]>
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", | |
185 | [FRAME_PTR] = "fp", | |
186 | [PTR_TO_STACK] = "fp", | |
187 | [CONST_IMM] = "imm", | |
969bf05e AS |
188 | [PTR_TO_PACKET] = "pkt", |
189 | [PTR_TO_PACKET_END] = "pkt_end", | |
17a52670 AS |
190 | }; |
191 | ||
58e2af8b | 192 | static void print_verifier_state(struct bpf_verifier_state *state) |
17a52670 | 193 | { |
58e2af8b | 194 | struct bpf_reg_state *reg; |
17a52670 AS |
195 | enum bpf_reg_type t; |
196 | int i; | |
197 | ||
198 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
199 | reg = &state->regs[i]; |
200 | t = reg->type; | |
17a52670 AS |
201 | if (t == NOT_INIT) |
202 | continue; | |
203 | verbose(" R%d=%s", i, reg_type_str[t]); | |
204 | if (t == CONST_IMM || t == PTR_TO_STACK) | |
969bf05e AS |
205 | verbose("%lld", reg->imm); |
206 | else if (t == PTR_TO_PACKET) | |
207 | verbose("(id=%d,off=%d,r=%d)", | |
208 | reg->id, reg->off, reg->range); | |
209 | else if (t == UNKNOWN_VALUE && reg->imm) | |
210 | verbose("%lld", reg->imm); | |
17a52670 AS |
211 | else if (t == CONST_PTR_TO_MAP || t == PTR_TO_MAP_VALUE || |
212 | t == PTR_TO_MAP_VALUE_OR_NULL) | |
213 | verbose("(ks=%d,vs=%d)", | |
1a0dc1ac AS |
214 | reg->map_ptr->key_size, |
215 | reg->map_ptr->value_size); | |
17a52670 | 216 | } |
9c399760 | 217 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { |
1a0dc1ac | 218 | if (state->stack_slot_type[i] == STACK_SPILL) |
17a52670 | 219 | verbose(" fp%d=%s", -MAX_BPF_STACK + i, |
1a0dc1ac | 220 | reg_type_str[state->spilled_regs[i / BPF_REG_SIZE].type]); |
17a52670 AS |
221 | } |
222 | verbose("\n"); | |
223 | } | |
224 | ||
cbd35700 AS |
225 | static const char *const bpf_class_string[] = { |
226 | [BPF_LD] = "ld", | |
227 | [BPF_LDX] = "ldx", | |
228 | [BPF_ST] = "st", | |
229 | [BPF_STX] = "stx", | |
230 | [BPF_ALU] = "alu", | |
231 | [BPF_JMP] = "jmp", | |
232 | [BPF_RET] = "BUG", | |
233 | [BPF_ALU64] = "alu64", | |
234 | }; | |
235 | ||
687f0715 | 236 | static const char *const bpf_alu_string[16] = { |
cbd35700 AS |
237 | [BPF_ADD >> 4] = "+=", |
238 | [BPF_SUB >> 4] = "-=", | |
239 | [BPF_MUL >> 4] = "*=", | |
240 | [BPF_DIV >> 4] = "/=", | |
241 | [BPF_OR >> 4] = "|=", | |
242 | [BPF_AND >> 4] = "&=", | |
243 | [BPF_LSH >> 4] = "<<=", | |
244 | [BPF_RSH >> 4] = ">>=", | |
245 | [BPF_NEG >> 4] = "neg", | |
246 | [BPF_MOD >> 4] = "%=", | |
247 | [BPF_XOR >> 4] = "^=", | |
248 | [BPF_MOV >> 4] = "=", | |
249 | [BPF_ARSH >> 4] = "s>>=", | |
250 | [BPF_END >> 4] = "endian", | |
251 | }; | |
252 | ||
253 | static const char *const bpf_ldst_string[] = { | |
254 | [BPF_W >> 3] = "u32", | |
255 | [BPF_H >> 3] = "u16", | |
256 | [BPF_B >> 3] = "u8", | |
257 | [BPF_DW >> 3] = "u64", | |
258 | }; | |
259 | ||
687f0715 | 260 | static const char *const bpf_jmp_string[16] = { |
cbd35700 AS |
261 | [BPF_JA >> 4] = "jmp", |
262 | [BPF_JEQ >> 4] = "==", | |
263 | [BPF_JGT >> 4] = ">", | |
264 | [BPF_JGE >> 4] = ">=", | |
265 | [BPF_JSET >> 4] = "&", | |
266 | [BPF_JNE >> 4] = "!=", | |
267 | [BPF_JSGT >> 4] = "s>", | |
268 | [BPF_JSGE >> 4] = "s>=", | |
269 | [BPF_CALL >> 4] = "call", | |
270 | [BPF_EXIT >> 4] = "exit", | |
271 | }; | |
272 | ||
273 | static void print_bpf_insn(struct bpf_insn *insn) | |
274 | { | |
275 | u8 class = BPF_CLASS(insn->code); | |
276 | ||
277 | if (class == BPF_ALU || class == BPF_ALU64) { | |
278 | if (BPF_SRC(insn->code) == BPF_X) | |
279 | verbose("(%02x) %sr%d %s %sr%d\n", | |
280 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
281 | insn->dst_reg, | |
282 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
283 | class == BPF_ALU ? "(u32) " : "", | |
284 | insn->src_reg); | |
285 | else | |
286 | verbose("(%02x) %sr%d %s %s%d\n", | |
287 | insn->code, class == BPF_ALU ? "(u32) " : "", | |
288 | insn->dst_reg, | |
289 | bpf_alu_string[BPF_OP(insn->code) >> 4], | |
290 | class == BPF_ALU ? "(u32) " : "", | |
291 | insn->imm); | |
292 | } else if (class == BPF_STX) { | |
293 | if (BPF_MODE(insn->code) == BPF_MEM) | |
294 | verbose("(%02x) *(%s *)(r%d %+d) = r%d\n", | |
295 | insn->code, | |
296 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
297 | insn->dst_reg, | |
298 | insn->off, insn->src_reg); | |
299 | else if (BPF_MODE(insn->code) == BPF_XADD) | |
300 | verbose("(%02x) lock *(%s *)(r%d %+d) += r%d\n", | |
301 | insn->code, | |
302 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
303 | insn->dst_reg, insn->off, | |
304 | insn->src_reg); | |
305 | else | |
306 | verbose("BUG_%02x\n", insn->code); | |
307 | } else if (class == BPF_ST) { | |
308 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
309 | verbose("BUG_st_%02x\n", insn->code); | |
310 | return; | |
311 | } | |
312 | verbose("(%02x) *(%s *)(r%d %+d) = %d\n", | |
313 | insn->code, | |
314 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
315 | insn->dst_reg, | |
316 | insn->off, insn->imm); | |
317 | } else if (class == BPF_LDX) { | |
318 | if (BPF_MODE(insn->code) != BPF_MEM) { | |
319 | verbose("BUG_ldx_%02x\n", insn->code); | |
320 | return; | |
321 | } | |
322 | verbose("(%02x) r%d = *(%s *)(r%d %+d)\n", | |
323 | insn->code, insn->dst_reg, | |
324 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
325 | insn->src_reg, insn->off); | |
326 | } else if (class == BPF_LD) { | |
327 | if (BPF_MODE(insn->code) == BPF_ABS) { | |
328 | verbose("(%02x) r0 = *(%s *)skb[%d]\n", | |
329 | insn->code, | |
330 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
331 | insn->imm); | |
332 | } else if (BPF_MODE(insn->code) == BPF_IND) { | |
333 | verbose("(%02x) r0 = *(%s *)skb[r%d + %d]\n", | |
334 | insn->code, | |
335 | bpf_ldst_string[BPF_SIZE(insn->code) >> 3], | |
336 | insn->src_reg, insn->imm); | |
337 | } else if (BPF_MODE(insn->code) == BPF_IMM) { | |
338 | verbose("(%02x) r%d = 0x%x\n", | |
339 | insn->code, insn->dst_reg, insn->imm); | |
340 | } else { | |
341 | verbose("BUG_ld_%02x\n", insn->code); | |
342 | return; | |
343 | } | |
344 | } else if (class == BPF_JMP) { | |
345 | u8 opcode = BPF_OP(insn->code); | |
346 | ||
347 | if (opcode == BPF_CALL) { | |
348 | verbose("(%02x) call %d\n", insn->code, insn->imm); | |
349 | } else if (insn->code == (BPF_JMP | BPF_JA)) { | |
350 | verbose("(%02x) goto pc%+d\n", | |
351 | insn->code, insn->off); | |
352 | } else if (insn->code == (BPF_JMP | BPF_EXIT)) { | |
353 | verbose("(%02x) exit\n", insn->code); | |
354 | } else if (BPF_SRC(insn->code) == BPF_X) { | |
355 | verbose("(%02x) if r%d %s r%d goto pc%+d\n", | |
356 | insn->code, insn->dst_reg, | |
357 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
358 | insn->src_reg, insn->off); | |
359 | } else { | |
360 | verbose("(%02x) if r%d %s 0x%x goto pc%+d\n", | |
361 | insn->code, insn->dst_reg, | |
362 | bpf_jmp_string[BPF_OP(insn->code) >> 4], | |
363 | insn->imm, insn->off); | |
364 | } | |
365 | } else { | |
366 | verbose("(%02x) %s\n", insn->code, bpf_class_string[class]); | |
367 | } | |
368 | } | |
369 | ||
58e2af8b | 370 | static int pop_stack(struct bpf_verifier_env *env, int *prev_insn_idx) |
17a52670 | 371 | { |
58e2af8b | 372 | struct bpf_verifier_stack_elem *elem; |
17a52670 AS |
373 | int insn_idx; |
374 | ||
375 | if (env->head == NULL) | |
376 | return -1; | |
377 | ||
378 | memcpy(&env->cur_state, &env->head->st, sizeof(env->cur_state)); | |
379 | insn_idx = env->head->insn_idx; | |
380 | if (prev_insn_idx) | |
381 | *prev_insn_idx = env->head->prev_insn_idx; | |
382 | elem = env->head->next; | |
383 | kfree(env->head); | |
384 | env->head = elem; | |
385 | env->stack_size--; | |
386 | return insn_idx; | |
387 | } | |
388 | ||
58e2af8b JK |
389 | static struct bpf_verifier_state *push_stack(struct bpf_verifier_env *env, |
390 | int insn_idx, int prev_insn_idx) | |
17a52670 | 391 | { |
58e2af8b | 392 | struct bpf_verifier_stack_elem *elem; |
17a52670 | 393 | |
58e2af8b | 394 | elem = kmalloc(sizeof(struct bpf_verifier_stack_elem), GFP_KERNEL); |
17a52670 AS |
395 | if (!elem) |
396 | goto err; | |
397 | ||
398 | memcpy(&elem->st, &env->cur_state, sizeof(env->cur_state)); | |
399 | elem->insn_idx = insn_idx; | |
400 | elem->prev_insn_idx = prev_insn_idx; | |
401 | elem->next = env->head; | |
402 | env->head = elem; | |
403 | env->stack_size++; | |
07016151 | 404 | if (env->stack_size > BPF_COMPLEXITY_LIMIT_STACK) { |
17a52670 AS |
405 | verbose("BPF program is too complex\n"); |
406 | goto err; | |
407 | } | |
408 | return &elem->st; | |
409 | err: | |
410 | /* pop all elements and return */ | |
411 | while (pop_stack(env, NULL) >= 0); | |
412 | return NULL; | |
413 | } | |
414 | ||
415 | #define CALLER_SAVED_REGS 6 | |
416 | static const int caller_saved[CALLER_SAVED_REGS] = { | |
417 | BPF_REG_0, BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4, BPF_REG_5 | |
418 | }; | |
419 | ||
58e2af8b | 420 | static void init_reg_state(struct bpf_reg_state *regs) |
17a52670 AS |
421 | { |
422 | int i; | |
423 | ||
424 | for (i = 0; i < MAX_BPF_REG; i++) { | |
425 | regs[i].type = NOT_INIT; | |
426 | regs[i].imm = 0; | |
17a52670 AS |
427 | } |
428 | ||
429 | /* frame pointer */ | |
430 | regs[BPF_REG_FP].type = FRAME_PTR; | |
431 | ||
432 | /* 1st arg to a function */ | |
433 | regs[BPF_REG_1].type = PTR_TO_CTX; | |
434 | } | |
435 | ||
58e2af8b | 436 | static void mark_reg_unknown_value(struct bpf_reg_state *regs, u32 regno) |
17a52670 AS |
437 | { |
438 | BUG_ON(regno >= MAX_BPF_REG); | |
439 | regs[regno].type = UNKNOWN_VALUE; | |
440 | regs[regno].imm = 0; | |
17a52670 AS |
441 | } |
442 | ||
443 | enum reg_arg_type { | |
444 | SRC_OP, /* register is used as source operand */ | |
445 | DST_OP, /* register is used as destination operand */ | |
446 | DST_OP_NO_MARK /* same as above, check only, don't mark */ | |
447 | }; | |
448 | ||
58e2af8b | 449 | static int check_reg_arg(struct bpf_reg_state *regs, u32 regno, |
17a52670 AS |
450 | enum reg_arg_type t) |
451 | { | |
452 | if (regno >= MAX_BPF_REG) { | |
453 | verbose("R%d is invalid\n", regno); | |
454 | return -EINVAL; | |
455 | } | |
456 | ||
457 | if (t == SRC_OP) { | |
458 | /* check whether register used as source operand can be read */ | |
459 | if (regs[regno].type == NOT_INIT) { | |
460 | verbose("R%d !read_ok\n", regno); | |
461 | return -EACCES; | |
462 | } | |
463 | } else { | |
464 | /* check whether register used as dest operand can be written to */ | |
465 | if (regno == BPF_REG_FP) { | |
466 | verbose("frame pointer is read only\n"); | |
467 | return -EACCES; | |
468 | } | |
469 | if (t == DST_OP) | |
470 | mark_reg_unknown_value(regs, regno); | |
471 | } | |
472 | return 0; | |
473 | } | |
474 | ||
475 | static int bpf_size_to_bytes(int bpf_size) | |
476 | { | |
477 | if (bpf_size == BPF_W) | |
478 | return 4; | |
479 | else if (bpf_size == BPF_H) | |
480 | return 2; | |
481 | else if (bpf_size == BPF_B) | |
482 | return 1; | |
483 | else if (bpf_size == BPF_DW) | |
484 | return 8; | |
485 | else | |
486 | return -EINVAL; | |
487 | } | |
488 | ||
1be7f75d AS |
489 | static bool is_spillable_regtype(enum bpf_reg_type type) |
490 | { | |
491 | switch (type) { | |
492 | case PTR_TO_MAP_VALUE: | |
493 | case PTR_TO_MAP_VALUE_OR_NULL: | |
494 | case PTR_TO_STACK: | |
495 | case PTR_TO_CTX: | |
969bf05e AS |
496 | case PTR_TO_PACKET: |
497 | case PTR_TO_PACKET_END: | |
1be7f75d AS |
498 | case FRAME_PTR: |
499 | case CONST_PTR_TO_MAP: | |
500 | return true; | |
501 | default: | |
502 | return false; | |
503 | } | |
504 | } | |
505 | ||
17a52670 AS |
506 | /* check_stack_read/write functions track spill/fill of registers, |
507 | * stack boundary and alignment are checked in check_mem_access() | |
508 | */ | |
58e2af8b JK |
509 | static int check_stack_write(struct bpf_verifier_state *state, int off, |
510 | int size, int value_regno) | |
17a52670 | 511 | { |
17a52670 | 512 | int i; |
9c399760 AS |
513 | /* caller checked that off % size == 0 and -MAX_BPF_STACK <= off < 0, |
514 | * so it's aligned access and [off, off + size) are within stack limits | |
515 | */ | |
17a52670 AS |
516 | |
517 | if (value_regno >= 0 && | |
1be7f75d | 518 | is_spillable_regtype(state->regs[value_regno].type)) { |
17a52670 AS |
519 | |
520 | /* register containing pointer is being spilled into stack */ | |
9c399760 | 521 | if (size != BPF_REG_SIZE) { |
17a52670 AS |
522 | verbose("invalid size of register spill\n"); |
523 | return -EACCES; | |
524 | } | |
525 | ||
17a52670 | 526 | /* save register state */ |
9c399760 AS |
527 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
528 | state->regs[value_regno]; | |
17a52670 | 529 | |
9c399760 AS |
530 | for (i = 0; i < BPF_REG_SIZE; i++) |
531 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_SPILL; | |
532 | } else { | |
17a52670 | 533 | /* regular write of data into stack */ |
9c399760 | 534 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE] = |
58e2af8b | 535 | (struct bpf_reg_state) {}; |
9c399760 AS |
536 | |
537 | for (i = 0; i < size; i++) | |
538 | state->stack_slot_type[MAX_BPF_STACK + off + i] = STACK_MISC; | |
17a52670 AS |
539 | } |
540 | return 0; | |
541 | } | |
542 | ||
58e2af8b | 543 | static int check_stack_read(struct bpf_verifier_state *state, int off, int size, |
17a52670 AS |
544 | int value_regno) |
545 | { | |
9c399760 | 546 | u8 *slot_type; |
17a52670 | 547 | int i; |
17a52670 | 548 | |
9c399760 | 549 | slot_type = &state->stack_slot_type[MAX_BPF_STACK + off]; |
17a52670 | 550 | |
9c399760 AS |
551 | if (slot_type[0] == STACK_SPILL) { |
552 | if (size != BPF_REG_SIZE) { | |
17a52670 AS |
553 | verbose("invalid size of register spill\n"); |
554 | return -EACCES; | |
555 | } | |
9c399760 AS |
556 | for (i = 1; i < BPF_REG_SIZE; i++) { |
557 | if (slot_type[i] != STACK_SPILL) { | |
17a52670 AS |
558 | verbose("corrupted spill memory\n"); |
559 | return -EACCES; | |
560 | } | |
561 | } | |
562 | ||
563 | if (value_regno >= 0) | |
564 | /* restore register state from stack */ | |
9c399760 AS |
565 | state->regs[value_regno] = |
566 | state->spilled_regs[(MAX_BPF_STACK + off) / BPF_REG_SIZE]; | |
17a52670 AS |
567 | return 0; |
568 | } else { | |
569 | for (i = 0; i < size; i++) { | |
9c399760 | 570 | if (slot_type[i] != STACK_MISC) { |
17a52670 AS |
571 | verbose("invalid read from stack off %d+%d size %d\n", |
572 | off, i, size); | |
573 | return -EACCES; | |
574 | } | |
575 | } | |
576 | if (value_regno >= 0) | |
577 | /* have read misc data from the stack */ | |
578 | mark_reg_unknown_value(state->regs, value_regno); | |
579 | return 0; | |
580 | } | |
581 | } | |
582 | ||
583 | /* check read/write into map element returned by bpf_map_lookup_elem() */ | |
58e2af8b | 584 | static int check_map_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
585 | int size) |
586 | { | |
587 | struct bpf_map *map = env->cur_state.regs[regno].map_ptr; | |
588 | ||
589 | if (off < 0 || off + size > map->value_size) { | |
590 | verbose("invalid access to map value, value_size=%d off=%d size=%d\n", | |
591 | map->value_size, off, size); | |
592 | return -EACCES; | |
593 | } | |
594 | return 0; | |
595 | } | |
596 | ||
969bf05e AS |
597 | #define MAX_PACKET_OFF 0xffff |
598 | ||
58e2af8b | 599 | static bool may_access_direct_pkt_data(struct bpf_verifier_env *env, |
36bbef52 | 600 | const struct bpf_call_arg_meta *meta) |
4acf6c0b | 601 | { |
36bbef52 DB |
602 | switch (env->prog->type) { |
603 | case BPF_PROG_TYPE_SCHED_CLS: | |
604 | case BPF_PROG_TYPE_SCHED_ACT: | |
4acf6c0b | 605 | case BPF_PROG_TYPE_XDP: |
36bbef52 DB |
606 | if (meta) |
607 | return meta->pkt_access; | |
608 | ||
609 | env->seen_direct_write = true; | |
4acf6c0b BB |
610 | return true; |
611 | default: | |
612 | return false; | |
613 | } | |
614 | } | |
615 | ||
58e2af8b | 616 | static int check_packet_access(struct bpf_verifier_env *env, u32 regno, int off, |
969bf05e AS |
617 | int size) |
618 | { | |
58e2af8b JK |
619 | struct bpf_reg_state *regs = env->cur_state.regs; |
620 | struct bpf_reg_state *reg = ®s[regno]; | |
969bf05e | 621 | |
d91b28ed | 622 | off += reg->off; |
b399cf64 | 623 | if (off < 0 || size <= 0 || off + size > reg->range) { |
d91b28ed AS |
624 | verbose("invalid access to packet, off=%d size=%d, R%d(id=%d,off=%d,r=%d)\n", |
625 | off, size, regno, reg->id, reg->off, reg->range); | |
969bf05e AS |
626 | return -EACCES; |
627 | } | |
628 | return 0; | |
629 | } | |
630 | ||
17a52670 | 631 | /* check access to 'struct bpf_context' fields */ |
58e2af8b | 632 | static int check_ctx_access(struct bpf_verifier_env *env, int off, int size, |
19de99f7 | 633 | enum bpf_access_type t, enum bpf_reg_type *reg_type) |
17a52670 AS |
634 | { |
635 | if (env->prog->aux->ops->is_valid_access && | |
19de99f7 | 636 | env->prog->aux->ops->is_valid_access(off, size, t, reg_type)) { |
32bbe007 AS |
637 | /* remember the offset of last byte accessed in ctx */ |
638 | if (env->prog->aux->max_ctx_offset < off + size) | |
639 | env->prog->aux->max_ctx_offset = off + size; | |
17a52670 | 640 | return 0; |
32bbe007 | 641 | } |
17a52670 AS |
642 | |
643 | verbose("invalid bpf_context access off=%d size=%d\n", off, size); | |
644 | return -EACCES; | |
645 | } | |
646 | ||
58e2af8b | 647 | static bool is_pointer_value(struct bpf_verifier_env *env, int regno) |
1be7f75d AS |
648 | { |
649 | if (env->allow_ptr_leaks) | |
650 | return false; | |
651 | ||
652 | switch (env->cur_state.regs[regno].type) { | |
653 | case UNKNOWN_VALUE: | |
654 | case CONST_IMM: | |
655 | return false; | |
656 | default: | |
657 | return true; | |
658 | } | |
659 | } | |
660 | ||
58e2af8b JK |
661 | static int check_ptr_alignment(struct bpf_verifier_env *env, |
662 | struct bpf_reg_state *reg, int off, int size) | |
969bf05e AS |
663 | { |
664 | if (reg->type != PTR_TO_PACKET) { | |
665 | if (off % size != 0) { | |
58e2af8b JK |
666 | verbose("misaligned access off %d size %d\n", |
667 | off, size); | |
969bf05e AS |
668 | return -EACCES; |
669 | } else { | |
670 | return 0; | |
671 | } | |
672 | } | |
673 | ||
674 | switch (env->prog->type) { | |
675 | case BPF_PROG_TYPE_SCHED_CLS: | |
676 | case BPF_PROG_TYPE_SCHED_ACT: | |
6a773a15 | 677 | case BPF_PROG_TYPE_XDP: |
969bf05e AS |
678 | break; |
679 | default: | |
680 | verbose("verifier is misconfigured\n"); | |
681 | return -EACCES; | |
682 | } | |
683 | ||
684 | if (IS_ENABLED(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS)) | |
685 | /* misaligned access to packet is ok on x86,arm,arm64 */ | |
686 | return 0; | |
687 | ||
688 | if (reg->id && size != 1) { | |
689 | verbose("Unknown packet alignment. Only byte-sized access allowed\n"); | |
690 | return -EACCES; | |
691 | } | |
692 | ||
693 | /* skb->data is NET_IP_ALIGN-ed */ | |
694 | if ((NET_IP_ALIGN + reg->off + off) % size != 0) { | |
695 | verbose("misaligned packet access off %d+%d+%d size %d\n", | |
696 | NET_IP_ALIGN, reg->off, off, size); | |
697 | return -EACCES; | |
698 | } | |
699 | return 0; | |
700 | } | |
701 | ||
17a52670 AS |
702 | /* check whether memory at (regno + off) is accessible for t = (read | write) |
703 | * if t==write, value_regno is a register which value is stored into memory | |
704 | * if t==read, value_regno is a register which will receive the value from memory | |
705 | * if t==write && value_regno==-1, some unknown value is stored into memory | |
706 | * if t==read && value_regno==-1, don't care what we read from memory | |
707 | */ | |
58e2af8b | 708 | static int check_mem_access(struct bpf_verifier_env *env, u32 regno, int off, |
17a52670 AS |
709 | int bpf_size, enum bpf_access_type t, |
710 | int value_regno) | |
711 | { | |
58e2af8b JK |
712 | struct bpf_verifier_state *state = &env->cur_state; |
713 | struct bpf_reg_state *reg = &state->regs[regno]; | |
17a52670 AS |
714 | int size, err = 0; |
715 | ||
1a0dc1ac AS |
716 | if (reg->type == PTR_TO_STACK) |
717 | off += reg->imm; | |
24b4d2ab | 718 | |
17a52670 AS |
719 | size = bpf_size_to_bytes(bpf_size); |
720 | if (size < 0) | |
721 | return size; | |
722 | ||
969bf05e AS |
723 | err = check_ptr_alignment(env, reg, off, size); |
724 | if (err) | |
725 | return err; | |
17a52670 | 726 | |
1a0dc1ac | 727 | if (reg->type == PTR_TO_MAP_VALUE) { |
1be7f75d AS |
728 | if (t == BPF_WRITE && value_regno >= 0 && |
729 | is_pointer_value(env, value_regno)) { | |
730 | verbose("R%d leaks addr into map\n", value_regno); | |
731 | return -EACCES; | |
732 | } | |
17a52670 AS |
733 | err = check_map_access(env, regno, off, size); |
734 | if (!err && t == BPF_READ && value_regno >= 0) | |
735 | mark_reg_unknown_value(state->regs, value_regno); | |
736 | ||
1a0dc1ac | 737 | } else if (reg->type == PTR_TO_CTX) { |
19de99f7 AS |
738 | enum bpf_reg_type reg_type = UNKNOWN_VALUE; |
739 | ||
1be7f75d AS |
740 | if (t == BPF_WRITE && value_regno >= 0 && |
741 | is_pointer_value(env, value_regno)) { | |
742 | verbose("R%d leaks addr into ctx\n", value_regno); | |
743 | return -EACCES; | |
744 | } | |
19de99f7 | 745 | err = check_ctx_access(env, off, size, t, ®_type); |
969bf05e | 746 | if (!err && t == BPF_READ && value_regno >= 0) { |
17a52670 | 747 | mark_reg_unknown_value(state->regs, value_regno); |
19de99f7 | 748 | if (env->allow_ptr_leaks) |
969bf05e | 749 | /* note that reg.[id|off|range] == 0 */ |
19de99f7 | 750 | state->regs[value_regno].type = reg_type; |
969bf05e | 751 | } |
17a52670 | 752 | |
1a0dc1ac | 753 | } else if (reg->type == FRAME_PTR || reg->type == PTR_TO_STACK) { |
17a52670 AS |
754 | if (off >= 0 || off < -MAX_BPF_STACK) { |
755 | verbose("invalid stack off=%d size=%d\n", off, size); | |
756 | return -EACCES; | |
757 | } | |
1be7f75d AS |
758 | if (t == BPF_WRITE) { |
759 | if (!env->allow_ptr_leaks && | |
760 | state->stack_slot_type[MAX_BPF_STACK + off] == STACK_SPILL && | |
761 | size != BPF_REG_SIZE) { | |
762 | verbose("attempt to corrupt spilled pointer on stack\n"); | |
763 | return -EACCES; | |
764 | } | |
17a52670 | 765 | err = check_stack_write(state, off, size, value_regno); |
1be7f75d | 766 | } else { |
17a52670 | 767 | err = check_stack_read(state, off, size, value_regno); |
1be7f75d | 768 | } |
969bf05e | 769 | } else if (state->regs[regno].type == PTR_TO_PACKET) { |
36bbef52 | 770 | if (t == BPF_WRITE && !may_access_direct_pkt_data(env, NULL)) { |
969bf05e AS |
771 | verbose("cannot write into packet\n"); |
772 | return -EACCES; | |
773 | } | |
4acf6c0b BB |
774 | if (t == BPF_WRITE && value_regno >= 0 && |
775 | is_pointer_value(env, value_regno)) { | |
776 | verbose("R%d leaks addr into packet\n", value_regno); | |
777 | return -EACCES; | |
778 | } | |
969bf05e AS |
779 | err = check_packet_access(env, regno, off, size); |
780 | if (!err && t == BPF_READ && value_regno >= 0) | |
781 | mark_reg_unknown_value(state->regs, value_regno); | |
17a52670 AS |
782 | } else { |
783 | verbose("R%d invalid mem access '%s'\n", | |
1a0dc1ac | 784 | regno, reg_type_str[reg->type]); |
17a52670 AS |
785 | return -EACCES; |
786 | } | |
969bf05e AS |
787 | |
788 | if (!err && size <= 2 && value_regno >= 0 && env->allow_ptr_leaks && | |
789 | state->regs[value_regno].type == UNKNOWN_VALUE) { | |
790 | /* 1 or 2 byte load zero-extends, determine the number of | |
791 | * zero upper bits. Not doing it fo 4 byte load, since | |
792 | * such values cannot be added to ptr_to_packet anyway. | |
793 | */ | |
794 | state->regs[value_regno].imm = 64 - size * 8; | |
795 | } | |
17a52670 AS |
796 | return err; |
797 | } | |
798 | ||
58e2af8b | 799 | static int check_xadd(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 800 | { |
58e2af8b | 801 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
802 | int err; |
803 | ||
804 | if ((BPF_SIZE(insn->code) != BPF_W && BPF_SIZE(insn->code) != BPF_DW) || | |
805 | insn->imm != 0) { | |
806 | verbose("BPF_XADD uses reserved fields\n"); | |
807 | return -EINVAL; | |
808 | } | |
809 | ||
810 | /* check src1 operand */ | |
811 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
812 | if (err) | |
813 | return err; | |
814 | ||
815 | /* check src2 operand */ | |
816 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
817 | if (err) | |
818 | return err; | |
819 | ||
820 | /* check whether atomic_add can read the memory */ | |
821 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
822 | BPF_SIZE(insn->code), BPF_READ, -1); | |
823 | if (err) | |
824 | return err; | |
825 | ||
826 | /* check whether atomic_add can write into the same memory */ | |
827 | return check_mem_access(env, insn->dst_reg, insn->off, | |
828 | BPF_SIZE(insn->code), BPF_WRITE, -1); | |
829 | } | |
830 | ||
831 | /* when register 'regno' is passed into function that will read 'access_size' | |
832 | * bytes from that pointer, make sure that it's within stack boundary | |
833 | * and all elements of stack are initialized | |
834 | */ | |
58e2af8b | 835 | static int check_stack_boundary(struct bpf_verifier_env *env, int regno, |
435faee1 DB |
836 | int access_size, bool zero_size_allowed, |
837 | struct bpf_call_arg_meta *meta) | |
17a52670 | 838 | { |
58e2af8b JK |
839 | struct bpf_verifier_state *state = &env->cur_state; |
840 | struct bpf_reg_state *regs = state->regs; | |
17a52670 AS |
841 | int off, i; |
842 | ||
8e2fe1d9 DB |
843 | if (regs[regno].type != PTR_TO_STACK) { |
844 | if (zero_size_allowed && access_size == 0 && | |
845 | regs[regno].type == CONST_IMM && | |
846 | regs[regno].imm == 0) | |
847 | return 0; | |
848 | ||
849 | verbose("R%d type=%s expected=%s\n", regno, | |
850 | reg_type_str[regs[regno].type], | |
851 | reg_type_str[PTR_TO_STACK]); | |
17a52670 | 852 | return -EACCES; |
8e2fe1d9 | 853 | } |
17a52670 AS |
854 | |
855 | off = regs[regno].imm; | |
856 | if (off >= 0 || off < -MAX_BPF_STACK || off + access_size > 0 || | |
857 | access_size <= 0) { | |
858 | verbose("invalid stack type R%d off=%d access_size=%d\n", | |
859 | regno, off, access_size); | |
860 | return -EACCES; | |
861 | } | |
862 | ||
435faee1 DB |
863 | if (meta && meta->raw_mode) { |
864 | meta->access_size = access_size; | |
865 | meta->regno = regno; | |
866 | return 0; | |
867 | } | |
868 | ||
17a52670 | 869 | for (i = 0; i < access_size; i++) { |
9c399760 | 870 | if (state->stack_slot_type[MAX_BPF_STACK + off + i] != STACK_MISC) { |
17a52670 AS |
871 | verbose("invalid indirect read from stack off %d+%d size %d\n", |
872 | off, i, access_size); | |
873 | return -EACCES; | |
874 | } | |
875 | } | |
876 | return 0; | |
877 | } | |
878 | ||
58e2af8b | 879 | static int check_func_arg(struct bpf_verifier_env *env, u32 regno, |
33ff9823 DB |
880 | enum bpf_arg_type arg_type, |
881 | struct bpf_call_arg_meta *meta) | |
17a52670 | 882 | { |
58e2af8b | 883 | struct bpf_reg_state *regs = env->cur_state.regs, *reg = ®s[regno]; |
6841de8b | 884 | enum bpf_reg_type expected_type, type = reg->type; |
17a52670 AS |
885 | int err = 0; |
886 | ||
80f1d68c | 887 | if (arg_type == ARG_DONTCARE) |
17a52670 AS |
888 | return 0; |
889 | ||
6841de8b | 890 | if (type == NOT_INIT) { |
17a52670 AS |
891 | verbose("R%d !read_ok\n", regno); |
892 | return -EACCES; | |
893 | } | |
894 | ||
1be7f75d AS |
895 | if (arg_type == ARG_ANYTHING) { |
896 | if (is_pointer_value(env, regno)) { | |
897 | verbose("R%d leaks addr into helper function\n", regno); | |
898 | return -EACCES; | |
899 | } | |
80f1d68c | 900 | return 0; |
1be7f75d | 901 | } |
80f1d68c | 902 | |
36bbef52 DB |
903 | if (type == PTR_TO_PACKET && !may_access_direct_pkt_data(env, meta)) { |
904 | verbose("helper access to the packet is not allowed\n"); | |
6841de8b AS |
905 | return -EACCES; |
906 | } | |
907 | ||
8e2fe1d9 | 908 | if (arg_type == ARG_PTR_TO_MAP_KEY || |
17a52670 AS |
909 | arg_type == ARG_PTR_TO_MAP_VALUE) { |
910 | expected_type = PTR_TO_STACK; | |
6841de8b AS |
911 | if (type != PTR_TO_PACKET && type != expected_type) |
912 | goto err_type; | |
8e2fe1d9 DB |
913 | } else if (arg_type == ARG_CONST_STACK_SIZE || |
914 | arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { | |
17a52670 | 915 | expected_type = CONST_IMM; |
6841de8b AS |
916 | if (type != expected_type) |
917 | goto err_type; | |
17a52670 AS |
918 | } else if (arg_type == ARG_CONST_MAP_PTR) { |
919 | expected_type = CONST_PTR_TO_MAP; | |
6841de8b AS |
920 | if (type != expected_type) |
921 | goto err_type; | |
608cd71a AS |
922 | } else if (arg_type == ARG_PTR_TO_CTX) { |
923 | expected_type = PTR_TO_CTX; | |
6841de8b AS |
924 | if (type != expected_type) |
925 | goto err_type; | |
435faee1 DB |
926 | } else if (arg_type == ARG_PTR_TO_STACK || |
927 | arg_type == ARG_PTR_TO_RAW_STACK) { | |
8e2fe1d9 DB |
928 | expected_type = PTR_TO_STACK; |
929 | /* One exception here. In case function allows for NULL to be | |
930 | * passed in as argument, it's a CONST_IMM type. Final test | |
931 | * happens during stack boundary checking. | |
932 | */ | |
6841de8b AS |
933 | if (type == CONST_IMM && reg->imm == 0) |
934 | /* final test in check_stack_boundary() */; | |
935 | else if (type != PTR_TO_PACKET && type != expected_type) | |
936 | goto err_type; | |
435faee1 | 937 | meta->raw_mode = arg_type == ARG_PTR_TO_RAW_STACK; |
17a52670 AS |
938 | } else { |
939 | verbose("unsupported arg_type %d\n", arg_type); | |
940 | return -EFAULT; | |
941 | } | |
942 | ||
17a52670 AS |
943 | if (arg_type == ARG_CONST_MAP_PTR) { |
944 | /* bpf_map_xxx(map_ptr) call: remember that map_ptr */ | |
33ff9823 | 945 | meta->map_ptr = reg->map_ptr; |
17a52670 AS |
946 | } else if (arg_type == ARG_PTR_TO_MAP_KEY) { |
947 | /* bpf_map_xxx(..., map_ptr, ..., key) call: | |
948 | * check that [key, key + map->key_size) are within | |
949 | * stack limits and initialized | |
950 | */ | |
33ff9823 | 951 | if (!meta->map_ptr) { |
17a52670 AS |
952 | /* in function declaration map_ptr must come before |
953 | * map_key, so that it's verified and known before | |
954 | * we have to check map_key here. Otherwise it means | |
955 | * that kernel subsystem misconfigured verifier | |
956 | */ | |
957 | verbose("invalid map_ptr to access map->key\n"); | |
958 | return -EACCES; | |
959 | } | |
6841de8b AS |
960 | if (type == PTR_TO_PACKET) |
961 | err = check_packet_access(env, regno, 0, | |
962 | meta->map_ptr->key_size); | |
963 | else | |
964 | err = check_stack_boundary(env, regno, | |
965 | meta->map_ptr->key_size, | |
966 | false, NULL); | |
17a52670 AS |
967 | } else if (arg_type == ARG_PTR_TO_MAP_VALUE) { |
968 | /* bpf_map_xxx(..., map_ptr, ..., value) call: | |
969 | * check [value, value + map->value_size) validity | |
970 | */ | |
33ff9823 | 971 | if (!meta->map_ptr) { |
17a52670 AS |
972 | /* kernel subsystem misconfigured verifier */ |
973 | verbose("invalid map_ptr to access map->value\n"); | |
974 | return -EACCES; | |
975 | } | |
6841de8b AS |
976 | if (type == PTR_TO_PACKET) |
977 | err = check_packet_access(env, regno, 0, | |
978 | meta->map_ptr->value_size); | |
979 | else | |
980 | err = check_stack_boundary(env, regno, | |
981 | meta->map_ptr->value_size, | |
982 | false, NULL); | |
8e2fe1d9 DB |
983 | } else if (arg_type == ARG_CONST_STACK_SIZE || |
984 | arg_type == ARG_CONST_STACK_SIZE_OR_ZERO) { | |
985 | bool zero_size_allowed = (arg_type == ARG_CONST_STACK_SIZE_OR_ZERO); | |
17a52670 | 986 | |
17a52670 AS |
987 | /* bpf_xxx(..., buf, len) call will access 'len' bytes |
988 | * from stack pointer 'buf'. Check it | |
989 | * note: regno == len, regno - 1 == buf | |
990 | */ | |
991 | if (regno == 0) { | |
992 | /* kernel subsystem misconfigured verifier */ | |
993 | verbose("ARG_CONST_STACK_SIZE cannot be first argument\n"); | |
994 | return -EACCES; | |
995 | } | |
6841de8b AS |
996 | if (regs[regno - 1].type == PTR_TO_PACKET) |
997 | err = check_packet_access(env, regno - 1, 0, reg->imm); | |
998 | else | |
999 | err = check_stack_boundary(env, regno - 1, reg->imm, | |
1000 | zero_size_allowed, meta); | |
17a52670 AS |
1001 | } |
1002 | ||
1003 | return err; | |
6841de8b AS |
1004 | err_type: |
1005 | verbose("R%d type=%s expected=%s\n", regno, | |
1006 | reg_type_str[type], reg_type_str[expected_type]); | |
1007 | return -EACCES; | |
17a52670 AS |
1008 | } |
1009 | ||
35578d79 KX |
1010 | static int check_map_func_compatibility(struct bpf_map *map, int func_id) |
1011 | { | |
35578d79 KX |
1012 | if (!map) |
1013 | return 0; | |
1014 | ||
6aff67c8 AS |
1015 | /* We need a two way check, first is from map perspective ... */ |
1016 | switch (map->map_type) { | |
1017 | case BPF_MAP_TYPE_PROG_ARRAY: | |
1018 | if (func_id != BPF_FUNC_tail_call) | |
1019 | goto error; | |
1020 | break; | |
1021 | case BPF_MAP_TYPE_PERF_EVENT_ARRAY: | |
1022 | if (func_id != BPF_FUNC_perf_event_read && | |
1023 | func_id != BPF_FUNC_perf_event_output) | |
1024 | goto error; | |
1025 | break; | |
1026 | case BPF_MAP_TYPE_STACK_TRACE: | |
1027 | if (func_id != BPF_FUNC_get_stackid) | |
1028 | goto error; | |
1029 | break; | |
4ed8ec52 | 1030 | case BPF_MAP_TYPE_CGROUP_ARRAY: |
60747ef4 | 1031 | if (func_id != BPF_FUNC_skb_under_cgroup && |
60d20f91 | 1032 | func_id != BPF_FUNC_current_task_under_cgroup) |
4a482f34 MKL |
1033 | goto error; |
1034 | break; | |
6aff67c8 AS |
1035 | default: |
1036 | break; | |
1037 | } | |
1038 | ||
1039 | /* ... and second from the function itself. */ | |
1040 | switch (func_id) { | |
1041 | case BPF_FUNC_tail_call: | |
1042 | if (map->map_type != BPF_MAP_TYPE_PROG_ARRAY) | |
1043 | goto error; | |
1044 | break; | |
1045 | case BPF_FUNC_perf_event_read: | |
1046 | case BPF_FUNC_perf_event_output: | |
1047 | if (map->map_type != BPF_MAP_TYPE_PERF_EVENT_ARRAY) | |
1048 | goto error; | |
1049 | break; | |
1050 | case BPF_FUNC_get_stackid: | |
1051 | if (map->map_type != BPF_MAP_TYPE_STACK_TRACE) | |
1052 | goto error; | |
1053 | break; | |
60d20f91 | 1054 | case BPF_FUNC_current_task_under_cgroup: |
747ea55e | 1055 | case BPF_FUNC_skb_under_cgroup: |
4a482f34 MKL |
1056 | if (map->map_type != BPF_MAP_TYPE_CGROUP_ARRAY) |
1057 | goto error; | |
1058 | break; | |
6aff67c8 AS |
1059 | default: |
1060 | break; | |
35578d79 KX |
1061 | } |
1062 | ||
1063 | return 0; | |
6aff67c8 AS |
1064 | error: |
1065 | verbose("cannot pass map_type %d into func %d\n", | |
1066 | map->map_type, func_id); | |
1067 | return -EINVAL; | |
35578d79 KX |
1068 | } |
1069 | ||
435faee1 DB |
1070 | static int check_raw_mode(const struct bpf_func_proto *fn) |
1071 | { | |
1072 | int count = 0; | |
1073 | ||
1074 | if (fn->arg1_type == ARG_PTR_TO_RAW_STACK) | |
1075 | count++; | |
1076 | if (fn->arg2_type == ARG_PTR_TO_RAW_STACK) | |
1077 | count++; | |
1078 | if (fn->arg3_type == ARG_PTR_TO_RAW_STACK) | |
1079 | count++; | |
1080 | if (fn->arg4_type == ARG_PTR_TO_RAW_STACK) | |
1081 | count++; | |
1082 | if (fn->arg5_type == ARG_PTR_TO_RAW_STACK) | |
1083 | count++; | |
1084 | ||
1085 | return count > 1 ? -EINVAL : 0; | |
1086 | } | |
1087 | ||
58e2af8b | 1088 | static void clear_all_pkt_pointers(struct bpf_verifier_env *env) |
969bf05e | 1089 | { |
58e2af8b JK |
1090 | struct bpf_verifier_state *state = &env->cur_state; |
1091 | struct bpf_reg_state *regs = state->regs, *reg; | |
969bf05e AS |
1092 | int i; |
1093 | ||
1094 | for (i = 0; i < MAX_BPF_REG; i++) | |
1095 | if (regs[i].type == PTR_TO_PACKET || | |
1096 | regs[i].type == PTR_TO_PACKET_END) | |
1097 | mark_reg_unknown_value(regs, i); | |
1098 | ||
1099 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1100 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1101 | continue; | |
1102 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1103 | if (reg->type != PTR_TO_PACKET && | |
1104 | reg->type != PTR_TO_PACKET_END) | |
1105 | continue; | |
1106 | reg->type = UNKNOWN_VALUE; | |
1107 | reg->imm = 0; | |
1108 | } | |
1109 | } | |
1110 | ||
58e2af8b | 1111 | static int check_call(struct bpf_verifier_env *env, int func_id) |
17a52670 | 1112 | { |
58e2af8b | 1113 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 1114 | const struct bpf_func_proto *fn = NULL; |
58e2af8b JK |
1115 | struct bpf_reg_state *regs = state->regs; |
1116 | struct bpf_reg_state *reg; | |
33ff9823 | 1117 | struct bpf_call_arg_meta meta; |
969bf05e | 1118 | bool changes_data; |
17a52670 AS |
1119 | int i, err; |
1120 | ||
1121 | /* find function prototype */ | |
1122 | if (func_id < 0 || func_id >= __BPF_FUNC_MAX_ID) { | |
1123 | verbose("invalid func %d\n", func_id); | |
1124 | return -EINVAL; | |
1125 | } | |
1126 | ||
1127 | if (env->prog->aux->ops->get_func_proto) | |
1128 | fn = env->prog->aux->ops->get_func_proto(func_id); | |
1129 | ||
1130 | if (!fn) { | |
1131 | verbose("unknown func %d\n", func_id); | |
1132 | return -EINVAL; | |
1133 | } | |
1134 | ||
1135 | /* eBPF programs must be GPL compatible to use GPL-ed functions */ | |
24701ece | 1136 | if (!env->prog->gpl_compatible && fn->gpl_only) { |
17a52670 AS |
1137 | verbose("cannot call GPL only function from proprietary program\n"); |
1138 | return -EINVAL; | |
1139 | } | |
1140 | ||
969bf05e AS |
1141 | changes_data = bpf_helper_changes_skb_data(fn->func); |
1142 | ||
33ff9823 | 1143 | memset(&meta, 0, sizeof(meta)); |
36bbef52 | 1144 | meta.pkt_access = fn->pkt_access; |
33ff9823 | 1145 | |
435faee1 DB |
1146 | /* We only support one arg being in raw mode at the moment, which |
1147 | * is sufficient for the helper functions we have right now. | |
1148 | */ | |
1149 | err = check_raw_mode(fn); | |
1150 | if (err) { | |
1151 | verbose("kernel subsystem misconfigured func %d\n", func_id); | |
1152 | return err; | |
1153 | } | |
1154 | ||
17a52670 | 1155 | /* check args */ |
33ff9823 | 1156 | err = check_func_arg(env, BPF_REG_1, fn->arg1_type, &meta); |
17a52670 AS |
1157 | if (err) |
1158 | return err; | |
33ff9823 | 1159 | err = check_func_arg(env, BPF_REG_2, fn->arg2_type, &meta); |
17a52670 AS |
1160 | if (err) |
1161 | return err; | |
33ff9823 | 1162 | err = check_func_arg(env, BPF_REG_3, fn->arg3_type, &meta); |
17a52670 AS |
1163 | if (err) |
1164 | return err; | |
33ff9823 | 1165 | err = check_func_arg(env, BPF_REG_4, fn->arg4_type, &meta); |
17a52670 AS |
1166 | if (err) |
1167 | return err; | |
33ff9823 | 1168 | err = check_func_arg(env, BPF_REG_5, fn->arg5_type, &meta); |
17a52670 AS |
1169 | if (err) |
1170 | return err; | |
1171 | ||
435faee1 DB |
1172 | /* Mark slots with STACK_MISC in case of raw mode, stack offset |
1173 | * is inferred from register state. | |
1174 | */ | |
1175 | for (i = 0; i < meta.access_size; i++) { | |
1176 | err = check_mem_access(env, meta.regno, i, BPF_B, BPF_WRITE, -1); | |
1177 | if (err) | |
1178 | return err; | |
1179 | } | |
1180 | ||
17a52670 AS |
1181 | /* reset caller saved regs */ |
1182 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
1183 | reg = regs + caller_saved[i]; | |
1184 | reg->type = NOT_INIT; | |
1185 | reg->imm = 0; | |
1186 | } | |
1187 | ||
1188 | /* update return register */ | |
1189 | if (fn->ret_type == RET_INTEGER) { | |
1190 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
1191 | } else if (fn->ret_type == RET_VOID) { | |
1192 | regs[BPF_REG_0].type = NOT_INIT; | |
1193 | } else if (fn->ret_type == RET_PTR_TO_MAP_VALUE_OR_NULL) { | |
1194 | regs[BPF_REG_0].type = PTR_TO_MAP_VALUE_OR_NULL; | |
1195 | /* remember map_ptr, so that check_map_access() | |
1196 | * can check 'value_size' boundary of memory access | |
1197 | * to map element returned from bpf_map_lookup_elem() | |
1198 | */ | |
33ff9823 | 1199 | if (meta.map_ptr == NULL) { |
17a52670 AS |
1200 | verbose("kernel subsystem misconfigured verifier\n"); |
1201 | return -EINVAL; | |
1202 | } | |
33ff9823 | 1203 | regs[BPF_REG_0].map_ptr = meta.map_ptr; |
17a52670 AS |
1204 | } else { |
1205 | verbose("unknown return type %d of func %d\n", | |
1206 | fn->ret_type, func_id); | |
1207 | return -EINVAL; | |
1208 | } | |
04fd61ab | 1209 | |
33ff9823 | 1210 | err = check_map_func_compatibility(meta.map_ptr, func_id); |
35578d79 KX |
1211 | if (err) |
1212 | return err; | |
04fd61ab | 1213 | |
969bf05e AS |
1214 | if (changes_data) |
1215 | clear_all_pkt_pointers(env); | |
1216 | return 0; | |
1217 | } | |
1218 | ||
58e2af8b JK |
1219 | static int check_packet_ptr_add(struct bpf_verifier_env *env, |
1220 | struct bpf_insn *insn) | |
969bf05e | 1221 | { |
58e2af8b JK |
1222 | struct bpf_reg_state *regs = env->cur_state.regs; |
1223 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1224 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
1225 | struct bpf_reg_state tmp_reg; | |
969bf05e AS |
1226 | s32 imm; |
1227 | ||
1228 | if (BPF_SRC(insn->code) == BPF_K) { | |
1229 | /* pkt_ptr += imm */ | |
1230 | imm = insn->imm; | |
1231 | ||
1232 | add_imm: | |
1233 | if (imm <= 0) { | |
1234 | verbose("addition of negative constant to packet pointer is not allowed\n"); | |
1235 | return -EACCES; | |
1236 | } | |
1237 | if (imm >= MAX_PACKET_OFF || | |
1238 | imm + dst_reg->off >= MAX_PACKET_OFF) { | |
1239 | verbose("constant %d is too large to add to packet pointer\n", | |
1240 | imm); | |
1241 | return -EACCES; | |
1242 | } | |
1243 | /* a constant was added to pkt_ptr. | |
1244 | * Remember it while keeping the same 'id' | |
1245 | */ | |
1246 | dst_reg->off += imm; | |
1247 | } else { | |
1b9b69ec AS |
1248 | if (src_reg->type == PTR_TO_PACKET) { |
1249 | /* R6=pkt(id=0,off=0,r=62) R7=imm22; r7 += r6 */ | |
1250 | tmp_reg = *dst_reg; /* save r7 state */ | |
1251 | *dst_reg = *src_reg; /* copy pkt_ptr state r6 into r7 */ | |
1252 | src_reg = &tmp_reg; /* pretend it's src_reg state */ | |
1253 | /* if the checks below reject it, the copy won't matter, | |
1254 | * since we're rejecting the whole program. If all ok, | |
1255 | * then imm22 state will be added to r7 | |
1256 | * and r7 will be pkt(id=0,off=22,r=62) while | |
1257 | * r6 will stay as pkt(id=0,off=0,r=62) | |
1258 | */ | |
1259 | } | |
1260 | ||
969bf05e AS |
1261 | if (src_reg->type == CONST_IMM) { |
1262 | /* pkt_ptr += reg where reg is known constant */ | |
1263 | imm = src_reg->imm; | |
1264 | goto add_imm; | |
1265 | } | |
1266 | /* disallow pkt_ptr += reg | |
1267 | * if reg is not uknown_value with guaranteed zero upper bits | |
1268 | * otherwise pkt_ptr may overflow and addition will become | |
1269 | * subtraction which is not allowed | |
1270 | */ | |
1271 | if (src_reg->type != UNKNOWN_VALUE) { | |
1272 | verbose("cannot add '%s' to ptr_to_packet\n", | |
1273 | reg_type_str[src_reg->type]); | |
1274 | return -EACCES; | |
1275 | } | |
1276 | if (src_reg->imm < 48) { | |
1277 | verbose("cannot add integer value with %lld upper zero bits to ptr_to_packet\n", | |
1278 | src_reg->imm); | |
1279 | return -EACCES; | |
1280 | } | |
1281 | /* dst_reg stays as pkt_ptr type and since some positive | |
1282 | * integer value was added to the pointer, increment its 'id' | |
1283 | */ | |
1f415a74 | 1284 | dst_reg->id = ++env->id_gen; |
969bf05e AS |
1285 | |
1286 | /* something was added to pkt_ptr, set range and off to zero */ | |
1287 | dst_reg->off = 0; | |
1288 | dst_reg->range = 0; | |
1289 | } | |
1290 | return 0; | |
1291 | } | |
1292 | ||
58e2af8b | 1293 | static int evaluate_reg_alu(struct bpf_verifier_env *env, struct bpf_insn *insn) |
969bf05e | 1294 | { |
58e2af8b JK |
1295 | struct bpf_reg_state *regs = env->cur_state.regs; |
1296 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
969bf05e AS |
1297 | u8 opcode = BPF_OP(insn->code); |
1298 | s64 imm_log2; | |
1299 | ||
1300 | /* for type == UNKNOWN_VALUE: | |
1301 | * imm > 0 -> number of zero upper bits | |
1302 | * imm == 0 -> don't track which is the same as all bits can be non-zero | |
1303 | */ | |
1304 | ||
1305 | if (BPF_SRC(insn->code) == BPF_X) { | |
58e2af8b | 1306 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; |
969bf05e AS |
1307 | |
1308 | if (src_reg->type == UNKNOWN_VALUE && src_reg->imm > 0 && | |
1309 | dst_reg->imm && opcode == BPF_ADD) { | |
1310 | /* dreg += sreg | |
1311 | * where both have zero upper bits. Adding them | |
1312 | * can only result making one more bit non-zero | |
1313 | * in the larger value. | |
1314 | * Ex. 0xffff (imm=48) + 1 (imm=63) = 0x10000 (imm=47) | |
1315 | * 0xffff (imm=48) + 0xffff = 0x1fffe (imm=47) | |
1316 | */ | |
1317 | dst_reg->imm = min(dst_reg->imm, src_reg->imm); | |
1318 | dst_reg->imm--; | |
1319 | return 0; | |
1320 | } | |
1321 | if (src_reg->type == CONST_IMM && src_reg->imm > 0 && | |
1322 | dst_reg->imm && opcode == BPF_ADD) { | |
1323 | /* dreg += sreg | |
1324 | * where dreg has zero upper bits and sreg is const. | |
1325 | * Adding them can only result making one more bit | |
1326 | * non-zero in the larger value. | |
1327 | */ | |
1328 | imm_log2 = __ilog2_u64((long long)src_reg->imm); | |
1329 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1330 | dst_reg->imm--; | |
1331 | return 0; | |
1332 | } | |
1333 | /* all other cases non supported yet, just mark dst_reg */ | |
1334 | dst_reg->imm = 0; | |
1335 | return 0; | |
1336 | } | |
1337 | ||
1338 | /* sign extend 32-bit imm into 64-bit to make sure that | |
1339 | * negative values occupy bit 63. Note ilog2() would have | |
1340 | * been incorrect, since sizeof(insn->imm) == 4 | |
1341 | */ | |
1342 | imm_log2 = __ilog2_u64((long long)insn->imm); | |
1343 | ||
1344 | if (dst_reg->imm && opcode == BPF_LSH) { | |
1345 | /* reg <<= imm | |
1346 | * if reg was a result of 2 byte load, then its imm == 48 | |
1347 | * which means that upper 48 bits are zero and shifting this reg | |
1348 | * left by 4 would mean that upper 44 bits are still zero | |
1349 | */ | |
1350 | dst_reg->imm -= insn->imm; | |
1351 | } else if (dst_reg->imm && opcode == BPF_MUL) { | |
1352 | /* reg *= imm | |
1353 | * if multiplying by 14 subtract 4 | |
1354 | * This is conservative calculation of upper zero bits. | |
1355 | * It's not trying to special case insn->imm == 1 or 0 cases | |
1356 | */ | |
1357 | dst_reg->imm -= imm_log2 + 1; | |
1358 | } else if (opcode == BPF_AND) { | |
1359 | /* reg &= imm */ | |
1360 | dst_reg->imm = 63 - imm_log2; | |
1361 | } else if (dst_reg->imm && opcode == BPF_ADD) { | |
1362 | /* reg += imm */ | |
1363 | dst_reg->imm = min(dst_reg->imm, 63 - imm_log2); | |
1364 | dst_reg->imm--; | |
1365 | } else if (opcode == BPF_RSH) { | |
1366 | /* reg >>= imm | |
1367 | * which means that after right shift, upper bits will be zero | |
1368 | * note that verifier already checked that | |
1369 | * 0 <= imm < 64 for shift insn | |
1370 | */ | |
1371 | dst_reg->imm += insn->imm; | |
1372 | if (unlikely(dst_reg->imm > 64)) | |
1373 | /* some dumb code did: | |
1374 | * r2 = *(u32 *)mem; | |
1375 | * r2 >>= 32; | |
1376 | * and all bits are zero now */ | |
1377 | dst_reg->imm = 64; | |
1378 | } else { | |
1379 | /* all other alu ops, means that we don't know what will | |
1380 | * happen to the value, mark it with unknown number of zero bits | |
1381 | */ | |
1382 | dst_reg->imm = 0; | |
1383 | } | |
1384 | ||
1385 | if (dst_reg->imm < 0) { | |
1386 | /* all 64 bits of the register can contain non-zero bits | |
1387 | * and such value cannot be added to ptr_to_packet, since it | |
1388 | * may overflow, mark it as unknown to avoid further eval | |
1389 | */ | |
1390 | dst_reg->imm = 0; | |
1391 | } | |
1392 | return 0; | |
1393 | } | |
1394 | ||
58e2af8b JK |
1395 | static int evaluate_reg_imm_alu(struct bpf_verifier_env *env, |
1396 | struct bpf_insn *insn) | |
969bf05e | 1397 | { |
58e2af8b JK |
1398 | struct bpf_reg_state *regs = env->cur_state.regs; |
1399 | struct bpf_reg_state *dst_reg = ®s[insn->dst_reg]; | |
1400 | struct bpf_reg_state *src_reg = ®s[insn->src_reg]; | |
969bf05e AS |
1401 | u8 opcode = BPF_OP(insn->code); |
1402 | ||
1403 | /* dst_reg->type == CONST_IMM here, simulate execution of 'add' insn. | |
1404 | * Don't care about overflow or negative values, just add them | |
1405 | */ | |
1406 | if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_K) | |
1407 | dst_reg->imm += insn->imm; | |
1408 | else if (opcode == BPF_ADD && BPF_SRC(insn->code) == BPF_X && | |
1409 | src_reg->type == CONST_IMM) | |
1410 | dst_reg->imm += src_reg->imm; | |
1411 | else | |
1412 | mark_reg_unknown_value(regs, insn->dst_reg); | |
17a52670 AS |
1413 | return 0; |
1414 | } | |
1415 | ||
1416 | /* check validity of 32-bit and 64-bit arithmetic operations */ | |
58e2af8b | 1417 | static int check_alu_op(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 1418 | { |
58e2af8b | 1419 | struct bpf_reg_state *regs = env->cur_state.regs, *dst_reg; |
17a52670 AS |
1420 | u8 opcode = BPF_OP(insn->code); |
1421 | int err; | |
1422 | ||
1423 | if (opcode == BPF_END || opcode == BPF_NEG) { | |
1424 | if (opcode == BPF_NEG) { | |
1425 | if (BPF_SRC(insn->code) != 0 || | |
1426 | insn->src_reg != BPF_REG_0 || | |
1427 | insn->off != 0 || insn->imm != 0) { | |
1428 | verbose("BPF_NEG uses reserved fields\n"); | |
1429 | return -EINVAL; | |
1430 | } | |
1431 | } else { | |
1432 | if (insn->src_reg != BPF_REG_0 || insn->off != 0 || | |
1433 | (insn->imm != 16 && insn->imm != 32 && insn->imm != 64)) { | |
1434 | verbose("BPF_END uses reserved fields\n"); | |
1435 | return -EINVAL; | |
1436 | } | |
1437 | } | |
1438 | ||
1439 | /* check src operand */ | |
1440 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1441 | if (err) | |
1442 | return err; | |
1443 | ||
1be7f75d AS |
1444 | if (is_pointer_value(env, insn->dst_reg)) { |
1445 | verbose("R%d pointer arithmetic prohibited\n", | |
1446 | insn->dst_reg); | |
1447 | return -EACCES; | |
1448 | } | |
1449 | ||
17a52670 AS |
1450 | /* check dest operand */ |
1451 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1452 | if (err) | |
1453 | return err; | |
1454 | ||
1455 | } else if (opcode == BPF_MOV) { | |
1456 | ||
1457 | if (BPF_SRC(insn->code) == BPF_X) { | |
1458 | if (insn->imm != 0 || insn->off != 0) { | |
1459 | verbose("BPF_MOV uses reserved fields\n"); | |
1460 | return -EINVAL; | |
1461 | } | |
1462 | ||
1463 | /* check src operand */ | |
1464 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1465 | if (err) | |
1466 | return err; | |
1467 | } else { | |
1468 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1469 | verbose("BPF_MOV uses reserved fields\n"); | |
1470 | return -EINVAL; | |
1471 | } | |
1472 | } | |
1473 | ||
1474 | /* check dest operand */ | |
1475 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1476 | if (err) | |
1477 | return err; | |
1478 | ||
1479 | if (BPF_SRC(insn->code) == BPF_X) { | |
1480 | if (BPF_CLASS(insn->code) == BPF_ALU64) { | |
1481 | /* case: R1 = R2 | |
1482 | * copy register state to dest reg | |
1483 | */ | |
1484 | regs[insn->dst_reg] = regs[insn->src_reg]; | |
1485 | } else { | |
1be7f75d AS |
1486 | if (is_pointer_value(env, insn->src_reg)) { |
1487 | verbose("R%d partial copy of pointer\n", | |
1488 | insn->src_reg); | |
1489 | return -EACCES; | |
1490 | } | |
17a52670 AS |
1491 | regs[insn->dst_reg].type = UNKNOWN_VALUE; |
1492 | regs[insn->dst_reg].map_ptr = NULL; | |
1493 | } | |
1494 | } else { | |
1495 | /* case: R = imm | |
1496 | * remember the value we stored into this reg | |
1497 | */ | |
1498 | regs[insn->dst_reg].type = CONST_IMM; | |
1499 | regs[insn->dst_reg].imm = insn->imm; | |
1500 | } | |
1501 | ||
1502 | } else if (opcode > BPF_END) { | |
1503 | verbose("invalid BPF_ALU opcode %x\n", opcode); | |
1504 | return -EINVAL; | |
1505 | ||
1506 | } else { /* all other ALU ops: and, sub, xor, add, ... */ | |
1507 | ||
17a52670 AS |
1508 | if (BPF_SRC(insn->code) == BPF_X) { |
1509 | if (insn->imm != 0 || insn->off != 0) { | |
1510 | verbose("BPF_ALU uses reserved fields\n"); | |
1511 | return -EINVAL; | |
1512 | } | |
1513 | /* check src1 operand */ | |
1514 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1515 | if (err) | |
1516 | return err; | |
1517 | } else { | |
1518 | if (insn->src_reg != BPF_REG_0 || insn->off != 0) { | |
1519 | verbose("BPF_ALU uses reserved fields\n"); | |
1520 | return -EINVAL; | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | /* check src2 operand */ | |
1525 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1526 | if (err) | |
1527 | return err; | |
1528 | ||
1529 | if ((opcode == BPF_MOD || opcode == BPF_DIV) && | |
1530 | BPF_SRC(insn->code) == BPF_K && insn->imm == 0) { | |
1531 | verbose("div by zero\n"); | |
1532 | return -EINVAL; | |
1533 | } | |
1534 | ||
229394e8 RV |
1535 | if ((opcode == BPF_LSH || opcode == BPF_RSH || |
1536 | opcode == BPF_ARSH) && BPF_SRC(insn->code) == BPF_K) { | |
1537 | int size = BPF_CLASS(insn->code) == BPF_ALU64 ? 64 : 32; | |
1538 | ||
1539 | if (insn->imm < 0 || insn->imm >= size) { | |
1540 | verbose("invalid shift %d\n", insn->imm); | |
1541 | return -EINVAL; | |
1542 | } | |
1543 | } | |
1544 | ||
1a0dc1ac AS |
1545 | /* check dest operand */ |
1546 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
1547 | if (err) | |
1548 | return err; | |
1549 | ||
1550 | dst_reg = ®s[insn->dst_reg]; | |
1551 | ||
17a52670 AS |
1552 | /* pattern match 'bpf_add Rx, imm' instruction */ |
1553 | if (opcode == BPF_ADD && BPF_CLASS(insn->code) == BPF_ALU64 && | |
1a0dc1ac AS |
1554 | dst_reg->type == FRAME_PTR && BPF_SRC(insn->code) == BPF_K) { |
1555 | dst_reg->type = PTR_TO_STACK; | |
1556 | dst_reg->imm = insn->imm; | |
1557 | return 0; | |
969bf05e AS |
1558 | } else if (opcode == BPF_ADD && |
1559 | BPF_CLASS(insn->code) == BPF_ALU64 && | |
1b9b69ec AS |
1560 | (dst_reg->type == PTR_TO_PACKET || |
1561 | (BPF_SRC(insn->code) == BPF_X && | |
1562 | regs[insn->src_reg].type == PTR_TO_PACKET))) { | |
969bf05e AS |
1563 | /* ptr_to_packet += K|X */ |
1564 | return check_packet_ptr_add(env, insn); | |
1565 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1566 | dst_reg->type == UNKNOWN_VALUE && | |
1567 | env->allow_ptr_leaks) { | |
1568 | /* unknown += K|X */ | |
1569 | return evaluate_reg_alu(env, insn); | |
1570 | } else if (BPF_CLASS(insn->code) == BPF_ALU64 && | |
1571 | dst_reg->type == CONST_IMM && | |
1572 | env->allow_ptr_leaks) { | |
1573 | /* reg_imm += K|X */ | |
1574 | return evaluate_reg_imm_alu(env, insn); | |
1be7f75d AS |
1575 | } else if (is_pointer_value(env, insn->dst_reg)) { |
1576 | verbose("R%d pointer arithmetic prohibited\n", | |
1577 | insn->dst_reg); | |
1578 | return -EACCES; | |
1579 | } else if (BPF_SRC(insn->code) == BPF_X && | |
1580 | is_pointer_value(env, insn->src_reg)) { | |
1581 | verbose("R%d pointer arithmetic prohibited\n", | |
1582 | insn->src_reg); | |
1583 | return -EACCES; | |
1584 | } | |
17a52670 | 1585 | |
1a0dc1ac AS |
1586 | /* mark dest operand */ |
1587 | mark_reg_unknown_value(regs, insn->dst_reg); | |
17a52670 AS |
1588 | } |
1589 | ||
1590 | return 0; | |
1591 | } | |
1592 | ||
58e2af8b JK |
1593 | static void find_good_pkt_pointers(struct bpf_verifier_state *state, |
1594 | struct bpf_reg_state *dst_reg) | |
969bf05e | 1595 | { |
58e2af8b | 1596 | struct bpf_reg_state *regs = state->regs, *reg; |
969bf05e | 1597 | int i; |
2d2be8ca DB |
1598 | |
1599 | /* LLVM can generate two kind of checks: | |
1600 | * | |
1601 | * Type 1: | |
1602 | * | |
1603 | * r2 = r3; | |
1604 | * r2 += 8; | |
1605 | * if (r2 > pkt_end) goto <handle exception> | |
1606 | * <access okay> | |
1607 | * | |
1608 | * Where: | |
1609 | * r2 == dst_reg, pkt_end == src_reg | |
1610 | * r2=pkt(id=n,off=8,r=0) | |
1611 | * r3=pkt(id=n,off=0,r=0) | |
1612 | * | |
1613 | * Type 2: | |
1614 | * | |
1615 | * r2 = r3; | |
1616 | * r2 += 8; | |
1617 | * if (pkt_end >= r2) goto <access okay> | |
1618 | * <handle exception> | |
1619 | * | |
1620 | * Where: | |
1621 | * pkt_end == dst_reg, r2 == src_reg | |
1622 | * r2=pkt(id=n,off=8,r=0) | |
1623 | * r3=pkt(id=n,off=0,r=0) | |
1624 | * | |
1625 | * Find register r3 and mark its range as r3=pkt(id=n,off=0,r=8) | |
1626 | * so that range of bytes [r3, r3 + 8) is safe to access. | |
969bf05e | 1627 | */ |
2d2be8ca | 1628 | |
969bf05e AS |
1629 | for (i = 0; i < MAX_BPF_REG; i++) |
1630 | if (regs[i].type == PTR_TO_PACKET && regs[i].id == dst_reg->id) | |
1631 | regs[i].range = dst_reg->off; | |
1632 | ||
1633 | for (i = 0; i < MAX_BPF_STACK; i += BPF_REG_SIZE) { | |
1634 | if (state->stack_slot_type[i] != STACK_SPILL) | |
1635 | continue; | |
1636 | reg = &state->spilled_regs[i / BPF_REG_SIZE]; | |
1637 | if (reg->type == PTR_TO_PACKET && reg->id == dst_reg->id) | |
1638 | reg->range = dst_reg->off; | |
1639 | } | |
1640 | } | |
1641 | ||
58e2af8b | 1642 | static int check_cond_jmp_op(struct bpf_verifier_env *env, |
17a52670 AS |
1643 | struct bpf_insn *insn, int *insn_idx) |
1644 | { | |
58e2af8b JK |
1645 | struct bpf_verifier_state *other_branch, *this_branch = &env->cur_state; |
1646 | struct bpf_reg_state *regs = this_branch->regs, *dst_reg; | |
17a52670 AS |
1647 | u8 opcode = BPF_OP(insn->code); |
1648 | int err; | |
1649 | ||
1650 | if (opcode > BPF_EXIT) { | |
1651 | verbose("invalid BPF_JMP opcode %x\n", opcode); | |
1652 | return -EINVAL; | |
1653 | } | |
1654 | ||
1655 | if (BPF_SRC(insn->code) == BPF_X) { | |
1656 | if (insn->imm != 0) { | |
1657 | verbose("BPF_JMP uses reserved fields\n"); | |
1658 | return -EINVAL; | |
1659 | } | |
1660 | ||
1661 | /* check src1 operand */ | |
1662 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1663 | if (err) | |
1664 | return err; | |
1be7f75d AS |
1665 | |
1666 | if (is_pointer_value(env, insn->src_reg)) { | |
1667 | verbose("R%d pointer comparison prohibited\n", | |
1668 | insn->src_reg); | |
1669 | return -EACCES; | |
1670 | } | |
17a52670 AS |
1671 | } else { |
1672 | if (insn->src_reg != BPF_REG_0) { | |
1673 | verbose("BPF_JMP uses reserved fields\n"); | |
1674 | return -EINVAL; | |
1675 | } | |
1676 | } | |
1677 | ||
1678 | /* check src2 operand */ | |
1679 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
1680 | if (err) | |
1681 | return err; | |
1682 | ||
1a0dc1ac AS |
1683 | dst_reg = ®s[insn->dst_reg]; |
1684 | ||
17a52670 AS |
1685 | /* detect if R == 0 where R was initialized to zero earlier */ |
1686 | if (BPF_SRC(insn->code) == BPF_K && | |
1687 | (opcode == BPF_JEQ || opcode == BPF_JNE) && | |
1a0dc1ac | 1688 | dst_reg->type == CONST_IMM && dst_reg->imm == insn->imm) { |
17a52670 AS |
1689 | if (opcode == BPF_JEQ) { |
1690 | /* if (imm == imm) goto pc+off; | |
1691 | * only follow the goto, ignore fall-through | |
1692 | */ | |
1693 | *insn_idx += insn->off; | |
1694 | return 0; | |
1695 | } else { | |
1696 | /* if (imm != imm) goto pc+off; | |
1697 | * only follow fall-through branch, since | |
1698 | * that's where the program will go | |
1699 | */ | |
1700 | return 0; | |
1701 | } | |
1702 | } | |
1703 | ||
1704 | other_branch = push_stack(env, *insn_idx + insn->off + 1, *insn_idx); | |
1705 | if (!other_branch) | |
1706 | return -EFAULT; | |
1707 | ||
58e2af8b | 1708 | /* detect if R == 0 where R is returned from bpf_map_lookup_elem() */ |
17a52670 | 1709 | if (BPF_SRC(insn->code) == BPF_K && |
1a0dc1ac AS |
1710 | insn->imm == 0 && (opcode == BPF_JEQ || opcode == BPF_JNE) && |
1711 | dst_reg->type == PTR_TO_MAP_VALUE_OR_NULL) { | |
17a52670 AS |
1712 | if (opcode == BPF_JEQ) { |
1713 | /* next fallthrough insn can access memory via | |
1714 | * this register | |
1715 | */ | |
1716 | regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; | |
1717 | /* branch targer cannot access it, since reg == 0 */ | |
735b4333 AS |
1718 | mark_reg_unknown_value(other_branch->regs, |
1719 | insn->dst_reg); | |
17a52670 AS |
1720 | } else { |
1721 | other_branch->regs[insn->dst_reg].type = PTR_TO_MAP_VALUE; | |
735b4333 | 1722 | mark_reg_unknown_value(regs, insn->dst_reg); |
17a52670 | 1723 | } |
969bf05e AS |
1724 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGT && |
1725 | dst_reg->type == PTR_TO_PACKET && | |
1726 | regs[insn->src_reg].type == PTR_TO_PACKET_END) { | |
2d2be8ca DB |
1727 | find_good_pkt_pointers(this_branch, dst_reg); |
1728 | } else if (BPF_SRC(insn->code) == BPF_X && opcode == BPF_JGE && | |
1729 | dst_reg->type == PTR_TO_PACKET_END && | |
1730 | regs[insn->src_reg].type == PTR_TO_PACKET) { | |
1731 | find_good_pkt_pointers(other_branch, ®s[insn->src_reg]); | |
1be7f75d AS |
1732 | } else if (is_pointer_value(env, insn->dst_reg)) { |
1733 | verbose("R%d pointer comparison prohibited\n", insn->dst_reg); | |
1734 | return -EACCES; | |
17a52670 AS |
1735 | } |
1736 | if (log_level) | |
2d2be8ca | 1737 | print_verifier_state(this_branch); |
17a52670 AS |
1738 | return 0; |
1739 | } | |
1740 | ||
0246e64d AS |
1741 | /* return the map pointer stored inside BPF_LD_IMM64 instruction */ |
1742 | static struct bpf_map *ld_imm64_to_map_ptr(struct bpf_insn *insn) | |
1743 | { | |
1744 | u64 imm64 = ((u64) (u32) insn[0].imm) | ((u64) (u32) insn[1].imm) << 32; | |
1745 | ||
1746 | return (struct bpf_map *) (unsigned long) imm64; | |
1747 | } | |
1748 | ||
17a52670 | 1749 | /* verify BPF_LD_IMM64 instruction */ |
58e2af8b | 1750 | static int check_ld_imm(struct bpf_verifier_env *env, struct bpf_insn *insn) |
17a52670 | 1751 | { |
58e2af8b | 1752 | struct bpf_reg_state *regs = env->cur_state.regs; |
17a52670 AS |
1753 | int err; |
1754 | ||
1755 | if (BPF_SIZE(insn->code) != BPF_DW) { | |
1756 | verbose("invalid BPF_LD_IMM insn\n"); | |
1757 | return -EINVAL; | |
1758 | } | |
1759 | if (insn->off != 0) { | |
1760 | verbose("BPF_LD_IMM64 uses reserved fields\n"); | |
1761 | return -EINVAL; | |
1762 | } | |
1763 | ||
1764 | err = check_reg_arg(regs, insn->dst_reg, DST_OP); | |
1765 | if (err) | |
1766 | return err; | |
1767 | ||
1768 | if (insn->src_reg == 0) | |
1769 | /* generic move 64-bit immediate into a register */ | |
1770 | return 0; | |
1771 | ||
1772 | /* replace_map_fd_with_map_ptr() should have caught bad ld_imm64 */ | |
1773 | BUG_ON(insn->src_reg != BPF_PSEUDO_MAP_FD); | |
1774 | ||
1775 | regs[insn->dst_reg].type = CONST_PTR_TO_MAP; | |
1776 | regs[insn->dst_reg].map_ptr = ld_imm64_to_map_ptr(insn); | |
1777 | return 0; | |
1778 | } | |
1779 | ||
96be4325 DB |
1780 | static bool may_access_skb(enum bpf_prog_type type) |
1781 | { | |
1782 | switch (type) { | |
1783 | case BPF_PROG_TYPE_SOCKET_FILTER: | |
1784 | case BPF_PROG_TYPE_SCHED_CLS: | |
94caee8c | 1785 | case BPF_PROG_TYPE_SCHED_ACT: |
96be4325 DB |
1786 | return true; |
1787 | default: | |
1788 | return false; | |
1789 | } | |
1790 | } | |
1791 | ||
ddd872bc AS |
1792 | /* verify safety of LD_ABS|LD_IND instructions: |
1793 | * - they can only appear in the programs where ctx == skb | |
1794 | * - since they are wrappers of function calls, they scratch R1-R5 registers, | |
1795 | * preserve R6-R9, and store return value into R0 | |
1796 | * | |
1797 | * Implicit input: | |
1798 | * ctx == skb == R6 == CTX | |
1799 | * | |
1800 | * Explicit input: | |
1801 | * SRC == any register | |
1802 | * IMM == 32-bit immediate | |
1803 | * | |
1804 | * Output: | |
1805 | * R0 - 8/16/32-bit skb data converted to cpu endianness | |
1806 | */ | |
58e2af8b | 1807 | static int check_ld_abs(struct bpf_verifier_env *env, struct bpf_insn *insn) |
ddd872bc | 1808 | { |
58e2af8b | 1809 | struct bpf_reg_state *regs = env->cur_state.regs; |
ddd872bc | 1810 | u8 mode = BPF_MODE(insn->code); |
58e2af8b | 1811 | struct bpf_reg_state *reg; |
ddd872bc AS |
1812 | int i, err; |
1813 | ||
24701ece | 1814 | if (!may_access_skb(env->prog->type)) { |
1a0dc1ac | 1815 | verbose("BPF_LD_[ABS|IND] instructions not allowed for this program type\n"); |
ddd872bc AS |
1816 | return -EINVAL; |
1817 | } | |
1818 | ||
1819 | if (insn->dst_reg != BPF_REG_0 || insn->off != 0 || | |
d82bccc6 | 1820 | BPF_SIZE(insn->code) == BPF_DW || |
ddd872bc | 1821 | (mode == BPF_ABS && insn->src_reg != BPF_REG_0)) { |
1a0dc1ac | 1822 | verbose("BPF_LD_[ABS|IND] uses reserved fields\n"); |
ddd872bc AS |
1823 | return -EINVAL; |
1824 | } | |
1825 | ||
1826 | /* check whether implicit source operand (register R6) is readable */ | |
1827 | err = check_reg_arg(regs, BPF_REG_6, SRC_OP); | |
1828 | if (err) | |
1829 | return err; | |
1830 | ||
1831 | if (regs[BPF_REG_6].type != PTR_TO_CTX) { | |
1832 | verbose("at the time of BPF_LD_ABS|IND R6 != pointer to skb\n"); | |
1833 | return -EINVAL; | |
1834 | } | |
1835 | ||
1836 | if (mode == BPF_IND) { | |
1837 | /* check explicit source operand */ | |
1838 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
1839 | if (err) | |
1840 | return err; | |
1841 | } | |
1842 | ||
1843 | /* reset caller saved regs to unreadable */ | |
1844 | for (i = 0; i < CALLER_SAVED_REGS; i++) { | |
1845 | reg = regs + caller_saved[i]; | |
1846 | reg->type = NOT_INIT; | |
1847 | reg->imm = 0; | |
1848 | } | |
1849 | ||
1850 | /* mark destination R0 register as readable, since it contains | |
1851 | * the value fetched from the packet | |
1852 | */ | |
1853 | regs[BPF_REG_0].type = UNKNOWN_VALUE; | |
1854 | return 0; | |
1855 | } | |
1856 | ||
475fb78f AS |
1857 | /* non-recursive DFS pseudo code |
1858 | * 1 procedure DFS-iterative(G,v): | |
1859 | * 2 label v as discovered | |
1860 | * 3 let S be a stack | |
1861 | * 4 S.push(v) | |
1862 | * 5 while S is not empty | |
1863 | * 6 t <- S.pop() | |
1864 | * 7 if t is what we're looking for: | |
1865 | * 8 return t | |
1866 | * 9 for all edges e in G.adjacentEdges(t) do | |
1867 | * 10 if edge e is already labelled | |
1868 | * 11 continue with the next edge | |
1869 | * 12 w <- G.adjacentVertex(t,e) | |
1870 | * 13 if vertex w is not discovered and not explored | |
1871 | * 14 label e as tree-edge | |
1872 | * 15 label w as discovered | |
1873 | * 16 S.push(w) | |
1874 | * 17 continue at 5 | |
1875 | * 18 else if vertex w is discovered | |
1876 | * 19 label e as back-edge | |
1877 | * 20 else | |
1878 | * 21 // vertex w is explored | |
1879 | * 22 label e as forward- or cross-edge | |
1880 | * 23 label t as explored | |
1881 | * 24 S.pop() | |
1882 | * | |
1883 | * convention: | |
1884 | * 0x10 - discovered | |
1885 | * 0x11 - discovered and fall-through edge labelled | |
1886 | * 0x12 - discovered and fall-through and branch edges labelled | |
1887 | * 0x20 - explored | |
1888 | */ | |
1889 | ||
1890 | enum { | |
1891 | DISCOVERED = 0x10, | |
1892 | EXPLORED = 0x20, | |
1893 | FALLTHROUGH = 1, | |
1894 | BRANCH = 2, | |
1895 | }; | |
1896 | ||
58e2af8b | 1897 | #define STATE_LIST_MARK ((struct bpf_verifier_state_list *) -1L) |
f1bca824 | 1898 | |
475fb78f AS |
1899 | static int *insn_stack; /* stack of insns to process */ |
1900 | static int cur_stack; /* current stack index */ | |
1901 | static int *insn_state; | |
1902 | ||
1903 | /* t, w, e - match pseudo-code above: | |
1904 | * t - index of current instruction | |
1905 | * w - next instruction | |
1906 | * e - edge | |
1907 | */ | |
58e2af8b | 1908 | static int push_insn(int t, int w, int e, struct bpf_verifier_env *env) |
475fb78f AS |
1909 | { |
1910 | if (e == FALLTHROUGH && insn_state[t] >= (DISCOVERED | FALLTHROUGH)) | |
1911 | return 0; | |
1912 | ||
1913 | if (e == BRANCH && insn_state[t] >= (DISCOVERED | BRANCH)) | |
1914 | return 0; | |
1915 | ||
1916 | if (w < 0 || w >= env->prog->len) { | |
1917 | verbose("jump out of range from insn %d to %d\n", t, w); | |
1918 | return -EINVAL; | |
1919 | } | |
1920 | ||
f1bca824 AS |
1921 | if (e == BRANCH) |
1922 | /* mark branch target for state pruning */ | |
1923 | env->explored_states[w] = STATE_LIST_MARK; | |
1924 | ||
475fb78f AS |
1925 | if (insn_state[w] == 0) { |
1926 | /* tree-edge */ | |
1927 | insn_state[t] = DISCOVERED | e; | |
1928 | insn_state[w] = DISCOVERED; | |
1929 | if (cur_stack >= env->prog->len) | |
1930 | return -E2BIG; | |
1931 | insn_stack[cur_stack++] = w; | |
1932 | return 1; | |
1933 | } else if ((insn_state[w] & 0xF0) == DISCOVERED) { | |
1934 | verbose("back-edge from insn %d to %d\n", t, w); | |
1935 | return -EINVAL; | |
1936 | } else if (insn_state[w] == EXPLORED) { | |
1937 | /* forward- or cross-edge */ | |
1938 | insn_state[t] = DISCOVERED | e; | |
1939 | } else { | |
1940 | verbose("insn state internal bug\n"); | |
1941 | return -EFAULT; | |
1942 | } | |
1943 | return 0; | |
1944 | } | |
1945 | ||
1946 | /* non-recursive depth-first-search to detect loops in BPF program | |
1947 | * loop == back-edge in directed graph | |
1948 | */ | |
58e2af8b | 1949 | static int check_cfg(struct bpf_verifier_env *env) |
475fb78f AS |
1950 | { |
1951 | struct bpf_insn *insns = env->prog->insnsi; | |
1952 | int insn_cnt = env->prog->len; | |
1953 | int ret = 0; | |
1954 | int i, t; | |
1955 | ||
1956 | insn_state = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
1957 | if (!insn_state) | |
1958 | return -ENOMEM; | |
1959 | ||
1960 | insn_stack = kcalloc(insn_cnt, sizeof(int), GFP_KERNEL); | |
1961 | if (!insn_stack) { | |
1962 | kfree(insn_state); | |
1963 | return -ENOMEM; | |
1964 | } | |
1965 | ||
1966 | insn_state[0] = DISCOVERED; /* mark 1st insn as discovered */ | |
1967 | insn_stack[0] = 0; /* 0 is the first instruction */ | |
1968 | cur_stack = 1; | |
1969 | ||
1970 | peek_stack: | |
1971 | if (cur_stack == 0) | |
1972 | goto check_state; | |
1973 | t = insn_stack[cur_stack - 1]; | |
1974 | ||
1975 | if (BPF_CLASS(insns[t].code) == BPF_JMP) { | |
1976 | u8 opcode = BPF_OP(insns[t].code); | |
1977 | ||
1978 | if (opcode == BPF_EXIT) { | |
1979 | goto mark_explored; | |
1980 | } else if (opcode == BPF_CALL) { | |
1981 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
1982 | if (ret == 1) | |
1983 | goto peek_stack; | |
1984 | else if (ret < 0) | |
1985 | goto err_free; | |
07016151 DB |
1986 | if (t + 1 < insn_cnt) |
1987 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
1988 | } else if (opcode == BPF_JA) { |
1989 | if (BPF_SRC(insns[t].code) != BPF_K) { | |
1990 | ret = -EINVAL; | |
1991 | goto err_free; | |
1992 | } | |
1993 | /* unconditional jump with single edge */ | |
1994 | ret = push_insn(t, t + insns[t].off + 1, | |
1995 | FALLTHROUGH, env); | |
1996 | if (ret == 1) | |
1997 | goto peek_stack; | |
1998 | else if (ret < 0) | |
1999 | goto err_free; | |
f1bca824 AS |
2000 | /* tell verifier to check for equivalent states |
2001 | * after every call and jump | |
2002 | */ | |
c3de6317 AS |
2003 | if (t + 1 < insn_cnt) |
2004 | env->explored_states[t + 1] = STATE_LIST_MARK; | |
475fb78f AS |
2005 | } else { |
2006 | /* conditional jump with two edges */ | |
2007 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2008 | if (ret == 1) | |
2009 | goto peek_stack; | |
2010 | else if (ret < 0) | |
2011 | goto err_free; | |
2012 | ||
2013 | ret = push_insn(t, t + insns[t].off + 1, BRANCH, env); | |
2014 | if (ret == 1) | |
2015 | goto peek_stack; | |
2016 | else if (ret < 0) | |
2017 | goto err_free; | |
2018 | } | |
2019 | } else { | |
2020 | /* all other non-branch instructions with single | |
2021 | * fall-through edge | |
2022 | */ | |
2023 | ret = push_insn(t, t + 1, FALLTHROUGH, env); | |
2024 | if (ret == 1) | |
2025 | goto peek_stack; | |
2026 | else if (ret < 0) | |
2027 | goto err_free; | |
2028 | } | |
2029 | ||
2030 | mark_explored: | |
2031 | insn_state[t] = EXPLORED; | |
2032 | if (cur_stack-- <= 0) { | |
2033 | verbose("pop stack internal bug\n"); | |
2034 | ret = -EFAULT; | |
2035 | goto err_free; | |
2036 | } | |
2037 | goto peek_stack; | |
2038 | ||
2039 | check_state: | |
2040 | for (i = 0; i < insn_cnt; i++) { | |
2041 | if (insn_state[i] != EXPLORED) { | |
2042 | verbose("unreachable insn %d\n", i); | |
2043 | ret = -EINVAL; | |
2044 | goto err_free; | |
2045 | } | |
2046 | } | |
2047 | ret = 0; /* cfg looks good */ | |
2048 | ||
2049 | err_free: | |
2050 | kfree(insn_state); | |
2051 | kfree(insn_stack); | |
2052 | return ret; | |
2053 | } | |
2054 | ||
969bf05e AS |
2055 | /* the following conditions reduce the number of explored insns |
2056 | * from ~140k to ~80k for ultra large programs that use a lot of ptr_to_packet | |
2057 | */ | |
58e2af8b JK |
2058 | static bool compare_ptrs_to_packet(struct bpf_reg_state *old, |
2059 | struct bpf_reg_state *cur) | |
969bf05e AS |
2060 | { |
2061 | if (old->id != cur->id) | |
2062 | return false; | |
2063 | ||
2064 | /* old ptr_to_packet is more conservative, since it allows smaller | |
2065 | * range. Ex: | |
2066 | * old(off=0,r=10) is equal to cur(off=0,r=20), because | |
2067 | * old(off=0,r=10) means that with range=10 the verifier proceeded | |
2068 | * further and found no issues with the program. Now we're in the same | |
2069 | * spot with cur(off=0,r=20), so we're safe too, since anything further | |
2070 | * will only be looking at most 10 bytes after this pointer. | |
2071 | */ | |
2072 | if (old->off == cur->off && old->range < cur->range) | |
2073 | return true; | |
2074 | ||
2075 | /* old(off=20,r=10) is equal to cur(off=22,re=22 or 5 or 0) | |
2076 | * since both cannot be used for packet access and safe(old) | |
2077 | * pointer has smaller off that could be used for further | |
2078 | * 'if (ptr > data_end)' check | |
2079 | * Ex: | |
2080 | * old(off=20,r=10) and cur(off=22,r=22) and cur(off=22,r=0) mean | |
2081 | * that we cannot access the packet. | |
2082 | * The safe range is: | |
2083 | * [ptr, ptr + range - off) | |
2084 | * so whenever off >=range, it means no safe bytes from this pointer. | |
2085 | * When comparing old->off <= cur->off, it means that older code | |
2086 | * went with smaller offset and that offset was later | |
2087 | * used to figure out the safe range after 'if (ptr > data_end)' check | |
2088 | * Say, 'old' state was explored like: | |
2089 | * ... R3(off=0, r=0) | |
2090 | * R4 = R3 + 20 | |
2091 | * ... now R4(off=20,r=0) <-- here | |
2092 | * if (R4 > data_end) | |
2093 | * ... R4(off=20,r=20), R3(off=0,r=20) and R3 can be used to access. | |
2094 | * ... the code further went all the way to bpf_exit. | |
2095 | * Now the 'cur' state at the mark 'here' has R4(off=30,r=0). | |
2096 | * old_R4(off=20,r=0) equal to cur_R4(off=30,r=0), since if the verifier | |
2097 | * goes further, such cur_R4 will give larger safe packet range after | |
2098 | * 'if (R4 > data_end)' and all further insn were already good with r=20, | |
2099 | * so they will be good with r=30 and we can prune the search. | |
2100 | */ | |
2101 | if (old->off <= cur->off && | |
2102 | old->off >= old->range && cur->off >= cur->range) | |
2103 | return true; | |
2104 | ||
2105 | return false; | |
2106 | } | |
2107 | ||
f1bca824 AS |
2108 | /* compare two verifier states |
2109 | * | |
2110 | * all states stored in state_list are known to be valid, since | |
2111 | * verifier reached 'bpf_exit' instruction through them | |
2112 | * | |
2113 | * this function is called when verifier exploring different branches of | |
2114 | * execution popped from the state stack. If it sees an old state that has | |
2115 | * more strict register state and more strict stack state then this execution | |
2116 | * branch doesn't need to be explored further, since verifier already | |
2117 | * concluded that more strict state leads to valid finish. | |
2118 | * | |
2119 | * Therefore two states are equivalent if register state is more conservative | |
2120 | * and explored stack state is more conservative than the current one. | |
2121 | * Example: | |
2122 | * explored current | |
2123 | * (slot1=INV slot2=MISC) == (slot1=MISC slot2=MISC) | |
2124 | * (slot1=MISC slot2=MISC) != (slot1=INV slot2=MISC) | |
2125 | * | |
2126 | * In other words if current stack state (one being explored) has more | |
2127 | * valid slots than old one that already passed validation, it means | |
2128 | * the verifier can stop exploring and conclude that current state is valid too | |
2129 | * | |
2130 | * Similarly with registers. If explored state has register type as invalid | |
2131 | * whereas register type in current state is meaningful, it means that | |
2132 | * the current state will reach 'bpf_exit' instruction safely | |
2133 | */ | |
58e2af8b JK |
2134 | static bool states_equal(struct bpf_verifier_state *old, |
2135 | struct bpf_verifier_state *cur) | |
f1bca824 | 2136 | { |
58e2af8b | 2137 | struct bpf_reg_state *rold, *rcur; |
f1bca824 AS |
2138 | int i; |
2139 | ||
2140 | for (i = 0; i < MAX_BPF_REG; i++) { | |
1a0dc1ac AS |
2141 | rold = &old->regs[i]; |
2142 | rcur = &cur->regs[i]; | |
2143 | ||
2144 | if (memcmp(rold, rcur, sizeof(*rold)) == 0) | |
2145 | continue; | |
2146 | ||
2147 | if (rold->type == NOT_INIT || | |
2148 | (rold->type == UNKNOWN_VALUE && rcur->type != NOT_INIT)) | |
2149 | continue; | |
2150 | ||
969bf05e AS |
2151 | if (rold->type == PTR_TO_PACKET && rcur->type == PTR_TO_PACKET && |
2152 | compare_ptrs_to_packet(rold, rcur)) | |
2153 | continue; | |
2154 | ||
1a0dc1ac | 2155 | return false; |
f1bca824 AS |
2156 | } |
2157 | ||
2158 | for (i = 0; i < MAX_BPF_STACK; i++) { | |
9c399760 AS |
2159 | if (old->stack_slot_type[i] == STACK_INVALID) |
2160 | continue; | |
2161 | if (old->stack_slot_type[i] != cur->stack_slot_type[i]) | |
2162 | /* Ex: old explored (safe) state has STACK_SPILL in | |
2163 | * this stack slot, but current has has STACK_MISC -> | |
2164 | * this verifier states are not equivalent, | |
2165 | * return false to continue verification of this path | |
2166 | */ | |
f1bca824 | 2167 | return false; |
9c399760 AS |
2168 | if (i % BPF_REG_SIZE) |
2169 | continue; | |
2170 | if (memcmp(&old->spilled_regs[i / BPF_REG_SIZE], | |
2171 | &cur->spilled_regs[i / BPF_REG_SIZE], | |
2172 | sizeof(old->spilled_regs[0]))) | |
2173 | /* when explored and current stack slot types are | |
2174 | * the same, check that stored pointers types | |
2175 | * are the same as well. | |
2176 | * Ex: explored safe path could have stored | |
58e2af8b | 2177 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -8} |
9c399760 | 2178 | * but current path has stored: |
58e2af8b | 2179 | * (bpf_reg_state) {.type = PTR_TO_STACK, .imm = -16} |
9c399760 AS |
2180 | * such verifier states are not equivalent. |
2181 | * return false to continue verification of this path | |
2182 | */ | |
2183 | return false; | |
2184 | else | |
2185 | continue; | |
f1bca824 AS |
2186 | } |
2187 | return true; | |
2188 | } | |
2189 | ||
58e2af8b | 2190 | static int is_state_visited(struct bpf_verifier_env *env, int insn_idx) |
f1bca824 | 2191 | { |
58e2af8b JK |
2192 | struct bpf_verifier_state_list *new_sl; |
2193 | struct bpf_verifier_state_list *sl; | |
f1bca824 AS |
2194 | |
2195 | sl = env->explored_states[insn_idx]; | |
2196 | if (!sl) | |
2197 | /* this 'insn_idx' instruction wasn't marked, so we will not | |
2198 | * be doing state search here | |
2199 | */ | |
2200 | return 0; | |
2201 | ||
2202 | while (sl != STATE_LIST_MARK) { | |
2203 | if (states_equal(&sl->state, &env->cur_state)) | |
2204 | /* reached equivalent register/stack state, | |
2205 | * prune the search | |
2206 | */ | |
2207 | return 1; | |
2208 | sl = sl->next; | |
2209 | } | |
2210 | ||
2211 | /* there were no equivalent states, remember current one. | |
2212 | * technically the current state is not proven to be safe yet, | |
2213 | * but it will either reach bpf_exit (which means it's safe) or | |
2214 | * it will be rejected. Since there are no loops, we won't be | |
2215 | * seeing this 'insn_idx' instruction again on the way to bpf_exit | |
2216 | */ | |
58e2af8b | 2217 | new_sl = kmalloc(sizeof(struct bpf_verifier_state_list), GFP_USER); |
f1bca824 AS |
2218 | if (!new_sl) |
2219 | return -ENOMEM; | |
2220 | ||
2221 | /* add new state to the head of linked list */ | |
2222 | memcpy(&new_sl->state, &env->cur_state, sizeof(env->cur_state)); | |
2223 | new_sl->next = env->explored_states[insn_idx]; | |
2224 | env->explored_states[insn_idx] = new_sl; | |
2225 | return 0; | |
2226 | } | |
2227 | ||
58e2af8b | 2228 | static int do_check(struct bpf_verifier_env *env) |
17a52670 | 2229 | { |
58e2af8b | 2230 | struct bpf_verifier_state *state = &env->cur_state; |
17a52670 | 2231 | struct bpf_insn *insns = env->prog->insnsi; |
58e2af8b | 2232 | struct bpf_reg_state *regs = state->regs; |
17a52670 AS |
2233 | int insn_cnt = env->prog->len; |
2234 | int insn_idx, prev_insn_idx = 0; | |
2235 | int insn_processed = 0; | |
2236 | bool do_print_state = false; | |
2237 | ||
2238 | init_reg_state(regs); | |
2239 | insn_idx = 0; | |
2240 | for (;;) { | |
2241 | struct bpf_insn *insn; | |
2242 | u8 class; | |
2243 | int err; | |
2244 | ||
2245 | if (insn_idx >= insn_cnt) { | |
2246 | verbose("invalid insn idx %d insn_cnt %d\n", | |
2247 | insn_idx, insn_cnt); | |
2248 | return -EFAULT; | |
2249 | } | |
2250 | ||
2251 | insn = &insns[insn_idx]; | |
2252 | class = BPF_CLASS(insn->code); | |
2253 | ||
07016151 | 2254 | if (++insn_processed > BPF_COMPLEXITY_LIMIT_INSNS) { |
17a52670 AS |
2255 | verbose("BPF program is too large. Proccessed %d insn\n", |
2256 | insn_processed); | |
2257 | return -E2BIG; | |
2258 | } | |
2259 | ||
f1bca824 AS |
2260 | err = is_state_visited(env, insn_idx); |
2261 | if (err < 0) | |
2262 | return err; | |
2263 | if (err == 1) { | |
2264 | /* found equivalent state, can prune the search */ | |
2265 | if (log_level) { | |
2266 | if (do_print_state) | |
2267 | verbose("\nfrom %d to %d: safe\n", | |
2268 | prev_insn_idx, insn_idx); | |
2269 | else | |
2270 | verbose("%d: safe\n", insn_idx); | |
2271 | } | |
2272 | goto process_bpf_exit; | |
2273 | } | |
2274 | ||
17a52670 AS |
2275 | if (log_level && do_print_state) { |
2276 | verbose("\nfrom %d to %d:", prev_insn_idx, insn_idx); | |
1a0dc1ac | 2277 | print_verifier_state(&env->cur_state); |
17a52670 AS |
2278 | do_print_state = false; |
2279 | } | |
2280 | ||
2281 | if (log_level) { | |
2282 | verbose("%d: ", insn_idx); | |
2283 | print_bpf_insn(insn); | |
2284 | } | |
2285 | ||
2286 | if (class == BPF_ALU || class == BPF_ALU64) { | |
1be7f75d | 2287 | err = check_alu_op(env, insn); |
17a52670 AS |
2288 | if (err) |
2289 | return err; | |
2290 | ||
2291 | } else if (class == BPF_LDX) { | |
3df126f3 | 2292 | enum bpf_reg_type *prev_src_type, src_reg_type; |
9bac3d6d AS |
2293 | |
2294 | /* check for reserved fields is already done */ | |
2295 | ||
17a52670 AS |
2296 | /* check src operand */ |
2297 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2298 | if (err) | |
2299 | return err; | |
2300 | ||
2301 | err = check_reg_arg(regs, insn->dst_reg, DST_OP_NO_MARK); | |
2302 | if (err) | |
2303 | return err; | |
2304 | ||
725f9dcd AS |
2305 | src_reg_type = regs[insn->src_reg].type; |
2306 | ||
17a52670 AS |
2307 | /* check that memory (src_reg + off) is readable, |
2308 | * the state of dst_reg will be updated by this func | |
2309 | */ | |
2310 | err = check_mem_access(env, insn->src_reg, insn->off, | |
2311 | BPF_SIZE(insn->code), BPF_READ, | |
2312 | insn->dst_reg); | |
2313 | if (err) | |
2314 | return err; | |
2315 | ||
ea2e7ce5 AS |
2316 | if (BPF_SIZE(insn->code) != BPF_W && |
2317 | BPF_SIZE(insn->code) != BPF_DW) { | |
725f9dcd AS |
2318 | insn_idx++; |
2319 | continue; | |
2320 | } | |
9bac3d6d | 2321 | |
3df126f3 JK |
2322 | prev_src_type = &env->insn_aux_data[insn_idx].ptr_type; |
2323 | ||
2324 | if (*prev_src_type == NOT_INIT) { | |
9bac3d6d AS |
2325 | /* saw a valid insn |
2326 | * dst_reg = *(u32 *)(src_reg + off) | |
3df126f3 | 2327 | * save type to validate intersecting paths |
9bac3d6d | 2328 | */ |
3df126f3 | 2329 | *prev_src_type = src_reg_type; |
9bac3d6d | 2330 | |
3df126f3 | 2331 | } else if (src_reg_type != *prev_src_type && |
9bac3d6d | 2332 | (src_reg_type == PTR_TO_CTX || |
3df126f3 | 2333 | *prev_src_type == PTR_TO_CTX)) { |
9bac3d6d AS |
2334 | /* ABuser program is trying to use the same insn |
2335 | * dst_reg = *(u32*) (src_reg + off) | |
2336 | * with different pointer types: | |
2337 | * src_reg == ctx in one branch and | |
2338 | * src_reg == stack|map in some other branch. | |
2339 | * Reject it. | |
2340 | */ | |
2341 | verbose("same insn cannot be used with different pointers\n"); | |
2342 | return -EINVAL; | |
2343 | } | |
2344 | ||
17a52670 | 2345 | } else if (class == BPF_STX) { |
3df126f3 | 2346 | enum bpf_reg_type *prev_dst_type, dst_reg_type; |
d691f9e8 | 2347 | |
17a52670 AS |
2348 | if (BPF_MODE(insn->code) == BPF_XADD) { |
2349 | err = check_xadd(env, insn); | |
2350 | if (err) | |
2351 | return err; | |
2352 | insn_idx++; | |
2353 | continue; | |
2354 | } | |
2355 | ||
17a52670 AS |
2356 | /* check src1 operand */ |
2357 | err = check_reg_arg(regs, insn->src_reg, SRC_OP); | |
2358 | if (err) | |
2359 | return err; | |
2360 | /* check src2 operand */ | |
2361 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2362 | if (err) | |
2363 | return err; | |
2364 | ||
d691f9e8 AS |
2365 | dst_reg_type = regs[insn->dst_reg].type; |
2366 | ||
17a52670 AS |
2367 | /* check that memory (dst_reg + off) is writeable */ |
2368 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2369 | BPF_SIZE(insn->code), BPF_WRITE, | |
2370 | insn->src_reg); | |
2371 | if (err) | |
2372 | return err; | |
2373 | ||
3df126f3 JK |
2374 | prev_dst_type = &env->insn_aux_data[insn_idx].ptr_type; |
2375 | ||
2376 | if (*prev_dst_type == NOT_INIT) { | |
2377 | *prev_dst_type = dst_reg_type; | |
2378 | } else if (dst_reg_type != *prev_dst_type && | |
d691f9e8 | 2379 | (dst_reg_type == PTR_TO_CTX || |
3df126f3 | 2380 | *prev_dst_type == PTR_TO_CTX)) { |
d691f9e8 AS |
2381 | verbose("same insn cannot be used with different pointers\n"); |
2382 | return -EINVAL; | |
2383 | } | |
2384 | ||
17a52670 AS |
2385 | } else if (class == BPF_ST) { |
2386 | if (BPF_MODE(insn->code) != BPF_MEM || | |
2387 | insn->src_reg != BPF_REG_0) { | |
2388 | verbose("BPF_ST uses reserved fields\n"); | |
2389 | return -EINVAL; | |
2390 | } | |
2391 | /* check src operand */ | |
2392 | err = check_reg_arg(regs, insn->dst_reg, SRC_OP); | |
2393 | if (err) | |
2394 | return err; | |
2395 | ||
2396 | /* check that memory (dst_reg + off) is writeable */ | |
2397 | err = check_mem_access(env, insn->dst_reg, insn->off, | |
2398 | BPF_SIZE(insn->code), BPF_WRITE, | |
2399 | -1); | |
2400 | if (err) | |
2401 | return err; | |
2402 | ||
2403 | } else if (class == BPF_JMP) { | |
2404 | u8 opcode = BPF_OP(insn->code); | |
2405 | ||
2406 | if (opcode == BPF_CALL) { | |
2407 | if (BPF_SRC(insn->code) != BPF_K || | |
2408 | insn->off != 0 || | |
2409 | insn->src_reg != BPF_REG_0 || | |
2410 | insn->dst_reg != BPF_REG_0) { | |
2411 | verbose("BPF_CALL uses reserved fields\n"); | |
2412 | return -EINVAL; | |
2413 | } | |
2414 | ||
2415 | err = check_call(env, insn->imm); | |
2416 | if (err) | |
2417 | return err; | |
2418 | ||
2419 | } else if (opcode == BPF_JA) { | |
2420 | if (BPF_SRC(insn->code) != BPF_K || | |
2421 | insn->imm != 0 || | |
2422 | insn->src_reg != BPF_REG_0 || | |
2423 | insn->dst_reg != BPF_REG_0) { | |
2424 | verbose("BPF_JA uses reserved fields\n"); | |
2425 | return -EINVAL; | |
2426 | } | |
2427 | ||
2428 | insn_idx += insn->off + 1; | |
2429 | continue; | |
2430 | ||
2431 | } else if (opcode == BPF_EXIT) { | |
2432 | if (BPF_SRC(insn->code) != BPF_K || | |
2433 | insn->imm != 0 || | |
2434 | insn->src_reg != BPF_REG_0 || | |
2435 | insn->dst_reg != BPF_REG_0) { | |
2436 | verbose("BPF_EXIT uses reserved fields\n"); | |
2437 | return -EINVAL; | |
2438 | } | |
2439 | ||
2440 | /* eBPF calling convetion is such that R0 is used | |
2441 | * to return the value from eBPF program. | |
2442 | * Make sure that it's readable at this time | |
2443 | * of bpf_exit, which means that program wrote | |
2444 | * something into it earlier | |
2445 | */ | |
2446 | err = check_reg_arg(regs, BPF_REG_0, SRC_OP); | |
2447 | if (err) | |
2448 | return err; | |
2449 | ||
1be7f75d AS |
2450 | if (is_pointer_value(env, BPF_REG_0)) { |
2451 | verbose("R0 leaks addr as return value\n"); | |
2452 | return -EACCES; | |
2453 | } | |
2454 | ||
f1bca824 | 2455 | process_bpf_exit: |
17a52670 AS |
2456 | insn_idx = pop_stack(env, &prev_insn_idx); |
2457 | if (insn_idx < 0) { | |
2458 | break; | |
2459 | } else { | |
2460 | do_print_state = true; | |
2461 | continue; | |
2462 | } | |
2463 | } else { | |
2464 | err = check_cond_jmp_op(env, insn, &insn_idx); | |
2465 | if (err) | |
2466 | return err; | |
2467 | } | |
2468 | } else if (class == BPF_LD) { | |
2469 | u8 mode = BPF_MODE(insn->code); | |
2470 | ||
2471 | if (mode == BPF_ABS || mode == BPF_IND) { | |
ddd872bc AS |
2472 | err = check_ld_abs(env, insn); |
2473 | if (err) | |
2474 | return err; | |
2475 | ||
17a52670 AS |
2476 | } else if (mode == BPF_IMM) { |
2477 | err = check_ld_imm(env, insn); | |
2478 | if (err) | |
2479 | return err; | |
2480 | ||
2481 | insn_idx++; | |
2482 | } else { | |
2483 | verbose("invalid BPF_LD mode\n"); | |
2484 | return -EINVAL; | |
2485 | } | |
2486 | } else { | |
2487 | verbose("unknown insn class %d\n", class); | |
2488 | return -EINVAL; | |
2489 | } | |
2490 | ||
2491 | insn_idx++; | |
2492 | } | |
2493 | ||
1a0dc1ac | 2494 | verbose("processed %d insns\n", insn_processed); |
17a52670 AS |
2495 | return 0; |
2496 | } | |
2497 | ||
fdc15d38 AS |
2498 | static int check_map_prog_compatibility(struct bpf_map *map, |
2499 | struct bpf_prog *prog) | |
2500 | ||
2501 | { | |
2502 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT && | |
2503 | (map->map_type == BPF_MAP_TYPE_HASH || | |
2504 | map->map_type == BPF_MAP_TYPE_PERCPU_HASH) && | |
2505 | (map->map_flags & BPF_F_NO_PREALLOC)) { | |
2506 | verbose("perf_event programs can only use preallocated hash map\n"); | |
2507 | return -EINVAL; | |
2508 | } | |
2509 | return 0; | |
2510 | } | |
2511 | ||
0246e64d AS |
2512 | /* look for pseudo eBPF instructions that access map FDs and |
2513 | * replace them with actual map pointers | |
2514 | */ | |
58e2af8b | 2515 | static int replace_map_fd_with_map_ptr(struct bpf_verifier_env *env) |
0246e64d AS |
2516 | { |
2517 | struct bpf_insn *insn = env->prog->insnsi; | |
2518 | int insn_cnt = env->prog->len; | |
fdc15d38 | 2519 | int i, j, err; |
0246e64d AS |
2520 | |
2521 | for (i = 0; i < insn_cnt; i++, insn++) { | |
9bac3d6d | 2522 | if (BPF_CLASS(insn->code) == BPF_LDX && |
d691f9e8 | 2523 | (BPF_MODE(insn->code) != BPF_MEM || insn->imm != 0)) { |
9bac3d6d AS |
2524 | verbose("BPF_LDX uses reserved fields\n"); |
2525 | return -EINVAL; | |
2526 | } | |
2527 | ||
d691f9e8 AS |
2528 | if (BPF_CLASS(insn->code) == BPF_STX && |
2529 | ((BPF_MODE(insn->code) != BPF_MEM && | |
2530 | BPF_MODE(insn->code) != BPF_XADD) || insn->imm != 0)) { | |
2531 | verbose("BPF_STX uses reserved fields\n"); | |
2532 | return -EINVAL; | |
2533 | } | |
2534 | ||
0246e64d AS |
2535 | if (insn[0].code == (BPF_LD | BPF_IMM | BPF_DW)) { |
2536 | struct bpf_map *map; | |
2537 | struct fd f; | |
2538 | ||
2539 | if (i == insn_cnt - 1 || insn[1].code != 0 || | |
2540 | insn[1].dst_reg != 0 || insn[1].src_reg != 0 || | |
2541 | insn[1].off != 0) { | |
2542 | verbose("invalid bpf_ld_imm64 insn\n"); | |
2543 | return -EINVAL; | |
2544 | } | |
2545 | ||
2546 | if (insn->src_reg == 0) | |
2547 | /* valid generic load 64-bit imm */ | |
2548 | goto next_insn; | |
2549 | ||
2550 | if (insn->src_reg != BPF_PSEUDO_MAP_FD) { | |
2551 | verbose("unrecognized bpf_ld_imm64 insn\n"); | |
2552 | return -EINVAL; | |
2553 | } | |
2554 | ||
2555 | f = fdget(insn->imm); | |
c2101297 | 2556 | map = __bpf_map_get(f); |
0246e64d AS |
2557 | if (IS_ERR(map)) { |
2558 | verbose("fd %d is not pointing to valid bpf_map\n", | |
2559 | insn->imm); | |
0246e64d AS |
2560 | return PTR_ERR(map); |
2561 | } | |
2562 | ||
fdc15d38 AS |
2563 | err = check_map_prog_compatibility(map, env->prog); |
2564 | if (err) { | |
2565 | fdput(f); | |
2566 | return err; | |
2567 | } | |
2568 | ||
0246e64d AS |
2569 | /* store map pointer inside BPF_LD_IMM64 instruction */ |
2570 | insn[0].imm = (u32) (unsigned long) map; | |
2571 | insn[1].imm = ((u64) (unsigned long) map) >> 32; | |
2572 | ||
2573 | /* check whether we recorded this map already */ | |
2574 | for (j = 0; j < env->used_map_cnt; j++) | |
2575 | if (env->used_maps[j] == map) { | |
2576 | fdput(f); | |
2577 | goto next_insn; | |
2578 | } | |
2579 | ||
2580 | if (env->used_map_cnt >= MAX_USED_MAPS) { | |
2581 | fdput(f); | |
2582 | return -E2BIG; | |
2583 | } | |
2584 | ||
0246e64d AS |
2585 | /* hold the map. If the program is rejected by verifier, |
2586 | * the map will be released by release_maps() or it | |
2587 | * will be used by the valid program until it's unloaded | |
2588 | * and all maps are released in free_bpf_prog_info() | |
2589 | */ | |
92117d84 AS |
2590 | map = bpf_map_inc(map, false); |
2591 | if (IS_ERR(map)) { | |
2592 | fdput(f); | |
2593 | return PTR_ERR(map); | |
2594 | } | |
2595 | env->used_maps[env->used_map_cnt++] = map; | |
2596 | ||
0246e64d AS |
2597 | fdput(f); |
2598 | next_insn: | |
2599 | insn++; | |
2600 | i++; | |
2601 | } | |
2602 | } | |
2603 | ||
2604 | /* now all pseudo BPF_LD_IMM64 instructions load valid | |
2605 | * 'struct bpf_map *' into a register instead of user map_fd. | |
2606 | * These pointers will be used later by verifier to validate map access. | |
2607 | */ | |
2608 | return 0; | |
2609 | } | |
2610 | ||
2611 | /* drop refcnt of maps used by the rejected program */ | |
58e2af8b | 2612 | static void release_maps(struct bpf_verifier_env *env) |
0246e64d AS |
2613 | { |
2614 | int i; | |
2615 | ||
2616 | for (i = 0; i < env->used_map_cnt; i++) | |
2617 | bpf_map_put(env->used_maps[i]); | |
2618 | } | |
2619 | ||
2620 | /* convert pseudo BPF_LD_IMM64 into generic BPF_LD_IMM64 */ | |
58e2af8b | 2621 | static void convert_pseudo_ld_imm64(struct bpf_verifier_env *env) |
0246e64d AS |
2622 | { |
2623 | struct bpf_insn *insn = env->prog->insnsi; | |
2624 | int insn_cnt = env->prog->len; | |
2625 | int i; | |
2626 | ||
2627 | for (i = 0; i < insn_cnt; i++, insn++) | |
2628 | if (insn->code == (BPF_LD | BPF_IMM | BPF_DW)) | |
2629 | insn->src_reg = 0; | |
2630 | } | |
2631 | ||
9bac3d6d AS |
2632 | /* convert load instructions that access fields of 'struct __sk_buff' |
2633 | * into sequence of instructions that access fields of 'struct sk_buff' | |
2634 | */ | |
58e2af8b | 2635 | static int convert_ctx_accesses(struct bpf_verifier_env *env) |
9bac3d6d | 2636 | { |
36bbef52 | 2637 | const struct bpf_verifier_ops *ops = env->prog->aux->ops; |
3df126f3 | 2638 | const int insn_cnt = env->prog->len; |
36bbef52 | 2639 | struct bpf_insn insn_buf[16], *insn; |
9bac3d6d | 2640 | struct bpf_prog *new_prog; |
d691f9e8 | 2641 | enum bpf_access_type type; |
3df126f3 | 2642 | int i, cnt, delta = 0; |
9bac3d6d | 2643 | |
36bbef52 DB |
2644 | if (ops->gen_prologue) { |
2645 | cnt = ops->gen_prologue(insn_buf, env->seen_direct_write, | |
2646 | env->prog); | |
2647 | if (cnt >= ARRAY_SIZE(insn_buf)) { | |
2648 | verbose("bpf verifier is misconfigured\n"); | |
2649 | return -EINVAL; | |
2650 | } else if (cnt) { | |
2651 | new_prog = bpf_patch_insn_single(env->prog, 0, | |
2652 | insn_buf, cnt); | |
2653 | if (!new_prog) | |
2654 | return -ENOMEM; | |
2655 | env->prog = new_prog; | |
3df126f3 | 2656 | delta += cnt - 1; |
36bbef52 DB |
2657 | } |
2658 | } | |
2659 | ||
2660 | if (!ops->convert_ctx_access) | |
9bac3d6d AS |
2661 | return 0; |
2662 | ||
3df126f3 | 2663 | insn = env->prog->insnsi + delta; |
36bbef52 | 2664 | |
9bac3d6d | 2665 | for (i = 0; i < insn_cnt; i++, insn++) { |
ea2e7ce5 AS |
2666 | if (insn->code == (BPF_LDX | BPF_MEM | BPF_W) || |
2667 | insn->code == (BPF_LDX | BPF_MEM | BPF_DW)) | |
d691f9e8 | 2668 | type = BPF_READ; |
ea2e7ce5 AS |
2669 | else if (insn->code == (BPF_STX | BPF_MEM | BPF_W) || |
2670 | insn->code == (BPF_STX | BPF_MEM | BPF_DW)) | |
d691f9e8 AS |
2671 | type = BPF_WRITE; |
2672 | else | |
9bac3d6d AS |
2673 | continue; |
2674 | ||
3df126f3 | 2675 | if (env->insn_aux_data[i].ptr_type != PTR_TO_CTX) |
9bac3d6d | 2676 | continue; |
9bac3d6d | 2677 | |
36bbef52 DB |
2678 | cnt = ops->convert_ctx_access(type, insn->dst_reg, insn->src_reg, |
2679 | insn->off, insn_buf, env->prog); | |
9bac3d6d AS |
2680 | if (cnt == 0 || cnt >= ARRAY_SIZE(insn_buf)) { |
2681 | verbose("bpf verifier is misconfigured\n"); | |
2682 | return -EINVAL; | |
2683 | } | |
2684 | ||
3df126f3 JK |
2685 | new_prog = bpf_patch_insn_single(env->prog, i + delta, insn_buf, |
2686 | cnt); | |
9bac3d6d AS |
2687 | if (!new_prog) |
2688 | return -ENOMEM; | |
2689 | ||
3df126f3 | 2690 | delta += cnt - 1; |
9bac3d6d AS |
2691 | |
2692 | /* keep walking new program and skip insns we just inserted */ | |
2693 | env->prog = new_prog; | |
3df126f3 | 2694 | insn = new_prog->insnsi + i + delta; |
9bac3d6d AS |
2695 | } |
2696 | ||
2697 | return 0; | |
2698 | } | |
2699 | ||
58e2af8b | 2700 | static void free_states(struct bpf_verifier_env *env) |
f1bca824 | 2701 | { |
58e2af8b | 2702 | struct bpf_verifier_state_list *sl, *sln; |
f1bca824 AS |
2703 | int i; |
2704 | ||
2705 | if (!env->explored_states) | |
2706 | return; | |
2707 | ||
2708 | for (i = 0; i < env->prog->len; i++) { | |
2709 | sl = env->explored_states[i]; | |
2710 | ||
2711 | if (sl) | |
2712 | while (sl != STATE_LIST_MARK) { | |
2713 | sln = sl->next; | |
2714 | kfree(sl); | |
2715 | sl = sln; | |
2716 | } | |
2717 | } | |
2718 | ||
2719 | kfree(env->explored_states); | |
2720 | } | |
2721 | ||
9bac3d6d | 2722 | int bpf_check(struct bpf_prog **prog, union bpf_attr *attr) |
51580e79 | 2723 | { |
cbd35700 | 2724 | char __user *log_ubuf = NULL; |
58e2af8b | 2725 | struct bpf_verifier_env *env; |
51580e79 AS |
2726 | int ret = -EINVAL; |
2727 | ||
9bac3d6d | 2728 | if ((*prog)->len <= 0 || (*prog)->len > BPF_MAXINSNS) |
cbd35700 AS |
2729 | return -E2BIG; |
2730 | ||
58e2af8b | 2731 | /* 'struct bpf_verifier_env' can be global, but since it's not small, |
cbd35700 AS |
2732 | * allocate/free it every time bpf_check() is called |
2733 | */ | |
58e2af8b | 2734 | env = kzalloc(sizeof(struct bpf_verifier_env), GFP_KERNEL); |
cbd35700 AS |
2735 | if (!env) |
2736 | return -ENOMEM; | |
2737 | ||
3df126f3 JK |
2738 | env->insn_aux_data = vzalloc(sizeof(struct bpf_insn_aux_data) * |
2739 | (*prog)->len); | |
2740 | ret = -ENOMEM; | |
2741 | if (!env->insn_aux_data) | |
2742 | goto err_free_env; | |
9bac3d6d | 2743 | env->prog = *prog; |
0246e64d | 2744 | |
cbd35700 AS |
2745 | /* grab the mutex to protect few globals used by verifier */ |
2746 | mutex_lock(&bpf_verifier_lock); | |
2747 | ||
2748 | if (attr->log_level || attr->log_buf || attr->log_size) { | |
2749 | /* user requested verbose verifier output | |
2750 | * and supplied buffer to store the verification trace | |
2751 | */ | |
2752 | log_level = attr->log_level; | |
2753 | log_ubuf = (char __user *) (unsigned long) attr->log_buf; | |
2754 | log_size = attr->log_size; | |
2755 | log_len = 0; | |
2756 | ||
2757 | ret = -EINVAL; | |
2758 | /* log_* values have to be sane */ | |
2759 | if (log_size < 128 || log_size > UINT_MAX >> 8 || | |
2760 | log_level == 0 || log_ubuf == NULL) | |
3df126f3 | 2761 | goto err_unlock; |
cbd35700 AS |
2762 | |
2763 | ret = -ENOMEM; | |
2764 | log_buf = vmalloc(log_size); | |
2765 | if (!log_buf) | |
3df126f3 | 2766 | goto err_unlock; |
cbd35700 AS |
2767 | } else { |
2768 | log_level = 0; | |
2769 | } | |
2770 | ||
0246e64d AS |
2771 | ret = replace_map_fd_with_map_ptr(env); |
2772 | if (ret < 0) | |
2773 | goto skip_full_check; | |
2774 | ||
9bac3d6d | 2775 | env->explored_states = kcalloc(env->prog->len, |
58e2af8b | 2776 | sizeof(struct bpf_verifier_state_list *), |
f1bca824 AS |
2777 | GFP_USER); |
2778 | ret = -ENOMEM; | |
2779 | if (!env->explored_states) | |
2780 | goto skip_full_check; | |
2781 | ||
475fb78f AS |
2782 | ret = check_cfg(env); |
2783 | if (ret < 0) | |
2784 | goto skip_full_check; | |
2785 | ||
1be7f75d AS |
2786 | env->allow_ptr_leaks = capable(CAP_SYS_ADMIN); |
2787 | ||
17a52670 | 2788 | ret = do_check(env); |
cbd35700 | 2789 | |
0246e64d | 2790 | skip_full_check: |
17a52670 | 2791 | while (pop_stack(env, NULL) >= 0); |
f1bca824 | 2792 | free_states(env); |
0246e64d | 2793 | |
9bac3d6d AS |
2794 | if (ret == 0) |
2795 | /* program is valid, convert *(u32*)(ctx + off) accesses */ | |
2796 | ret = convert_ctx_accesses(env); | |
2797 | ||
cbd35700 AS |
2798 | if (log_level && log_len >= log_size - 1) { |
2799 | BUG_ON(log_len >= log_size); | |
2800 | /* verifier log exceeded user supplied buffer */ | |
2801 | ret = -ENOSPC; | |
2802 | /* fall through to return what was recorded */ | |
2803 | } | |
2804 | ||
2805 | /* copy verifier log back to user space including trailing zero */ | |
2806 | if (log_level && copy_to_user(log_ubuf, log_buf, log_len + 1) != 0) { | |
2807 | ret = -EFAULT; | |
2808 | goto free_log_buf; | |
2809 | } | |
2810 | ||
0246e64d AS |
2811 | if (ret == 0 && env->used_map_cnt) { |
2812 | /* if program passed verifier, update used_maps in bpf_prog_info */ | |
9bac3d6d AS |
2813 | env->prog->aux->used_maps = kmalloc_array(env->used_map_cnt, |
2814 | sizeof(env->used_maps[0]), | |
2815 | GFP_KERNEL); | |
0246e64d | 2816 | |
9bac3d6d | 2817 | if (!env->prog->aux->used_maps) { |
0246e64d AS |
2818 | ret = -ENOMEM; |
2819 | goto free_log_buf; | |
2820 | } | |
2821 | ||
9bac3d6d | 2822 | memcpy(env->prog->aux->used_maps, env->used_maps, |
0246e64d | 2823 | sizeof(env->used_maps[0]) * env->used_map_cnt); |
9bac3d6d | 2824 | env->prog->aux->used_map_cnt = env->used_map_cnt; |
0246e64d AS |
2825 | |
2826 | /* program is valid. Convert pseudo bpf_ld_imm64 into generic | |
2827 | * bpf_ld_imm64 instructions | |
2828 | */ | |
2829 | convert_pseudo_ld_imm64(env); | |
2830 | } | |
cbd35700 AS |
2831 | |
2832 | free_log_buf: | |
2833 | if (log_level) | |
2834 | vfree(log_buf); | |
9bac3d6d | 2835 | if (!env->prog->aux->used_maps) |
0246e64d AS |
2836 | /* if we didn't copy map pointers into bpf_prog_info, release |
2837 | * them now. Otherwise free_bpf_prog_info() will release them. | |
2838 | */ | |
2839 | release_maps(env); | |
9bac3d6d | 2840 | *prog = env->prog; |
3df126f3 | 2841 | err_unlock: |
cbd35700 | 2842 | mutex_unlock(&bpf_verifier_lock); |
3df126f3 JK |
2843 | vfree(env->insn_aux_data); |
2844 | err_free_env: | |
2845 | kfree(env); | |
51580e79 AS |
2846 | return ret; |
2847 | } |