 | for (;;) { |
2583 | struct bpf_insn *insn; | |
2584 | u8 class; | |
2585 | int err; | |
2586 | ||
2587 | if (insn_idx >= insn_cnt) { | |
2588 | verbose("invalid insn idx %d insn_cnt %d\n", | |
2589 | insn_idx, insn_cnt); | |
2590 | return -EFAULT; | |
2591 | } | |
2592 | ||
2593 | insn = &insns[insn_idx]; | |
2594 | class = BPF_CLASS(insn->code); | |
2595 | ||
07016151 | 2596 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
17a52670 AS |
2597 | verbose("BPF program is too large. Proccessed %d insn\n", |
2598 | insn_processed); | |
2599 | return -E2BIG; | |
2600 | } | |
2601 | ||
f1bca824 AS |
2602 | err = is_state_visited(env, insn_idx); |
2603 | if (err < 0) | |
2604 | return err; | |
2605 | if (err == 1) { | |
2606 | /* found equivalent state, can prune the search */ | |
2607 | if (log_level) { | |
2608 | if (do_print_state) | |
2609 | verbose("\nfrom %d to %d: safe\n", | |
2610 | prev_insn_idx, insn_idx); | |
2611 | else | |
2612 | verbose("%d: safe\n", insn_idx); | |
2613 | } | |
2614 | goto process_bpf_exit; | |
2615 | } | |
2616 | ||
17a52670 AS |
2617 | if (log_level && do_print_state) { |
2618 | verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); | |
1a0dc1ac | 2619 | print_verifier_state(&env->cur_state); |
17a52670 AS |
2620 | do_print_state = false; |
2621 | } | |
2622 | ||
2623 | if (log_level) { | |
2624 | verbose("%d: ", insn_idx); | |
2625 | print_bpf_insn(insn); | |
2626 | } | |
2627 | ||
13a27dfc JK |
2628 | err = ext_analyzer_insn_hook(env, insn_idx, prev_insn_idx); |
2629 | if (err) | |
2630 | return err; | |
2631 | ||
17a52670 | 2632 | if (class == BPF_ALU || class == BPF_ALU64) { |
1be7f75d | 2633 | err = check_alu_op(env, insn); |
17a52670 AS |
2634 | if (err) |
2635 | return err; | |
2636 | ||
2637 | } else if (class == BPF_LDX) { | |
3df126f3 | 2638 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
2639 | |
2640 | /* check for reserved fields is already done */ | |
2641 | ||
17a52670 AS |
2642 | /* check src operand */ |
2643 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2644 | if (err) | |
2645 | return err; | |
2646 | ||
2647 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
2648 | if (err) | |
2649 | return err; | |
2650 | ||
725f9dcd AS |
2651 | src_reg_type = regs[insn->src_reg].type; |
2652 | ||
17a52670 AS |
2653 | /* check that memory (src_reg + off) is readable, |
2654 | * the state of dst_reg will be updated by this func | |
2655 | */ | |
2656 | err = check_mem_access(env, insn->src_reg, insn->off, | |
2657 | BPF_SIZE(insn->code), BPF_READ, | |
2658 | insn->dst_reg); | |
2659 | if (err) | |
2660 | return err; | |
2661 | ||
48461135 | 2662 | reset_reg_range_values(regs, insn->dst_reg); |
ea2e7ce5 AS |
2663 | if (BPF_SIZE(insn->code) != BPF_W && |
2664 | BPF_SIZE(insn->code) != BPF_DW) { | |
725f9dcd AS |
2665 | insn_idx++; |
2666 | continue; | |
2667 | } | |
9bac3d6d | 2668 | |
3df126f3 JK |
2669 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
2670 | ||
2671 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
2672 | /* saw a valid insn |
2673 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 2674 | * save type to validate intersecting paths |
9bac3d6d | 2675 | */ |
3df126f3 | 2676 | *prev_src_type = src_reg_type; |
9bac3d6d | 2677 | |
3df126f3 | 2678 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 2679 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 2680 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
2681 | /* ABuser program is trying to use the same insn |
2682 | * dst_reg = *(u32*) (src_reg + off) | |
2683 | * with different pointer types: | |
2684 | * src_reg == ctx in one branch and | |
2685 | * src_reg == stack|map in some other branch. | |
2686 | * Reject it. | |
2687 | */ | |
2688 | verbose("same insn cannot be used with different pointers\n"); | |
2689 | return -EINVAL; | |
2690 | } | |
2691 | ||
17a52670 | 2692 | } else if (class == BPF_STX) { |
3df126f3 | 2693 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 2694 | |
17a52670 AS |
2695 | if (BPF_MODE(insn->code) == BPF_XADD) { |
2696 | err = check_xadd(env, insn); | |
2697 | if (err) | |
2698 | return err; | |
2699 | insn_idx++; | |
2700 | continue; | |
2701 | } | |
2702 | ||
17a52670 AS |
2703 | /* check src1 operand */ |
2704 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2705 | if (err) | |
2706 | return err; | |
2707 | /* check src2 operand */ | |
2708 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2709 | if (err) | |
2710 | return err; | |
2711 | ||
d691f9e8 AS |
2712 | dst_reg_type = regs[insn->dst_reg].type; |
2713 | ||
17a52670 AS |
2714 | /* check that memory (dst_reg + off) is writeable */ |
2715 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2716 | BPF_SIZE(insn->code), BPF_WRITE, | |
2717 | insn->src_reg); | |
2718 | if (err) | |
2719 | return err; | |
2720 | ||
3df126f3 JK |
2721 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
2722 | ||
2723 | if (*prev_dst_type == NOT_INIT) { | |
2724 | *prev_dst_type = dst_reg_type; | |
2725 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 2726 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 2727 | *prev_dst_type == PTR_TO_CTX)) { |
d691f9e8 AS |
2728 | verbose("same insn cannot be used with different pointers\n"); |
2729 | return -EINVAL; | |
2730 | } | |
2731 | ||
17a52670 AS |
2732 | } else if (class == BPF_ST) { |
2733 | if (BPF_MODE(insn->code) != BPF_MEM || | |
2734 | insn->src_reg != BPF_REG_0) { | |
2735 | verbose("BPF_ST uses reserved fields\n"); | |
2736 | return -EINVAL; | |
2737 | } | |
2738 | /* check src operand */ | |
2739 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2740 | if (err) | |
2741 | return err; | |
2742 | ||
2743 | /* check that memory (dst_reg + off) is writeable */ | |
2744 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2745 | BPF_SIZE(insn->code), BPF_WRITE, | |
2746 | -1); | |
2747 | if (err) | |
2748 | return err; | |
2749 | ||
2750 | } else if (class == BPF_JMP) { | |
2751 | u8 opcode = BPF_OP(insn->code); | |
2752 | ||
2753 | if (opcode == BPF_CALL) { | |
2754 | if (BPF_SRC(insn->code) != BPF_K || | |
2755 | insn->off != 0 || | |
2756 | insn->src_reg != BPF_REG_0 || | |
2757 | insn->dst_reg != BPF_REG_0) { | |
2758 | verbose("BPF_CALL uses reserved fields\n"); | |
2759 | return -EINVAL; | |
2760 | } | |
2761 | ||
2762 | err = check_call(env, insn->imm); | |
2763 | if (err) | |
2764 | return err; | |
2765 | ||
2766 | } else if (opcode == BPF_JA) { | |
2767 | if (BPF_SRC(insn->code) != BPF_K || | |
2768 | insn->imm != 0 || | |
2769 | insn->src_reg != BPF_REG_0 || | |
2770 | insn->dst_reg != BPF_REG_0) { | |
2771 | verbose("BPF_JA uses reserved fields\n"); | |
2772 | return -EINVAL; | |
2773 | } | |
2774 | ||
2775 | insn_idx += insn->off + 1; | |
2776 | continue; | |
2777 | ||
2778 | } else if (opcode == BPF_EXIT) { | |
2779 | if (BPF_SRC(insn->code) != BPF_K || | |
2780 | insn->imm != 0 || | |
2781 | insn->src_reg != BPF_REG_0 || | |
2782 | insn->dst_reg != BPF_REG_0) { | |
2783 | verbose("BPF_EXIT uses reserved fields\n"); | |
2784 | return -EINVAL; | |
2785 | } | |
2786 | ||
2787 | /* eBPF calling convetion is such that R0 is used | |
2788 | * to return the value from eBPF program. | |
2789 | * Make sure that it's readable at this time | |
2790 | * of bpf_exit, which means that program wrote | |
2791 | * something into it earlier | |
2792 | */ | |
2793 | err = check_reg_arg(regs, BPF_REG_0, SRC_OP); | |
2794 | if (err) | |
2795 | return err; | |
2796 | ||
1be7f75d AS |
2797 | if (is_pointer_value(env, BPF_REG_0)) { |
2798 | verbose("R0 leaks addr as return value\n"); | |
2799 | return -EACCES; | |
2800 | } | |
2801 | ||
f1bca824 | 2802 | process_bpf_exit: |
17a52670 AS |
2803 | insn_idx = pop_stack(env, &prev_insn_idx); |
2804 | if (insn_idx < 0) { | |
2805 | break; | |
2806 | } else { | |
2807 | do_print_state = true; | |
2808 | continue; | |
2809 | } | |
2810 | } else { | |
2811 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
2812 | if (err) | |
2813 | return err; | |
2814 | } | |
2815 | } else if (class == BPF_LD) { | |
2816 | u8 mode = BPF_MODE(insn->code); | |
2817 | ||
2818 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
2819 | err = check_ld_abs(env, insn); |
2820 | if (err) | |
2821 | return err; | |
2822 | ||
17a52670 AS |
2823 | } else if (mode == BPF_IMM) { |
2824 | err = check_ld_imm(env, insn); | |
2825 | if (err) | |
2826 | return err; | |
2827 | ||
2828 | insn_idx++; | |
2829 | } else { | |
2830 | verbose("invalid BPF_LD mode\n"); | |
2831 | return -EINVAL; | |
2832 | } | |
48461135 | 2833 | reset_reg_range_values(regs, insn->dst_reg); |
17a52670 AS |
2834 | } else { |
2835 | verbose("unknown insn class %d\n", class); | |
2836 | return -EINVAL; | |
2837 | } | |
2838 | ||
2839 | insn_idx++; | |
2840 | } | |
2841 | ||
1a0dc1ac | 2842 | verbose("processed %d insns\n", insn_processed); |
17a52670 AS |
2843 | return 0; |
2844 | } | |
2845 | ||
fdc15d38 AS |
2846 | static int check_map_prog_compatibility(struct bpf_map *map, |
2847 | struct bpf_prog *prog) | |
2848 | ||
2849 | { | |
2850 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT && | |
2851 | (map->map_type == BPF_MAP_TYPE_HASH || | |
2852 | map->map_type == BPF_MAP_TYPE_PERCPU_HASH) && | |
2853 | (map->map_flags & BPF_F_NO_PREALLOC)) { | |
2854 | verbose("perf_event programs can only use preallocated hash map\n"); | |
2855 | return -EINVAL; | |
2856 | } | |
2857 | return 0; | |
2858 | } | |
2859 | ||
0246e64d AS |
2860 | /* look for pseudo eBPF instructions that access map FDs and |
2861 | * replace them with actual map pointers | |
2862 | */ | |
58e2af8b | 2863 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
2864 | { |
2865 | struct bpf_insn *insn = env->prog->insnsi; | |
2866 | int insn_cnt = env->prog->len; | |
fdc15d38 | 2867 | int i, j, err; |
0246e64d AS |
2868 | |
2869 | for (i = 0; i < insn_cnt; i++, insn++) { | |
9bac3d6d | 2870 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 2871 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
9bac3d6d AS |
2872 | verbose("BPF_LDX uses reserved fields\n"); |
2873 | return -EINVAL; | |
2874 | } | |
2875 | ||
d691f9e8 AS |
2876 | if (BPF_CLASS(insn->code) == BPF_STX && |
2877 | ((BPF_MODE(insn->code) != BPF_MEM && | |
2878 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
2879 | verbose("BPF_STX uses reserved fields\n"); | |
2880 | return -EINVAL; | |
2881 | } | |
2882 | ||
0246e64d AS |
2883 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
2884 | struct bpf_map *map; | |
2885 | struct fd f; | |
2886 | ||
2887 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
2888 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
2889 | insn[1].off != 0) { | |
2890 | verbose("invalid bpf_ld_imm64 insn\n"); | |
2891 | return -EINVAL; | |
2892 | } | |
2893 | ||
2894 | if (insn->src_reg == 0) | |
2895 | /* valid generic load 64-bit imm */ | |
2896 | goto next_insn; | |
2897 | ||
2898 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
2899 | verbose("unrecognized bpf_ld_imm64 insn\n"); | |
2900 | return -EINVAL; | |
2901 | } | |
2902 | ||
2903 | f = fdget(insn->imm); | |
c2101297 | 2904 | map = __bpf_map_get(f); |
0246e64d AS |
2905 | if (IS_ERR(map)) { |
2906 | verbose("fd %d is not pointing to valid bpf_map\n", | |
2907 | insn->imm); | |
0246e64d AS |
2908 | return PTR_ERR(map); |
2909 | } | |
2910 | ||
fdc15d38 AS |
2911 | err = check_map_prog_compatibility(map, env->prog); |
2912 | if (err) { | |
2913 | fdput(f); | |
2914 | return err; | |
2915 | } | |
2916 | ||
0246e64d AS |
2917 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
2918 | insn[0].imm = (u32) (unsigned long) map; | |
2919 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
2920 | ||
2921 | /* check whether we recorded this map already */ | |
2922 | for (j = 0; j < env->used_map_cnt; j++) | |
2923 | if (env->used_maps[j] == map) { | |
2924 | fdput(f); | |
2925 | goto next_insn; | |
2926 | } | |
2927 | ||
2928 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
2929 | fdput(f); | |
2930 | return -E2BIG; | |
2931 | } | |
2932 | ||
0246e64d AS |
2933 | /* hold the map. If the program is rejected by verifier, |
2934 | * the map will be released by release_maps() or it | |
2935 | * will be used by the valid program until it's unloaded | |
2936 | * and all maps are released in free_bpf_prog_info() | |
2937 | */ | |
92117d84 AS |
2938 | map = bpf_map_inc(map, false); |
2939 | if (IS_ERR(map)) { | |
2940 | fdput(f); | |
2941 | return PTR_ERR(map); | |
2942 | } | |
2943 | env->used_maps[env->used_map_cnt++] = map; | |
2944 | ||
0246e64d AS |
2945 | fdput(f); |
2946 | next_insn: | |
2947 | insn++; | |
2948 | i++; | |
2949 | } | |
2950 | } | |
2951 | ||
2952 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
2953 | * 'struct bpf_map *' into a register instead of user map_fd. | |
2954 | * These pointers will be used later by verifier to validate map access. | |
2955 | */ | |
2956 | return 0; | |
2957 | } | |
2958 | ||
2959 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 2960 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
2961 | { |
2962 | int i; | |
2963 | ||
2964 | for (i = 0; i < env->used_map_cnt; i++) | |
2965 | bpf_map_put(env->used_maps[i]); | |
2966 | } | |
2967 | ||
2968 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 2969 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
2970 | { |
2971 | struct bpf_insn *insn = env->prog->insnsi; | |
2972 | int insn_cnt = env->prog->len; | |
2973 | int i; | |
2974 | ||
2975 | for (i = 0; i < insn_cnt; i++, insn++) | |
2976 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
2977 | insn->src_reg = 0; | |
2978 | } | |
2979 | ||
9bac3d6d AS |
2980 | /* convert load instructions that access fields of 'struct __sk_buff' |
2981 | * into sequence of instructions that access fields of 'struct sk_buff' | |
2982 | */ | |
58e2af8b | 2983 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 2984 | { |
36bbef52 | 2985 | const struct bpf_verifier_ops *ops = env->prog->aux->ops; |
3df126f3 | 2986 | const int insn_cnt = env->prog->len; |
36bbef52 | 2987 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 2988 | struct bpf_prog *new_prog; |
d691f9e8 | 2989 | enum bpf_access_type type; |
3df126f3 | 2990 | int i, cnt, delta = 0; |
9bac3d6d | 2991 | |
36bbef52 DB |
2992 | if (ops->gen_prologue) { |
2993 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
2994 | env->prog); | |
2995 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
2996 | verbose("bpf verifier is misconfigured\n"); | |
2997 | return -EINVAL; | |
2998 | } else if (cnt) { | |
2999 | new_prog = bpf_patch_insn_single(env->prog, 0, | |
3000 | insn_buf, cnt); | |
3001 | if (!new_prog) | |
3002 | return -ENOMEM; | |
3003 | env->prog = new_prog; | |
3df126f3 | 3004 | delta += cnt - 1; |
36bbef52 DB |
3005 | } |
3006 | } | |
3007 | ||
3008 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
3009 | return 0; |
3010 | ||
3df126f3 | 3011 | insn = env->prog->insnsi + delta; |
36bbef52 | 3012 | |
9bac3d6d | 3013 | for (i = 0; i < insn_cnt; i++, insn++) { |
ea2e7ce5 AS |
3014 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) || |
3015 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) | |
d691f9e8 | 3016 | type = BPF_READ; |
ea2e7ce5 AS |
3017 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) || |
3018 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) | |
d691f9e8 AS |
3019 | type = BPF_WRITE; |
3020 | else | |
9bac3d6d AS |
3021 | continue; |
3022 | ||
3df126f3 | 3023 | if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX) |
9bac3d6d | 3024 | continue; |
9bac3d6d | 3025 | |
36bbef52 DB |
3026 | cnt = ops->convert_ctx_access(type, insn->dst_reg, insn->src_reg, |
3027 | insn->off, insn_buf, env->prog); | |
9bac3d6d AS |
3028 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { |
3029 | verbose("bpf verifier is misconfigured\n"); | |
3030 | return -EINVAL; | |
3031 | } | |
3032 | ||
3df126f3 JK |
3033 | new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf, |
3034 | cnt); | |
9bac3d6d AS |
3035 | if (!new_prog) |
3036 | return -ENOMEM; | |
3037 | ||
3df126f3 | 3038 | delta += cnt - 1; |
9bac3d6d AS |
3039 | |
3040 | /* keep walking new program and skip insns we just inserted */ | |
3041 | env->prog = new_prog; | |
3df126f3 | 3042 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
3043 | } |
3044 | ||
3045 | return 0; | |
3046 | } | |
3047 | ||
58e2af8b | 3048 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 3049 | { |
58e2af8b | 3050 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
3051 | int i; |
3052 | ||
3053 | if (!env->explored_states) | |
3054 | return; | |
3055 | ||
3056 | for (i = 0; i < env->prog->len; i++) { | |
3057 | sl = env->explored_states[i]; | |
3058 | ||
3059 | if (sl) | |
3060 | while (sl != STATE_LIST_MARK) { | |
3061 | sln = sl->next; | |
3062 | kfree(sl); | |
3063 | sl = sln; | |
3064 | } | |
3065 | } | |
3066 | ||
3067 | kfree(env->explored_states); | |
3068 | } | |
3069 | ||
9bac3d6d | 3070 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 3071 | { |
cbd35700 | 3072 | char __user *log_ubuf = NULL; |
58e2af8b | 3073 | struct bpf_verifier_env *env; |
51580e79 AS |
3074 | int ret = -EINVAL; |
3075 | ||
9bac3d6d | 3076 | if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS) |
cbd35700 AS |
3077 | return -E2BIG; |
3078 | ||
58e2af8b | 3079 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
3080 | * allocate/free it every time bpf_check() is called |
3081 | */ | |
58e2af8b | 3082 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
3083 | if (!env) |
3084 | return -ENOMEM; | |
3085 | ||
3df126f3 JK |
3086 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
3087 | (*prog)->len); | |
3088 | ret = -ENOMEM; | |
3089 | if (!env->insn_aux_data) | |
3090 | goto err_free_env; | |
9bac3d6d | 3091 | env->prog = *prog; |
0246e64d | 3092 | |
cbd35700 AS |
3093 | /* grab the mutex to protect few globals used by verifier */ |
3094 | mutex_lock(&bpf_verifier_lock); | |
3095 | ||
3096 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
3097 | /* user requested verbose verifier output | |
3098 | * and supplied buffer to store the verification trace | |
3099 | */ | |
3100 | log_level = attr->log_level; | |
3101 | log_ubuf = (char __user *) (unsigned long) attr->log_buf; | |
3102 | log_size = attr->log_size; | |
3103 | log_len = 0; | |
3104 | ||
3105 | ret = -EINVAL; | |
3106 | /* log_* values have to be sane */ | |
3107 | if (log_size < 128 || log_size > UINT_MAX >> 8 || | |
3108 | log_level == 0 || log_ubuf == NULL) | |
3df126f3 | 3109 | goto err_unlock; |
cbd35700 AS |
3110 | |
3111 | ret = -ENOMEM; | |
3112 | log_buf = vmalloc(log_size); | |
3113 | if (!log_buf) | |
3df126f3 | 3114 | goto err_unlock; |
cbd35700 AS |
3115 | } else { |
3116 | log_level = 0; | |
3117 | } | |
3118 | ||
0246e64d AS |
3119 | ret = replace_map_fd_with_map_ptr(env); |
3120 | if (ret < 0) | |
3121 | goto skip_full_check; | |
3122 | ||
9bac3d6d | 3123 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 3124 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
3125 | GFP_USER); |
3126 | ret = -ENOMEM; | |
3127 | if (!env->explored_states) | |
3128 | goto skip_full_check; | |
3129 | ||
475fb78f AS |
3130 | ret = check_cfg(env); |
3131 | if (ret < 0) | |
3132 | goto skip_full_check; | |
3133 | ||
1be7f75d AS |
3134 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
3135 | ||
17a52670 | 3136 | ret = do_check(env); |
cbd35700 | 3137 | |
0246e64d | 3138 | skip_full_check: |
17a52670 | 3139 | while (pop_stack(env, NULL) >= 0); |
f1bca824 | 3140 | free_states(env); |
0246e64d | 3141 | |
9bac3d6d AS |
3142 | if (ret == 0) |
3143 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
3144 | ret = convert_ctx_accesses(env); | |
3145 | ||
cbd35700 AS |
3146 | if (log_level && log_len >= log_size - 1) { |
3147 | BUG_ON(log_len >= log_size); | |
3148 | /* verifier log exceeded user supplied buffer */ | |
3149 | ret = -ENOSPC; | |
3150 | /* fall through to return what was recorded */ | |
3151 | } | |
3152 | ||
3153 | /* copy verifier log back to user space including trailing zero */ | |
3154 | if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { | |
3155 | ret = -EFAULT; | |
3156 | goto free_log_buf; | |
3157 | } | |
3158 | ||
0246e64d AS |
3159 | if (ret == 0 && env->used_map_cnt) { |
3160 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
3161 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
3162 | sizeof(env->used_maps[0]), | |
3163 | GFP_KERNEL); | |
0246e64d | 3164 | |
9bac3d6d | 3165 | if (!env->prog->aux->used_maps) { |
0246e64d AS |
3166 | ret = -ENOMEM; |
3167 | goto free_log_buf; | |
3168 | } | |
3169 | ||
9bac3d6d | 3170 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 3171 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 3172 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
3173 | |
3174 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
3175 | * bpf_ld_imm64 instructions | |
3176 | */ | |
3177 | convert_pseudo_ld_imm64(env); | |
3178 | } | |
cbd35700 AS |
3179 | |
3180 | free_log_buf: | |
3181 | if (log_level) | |
3182 | vfree(log_buf); | |
9bac3d6d | 3183 | if (!env->prog->aux->used_maps) |
0246e64d AS |
3184 | /* if we didn't copy map pointers into bpf_prog_info, release |
3185 | * them now. Otherwise free_bpf_prog_info() will release them. | |
3186 | */ | |
3187 | release_maps(env); | |
9bac3d6d | 3188 | *prog = env->prog; |
3df126f3 | 3189 | err_unlock: |
cbd35700 | 3190 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
3191 | vfree(env->insn_aux_data); |
3192 | err_free_env: | |
3193 | kfree(env); | |
51580e79 AS |
3194 | return ret; |
3195 | } | |
13a27dfc JK |
3196 | |
3197 | int bpf_analyzer(struct bpf_prog *prog, const struct bpf_ext_analyzer_ops *ops, | |
3198 | void *priv) | |
3199 | { | |
3200 | struct bpf_verifier_env *env; | |
3201 | int ret; | |
3202 | ||
3203 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); | |
3204 | if (!env) | |
3205 | return -ENOMEM; | |
3206 | ||
3207 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * | |
3208 | prog->len); | |
3209 | ret = -ENOMEM; | |
3210 | if (!env->insn_aux_data) | |
3211 | goto err_free_env; | |
3212 | env->prog = prog; | |
3213 | env->analyzer_ops = ops; | |
3214 | env->analyzer_priv = priv; | |
3215 | ||
3216 | /* grab the mutex to protect few globals used by verifier */ | |
3217 | mutex_lock(&bpf_verifier_lock); | |
3218 | ||
3219 | log_level = 0; | |
3220 | ||
3221 | env->explored_states = kcalloc(env->prog->len, | |
3222 | sizeof(struct bpf_verifier_state_list *), | |
3223 | GFP_KERNEL); | |
3224 | ret = -ENOMEM; | |
3225 | if (!env->explored_states) | |
3226 | goto skip_full_check; | |
3227 | ||
3228 | ret = check_cfg(env); | |
3229 | if (ret < 0) | |
3230 | goto skip_full_check; | |
3231 | ||
3232 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); | |
3233 | ||
3234 | ret = do_check(env); | |
3235 | ||
3236 | skip_full_check: | |
3237 | while (pop_stack(env, NULL) >= 0); | |
3238 | free_states(env); | |
3239 | ||
3240 | mutex_unlock(&bpf_verifier_lock); | |
3241 | vfree(env->insn_aux_data); | |
3242 | err_free_env: | |
3243 | kfree(env); | |
3244 | return ret; | |
3245 | } | |
3246 | EXPORT_SYMBOL_GPL(bpf_analyzer); |